paperID
stringlengths
36
36
pwc_id
stringlengths
8
47
arxiv_id
stringlengths
6
16
nips_id
float64
url_abs
stringlengths
18
329
url_pdf
stringlengths
18
742
title
stringlengths
8
325
abstract
stringlengths
1
7.27k
authors
stringlengths
2
7.06k
published
stringlengths
10
10
conference
stringlengths
12
47
conference_url_abs
stringlengths
16
198
conference_url_pdf
stringlengths
27
199
proceeding
stringlengths
6
47
taskID
stringlengths
7
1.44k
areaID
stringclasses
688 values
embedding
stringlengths
9.26k
12.5k
umap_embedding
stringlengths
29
44
91de2389-e4fa-47a8-b1d8-e711c5f1c68f
neural-concept-formation-in-knowledge-graphs
null
null
https://openreview.net/forum?id=V61-62OS4mZ
https://openreview.net/pdf?id=V61-62OS4mZ
Neural Concept Formation in Knowledge Graphs
In this work, we investigate how to learn novel concepts in Knowledge Graphs (KGs) in a principled way, and how to effectively exploit them to produce more accurate neural link prediction models. Specifically, we show how concept membership relationships learned via unsupervised clustering of entities can be reified and used to augment a KG. In a thorough set of experiments, we confirm that neural link predictors trained on these augmented KGs, or in a joint Expectation-Maximization iterative scheme, can generalize better and produce more accurate predictions for infrequent relationships. For instance, our method yields relative improvements of up to 8.6% MRR on WN18RR for rare predicates, and up to 82% in small-data regimes, where the model has access to just a small subset of the training triples. Furthermore, our proposed models are able to learn meaningful concepts.
['Pasquale Minervini', 'Antonio Vergari', 'Agnieszka Dobrowolska']
2021-06-22
null
null
null
akbc-2021-10
['novel-concepts']
['reasoning']
[ 1.54296324e-01 8.67769837e-01 -6.80537403e-01 -6.45391524e-01 -2.36760721e-01 -3.32420886e-01 3.21055084e-01 4.21708882e-01 -3.44073683e-01 1.20781672e+00 1.37622237e-01 -2.46026158e-01 -3.62160653e-01 -1.14378023e+00 -1.16703248e+00 -1.55001879e-01 -5.35619795e-01 7.03670561e-01 2.03727037e-01 -1.64004818e-01 -3.04270357e-01 3.12875748e-01 -1.26758432e+00 1.52926981e-01 1.31287313e+00 7.20617175e-01 -1.09426551e-01 2.76628315e-01 -7.17689320e-02 8.86690259e-01 -1.94362357e-01 -7.91421711e-01 -3.12800333e-02 -6.36593476e-02 -1.04167628e+00 -3.78318071e-01 4.07629371e-01 -1.51912794e-01 -4.75584954e-01 9.76112962e-01 9.73426402e-02 4.17182267e-01 6.79234982e-01 -9.45565283e-01 -1.05270863e+00 1.45925033e+00 -5.18380880e-01 3.15076470e-01 1.30442619e-01 -4.22742695e-01 1.61547422e+00 -9.14651453e-01 9.03729916e-01 1.07008874e+00 7.71530569e-01 4.86143947e-01 -1.49955952e+00 -8.76754344e-01 3.35886389e-01 4.21128839e-01 -1.37951922e+00 -3.04418683e-01 6.91665113e-01 -1.40967324e-01 1.10036898e+00 -1.47485942e-01 6.11174881e-01 8.83570611e-01 -2.07112491e-01 7.68143296e-01 6.13160074e-01 -4.40173090e-01 4.40081917e-02 2.02845171e-01 5.03733337e-01 8.19510043e-01 9.97105539e-01 1.28607079e-02 -8.50353420e-01 -7.48232082e-02 4.36459810e-01 -3.25672150e-01 -2.30837017e-01 -3.84809136e-01 -1.03440022e+00 7.87420809e-01 8.80579710e-01 3.34300369e-01 -3.43248546e-01 3.60578001e-01 5.27294353e-02 7.79926032e-02 6.21679604e-01 8.45471740e-01 -8.41132820e-01 1.47370607e-01 -7.24724770e-01 2.71667819e-02 7.90864646e-01 1.06612074e+00 9.77211475e-01 1.85516886e-02 6.27979338e-02 7.68414617e-01 1.94397017e-01 3.10206950e-01 2.56271183e-01 -6.02345586e-01 6.16449296e-01 4.77654159e-01 -1.91528946e-01 -1.02812564e+00 -5.25404394e-01 -7.59335101e-01 -5.87106884e-01 -6.03836834e-01 2.17561081e-01 -4.65853423e-01 -9.66696322e-01 2.04345345e+00 1.50942802e-01 5.17866075e-01 4.00926024e-01 3.01904559e-01 7.77806520e-01 3.28952134e-01 3.65907848e-01 -1.96941271e-01 9.30962086e-01 -5.95543325e-01 -5.41981816e-01 -2.33600602e-01 9.91550744e-01 -8.27136729e-03 6.31804049e-01 1.82377279e-01 -9.30660248e-01 -4.60798115e-01 -1.07936931e+00 3.40950377e-02 -5.14004827e-01 -3.65788490e-02 1.26865208e+00 4.62745547e-01 -1.10847092e+00 8.77874911e-01 -8.69197667e-01 -4.82717782e-01 8.62363040e-01 7.14474678e-01 -3.23161334e-01 -1.86642066e-01 -1.58791888e+00 8.14880192e-01 1.20535064e+00 -5.23759723e-02 -5.09980917e-01 -9.28665102e-01 -8.35100055e-01 3.43633145e-01 6.42029226e-01 -8.62521708e-01 6.27496004e-01 -5.37919700e-01 -1.04599440e+00 5.04348338e-01 -1.45569175e-01 -8.74333978e-01 -1.43613338e-01 -4.02217209e-01 -7.43405998e-01 1.36340320e-01 -7.74390176e-02 8.94989908e-01 3.86455685e-01 -1.30753803e+00 -7.88181961e-01 -2.49446020e-01 5.09227440e-02 1.61409289e-01 -8.04922163e-01 -4.68311995e-01 -2.91059136e-01 -5.53481460e-01 1.76352739e-01 -9.02530432e-01 -2.24313065e-02 -5.77306271e-01 -7.02638030e-01 -3.60121846e-01 4.68798280e-01 -5.36043942e-01 1.11379886e+00 -1.81866813e+00 1.95922747e-01 7.23736763e-01 5.10090351e-01 1.94408774e-01 -1.31532913e-02 1.62100479e-01 -1.97800353e-01 3.95630598e-01 -2.61885691e-02 2.04668287e-02 -7.90916830e-02 5.08852065e-01 -3.69455844e-01 -2.30955869e-01 3.25003028e-01 1.28797996e+00 -1.00783503e+00 -3.01557630e-01 -4.68648106e-01 1.23460189e-01 -6.54415190e-01 -2.86641836e-01 -4.41590607e-01 3.13007176e-01 -5.49597025e-01 6.68556511e-01 2.70766467e-01 -5.48367858e-01 7.51038730e-01 -3.14791858e-01 6.59715414e-01 4.54471290e-01 -8.24497521e-01 1.45373321e+00 -1.20974056e-01 6.72743678e-01 -6.43171966e-01 -1.38152456e+00 9.33336675e-01 4.73608309e-03 3.84570032e-01 -4.42687869e-01 -2.21639216e-01 2.02256456e-01 2.14500889e-01 -2.38965303e-01 5.94687819e-01 -9.98684540e-02 2.27492034e-01 2.21830308e-01 4.33053583e-01 4.94519830e-01 3.57395291e-01 5.06468236e-01 1.07102311e+00 -1.14398122e-01 3.67573023e-01 -1.16704486e-01 1.14721619e-01 4.19650562e-02 7.10421264e-01 8.32538605e-01 2.70909697e-01 -1.37723774e-01 4.99238759e-01 -1.91684350e-01 -7.08858490e-01 -1.20476747e+00 -2.49985337e-01 1.02314448e+00 5.10911569e-02 -4.05840158e-01 -1.13610208e-01 -9.99766946e-01 4.59419042e-01 1.05909431e+00 -5.72944820e-01 -6.10589683e-01 -5.02169073e-01 -1.01878309e+00 7.17326105e-01 9.68356133e-01 3.88936669e-01 -8.61568630e-01 2.62872219e-01 1.13766767e-01 1.02514826e-01 -1.25865257e+00 1.83710784e-01 4.58674014e-01 -1.26058197e+00 -9.43304598e-01 -5.89302368e-02 -8.01832438e-01 9.06691194e-01 -1.88905209e-01 1.24588275e+00 1.49115533e-01 1.42888993e-01 1.82237208e-01 -4.32926685e-01 -2.27361023e-01 -1.09749354e-01 3.93429935e-01 4.00368392e-01 -1.90480947e-01 6.45508707e-01 -9.80652571e-01 -9.14673582e-02 -5.65535687e-02 -5.86565554e-01 -9.32201296e-02 7.59984612e-01 9.54744756e-01 5.15304089e-01 4.26933765e-01 1.13403392e+00 -1.57007623e+00 5.62486291e-01 -6.95050597e-01 -3.30995291e-01 5.01288414e-01 -1.23084748e+00 3.75261724e-01 4.41573203e-01 -3.76560539e-01 -1.13326442e+00 -1.95027851e-02 1.94081962e-01 -1.87404394e-01 2.91815810e-02 1.17867815e+00 -1.09369077e-01 -1.19122401e-01 7.84758449e-01 -6.08246848e-02 -4.76921827e-01 -4.08787698e-01 9.31821227e-01 2.41673291e-01 6.30088329e-01 -9.80669677e-01 1.20522916e+00 2.30419621e-01 6.64009377e-02 -4.72127557e-01 -1.04533947e+00 -2.78203607e-01 -8.91273022e-01 1.51716396e-01 5.55257857e-01 -1.06172097e+00 -5.65568566e-01 -8.92939791e-02 -6.94038808e-01 -3.63962263e-01 -2.97772855e-01 7.93357790e-01 -1.80166990e-01 1.71667278e-01 -5.90228856e-01 -3.91574383e-01 -1.75156951e-01 -2.76097000e-01 2.53751934e-01 5.12196124e-01 -1.41901657e-01 -1.28443336e+00 -1.53903857e-01 4.18394715e-01 -2.59381570e-02 4.86373417e-02 1.47316492e+00 -1.27834928e+00 -6.74186289e-01 3.04106716e-02 -3.29404473e-01 1.25903085e-01 3.88705194e-01 -9.04032961e-02 -8.30075085e-01 -1.09632544e-01 -8.62295330e-01 -4.33192015e-01 1.29071057e+00 1.74776539e-01 1.23253644e+00 -3.04897338e-01 -9.52184498e-01 4.93061393e-01 1.15610719e+00 9.89194512e-02 4.49044019e-01 -5.40439300e-02 9.82153535e-01 4.69989717e-01 5.79279065e-01 -5.95825166e-02 6.24829471e-01 4.12507474e-01 7.48787671e-02 1.32651240e-01 -2.24249288e-02 -5.93027651e-01 4.14429158e-02 8.15588772e-01 -5.42255640e-01 -1.30858913e-01 -1.00525105e+00 7.40320206e-01 -1.97961426e+00 -8.60420823e-01 -8.69478285e-02 2.05713773e+00 1.40398097e+00 3.02619755e-01 -1.42530024e-01 -2.39827514e-01 6.78513348e-01 -1.99671566e-01 -7.08040118e-01 1.05778769e-01 -2.64913857e-01 4.85145956e-01 6.77855670e-01 3.18840325e-01 -9.44445312e-01 1.35345662e+00 7.08312511e+00 6.73294306e-01 -5.95649600e-01 -2.96677854e-02 3.53818387e-01 1.63599011e-02 -6.42089248e-01 1.66712403e-01 -9.88406420e-01 7.53943473e-02 9.96413112e-01 -2.68907189e-01 4.44938183e-01 7.85928130e-01 -3.94810557e-01 6.89831227e-02 -1.34594655e+00 5.64482152e-01 3.73541526e-02 -1.57672620e+00 3.19259524e-01 7.48553500e-02 1.21299374e+00 2.81192549e-02 2.08814722e-03 6.60464644e-01 8.88658285e-01 -1.05016327e+00 4.90959063e-02 6.20189309e-01 3.70458782e-01 -9.39867675e-01 7.14664042e-01 2.07789108e-01 -9.38364923e-01 -1.23287246e-01 -5.76673090e-01 1.54706359e-01 -7.37582967e-02 8.69483292e-01 -1.39433372e+00 9.37294960e-01 6.01992607e-01 1.08688664e+00 -7.39404619e-01 8.16554487e-01 -5.44497728e-01 9.57522929e-01 -3.42500776e-01 -4.31663431e-02 -8.05666596e-02 2.06994981e-01 3.58432025e-01 9.13303137e-01 1.63250998e-01 2.11941361e-01 5.70286922e-02 9.96942461e-01 -6.97220802e-01 4.33510020e-02 -5.58266819e-01 -5.24971604e-01 7.08471715e-01 1.07956135e+00 -5.83750069e-01 -3.82000744e-01 -4.40154821e-01 5.64417720e-01 1.06346142e+00 8.27485025e-01 -6.86698973e-01 -3.63645315e-01 4.93245721e-01 -9.92116407e-02 6.11651063e-01 -2.56308287e-01 -2.63826758e-01 -1.25542200e+00 -1.71062350e-01 -2.99697906e-01 6.45431399e-01 -7.45586395e-01 -1.47498977e+00 1.87445119e-01 2.14035988e-01 -4.94934112e-01 -3.54823381e-01 -5.97458482e-01 -5.02124786e-01 5.44774175e-01 -1.69775343e+00 -1.18025732e+00 4.02582213e-02 4.59502250e-01 -1.96929350e-01 -4.06777531e-01 7.57498205e-01 3.21255147e-01 -5.54865956e-01 9.62671816e-01 2.62274981e-01 4.49635327e-01 5.38924098e-01 -1.38547945e+00 2.54017264e-01 7.96475768e-01 8.35458219e-01 9.81095433e-01 4.75635022e-01 -1.09748077e+00 -9.66505110e-01 -1.34278750e+00 9.52201784e-01 -4.46191311e-01 8.06317091e-01 -1.16252832e-01 -1.20312989e+00 1.26185572e+00 -2.93314606e-01 1.30156785e-01 9.23527598e-01 1.10293770e+00 -6.50380850e-01 -2.22976774e-01 -9.18996096e-01 5.33542991e-01 1.44935024e+00 -4.90660012e-01 -9.06370819e-01 3.89639974e-01 9.86829460e-01 -2.10070089e-01 -1.32389522e+00 7.24486113e-01 2.85405427e-01 -4.26681221e-01 1.12417591e+00 -1.27767611e+00 3.83528501e-01 -2.18110293e-01 -1.94154844e-01 -1.49427414e+00 -3.59053284e-01 -2.92581737e-01 -8.06422353e-01 1.31825745e+00 1.07479000e+00 -7.94658184e-01 9.69575405e-01 4.68435764e-01 -1.14480570e-01 -7.61017263e-01 -5.83444417e-01 -8.00133348e-01 5.25152199e-02 -3.71922493e-01 5.36576450e-01 1.35025561e+00 3.49829584e-01 4.41470087e-01 -2.45611489e-01 4.71508175e-01 6.68380797e-01 1.46476433e-01 4.02039200e-01 -1.63718617e+00 -4.17120069e-01 -7.55878687e-02 -4.84834641e-01 -9.68418479e-01 6.44935191e-01 -1.37089276e+00 -2.21747056e-01 -1.60218894e+00 4.37171489e-01 -8.28386426e-01 -6.96125388e-01 8.90597343e-01 -5.53557575e-01 1.96182698e-01 -1.26641899e-01 8.43285322e-02 -6.74411595e-01 5.01221120e-01 1.00946665e+00 -2.80913591e-01 -2.73510277e-01 -6.45409599e-02 -1.08305252e+00 6.70156717e-01 7.44541049e-01 -5.66651940e-01 -8.06323826e-01 -2.67252326e-01 6.82683766e-01 -3.49197924e-01 3.48483920e-01 -8.29589069e-01 3.66674691e-01 -7.15344101e-02 7.11704433e-01 -4.60257471e-01 3.52368861e-01 -5.20030975e-01 -1.52743766e-02 1.69368714e-01 -5.84572971e-01 -6.24992132e-01 2.79017925e-01 1.14597070e+00 -8.97745416e-02 -1.92695335e-01 2.35125914e-01 1.28646165e-01 -9.48861659e-01 2.37733513e-01 2.47301668e-01 2.78479904e-01 8.73433948e-01 -6.89989179e-02 -5.99299669e-01 -3.44419897e-01 -1.12681186e+00 4.47020501e-01 1.59178879e-02 2.92811215e-01 4.95429695e-01 -1.54005945e+00 -4.87403393e-01 -1.52952388e-01 2.35162929e-01 1.23846568e-01 -2.33451594e-02 7.23745942e-01 2.81478814e-03 6.68254972e-01 -2.70230789e-02 -2.88260788e-01 -9.04641330e-01 4.81456101e-01 1.36109233e-01 -3.05560052e-01 -5.91322184e-01 1.09426725e+00 -2.30183899e-02 -3.71556967e-01 1.21164225e-01 -5.31019807e-01 -3.54247123e-01 -5.00646606e-03 1.14729084e-01 3.06716021e-02 -1.34680584e-01 -3.26807499e-01 -2.74597287e-01 2.62081534e-01 -5.15764534e-01 3.43700886e-01 1.46795702e+00 9.15250406e-02 2.66180728e-02 3.29140067e-01 8.28970134e-01 1.71726003e-01 -9.23688054e-01 -6.11876130e-01 4.13673520e-01 -1.16939045e-01 1.01354485e-02 -9.56775308e-01 -1.06665838e+00 2.95451552e-01 -1.16858445e-01 1.51363283e-01 8.96863818e-01 4.24597412e-01 6.72755897e-01 1.01220989e+00 3.42504978e-01 -1.00818384e+00 -9.78352949e-02 4.04584169e-01 2.96439052e-01 -1.13240910e+00 2.15809584e-01 -8.34400356e-01 -5.33152997e-01 9.03493345e-01 7.60529160e-01 6.26539737e-02 7.43508875e-01 -1.69315204e-01 -6.46621466e-01 -3.40695024e-01 -8.76147807e-01 -3.18750352e-01 6.22526228e-01 9.33897316e-01 2.59395123e-01 2.29645669e-01 -3.21698897e-02 8.84250641e-01 -3.22278649e-01 -1.15301743e-01 3.41280222e-01 4.52435374e-01 -5.20468473e-01 -1.04318523e+00 2.18967631e-01 9.36176836e-01 -1.98886320e-01 -3.69702935e-01 -4.25455540e-01 8.59742522e-01 9.27565545e-02 7.64972091e-01 2.55066180e-03 -6.81974113e-01 2.15373635e-02 4.73074853e-01 3.95681113e-01 -9.20711756e-01 2.20516864e-02 -5.97416639e-01 6.34780288e-01 -3.86187017e-01 -6.01859868e-01 -5.40972769e-01 -1.65364695e+00 -2.44568005e-01 -5.61599672e-01 1.68804213e-01 8.18842426e-02 1.25267923e+00 4.31552380e-01 6.51032031e-01 1.66343600e-01 3.28723341e-02 -3.05556953e-01 -9.42459345e-01 -8.01519036e-01 2.52806723e-01 -1.65566877e-01 -9.11929011e-01 -9.35374573e-02 5.56564108e-02]
[8.88937759399414, 7.979780673980713]
5657ebcc-0485-4be6-9c31-8ca3fb8c111a
a-large-scale-study-of-language-models-for
1804.01849
null
http://arxiv.org/abs/1804.01849v1
http://arxiv.org/pdf/1804.01849v1.pdf
A Large-Scale Study of Language Models for Chord Prediction
We conduct a large-scale study of language models for chord prediction. Specifically, we compare N-gram models to various flavours of recurrent neural networks on a comprehensive dataset comprising all publicly available datasets of annotated chords known to us. This large amount of data allows us to systematically explore hyper-parameter settings for the recurrent neural networks---a crucial step in achieving good results with this model class. Our results show not only a quantitative difference between the models, but also a qualitative one: in contrast to static N-gram models, certain RNN configurations adapt to the songs at test time. This finding constitutes a further step towards the development of chord recognition systems that are more aware of local musical context than what was previously possible.
['Filip Korzeniowski', 'David R. W. Sears', 'Gerhard Widmer']
2018-04-05
null
null
null
null
['chord-recognition']
['audio']
[ 1.31269753e-01 -1.49378553e-01 -1.12563297e-02 -5.72638437e-02 -6.83169484e-01 -1.00801635e+00 3.60530078e-01 -4.51352932e-02 -5.17974138e-01 4.59509373e-01 5.72915971e-01 -2.67934382e-01 -1.70638099e-01 -6.52473748e-01 -2.88768828e-01 -5.65615356e-01 -3.11165273e-01 6.61789775e-01 5.04404008e-01 -7.02310681e-01 4.10270900e-01 5.75588405e-01 -1.48328602e+00 4.05647784e-01 -7.73190930e-02 5.12066960e-01 1.17409259e-01 1.04943848e+00 2.16723934e-01 8.93398046e-01 -5.56805134e-01 -2.02781036e-01 2.07842022e-01 -7.14278162e-01 -9.40511882e-01 -4.52200830e-01 1.72453314e-01 2.57465690e-01 -3.03477645e-01 7.09769487e-01 7.43931890e-01 4.58440155e-01 5.56505919e-01 -4.78914171e-01 -3.54028970e-01 1.25539529e+00 -1.11388443e-02 3.21812689e-01 1.89961195e-01 2.43172765e-01 1.46057296e+00 -5.38937986e-01 7.32476175e-01 8.33171487e-01 1.01066327e+00 6.32754445e-01 -1.08668840e+00 -6.31997168e-01 -2.42333099e-01 1.52848348e-01 -1.28907251e+00 -4.97175694e-01 9.56564546e-01 -4.10863191e-01 1.11759186e+00 3.86183262e-01 9.78722215e-01 1.09604633e+00 -6.82775006e-02 5.41351736e-01 9.20548081e-01 -6.83233678e-01 -8.68554935e-02 -2.07361609e-01 1.25843525e-01 3.75627398e-01 -3.64612699e-01 2.37540185e-01 -7.75011003e-01 -1.15818709e-01 9.25716758e-01 -5.61442435e-01 -1.81221485e-01 -2.28258207e-01 -1.30872965e+00 6.02233946e-01 1.07921489e-01 7.85908222e-01 -2.01235622e-01 1.69225007e-01 8.56428504e-01 5.39795220e-01 6.70252219e-02 1.01381326e+00 -7.46257603e-01 -7.83511937e-01 -1.04718947e+00 3.43406677e-01 9.68120754e-01 2.94854850e-01 1.75931394e-01 4.91360992e-01 -4.52591032e-02 1.04295957e+00 -1.34237587e-01 -1.68231111e-02 9.45104241e-01 -9.13604319e-01 1.33674458e-01 3.14702779e-01 -5.50451159e-01 -7.91766524e-01 -4.97685701e-01 -8.62807214e-01 -5.97969592e-01 7.03033581e-02 6.68030500e-01 9.06473026e-04 -4.53650773e-01 1.85672808e+00 -1.96430430e-01 8.07841643e-05 -2.83565577e-02 6.05708420e-01 5.20456493e-01 3.73366296e-01 -2.39853665e-01 -1.61607891e-01 1.05505860e+00 -7.78208792e-01 -3.00843954e-01 1.48419246e-01 7.49439418e-01 -9.21212554e-01 1.33843100e+00 7.64374793e-01 -1.06492615e+00 -6.63830340e-01 -1.03292084e+00 1.19196475e-01 -2.18298241e-01 3.51417325e-02 6.90066636e-01 6.33679271e-01 -1.10804474e+00 9.35379028e-01 -4.85601872e-01 -4.15473491e-01 -2.86976337e-01 4.23222244e-01 -1.18279994e-01 6.62367046e-01 -1.28174007e+00 7.28535354e-01 6.60140872e-01 1.18579485e-01 -7.62506425e-01 -4.29677695e-01 -2.94580877e-01 1.44729227e-01 4.31812197e-01 -2.07738236e-01 1.69718277e+00 -9.34323132e-01 -1.63722897e+00 8.35450828e-01 1.76936179e-01 -5.94049215e-01 2.39139780e-01 -5.00658937e-02 -3.07195812e-01 -2.15690434e-01 -3.82817447e-01 3.91357422e-01 2.51582146e-01 -9.06759024e-01 -3.22900534e-01 -9.02993754e-02 -1.65730566e-01 8.51861835e-02 -2.94407040e-01 2.31358141e-01 -4.36261892e-01 -1.01730382e+00 2.81425156e-02 -1.23689067e+00 -2.92337954e-01 -7.01651454e-01 -5.30391216e-01 -2.60758042e-01 3.46663535e-01 -5.02110064e-01 1.62622869e+00 -1.90039551e+00 3.46180618e-01 3.86857510e-01 -2.19816282e-01 3.65125477e-01 -1.53313756e-01 7.70112514e-01 -2.67436892e-01 2.75151227e-02 -3.52043547e-02 -9.20425132e-02 -1.72214225e-01 1.05947345e-01 -6.40409052e-01 -4.75772917e-02 -2.46950328e-01 8.10916364e-01 -5.68893194e-01 -1.92889050e-01 -1.00450426e-01 3.93333852e-01 -6.09071732e-01 7.20130354e-02 -4.35132653e-01 3.96552771e-01 -7.13579953e-02 4.15550113e-01 -4.38571334e-01 -7.98689872e-02 4.80161756e-01 1.30411357e-01 -1.94330007e-01 7.39332974e-01 -9.54098642e-01 1.67928040e+00 -2.78918326e-01 7.40936100e-01 -5.71908057e-01 -8.07934761e-01 1.04196012e+00 5.75052917e-01 4.68866587e-01 -4.35469419e-01 1.51401252e-01 2.85347342e-01 6.72387838e-01 -1.21668115e-01 1.00752664e+00 -3.43690187e-01 -2.81916618e-01 5.77535987e-01 2.24555969e-01 6.07091449e-02 3.63688827e-01 -1.53261587e-01 1.03517532e+00 2.37000704e-01 3.40056062e-01 -1.37236759e-01 3.54095191e-01 -1.91935956e-01 6.24972939e-01 9.50802624e-01 1.11478731e-01 7.66633928e-01 5.56288481e-01 -5.71411431e-01 -1.04244328e+00 -6.75786614e-01 3.62609513e-02 1.45484245e+00 -6.73551619e-01 -8.64257991e-01 -6.41723931e-01 -2.46246696e-01 -3.98431659e-01 3.36100370e-01 -7.53008842e-01 -6.83619082e-02 -8.86928499e-01 -4.74218041e-01 1.28907764e+00 6.87972128e-01 -1.40085826e-02 -1.68874574e+00 -6.08267546e-01 3.48665357e-01 -1.20301964e-02 -8.26220810e-01 -2.77439386e-01 5.85840702e-01 -8.87646258e-01 -1.09604967e+00 -5.43635309e-01 -7.75441527e-01 -1.73591152e-01 -2.37055361e-01 1.29162216e+00 1.79257452e-01 -1.31619588e-01 4.23982106e-02 -4.41678762e-01 -3.89242917e-01 -6.05476499e-01 7.28075385e-01 1.28893465e-01 -4.91069645e-01 2.37437233e-01 -8.88184011e-01 -2.17262372e-01 1.70029446e-01 -6.84972346e-01 -1.17349967e-01 5.13470888e-01 6.23253286e-01 5.68349004e-01 -1.46878988e-01 6.37597442e-01 -9.32132661e-01 9.64036226e-01 -1.83061033e-01 -4.69675690e-01 2.97120005e-01 -4.46417749e-01 -3.71290706e-02 8.16959023e-01 -7.14485765e-01 -4.78425890e-01 1.20366784e-02 -4.29954588e-01 -1.29997984e-01 -2.70289667e-02 7.08586097e-01 9.93736014e-02 7.82076493e-02 8.23793948e-01 3.16028327e-01 -3.50252122e-01 -7.11565077e-01 2.12322235e-01 4.77960408e-01 8.43864858e-01 -8.30152750e-01 7.26119399e-01 -1.31604418e-01 4.79460731e-02 -1.04737043e+00 -5.30141652e-01 -3.49541247e-01 -9.61238086e-01 -3.39869678e-01 5.82751751e-01 -5.37012458e-01 -8.16820800e-01 3.72393847e-01 -7.17267096e-01 -6.28719032e-01 -4.37180579e-01 3.18538249e-01 -7.68511176e-01 3.46652836e-01 -1.02327251e+00 -9.52465951e-01 -3.21774423e-01 -9.87948298e-01 4.55711961e-01 7.10207149e-02 -8.12365949e-01 -1.01405632e+00 7.39211857e-01 1.31844863e-01 4.00052249e-01 -5.02625406e-02 1.07993937e+00 -1.00088215e+00 -3.87870342e-01 -1.27801597e-01 3.72530282e-01 2.38712966e-01 2.41400618e-02 3.54299575e-01 -1.05175054e+00 -1.25978604e-01 -2.64176071e-01 -5.76773703e-01 8.65834773e-01 2.47307256e-01 1.15911746e+00 4.92239743e-02 1.31307468e-01 5.02974987e-01 9.87766087e-01 3.10376912e-01 4.64264631e-01 6.18092537e-01 5.52635372e-01 3.70370358e-01 4.97695625e-01 2.58715421e-01 1.49425283e-01 9.69816685e-01 5.90875633e-02 2.45660678e-01 -6.71810806e-02 -4.67633575e-01 3.17761928e-01 1.43757463e+00 -4.91490573e-01 -6.48636594e-02 -1.10645211e+00 6.52202904e-01 -1.77166343e+00 -1.11391151e+00 2.10660651e-01 2.28169131e+00 9.28531468e-01 4.84306037e-01 8.08876038e-01 6.88983083e-01 1.98852181e-01 3.27433228e-01 -2.51252741e-01 -5.71826398e-01 -1.96424335e-01 5.24115860e-01 9.52888876e-02 2.99970359e-01 -8.80585134e-01 1.18838775e+00 7.55928850e+00 8.39303970e-01 -1.27428973e+00 -3.88047427e-01 1.69180840e-01 -3.80335689e-01 -2.53506154e-01 1.27154395e-01 -6.28138244e-01 1.87303409e-01 1.38823044e+00 -3.39375176e-02 7.20916152e-01 6.40985131e-01 1.03702746e-01 3.53453577e-01 -1.02559662e+00 4.82551485e-01 2.92912107e-02 -1.55775607e+00 9.13169608e-02 1.32509479e-02 3.93635333e-01 3.39759052e-01 -9.62274848e-04 4.83925819e-01 3.67505044e-01 -1.09979773e+00 7.89215207e-01 5.43741763e-01 5.57027459e-01 -9.38796282e-01 4.18402046e-01 3.42730343e-01 -1.18788123e+00 -1.59711331e-01 -2.39133105e-01 -3.07849795e-01 1.63378894e-01 7.58380666e-02 -9.83530283e-01 3.48070592e-01 6.71414375e-01 6.09257102e-01 -7.50527442e-01 9.89582717e-01 -1.27165481e-01 1.04835403e+00 -1.72469050e-01 -1.09156139e-01 7.97745734e-02 -9.20548365e-02 6.72003865e-01 1.43422687e+00 2.01341629e-01 -1.23447493e-01 -1.01718847e-02 4.02876645e-01 6.99502081e-02 3.57700139e-01 -5.80821395e-01 -4.07136738e-01 3.74957442e-01 1.00806022e+00 -8.68623853e-01 -3.92725244e-02 -5.55245718e-03 5.36551952e-01 6.12151980e-01 7.98527524e-02 -3.01592112e-01 -3.43313724e-01 5.09807825e-01 7.92261735e-02 3.27811599e-01 -5.92334926e-01 -1.70956448e-01 -9.09152508e-01 -2.00780973e-01 -1.27434647e+00 4.73877251e-01 -6.84005737e-01 -9.88519073e-01 8.59604418e-01 -1.90172806e-01 -1.06773221e+00 -8.17373037e-01 -6.09241247e-01 -8.39829028e-01 6.58229232e-01 -9.38885629e-01 -1.19272768e+00 4.21873510e-01 3.51590723e-01 5.33148408e-01 -5.53187788e-01 1.32440269e+00 1.67767763e-01 -3.93989682e-01 6.17184997e-01 1.79106399e-04 4.72655654e-01 7.28653729e-01 -1.23858893e+00 6.97629094e-01 5.98264337e-01 1.01370561e+00 9.56723034e-01 8.02963972e-01 -3.76265883e-01 -9.52061892e-01 -5.37969589e-01 9.60451066e-01 -4.78487790e-01 8.89047265e-01 -2.75875181e-01 -1.17277741e+00 8.22746575e-01 1.67076111e-01 -4.86440748e-01 1.13518798e+00 9.38667417e-01 -3.92283440e-01 1.04052924e-01 -1.92995802e-01 6.95086896e-01 9.71167505e-01 -1.01573765e+00 -7.84851909e-01 -8.43967572e-02 5.85991740e-01 -3.31358045e-01 -8.97217333e-01 5.97049057e-01 1.01893449e+00 -1.23872781e+00 8.72541904e-01 -9.03329253e-01 4.13562268e-01 -1.60063550e-01 -2.74443120e-01 -1.19440496e+00 -3.62161666e-01 -7.26280093e-01 4.81784418e-02 1.29682052e+00 6.42105997e-01 -1.16100617e-01 8.83654892e-01 6.74606338e-02 -2.82441616e-01 -8.22256267e-01 -8.00681651e-01 -6.99531853e-01 3.33316743e-01 -8.15773547e-01 3.44906628e-01 9.74268377e-01 3.18335950e-01 3.67750138e-01 -5.30321956e-01 -3.50940138e-01 1.07630098e-03 2.20398858e-01 8.63908589e-01 -1.30835080e+00 -9.06684756e-01 -7.54862368e-01 -4.61777747e-01 -7.83671856e-01 2.46254325e-01 -9.48615611e-01 -1.18338346e-01 -7.23545253e-01 5.96893616e-02 -4.11469728e-01 -7.62201667e-01 5.83679736e-01 1.13261119e-01 6.48772478e-01 3.41531724e-01 2.32894674e-01 -4.62056369e-01 2.45828047e-01 7.96937764e-01 2.24757209e-01 -4.61877704e-01 2.75065243e-01 -6.26478374e-01 9.54468191e-01 1.06193781e+00 -4.47629899e-01 -4.60998237e-01 -1.36031032e-01 6.89333975e-01 2.59438127e-01 2.50643380e-02 -1.25274551e+00 2.90019393e-01 -5.60038127e-02 1.89638123e-01 -6.72077656e-01 4.59571749e-01 -3.36191446e-01 1.91931814e-01 3.25402886e-01 -8.31559181e-01 2.88229078e-01 3.35138917e-01 3.11683714e-01 -2.60902852e-01 -3.29675347e-01 5.26366949e-01 -3.10753971e-01 -6.28391087e-01 -1.30236819e-01 -6.72412574e-01 5.86912669e-02 4.23199624e-01 -2.28094757e-01 9.60236862e-02 -5.09791374e-01 -9.74203885e-01 -3.28663975e-01 4.66290027e-01 5.98022461e-01 1.15824610e-01 -1.21648145e+00 -4.98653859e-01 2.27820054e-01 1.03488810e-01 -4.07822728e-01 2.09922642e-02 5.74408650e-01 -2.51937836e-01 5.01643360e-01 -2.90650725e-01 -3.24429780e-01 -1.62317848e+00 2.16879576e-01 3.57554018e-01 -5.06654739e-01 -5.72949350e-01 7.55040407e-01 -3.00138086e-01 -7.85525084e-01 2.51042128e-01 -2.59388566e-01 -4.26375389e-01 1.00682028e-01 3.42411101e-01 1.68746002e-02 -1.45643363e-02 -7.48153687e-01 -2.74724197e-02 5.26893318e-01 -4.53017242e-02 -3.36473346e-01 1.34708679e+00 3.15012008e-01 -1.29530460e-01 1.43848121e+00 6.52405858e-01 3.36430907e-01 -8.94677520e-01 -2.01316103e-01 3.60525936e-01 -4.65032570e-02 -3.82473290e-01 -8.56681943e-01 -7.72314906e-01 8.77527058e-01 1.84073180e-01 3.53127569e-01 9.47556376e-01 -1.20689698e-01 5.43528736e-01 7.21732199e-01 2.07541063e-01 -1.09010756e+00 3.67250778e-02 1.00420213e+00 7.63144374e-01 -7.04603553e-01 -1.33082688e-01 3.38183105e-01 -6.97906375e-01 1.24561524e+00 4.04177845e-01 -3.12953860e-01 4.16830540e-01 2.12925196e-01 3.86801958e-01 -1.22123308e-01 -1.17795289e+00 -1.12720490e-01 4.39107269e-01 2.82877505e-01 9.95818913e-01 -1.13908820e-01 -1.56650499e-01 6.82784200e-01 -9.07116890e-01 -7.10807294e-02 5.29217720e-01 5.20380735e-01 -3.43502790e-01 -1.51848376e+00 -1.31915957e-01 1.91757634e-01 -6.79947436e-01 -7.04689696e-02 -8.03751647e-01 9.47030008e-01 -1.82695404e-01 6.27866030e-01 -1.75879925e-01 -7.80264020e-01 2.45638371e-01 4.12133574e-01 4.40176100e-01 -5.75708926e-01 -1.30729449e+00 1.46586940e-01 2.46196881e-01 -2.57506311e-01 -3.02654803e-01 -7.00523913e-01 -1.12439835e+00 -5.71960509e-02 -2.42751345e-01 3.61497730e-01 4.89206761e-01 7.88432837e-01 -7.64882639e-02 7.07521200e-01 4.37237352e-01 -9.95050251e-01 -4.30039406e-01 -1.06719518e+00 -6.58716321e-01 1.90137908e-01 1.03192382e-01 -1.49010435e-01 -1.97172523e-01 -1.08706973e-01]
[15.893318176269531, 5.330941200256348]
6bb64aa4-f278-433d-8d44-b75d3ffadc49
consistent-and-symmetry-preserving-data
2104.11578
null
https://arxiv.org/abs/2104.11578v1
https://arxiv.org/pdf/2104.11578v1.pdf
Consistent and symmetry preserving data-driven interface reconstruction for the level-set method
Recently, machine learning has been used to substitute parts of conventional computational fluid dynamics, e.g. the cell-face reconstruction in finite-volume solvers or the curvature computation in the Volume-of-Fluid (VOF) method. The latter showed improvements in terms of accuracy for coarsely resolved interfaces, however at the expense of convergence and symmetry. In this work, a combined approach is proposed, adressing the aforementioned shortcomings. We focus on interface reconstruction (IR) in the level-set method, i.e. the computation of the volume fraction and apertures. The combined model consists of a classification neural network, that chooses between the conventional (linear) IR and the neural network IR depending on the local interface resolution. The proposed approach improves accuracy for coarsely resolved interfaces and recovers the conventional IR for high resolutions, yielding first order overall convergence. Symmetry is preserved by mirroring and rotating the input level-set grid and subsequently averaging the predictions. The combined model is implemented into a CFD solver and demonstrated for two-phase flows. Furthermore, we provide details of floating point symmetric implementation and computational efficiency.
['Nikolaus Adams', 'Deniz A. Bezgin', 'Aaron B. Buhendwa']
2021-04-23
null
null
null
null
['face-reconstruction']
['computer-vision']
[ 1.12194330e-01 -1.17852084e-01 4.91536885e-01 2.18671620e-01 -6.04186475e-01 -1.84181243e-01 6.49611294e-01 3.24491858e-01 -3.71626735e-01 9.88221288e-01 -2.43738443e-01 -3.24923217e-01 -3.40482146e-01 -9.80118155e-01 -4.43558455e-01 -1.04268193e+00 1.76164676e-02 7.28078723e-01 1.59679070e-01 -2.62490511e-01 3.82438898e-01 8.94627273e-01 -1.89849782e+00 3.27333897e-01 1.10450327e+00 1.22653198e+00 -2.15366483e-01 6.90595329e-01 -3.64773512e-01 6.21974707e-01 -8.50257427e-02 1.42021403e-01 2.05133572e-01 -2.13170886e-01 -6.50380492e-01 -1.56142637e-01 4.37544197e-01 -1.95397720e-01 2.90798634e-01 5.66363811e-01 4.33526844e-01 3.89648944e-01 9.61053610e-01 -7.73768306e-01 1.29567146e-01 -1.65334642e-02 -3.96267146e-01 -9.27667171e-02 3.99685472e-01 -3.48245562e-03 5.72766006e-01 -1.27787817e+00 5.43397427e-01 8.96344304e-01 9.03654635e-01 2.14743987e-01 -1.34385800e+00 -1.31554723e-01 -3.64412695e-01 -3.08443040e-01 -1.40134358e+00 -1.99051753e-01 9.42138553e-01 -1.15583324e+00 1.02973413e+00 5.01426399e-01 7.79581904e-01 1.84860945e-01 4.99231398e-01 -1.88859403e-01 1.22431850e+00 -5.04071951e-01 5.42737901e-01 3.79016727e-01 -4.70780395e-02 4.85610038e-01 5.46492696e-01 3.26729476e-01 -1.08214796e-01 -4.79813367e-01 9.95769262e-01 -2.97847509e-01 -3.73061538e-01 -6.70880616e-01 -6.52650952e-01 8.95210862e-01 2.29594335e-01 6.68829322e-01 -3.32514018e-01 -3.45846862e-01 2.55787343e-01 -7.71080563e-03 9.35268700e-01 8.22741210e-01 -4.56334472e-01 2.04645768e-02 -1.30791402e+00 6.74820304e-01 1.08451462e+00 1.18653819e-01 9.31366265e-01 2.97377348e-01 3.73376422e-02 4.60776269e-01 2.35013634e-01 3.01102728e-01 2.78150946e-01 -1.10331893e+00 3.33109200e-02 8.19178164e-01 4.37752932e-01 -1.07415259e+00 -5.07425606e-01 -3.52176458e-01 -9.70098555e-01 9.04226899e-01 7.03760445e-01 -3.15561473e-01 -5.10869563e-01 9.92363572e-01 7.77220666e-01 2.34230205e-01 1.59028485e-01 9.85309422e-01 6.93901241e-01 6.29844546e-01 -3.03210132e-02 -4.25705492e-01 9.71804202e-01 -9.51465189e-01 -6.38393998e-01 3.59359294e-01 8.11798453e-01 -7.13849247e-01 5.00077069e-01 4.24119532e-01 -1.20716250e+00 -6.52891338e-01 -8.87475610e-01 2.71202996e-02 -5.46924472e-01 7.19068274e-02 2.50991881e-01 4.40158606e-01 -1.14178061e+00 1.37665772e+00 -6.76310718e-01 1.13667533e-01 -8.00511837e-02 2.65640944e-01 -3.17656726e-01 5.26178956e-01 -9.37568724e-01 7.74172068e-01 -9.16509479e-02 1.30205497e-01 -2.43508201e-02 -1.25612938e+00 -9.13988888e-01 5.58354631e-02 -2.03486323e-01 -7.91058958e-01 8.47517908e-01 -1.01203489e+00 -1.87394130e+00 6.88349903e-01 -3.43091160e-01 -3.24416369e-01 1.07059586e+00 -9.16980058e-02 4.90591154e-02 1.66835994e-01 -1.76498398e-01 1.33982733e-01 6.11335039e-01 -1.56495810e+00 -3.23858678e-01 -1.30515575e-01 -2.61280417e-01 1.40130475e-01 1.69072255e-01 -5.47362804e-01 3.67898285e-01 -3.36917102e-01 1.57858893e-01 -6.49264991e-01 -3.61759156e-01 4.75339182e-02 1.22209385e-01 -5.08567095e-02 7.75145590e-01 -8.39189768e-01 1.19119000e+00 -1.70974839e+00 2.24859998e-01 3.50104958e-01 2.68997788e-01 4.30273086e-01 5.06566763e-01 5.52715898e-01 -3.50543141e-01 -2.03106329e-01 -5.84249973e-01 -4.54592913e-01 -4.22007501e-01 -5.05739935e-02 -1.98942333e-01 7.19124734e-01 2.95676112e-01 4.87152040e-01 -4.86600131e-01 -3.17172021e-01 5.03639042e-01 8.65257680e-01 -7.36627042e-01 2.22098425e-01 5.97265139e-02 1.06279719e+00 -1.79102540e-01 1.84381008e-01 1.03979683e+00 -1.86947271e-01 7.82901794e-02 -2.45503724e-01 -9.60739791e-01 1.67131528e-01 -1.54734588e+00 1.13150799e+00 -6.91898942e-01 2.12811157e-01 6.38716161e-01 -1.14134967e+00 1.15341783e+00 5.09080470e-01 8.87130678e-01 -5.70134878e-01 1.47258595e-01 7.56048918e-01 -6.96991235e-02 -3.58806878e-01 5.47833145e-01 -4.12492096e-01 3.56232017e-01 4.11766171e-02 -2.44372889e-01 -3.04798573e-01 2.18763370e-02 -3.46086621e-01 5.11733413e-01 3.67215186e-01 5.16053855e-01 -7.58919716e-01 1.21955407e+00 2.28386834e-01 1.02629520e-01 2.45701551e-01 1.56602576e-01 7.11475074e-01 5.02999842e-01 -7.54022360e-01 -1.15166295e+00 -6.07390285e-01 -6.29577637e-01 4.64175910e-01 1.22591652e-01 -4.60621528e-02 -9.53781247e-01 1.65910631e-01 3.42684031e-01 4.69855398e-01 -4.61129785e-01 2.13651165e-01 -9.79862094e-01 -5.54182351e-01 2.32918616e-02 1.77698061e-01 3.07579368e-01 -9.51329768e-01 -8.02128196e-01 3.89127165e-01 2.59563297e-01 -8.77988815e-01 2.95240968e-01 5.49443252e-02 -1.20619023e+00 -1.17076540e+00 -6.79642677e-01 -4.54469323e-01 3.37697446e-01 -2.32508644e-01 1.08172274e+00 3.08061928e-01 -1.23417050e-01 2.42087647e-01 -9.62030515e-02 2.20233947e-01 -5.47651768e-01 -1.83144503e-03 5.89210875e-02 8.47118422e-02 -2.84474134e-01 -6.62954688e-01 -5.42983592e-01 7.28274658e-02 -5.95667720e-01 1.68128356e-01 9.12860632e-02 8.61060143e-01 4.83620048e-01 -1.96481854e-01 4.65341151e-01 -8.37299287e-01 3.35864842e-01 -2.88597763e-01 -8.81592453e-01 -2.63867676e-01 -6.98983610e-01 -3.28707770e-02 8.59770656e-01 -2.33295694e-01 -1.09794378e+00 -4.23294492e-02 -5.41340530e-01 -5.03769577e-01 -2.70428151e-01 1.66551456e-01 1.27486527e-01 -6.62740171e-01 4.40303952e-01 8.83566141e-02 1.31633386e-01 -7.41711497e-01 -2.29484528e-01 4.80307192e-01 1.07400388e-01 -8.29617798e-01 5.44320524e-01 4.97001678e-01 4.47654128e-01 -1.17002559e+00 -1.64283179e-02 -4.20632899e-01 -8.97548795e-01 -3.65801036e-01 6.77710176e-01 -5.00321388e-01 -8.81947875e-01 5.77338636e-01 -1.31250751e+00 -2.92710632e-01 -6.39462471e-01 4.98432487e-01 -5.00759721e-01 3.04313660e-01 -8.16922903e-01 -1.05679178e+00 -3.96097779e-01 -1.29219139e+00 1.13629031e+00 2.70543337e-01 -1.70054168e-01 -1.27007353e+00 3.24479610e-01 7.02790767e-02 6.80333555e-01 6.60041630e-01 8.39184940e-01 -3.37322176e-01 -2.97749251e-01 1.89847443e-02 -5.99917173e-02 4.64505656e-03 -9.41178650e-02 1.83734670e-01 -1.08590186e+00 -2.42498145e-01 3.53382438e-01 3.28316242e-01 7.27457225e-01 6.13495708e-01 7.31608868e-01 -3.37363780e-01 -4.67186660e-01 7.12076604e-01 1.84181523e+00 1.92569822e-01 3.59282136e-01 1.04664095e-01 5.73020339e-01 7.29007602e-01 3.13693911e-01 6.67870164e-01 -1.12716913e-01 8.22502136e-01 7.35613629e-02 -4.41821128e-01 -2.59115696e-02 1.72452942e-01 -1.03662260e-01 6.37648523e-01 -5.11417389e-01 1.87399954e-01 -1.04547334e+00 2.72228807e-01 -1.43303549e+00 -6.55588865e-01 -5.66073179e-01 2.31816769e+00 5.90651929e-01 1.50842369e-02 -9.43322554e-02 4.61712569e-01 5.36690533e-01 -1.03423998e-01 -1.96831107e-01 -9.64996457e-01 1.69985309e-01 4.59779054e-01 3.28610778e-01 1.23098481e+00 -1.09935427e+00 3.92361462e-01 5.46131992e+00 4.90444630e-01 -1.53164375e+00 1.45409882e-01 5.82243025e-01 1.78011909e-01 -1.16623163e-01 -1.74111605e-01 -7.99925268e-01 4.01002139e-01 9.02090549e-01 1.22184142e-01 4.76617396e-01 7.04280376e-01 4.13623989e-01 -3.17235023e-01 -6.49363279e-01 5.59578180e-01 -1.52069628e-01 -1.49597108e+00 7.46134948e-03 -3.31947557e-03 6.44937098e-01 -3.83251935e-01 -3.94848585e-01 1.23218365e-01 -5.77891946e-01 -1.02660811e+00 4.91897583e-01 7.46535301e-01 1.01386750e+00 -5.18141150e-01 9.40890491e-01 4.55588132e-01 -1.56870842e+00 1.56509534e-01 1.58290327e-01 -5.26232183e-01 2.83630669e-01 6.80291891e-01 -4.70221251e-01 8.24147999e-01 5.31601012e-01 4.03347462e-01 -1.04442790e-01 8.44738603e-01 6.48158908e-01 2.93361962e-01 -4.78617281e-01 1.57522857e-01 2.06377119e-01 -9.61468816e-01 7.09875286e-01 1.09786582e+00 6.12812757e-01 1.33527577e-01 2.44233832e-01 1.09231710e+00 5.87625265e-01 3.71902138e-01 -5.50471187e-01 4.53880578e-01 -1.72714721e-02 1.35677111e+00 -7.10014403e-01 -1.98679194e-01 -2.15401977e-01 3.19189698e-01 1.55200496e-01 3.45572680e-01 -5.44295251e-01 -1.59499004e-01 8.30243289e-01 8.71468902e-01 3.02242428e-01 -1.92760572e-01 -4.74809557e-01 -7.65425444e-01 -1.09394915e-01 -3.40819627e-01 6.76535070e-02 -4.18290973e-01 -1.13375556e+00 7.03493834e-01 1.12286121e-01 -1.07084966e+00 -2.74764866e-01 -9.73246753e-01 -5.31640708e-01 1.24298632e+00 -1.74723339e+00 -8.38295043e-01 -2.05485955e-01 1.68374628e-01 1.88627660e-01 2.02287331e-01 7.98647940e-01 4.97321069e-01 -2.03097805e-01 1.26976371e-01 1.77945077e-01 -9.86803770e-02 2.96676695e-01 -1.28542149e+00 -1.72281221e-01 3.59918714e-01 -8.67231727e-01 3.87428284e-01 9.00365114e-01 -7.89895535e-01 -1.28013134e+00 -9.27214861e-01 1.00286746e+00 2.02254519e-01 4.29718167e-01 -3.94165590e-02 -1.51426983e+00 3.85439619e-02 5.00513092e-02 3.05326939e-01 2.79071003e-01 -3.52427155e-01 1.65047497e-01 -1.01714976e-01 -1.35573816e+00 2.70764083e-01 3.65263551e-01 -1.99469343e-01 -4.30824310e-01 -1.19760484e-02 3.71041775e-01 -7.54552722e-01 -1.07970548e+00 7.27293551e-01 5.12830317e-01 -1.18713188e+00 9.28911448e-01 -2.35626221e-01 4.19851601e-01 -3.96170080e-01 1.61671326e-01 -9.38355684e-01 -2.01207936e-01 -4.33934987e-01 -8.21843371e-02 1.02226472e+00 1.92949578e-01 -1.09040034e+00 6.37931705e-01 5.45643449e-01 -1.33440599e-01 -1.23076415e+00 -1.50348830e+00 -1.85672075e-01 5.45462847e-01 -7.88531080e-02 1.64044738e-01 9.78516281e-01 -5.66606596e-02 -4.83253747e-02 -1.15459785e-01 2.29567513e-01 5.37786901e-01 3.24522883e-01 3.62445176e-01 -1.67141914e+00 -1.95723385e-01 -4.54733849e-01 -1.30179031e-02 -5.30850112e-01 1.69312015e-01 -7.15298951e-01 -7.92973787e-02 -1.22251797e+00 -5.58133125e-01 -7.49243259e-01 2.71045268e-01 -1.71806365e-01 1.03073306e-01 2.36479729e-01 -2.08523408e-01 8.43345374e-02 1.41407013e-01 3.87220562e-01 1.42852592e+00 3.67585570e-01 -6.17436349e-01 -9.67595801e-02 2.20067203e-02 7.76267707e-01 8.57943892e-01 -1.49339184e-01 3.08892727e-01 2.60464996e-01 1.13701485e-02 3.22793752e-01 3.93094957e-01 -1.39075053e+00 2.14461610e-01 1.41836241e-01 5.16812086e-01 -4.55879182e-01 3.74019653e-01 -9.14206624e-01 3.86210740e-01 6.48024201e-01 -1.53398767e-01 -8.01497772e-02 4.73752767e-01 1.25806570e-01 -1.93179250e-01 -2.06163123e-01 1.18537593e+00 -2.51288414e-01 -6.61949962e-02 -1.06922254e-01 -4.79566514e-01 -1.87860489e-01 9.27814066e-01 -4.48647052e-01 5.46538047e-02 1.61490008e-01 -5.93275487e-01 -2.85609126e-01 6.42991602e-01 -3.23932618e-01 3.09392065e-01 -1.07370591e+00 -6.42830372e-01 6.15677893e-01 -4.98257101e-01 2.23002389e-01 4.08576012e-01 1.19373178e+00 -1.05675745e+00 3.46334219e-01 -1.31565258e-01 -7.06727684e-01 -1.02405214e+00 4.20566052e-01 1.13603246e+00 -6.92758501e-01 -6.18522763e-01 4.14001405e-01 4.89958711e-02 -6.53132319e-01 -2.81174779e-01 -3.36685926e-01 -6.34095550e-01 2.42565379e-01 3.84692699e-01 7.99124122e-01 3.92030686e-01 -9.22247112e-01 -1.61986172e-01 1.25146675e+00 6.57809496e-01 9.93543789e-02 1.21604097e+00 5.17997630e-02 -4.08599436e-01 5.82356751e-01 1.15384555e+00 2.30988741e-01 -1.32943654e+00 2.30379462e-01 -2.81574905e-01 -2.21730530e-01 1.92570135e-01 -3.70081991e-01 -8.80840957e-01 1.06058788e+00 4.91271108e-01 2.70369679e-01 8.71696830e-01 -4.89155024e-01 6.96866155e-01 -1.04907311e-01 2.16218308e-01 -7.63318956e-01 -6.29301012e-01 5.80501795e-01 8.21216166e-01 -8.09940994e-01 2.22646490e-01 -8.03378880e-01 -1.11704646e-02 1.53565180e+00 4.68739748e-01 -4.76716042e-01 8.96569192e-01 8.19221854e-01 -7.05150738e-02 9.68679786e-02 -4.04017299e-01 1.65897265e-01 3.42485547e-01 6.96283579e-02 7.16399074e-01 -2.56545860e-02 -7.22663522e-01 3.58402848e-01 1.11811429e-01 1.14464357e-01 1.02217764e-01 7.53003418e-01 -3.93208206e-01 -8.76279175e-01 -7.26872683e-01 2.91914701e-01 -2.84662962e-01 1.32618798e-02 1.73104391e-01 8.12457085e-01 5.49264073e-01 6.78784370e-01 4.43998426e-01 -4.61646728e-02 2.90724903e-01 3.05567592e-01 3.61134052e-01 -1.99070707e-01 -9.45801198e-01 1.10453561e-01 3.16949119e-03 -4.29895639e-01 -3.59274805e-01 -6.26548052e-01 -1.37839413e+00 -2.07545698e-01 -1.56556159e-01 7.13263869e-01 4.77473378e-01 9.85494792e-01 2.15730384e-01 5.39482892e-01 4.43992347e-01 -1.71669555e+00 -8.85231197e-02 -9.18256104e-01 -7.53582895e-01 2.66955853e-01 5.34698009e-01 -9.92743850e-01 -8.68624926e-01 -1.45767584e-01]
[6.387056827545166, 3.3242926597595215]
3963eb53-5252-41fe-a220-3e7e72c7c72f
resources-and-evaluations-for-multi
2306.12601
null
https://arxiv.org/abs/2306.12601v1
https://arxiv.org/pdf/2306.12601v1.pdf
Resources and Evaluations for Multi-Distribution Dense Information Retrieval
We introduce and define the novel problem of multi-distribution information retrieval (IR) where given a query, systems need to retrieve passages from within multiple collections, each drawn from a different distribution. Some of these collections and distributions might not be available at training time. To evaluate methods for multi-distribution retrieval, we design three benchmarks for this task from existing single-distribution datasets, namely, a dataset based on question answering and two based on entity matching. We propose simple methods for this task which allocate the fixed retrieval budget (top-k passages) strategically across domains to prevent the known domains from consuming most of the budget. We show that our methods lead to an average of 3.8+ and up to 8.0 points improvements in Recall@100 across the datasets and that improvements are consistent when fine-tuning different base retrieval models. Our benchmarks are made publicly available.
['Simran Arora', 'Omar Khattab', 'Soumya Chatterjee']
2023-06-21
null
null
null
null
['retrieval', 'question-answering', 'information-retrieval']
['methodology', 'natural-language-processing', 'natural-language-processing']
[-2.10464269e-01 -5.10327697e-01 -5.26404142e-01 -2.33647972e-01 -1.98825192e+00 -1.10904813e+00 6.78387940e-01 5.10692894e-01 -5.34735560e-01 9.06216025e-01 1.99299380e-01 -8.69033709e-02 -6.29670799e-01 -7.66481400e-01 -6.29947305e-01 -3.76926154e-01 1.20171323e-01 1.33626032e+00 6.35550082e-01 -3.41625720e-01 4.74821061e-01 5.02526760e-01 -1.44645298e+00 5.30076087e-01 6.28048897e-01 9.49928164e-01 3.17754522e-02 8.14154923e-01 -3.53760511e-01 6.21657252e-01 -1.09526861e+00 -2.37133235e-01 4.58187640e-01 -4.30445597e-02 -1.02265215e+00 -5.20315230e-01 6.89613879e-01 -5.95242500e-01 -6.05933547e-01 7.11037576e-01 6.96214318e-01 4.30305600e-01 1.08636510e+00 -1.09436989e+00 -9.92352009e-01 3.66921127e-01 -6.86262429e-01 5.66915810e-01 6.70047998e-01 -4.42831308e-01 1.31119549e+00 -5.66285372e-01 8.23876143e-01 1.25270343e+00 1.78994775e-01 4.36694443e-01 -1.18605244e+00 -5.20595551e-01 -9.81143042e-02 4.66380157e-02 -1.62287390e+00 -3.95600498e-01 1.75908074e-01 -1.73716128e-01 1.01544213e+00 3.96992385e-01 -1.54837728e-01 6.79993093e-01 -1.34873360e-01 8.76143873e-01 5.87220252e-01 -4.88195896e-01 1.24982327e-01 3.89184415e-01 7.27203131e-01 -1.89261898e-01 5.40669560e-01 -3.10237139e-01 -3.30013424e-01 -9.24172759e-01 1.81139827e-01 -3.00491564e-02 -3.08684349e-01 -2.65292227e-01 -7.45120823e-01 7.78861165e-01 4.02308702e-02 2.55580693e-01 -2.67910331e-01 8.63225535e-02 2.45122612e-01 6.90499485e-01 4.71157789e-01 7.28127539e-01 -7.56554067e-01 -1.11610340e-02 -8.25352371e-01 8.48132610e-01 1.19710636e+00 1.46993566e+00 9.72282887e-01 -9.63652194e-01 -5.62637329e-01 1.28200006e+00 3.71703543e-02 1.13023424e+00 4.22991037e-01 -8.90456617e-01 7.13556767e-01 4.73915488e-01 6.49003386e-01 -8.80567312e-01 -7.98443779e-02 1.56256035e-01 -2.55017608e-01 -7.31111526e-01 6.10755861e-01 -3.70432697e-02 -8.11471999e-01 1.60783911e+00 1.80142671e-01 -2.15675935e-01 2.52517276e-02 7.05952644e-01 6.94130003e-01 8.30986917e-01 -5.43106534e-02 -1.23115070e-02 1.27462888e+00 -8.16395819e-01 -3.22359085e-01 -5.00997752e-02 6.92619860e-01 -9.16212857e-01 1.06358111e+00 2.47476265e-01 -1.07964981e+00 -1.89479783e-01 -7.32777059e-01 -1.72013849e-01 -5.39161265e-01 -2.12495878e-01 1.34355605e-01 5.63230813e-01 -1.10259485e+00 2.37279087e-01 -2.90870547e-01 -3.65123242e-01 3.68452258e-02 2.21890762e-01 -1.57688946e-01 -5.67805290e-01 -1.27228701e+00 7.94507205e-01 2.17892140e-01 -7.26059496e-01 -7.72283494e-01 -1.02078259e+00 -2.89490163e-01 2.17587575e-01 1.43994629e-01 -7.90662885e-01 1.51543057e+00 -6.15731299e-01 -7.84862638e-01 8.33298862e-01 -2.20011562e-01 -2.17534110e-01 1.98448390e-01 -5.72351813e-01 -3.03217649e-01 4.94579911e-01 2.35580653e-01 4.50444192e-01 5.25297761e-01 -1.14611304e+00 -7.19731510e-01 -3.83134216e-01 1.95312843e-01 3.60809147e-01 -5.72792292e-01 3.69572282e-01 -9.76906121e-01 -4.31799591e-01 -4.12905931e-01 -8.39144588e-01 1.10862322e-01 -2.95685947e-01 -2.59679347e-01 -4.30873215e-01 4.81163770e-01 -6.13275111e-01 1.54915380e+00 -2.00362873e+00 -1.07715338e-01 3.93912405e-01 5.46918102e-02 4.46975343e-02 -4.89291340e-01 7.22837448e-01 3.31476092e-01 3.12292188e-01 2.27727562e-01 -1.56645954e-01 1.66450202e-01 1.21114530e-01 -6.48328722e-01 2.50883847e-01 -2.39768699e-01 6.69876039e-01 -8.79350662e-01 -5.05356610e-01 -4.12306011e-01 1.80232003e-01 -5.47534049e-01 3.25833291e-01 -6.54382408e-01 -1.45861313e-01 -6.69885516e-01 6.50182962e-01 8.10724199e-01 -5.26731312e-01 9.78258774e-02 3.31609100e-01 5.11828303e-01 4.61714506e-01 -1.01205218e+00 1.66771770e+00 -3.29862475e-01 3.75749588e-01 -2.16442287e-01 -5.42708218e-01 6.93843067e-01 1.56576201e-01 5.37919343e-01 -1.18913031e+00 -2.99894750e-01 3.90320897e-01 -5.01990199e-01 -2.01132998e-01 1.22098088e+00 1.72276840e-01 -4.41761225e-01 8.73075604e-01 -4.98644039e-02 2.73141600e-02 5.01437783e-01 4.96156156e-01 1.50923181e+00 -6.70056462e-01 -1.60148457e-01 -3.03573102e-01 1.14897817e-01 3.11317116e-01 1.34242609e-01 1.31092119e+00 2.18271703e-01 6.40087605e-01 3.61114532e-01 -5.90663105e-02 -1.24730325e+00 -1.18495035e+00 -2.35594437e-01 1.58614254e+00 4.39371198e-01 -3.41923416e-01 -2.67061472e-01 -5.59668660e-01 5.04759192e-01 5.71873069e-01 -2.25201219e-01 -1.14639260e-01 -4.49783742e-01 -6.62979364e-01 6.15357757e-01 4.00216669e-01 1.58077672e-01 -7.16556847e-01 -1.58619538e-01 2.89469715e-02 -3.01820755e-01 -7.97332227e-01 -7.10460484e-01 -2.11486384e-01 -5.95567703e-01 -1.04508209e+00 -1.33613718e+00 -6.71377003e-01 3.49379361e-01 5.70682228e-01 1.65184200e+00 1.03098683e-01 -3.76437515e-01 9.19138253e-01 -3.45575988e-01 -3.29300106e-01 -2.58610904e-01 6.17093801e-01 -1.78743377e-01 -7.10070133e-01 8.91451061e-01 -2.46904671e-01 -8.32356572e-01 4.99647588e-01 -1.24159253e+00 -8.68189692e-01 4.30039823e-01 7.10525095e-01 6.16129100e-01 -1.27709225e-01 8.49907637e-01 -1.09186912e+00 1.13930035e+00 -1.05774331e+00 -5.91771066e-01 8.69489074e-01 -6.96328282e-01 2.10195675e-01 3.70663732e-01 -5.91599464e-01 -8.59669745e-01 -5.60195923e-01 3.40228915e-01 -4.40146327e-01 -1.00519925e-01 3.72622728e-01 1.30816504e-01 1.71220422e-01 1.06099153e+00 1.63540408e-01 -2.52203643e-01 -6.83416486e-01 5.96187949e-01 9.75769758e-01 1.26480192e-01 -1.25879419e+00 4.81483757e-01 1.32423267e-01 -6.37352049e-01 -6.25263810e-01 -8.47820818e-01 -1.31638348e+00 -1.83153376e-01 3.52972895e-01 2.59312898e-01 -1.03515553e+00 -3.49853039e-01 2.60046989e-01 -1.05989587e+00 -3.69141221e-01 -2.36371443e-01 9.23504159e-02 -3.12284261e-01 4.72331345e-01 -7.00798512e-01 -6.13853216e-01 -6.26217425e-01 -6.13228261e-01 1.36344886e+00 3.76329571e-01 -1.43894628e-01 -9.46080983e-01 6.85973346e-01 1.19876757e-01 8.21595609e-01 -3.11569184e-01 1.19515693e+00 -1.32472217e+00 -6.93227053e-01 -2.52399385e-01 -4.23869073e-01 1.95812210e-01 1.48612529e-01 -2.80216753e-01 -7.55165517e-01 -5.90859473e-01 -4.41465676e-01 -7.20275640e-01 9.08749223e-01 4.03964147e-02 1.14735389e+00 -3.39935213e-01 -7.56632149e-01 8.49612057e-02 1.62635589e+00 2.96487033e-01 6.10982060e-01 5.87682903e-01 2.29911739e-03 3.68519813e-01 7.17562437e-01 5.72927594e-01 4.92474794e-01 7.58156836e-01 -3.28489482e-01 2.99410343e-01 -6.19733660e-03 -2.16906741e-01 -8.32367688e-02 5.62258124e-01 5.53382933e-01 -6.92254364e-01 -9.80512559e-01 1.09816837e+00 -1.78680277e+00 -8.52440536e-01 4.18366373e-01 2.61397576e+00 1.16588652e+00 -1.89971328e-01 3.36811900e-01 -4.79109585e-01 6.25747800e-01 -8.40068534e-02 -7.78314471e-01 -1.80782557e-01 -1.88129827e-01 5.17807543e-01 6.48828268e-01 4.22958970e-01 -9.28617001e-01 6.39470994e-01 7.45815659e+00 1.00716126e+00 -7.47942805e-01 -3.09231997e-01 5.63163042e-01 -3.72852534e-01 -7.15625286e-01 -1.34991571e-01 -1.36025262e+00 5.17840743e-01 1.33377528e+00 -9.05253768e-01 3.57172996e-01 8.82199287e-01 -6.96579218e-01 -2.13350981e-01 -1.14004374e+00 9.19005990e-01 1.12540744e-01 -1.07029486e+00 4.02917117e-01 -1.01331644e-01 8.33236516e-01 1.94268584e-01 1.13194317e-01 6.72653735e-01 6.89140618e-01 -7.37011433e-01 1.80854082e-01 6.36747599e-01 7.46642232e-01 -7.83563316e-01 4.31370378e-01 4.33459967e-01 -7.68857598e-01 1.59735475e-02 -7.68346608e-01 5.86655855e-01 -6.88094646e-02 5.13676405e-01 -7.27968097e-01 6.71717048e-01 7.71723509e-01 2.52238065e-01 -4.70663309e-01 1.39218676e+00 3.11839461e-01 3.69565815e-01 -7.40254998e-01 -2.34735519e-01 -3.19619626e-02 2.26440415e-01 3.03183556e-01 1.40467155e+00 3.51251543e-01 4.68341261e-02 1.42149091e-01 3.94968927e-01 -4.85379189e-01 1.58604428e-01 -7.29548216e-01 2.36439090e-02 1.11442792e+00 8.91547441e-01 -1.05967872e-01 -5.63913226e-01 -4.70826834e-01 8.59810591e-01 4.29756254e-01 6.20129704e-01 -7.05494940e-01 -8.73808861e-01 7.88394213e-01 1.34144187e-01 4.25040811e-01 5.75608760e-02 2.47855276e-01 -1.14282179e+00 3.18554968e-01 -9.43027496e-01 1.06969810e+00 -5.24027586e-01 -2.08099604e+00 3.98416549e-01 5.39144814e-01 -9.40682113e-01 -5.76773584e-01 -4.59302753e-01 1.80102199e-01 1.04980659e+00 -1.95056701e+00 -4.86625195e-01 -1.09596729e-01 6.91616714e-01 1.61090344e-01 -1.03391141e-01 9.76172864e-01 8.90287697e-01 -6.16891012e-02 1.00361192e+00 9.13102150e-01 7.76434243e-02 1.35099208e+00 -1.33137631e+00 1.44735903e-01 1.94080785e-01 1.20838188e-01 9.74953175e-01 2.26453915e-01 -2.62378633e-01 -1.38401556e+00 -8.81601036e-01 9.30084586e-01 -9.22664165e-01 6.54977739e-01 -1.34167030e-01 -1.11887455e+00 6.60623729e-01 2.72697240e-01 -1.92255035e-01 9.36290503e-01 3.23555052e-01 -8.33326519e-01 -2.69518852e-01 -1.35229003e+00 1.97215796e-01 6.96875513e-01 -6.58407629e-01 -8.23459029e-01 6.30715847e-01 7.63152897e-01 -4.82298255e-01 -1.09173286e+00 1.42958760e-01 5.60032368e-01 -4.74633902e-01 1.28145730e+00 -9.11900640e-01 2.13612795e-01 -3.15082744e-02 -4.97729540e-01 -1.24728084e+00 -1.11996368e-01 -3.77096653e-01 -2.76944995e-01 1.18144333e+00 6.40274107e-01 -7.21047342e-01 5.44979274e-01 1.11265171e+00 4.67389643e-01 -3.19828242e-01 -6.60283983e-01 -1.03662503e+00 7.00481415e-01 4.46830243e-02 7.62175798e-01 7.55078435e-01 -1.24424711e-01 4.64362025e-01 4.13364470e-02 1.18307106e-01 4.32666153e-01 5.12163579e-01 9.49667454e-01 -1.04481232e+00 -4.31912273e-01 -4.14868802e-01 4.46262695e-02 -1.84266901e+00 6.64005652e-02 -7.00559556e-01 -6.38923049e-02 -1.53416860e+00 6.89626932e-01 -9.07898784e-01 -5.39141297e-01 3.77361149e-01 -2.63738602e-01 -1.91985250e-01 -5.65703698e-02 6.04536057e-01 -1.21363103e+00 3.01340759e-01 7.36844182e-01 -3.17933381e-01 -1.03564031e-01 -1.10534444e-01 -1.05854344e+00 -5.75078167e-02 6.03125513e-01 -6.51232123e-01 -6.42566264e-01 -7.97090352e-01 2.42334634e-01 2.80797184e-01 -1.10724024e-01 -6.63465619e-01 6.32508159e-01 -1.19314596e-01 1.68033808e-01 -6.94619477e-01 2.96932012e-01 -5.39940119e-01 -1.42199755e-01 -2.67436564e-01 -6.50119245e-01 3.07811797e-01 5.47794044e-01 7.64462054e-01 -4.15502846e-01 -2.51828641e-01 4.33942795e-01 -1.08195119e-01 -6.36404634e-01 4.10918802e-01 -8.50819945e-02 8.26346636e-01 7.80222058e-01 3.26398164e-01 -9.64292705e-01 -5.02419651e-01 -3.19649011e-01 5.94592154e-01 5.07871389e-01 4.89295661e-01 3.84374470e-01 -1.27305436e+00 -7.67602861e-01 -3.27545255e-01 6.38330996e-01 7.48595968e-02 1.50706246e-01 2.38000378e-01 -4.55914855e-01 8.49137902e-01 1.31229550e-01 -4.75324929e-01 -1.15788412e+00 6.70414209e-01 1.99481159e-01 -7.40425766e-01 -1.81254670e-01 6.95053339e-01 9.79521349e-02 -5.92109561e-01 4.18832690e-01 1.34482132e-02 9.28795785e-02 -5.43609783e-02 9.46446836e-01 3.92488033e-01 1.61019608e-01 5.13414033e-02 -3.19842964e-01 3.81760299e-01 -7.82013893e-01 -3.27741235e-01 1.16475153e+00 -1.04393713e-01 -2.03342974e-01 2.48969004e-01 1.54656672e+00 2.84143742e-02 -5.45943022e-01 -7.51376510e-01 4.75865692e-01 -8.01338494e-01 -3.07374775e-01 -1.00976324e+00 -7.21379876e-01 4.78264898e-01 2.95737565e-01 4.23808843e-01 1.24462032e+00 1.85980976e-01 1.05436969e+00 9.49855208e-01 7.25983202e-01 -1.01303220e+00 1.59685001e-01 5.13992369e-01 6.78313613e-01 -9.39490795e-01 -5.32279462e-02 1.71374992e-01 -2.95503616e-01 8.01475585e-01 5.68838775e-01 -1.50028661e-01 8.58451962e-01 1.17560983e-01 -3.97195108e-02 -2.59681731e-01 -1.08548295e+00 -1.33495227e-01 3.39597285e-01 4.53502417e-01 3.08308482e-01 -1.59981176e-01 -2.91802198e-01 4.70817357e-01 1.98266521e-01 -1.64362207e-01 6.88529387e-02 1.10753727e+00 -6.31344795e-01 -1.45260918e+00 -2.56700367e-01 7.12334096e-01 -7.25854933e-01 -2.24797949e-01 -5.13515770e-01 7.93828666e-01 -7.76638627e-01 8.71950328e-01 1.86477616e-01 -5.68910642e-03 4.95715648e-01 5.58735244e-02 4.50885564e-01 -6.55357063e-01 -3.41234237e-01 -1.83576420e-01 2.58565277e-01 -3.93429786e-01 -2.03580543e-01 -5.34708560e-01 -8.90202582e-01 -4.83454376e-01 -2.80722678e-01 5.70730627e-01 2.74991125e-01 4.32385147e-01 8.58044744e-01 -1.44281015e-01 6.66110218e-01 -1.78667028e-02 -1.05060303e+00 -1.00882506e+00 -7.46104121e-01 6.41465127e-01 2.86599666e-01 -4.29311901e-01 -4.53520358e-01 -2.63552397e-01]
[11.454911231994629, 7.704309463500977]
b328b38b-0cbc-44e0-b008-7896e324eaa0
chili-pepper-disease-diagnosis-via-image
2306.12057
null
https://arxiv.org/abs/2306.12057v1
https://arxiv.org/pdf/2306.12057v1.pdf
Chili Pepper Disease Diagnosis via Image Reconstruction Using GrabCut and Generative Adversarial Serial Autoencoder
With the recent development of smart farms, researchers are very interested in such fields. In particular, the field of disease diagnosis is the most important factor. Disease diagnosis belongs to the field of anomaly detection and aims to distinguish whether plants or fruits are normal or abnormal. The problem can be solved by binary or multi-classification based on CNN, but it can also be solved by image reconstruction. However, due to the limitation of the performance of image generation, SOTA's methods propose a score calculation method using a latent vector error. In this paper, we propose a network that focuses on chili peppers and proceeds with background removal through Grabcut. It shows high performance through image-based score calculation method. Due to the difficulty of reconstructing the input image, the difference between the input and output images is large. However, the serial autoencoder proposed in this paper uses the difference between the two fake images except for the actual input as a score. We propose a method of generating meaningful images using the GAN structure and classifying three results simultaneously by one discriminator. The proposed method showed higher performance than previous researches, and image-based scores showed the best performanc
['Sungyoung Kim', 'Jongwook Si']
2023-06-21
null
null
null
null
['image-reconstruction', 'anomaly-detection']
['computer-vision', 'methodology']
[ 3.03958982e-01 -3.16793501e-01 1.62073508e-01 -1.66590855e-01 -6.73410371e-02 -2.61246473e-01 1.68067962e-01 -1.79791704e-01 2.76662922e-03 4.86118883e-01 -4.02705550e-01 6.00404851e-02 -1.01132356e-01 -1.41629064e+00 -4.16858345e-01 -1.01171851e+00 3.96085948e-01 1.16422191e-01 6.36012927e-02 -2.81823158e-01 1.10325798e-01 4.16846305e-01 -1.43413019e+00 3.09709191e-01 1.06688607e+00 1.09913659e+00 4.15244251e-01 4.76284802e-01 -4.45358843e-01 6.64755881e-01 -1.15306151e+00 -2.56817043e-01 3.49206924e-01 -1.20515418e+00 -3.62644255e-01 1.76843092e-01 -4.52688336e-02 -6.44359231e-01 3.48820090e-02 1.43946350e+00 4.71086919e-01 -3.73298347e-01 6.93641007e-01 -1.46339977e+00 -1.11075723e+00 4.51759696e-01 -6.32218838e-01 -9.54380035e-02 -1.17159851e-01 -7.71759450e-02 4.75447267e-01 -2.73902863e-01 3.14659208e-01 9.62934434e-01 4.75038946e-01 4.46407467e-01 -1.08917058e+00 -7.45732367e-01 -3.38602632e-01 6.53143406e-01 -1.17744231e+00 2.86941171e-01 9.82470751e-01 -4.15290922e-01 3.66623968e-01 2.18197942e-01 9.11033034e-01 9.47407961e-01 3.59739780e-01 6.81345344e-01 1.26322722e+00 -4.78309780e-01 1.64923728e-01 8.30114633e-02 -5.82066178e-02 6.47063255e-01 3.35011274e-01 1.73864271e-02 2.12761819e-01 2.15423062e-01 8.16055477e-01 3.93414557e-01 -3.87796849e-01 5.44072837e-02 -1.02337563e+00 1.05728352e+00 6.37933135e-01 9.73192155e-01 -4.32672679e-01 -3.01857740e-01 9.32451412e-02 3.06063056e-01 2.09022388e-01 3.04585725e-01 -3.40182483e-01 3.06725383e-01 -1.03775597e+00 -3.09688658e-01 6.86972678e-01 4.18878824e-01 3.59118432e-01 5.88346660e-01 -1.85894027e-01 6.83250010e-01 8.42114240e-02 6.49359822e-01 8.17216575e-01 -5.82523525e-01 -2.30881795e-02 7.74557769e-01 -2.52807379e-01 -1.48524618e+00 -6.80819228e-02 -4.15291190e-01 -1.33646441e+00 5.38109362e-01 2.92535663e-01 -2.31429413e-01 -1.09781098e+00 1.53341842e+00 1.74478292e-01 6.72251880e-02 2.63095617e-01 9.04898584e-01 9.20001447e-01 1.22931397e+00 -1.51584581e-01 -3.12319249e-01 1.07223785e+00 -1.02593637e+00 -1.18475950e+00 2.69368947e-01 3.02214533e-01 -7.15787470e-01 6.36161149e-01 8.56158257e-01 -6.16515458e-01 -7.69321978e-01 -1.27699971e+00 4.33041722e-01 -4.62456614e-01 7.00011075e-01 4.69476908e-01 5.55380404e-01 -7.77606905e-01 7.10776329e-01 -5.59031963e-01 -4.49081600e-01 2.37523764e-01 9.81822237e-03 -4.24101114e-01 1.66494116e-01 -9.54662144e-01 9.19442892e-01 6.40508056e-01 3.18686754e-01 -6.45097375e-01 -9.42354351e-02 -4.80227321e-01 3.92412394e-01 1.26003236e-01 -1.39812231e-01 7.09235549e-01 -1.64838958e+00 -1.69761896e+00 3.77516568e-01 3.77957016e-01 -2.50504076e-01 3.09902430e-01 -4.91620526e-02 -8.01382303e-01 2.99439996e-01 8.43487754e-02 4.14541930e-01 8.26946557e-01 -1.05899906e+00 -4.73472595e-01 -4.01220769e-01 -1.77741453e-01 -1.83657914e-01 -3.89186293e-01 -1.57592282e-01 1.43728644e-01 -7.06207514e-01 3.90035003e-01 -5.21961927e-01 8.41345340e-02 1.38664886e-01 -3.17879885e-01 1.60527807e-02 1.36640906e+00 -1.21358645e+00 8.82984042e-01 -2.26600361e+00 -1.05197234e-02 1.47529855e-01 3.52263055e-03 4.93451715e-01 -2.20700517e-01 6.67188913e-02 -7.81744570e-02 8.16676095e-02 -3.58116835e-01 5.43707132e-01 -3.69768620e-01 2.58911669e-01 -5.98544721e-03 1.04180105e-01 5.08820653e-01 7.17929721e-01 -5.60587049e-01 -5.28641880e-01 2.58418798e-01 5.93499899e-01 -1.14113785e-01 3.94500166e-01 -6.81650490e-02 2.36834511e-01 -4.47565764e-01 6.61260247e-01 1.03832293e+00 -1.77288130e-01 1.31809250e-01 -5.38080931e-01 -1.80850700e-02 -4.90443259e-01 -1.27161014e+00 1.25872135e+00 -2.87716925e-01 5.09561419e-01 1.08046360e-01 -1.41678929e+00 1.24465883e+00 3.68412852e-01 4.53698963e-01 -7.02986062e-01 3.77277523e-01 2.84991600e-02 1.53362438e-01 -7.57504940e-01 2.48993910e-03 7.45929107e-02 3.88173610e-01 1.35982081e-01 9.02555510e-02 -1.84379548e-01 1.52768388e-01 -9.45955738e-02 1.03395021e+00 2.83460736e-01 3.67615074e-01 7.41328821e-02 5.85763872e-01 1.91816598e-01 8.03063810e-01 2.42481470e-01 -1.33603029e-02 5.22858262e-01 5.80750942e-01 -4.15627003e-01 -9.36186612e-01 -8.69971395e-01 -1.27452567e-01 3.50668520e-01 2.28678614e-01 1.36754364e-01 -9.30163562e-01 -7.22151399e-01 -1.74271077e-01 6.31399810e-01 -5.18747687e-01 -2.73474842e-01 -3.88544738e-01 -9.42774773e-01 5.65615475e-01 4.13773835e-01 1.43153358e+00 -1.40916264e+00 -6.67997837e-01 3.64032924e-01 -2.32024118e-01 -6.36384130e-01 1.54140458e-01 2.69378126e-01 -9.28205490e-01 -9.86715794e-01 -9.07113016e-01 -9.90850508e-01 7.87073910e-01 1.89230502e-01 7.14298248e-01 2.60741413e-01 -3.69853139e-01 -2.78863937e-01 -7.06435621e-01 -3.86072606e-01 -6.12582326e-01 -1.18433371e-01 -4.48247135e-01 2.80496404e-02 2.98371226e-01 -3.00876170e-01 -4.65001553e-01 1.82094261e-01 -1.18816161e+00 1.05535187e-01 9.12716985e-01 1.10575640e+00 3.70639592e-01 6.09805346e-01 5.98185122e-01 -7.75990903e-01 3.56817901e-01 -3.00226182e-01 -7.40445912e-01 4.19820189e-01 -5.89179695e-01 5.80797978e-02 8.56703460e-01 -4.12101686e-01 -1.19884038e+00 -2.02574264e-02 -2.69711345e-01 -2.01302320e-01 -5.13592362e-01 3.42294246e-01 -4.51721072e-01 2.53306087e-02 4.07258630e-01 2.90143102e-01 2.57868290e-01 -2.87900269e-01 2.46800128e-02 7.99461663e-01 2.97863871e-01 2.50377089e-01 6.18116915e-01 1.78007230e-01 1.39012769e-01 -6.94970429e-01 -4.42136735e-01 1.08454436e-01 -3.46618921e-01 -3.37145448e-01 1.33564830e+00 -5.14538169e-01 -6.84253812e-01 8.09814572e-01 -1.25947046e+00 1.56838983e-01 -1.67397961e-01 5.92052519e-01 -5.64602986e-02 3.50376576e-01 -8.63836348e-01 -5.27630806e-01 -5.84941804e-01 -1.06024063e+00 6.13198698e-01 4.23320144e-01 3.39571029e-01 -6.02279723e-01 -1.61455780e-01 6.74563497e-02 5.27561724e-01 5.27272880e-01 1.12933993e+00 -4.26548481e-01 -4.87074733e-01 -3.33538324e-01 -2.12144002e-01 9.29563165e-01 3.99210721e-01 2.24840134e-01 -8.02023709e-01 -2.42829714e-02 5.77750564e-01 -1.84942544e-01 7.35792220e-01 4.27710205e-01 1.25276804e+00 -4.44648653e-01 -1.10747747e-01 5.07614493e-01 1.81298256e+00 7.81981766e-01 9.30466354e-01 1.90691665e-01 6.67594790e-01 4.13315117e-01 4.92707521e-01 1.32536620e-01 -2.38581195e-01 3.71923983e-01 6.06191754e-01 -4.21550304e-01 -1.73585013e-01 2.88419034e-02 2.83074498e-01 1.10081995e+00 -4.65968736e-02 -4.06761855e-01 -2.85116851e-01 2.98881114e-01 -1.49972570e+00 -1.41018617e+00 -3.89143556e-01 1.87211406e+00 5.05041540e-01 -1.07854761e-01 -3.21918160e-01 6.14752471e-01 1.00632644e+00 -1.34065002e-01 -5.70552588e-01 -4.56284404e-01 -4.18397188e-01 3.84589076e-01 3.43597084e-01 -6.21838234e-02 -8.87473464e-01 4.13723648e-01 5.77344465e+00 9.48017418e-01 -1.48915946e+00 1.47175267e-01 6.76606357e-01 4.03962910e-01 7.14841411e-02 -1.93533272e-01 -2.69206494e-01 7.59365916e-01 5.02174616e-01 3.39177251e-01 3.66061896e-01 9.24548745e-01 -1.82901219e-01 -2.95605689e-01 -5.50759375e-01 9.00606871e-01 4.08769876e-01 -7.63559997e-01 7.90821314e-02 -1.52924269e-01 7.63999522e-01 -5.19795239e-01 -1.58184305e-01 7.32145160e-02 3.26882228e-02 -8.14664125e-01 2.76727736e-01 4.50183272e-01 4.95765924e-01 -5.44589937e-01 1.04787683e+00 4.04297441e-01 -9.75602806e-01 -9.96776968e-02 -5.17122805e-01 1.17302358e-01 -2.57600129e-01 9.92419124e-01 -5.02106607e-01 7.20276892e-01 6.28242016e-01 5.79738081e-01 -4.70647782e-01 9.52828944e-01 -4.64675575e-01 7.53321767e-01 -3.10991764e-01 -1.66686952e-01 -9.31006819e-02 -6.52268291e-01 1.35598257e-01 6.98911667e-01 9.49169278e-01 8.85810778e-02 -5.27846254e-02 1.16113126e+00 1.86997458e-01 2.95090199e-01 -7.13639677e-01 -1.51098356e-01 7.19815260e-03 1.43468857e+00 -9.63601530e-01 -4.48798776e-01 -2.65440583e-01 1.39513230e+00 -2.26026177e-01 -2.07947684e-03 -8.93087268e-01 -8.96711707e-01 -2.63222098e-01 -2.77902007e-01 4.79567260e-01 1.25476420e-01 -4.41574231e-02 -1.09922063e+00 5.06058782e-02 -8.41606319e-01 1.89120099e-01 -1.07265580e+00 -1.06006777e+00 6.80537283e-01 -3.36075157e-01 -1.21209764e+00 -1.01816311e-01 -6.70850873e-01 -7.07626522e-01 7.78792500e-01 -1.13319886e+00 -1.07166016e+00 -8.00868571e-01 3.54475707e-01 5.02115190e-01 -2.47035220e-01 1.09557247e+00 4.06224728e-01 -6.18048310e-01 2.36406252e-01 4.48306173e-01 3.29499424e-01 4.44106519e-01 -1.10561478e+00 -2.01642692e-01 8.75242829e-01 1.17986791e-01 -2.07917884e-01 3.18047673e-01 -7.85872340e-01 -7.98411191e-01 -7.80946195e-01 6.36342704e-01 3.08299184e-01 -1.92361642e-02 2.18345195e-01 -8.86027157e-01 1.61853582e-01 5.20084620e-01 -2.08832592e-01 3.80287141e-01 -5.79654992e-01 5.93066253e-02 -4.39425886e-01 -1.67367530e+00 8.02929848e-02 4.02355224e-01 -2.79966332e-02 -4.56078649e-01 9.52068418e-02 4.76096362e-01 -5.29555455e-02 -6.93382502e-01 5.52305222e-01 4.17210877e-01 -9.77603316e-01 5.10391295e-01 4.46671508e-02 7.60608256e-01 -5.79438567e-01 -2.01056357e-02 -1.45127690e+00 -5.37475288e-01 3.07323277e-01 3.01914334e-01 1.42316341e+00 5.21426462e-02 -5.40503621e-01 6.37439787e-01 -2.60373533e-01 2.61910945e-01 -4.78312522e-01 -3.53992581e-01 -6.61059737e-01 -2.51100600e-01 2.27383465e-01 8.49085808e-01 1.11448646e+00 -4.84173685e-01 2.73367196e-01 -3.57934147e-01 1.49869755e-01 4.25808936e-01 3.00014913e-01 3.56332719e-01 -1.42173660e+00 -3.92834932e-01 -2.19328016e-01 -6.28482282e-01 -4.48409379e-01 -2.02290818e-01 -7.87854135e-01 2.86975615e-02 -1.76119316e+00 5.46369739e-02 -9.01202485e-02 -2.90273994e-01 4.97690201e-01 -4.29797508e-02 3.02803993e-01 3.57037663e-01 1.60952985e-01 9.03072283e-02 4.08572286e-01 1.44376218e+00 -4.94673520e-01 -1.43686056e-01 -6.58959448e-02 -3.68468851e-01 6.02971196e-01 1.32515025e+00 -4.69243824e-01 -3.46768856e-01 -5.38816988e-01 4.43265662e-02 2.88677722e-01 5.24126470e-01 -1.38559592e+00 -7.44238421e-02 -7.72674978e-02 9.45284963e-01 -7.03670621e-01 1.45317271e-01 -1.21787488e+00 4.71352458e-01 9.12581146e-01 -1.32041126e-01 2.35653281e-01 -1.29331931e-01 2.16538057e-01 -3.72613102e-01 -7.14441299e-01 7.50160098e-01 -3.34175438e-01 -6.20841861e-01 -7.87889138e-02 -3.35341245e-01 -5.38399875e-01 1.28464210e+00 -3.12201619e-01 -3.78573060e-01 -2.61734009e-01 -5.94901204e-01 -2.84150451e-01 1.88350119e-02 1.80879653e-01 6.95554852e-01 -1.46357393e+00 -5.35495996e-01 5.37998915e-01 -2.41478607e-01 -8.46219361e-02 2.58665502e-01 4.44652826e-01 -9.91656601e-01 8.42300057e-02 -9.07253742e-01 -5.58839977e-01 -1.14141107e+00 4.20077920e-01 2.56567955e-01 -2.09749639e-01 -4.54166979e-01 2.54584163e-01 -2.54811533e-02 -7.95510933e-02 -6.23307237e-03 -1.40843019e-01 -6.44503057e-01 6.02000207e-02 3.64252716e-01 5.04735768e-01 1.12503983e-01 -4.79916990e-01 1.25314787e-01 5.88177323e-01 1.82105899e-01 1.55589402e-01 1.30720329e+00 4.00386244e-01 -4.37227398e-01 1.52087480e-01 1.02488935e+00 -2.45742857e-01 -6.38971269e-01 3.47638428e-01 -4.83566344e-01 -4.16334063e-01 2.08627447e-01 -1.11329663e+00 -1.69151235e+00 9.07766104e-01 1.28635263e+00 6.27478004e-01 1.60212028e+00 -5.46081722e-01 8.65638375e-01 3.50304633e-01 2.98584789e-01 -1.01918817e+00 1.10810854e-01 1.71667431e-02 6.35292649e-01 -1.20463204e+00 -2.37035453e-01 -3.85125339e-01 -4.66877550e-01 1.27222669e+00 8.77031386e-01 -1.69434771e-01 5.31115294e-01 2.77392566e-01 2.88242191e-01 5.61249405e-02 -6.76667914e-02 -1.55401751e-01 -9.50584114e-02 7.82151878e-01 2.01748326e-01 1.34929299e-01 -5.85181475e-01 6.18780315e-01 -8.30535404e-03 2.16912717e-01 4.74409014e-01 8.26369226e-01 -4.97458667e-01 -1.12149227e+00 -7.26630270e-01 7.45873392e-01 -5.51381290e-01 7.61886388e-02 -5.96702397e-02 5.47514856e-01 6.35460079e-01 1.01353014e+00 5.82852103e-02 -5.17646492e-01 1.51230469e-01 -3.47045902e-03 3.76084477e-01 -1.47326276e-01 -4.90357935e-01 -1.01544872e-01 -5.28088868e-01 -2.49742001e-01 -6.21159017e-01 -1.70479953e-01 -1.07337296e+00 -2.59377629e-01 -8.44462395e-01 2.04366431e-01 9.55939233e-01 7.42762744e-01 5.02109453e-02 7.77924418e-01 8.47442389e-01 -3.02923411e-01 -5.61142623e-01 -1.01399004e+00 -7.72976577e-01 6.98014915e-01 -7.91724473e-02 -2.95957386e-01 -3.26246381e-01 1.62072062e-01]
[7.676959991455078, 1.907288908958435]
c13f7897-3384-4256-85c6-222f39ed7c89
channel-recurrent-attention-networks-for
2010.03108
null
https://arxiv.org/abs/2010.03108v1
https://arxiv.org/pdf/2010.03108v1.pdf
Channel Recurrent Attention Networks for Video Pedestrian Retrieval
Full attention, which generates an attention value per element of the input feature maps, has been successfully demonstrated to be beneficial in visual tasks. In this work, we propose a fully attentional network, termed {\it channel recurrent attention network}, for the task of video pedestrian retrieval. The main attention unit, \textit{channel recurrent attention}, identifies attention maps at the frame level by jointly leveraging spatial and channel patterns via a recurrent neural network. This channel recurrent attention is designed to build a global receptive field by recurrently receiving and learning the spatial vectors. Then, a \textit{set aggregation} cell is employed to generate a compact video representation. Empirical experimental results demonstrate the superior performance of the proposed deep network, outperforming current state-of-the-art results across standard video person retrieval benchmarks, and a thorough ablation study shows the effectiveness of the proposed units.
['Mehrtash Harandi', 'Lars Petersson', 'Jieming Zhou', 'Pan Ji', 'Pengfei Fang']
2020-10-07
null
null
null
null
['person-retrieval']
['computer-vision']
[ 3.31758559e-01 -4.68979299e-01 -1.05405629e-01 -6.46921322e-02 -6.49204075e-01 -2.54998449e-02 6.67910695e-01 -3.15437496e-01 -3.57983410e-01 5.90136170e-01 6.06874585e-01 1.86678290e-01 1.51827484e-01 -5.35718501e-01 -8.61878991e-01 -7.90900230e-01 -8.10200050e-02 -2.36137047e-01 9.76259857e-02 5.31656183e-02 1.66045710e-01 1.85503915e-01 -1.78323710e+00 6.54671311e-01 5.30960917e-01 1.37397957e+00 4.62384909e-01 7.65930533e-01 1.72042683e-01 1.27195692e+00 -5.97364604e-01 -2.40849674e-01 1.72969699e-02 -2.99013674e-01 -5.15770674e-01 -1.24579743e-01 4.61369187e-01 -7.20947623e-01 -1.12005746e+00 7.20463574e-01 7.06781924e-01 4.39777046e-01 8.03683400e-01 -8.37787747e-01 -1.27053511e+00 4.55318987e-01 -5.71735203e-01 8.28846335e-01 5.04759789e-01 2.21892580e-01 1.13395238e+00 -1.27254629e+00 4.42217529e-01 1.35545611e+00 3.70649338e-01 5.31252980e-01 -6.36750579e-01 -7.04881012e-01 5.10400414e-01 5.34082174e-01 -1.57326961e+00 -3.38898182e-01 5.36602139e-01 -3.85908067e-01 1.32602966e+00 2.28667170e-01 8.96827400e-01 1.36251545e+00 3.68034840e-01 1.38195455e+00 4.65168446e-01 -1.50739595e-01 -1.79835960e-01 -4.03938860e-01 2.86220342e-01 5.69570601e-01 1.54663220e-01 -4.79262806e-02 -6.42467797e-01 1.30327344e-01 1.11391568e+00 6.11045361e-01 -3.79978359e-01 -5.65207191e-02 -1.05234671e+00 5.92715025e-01 1.02713966e+00 1.23757228e-01 -7.30051875e-01 8.11306596e-01 5.31709790e-01 -9.08276141e-02 3.47688079e-01 -1.05489463e-01 2.84898371e-01 4.62614782e-02 -6.47924662e-01 1.98573411e-01 -9.41105187e-02 1.37553179e+00 3.90717834e-01 1.13905333e-01 -1.10755980e+00 8.22113037e-01 3.25612783e-01 6.47158742e-01 2.85227060e-01 -7.69902647e-01 5.76689661e-01 4.62697327e-01 2.09996074e-01 -9.97038066e-01 -2.61932351e-02 -4.60300624e-01 -1.03910327e+00 -5.38742602e-01 -1.68063521e-01 -5.25146872e-02 -1.06746137e+00 1.41548896e+00 -2.85615474e-01 4.02631372e-01 -8.94120038e-02 1.26898086e+00 1.23716497e+00 7.98462331e-01 4.68934625e-01 2.03209165e-02 1.22194135e+00 -1.37442064e+00 -6.95151508e-01 -9.03852135e-02 2.67415158e-02 -4.44087148e-01 7.74941742e-01 -4.26742435e-02 -1.32496512e+00 -9.53461528e-01 -8.47656608e-01 -3.63437325e-01 -3.32657725e-01 4.16273564e-01 6.23970926e-01 2.11876661e-01 -1.07771444e+00 1.36147723e-01 -5.61324835e-01 -3.03885937e-01 7.88899064e-01 3.51004273e-01 -1.17599912e-01 -2.35436127e-01 -1.26324809e+00 2.72586882e-01 8.67248103e-02 6.79696620e-01 -1.13087749e+00 -4.58244234e-01 -9.73761380e-01 5.83157837e-01 1.80266239e-02 -1.00644100e+00 8.62911642e-01 -7.85013378e-01 -1.20361137e+00 5.98375559e-01 -3.60758662e-01 -7.38174379e-01 1.52159214e-01 -6.66613042e-01 -1.67543098e-01 5.31284451e-01 7.72281364e-02 1.09389853e+00 1.22835958e+00 -9.79743302e-01 -6.23268425e-01 -2.07092047e-01 1.05009172e-02 3.28697026e-01 -6.31854534e-01 1.89416349e-01 -1.26209843e+00 -1.12961328e+00 -4.49442178e-01 -7.08898783e-01 -2.57207900e-01 -3.39722455e-01 -3.56507748e-01 -4.62415904e-01 6.58003032e-01 -6.48955107e-01 1.72670758e+00 -2.05465817e+00 5.06599724e-01 2.11050838e-01 3.79946202e-01 4.13666695e-01 -4.49644476e-02 1.49759185e-02 -7.54321888e-02 9.00064781e-03 8.36687535e-02 -3.61666799e-01 -1.07648805e-01 -1.86326906e-01 -5.51490068e-01 3.16604316e-01 3.92385125e-01 1.48640215e+00 -7.97562003e-01 -4.31003600e-01 4.15103883e-01 9.16530073e-01 -6.78185582e-01 3.99139225e-01 -5.90936560e-03 1.22878104e-01 -7.70385504e-01 6.84439361e-01 2.61175632e-01 -5.56139827e-01 -2.26139575e-01 -2.01451778e-01 -1.05433213e-02 -2.23006099e-01 -5.01880050e-01 1.78631115e+00 -1.60972804e-01 9.79705989e-01 -4.51983631e-01 -7.99657285e-01 6.87444746e-01 1.57740355e-01 4.03488040e-01 -1.16002548e+00 3.44225585e-01 -2.78315544e-01 -3.38425070e-01 -4.95276511e-01 8.75390410e-01 9.60133374e-01 -2.28289679e-01 3.67770672e-01 1.64228976e-01 8.80720496e-01 -4.71945927e-02 3.53392333e-01 1.01563334e+00 2.31823027e-01 -3.20543982e-02 -1.98876336e-01 6.82313144e-01 -5.04500568e-01 1.91139981e-01 1.01063609e+00 -2.15589598e-01 9.41270888e-01 2.34763697e-01 -6.26080930e-01 -1.15620863e+00 -7.22391725e-01 1.08727612e-01 1.50722468e+00 3.67341757e-01 -6.13169134e-01 -8.42895687e-01 -3.81017685e-01 -1.08555607e-01 6.85253814e-02 -1.08809185e+00 -2.50407070e-01 -8.21140409e-01 -5.12175262e-01 5.06235957e-01 1.02286255e+00 9.21106577e-01 -1.50192213e+00 -1.02249634e+00 9.89328921e-02 -3.43466729e-01 -1.10080588e+00 -8.53333116e-01 -3.58952612e-01 -1.81251228e-01 -7.46518195e-01 -1.50455236e+00 -8.46463501e-01 6.52038634e-01 6.92918718e-01 1.00520909e+00 3.75182599e-01 -6.32318079e-01 7.50351727e-01 -4.37035918e-01 -2.55322546e-01 6.23989403e-01 1.51141524e-01 -3.58333826e-01 3.64811659e-01 4.45842892e-01 -3.01664397e-02 -1.26246881e+00 1.59494027e-01 -8.60402882e-01 -1.96675897e-01 7.30599999e-01 1.04123604e+00 7.61534452e-01 -4.03154254e-01 5.39897859e-01 -6.32511079e-01 5.77637374e-01 -5.24454355e-01 -1.84548229e-01 3.24704140e-01 2.42715120e-01 -1.05309799e-01 3.43714833e-01 -1.53498381e-01 -9.72141743e-01 -3.84473689e-02 1.65502027e-01 -8.65880251e-01 1.01661704e-01 2.46194914e-01 3.57844830e-02 -2.34993733e-03 2.52561778e-01 4.31281954e-01 -2.04403803e-01 -2.05984205e-01 2.55335301e-01 7.01008618e-01 5.02269626e-01 -4.28934455e-01 1.54917583e-01 4.30505246e-01 -3.61489534e-01 -8.35864842e-01 -6.98389053e-01 -6.19458914e-01 -4.58922148e-01 -5.07860124e-01 1.28290772e+00 -1.29481459e+00 -8.76434386e-01 4.69353318e-01 -1.17963994e+00 -3.09769928e-01 2.06827633e-02 2.59698570e-01 -5.13814092e-01 1.59175694e-01 -5.75371742e-01 -8.60857129e-01 -8.08076799e-01 -1.33008873e+00 1.42339611e+00 5.60746193e-01 8.38587955e-02 -6.43591762e-01 -5.31289577e-01 2.08214149e-01 5.07481694e-01 4.28666025e-02 4.59283501e-01 -2.10567906e-01 -1.09353626e+00 -9.25810486e-02 -7.27516234e-01 -5.55787012e-02 -1.99980408e-01 -1.82325229e-01 -1.08279431e+00 -4.70733374e-01 -7.33150363e-01 -3.34383547e-01 1.64706230e+00 8.20927978e-01 1.82209134e+00 -3.59523408e-02 -6.12609625e-01 6.21727526e-01 1.12650061e+00 2.48790737e-02 1.12366831e+00 2.83734381e-01 9.13727820e-01 1.06250465e-01 4.06604499e-01 5.79496264e-01 4.07261014e-01 7.06364512e-01 2.49879509e-01 -2.92225003e-01 -4.80645806e-01 -8.09509009e-02 2.46929660e-01 3.80532771e-01 -4.99972969e-01 -4.91096407e-01 -6.30429983e-01 6.47782505e-01 -2.14899182e+00 -1.40500247e+00 4.15681720e-01 2.03573704e+00 7.69078135e-02 5.22946902e-02 1.05358124e-01 -2.48030469e-01 7.78960347e-01 3.01208645e-01 -5.56712508e-01 -2.04665035e-01 -2.75898695e-01 2.23625481e-01 3.28348726e-01 1.58725455e-01 -1.44275033e+00 1.07687509e+00 6.57635117e+00 7.23224759e-01 -9.07915592e-01 -1.76186845e-01 8.84682417e-01 -2.07106695e-01 -1.44640401e-01 -5.92075109e-01 -9.69374061e-01 5.90022445e-01 5.06725311e-01 5.48643656e-02 3.61505955e-01 6.40498579e-01 6.32264465e-02 6.54165149e-02 -8.98351252e-01 1.33640337e+00 6.35904074e-01 -1.54244745e+00 6.83505416e-01 -2.35469230e-02 7.84331739e-01 -6.85077459e-02 5.44752002e-01 4.49306548e-01 8.52441508e-03 -1.38392293e+00 6.89370453e-01 1.13807130e+00 9.61455941e-01 -8.91221941e-01 7.01126337e-01 -2.74630666e-01 -1.74365604e+00 -5.82285345e-01 -5.12309611e-01 1.25861824e-01 5.69179319e-02 -1.88079283e-01 -2.02286795e-01 4.23689336e-01 1.29385889e+00 1.34930706e+00 -8.62128139e-01 1.33436882e+00 1.55707136e-01 4.09399867e-01 7.52733573e-02 -1.93803415e-01 4.58541542e-01 1.67370275e-01 5.15815496e-01 1.73505032e+00 1.20850854e-01 2.68600553e-01 1.07632384e-01 6.37858629e-01 -3.08062673e-01 -6.59214556e-02 -6.37579441e-01 1.45909503e-01 3.15088004e-01 1.02107096e+00 -3.93751085e-01 -5.59050918e-01 -4.72180247e-01 1.13417339e+00 4.54120606e-01 9.78910923e-01 -1.06019580e+00 -5.01785934e-01 5.89986086e-01 -1.73068255e-01 9.82222080e-01 -2.79215630e-02 1.08636625e-01 -1.03741944e+00 1.80675223e-01 -5.61445117e-01 6.25794113e-01 -1.10425448e+00 -1.08281529e+00 8.20262313e-01 -6.29499853e-02 -1.17880940e+00 -2.35495150e-01 -4.39745843e-01 -5.00537992e-01 9.54470694e-01 -1.66423619e+00 -1.38694060e+00 -7.47426212e-01 9.03217316e-01 9.30176020e-01 -5.75608432e-01 5.46609402e-01 4.37166899e-01 -8.45605612e-01 1.05114424e+00 -1.57326251e-01 4.06522661e-01 5.09808004e-01 -8.84015143e-01 3.25993598e-01 7.84930527e-01 -1.50636747e-01 9.52742398e-01 5.15421964e-02 -5.32029569e-01 -1.55284762e+00 -1.53689206e+00 3.47281337e-01 -2.58335203e-01 3.74112636e-01 -3.50391507e-01 -6.65450990e-01 7.81127751e-01 5.24241447e-01 3.57592016e-01 4.73981470e-01 -2.97011763e-01 -3.54026705e-01 -6.37827814e-02 -5.07461369e-01 5.41079700e-01 1.31155980e+00 -6.40697420e-01 -3.19792807e-01 1.12854332e-01 8.46568167e-01 -4.18904722e-01 -7.24616706e-01 2.75467664e-01 9.34356153e-01 -6.79265797e-01 1.47088754e+00 -7.01831281e-01 5.97075582e-01 -3.20396990e-01 -2.40601957e-01 -8.09167802e-01 -9.04914021e-01 -6.88043058e-01 -3.99675518e-01 1.00566161e+00 1.42063096e-01 -1.07955791e-01 5.56351364e-01 4.07585025e-01 -2.65179217e-01 -8.02576244e-01 -8.00228715e-01 -2.49613866e-01 -6.64203689e-02 -1.50862500e-01 4.22437698e-01 2.47099981e-01 -2.80382812e-01 4.01431680e-01 -6.78564548e-01 3.52993235e-02 6.63296402e-01 -7.44086877e-02 6.00877702e-01 -7.01311946e-01 -5.18631637e-02 -6.59045994e-01 -6.77394629e-01 -1.71774936e+00 2.01197669e-01 -6.11249328e-01 -2.79996656e-02 -1.67988646e+00 6.62375271e-01 -2.62789339e-01 -6.78028882e-01 2.02802539e-01 -5.91276348e-01 7.23249912e-01 4.04075235e-01 2.54898489e-01 -1.40356886e+00 7.09218144e-01 1.15069687e+00 -4.88381505e-01 -1.36153087e-01 -3.09692115e-01 -6.73621833e-01 3.26194882e-01 2.53818780e-01 2.88920999e-01 -2.84196019e-01 -8.97199929e-01 -9.61497147e-03 -1.44344181e-01 6.84359312e-01 -1.11517012e+00 5.12045741e-01 3.60571712e-01 1.06352365e+00 -9.42921221e-01 5.68444133e-01 -6.24335945e-01 -2.42475718e-01 1.39584005e-01 -6.84775889e-01 1.63789734e-01 1.73854724e-01 8.82949889e-01 -2.64582038e-01 3.48760635e-01 2.64807165e-01 -4.18964103e-02 -1.18497252e+00 7.27191508e-01 -3.25148612e-01 -8.46192911e-02 1.00692284e+00 -1.47854388e-01 -2.79219270e-01 -2.53620058e-01 -7.34961331e-01 5.20649672e-01 -1.43302545e-01 7.03916788e-01 1.10864222e+00 -1.72513378e+00 -7.32331872e-01 3.06029856e-01 3.21897328e-01 -1.16468966e-01 6.63386524e-01 4.68648344e-01 -2.20501304e-01 1.01027560e+00 -2.91028082e-01 -8.40894163e-01 -1.10457671e+00 7.48034775e-01 2.50949800e-01 -2.19733473e-02 -1.00679517e+00 1.03462267e+00 6.03493214e-01 6.82816148e-01 6.71912193e-01 -2.28408381e-01 -8.22257757e-01 -5.99515103e-02 1.12279975e+00 2.41319448e-01 -2.68967450e-01 -8.28319252e-01 -1.34563878e-01 7.40891933e-01 -3.87071580e-01 3.29975158e-01 1.15806925e+00 -3.70794117e-01 1.49831697e-01 -8.86603072e-02 1.16513705e+00 -4.80817705e-01 -1.52212238e+00 -1.86383530e-01 -5.69046438e-01 -4.97116894e-01 -1.17778862e-02 -2.94272244e-01 -1.08762193e+00 9.26757276e-01 5.53696632e-01 -7.95304105e-02 1.18145216e+00 -5.29500656e-02 9.62673187e-01 4.70929503e-01 1.48994759e-01 -9.65572655e-01 4.23335582e-01 5.18200159e-01 1.17205858e+00 -1.05269563e+00 -1.07310176e-01 -2.28873074e-01 -6.91443384e-01 9.04085457e-01 8.83259535e-01 -5.38289607e-01 5.45215130e-01 -1.22813292e-01 -5.22259951e-01 -9.65090618e-02 -7.85133898e-01 -4.92332935e-01 5.34055591e-01 5.89682221e-01 4.75510925e-01 -4.63496484e-02 1.15519859e-01 6.36983156e-01 2.47345626e-01 8.22252557e-02 6.77122734e-04 7.03432977e-01 -4.38431293e-01 -4.61615950e-01 -2.78446883e-01 5.64855754e-01 -4.01809603e-01 -3.68664533e-01 -2.76899248e-01 4.10453975e-01 -1.43120930e-01 6.63219154e-01 4.89549190e-01 -2.71483034e-01 2.26197019e-01 -3.93880248e-01 4.37555015e-01 -2.29218230e-01 -5.08149564e-01 2.62189239e-01 -2.27557838e-01 -8.27680945e-01 -5.65719604e-01 -5.54526031e-01 -1.01555288e+00 2.77957302e-02 7.09498972e-02 1.72321945e-02 -9.26582292e-02 8.09075832e-01 7.01692283e-01 1.08174038e+00 4.76707548e-01 -1.13575602e+00 1.00181259e-01 -9.06694949e-01 7.54621476e-02 4.60671633e-01 4.31188345e-01 -9.65963542e-01 2.87881851e-01 2.42348164e-01]
[9.441946983337402, 0.6835160255432129]
07343c04-d472-4117-93fd-aaedd6793ec2
can-neural-networks-do-arithmetic-a-survey-on
2303.07735
null
https://arxiv.org/abs/2303.07735v1
https://arxiv.org/pdf/2303.07735v1.pdf
Can neural networks do arithmetic? A survey on the elementary numerical skills of state-of-the-art deep learning models
Creating learning models that can exhibit sophisticated reasoning skills is one of the greatest challenges in deep learning research, and mathematics is rapidly becoming one of the target domains for assessing scientific progress in this direction. In the past few years there has been an explosion of neural network architectures, data sets, and benchmarks specifically designed to tackle mathematical problems, reporting notable success in disparate fields such as automated theorem proving, numerical integration, and discovery of new conjectures or matrix multiplication algorithms. However, despite these impressive achievements it is still unclear whether deep learning models possess an elementary understanding of quantities and symbolic numbers. In this survey we critically examine the recent literature, concluding that even state-of-the-art architectures often fall short when probed with relatively simple tasks designed to test basic numerical and arithmetic knowledge.
['Alberto Testolin']
2023-03-14
null
null
null
null
['numerical-integration', 'automated-theorem-proving', 'automated-theorem-proving']
['miscellaneous', 'miscellaneous', 'reasoning']
[-1.94684893e-01 -1.03545956e-01 -2.23597452e-01 -2.23596275e-01 -2.01922163e-01 -6.41880989e-01 7.34340131e-01 5.51803887e-01 -4.35138553e-01 8.27689946e-01 -2.87558585e-01 -9.73234534e-01 -4.46500242e-01 -1.21110022e+00 -7.15499222e-01 -2.55844891e-01 -3.75246882e-01 5.26835799e-01 -2.19003975e-01 -5.23171842e-01 5.66715717e-01 7.35024750e-01 -1.34258640e+00 8.90057236e-02 8.95370960e-01 1.14068627e+00 -5.63145936e-01 7.82502770e-01 -3.47819000e-01 1.35976529e+00 -5.87597668e-01 -6.86285436e-01 2.14038014e-01 -3.53685528e-01 -9.71815825e-01 -8.58779728e-01 5.89076579e-01 -3.99743825e-01 -5.18593788e-01 1.15479875e+00 8.26887861e-02 6.68514818e-02 3.96888673e-01 -1.49970329e+00 -8.68238091e-01 7.50834882e-01 -2.18655810e-01 5.37041664e-01 1.54518858e-01 2.58585483e-01 1.23982167e+00 -4.63477015e-01 2.28397295e-01 1.06942832e+00 9.03282762e-01 3.30524653e-01 -1.27727914e+00 -8.88032794e-01 -2.00579226e-01 6.11452937e-01 -1.06507051e+00 -2.03544155e-01 7.14670539e-01 -3.81650984e-01 9.50039804e-01 8.03056732e-02 9.51819718e-01 5.53608716e-01 4.14165884e-01 6.02828860e-01 9.44259167e-01 -3.43557954e-01 3.22474509e-01 -1.89598933e-01 3.09556752e-01 7.87036777e-01 5.52626014e-01 9.90807340e-02 -3.06288928e-01 1.99838653e-01 1.05948532e+00 -3.90113771e-01 3.43147740e-02 -4.08415198e-01 -1.51690888e+00 1.07360387e+00 4.57908094e-01 5.25841653e-01 -5.06422594e-02 5.62164307e-01 6.11999512e-01 5.64512789e-01 -2.83016209e-02 1.20214176e+00 -5.99097908e-01 -3.97468626e-01 -9.62200642e-01 7.94652522e-01 9.79053020e-01 4.93842632e-01 5.12476623e-01 4.04175043e-01 2.67422050e-01 9.96632054e-02 -2.26545289e-01 1.49206653e-01 1.92603528e-01 -1.13211024e+00 4.55090642e-01 6.63539350e-01 -1.56755880e-01 -1.21343291e+00 -4.84327257e-01 -7.19065547e-01 -1.12661982e+00 4.26955193e-01 8.33364069e-01 -1.69891939e-01 -4.03449863e-01 1.65239465e+00 7.56399287e-03 2.54185170e-01 7.05354065e-02 5.43031156e-01 8.92309546e-01 6.06039107e-01 -1.24877598e-02 1.63785815e-01 8.28852117e-01 -4.89289403e-01 -3.15627605e-01 -7.42847621e-02 8.16166520e-01 -4.19056773e-01 6.43507957e-01 7.46353209e-01 -1.37068021e+00 -6.87323272e-01 -1.20905709e+00 -4.14563596e-01 -6.46072388e-01 -3.40602219e-01 1.44824946e+00 6.85209632e-01 -1.08585441e+00 9.25030410e-01 -6.42258465e-01 1.62943393e-01 7.33924031e-01 5.17916262e-01 -7.35655352e-02 -3.78729030e-02 -1.45513737e+00 1.19313443e+00 6.17761016e-01 -1.14886360e-02 -6.22793853e-01 -1.26095724e+00 -7.79859424e-01 4.80309844e-01 2.43689582e-01 -5.74853957e-01 1.31369090e+00 -6.45823777e-01 -1.25460947e+00 6.64979160e-01 2.64522403e-01 -9.08035278e-01 4.10101712e-01 6.74001547e-03 -3.58637065e-01 -7.60677010e-02 -2.73578137e-01 5.03144860e-01 2.21518829e-01 -6.65269315e-01 -4.86147761e-01 -2.58809090e-01 5.17132401e-01 -1.17720447e-01 -2.01087683e-01 -2.95953937e-02 3.56114507e-01 -4.54039603e-01 1.63962040e-02 -3.39329153e-01 -1.40229806e-01 1.20312333e-01 -2.00936440e-02 -5.68388522e-01 2.92102963e-01 -3.36859286e-01 1.03205872e+00 -1.71081030e+00 2.55172938e-01 8.45040753e-02 5.98612309e-01 3.56066823e-01 7.21440911e-02 1.75692409e-01 -3.00138623e-01 3.09637517e-01 2.96096746e-02 4.57129538e-01 1.72768742e-01 -1.18130453e-01 -4.92963880e-01 3.86430144e-01 3.81242007e-01 1.25136960e+00 -1.19382989e+00 -4.18769240e-01 3.13678175e-01 1.36495009e-01 -5.62900782e-01 -2.11017191e-01 -6.65329933e-01 2.63869345e-01 -2.42343187e-01 4.58578885e-01 3.95723343e-01 -4.80384082e-01 1.79612264e-01 1.42986029e-01 -7.77156577e-02 5.58995724e-01 -1.22188461e+00 1.34355140e+00 -2.65753895e-01 1.11161745e+00 -2.19980285e-01 -1.51534843e+00 6.12476766e-01 1.73467740e-01 4.05426800e-01 -8.99737060e-01 4.25274491e-01 3.05796653e-01 6.93282485e-01 -2.86233574e-01 4.12712783e-01 -2.56289184e-01 7.01156855e-02 4.24928993e-01 3.40987854e-02 -5.75670004e-01 5.73534667e-01 4.60780077e-02 1.01412773e+00 2.56169010e-02 3.67868602e-01 -8.65734220e-02 7.62889326e-01 3.78358305e-01 2.86481500e-01 6.68664634e-01 -1.45372331e-01 -1.02013409e-01 7.62779951e-01 -9.23740506e-01 -1.19298697e+00 -1.08121276e+00 -2.10849196e-01 9.97251868e-01 -2.97045439e-01 -2.41198316e-01 -4.80049133e-01 2.53944471e-03 2.07988858e-01 7.34520078e-01 -5.81469238e-01 -1.64211750e-01 -8.37070465e-01 -5.12681127e-01 9.34927106e-01 8.26902330e-01 8.06959987e-01 -1.11821771e+00 -6.69002116e-01 2.01958194e-01 1.74690306e-01 -1.01320744e+00 6.43107235e-01 3.21931869e-01 -1.19131219e+00 -1.34328914e+00 -4.35793251e-01 -9.72936571e-01 2.45537996e-01 -1.69607326e-01 1.50330198e+00 3.03013474e-01 -3.40873420e-01 -3.32727581e-02 1.23981573e-01 -4.71422821e-01 -5.99945664e-01 1.60297588e-01 6.11609668e-02 -7.28162706e-01 4.57310021e-01 -6.04092300e-01 -1.00781895e-01 -4.35295194e-01 -8.39391291e-01 1.34192640e-02 4.25046295e-01 8.35977912e-01 -2.71845870e-02 3.81105334e-01 6.73812449e-01 -4.23545241e-01 7.44171560e-01 -3.05085957e-01 -9.28272367e-01 1.96960583e-01 -4.56045777e-01 1.69965088e-01 1.11783755e+00 -2.57021338e-01 -4.77904141e-01 -6.90089881e-01 -5.61487749e-02 -2.04865970e-02 -1.51205435e-01 8.67239714e-01 2.52871096e-01 -3.28946084e-01 6.61610305e-01 1.89305112e-01 -3.24851833e-02 -1.78665540e-03 3.47858280e-01 -3.26816700e-02 6.99927449e-01 -1.03933895e+00 9.99807239e-01 5.52479289e-02 6.06932163e-01 -7.14029372e-01 -9.79007840e-01 1.16610564e-01 -5.42505980e-01 -7.10438984e-03 5.43038189e-01 -5.54384768e-01 -1.31238925e+00 4.71387923e-01 -1.12252355e+00 -4.70368594e-01 -1.79367244e-01 4.62012082e-01 -5.49689889e-01 9.72583964e-02 -7.70140350e-01 -6.06751978e-01 -1.62627757e-01 -1.05449569e+00 1.92716092e-01 2.31127441e-01 -4.01184529e-01 -1.35081089e+00 1.46460468e-02 1.66277498e-01 6.01907969e-01 3.98148358e-01 1.70807123e+00 -6.23244941e-01 -6.97004855e-01 -2.88480848e-01 -4.84492868e-01 4.08814937e-01 -2.03765929e-01 5.87094389e-02 -8.02718043e-01 1.06939167e-01 -1.08622439e-01 -7.15996265e-01 6.86586499e-01 1.91021815e-01 1.55477071e+00 -1.15781970e-01 1.04326651e-01 6.70225024e-01 1.24610507e+00 3.51512693e-02 7.12277949e-01 4.52391893e-01 2.78739363e-01 2.34821320e-01 7.38974437e-02 3.91655341e-02 3.62197727e-01 1.64853092e-02 4.27975774e-01 2.55493701e-01 1.61669046e-01 -3.54686193e-02 -1.20347023e-01 6.19508803e-01 -3.04896921e-01 2.04558223e-01 -1.37198973e+00 4.13953632e-01 -1.32585061e+00 -1.32079566e+00 -9.01678577e-03 1.99118316e+00 1.16604733e+00 6.03558660e-01 -4.39464413e-02 7.09702432e-01 2.03015506e-01 1.09502010e-01 -6.35857582e-01 -6.28984034e-01 8.84718895e-02 9.52758610e-01 6.73450604e-02 2.76727825e-01 -9.63133216e-01 9.03758526e-01 6.92445946e+00 5.45904994e-01 -1.08267128e+00 -6.08969867e-01 7.64953494e-01 2.57952094e-01 -2.30977282e-01 -3.32204342e-01 -5.62483847e-01 4.95057292e-02 1.04735565e+00 -3.10043365e-01 5.76506019e-01 8.18642020e-01 -1.89475507e-01 -8.65726694e-02 -1.53680444e+00 9.04714942e-01 -1.60714179e-01 -1.92928803e+00 3.78976017e-02 -1.00327283e-01 7.03449845e-01 -2.26953998e-01 3.81066829e-01 8.00736129e-01 4.97496575e-01 -1.81240606e+00 3.77182305e-01 3.25441599e-01 7.70641923e-01 -8.63484621e-01 6.83768570e-01 2.81192303e-01 -8.38603497e-01 -1.22167662e-01 -2.48222321e-01 -9.90570366e-01 -5.07298291e-01 4.07533556e-01 -5.58688164e-01 1.74374133e-01 4.87858802e-01 5.60671806e-01 -5.73145390e-01 1.13312769e+00 -1.73299208e-01 6.45325482e-01 -2.09351376e-01 -5.11257470e-01 4.13773865e-01 -2.25807145e-01 3.22770998e-02 8.17214787e-01 -1.20897539e-01 2.00182572e-01 -2.54020482e-01 1.23208857e+00 -3.25553894e-01 -1.54141113e-01 -4.11747307e-01 -4.99495000e-01 3.65372926e-01 9.66262221e-01 -8.15497041e-01 -4.39678103e-01 -4.70692664e-01 1.33868068e-01 5.31728327e-01 2.03114256e-01 -9.76392388e-01 -5.53903520e-01 8.01484883e-01 -2.02066526e-01 1.71103515e-02 -8.21953773e-01 -1.08802664e+00 -1.12400699e+00 -1.74698055e-01 -1.29126072e+00 1.24429993e-01 -5.94417155e-01 -9.92140651e-01 -1.16053067e-01 -1.32854562e-02 -5.73571801e-01 -2.73252696e-01 -1.14246452e+00 -6.46107614e-01 8.50034833e-01 -1.38819087e+00 -4.37212050e-01 -2.63707429e-01 2.77214795e-01 1.37478590e-01 -2.95340002e-01 1.01945794e+00 6.40655085e-02 -2.54800677e-01 5.10530889e-01 2.71717250e-01 6.83023453e-01 9.80647802e-02 -1.11866570e+00 5.41063070e-01 5.16699970e-01 2.50776291e-01 9.16175842e-01 6.76044166e-01 -2.42625549e-01 -1.63459325e+00 -5.10348856e-01 8.80337179e-01 -3.34705293e-01 1.13967443e+00 -2.77789116e-01 -9.43936467e-01 6.72505677e-01 -8.06912035e-02 -3.11963893e-02 4.96905297e-01 3.75072837e-01 -6.86149120e-01 -2.10089296e-01 -1.00371444e+00 7.04473972e-01 5.93519211e-01 -4.89746392e-01 -8.25139940e-01 2.13589981e-01 5.40738225e-01 -6.94543958e-01 -9.48806942e-01 4.35727775e-01 6.77770197e-01 -1.02266991e+00 1.17006493e+00 -1.08493376e+00 1.09881079e+00 9.95428488e-02 1.08992690e-02 -9.98165905e-01 -2.35185042e-01 -3.80336136e-01 -4.91539896e-01 7.05792427e-01 1.09111175e-01 -4.75240856e-01 1.01891589e+00 5.27334630e-01 4.43191230e-02 -1.04067814e+00 -5.60529053e-01 -6.88255966e-01 1.08476222e+00 -5.80303669e-01 6.12146139e-01 1.14168251e+00 2.35470384e-01 3.55974287e-01 4.00516421e-01 -3.11246544e-01 4.44130242e-01 3.39507163e-01 7.64159501e-01 -1.58522952e+00 -1.28237486e-01 -1.26797736e+00 -7.70754695e-01 -8.17487538e-01 4.18072343e-01 -1.08789504e+00 -5.61012387e-01 -1.62167382e+00 -1.70072224e-02 -4.37077641e-01 -2.18751684e-01 4.64178920e-01 8.87069851e-02 2.63508201e-01 2.81574354e-02 -4.53058004e-01 -4.35055941e-01 3.06151420e-01 1.28303552e+00 -3.08630377e-01 2.01296702e-01 -4.43177819e-02 -8.02285612e-01 8.99110258e-01 1.01969457e+00 1.03614755e-01 -1.31177366e-01 -4.96986300e-01 1.01318049e+00 7.59388581e-02 7.97202826e-01 -1.54440665e+00 2.73741663e-01 -4.72214341e-01 9.13076818e-01 -6.21513665e-01 3.92735861e-02 -5.18736064e-01 -4.87625062e-01 6.27665520e-01 -6.53140545e-01 3.14605474e-01 6.25449836e-01 4.52183709e-02 -1.30188629e-01 -2.00609237e-01 7.83858657e-01 -2.55468160e-01 -8.01632762e-01 1.53815687e-01 -1.43172279e-01 5.98695397e-01 8.14545333e-01 -1.64541677e-02 -3.78790438e-01 -3.65668565e-01 -3.11874449e-01 2.92119205e-01 1.41237855e-01 2.73261040e-01 4.95783687e-01 -1.07167172e+00 -6.73620701e-01 -1.57074764e-01 -3.76373768e-01 2.73492873e-01 -1.38577357e-01 5.53243577e-01 -1.01651692e+00 9.45496023e-01 -4.58454669e-01 -1.65571064e-01 -6.90270722e-01 6.27803147e-01 5.91557741e-01 -5.31632960e-01 -3.94146174e-01 8.14563870e-01 -2.50400484e-01 -4.86111909e-01 5.22270858e-01 -7.64574349e-01 1.50756100e-02 -2.86734670e-01 5.14240682e-01 7.31962323e-01 1.24831028e-01 -1.41576387e-03 -1.70925975e-01 2.33060256e-01 9.93102491e-02 3.12427968e-01 1.46521330e+00 7.20143676e-01 -3.69525224e-01 6.02409482e-01 9.70574617e-01 -4.26758021e-01 -6.02748871e-01 -9.15872604e-02 5.04212156e-02 2.05525368e-01 3.25027184e-04 -8.69976521e-01 -7.79016793e-01 1.38715136e+00 -7.00996071e-02 3.16917062e-01 8.52125704e-01 -4.80685204e-01 7.21144974e-01 1.02105236e+00 5.06632805e-01 -8.31432641e-01 1.43523097e-01 1.24899232e+00 5.90173960e-01 -1.28155434e+00 4.41378772e-01 3.97902206e-02 1.13970034e-01 1.62077022e+00 6.80786729e-01 -3.38118345e-01 6.45864248e-01 3.30220371e-01 -4.75933552e-01 -6.17659204e-02 -5.54962695e-01 1.16343275e-01 2.47254789e-01 3.97594452e-01 8.34535241e-01 -2.05193013e-01 4.39371765e-02 3.41036290e-01 -9.04473007e-01 2.96654195e-01 5.32280862e-01 7.22127914e-01 -4.95296717e-01 -7.78560638e-01 -3.12557369e-01 6.57892466e-01 -4.88904893e-01 -4.14681405e-01 -1.41164020e-01 1.15465844e+00 5.11206426e-02 5.75505614e-01 2.30568469e-01 -1.37388542e-01 -2.34990925e-01 3.45081151e-01 1.03291750e+00 -4.38613832e-01 -6.99865758e-01 -8.46018076e-01 -1.00178018e-01 -2.52172828e-01 -1.68465465e-01 -4.45633173e-01 -1.44674027e+00 -9.61880803e-01 5.66422865e-02 1.45393163e-01 4.88139033e-01 1.18881905e+00 -1.53308481e-01 7.50155509e-01 3.37892994e-02 -6.20039105e-01 -9.56502199e-01 -8.63433838e-01 -3.27866286e-01 -8.84958655e-02 4.92377639e-01 -5.73641658e-01 -1.87171668e-01 -1.81466550e-01]
[9.254510879516602, 7.161590099334717]
372e1a7f-7a3a-4cdf-af48-6fd0413ca8d8
pac-assisted-value-factorisation-with
2206.11420
null
https://arxiv.org/abs/2206.11420v3
https://arxiv.org/pdf/2206.11420v3.pdf
PAC: Assisted Value Factorisation with Counterfactual Predictions in Multi-Agent Reinforcement Learning
Multi-agent reinforcement learning (MARL) has witnessed significant progress with the development of value function factorization methods. It allows optimizing a joint action-value function through the maximization of factorized per-agent utilities due to monotonicity. In this paper, we show that in partially observable MARL problems, an agent's ordering over its own actions could impose concurrent constraints (across different states) on the representable function class, causing significant estimation error during training. We tackle this limitation and propose PAC, a new framework leveraging Assistive information generated from Counterfactual Predictions of optimal joint action selection, which enable explicit assistance to value function factorization through a novel counterfactual loss. A variational inference-based information encoding method is developed to collect and encode the counterfactual predictions from an estimated baseline. To enable decentralized execution, we also derive factorized per-agent policies inspired by a maximum-entropy MARL framework. We evaluate the proposed PAC on multi-agent predator-prey and a set of StarCraft II micromanagement tasks. Empirical results demonstrate improved results of PAC over state-of-the-art value-based and policy-based multi-agent reinforcement learning algorithms on all benchmarks.
['Vaneet Aggarwal', 'Tian Lan', 'Hanhan Zhou']
2022-06-22
null
null
null
null
['starcraft-ii']
['playing-games']
[ 9.33378178e-04 3.37637067e-01 -7.03318775e-01 -7.28595704e-02 -1.04379749e+00 -4.86216396e-01 7.58728504e-01 1.25225976e-01 -8.62132728e-01 1.56500614e+00 4.47000980e-01 -1.93858057e-01 -4.48424280e-01 -7.45422244e-01 -8.85571718e-01 -9.10247803e-01 -6.83379650e-01 7.03703523e-01 -2.49959201e-01 -3.14699680e-01 1.89516563e-02 -3.92066725e-02 -1.31687689e+00 3.90387863e-01 7.54361808e-01 9.52060044e-01 8.90948176e-02 6.61737859e-01 3.20014536e-01 1.33430743e+00 -6.44056678e-01 -4.94902045e-01 5.52560925e-01 -3.72807205e-01 -5.86950481e-01 -1.54029906e-01 -1.58371627e-01 -9.40099418e-01 -2.42287472e-01 9.04352486e-01 4.39851731e-01 4.73658800e-01 6.73738301e-01 -1.83149004e+00 -3.37780595e-01 1.17881715e+00 -5.31198561e-01 -1.75321057e-01 2.00210989e-01 6.36239350e-01 1.39208126e+00 4.43545468e-02 5.32073319e-01 1.66264749e+00 2.60085315e-01 8.45506847e-01 -1.45269680e+00 -3.12609583e-01 4.46296781e-01 2.69246399e-01 -5.45602739e-01 -4.82851192e-02 5.41304290e-01 -1.22144274e-01 1.28955257e+00 4.91448939e-02 9.68147695e-01 1.23477244e+00 6.27071977e-01 1.37311864e+00 1.28786016e+00 -9.97485146e-02 7.30385900e-01 -1.66452482e-01 -7.49645889e-01 6.63807034e-01 2.51128674e-01 6.90079510e-01 -5.55276155e-01 -7.26344049e-01 5.95860183e-01 2.43498031e-02 7.34889656e-02 -6.75348938e-01 -1.15248823e+00 1.09791851e+00 1.23441242e-01 -4.89854544e-01 -8.97097051e-01 7.07553566e-01 5.39891958e-01 5.77562153e-01 2.66708583e-01 6.78598702e-01 -8.23552370e-01 -4.41485137e-01 -4.32931870e-01 8.67719829e-01 9.18038368e-01 5.06323040e-01 6.20160639e-01 3.45473915e-01 -6.06903374e-01 4.67553824e-01 5.68201244e-01 7.01789260e-01 5.75752079e-01 -1.54333842e+00 5.68783462e-01 3.10773492e-01 6.15254462e-01 -2.76182264e-01 -3.51213902e-01 -2.17902213e-01 -3.51948768e-01 7.88061738e-01 2.88534850e-01 -7.79579878e-01 -5.63349962e-01 2.12380505e+00 5.04250586e-01 1.83737800e-01 5.09576201e-01 7.27732360e-01 -2.30757415e-01 5.75448513e-01 1.68056801e-01 -7.08830893e-01 1.03480399e+00 -9.52578902e-01 -7.07423449e-01 -4.43271995e-02 5.85742891e-01 4.26854119e-02 5.48841238e-01 4.44755971e-01 -1.10491216e+00 1.59898758e-01 -9.11768675e-01 6.50072575e-01 1.03171468e-01 -3.72921646e-01 9.94818926e-01 5.74614704e-01 -7.95354724e-01 8.47948074e-01 -1.11341822e+00 4.22253072e-01 6.88251197e-01 4.49884146e-01 -1.05356358e-01 3.31653357e-01 -1.16013098e+00 1.03891122e+00 5.14834166e-01 -3.71881634e-01 -1.74040949e+00 -8.97633195e-01 -8.24619889e-01 1.79158673e-01 1.14615273e+00 -7.83721626e-01 1.72279561e+00 -1.12395585e+00 -2.19870710e+00 -8.04744139e-02 4.97983068e-01 -1.29634833e+00 7.25045204e-01 -2.80800611e-01 1.82695035e-02 2.05180556e-01 1.78009570e-01 6.62814796e-01 1.08631921e+00 -1.09034920e+00 -9.64785993e-01 -1.39334694e-01 6.20362103e-01 5.05293429e-01 -1.37928993e-01 -5.02657712e-01 5.87661743e-01 -5.58714330e-01 -1.17985570e+00 -9.60332334e-01 -6.37093306e-01 -1.94511920e-01 1.30440397e-02 -3.08873087e-01 3.07234168e-01 -4.75784093e-01 1.06780303e+00 -1.65924716e+00 5.12117863e-01 -1.18018143e-01 6.60343915e-02 2.37165261e-02 -4.94971484e-01 5.81305385e-01 3.67470354e-01 -1.12707712e-01 -3.91453087e-01 -3.49867314e-01 6.45905197e-01 4.42358732e-01 -4.67689216e-01 6.11498773e-01 7.66036436e-02 8.77340138e-01 -1.15771735e+00 -2.67007440e-01 1.98932275e-01 4.94866341e-04 -1.10046911e+00 3.82579446e-01 -9.77307796e-01 1.45085305e-01 -5.58685422e-01 2.93477774e-01 4.39755172e-01 2.20940962e-01 7.20639467e-01 3.34149688e-01 -6.34455159e-02 7.19335824e-02 -1.14301550e+00 1.75555921e+00 -6.69028938e-01 -3.76383886e-02 5.19005992e-02 -9.80142236e-01 2.94235289e-01 3.57025892e-01 1.01809931e+00 -6.20204806e-01 2.01902896e-01 2.42220866e-03 1.41893119e-01 -2.14106500e-01 5.24940372e-01 -2.96175271e-01 -3.94665807e-01 7.93282688e-01 3.07090372e-01 -4.86099310e-02 3.95333380e-01 1.17781594e-01 1.05915248e+00 6.44609809e-01 8.15406919e-01 -2.69761592e-01 3.32937509e-01 -6.31527882e-03 9.72520828e-01 1.03548527e+00 -3.67097616e-01 -3.37529331e-01 8.88695836e-01 -4.11765218e-01 -9.17648077e-01 -1.03934050e+00 3.90757352e-01 1.21984708e+00 -1.50751933e-01 -3.33162278e-01 -6.05316162e-01 -1.01618981e+00 6.82485282e-01 1.07375193e+00 -8.68065298e-01 -1.05153620e-01 -4.47150111e-01 -9.74517345e-01 3.56750667e-01 3.48945111e-01 3.94410223e-01 -1.08815515e+00 -1.33206987e+00 5.69484591e-01 1.18385486e-01 -5.04330456e-01 -2.77196497e-01 1.84975907e-01 -6.00475907e-01 -1.05502963e+00 -5.66541076e-01 8.66818801e-02 2.19354436e-01 -2.95736015e-01 9.36616123e-01 -4.50458914e-01 7.02650100e-02 7.71080852e-01 -1.74661651e-01 -3.92156869e-01 -5.08075178e-01 -2.16846123e-01 4.77385223e-01 2.46933755e-02 -4.34817746e-02 -4.86553192e-01 -6.70408905e-01 -5.41488640e-02 -8.44975054e-01 1.88379940e-02 5.43002248e-01 1.30379689e+00 4.07310873e-01 -1.19676217e-02 8.97384405e-01 -4.95710254e-01 9.34372604e-01 -7.73312032e-01 -1.12557852e+00 5.10201991e-01 -6.28230512e-01 8.34768772e-01 9.04153228e-01 -5.13485849e-01 -1.38842607e+00 -2.61453595e-02 2.86631346e-01 -2.30351195e-01 2.95662999e-01 3.58871222e-01 -5.20854965e-02 4.52448130e-01 3.23657244e-01 2.53950864e-01 4.76955324e-01 -1.72330797e-01 6.72888279e-01 4.16205019e-01 7.43082389e-02 -1.17240787e+00 4.05213118e-01 4.42564696e-01 2.19400108e-01 -2.89888173e-01 -5.19881010e-01 9.15511325e-02 1.04020268e-01 -2.50611812e-01 6.00326896e-01 -1.08476985e+00 -1.71258080e+00 3.57876688e-01 -9.49642062e-01 -7.69085705e-01 -6.90497875e-01 6.64726079e-01 -1.36322951e+00 2.29903638e-01 -3.22702318e-01 -1.32389235e+00 -2.35469520e-01 -1.20583928e+00 8.04352939e-01 1.77967548e-01 2.54936099e-01 -8.95778418e-01 5.83407342e-01 7.18432814e-02 2.45159149e-01 2.84066230e-01 6.61835074e-01 -4.08959836e-01 -5.81555426e-01 3.98664117e-01 4.24907684e-01 2.69356757e-01 -5.76924160e-02 -3.27872664e-01 -4.96919483e-01 -7.52484858e-01 -1.36061966e-01 -7.45107591e-01 8.34690392e-01 4.82993692e-01 6.05009854e-01 -1.10718501e+00 -1.13598771e-01 3.23695391e-01 1.52646339e+00 3.84099662e-01 9.09742564e-02 4.78570849e-01 1.05006382e-01 2.42528707e-01 8.62053037e-01 1.39929581e+00 5.89135885e-01 6.97100699e-01 9.21071708e-01 9.42356050e-01 4.51550752e-01 -4.78309631e-01 1.08287609e+00 -8.30701888e-02 -1.56990007e-01 -2.75089860e-01 -3.28742981e-01 5.43333292e-01 -2.39215922e+00 -1.33843708e+00 6.99287295e-01 2.24071455e+00 1.05038488e+00 -1.65671274e-01 6.83962107e-01 -4.72255796e-01 1.99312150e-01 1.83704287e-01 -1.29198563e+00 -4.79869843e-01 9.67422500e-02 -1.97299849e-03 7.99122572e-01 8.06339622e-01 -1.02540302e+00 7.68131196e-01 5.60291910e+00 9.44094956e-01 -6.74134552e-01 2.24010482e-01 3.71483684e-01 -6.67146564e-01 -5.41683972e-01 -1.74038321e-01 -7.09798872e-01 4.98234659e-01 9.16520536e-01 -6.50132716e-01 1.08010411e+00 1.11298239e+00 2.87472993e-01 -3.55167419e-01 -1.08264208e+00 5.95241308e-01 -2.94228286e-01 -1.33786511e+00 -1.98860690e-02 1.47271276e-01 9.83794868e-01 2.03162163e-01 1.05373248e-01 6.97981000e-01 1.29086137e+00 -7.46483445e-01 9.34355497e-01 3.95284861e-01 4.75735188e-01 -1.16478431e+00 4.82214242e-01 4.78270799e-01 -9.67157543e-01 -8.38496685e-01 -3.60108614e-01 -5.07923484e-01 1.35277525e-01 -1.28475986e-02 -9.68096673e-01 5.98078549e-01 1.17733508e-01 5.62845886e-01 1.62450120e-01 6.40745640e-01 -3.83026749e-01 3.51536006e-01 -3.32445055e-01 -3.65023911e-01 5.65423369e-01 -1.93183348e-01 7.64570892e-01 6.40106976e-01 1.14213988e-01 -1.84291616e-01 4.80858773e-01 7.42711663e-01 -6.63542524e-02 -1.51923358e-01 -4.63505000e-01 -2.11606145e-01 3.32944036e-01 1.08457804e+00 -2.39867881e-01 -2.39888921e-01 -3.06333303e-01 7.11900711e-01 5.04762530e-01 2.55925059e-01 -1.04163110e+00 2.11158186e-01 1.40298128e+00 -5.74831128e-01 4.28957164e-01 -4.43589650e-02 2.31407240e-01 -1.44323552e+00 -1.95340782e-01 -1.10680342e+00 6.51096463e-01 -6.87343627e-02 -1.32454038e+00 1.46991923e-01 2.93420076e-01 -1.04093802e+00 -1.09745824e+00 -5.38385689e-01 -3.89247030e-01 4.13201779e-01 -1.56304073e+00 -8.93261492e-01 7.41001308e-01 4.57495332e-01 5.52422643e-01 -5.88329196e-01 8.31722677e-01 -2.18834579e-01 -4.45681393e-01 4.99103457e-01 6.02870107e-01 -3.42094362e-01 2.81258762e-01 -1.53269744e+00 1.67083386e-02 5.40048718e-01 -9.67810303e-02 2.95924582e-02 7.35498309e-01 -6.44888937e-01 -1.80979502e+00 -9.16302741e-01 -1.23394012e-01 -1.70172408e-01 8.97175491e-01 -2.62052827e-02 -1.06859230e-01 6.71842456e-01 4.97390807e-01 -2.14433610e-01 4.50182855e-01 -1.07136607e-01 -7.15269670e-02 -1.24035418e-01 -1.35587704e+00 7.46059358e-01 7.57285476e-01 -7.02171698e-02 -6.60639107e-01 1.12594873e-01 8.71695757e-01 -1.98518664e-01 -8.75532806e-01 2.08732672e-02 7.41121948e-01 -7.91386366e-01 8.70224416e-01 -1.23165011e+00 5.97655535e-01 -1.19542174e-01 -3.59385133e-01 -1.93340254e+00 -1.14691265e-01 -1.15275264e+00 -5.85125744e-01 6.96111262e-01 2.14470774e-01 -7.61599839e-01 7.20213592e-01 5.89311361e-01 8.78902748e-02 -8.36403072e-01 -1.36487031e+00 -1.01184762e+00 2.54169077e-01 -2.96592325e-01 7.68500686e-01 5.24155915e-01 3.64793748e-01 8.53901356e-02 -9.46587443e-01 8.56445637e-03 1.16261804e+00 1.79251462e-01 6.97226882e-01 -6.09041870e-01 -1.03938997e+00 -4.33294892e-01 1.36379488e-02 -8.15634847e-01 6.46081865e-01 -8.26641023e-01 6.43978419e-04 -1.26473927e+00 2.10695878e-01 -3.37241627e-02 -4.05289948e-01 6.98035955e-01 -9.23118591e-02 -5.13833404e-01 8.44970226e-01 -3.64275157e-01 -9.73299146e-01 1.18349040e+00 1.28106689e+00 -3.07166487e-01 -1.96088210e-01 -1.00426748e-02 -5.28979301e-01 5.61940432e-01 8.49689066e-01 -5.32867074e-01 -6.74673796e-01 -1.63926005e-01 4.17685002e-01 7.60071099e-01 2.28533104e-01 -6.45271480e-01 -1.43355474e-01 -8.88990343e-01 -8.03364590e-02 -9.63267386e-02 4.03724790e-01 -7.95816720e-01 -2.39696391e-02 9.80218291e-01 -6.14052892e-01 6.87014684e-02 1.42594233e-01 9.72351134e-01 1.04825027e-01 -2.25996315e-01 5.42287171e-01 -3.03989440e-01 -6.54312372e-01 3.08042318e-01 -5.99920630e-01 2.12201238e-01 1.34658110e+00 5.79137266e-01 -3.58167291e-01 -4.72638845e-01 -3.70442539e-01 7.84955263e-01 1.05784863e-01 1.74735740e-01 5.46952546e-01 -1.35905182e+00 -9.66397285e-01 -9.70876887e-02 -2.87527114e-01 -4.39103037e-01 3.83265972e-01 4.70035881e-01 1.05420671e-01 2.71450073e-01 -6.89276695e-01 8.49629045e-02 -8.29861701e-01 5.27653456e-01 4.31635678e-01 -1.00481570e+00 -2.36133039e-01 4.12345499e-01 -8.18188637e-02 -3.35972279e-01 1.49157736e-02 -2.85659909e-01 5.61473742e-02 2.57235974e-01 5.77340424e-01 6.59244359e-01 -6.36461139e-01 -6.11579493e-02 -1.69395670e-01 -3.33849639e-01 -1.66016638e-01 -7.68898904e-01 1.71007776e+00 6.38932362e-02 1.94887683e-01 1.17384732e-01 7.61449337e-01 -4.26690936e-01 -2.16701674e+00 -1.31656662e-01 -1.20615214e-01 -4.35063332e-01 3.55943665e-02 -1.14783597e+00 -7.77451038e-01 4.03203487e-01 3.14866513e-01 1.44622833e-01 7.97676504e-01 -4.89541501e-01 5.08008480e-01 6.79156840e-01 8.55292857e-01 -1.40780234e+00 8.52802768e-03 4.15512025e-01 8.91213059e-01 -1.26981246e+00 1.01627141e-01 5.49201667e-01 -1.20611846e+00 9.05010283e-01 4.30087179e-01 -2.71876574e-01 3.66365492e-01 3.83643597e-01 -4.21863019e-01 3.52783054e-01 -1.50564778e+00 -2.55090803e-01 -1.80136397e-01 7.45104730e-01 -1.05019420e-01 5.85931659e-01 -2.99183398e-01 6.92234159e-01 6.81456402e-02 1.02622606e-01 7.15881169e-01 9.69927549e-01 -3.49842429e-01 -1.37356448e+00 -2.25601256e-01 5.83191037e-01 -4.91879165e-01 6.43225983e-02 1.87986329e-01 5.86246669e-01 -2.21974298e-01 8.26819539e-01 9.67953820e-03 -3.77523638e-02 -1.08718477e-01 -8.41557235e-02 8.67223680e-01 -2.46206850e-01 -6.90250635e-01 -1.15186945e-01 1.81923479e-01 -8.27576160e-01 -5.16867220e-01 -8.86951983e-01 -1.19284105e+00 -2.93250859e-01 1.55884862e-01 3.88563484e-01 4.90961194e-01 9.81453240e-01 3.58391136e-01 6.44516766e-01 7.20746517e-01 -8.15241396e-01 -1.71348548e+00 -7.75450706e-01 -5.55872679e-01 2.06164271e-01 5.72068214e-01 -1.00535619e+00 -1.78484961e-01 -4.57110435e-01]
[3.7670390605926514, 2.0681231021881104]
86976613-7bf1-456b-82c5-d500533d2921
monocular-3d-object-detection-using-multi
2212.11804
null
https://arxiv.org/abs/2212.11804v1
https://arxiv.org/pdf/2212.11804v1.pdf
Monocular 3D Object Detection using Multi-Stage Approaches with Attention and Slicing aided hyper inference
3D object detection is vital as it would enable us to capture objects' sizes, orientation, and position in the world. As a result, we would be able to use this 3D detection in real-world applications such as Augmented Reality (AR), self-driving cars, and robotics which perceive the world the same way we do as humans. Monocular 3D Object Detection is the task to draw 3D bounding box around objects in a single 2D RGB image. It is localization task but without any extra information like depth or other sensors or multiple images. Monocular 3D object detection is an important yet challenging task. Beyond the significant progress in image-based 2D object detection, 3D understanding of real-world objects is an open challenge that has not been explored extensively thus far. In addition to the most closely related studies.
['Ashish Patel', 'Abonia Sojasingarayar']
2022-12-22
null
null
null
null
['monocular-3d-object-detection']
['computer-vision']
[ 9.89828184e-02 -1.63990825e-01 2.02761710e-01 -2.35028028e-01 9.24237967e-02 -7.84826398e-01 4.96305585e-01 -2.50436049e-02 -4.58533257e-01 2.91924417e-01 -5.47745168e-01 -4.77302819e-01 4.83384699e-01 -6.86647594e-01 -6.26309335e-01 -3.28630507e-01 1.65992066e-01 5.87148786e-01 9.19364214e-01 -2.90166825e-01 3.80129874e-01 1.23715281e+00 -1.79464543e+00 -2.30604485e-01 1.47022709e-01 1.24738967e+00 5.61934471e-01 9.15202320e-01 -6.19707219e-02 2.75320232e-01 -3.07790279e-01 -1.01208396e-01 4.97306108e-01 2.72824485e-02 -1.29995018e-01 3.38278443e-01 4.68838543e-01 -6.34278953e-01 -3.99475634e-01 1.03968060e+00 1.94326997e-01 -9.45656840e-03 4.10432816e-01 -1.28548002e+00 -3.42275441e-01 -4.74266976e-01 -7.74313807e-01 6.98203743e-02 8.45731854e-01 4.99429330e-02 2.40629956e-01 -9.33380306e-01 6.13104343e-01 1.38215435e+00 2.00562581e-01 4.14714187e-01 -8.07822227e-01 -4.03183430e-01 -1.78769249e-02 1.62667349e-01 -1.29212844e+00 -1.05244279e-01 9.63974774e-01 -2.69819468e-01 9.09391880e-01 4.31095511e-01 6.65184081e-01 6.18712485e-01 1.09042093e-01 6.82853997e-01 1.26285887e+00 -6.06583893e-01 1.62831724e-01 5.30719161e-01 4.47686203e-02 6.05355263e-01 7.50368178e-01 2.10646912e-01 -3.18810910e-01 2.91132092e-01 1.10427904e+00 4.20270830e-01 -7.69366547e-02 -1.08907783e+00 -1.39966261e+00 3.96104604e-01 7.56373584e-01 1.03753708e-01 -2.38640457e-01 2.16805175e-01 -1.86139435e-01 -1.50657333e-02 1.75696582e-01 3.95479232e-01 -4.30872351e-01 -1.91562325e-01 -1.92124948e-01 1.63145319e-01 3.96898657e-01 1.29211652e+00 8.73116970e-01 -1.13254212e-01 7.16016352e-01 1.59712479e-01 4.79325891e-01 9.66049194e-01 2.09081113e-01 -1.00240147e+00 3.97469759e-01 1.13636875e+00 5.77457964e-01 -9.85809863e-01 -4.93972301e-01 -1.17130503e-01 -4.87836301e-01 9.53754663e-01 4.28888440e-01 1.35976702e-01 -8.81526887e-01 9.13458169e-01 7.66047001e-01 -1.67783067e-01 -1.62603706e-01 1.28967762e+00 8.92805696e-01 3.32516670e-01 -6.85304999e-01 2.09516674e-01 1.61491501e+00 -4.06413585e-01 -4.63425219e-01 -6.66247547e-01 3.65786135e-01 -8.92942548e-01 7.70284772e-01 4.61037844e-01 -1.03524244e+00 -4.99213189e-01 -1.19219363e+00 -3.93225014e-01 -7.66642034e-01 3.80063779e-03 9.03358698e-01 8.53989720e-01 -6.59385741e-01 -9.79079530e-02 -8.84943604e-01 -4.99639481e-01 1.81026757e-01 4.11366135e-01 -7.77124524e-01 -2.33778179e-01 -7.19502270e-01 1.42608249e+00 4.49418277e-01 2.02739894e-01 -3.18250686e-01 -2.75683522e-01 -1.12588418e+00 -3.98238868e-01 7.11759627e-01 -5.71850061e-01 1.18368423e+00 -2.05944642e-01 -1.44152057e+00 1.32878006e+00 -3.43890578e-01 -3.34545434e-01 5.30427158e-01 -3.63103956e-01 -6.10496886e-02 8.13797340e-02 -1.85880035e-01 7.50687003e-01 5.98856390e-01 -1.50728786e+00 -6.53452277e-01 -1.10093844e+00 1.58819228e-01 5.42697310e-01 3.57189357e-01 -2.75520217e-02 -2.79547215e-01 1.09912559e-01 9.88515258e-01 -1.00491285e+00 -3.18995416e-01 6.60965145e-01 -2.00093165e-01 -3.73961893e-03 1.15534246e+00 -6.04411662e-02 3.41785789e-01 -2.02185869e+00 -2.28717208e-01 -1.71328261e-01 2.52493232e-01 3.28465521e-01 3.66584867e-01 1.62618250e-01 3.13592166e-01 -1.48376271e-01 2.30744585e-01 -2.44626716e-01 -3.39177460e-03 6.02739118e-02 -2.35341653e-01 7.30477870e-01 1.01302154e-01 1.00002658e+00 -8.85741115e-01 -3.11430037e-01 9.33082402e-01 5.80792069e-01 -8.45682099e-02 1.15578659e-01 6.39617397e-03 3.00319374e-01 -4.64936137e-01 7.62275517e-01 9.48179185e-01 -1.80245023e-02 -2.43844122e-01 -3.08895335e-02 -3.04390073e-01 1.42285243e-01 -1.50673711e+00 1.49509251e+00 -3.17391276e-01 9.63978231e-01 -2.22204756e-02 -7.04239905e-01 1.29268682e+00 1.62679777e-01 1.39895082e-01 -1.00465500e+00 1.20640412e-01 2.61402100e-01 -1.71884939e-01 -3.32965195e-01 8.42371762e-01 -1.37531847e-01 -1.47266760e-01 2.35974416e-01 -5.97886801e-01 -8.30265641e-01 -2.52632618e-01 5.01463152e-02 9.65476274e-01 1.04367211e-01 6.92148149e-01 4.48307395e-01 2.43340909e-01 2.42693275e-01 8.72032940e-02 7.26805627e-01 -2.68062413e-01 6.47645533e-01 -1.54039174e-01 -5.24936080e-01 -1.07064617e+00 -1.24941528e+00 -2.94731498e-01 2.56452560e-01 8.75368595e-01 2.24268466e-01 2.99054906e-02 -2.87675112e-01 3.62782180e-01 4.09077048e-01 -4.56954241e-01 1.49804354e-01 -4.86489236e-01 -3.11380863e-01 -1.93650246e-01 6.14880383e-01 3.53073329e-01 -8.18812430e-01 -1.72625721e+00 -6.41069561e-03 3.28729361e-01 -1.53023648e+00 4.84520160e-02 3.59416574e-01 -9.43788052e-01 -1.02504373e+00 -6.65745676e-01 -4.67075974e-01 5.99708557e-01 1.14589286e+00 1.00635242e+00 -3.04744005e-01 -6.65737689e-01 4.85947311e-01 -4.52914178e-01 -9.95991349e-01 7.89618418e-02 -5.59195936e-01 2.16916233e-01 -2.42530093e-01 6.45901322e-01 -2.23949119e-01 -6.93622768e-01 6.14799440e-01 -4.14617896e-01 1.56515896e-01 5.95206618e-01 1.44596338e-01 5.20182312e-01 -6.83974102e-02 3.59715410e-02 -3.77084523e-01 -5.87611087e-02 -6.76469058e-02 -1.04344738e+00 -1.43416628e-01 8.99133179e-03 -5.34032166e-01 9.55686998e-03 -4.15223777e-01 -7.84067035e-01 5.42647362e-01 4.03569192e-01 -3.85881811e-01 -8.45284104e-01 -1.79088935e-01 -1.71989158e-01 -1.82729572e-01 7.06326842e-01 1.59286648e-01 -1.24373771e-02 -4.85587358e-01 2.52189100e-01 9.54394341e-01 6.59380257e-01 6.48297742e-02 8.31591785e-01 1.03555179e+00 3.78087997e-01 -1.15195656e+00 -7.05837786e-01 -9.04904664e-01 -1.17487836e+00 -2.43676096e-01 7.79697657e-01 -8.94972861e-01 -1.03942358e+00 1.79413378e-01 -1.53107953e+00 2.99681835e-02 -2.97622949e-01 7.92540252e-01 -5.94185770e-01 2.70329803e-01 -6.31916299e-02 -1.40587020e+00 2.72888929e-01 -1.00091219e+00 1.39689386e+00 4.49531436e-01 -4.20951098e-02 -5.40753841e-01 -3.42902422e-01 4.75547165e-01 6.12745844e-02 4.49055344e-01 3.13199133e-01 -1.00771859e-01 -1.08374560e+00 -6.67996585e-01 -4.19983119e-01 -1.89339131e-01 1.86518684e-01 -1.78223923e-01 -9.68736351e-01 1.06194519e-01 2.13683948e-01 -2.86640506e-02 4.60984826e-01 2.60296077e-01 6.08674765e-01 5.00213861e-01 -4.97490913e-01 3.65653723e-01 1.25445199e+00 7.09848583e-01 4.30527180e-01 3.87700945e-01 5.42865872e-01 5.08483589e-01 1.03635764e+00 1.82294190e-01 3.31150264e-01 1.06409836e+00 9.48722482e-01 -2.32376590e-01 -2.11442277e-01 -9.77462754e-02 -7.28338137e-02 3.02512497e-01 -5.23554683e-02 3.04354671e-02 -1.06601143e+00 2.28256986e-01 -1.56670988e+00 -7.03655422e-01 -7.17432976e-01 2.38818789e+00 2.43855104e-01 3.36478442e-01 -3.83762196e-02 4.55943763e-01 4.14828718e-01 -4.80065197e-01 -8.63987327e-01 -2.78267235e-01 -1.70353785e-01 -1.70149758e-01 5.87779045e-01 2.93675661e-01 -9.46630418e-01 8.13804984e-01 6.10014391e+00 4.29335283e-03 -1.20270538e+00 -1.97102696e-01 8.25014040e-02 1.10757716e-01 1.03534997e-01 -4.34279405e-02 -1.05265510e+00 1.40969634e-01 2.20849499e-01 1.83512419e-01 1.62638471e-01 8.97400677e-01 6.18965849e-02 -9.44080770e-01 -1.14255452e+00 1.39729881e+00 1.41041175e-01 -8.45635355e-01 -4.63727027e-01 3.47424299e-01 5.27832985e-01 -2.31635440e-02 8.70263390e-03 -1.81230739e-01 8.10011402e-02 -6.72165513e-01 7.75125921e-01 3.89089465e-01 6.28998399e-01 -3.55420381e-01 7.37081349e-01 8.94498944e-01 -1.12008095e+00 -5.92407472e-02 -7.13972747e-01 -4.95583653e-01 3.09276372e-01 6.86252892e-01 -1.30925739e+00 1.84777781e-01 6.43408895e-01 4.20477748e-01 -5.90819359e-01 1.39805484e+00 -2.53432691e-01 -1.33569628e-01 -5.76143384e-01 -3.19073826e-01 -9.69595611e-02 -8.38915706e-02 5.59829473e-01 7.46799707e-01 3.78651828e-01 3.71605903e-01 -6.74181210e-04 8.06064844e-01 2.75960267e-01 -3.16066474e-01 -1.13784707e+00 3.85543734e-01 3.28750223e-01 1.03783917e+00 -1.00151432e+00 -1.13712490e-01 -5.39315403e-01 9.32545125e-01 -1.38112694e-01 1.67646989e-01 -4.58802611e-01 -4.25381154e-01 4.56540048e-01 3.89777869e-01 3.38879883e-01 -9.51399624e-01 -3.30475777e-01 -1.12381136e+00 1.99327141e-01 -1.80086970e-01 -2.90396065e-01 -1.43457913e+00 -8.36209118e-01 4.72210646e-01 1.11458078e-01 -1.44224024e+00 -9.25438926e-02 -1.16368520e+00 -5.05599491e-02 8.07836652e-01 -1.59489906e+00 -9.73667324e-01 -6.48715138e-01 4.33394670e-01 3.92487407e-01 2.74009317e-01 6.24429822e-01 -5.65153286e-02 -2.96007264e-02 -1.33796409e-01 -2.22989142e-01 -1.05183981e-01 4.14280921e-01 -1.15883732e+00 5.24285078e-01 7.07527757e-01 3.57940555e-01 4.21788335e-01 6.95248008e-01 -5.34787118e-01 -1.92218554e+00 -4.91386771e-01 5.10407925e-01 -1.08456540e+00 3.27806532e-01 -7.84875214e-01 -6.34742677e-01 5.84169984e-01 -3.58136475e-01 4.96110082e-01 1.75232217e-01 -1.65170074e-01 -2.87071615e-01 -1.28785416e-01 -1.18400574e+00 3.25483292e-01 1.18151772e+00 -3.89306456e-01 -5.95474482e-01 9.53933820e-02 6.86073363e-01 -1.00435305e+00 -4.55754668e-01 4.35425878e-01 7.65202701e-01 -1.08465648e+00 1.39145041e+00 -2.61628002e-01 -2.60184377e-01 -7.16015697e-01 -4.18656677e-01 -8.10469270e-01 1.90480456e-01 -2.80519605e-01 -3.98797959e-01 6.56937242e-01 9.16160122e-02 -7.92577147e-01 1.02292442e+00 7.72642195e-01 4.14555669e-02 -3.60031843e-01 -9.93365467e-01 -9.06415522e-01 -5.41740477e-01 -5.42499602e-01 4.93574977e-01 4.10534620e-01 -2.50017613e-01 2.35797748e-01 -6.59031197e-02 4.55745041e-01 7.37053454e-01 6.76480114e-01 1.26661170e+00 -1.61564422e+00 1.67168602e-01 -2.47588173e-01 -1.13792133e+00 -1.44910717e+00 -3.77540559e-01 -4.12192225e-01 1.81822218e-02 -1.46936262e+00 3.20329219e-02 -6.06613219e-01 2.14678958e-01 -6.36852980e-02 4.57182266e-02 6.30554378e-01 4.38567221e-01 4.53412794e-02 -6.05870187e-01 3.38215679e-01 1.45355701e+00 2.69490302e-01 -1.08231999e-01 2.58189321e-01 -3.20690423e-01 8.61795187e-01 7.86392212e-01 -3.51804644e-01 -2.60458112e-01 -3.08924794e-01 2.04024419e-01 2.55209982e-01 7.10166156e-01 -1.03461027e+00 3.31732243e-01 -2.89445907e-01 7.52551794e-01 -1.34480608e+00 1.12276793e+00 -1.19221318e+00 -1.13379955e-01 2.88537234e-01 5.18634617e-01 1.48871750e-01 1.46203622e-01 4.19791341e-01 3.41650397e-02 -2.46501416e-01 5.38404465e-01 -6.40824437e-01 -1.05105114e+00 7.66621903e-02 -2.52846599e-01 -3.48956078e-01 1.38028252e+00 -9.66118574e-01 -2.41010904e-01 -3.39342326e-01 -6.69874489e-01 4.97728260e-03 8.15359235e-01 4.91302371e-01 1.02350688e+00 -1.12143159e+00 -2.96116412e-01 4.35013086e-01 2.28739515e-01 4.51944053e-01 -1.20571163e-02 5.20373166e-01 -6.26091063e-01 8.71086299e-01 -1.71185642e-01 -1.12639070e+00 -1.32879245e+00 8.87144923e-01 2.10309446e-01 4.09764528e-01 -6.25804961e-01 4.86222297e-01 3.24628860e-01 -5.01915932e-01 2.07419544e-01 -4.67172593e-01 1.33524016e-01 -4.23066288e-01 5.58509886e-01 3.30351770e-01 1.35284096e-01 -6.71326816e-01 -4.07785773e-01 1.00146413e+00 1.77799851e-01 -1.12608664e-01 1.10873854e+00 -4.95685399e-01 2.55378813e-01 8.17756116e-01 8.93166840e-01 -2.39807442e-01 -1.26472318e+00 -2.27829620e-01 -8.76457617e-02 -1.05746961e+00 7.18456088e-03 -5.10875463e-01 -4.85686600e-01 1.43428457e+00 7.10105121e-01 3.53394449e-01 8.45767438e-01 2.93024719e-01 3.66160363e-01 5.70524991e-01 9.78606820e-01 -6.77535117e-01 3.10902923e-01 4.71035719e-01 7.88828671e-01 -1.64383256e+00 1.88488513e-01 -6.55140221e-01 -3.18559289e-01 1.04258013e+00 8.28895748e-01 9.17029157e-02 4.76778209e-01 3.57434958e-01 6.00680076e-02 -3.24176759e-01 -3.42558116e-01 -5.85570455e-01 2.90996909e-01 8.85687888e-01 1.34736761e-01 2.29286104e-01 2.64532000e-01 -1.97666317e-01 -9.53592509e-02 -2.21555397e-01 6.18064582e-01 1.15122998e+00 -7.24608064e-01 -8.12770307e-01 -8.07936370e-01 2.91343704e-02 1.52582437e-01 5.78174591e-01 -4.15729254e-01 9.84591484e-01 4.81373689e-04 9.15134847e-01 2.88666040e-01 1.82830449e-02 7.50860333e-01 -2.18009546e-01 8.74217033e-01 -7.98695743e-01 1.22716501e-01 -1.44985646e-01 -4.48649287e-01 -5.27108252e-01 -2.66394496e-01 -6.59921587e-01 -1.33121765e+00 -1.77611895e-02 -6.27578318e-01 -4.78181452e-01 1.49152422e+00 5.57521045e-01 1.48811594e-01 -2.83949450e-02 5.64953804e-01 -1.44993734e+00 -1.09531589e-01 -4.71559227e-01 -6.07055128e-01 1.67976663e-01 6.16993368e-01 -8.80258501e-01 -1.41506493e-01 -2.88056433e-01]
[7.7036943435668945, -2.5074880123138428]
252dfae8-3c79-4859-8731-65362d70fa17
towards-a-better-understanding-of
2305.18491
null
https://arxiv.org/abs/2305.18491v1
https://arxiv.org/pdf/2305.18491v1.pdf
Towards a Better Understanding of Representation Dynamics under TD-learning
TD-learning is a foundation reinforcement learning (RL) algorithm for value prediction. Critical to the accuracy of value predictions is the quality of state representations. In this work, we consider the question: how does end-to-end TD-learning impact the representation over time? Complementary to prior work, we provide a set of analysis that sheds further light on the representation dynamics under TD-learning. We first show that when the environments are reversible, end-to-end TD-learning strictly decreases the value approximation error over time. Under further assumptions on the environments, we can connect the representation dynamics with spectral decomposition over the transition matrix. This latter finding establishes fitting multiple value functions from randomly generated rewards as a useful auxiliary task for representation learning, as we empirically validate on both tabular and Atari game suites.
['Rémi Munos', 'Yunhao Tang']
2023-05-29
null
null
null
null
['value-prediction']
['computer-code']
[ 3.80673148e-02 2.57456988e-01 -7.59773910e-01 -1.18984714e-01 -8.37758243e-01 -7.46941507e-01 5.37485898e-01 3.18429887e-01 -4.77282286e-01 1.03467464e+00 3.78732532e-01 -5.25980055e-01 -5.67838490e-01 -5.89721203e-01 -7.18707085e-01 -5.01070082e-01 -6.04621530e-01 4.44654077e-01 -6.87963970e-04 -5.53710878e-01 3.91249061e-01 2.79594153e-01 -1.45029783e+00 1.58292890e-01 6.62259221e-01 8.60214949e-01 -1.50202572e-01 7.26712406e-01 1.90846041e-01 1.40294504e+00 -5.05558014e-01 -4.22862284e-02 4.02510583e-01 -4.99859959e-01 -8.95516455e-01 -1.01493932e-02 -2.99152702e-01 -6.48896754e-01 -5.93281746e-01 7.77311206e-01 4.19166237e-01 6.81637347e-01 7.72844434e-01 -1.27633119e+00 -3.55051637e-01 8.64426851e-01 -2.72892684e-01 2.79180557e-01 4.01785046e-01 5.51923931e-01 1.16538286e+00 -1.65450647e-01 6.82727873e-01 1.19862902e+00 6.75402522e-01 6.14339888e-01 -1.57255006e+00 -3.23363513e-01 3.31809640e-01 2.89703518e-01 -8.68991375e-01 -4.97960389e-01 6.08908355e-01 -4.53580379e-01 1.04406381e+00 -1.48476705e-01 8.88034225e-01 1.05621600e+00 2.12434098e-01 9.97765303e-01 1.22816348e+00 -4.65785921e-01 6.42826557e-01 -1.84881464e-01 2.04167038e-01 3.56345087e-01 2.64547288e-01 8.08484614e-01 -4.82897103e-01 -1.97181106e-01 8.03001285e-01 -2.17989951e-01 1.43145576e-01 -8.78617525e-01 -7.93631613e-01 9.19541895e-01 6.69075623e-02 -1.36664718e-01 -3.82018983e-01 8.03466618e-01 6.20818555e-01 7.77819276e-01 2.72094965e-01 7.06955135e-01 -4.66924012e-01 -1.05813575e+00 -4.82259989e-01 6.68047786e-01 7.11752534e-01 6.96010172e-01 5.73175490e-01 4.52575445e-01 -4.26660478e-01 5.04254282e-01 1.65916264e-01 4.99914050e-01 4.30846602e-01 -1.38789475e+00 3.33618701e-01 1.86104655e-01 5.33237278e-01 -3.58315945e-01 -3.82359058e-01 -4.97905672e-01 5.76081797e-02 3.20303112e-01 8.65058064e-01 -5.78202128e-01 -6.83918595e-01 2.03439450e+00 5.28369024e-02 1.00339547e-01 2.65309304e-01 5.07428288e-01 -1.18515864e-01 4.18843180e-01 -6.25875778e-03 -4.28561509e-01 5.15016913e-01 -4.72355574e-01 -7.56641090e-01 -1.79564893e-01 1.15978694e+00 -1.37993664e-01 1.08134973e+00 5.07227480e-01 -1.23126924e+00 -2.61617899e-01 -7.92036235e-01 4.03068125e-01 1.19681373e-01 -3.34471405e-01 8.02063882e-01 5.59153616e-01 -1.00081229e+00 9.50474381e-01 -1.01103044e+00 -2.78428376e-01 3.37630272e-01 3.95080239e-01 3.11637431e-01 -3.80227366e-03 -1.27275133e+00 1.26452816e+00 3.81973326e-01 -4.05846655e-01 -1.44415152e+00 -4.05913204e-01 -5.66489041e-01 6.24105558e-02 6.18825912e-01 -2.93697357e-01 1.96638048e+00 -1.07732010e+00 -1.70285749e+00 2.83633947e-01 7.30588958e-02 -8.93574655e-01 4.10222888e-01 1.68444943e-02 -1.33980930e-01 -9.81900468e-02 -2.39392459e-01 2.83432454e-01 7.17412174e-01 -1.24512136e+00 -5.00271857e-01 -2.28593990e-01 3.94405216e-01 2.48121232e-01 -1.78601667e-02 -4.09722567e-01 2.61138767e-01 -4.09318119e-01 -3.20135593e-01 -1.07403791e+00 -3.69664907e-01 -3.12406749e-01 1.24730663e-02 -1.72545552e-01 1.53849825e-01 -2.75912881e-01 1.25856936e+00 -1.91025555e+00 1.11368291e-01 4.58445728e-01 -3.48398276e-02 -4.60051298e-02 -3.22523981e-01 7.48011827e-01 -2.87165463e-01 -1.15438335e-01 1.31639361e-01 -4.81262952e-02 3.58432770e-01 2.67021805e-01 -6.30686820e-01 5.02956927e-01 -8.28297138e-02 1.00286484e+00 -1.02423882e+00 -2.84779929e-02 9.29817781e-02 -6.73002452e-02 -7.40045726e-01 2.90781677e-01 -6.31797493e-01 2.47905046e-01 -4.79068995e-01 1.86406255e-01 1.06670462e-01 2.10273638e-02 5.51409364e-01 5.11704385e-01 -8.28981679e-03 6.46549463e-01 -1.11726332e+00 1.63333130e+00 -1.60742536e-01 3.74061614e-01 -3.95507157e-01 -1.14933085e+00 8.14868212e-01 -5.34603335e-02 6.91739261e-01 -1.10451043e+00 1.69406250e-01 -2.12287493e-02 4.11519021e-01 -2.80992270e-01 6.17374897e-01 -2.98184991e-01 -3.35978121e-01 8.95126343e-01 -9.51808915e-02 -4.87682968e-02 2.63206631e-01 1.60371661e-01 1.11806738e+00 5.08038163e-01 3.19736034e-01 -1.38012484e-01 -2.71406710e-01 4.52027142e-01 5.76725900e-01 9.60283935e-01 -2.33196557e-01 1.71383899e-02 1.13674879e+00 -3.10696334e-01 -1.05356145e+00 -1.14435112e+00 3.68037187e-02 1.47605669e+00 -6.73940480e-02 -4.76975322e-01 -5.75310349e-01 -4.62228328e-01 2.64823318e-01 1.00195992e+00 -8.43419015e-01 -7.65979886e-01 -2.48746663e-01 -5.09787798e-01 6.09720230e-01 8.23104024e-01 -7.67151862e-02 -9.36309457e-01 -7.93115079e-01 5.05811512e-01 1.72382638e-01 -5.16914606e-01 -2.84195006e-01 6.98377788e-01 -1.12022722e+00 -9.52619791e-01 -3.46545130e-01 -3.01898003e-01 1.52048290e-01 8.02575275e-02 1.01119423e+00 -1.81485444e-01 2.75873601e-01 9.23847079e-01 -3.01222980e-01 -3.68764222e-01 -6.07464969e-01 3.39713879e-02 2.34698474e-01 -5.99944890e-01 1.56309634e-01 -6.44102812e-01 -3.89819145e-01 3.93871740e-02 -5.69597006e-01 -1.59907565e-01 3.82523745e-01 9.32343721e-01 5.03156066e-01 -1.08258403e-03 7.04106688e-01 -7.60377109e-01 1.19971633e+00 -6.19518042e-01 -7.29956448e-01 1.95618913e-01 -9.50589120e-01 6.80695891e-01 5.45852005e-01 -5.97446144e-01 -8.54445279e-01 -1.02400124e-01 2.25548729e-01 -3.87195587e-01 1.92698345e-01 8.14325154e-01 3.36826324e-01 3.11482340e-01 8.92521977e-01 4.73539978e-01 4.14383203e-01 -1.39465034e-01 5.14117956e-01 1.94824263e-01 2.05637529e-01 -1.20855117e+00 6.32791340e-01 -3.36061306e-02 3.80886793e-02 -3.82014543e-01 -6.92203820e-01 9.12660733e-03 -2.71497935e-01 -2.96473503e-01 3.74936521e-01 -9.48299348e-01 -1.16705930e+00 1.68507516e-01 -6.06996536e-01 -1.36225760e+00 -7.73235798e-01 2.97861189e-01 -1.33145082e+00 1.12976238e-01 -4.91048008e-01 -1.31704414e+00 1.50081605e-01 -1.04002917e+00 3.62745345e-01 -1.58686694e-02 -1.84767365e-01 -1.13730967e+00 4.77648258e-01 -6.73738793e-02 3.47964406e-01 1.50665537e-01 1.14834368e+00 -6.10889375e-01 -3.69903475e-01 1.55836269e-01 2.26979077e-01 1.32663444e-01 -1.45665482e-01 -2.62504518e-01 -8.49658370e-01 -5.12900710e-01 -2.21065432e-01 -7.78185189e-01 7.55752027e-01 5.57053328e-01 1.10208356e+00 -2.98062861e-01 1.05986401e-01 2.79852837e-01 1.33807600e+00 5.73733985e-01 6.33902490e-01 7.30590105e-01 2.68817961e-01 5.40176034e-01 1.00097132e+00 1.03438652e+00 4.60125715e-01 5.82551420e-01 3.97444934e-01 4.68019247e-01 2.23489091e-01 -6.29403591e-01 8.17709863e-01 2.07137287e-01 -8.40084180e-02 -1.20599531e-02 -9.37941968e-01 4.97027576e-01 -2.21183658e+00 -1.35601461e+00 4.07669723e-01 2.62960815e+00 9.91784632e-01 4.83443618e-01 9.89752352e-01 5.59352979e-04 3.72695774e-01 2.10566238e-01 -1.23361993e+00 -6.47024989e-01 3.26539904e-01 3.58918518e-01 6.91065609e-01 4.02918249e-01 -5.18597364e-01 8.58765006e-01 7.70582581e+00 7.07486928e-01 -9.00993466e-01 -9.73780975e-02 5.12142122e-01 -4.14491534e-01 -4.82538164e-01 1.86696276e-02 -4.50060010e-01 3.54605675e-01 1.41561317e+00 -6.09408557e-01 1.05262995e+00 1.02859724e+00 5.71226180e-01 -1.31113052e-01 -1.31582332e+00 6.50725842e-01 -5.80967903e-01 -1.18162119e+00 -4.63388830e-01 2.77983934e-01 7.55112708e-01 4.05166335e-02 4.32469279e-01 9.01920557e-01 1.19352055e+00 -1.17044199e+00 9.21512306e-01 3.19858015e-01 8.21125805e-01 -1.24843311e+00 3.52123916e-01 4.20303702e-01 -9.40162301e-01 -5.97599924e-01 -3.92415762e-01 -4.91093189e-01 -2.34110638e-01 -6.89121708e-02 -9.53989327e-01 8.88069645e-02 -6.80796057e-02 8.58601570e-01 -4.11052287e-01 8.08722436e-01 -2.15743735e-01 8.56845796e-01 -9.32886750e-02 -1.93914901e-02 4.02216703e-01 -1.07068047e-01 2.33146578e-01 8.92810524e-01 7.76174292e-02 -5.60536608e-02 2.30775759e-01 9.30684090e-01 2.13147730e-01 -3.83818626e-01 -7.56162643e-01 -3.69757265e-01 5.94552934e-01 4.94490921e-01 -3.87611330e-01 -7.98225477e-02 5.46662845e-02 3.35950196e-01 5.03190935e-01 5.09904146e-01 -7.93764770e-01 -7.53320083e-02 9.89908516e-01 1.69935197e-01 3.29676926e-01 -4.88106877e-01 -3.43910545e-01 -9.85949755e-01 -4.37065691e-01 -1.19505346e+00 3.30458283e-01 -5.12194335e-01 -1.00649595e+00 -7.78167322e-02 2.17502668e-01 -1.35844696e+00 -1.10202444e+00 -3.70314211e-01 -4.93130922e-01 4.20666784e-01 -1.37325656e+00 -4.68296289e-01 2.95633554e-01 6.49408519e-01 3.73155296e-01 -1.53354213e-01 6.86924458e-01 -4.08066064e-01 -6.03422880e-01 7.31273353e-01 7.77297974e-01 -2.10758001e-02 4.62258577e-01 -1.40020049e+00 3.15335423e-01 5.43325245e-01 -6.43248409e-02 5.74585319e-01 1.06220984e+00 -4.40440893e-01 -1.83065224e+00 -8.47335279e-01 -1.57714680e-01 -4.22580063e-01 8.88270557e-01 -9.45084095e-02 -6.81755781e-01 8.35725009e-01 -5.20956889e-02 -1.99025825e-01 5.17285228e-01 5.06753087e-01 -2.75608242e-01 7.44828582e-02 -1.00775003e+00 5.84417224e-01 1.08448374e+00 -6.78418159e-01 -6.73501074e-01 -9.24868062e-02 7.27786899e-01 -6.08762383e-01 -8.19095135e-01 -2.53319442e-01 6.72558725e-01 -8.63700151e-01 8.64296079e-01 -1.16387725e+00 5.56963861e-01 5.68190441e-02 -2.87401855e-01 -1.73089635e+00 -3.23855996e-01 -8.13764811e-01 -5.11491537e-01 7.78375447e-01 3.00287485e-01 -5.58786690e-01 8.21383953e-01 9.29816008e-01 -4.81993444e-02 -8.34408700e-01 -7.54252255e-01 -1.26408327e+00 5.69105983e-01 -8.46326232e-01 4.61712748e-01 7.32019067e-01 4.98235583e-01 1.38890475e-01 -4.46049660e-01 -3.81874591e-01 6.92769051e-01 4.80496176e-02 7.71642685e-01 -1.12371993e+00 -7.30851889e-01 -3.44263434e-01 -1.37953684e-01 -1.17595851e+00 4.03427392e-01 -8.10782731e-01 3.47020999e-02 -1.29187989e+00 8.44349638e-02 -5.40590048e-01 -5.56117594e-01 6.42683148e-01 2.55516291e-01 -4.34819967e-01 4.54890162e-01 6.61990121e-02 -1.01284814e+00 7.88333118e-01 1.24837267e+00 1.18322372e-01 -5.19787550e-01 5.00845052e-02 -8.74045312e-01 3.06305498e-01 9.74037826e-01 -6.21332824e-01 -9.02608573e-01 -2.00964525e-01 4.86750692e-01 5.85162640e-01 -9.95542016e-03 -6.90735042e-01 -1.60506010e-01 -8.79692078e-01 9.25717950e-02 -2.07158163e-01 3.71817201e-01 -5.36436677e-01 -1.47371829e-01 6.50701344e-01 -9.80336368e-01 1.71954647e-01 5.19844770e-01 8.16603124e-01 3.32997710e-01 -3.10454398e-01 6.23435438e-01 -4.14511785e-02 -7.92642176e-01 1.32120818e-01 -9.18783903e-01 6.55769646e-01 9.22467232e-01 -3.31185430e-01 -3.07904154e-01 -8.76446545e-01 -7.95619249e-01 3.22302490e-01 5.12052238e-01 7.55068362e-02 5.29169381e-01 -1.42411745e+00 -3.60615373e-01 -3.37047167e-02 1.13790572e-01 -3.57799292e-01 -1.22404788e-02 7.04271913e-01 1.07310295e-01 2.29559287e-01 -4.19297218e-01 -2.06663415e-01 -8.06231499e-01 4.57832515e-01 5.13140857e-01 -4.62740809e-01 -2.75640666e-01 2.77796537e-01 -2.97468305e-01 -8.85646492e-02 3.13972533e-01 -4.66256529e-01 -3.19452360e-02 1.64408758e-01 3.13111991e-01 4.85717952e-01 -3.85054201e-01 -3.37194875e-02 2.51396243e-02 1.72331948e-02 -2.71826148e-01 -5.33661902e-01 1.60997653e+00 -1.73184335e-01 6.13304436e-01 7.92468727e-01 7.97209799e-01 -5.85394323e-01 -1.79901528e+00 -1.72655523e-01 1.68151304e-01 -4.95789170e-01 1.67012140e-01 -8.15206528e-01 -6.95562780e-01 7.25479245e-01 6.65408492e-01 2.71870703e-01 7.46411324e-01 -3.44781160e-01 4.35063004e-01 8.40210676e-01 6.96539700e-01 -1.38509464e+00 5.58174312e-01 8.63507390e-01 5.28671682e-01 -9.76045549e-01 -9.22730714e-02 4.52357739e-01 -1.17077434e+00 9.75949764e-01 5.15287936e-01 -2.85977334e-01 1.89580694e-01 1.45606741e-01 -2.87873656e-01 4.87864837e-02 -1.39403951e+00 -5.15839934e-01 -2.12826058e-01 1.03934538e+00 3.48126471e-01 4.54576202e-02 7.17670694e-02 4.98514980e-01 -1.27227411e-01 2.46800959e-01 9.25217688e-01 1.07790589e+00 -6.88775837e-01 -1.23555255e+00 -2.20588043e-01 5.24956524e-01 -2.20515519e-01 1.04295015e-01 -2.22902641e-01 8.16892803e-01 -5.50581098e-01 8.86534512e-01 6.72857091e-02 -5.62233984e-01 3.37137163e-01 1.69338658e-01 8.73762012e-01 -6.37038171e-01 -6.19434655e-01 -9.56145376e-02 2.91111380e-01 -6.46364450e-01 6.98618740e-02 -9.11398292e-01 -1.62692094e+00 -6.42023981e-01 -4.21863794e-02 7.79739320e-02 3.02705497e-01 9.80395555e-01 3.21824908e-01 5.20903409e-01 8.07703614e-01 -5.29839039e-01 -1.32104397e+00 -6.70369208e-01 -8.75632405e-01 2.81870365e-01 4.55187082e-01 -9.11629379e-01 -3.86647701e-01 -2.90610313e-01]
[4.067844867706299, 1.9328854084014893]
a4d27304-421c-46e7-8f4e-b6cd0beaa69a
multi-modal-page-stream-segmentation-with
null
null
https://link.springer.com/article/10.1007/s10579-019-09476-2
https://www.inf.uni-hamburg.de/en/inst/ab/lt/publications/2019-wiedemann-lre-pss.pdf
Multi-modal Page Stream Segmentation with Convolutional Neural Networks
In recent years, (retro-)digitizing paper-based files became a major undertaking for private and public archives as well as an important task in electronic mailroom applications. As first steps, the workflow usually involves batch scanning and optical character recognition (OCR) of documents. In the case of multi-page documents, the preservation of document contexts is a major requirement. To facilitate workflows involving very large amounts of paper scans, page stream segmentation (PSS) is the task to automatically separate a stream of scanned images into coherent multi-page documents. In a digitization project together with a German federal archive, we developed a novel approach for PSS based on convolutional neural networks (CNN). As a first project, we combine visual information from scanned images with semantic information from OCR-ed texts for this task. The multi-modal combination of features in a single classification architecture allows for major improvements towards optimal document separation. Further to multimodality, our PSS approach profits from transfer-learning and sequential page modeling. We achieve accuracy up to 95% on multi-page documents on our in-house dataset and up to 93% on a publicly available dataset.
['Gerhard Heyer', 'Gregor Wiedemann']
2019-09-27
null
null
null
lang-resources-evaluation-2019-9
['page-stream-segmentation']
['natural-language-processing']
[ 6.70916975e-01 -1.95391372e-01 1.54130861e-01 -2.95114279e-01 -1.16281581e+00 -8.68752360e-01 5.69551170e-01 4.09916759e-01 -4.00897682e-01 3.82297307e-01 -6.19797818e-02 -3.81673992e-01 -2.39493340e-01 -6.03458226e-01 -6.65960968e-01 -2.46021464e-01 3.54760557e-01 8.67896736e-01 3.68163407e-01 1.07595190e-01 5.19239783e-01 1.10770488e+00 -1.40636790e+00 9.17419136e-01 7.19632030e-01 8.65691960e-01 6.16234779e-01 1.15006709e+00 -8.91650319e-01 3.25680375e-01 -6.82069123e-01 -4.61209953e-01 1.03707209e-01 -2.37383887e-01 -1.15249753e+00 6.73698127e-01 7.37189949e-01 -2.41813689e-01 -1.31252892e-02 7.99045146e-01 2.35415891e-01 -1.32422656e-01 6.04295015e-01 -8.07358027e-01 -5.24065316e-01 3.69088590e-01 -4.41428304e-01 -2.79318020e-02 1.53390259e-01 -1.13259152e-01 1.09169221e+00 -6.29075646e-01 7.20682383e-01 1.03781760e+00 6.70941293e-01 8.17949604e-03 -1.35752118e+00 -2.74146553e-02 -1.44049138e-01 6.92183152e-02 -8.87268960e-01 -3.51540208e-01 6.34411991e-01 -5.13467550e-01 1.01510859e+00 4.35195684e-01 6.23851120e-01 6.29724383e-01 5.40107302e-02 1.14081228e+00 9.53085899e-01 -8.61701846e-01 7.33097494e-02 4.96015579e-01 3.17144871e-01 4.07974243e-01 1.97954476e-01 -8.34538519e-01 -2.25555703e-01 1.06867194e-01 7.17786610e-01 -9.21562612e-02 -1.13130897e-01 -4.00089234e-01 -9.56688106e-01 5.10910988e-01 -1.22303039e-01 6.05722666e-01 -2.71742374e-01 -2.27916032e-01 2.75362641e-01 1.82035625e-01 2.90955693e-01 5.64881265e-01 -2.91130483e-01 -3.84309500e-01 -1.45212817e+00 1.59764186e-01 9.29325998e-01 8.35402846e-01 5.70990324e-01 -3.44941914e-01 -3.47559117e-02 1.10932243e+00 2.51331151e-01 1.82685435e-01 3.95503610e-01 -7.45866835e-01 7.52825797e-01 7.54539013e-01 1.07457824e-01 -7.53686845e-01 -3.22545260e-01 -3.49657685e-01 -4.10773158e-01 1.19744979e-01 5.96499443e-01 3.34955305e-01 -1.11342263e+00 8.00785303e-01 -1.82737634e-02 -7.06249833e-01 -2.47663453e-01 6.22739613e-01 3.94169748e-01 6.73546374e-01 -1.00201145e-01 -5.62673211e-02 1.55788195e+00 -8.68057907e-01 -6.66785479e-01 -1.50857404e-01 6.06973231e-01 -1.19263053e+00 1.01372838e+00 9.01822329e-01 -1.16054833e+00 -3.74094844e-01 -1.19741714e+00 -1.53807908e-01 -5.84590793e-01 4.44113165e-01 3.62002522e-01 8.05436790e-01 -1.02638900e+00 7.41416931e-01 -5.87711573e-01 -8.25241089e-01 5.74507535e-01 4.42115903e-01 -5.42751789e-01 -2.48372525e-01 -5.34904242e-01 6.58326328e-01 2.98489243e-01 1.44586831e-01 -6.79685697e-02 -3.73215884e-01 -3.38343173e-01 2.49522343e-01 3.37089062e-01 -4.29339446e-02 1.14926136e+00 -1.04535353e+00 -1.38186407e+00 9.34449315e-01 1.08033970e-01 -1.65848896e-01 8.57684314e-01 -4.60890681e-01 -4.84345287e-01 5.77925384e-01 -5.93964159e-02 4.80268538e-01 8.47389698e-01 -1.37100184e+00 -6.04701400e-01 -4.35533822e-01 -4.42316890e-01 1.14708923e-01 -6.30324423e-01 2.55608857e-01 -9.88522530e-01 -4.33355212e-01 8.60331021e-03 -7.94692695e-01 4.37274963e-01 -2.39716679e-01 -4.00131077e-01 -4.78064455e-02 1.03522730e+00 -1.19057620e+00 1.03143263e+00 -1.99600172e+00 2.68967953e-02 4.03146356e-01 -7.85055533e-02 5.89080393e-01 -1.91624478e-01 5.39868832e-01 -6.74891993e-02 7.93180764e-02 -2.31177241e-01 -4.19148386e-01 8.95067230e-02 -2.01682940e-01 -1.74493656e-01 2.21154302e-01 3.45019281e-01 7.86719799e-01 -4.56158251e-01 -7.79698670e-01 1.42164811e-01 2.90314138e-01 -1.56806931e-01 4.69220169e-02 -2.39933565e-01 -1.09862559e-01 -1.23128511e-01 7.59342551e-01 6.18955553e-01 -3.06538403e-01 4.74496692e-01 -7.90725425e-02 -1.55426681e-01 -5.33126518e-02 -1.30480576e+00 1.59101975e+00 -3.52390975e-01 1.01858687e+00 3.06989014e-01 -8.33661556e-01 8.36208761e-01 1.31783128e-01 3.89033467e-01 -7.93265879e-01 7.93746114e-02 3.21894318e-01 -3.42281491e-01 -4.82705265e-01 1.03269207e+00 1.94878638e-01 1.78134531e-01 6.47276878e-01 4.57680635e-02 -8.96822587e-02 5.47775805e-01 2.36974120e-01 9.96162295e-01 2.30030626e-01 -2.00890794e-01 -3.92756425e-02 4.14940953e-01 3.85733634e-01 1.64749194e-02 5.96253395e-01 4.34187055e-02 1.02214491e+00 6.43324494e-01 -2.03588963e-01 -1.27046418e+00 -6.78864181e-01 -8.69567022e-02 7.83824027e-01 -4.39961195e-01 -2.52517372e-01 -1.08244419e+00 -5.05608499e-01 2.03884356e-02 5.70708156e-01 -1.84860125e-01 4.74148512e-01 -7.01974988e-01 -6.51698887e-01 5.16523957e-01 5.19919515e-01 1.99421093e-01 -1.04272151e+00 -3.56738210e-01 2.54999638e-01 1.10467598e-01 -1.14321506e+00 -3.12209755e-01 5.48920095e-01 -7.45615125e-01 -1.12866330e+00 -1.04155421e+00 -8.72454107e-01 5.96800327e-01 2.60829210e-01 7.29201198e-01 -1.84826493e-01 -6.87909305e-01 5.79346895e-01 -3.34002435e-01 -2.24067703e-01 -5.39272964e-01 3.26795489e-01 -3.58543664e-01 2.11254969e-01 2.50097603e-01 -1.20222278e-01 -2.36558542e-01 4.34825150e-03 -1.39526975e+00 -6.78346157e-02 8.59980345e-01 4.36035872e-01 4.27045971e-01 1.71031937e-01 8.72686952e-02 -9.46930528e-01 8.55437994e-01 -2.31460091e-02 -7.15905011e-01 5.65348268e-01 -6.62184119e-01 -8.85397289e-03 4.69148725e-01 -2.01777026e-01 -1.18079925e+00 1.38037562e-01 1.41534984e-01 -2.52581865e-01 -4.38591957e-01 2.63429850e-01 -3.69370252e-01 1.20435111e-01 2.15435207e-01 2.76402801e-01 4.31870110e-02 -8.51947367e-01 2.48830557e-01 1.28980613e+00 5.26907742e-01 -3.55670452e-01 5.27165890e-01 3.54153007e-01 -1.63876325e-01 -1.40571141e+00 -2.74820507e-01 -6.59721851e-01 -1.12884665e+00 -2.53640383e-01 1.00092959e+00 -5.46557546e-01 -3.38743836e-01 7.28236914e-01 -1.13495135e+00 -2.78387874e-01 1.08709179e-01 2.24562973e-01 -3.30061018e-01 8.71328890e-01 -5.49860775e-01 -6.75732613e-01 -1.38253868e-01 -1.04622173e+00 1.09339380e+00 2.16874212e-01 -4.73002315e-01 -6.97868764e-01 2.35071220e-02 9.33900237e-01 1.40772268e-01 -2.09782869e-01 1.18076575e+00 -9.55102682e-01 -7.79929399e-01 -6.57731533e-01 -4.90664482e-01 4.42914128e-01 1.40334532e-01 3.28871548e-01 -1.04308283e+00 -5.14089130e-02 -3.30614150e-01 -1.22254968e-01 8.99728417e-01 1.12369530e-01 1.08676314e+00 2.20206305e-01 -2.15606526e-01 3.08491766e-01 1.29195225e+00 4.02078807e-01 8.01118135e-01 8.87158692e-01 8.23902965e-01 8.74153316e-01 5.04190147e-01 1.66312099e-01 -4.70212288e-03 5.58923066e-01 -4.39723060e-02 1.60391867e-01 -2.38741636e-01 1.03252120e-01 1.10380627e-01 5.97126782e-01 2.21853495e-01 -4.84666944e-01 -1.11005473e+00 5.09854496e-01 -1.51591790e+00 -8.60367715e-01 -4.56539840e-01 2.10817599e+00 6.18775308e-01 2.94144064e-01 9.01580378e-02 2.72268355e-01 8.74960661e-01 -7.30334967e-02 -1.67250708e-01 -6.65550411e-01 -8.10132325e-02 -1.08706225e-02 4.79830623e-01 2.05462754e-01 -1.19423234e+00 6.91355348e-01 5.96249342e+00 8.51129413e-01 -9.92911041e-01 -2.99364269e-01 4.97645408e-01 -2.29095250e-01 -1.66856632e-01 -2.45026529e-01 -9.80981469e-01 5.57590187e-01 1.05079055e+00 4.37640578e-01 3.59408021e-01 7.05700874e-01 3.86093222e-02 -3.36583704e-01 -9.13674176e-01 8.96142006e-01 2.18879566e-01 -1.37866867e+00 1.20742910e-01 3.33040476e-01 4.02063489e-01 -1.27740592e-01 -1.39780328e-01 -1.64387316e-01 -9.36845466e-02 -9.44279611e-01 9.30941701e-01 4.80934739e-01 6.46913469e-01 -8.19782913e-01 6.42282069e-01 6.01780973e-02 -8.71344388e-01 2.21775174e-02 3.38425823e-02 5.64239085e-01 9.76865888e-02 6.00750804e-01 -7.96811998e-01 3.98815155e-01 6.91724718e-01 3.29772949e-01 -7.32987046e-01 1.06987011e+00 2.38588706e-01 2.91161656e-01 -1.65448800e-01 -8.68159086e-02 3.12148124e-01 -3.91580701e-01 2.55670696e-01 1.71918738e+00 1.46171808e-01 -2.44818822e-01 -1.98173195e-01 6.90009356e-01 -1.69634923e-01 1.71870783e-01 -1.16481021e-01 -7.94568896e-01 1.01740636e-01 1.59839416e+00 -1.48223639e+00 -2.30806738e-01 -3.13169897e-01 1.14870882e+00 1.63251758e-01 1.59400553e-01 -3.85978997e-01 -9.72243369e-01 1.53495833e-01 1.23870626e-01 7.21588135e-01 -4.94053245e-01 -5.56081772e-01 -8.23012531e-01 3.87971014e-01 -9.50793922e-01 1.74133442e-02 -6.78474844e-01 -7.83906162e-01 6.27910256e-01 -3.40630740e-01 -1.13593030e+00 6.77976012e-02 -8.70262206e-01 -4.26874220e-01 8.80470932e-01 -1.37822962e+00 -1.17476583e+00 -3.02825481e-01 1.70516744e-01 7.42294312e-01 -2.81682670e-01 5.58804214e-01 3.07771027e-01 -7.45509803e-01 1.44566581e-01 6.49997294e-01 2.51826763e-01 8.30434561e-01 -1.48895323e+00 3.79810065e-01 7.89242446e-01 3.73873025e-01 4.17232096e-01 3.36680353e-01 -8.27167034e-01 -1.54644835e+00 -8.82806122e-01 9.49800670e-01 -3.84810954e-01 5.89825273e-01 -4.53841776e-01 -1.07729006e+00 4.33234006e-01 4.23336476e-01 -7.09935367e-01 8.97005737e-01 -1.35194764e-01 -8.94399956e-02 2.17727050e-02 -8.74768615e-01 3.90629947e-01 3.57793629e-01 -6.16517663e-01 -7.17057407e-01 3.25263143e-01 1.56333104e-01 -2.65290707e-01 -7.34374225e-01 -3.97931784e-01 6.01096213e-01 -8.95177901e-01 6.69505298e-01 -3.45700920e-01 6.20121300e-01 -1.80757374e-01 7.70188286e-04 -8.85206997e-01 -1.48877591e-01 -5.42529464e-01 1.31755531e-01 1.58041334e+00 3.30708623e-01 -2.85532832e-01 8.61209273e-01 8.30604672e-01 -6.72521293e-02 -1.90808341e-01 -3.77129674e-01 -7.32849121e-01 -2.20782533e-01 -5.69716513e-01 3.52588326e-01 5.75586915e-01 -1.79297298e-01 3.00240442e-02 1.60994343e-02 -4.91634309e-02 6.05307162e-01 1.90904051e-01 6.28132284e-01 -1.41545582e+00 -3.47660154e-01 -6.96310997e-01 -1.16355464e-01 -6.54743135e-01 -8.29354301e-03 -9.06358242e-01 1.23625677e-02 -1.93166637e+00 2.40667298e-01 -8.54513571e-02 -1.52369514e-01 3.06777120e-01 2.36339882e-01 2.25686312e-01 4.67042744e-01 4.87528145e-01 -5.47590971e-01 -5.45827076e-02 7.49837220e-01 -3.11436981e-01 -3.29712957e-01 -1.60997838e-01 -4.63457197e-01 4.55722898e-01 7.12609172e-01 -3.33173573e-01 -8.64222087e-03 -4.90489304e-01 2.42836058e-01 -8.30321386e-02 1.72547787e-01 -9.64573324e-01 2.49775052e-01 1.00740351e-01 8.70051980e-01 -1.10789669e+00 2.82924920e-01 -7.53748059e-01 -1.87275801e-02 1.64670378e-01 -3.67856801e-01 -1.83727816e-01 3.82157415e-01 4.73772496e-01 -2.29599789e-01 -5.56669593e-01 5.96998274e-01 -3.29168737e-01 -6.04474723e-01 -3.11801553e-01 -6.70191288e-01 -4.21387851e-01 8.49620223e-01 -6.10669255e-01 -2.41383851e-01 2.35719769e-03 -5.59109271e-01 6.90711662e-02 6.37605965e-01 4.79598731e-01 4.40600604e-01 -7.85479486e-01 -3.15135241e-01 2.40561157e-01 -8.99953097e-02 5.54922670e-02 1.16642289e-01 7.16254711e-01 -1.14134026e+00 1.05759466e+00 -3.98027480e-01 -5.25138199e-01 -1.74912691e+00 1.27760079e-02 -1.08472563e-01 -1.73986971e-01 -5.95012546e-01 5.87025523e-01 -3.93872321e-01 -9.45211947e-02 2.94492632e-01 -1.83649600e-01 -2.88583159e-01 6.79727852e-01 6.75634205e-01 5.89709818e-01 6.88094199e-01 -2.86793232e-01 -1.29247695e-01 3.60960275e-01 -4.97338980e-01 -4.05405819e-01 1.77650321e+00 -5.43680601e-02 -2.30982795e-01 3.40667307e-01 1.38627100e+00 -4.42330055e-02 -1.14386737e+00 1.07295923e-02 4.47559357e-01 -3.61631334e-01 2.30020016e-01 -7.50842452e-01 -8.00446153e-01 9.80005205e-01 4.21254545e-01 4.61160123e-01 9.44502711e-01 -3.71198468e-02 6.98690712e-01 5.45175195e-01 -6.47687390e-02 -1.70968604e+00 1.18943147e-01 3.98085475e-01 7.24540234e-01 -1.11087644e+00 1.20587796e-01 -2.24253699e-01 -6.02500737e-01 1.74277067e+00 1.37336537e-01 2.82531679e-01 7.20450729e-02 2.28968471e-01 -3.65070738e-02 -2.05597162e-01 -2.41757125e-01 -8.91535208e-02 4.36940640e-01 4.19465184e-01 4.69789565e-01 -1.75036743e-01 -8.14484153e-03 5.46996832e-01 9.91025940e-02 8.25853944e-02 7.03808010e-01 1.40721583e+00 -6.14148378e-01 -1.29327261e+00 -8.06521058e-01 6.47769690e-01 -4.90462333e-01 -6.63712174e-02 -8.10941577e-01 8.09377789e-01 -3.50019664e-01 7.18095899e-01 3.62293005e-01 -1.84140597e-02 1.54448599e-01 5.53334415e-01 5.53797781e-01 -5.55854082e-01 -6.65913224e-01 4.33096468e-01 1.64215162e-01 -3.53386641e-01 -2.19032973e-01 -1.05430722e+00 -1.18095911e+00 -3.34157348e-02 -9.64412838e-02 -8.05741251e-02 1.24719715e+00 1.02661693e+00 3.55575114e-01 6.19611442e-01 2.44239941e-01 -9.37988043e-01 -3.87596071e-01 -6.95858300e-01 -8.53534162e-01 3.03733498e-01 1.35111846e-02 -9.33669433e-02 -1.35943651e-01 4.04616207e-01]
[11.750812530517578, 2.6896584033966064]
31bf57ad-19aa-4903-96d6-71fe643559c7
video-face-clustering-with-unknown-number-of
1908.03381
null
https://arxiv.org/abs/1908.03381v2
https://arxiv.org/pdf/1908.03381v2.pdf
Video Face Clustering with Unknown Number of Clusters
Understanding videos such as TV series and movies requires analyzing who the characters are and what they are doing. We address the challenging problem of clustering face tracks based on their identity. Different from previous work in this area, we choose to operate in a realistic and difficult setting where: (i) the number of characters is not known a priori; and (ii) face tracks belonging to minor or background characters are not discarded. To this end, we propose Ball Cluster Learning (BCL), a supervised approach to carve the embedding space into balls of equal size, one for each cluster. The learned ball radius is easily translated to a stopping criterion for iterative merging algorithms. This gives BCL the ability to estimate the number of clusters as well as their assignment, achieving promising results on commonly used datasets. We also present a thorough discussion of how existing metric learning literature can be adapted for this task.
['Sanja Fidler', 'Marc T. Law', 'Makarand Tapaswi']
2019-08-09
video-face-clustering-with-unknown-number-of-1
http://openaccess.thecvf.com/content_ICCV_2019/html/Tapaswi_Video_Face_Clustering_With_Unknown_Number_of_Clusters_ICCV_2019_paper.html
http://openaccess.thecvf.com/content_ICCV_2019/papers/Tapaswi_Video_Face_Clustering_With_Unknown_Number_of_Clusters_ICCV_2019_paper.pdf
iccv-2019-10
['face-clustering']
['computer-vision']
[ 6.90754205e-02 -6.53285980e-02 -2.68651247e-01 -3.28140646e-01 -5.03036916e-01 -8.36068809e-01 5.81524193e-01 2.08750412e-01 -4.23652768e-01 3.41593415e-01 -1.22059703e-01 -7.69765377e-02 -1.24244347e-01 -5.36969364e-01 -5.45166552e-01 -7.11296618e-01 -1.84486344e-01 8.21935058e-01 1.97814584e-01 1.63019478e-01 2.17853025e-01 6.68225884e-01 -1.63628864e+00 -9.54759249e-04 4.91583139e-01 8.11958075e-01 -1.15332134e-01 6.89902425e-01 7.82606527e-02 6.90635562e-01 -3.14777851e-01 -7.44517744e-01 3.62934977e-01 -6.88437700e-01 -8.27109456e-01 4.73708838e-01 5.31078994e-01 -1.20429300e-01 -9.25364345e-02 1.11080432e+00 2.00998172e-01 1.17579386e-01 8.84958744e-01 -1.49367893e+00 -1.59282044e-01 4.51136440e-01 -8.74767005e-01 2.53641754e-01 2.90704668e-01 -5.05403697e-01 1.16798413e+00 -9.55850124e-01 6.36828244e-01 1.10986066e+00 6.19185269e-01 6.15397274e-01 -1.46934223e+00 -6.01677179e-01 2.19574660e-01 5.03594637e-01 -1.78971708e+00 -7.07196474e-01 7.12023973e-01 -7.61112332e-01 1.03899278e-01 2.74592638e-01 5.75975239e-01 6.68046236e-01 -4.64951366e-01 6.09658718e-01 6.42889321e-01 -6.42260432e-01 3.68374527e-01 4.57699388e-01 7.12639689e-02 5.77532291e-01 2.44399771e-01 -4.24619526e-01 -4.78956282e-01 -3.00434232e-01 5.16524017e-01 -2.17072800e-01 -1.08598322e-01 -9.16120648e-01 -9.97844934e-01 1.09055829e+00 -6.95002452e-02 3.50024223e-01 -3.38157751e-02 8.64976048e-02 2.17327222e-01 3.36598545e-01 4.02656972e-01 2.10945383e-01 -1.19771548e-01 -7.53222406e-02 -1.29203022e+00 2.91420817e-01 9.26029325e-01 9.62017953e-01 8.39658558e-01 -3.57979596e-01 1.27025247e-01 6.91594899e-01 2.98497051e-01 6.71521202e-02 -4.73592710e-03 -1.05590463e+00 1.54397637e-01 3.87496084e-01 3.55128080e-01 -1.25710428e+00 -1.52431816e-01 2.49160212e-02 -3.83274734e-01 1.42840564e-01 8.66358399e-01 -7.36787319e-02 -5.40710509e-01 1.82675910e+00 6.67282701e-01 6.44644082e-01 -4.17229205e-01 7.08047509e-01 3.69207799e-01 2.98426419e-01 -3.85605723e-01 -3.15565944e-01 1.44016802e+00 -6.02453887e-01 -6.15547597e-01 1.82628334e-02 5.89569867e-01 -8.85193706e-01 6.32532418e-01 5.16884983e-01 -8.81092191e-01 -3.50582272e-01 -9.21288550e-01 2.33677387e-01 -1.93634555e-01 4.03953820e-01 5.56460321e-01 1.13937378e+00 -1.21807516e+00 6.64164543e-01 -9.20955896e-01 -5.62746048e-01 5.46640813e-01 5.59054255e-01 -4.31937695e-01 -1.12894915e-01 -5.26452065e-01 6.64814830e-01 2.15961456e-01 3.89886498e-02 -6.30907238e-01 -5.62079430e-01 -8.45458925e-01 -1.85057223e-01 4.32956666e-01 -2.64436692e-01 1.10827041e+00 -1.17464697e+00 -1.27307439e+00 1.16845834e+00 -3.88371319e-01 -2.37861961e-01 5.76361299e-01 4.83911932e-02 -2.47379646e-01 3.71990293e-01 1.08305939e-01 5.11048794e-01 1.01059544e+00 -1.25080514e+00 -7.41155982e-01 -5.78630745e-01 8.04230049e-02 1.13851957e-01 -7.76659310e-01 4.13742572e-01 -8.69487822e-01 -5.62877357e-01 1.31980717e-01 -1.06713796e+00 4.99205738e-02 2.86051154e-01 -2.49062285e-01 -4.07333553e-01 8.47159922e-01 -3.39532763e-01 1.35519338e+00 -2.12920856e+00 2.81935841e-01 3.42222720e-01 4.32436317e-01 -3.06958482e-02 1.10227115e-01 3.93919617e-01 -9.67795104e-02 -9.57694836e-03 -5.33359759e-02 -6.98131323e-01 4.70746011e-02 7.99786672e-02 -1.09228425e-01 9.91501868e-01 1.12235293e-01 2.00821310e-01 -9.02657986e-01 -7.10459352e-01 1.32505387e-01 2.75306880e-01 -6.45262361e-01 6.92409426e-02 2.31862351e-01 4.20896649e-01 -3.04230660e-01 5.48481584e-01 7.39438713e-01 -1.92039967e-01 3.10157061e-01 -2.21149344e-03 1.33828908e-01 -4.73162457e-02 -1.79929149e+00 1.36024642e+00 -1.26665672e-02 7.74866104e-01 3.71375352e-01 -1.07454503e+00 6.05093896e-01 2.85868943e-01 9.73941267e-01 7.32990652e-02 2.65287876e-01 1.28348973e-02 -4.20928523e-02 -3.06514353e-01 1.89357325e-01 -3.20817590e-01 1.98833555e-01 7.25677669e-01 -6.66446611e-03 1.46272361e-01 4.63771731e-01 2.40161762e-01 9.76925015e-01 -2.05130175e-01 1.48573592e-01 -3.20120811e-01 5.52265286e-01 -1.91098630e-01 6.97669625e-01 4.16134357e-01 -3.56561661e-01 7.63916373e-01 6.07359529e-01 -2.00862497e-01 -1.03498220e+00 -1.09875333e+00 -4.08200651e-01 1.16060877e+00 2.34026819e-01 -6.61571503e-01 -1.08065522e+00 -7.28665650e-01 -7.53217796e-03 2.19610959e-01 -9.17703331e-01 -6.29418641e-02 -5.69938600e-01 -6.35406733e-01 2.53923267e-01 4.84029114e-01 -3.06118757e-01 -7.48684168e-01 -2.68957198e-01 -3.35199498e-02 -1.91924706e-01 -1.15206981e+00 -7.84227550e-01 -9.52127129e-02 -5.44895947e-01 -1.36866212e+00 -5.08458614e-01 -8.96500647e-01 9.28203642e-01 2.93829829e-01 1.07119012e+00 5.76582588e-02 -3.44311029e-01 6.26761556e-01 -4.64919955e-01 -3.72593492e-01 -2.66648710e-01 -3.63524407e-02 2.50561833e-01 6.10943675e-01 7.11842299e-01 -3.66165668e-01 -4.69338059e-01 5.55129945e-01 -6.96872413e-01 -4.51215506e-01 3.99515852e-02 4.06257719e-01 3.61068249e-01 2.58228779e-01 2.70325691e-01 -8.41580927e-01 2.42218018e-01 -6.55742586e-01 -5.69423974e-01 1.32329509e-01 -3.60555589e-01 -3.05068910e-01 4.51989174e-01 -6.29355907e-01 -3.74413311e-01 4.05024707e-01 2.35893801e-01 -5.66565633e-01 -8.57593641e-02 -5.15422300e-02 -3.02064657e-01 -2.19595402e-01 3.73367876e-01 -2.66750474e-02 8.13993290e-02 -3.58569384e-01 3.27695489e-01 6.80904746e-01 4.18915659e-01 -5.29026270e-01 8.95445049e-01 8.44465971e-01 -1.35820299e-01 -8.12344432e-01 -5.63647389e-01 -8.58020723e-01 -1.14819384e+00 -4.05127972e-01 8.81698549e-01 -7.80061007e-01 -9.52759385e-01 4.74661887e-02 -8.73278916e-01 -1.37572750e-01 -1.48455605e-01 6.20579898e-01 -7.17082739e-01 6.63855493e-01 -4.34532434e-01 -9.19141710e-01 2.41024420e-01 -1.05559003e+00 9.59532738e-01 7.16180056e-02 -4.59051073e-01 -1.07051253e+00 1.07476473e-01 3.35132897e-01 -9.90058780e-02 2.42083073e-01 6.14768147e-01 -6.07868075e-01 -2.84832627e-01 -4.18771923e-01 2.28798017e-02 1.69140145e-01 2.98751831e-01 2.07127109e-01 -7.90865242e-01 -5.16629100e-01 5.52556962e-02 -2.91385740e-01 5.34060776e-01 3.61984611e-01 1.08421719e+00 -1.21653035e-01 -6.27998114e-01 5.14383376e-01 1.14530575e+00 4.66984734e-02 2.87978023e-01 1.44914925e-01 7.44710922e-01 8.87802482e-01 5.67516387e-01 6.67203426e-01 3.55534405e-01 1.02653980e+00 3.06216151e-01 2.78196037e-02 1.34189829e-01 -7.23710358e-02 3.42146963e-01 4.86097485e-01 1.23971567e-01 -1.13792807e-01 -8.57358456e-01 6.88077629e-01 -1.85472250e+00 -1.06709218e+00 -6.08066209e-02 2.50110412e+00 8.91334891e-01 6.06080294e-02 6.59898043e-01 3.68356556e-01 1.03422916e+00 -1.42035306e-01 -4.12124276e-01 -6.91424087e-02 9.73071083e-02 -7.44141564e-02 3.87740403e-01 3.77876580e-01 -1.29233325e+00 8.56165767e-01 6.52478933e+00 8.46960664e-01 -8.59491229e-01 -2.87699699e-03 6.35529816e-01 -3.55398029e-01 4.88993078e-02 1.65582582e-01 -1.02327013e+00 5.99594593e-01 7.39185274e-01 3.19210738e-02 4.99868512e-01 7.12740123e-01 1.74316198e-01 -4.52265479e-02 -1.54169023e+00 1.14973438e+00 4.24430341e-01 -9.47486281e-01 -1.29231170e-01 2.68908560e-01 6.11856282e-01 -3.44868422e-01 1.00223541e-01 -8.01625401e-02 3.80868971e-01 -1.08618498e+00 8.38349998e-01 2.94443727e-01 6.70452178e-01 -9.33163583e-01 3.11179608e-01 2.70819962e-01 -1.28670025e+00 -1.67836323e-01 -4.46105182e-01 8.24641287e-02 4.40989994e-02 4.57681805e-01 -7.27881074e-01 2.44756892e-01 7.14915395e-01 8.83118272e-01 -5.78463674e-01 1.15260983e+00 1.04200624e-01 7.92964816e-01 -5.60685337e-01 2.12278903e-01 8.67192298e-02 -6.48191690e-01 3.69922817e-01 9.81432438e-01 2.65550107e-01 -1.82916950e-02 2.71159559e-01 6.03731275e-01 -1.04072832e-01 3.13644409e-01 -5.44775486e-01 5.23918755e-02 6.70386910e-01 1.42940104e+00 -1.24441051e+00 -7.68957585e-02 -4.60363060e-01 9.75767434e-01 3.58683288e-01 1.14490882e-01 -8.07534218e-01 -4.00212824e-01 8.96920681e-01 4.04029548e-01 7.30551302e-01 -2.05532938e-01 2.57506792e-04 -1.06248152e+00 1.13534324e-01 -7.48027444e-01 6.14199162e-01 -2.65589118e-01 -1.23521841e+00 4.68596995e-01 -5.33645712e-02 -1.27802753e+00 -1.13180675e-01 -6.26488447e-01 -5.31119466e-01 2.74330109e-01 -1.21207631e+00 -9.86583948e-01 -6.31656125e-02 6.77616954e-01 2.63972759e-01 -1.26682892e-01 5.51954627e-01 4.97918725e-01 -8.41093242e-01 7.62981951e-01 3.28239381e-01 4.62117046e-01 8.22282255e-01 -1.28113115e+00 -8.08985531e-02 7.46175706e-01 5.90831697e-01 4.93944585e-01 8.35833430e-01 -2.68308252e-01 -1.21211362e+00 -9.53038096e-01 8.40692937e-01 -9.23906326e-01 7.56695330e-01 -9.43737388e-01 -6.09298885e-01 7.07447410e-01 -1.13449544e-01 2.38116056e-01 1.06173861e+00 2.91397363e-01 -3.47710222e-01 -2.16913208e-01 -1.08615768e+00 4.89189178e-01 9.73443210e-01 -3.65254909e-01 -1.76111042e-01 5.26696384e-01 2.80814152e-02 -3.77322175e-02 -8.22490811e-01 2.90943645e-02 4.98100579e-01 -9.41672146e-01 9.21371639e-01 -6.55667543e-01 2.73227133e-02 -5.03745735e-01 -1.11419089e-01 -8.75386119e-01 -2.64031261e-01 -7.40333796e-01 -2.72128582e-01 1.46435595e+00 9.87778157e-02 -1.36391804e-01 1.12186229e+00 5.06556928e-01 2.86346346e-01 -7.26899028e-01 -9.96467829e-01 -7.71796584e-01 1.61204580e-02 -3.11783344e-01 4.44884688e-01 1.18061078e+00 9.89439785e-02 2.27800086e-01 -4.61240262e-01 1.60611585e-01 7.47753620e-01 1.05567761e-01 8.42634737e-01 -1.56983244e+00 -1.52959615e-01 -5.64176977e-01 -6.97018862e-01 -8.59206140e-01 3.73894811e-01 -7.98556387e-01 -1.39240660e-02 -9.77938473e-01 4.26242828e-01 -5.00538409e-01 -9.43160504e-02 2.27582738e-01 -1.93531767e-01 6.14444852e-01 5.45005277e-02 3.30740839e-01 -8.06684315e-01 3.51510882e-01 7.01576948e-01 9.62237418e-02 2.05099285e-02 3.86613905e-01 -7.08172798e-01 8.22124422e-01 5.84545195e-01 -6.50002241e-01 -4.27266270e-01 -1.94640249e-01 1.48771331e-01 -2.30953738e-01 1.63825601e-01 -9.16623771e-01 3.01636219e-01 -1.24733098e-01 2.63205916e-01 -4.24124092e-01 4.37495708e-01 -9.78371620e-01 -3.54902223e-02 1.80632129e-01 -2.34588981e-01 7.19122812e-02 -4.77168150e-02 7.28331387e-01 -8.45256969e-02 -5.62289238e-01 9.97717798e-01 1.24334708e-01 -3.28128934e-01 5.07955909e-01 -4.08920109e-01 8.75587538e-02 1.48713732e+00 -6.41137719e-01 3.68274897e-01 -5.20567834e-01 -8.84784162e-01 2.69913435e-01 7.51421630e-01 3.87558937e-01 2.94454724e-01 -1.48428929e+00 -7.63984263e-01 1.21759072e-01 3.28104824e-01 -3.71328712e-01 -1.59514487e-01 8.89744818e-01 -4.21388805e-01 6.52272031e-02 1.37570098e-01 -6.42909467e-01 -1.66989481e+00 7.45220721e-01 2.68951923e-01 1.79696694e-01 -3.55146050e-01 8.43401432e-01 2.98323072e-02 -2.13596150e-01 4.12565202e-01 2.69455522e-01 -4.89814162e-01 5.49576700e-01 7.21504450e-01 4.38556463e-01 1.54655986e-02 -1.02258396e+00 -4.99590278e-01 7.30505943e-01 -2.25364342e-01 -7.18957260e-02 1.37904811e+00 -2.86869556e-01 -1.59423247e-01 4.93108183e-01 1.43936348e+00 2.06654623e-01 -1.22341800e+00 -2.78860509e-01 3.17540020e-01 -5.99758506e-01 -2.31854871e-01 1.17692024e-01 -1.10720563e+00 8.13728094e-01 5.74805856e-01 2.57455260e-01 1.02305186e+00 2.84649014e-01 5.19661129e-01 6.04989007e-02 2.07227856e-01 -1.22033978e+00 2.31850505e-01 1.70506269e-01 2.97823519e-01 -1.20683610e+00 4.75172177e-02 -4.45211083e-01 -4.50358003e-01 1.07750762e+00 3.42046708e-01 -1.44913301e-01 8.65985632e-01 2.75259852e-01 -7.69458264e-02 -2.32418448e-01 -4.59311068e-01 -2.36537993e-01 2.34651372e-01 6.63086176e-01 3.31714451e-01 -1.73766464e-01 -9.81074274e-02 2.95269728e-01 -3.02193105e-01 -2.30128765e-01 6.22003555e-01 8.79330397e-01 -4.69882488e-01 -1.15862870e+00 -5.39242148e-01 4.02577311e-01 -5.09743094e-01 2.44994491e-01 -6.69851124e-01 5.79325020e-01 3.24512511e-01 1.15183985e+00 3.03958327e-01 -2.47247368e-01 7.67323002e-02 -7.00880662e-02 6.36108339e-01 -8.74286711e-01 -1.27377391e-01 9.58589613e-02 -2.49200463e-01 -3.17905545e-01 -5.36267579e-01 -1.16216218e+00 -9.40047026e-01 -4.49236274e-01 -5.69339335e-01 4.08736825e-01 4.48640645e-01 7.72847295e-01 7.33170658e-02 -1.38238087e-01 9.11105275e-01 -7.64273882e-01 -3.96022141e-01 -7.59966731e-01 -8.52391779e-01 6.73628211e-01 2.64586926e-01 -8.76734614e-01 -3.77935946e-01 4.41927791e-01]
[13.462641716003418, 1.0532432794570923]
1fd2aa26-8c5c-4b6d-bea3-fb247190d80c
multi-view-subspace-clustering-via-partition
1912.01201
null
https://arxiv.org/abs/1912.01201v1
https://arxiv.org/pdf/1912.01201v1.pdf
Multi-view Subspace Clustering via Partition Fusion
Multi-view clustering is an important approach to analyze multi-view data in an unsupervised way. Among various methods, the multi-view subspace clustering approach has gained increasing attention due to its encouraging performance. Basically, it integrates multi-view information into graphs, which are then fed into spectral clustering algorithm for final result. However, its performance may degrade due to noises existing in each individual view or inconsistency between heterogeneous features. Orthogonal to current work, we propose to fuse multi-view information in a partition space, which enhances the robustness of Multi-view clustering. Specifically, we generate multiple partitions and integrate them to find the shared partition. The proposed model unifies graph learning, generation of basic partitions, and view weight learning. These three components co-evolve towards better quality outputs. We have conducted comprehensive experiments on benchmark datasets and our empirical results verify the effectiveness and robustness of our approach.
['Zenglin Xu', 'Boyu Wang', 'Zhao Kang', 'Juncheng Lv', 'Luping Ji']
2019-12-03
null
null
null
null
['multi-view-subspace-clustering']
['computer-vision']
[-1.57365635e-01 -5.27896643e-01 -1.45767763e-01 -5.70858158e-02 -5.12539804e-01 -7.55040526e-01 3.75528932e-01 7.99397826e-02 1.44503817e-01 2.03171283e-01 4.52410221e-01 1.74532682e-01 -4.09128547e-01 -7.21879840e-01 -1.36038601e-01 -9.43844259e-01 2.30492398e-01 2.18531981e-01 3.64398628e-01 1.33844435e-01 2.96466410e-01 1.69014886e-01 -1.37958312e+00 3.36331189e-01 1.04537046e+00 4.93329942e-01 1.54781461e-01 3.51318508e-01 8.63310508e-03 5.63896537e-01 -1.23103879e-01 -2.02345952e-01 1.27893493e-01 -4.28036213e-01 -5.78272879e-01 6.33885801e-01 -3.17890197e-01 2.20500290e-01 -7.89122954e-02 1.17281795e+00 6.05324626e-01 5.61744012e-02 6.88735545e-01 -1.39036250e+00 -3.46076429e-01 4.51104224e-01 -1.06953609e+00 5.72663322e-02 2.96456158e-01 -1.44384146e-01 1.06956291e+00 -8.87486994e-01 5.64091027e-01 1.24959540e+00 3.35865855e-01 2.84164050e-03 -1.38661599e+00 -5.75590849e-01 3.37497473e-01 3.38861912e-01 -1.48359179e+00 -3.30207795e-01 1.19367588e+00 -5.59316635e-01 4.02794272e-01 3.98462676e-02 6.83031023e-01 7.37039387e-01 1.89542606e-01 7.98249245e-01 1.17570436e+00 -1.77048266e-01 1.53805539e-01 3.67375873e-02 8.53815377e-02 5.13228476e-01 3.70609403e-01 -3.04209948e-01 -2.50430793e-01 -3.48327249e-01 2.63598382e-01 3.29493105e-01 -4.16278005e-01 -1.07926536e+00 -1.18059897e+00 7.36031294e-01 2.53316373e-01 3.13999414e-01 -3.57315540e-01 -5.60258627e-01 4.37545180e-01 6.95011988e-02 4.24130321e-01 -5.64786270e-02 -5.15758358e-02 1.62256300e-01 -7.43795276e-01 -1.52542576e-01 4.96543497e-01 8.42681110e-01 7.79340208e-01 -1.70961350e-01 2.00150222e-01 1.03382528e+00 4.25131291e-01 2.38387212e-01 5.25881827e-01 -8.04613173e-01 5.82596362e-01 1.20108914e+00 -2.41078123e-01 -1.46173918e+00 -5.21835327e-01 -4.94738787e-01 -1.23844838e+00 -1.77887261e-01 -5.99499904e-02 -1.94931373e-01 -6.57689035e-01 1.55478704e+00 4.61707652e-01 2.22859725e-01 2.93066740e-01 8.10544431e-01 7.95246303e-01 4.88848805e-01 -2.08518684e-01 -4.78099674e-01 1.12609315e+00 -1.00219357e+00 -5.91502428e-01 1.87740177e-01 2.43080840e-01 -8.18838537e-01 5.10580719e-01 6.32409692e-01 -7.95694172e-01 -5.98601937e-01 -1.12790263e+00 5.95567822e-01 -1.61943957e-01 1.30846590e-01 4.09249932e-01 5.87549984e-01 -8.82807910e-01 2.47712329e-01 -7.54734159e-01 -3.56402278e-01 2.30493471e-01 3.03490430e-01 -5.25403142e-01 -2.96777427e-01 -8.53218555e-01 2.02683181e-01 6.85879171e-01 -7.21031204e-02 -4.25624132e-01 -2.01286450e-01 -6.20680332e-01 4.30839099e-02 6.67718530e-01 -8.56714845e-01 3.80804658e-01 -5.73754668e-01 -1.05075622e+00 4.60536331e-01 -2.22568363e-01 -2.64728833e-02 5.90388589e-02 2.46816024e-01 -7.38394737e-01 3.53772819e-01 2.18449175e-01 2.66710848e-01 7.58609891e-01 -1.78760684e+00 -6.19644225e-01 -7.55236924e-01 -2.43447006e-01 4.97128695e-01 -4.47408319e-01 -1.57272488e-01 -9.19402480e-01 -4.47107315e-01 8.26219082e-01 -1.04420805e+00 -3.75432879e-01 -9.61951077e-01 -5.30669510e-01 -4.97336164e-02 1.03906655e+00 -3.75675231e-01 1.60997069e+00 -2.22948527e+00 6.90533280e-01 4.93118554e-01 5.45772195e-01 -6.04960062e-02 3.02793413e-01 7.27390766e-01 -3.11723799e-01 1.41806722e-01 -1.10664390e-01 1.23729790e-02 -2.86408454e-01 -7.29900971e-02 1.61086157e-01 4.63620842e-01 -2.40394041e-01 5.62646151e-01 -8.01358700e-01 -6.98255599e-01 2.86278576e-01 1.83604196e-01 -5.29030979e-01 1.56434804e-01 2.79424578e-01 4.88729477e-01 -5.19258440e-01 5.37623167e-01 8.22416663e-01 -6.21797442e-01 5.84412873e-01 -4.99693036e-01 8.02293345e-02 -4.43731904e-01 -1.73938572e+00 1.59459305e+00 -7.54524618e-02 -1.54466569e-01 1.91935614e-01 -1.09573615e+00 8.22699666e-01 4.31884587e-01 9.39497709e-01 -1.16285749e-01 7.14005083e-02 -4.30030972e-02 1.45551458e-01 -4.17308092e-01 3.22487444e-01 -1.46041140e-01 -2.64319833e-02 5.72544992e-01 -2.24102922e-02 2.68224657e-01 2.80578136e-01 4.84371811e-01 7.07605004e-01 -5.31071499e-02 5.37503362e-01 -1.49276793e-01 1.00592446e+00 -1.79991975e-01 7.09661245e-01 2.24539582e-02 -1.23603523e-01 7.41924942e-01 5.70178092e-01 -1.65355220e-01 -7.07324386e-01 -1.01352310e+00 3.25794250e-01 6.84075236e-01 3.95377487e-01 -6.90252721e-01 -6.36751592e-01 -8.20929646e-01 -2.48271704e-01 1.58015117e-01 -4.17476535e-01 -2.32115328e-01 -1.52300581e-01 -9.40799475e-01 -4.29267660e-02 4.66441244e-01 4.26094621e-01 -5.72973609e-01 -1.09305248e-01 1.62808448e-01 -4.41307873e-01 -1.00761223e+00 -4.30897474e-01 1.02825444e-02 -1.01853180e+00 -1.41022110e+00 -5.58501601e-01 -6.74407721e-01 7.07301974e-01 1.01858902e+00 8.68601799e-01 3.50115485e-02 1.27269089e-01 5.85651994e-01 -4.69108760e-01 1.94648132e-02 -2.13554591e-01 1.60615101e-01 1.90293506e-01 5.25907040e-01 3.08661044e-01 -9.07585084e-01 -6.42560840e-01 5.36349595e-01 -1.01542854e+00 -7.82186259e-03 4.71141070e-01 7.62965560e-01 7.78281271e-01 6.39592350e-01 5.47266245e-01 -1.04848015e+00 6.98673785e-01 -6.93905294e-01 -4.09983933e-01 3.14293772e-01 -7.86548793e-01 -1.05315097e-01 7.84086525e-01 3.37629020e-02 -1.00014675e+00 2.83810914e-01 3.39463502e-01 -8.39361250e-01 -2.05712095e-01 6.16254270e-01 -6.55167997e-01 1.94265097e-01 1.20721273e-01 4.57839340e-01 7.94490054e-02 -3.20273459e-01 4.53766584e-01 6.09319746e-01 1.52309880e-01 -3.99861038e-01 8.54087353e-01 6.37089729e-01 -4.09764349e-02 -6.52262151e-01 -5.63728631e-01 -8.46713603e-01 -9.42021906e-01 -3.83471519e-01 9.65216458e-01 -1.12304342e+00 -7.11752653e-01 3.71210873e-01 -5.93799114e-01 6.68460786e-01 4.04588848e-01 5.29790938e-01 -3.56352806e-01 8.61162663e-01 -2.36013502e-01 -7.39962637e-01 -1.79756328e-01 -1.33158946e+00 7.40020156e-01 2.77526319e-01 1.31317481e-01 -1.01891601e+00 4.76302020e-02 5.09792149e-01 -8.35200846e-02 2.94259816e-01 9.40637946e-01 -7.27460325e-01 -3.06496978e-01 -2.43115555e-02 -1.16406918e-01 8.43406469e-02 4.43495512e-01 1.19074561e-01 -8.26669216e-01 -5.43665767e-01 1.13296606e-01 -3.05655096e-02 8.24413300e-01 3.64974856e-01 1.07525361e+00 1.28029495e-01 -6.53495312e-01 5.02009451e-01 1.64529741e+00 4.15035963e-01 1.82501867e-01 2.08245844e-01 1.06002080e+00 6.40770495e-01 5.38376510e-01 4.75408763e-01 6.80714726e-01 5.03054142e-01 4.27079111e-01 -5.73184006e-02 3.82017940e-01 -1.31915793e-01 1.26143560e-01 1.48110533e+00 -1.69194147e-01 -3.29209834e-01 -9.39140618e-01 5.91603816e-01 -2.02683139e+00 -1.22279382e+00 -3.56133372e-01 2.03862023e+00 9.84540433e-02 1.55222937e-01 5.56912243e-01 3.10929507e-01 1.01604307e+00 3.85731220e-01 -4.11283016e-01 9.52263698e-02 -1.40237123e-01 -4.02440429e-01 1.36649996e-01 1.11654930e-01 -1.21735346e+00 6.69361591e-01 5.40288162e+00 5.56466699e-01 -8.40840757e-01 -3.96177173e-02 6.38749897e-01 7.49179572e-02 -4.52457190e-01 1.02736741e-01 -3.65266949e-01 4.35194343e-01 4.86589134e-01 -2.54837126e-01 4.73167837e-01 6.70919359e-01 3.10939819e-01 -2.19277255e-02 -6.58095479e-01 1.04987299e+00 3.32591146e-01 -9.10303831e-01 1.88739046e-01 2.14244142e-01 9.17916417e-01 -2.10190922e-01 -7.79666454e-02 4.85454947e-02 4.36504394e-01 -5.61619580e-01 1.74428135e-01 3.84795755e-01 1.93500713e-01 -1.32609105e+00 7.11925924e-01 5.60126364e-01 -1.59490359e+00 -2.12341815e-01 -2.96831131e-01 2.69569010e-01 7.77024776e-02 5.16090631e-01 -5.33462167e-01 1.50814390e+00 5.93260586e-01 1.00482631e+00 -7.78980613e-01 9.43681061e-01 1.04693927e-01 5.16035378e-01 1.97880752e-02 4.24495518e-01 2.17259899e-01 -7.87225366e-01 5.34348130e-01 6.55565739e-01 2.08094507e-01 2.03167155e-01 6.59956694e-01 4.80933070e-01 2.18136773e-01 4.10962075e-01 -8.10404539e-01 1.76717103e-01 3.69909286e-01 1.54698479e+00 -1.24163020e+00 -3.33118677e-01 -6.85703933e-01 7.55242944e-01 2.77426988e-01 3.70188951e-01 -6.97403491e-01 -1.03311695e-01 2.79715359e-01 -1.39368743e-01 5.25142670e-01 -1.56000361e-01 -3.06653440e-01 -1.39754093e+00 -2.12464891e-02 -1.00563788e+00 8.14075828e-01 -5.21253765e-01 -1.31624675e+00 4.98371303e-01 -4.15711962e-02 -1.79284108e+00 -1.05166003e-01 -6.89014122e-02 -5.07921636e-01 5.20562351e-01 -1.03847265e+00 -1.07358038e+00 -2.84500152e-01 7.92241335e-01 4.82274950e-01 -3.18032533e-01 5.81764162e-01 2.32301205e-01 -7.61058629e-01 2.62259513e-01 3.01853210e-01 7.06613138e-02 6.52638316e-01 -1.25552726e+00 -7.36888871e-02 1.06490505e+00 1.90635815e-01 7.25650966e-01 2.17760026e-01 -6.78950071e-01 -1.52559793e+00 -1.05485153e+00 2.17587456e-01 -2.26224959e-01 4.85503286e-01 -1.78801045e-01 -8.93269122e-01 4.38388407e-01 4.12108004e-01 -2.14066535e-01 1.16580379e+00 1.19308926e-01 -2.31679693e-01 -1.99898571e-01 -8.27117860e-01 5.55582404e-01 7.67908514e-01 -2.86465555e-01 -5.10803699e-01 -5.68649657e-02 4.30667937e-01 -1.19578689e-01 -1.15974438e+00 4.07040656e-01 4.21986252e-01 -1.42241204e+00 9.53588903e-01 -4.09186721e-01 4.23289835e-01 -7.43772566e-01 -2.72401750e-01 -1.57613778e+00 -7.30026484e-01 -2.62150377e-01 9.26444754e-02 1.61517608e+00 2.01336548e-01 -6.01796627e-01 6.03318810e-01 3.55142392e-02 7.02696294e-02 -6.15673542e-01 -4.56422687e-01 -5.06919801e-01 -3.68653327e-01 -1.16275482e-01 6.54433131e-01 1.15686822e+00 4.86409329e-02 8.09766829e-01 -4.10849899e-01 5.33602059e-01 8.38593900e-01 8.41335535e-01 8.14613461e-01 -1.42230654e+00 -3.34654093e-01 -4.40654904e-01 -3.26713771e-01 -4.49555337e-01 -2.45924685e-02 -9.38547254e-01 -2.39437699e-01 -1.69758630e+00 7.48279393e-01 -1.55356362e-01 -5.12847066e-01 1.78265700e-03 -7.33534515e-01 1.01803914e-01 2.84906119e-01 4.72884655e-01 -9.30189431e-01 5.17048597e-01 1.24468136e+00 -5.11413552e-02 -4.68993545e-01 1.21560127e-01 -9.29937422e-01 7.98656285e-01 6.95127189e-01 -3.50008011e-01 -7.67181814e-01 -3.08905020e-02 -1.19933538e-01 3.81337821e-01 -8.96486267e-02 -1.17987883e+00 3.23922247e-01 -1.10476315e-01 5.61159790e-01 -8.54709029e-01 1.09896705e-01 -1.18034232e+00 4.24681157e-01 2.61462390e-01 2.55330056e-01 3.07406366e-01 -1.21809430e-01 1.07055199e+00 -5.66329777e-01 1.97843581e-01 7.28584230e-01 -3.69851828e-01 -6.17385864e-01 2.53974199e-01 -1.58054709e-01 1.26883417e-01 1.12267959e+00 -3.83192748e-01 -2.91824881e-02 -4.24318224e-01 -8.61835003e-01 5.65082371e-01 7.19905555e-01 5.07241130e-01 4.52569723e-01 -1.56053221e+00 -4.35541064e-01 1.84840873e-01 3.14139158e-01 -3.86283472e-02 4.67321783e-01 9.94124770e-01 -4.05596048e-02 2.76563913e-01 -1.12302810e-01 -9.18012321e-01 -1.49740958e+00 1.01540649e+00 -1.62129834e-01 -3.13177526e-01 -3.82263988e-01 1.81789890e-01 2.64927447e-01 -3.64726841e-01 -7.51300603e-02 2.39807770e-01 -7.40960658e-01 4.98513132e-01 1.53284088e-01 4.68273431e-01 -8.33402276e-02 -8.75509322e-01 -3.62708956e-01 8.72587919e-01 -1.05117103e-02 5.06693721e-02 1.46721649e+00 -4.98421639e-01 -2.99606293e-01 6.32164061e-01 1.08442807e+00 2.27704227e-01 -7.44970500e-01 -1.19091034e-01 -6.82812696e-03 -3.21020722e-01 -1.41374364e-01 -2.65492201e-01 -1.25524437e+00 7.53513277e-01 3.45737606e-01 4.94454414e-01 1.45013976e+00 -1.29571185e-01 5.76620460e-01 -4.26058937e-03 2.35777482e-01 -1.01152170e+00 1.89607382e-01 1.54675022e-01 3.19742173e-01 -1.16910553e+00 2.58348823e-01 -6.97366357e-01 -8.70401144e-01 9.49732065e-01 7.24086404e-01 8.26929957e-02 7.44312763e-01 -3.93443070e-02 -6.66191801e-02 -4.31442976e-01 -7.33775198e-01 -2.14479178e-01 2.25818619e-01 4.27403718e-01 2.68308043e-01 4.15098257e-02 -3.01961750e-01 6.25954330e-01 7.57181942e-02 -3.30795616e-01 3.46735716e-01 7.03755915e-01 -3.65455627e-01 -1.16290295e+00 -4.93080348e-01 4.67912138e-01 -3.50357443e-01 1.84360459e-01 -5.55042088e-01 5.72183073e-01 1.71765909e-02 1.20502925e+00 -3.45780939e-01 -7.96166599e-01 3.40439349e-01 2.33724833e-01 8.10261816e-02 -5.65111220e-01 -4.84947413e-01 7.19362080e-01 -2.34273627e-01 -4.24821317e-01 -7.97772944e-01 -7.71734834e-01 -1.01336205e+00 -7.02513158e-02 -4.54532892e-01 4.31686074e-01 1.01860672e-01 7.07522333e-01 5.92139304e-01 5.62162161e-01 1.25190151e+00 -4.44912970e-01 -1.92742720e-01 -6.73242807e-01 -6.75438166e-01 5.00893414e-01 -1.76179245e-01 -7.43530810e-01 -2.74381757e-01 1.24570698e-01]
[8.217903137207031, 4.655970096588135]
e010b1af-75f1-40d7-8faa-1560020a6b0d
integration-of-workflow-and-pipeline-for
null
null
https://aclanthology.org/L14-1708
https://aclanthology.org/L14-1708.pdf
Integration of Workflow and Pipeline for Language Service Composition
Integrating language resources and language services is a critical part of building natural language processing applications. Service workflow and processing pipeline are two approaches for sharing and combining language resources. Workflow languages focus on expressive power of the languages to describe variety of workflow patterns to meet users{'} needs. Users can combine those language services in service workflows to meet their requirements. The workflows can be accessible in distributed manner and can be invoked independently of the platforms. However, workflow languages lack of pipelined execution support to improve performance of workflows. Whereas, the processing pipeline provides a straightforward way to create a sequence of linguistic processing to analyze large amounts of text data. It focuses on using pipelined execution and parallel execution to improve throughput of pipelines. However, the resulting pipelines are standalone applications, i.e., software tools that are accessible only via local machine and that can only be run with the processing pipeline platforms. In this paper we propose an integration framework of the two approaches so that each offests the disadvantages of the other. We then present a case study wherein two representative frameworks, the Language Grid and UIMA, are integrated.
['Trang Mai Xuan', 'Donghui Lin', 'Yohei Murakami', 'Toru Ishida']
2014-05-01
null
null
null
lrec-2014-5
['service-composition']
['miscellaneous']
[-2.33412459e-01 -9.92582738e-02 2.26164266e-01 -7.09883034e-01 -4.54200119e-01 -1.04721403e+00 7.60705411e-01 1.92833424e-01 -3.83341938e-01 1.70149565e-01 2.06915691e-01 -3.69870454e-01 -1.10844292e-01 -8.32143188e-01 9.92725790e-02 -4.15284008e-01 1.32810131e-01 5.96603572e-01 6.23472929e-01 -2.45969221e-01 -1.93176605e-02 8.12474012e-01 -1.83681953e+00 7.30973780e-01 8.87844861e-01 6.26631796e-01 6.23336852e-01 5.49136579e-01 -1.42791295e+00 6.63049221e-01 -3.03268969e-01 8.13262463e-02 2.49046355e-01 -1.06623515e-01 -9.22829390e-01 -5.75704314e-02 -4.86621439e-01 3.76107022e-02 4.80633557e-01 1.05390835e+00 3.71267617e-01 2.25271303e-02 6.13340884e-02 -1.66630030e+00 -1.22082524e-01 6.79158926e-01 -2.62348264e-01 -1.12170003e-01 7.82526433e-01 2.75452077e-01 5.71546555e-01 -8.04849088e-01 7.24574268e-01 1.27195668e+00 3.45390141e-01 4.93368864e-01 -7.66148925e-01 -3.82310450e-01 1.38126150e-01 -8.21617320e-02 -1.24585414e+00 -7.79787302e-01 2.93694109e-01 -5.68835199e-01 1.33774531e+00 6.81634903e-01 3.46982718e-01 3.10631871e-01 -1.47429360e-02 4.55324829e-01 1.00461113e+00 -6.61899686e-01 2.29562312e-01 3.27369988e-01 6.54509127e-01 5.53437352e-01 4.10131603e-01 -7.68151343e-01 -5.48918724e-01 -1.99153289e-01 1.94997057e-01 2.91413486e-01 -3.65131646e-02 1.09440327e-01 -9.57598627e-01 1.49538696e-01 -2.26409853e-01 8.03544283e-01 -4.98663276e-01 -4.12583739e-01 7.43153393e-01 2.15850934e-01 1.82145238e-01 1.94662735e-02 -6.31403923e-01 -2.07005203e-01 -9.03468907e-01 3.72036874e-01 1.27367735e+00 1.24020839e+00 9.24047232e-01 -3.18129987e-01 -2.76856333e-01 4.82277423e-01 7.13048518e-01 2.98636258e-01 5.59954405e-01 -6.07465804e-01 3.15535247e-01 1.35621560e+00 4.25469041e-01 -4.44019020e-01 -6.86828971e-01 4.05683219e-01 -3.55788589e-01 3.38467926e-01 5.21851003e-01 -4.64163395e-03 -7.81453311e-01 9.48559165e-01 6.77350223e-01 -4.42373872e-01 1.82726368e-01 8.49809229e-01 1.12099123e+00 6.24975383e-01 5.36219597e-01 -2.63011634e-01 2.00509381e+00 -1.01467490e+00 -1.11315143e+00 1.25241959e-02 7.84369707e-01 -1.03549838e+00 1.55608201e+00 1.70815706e-01 -1.35187542e+00 -2.63481379e-01 -5.34036875e-01 -2.88633734e-01 -8.36601555e-01 -1.79655328e-01 7.35085428e-01 5.09269357e-01 -1.17246664e+00 3.50257099e-01 -1.03606999e+00 -7.72044480e-01 -7.33361319e-02 2.86465436e-02 -2.33934239e-01 6.07855767e-02 -9.11856294e-01 7.38313913e-01 5.35827518e-01 5.74541315e-02 -4.52404499e-01 -5.99749982e-01 -6.81367576e-01 4.02168751e-01 3.41869771e-01 -6.08562768e-01 1.51709080e+00 -1.07698309e+00 -1.55674779e+00 8.93923223e-01 -4.13767844e-01 1.64576173e-01 6.34355307e-01 5.63662127e-02 -7.29624450e-01 -1.95935115e-01 4.19682562e-01 -1.69320464e-01 -2.89675035e-02 -8.33132565e-01 -1.26473737e+00 -4.44918245e-01 1.13889640e-02 1.18361995e-01 -2.30062932e-01 1.19957387e+00 -8.08052540e-01 2.24463150e-01 -6.92937225e-02 -4.06792223e-01 -4.57772613e-01 -2.63438791e-01 -4.88058217e-02 -5.00477314e-01 7.71138012e-01 -5.99730670e-01 1.38761687e+00 -2.03006363e+00 -4.39985931e-01 2.36440763e-01 -2.95561645e-02 2.33684361e-01 1.35865211e-01 1.04964030e+00 2.96878189e-01 5.63880444e-01 -1.10293590e-01 -1.58938646e-01 4.18657780e-01 4.59563762e-01 1.17157400e-01 -4.30513769e-02 1.80765539e-02 3.98531109e-01 -8.59259605e-01 -6.83385134e-01 -1.66434243e-01 3.08514893e-01 -3.89679819e-02 4.90704656e-01 -4.52209204e-01 4.33543861e-01 -7.65837371e-01 7.50983119e-01 9.31395173e-01 -1.11461490e-01 5.44965863e-01 -9.42846388e-03 -1.01579523e+00 4.83598858e-01 -1.65378070e+00 2.04104090e+00 -6.37561321e-01 -1.25297770e-01 6.77067935e-01 -5.19943058e-01 8.22097719e-01 9.10369635e-01 3.08459550e-01 -2.00203791e-01 -1.98685586e-01 3.95683795e-01 -1.99524853e-02 -1.17536044e+00 3.96528363e-01 2.40307003e-01 -2.20893264e-01 9.20818746e-01 -1.03548139e-01 1.32061765e-01 9.31422830e-01 -2.58459273e-04 1.03571522e+00 8.10814619e-01 5.55296123e-01 -4.27830070e-01 8.51855814e-01 3.40487480e-01 9.02324677e-01 3.99311841e-01 9.92091075e-02 -9.04652104e-02 4.84837353e-01 -8.04200947e-01 -1.04452944e+00 -7.55752146e-01 1.33644372e-01 1.75914741e+00 -2.14005411e-01 -8.83605421e-01 -5.56594849e-01 -3.66880894e-01 -4.59103435e-01 5.86867392e-01 2.48899996e-01 8.54622483e-01 -4.62948740e-01 -5.15042841e-01 6.47647500e-01 3.02795678e-01 4.74275619e-01 -1.28750372e+00 -1.18912148e+00 2.69846231e-01 -1.18812472e-01 -1.22064555e+00 -1.74409106e-01 -1.57806441e-01 -3.73521686e-01 -9.66336370e-01 1.46696970e-01 -5.73023617e-01 5.89848280e-01 3.17194968e-01 1.12624562e+00 3.22486699e-01 -1.56433001e-01 1.09783038e-01 -5.86054683e-01 -7.75127113e-01 -5.38715720e-01 3.45088243e-02 -9.86461863e-02 -4.36686315e-02 6.11858547e-01 -6.27335489e-01 -2.67923146e-01 1.61012203e-01 -1.30477870e+00 6.39858305e-01 1.61998793e-01 1.31903067e-01 3.85346383e-01 1.10805318e-01 3.62942547e-01 -1.00167620e+00 9.41655874e-01 -4.97046739e-01 -8.54123592e-01 5.93809009e-01 -5.42308569e-01 4.70771119e-02 9.12426889e-01 1.10143408e-01 -1.62759531e+00 2.45346114e-01 -1.11827895e-01 7.49085173e-02 -7.23602414e-01 8.49389374e-01 -7.03644335e-01 4.59772974e-01 1.79333746e-01 4.73246984e-02 -2.05563188e-01 -1.01067793e+00 3.81417274e-01 9.84547853e-01 1.77575275e-01 -7.95973420e-01 4.97979134e-01 5.50912976e-01 -2.67061919e-01 -4.60084677e-01 -7.57121220e-02 -7.18792319e-01 -6.64029717e-01 -1.95785254e-01 7.42166162e-01 -5.22587419e-01 -8.29773009e-01 -2.75977254e-02 -1.41254485e+00 -1.44518152e-01 -1.73096463e-01 1.16789229e-01 -1.74298763e-01 2.60964602e-01 -4.57586169e-01 -1.10135460e+00 -7.90536940e-01 -1.14008260e+00 9.38999951e-01 4.25567061e-01 -2.18014657e-01 -7.18302965e-01 -1.53617471e-01 -4.26850803e-02 6.66547835e-01 2.60134637e-01 5.51727295e-01 -1.06673324e+00 -3.82087618e-01 -2.98037589e-01 -2.09961340e-01 -1.87152714e-01 2.46244907e-01 5.66161752e-01 -1.07808685e+00 4.95215766e-02 -1.23900652e-01 3.07002574e-01 -8.13935474e-02 -3.62388372e-01 1.02577019e+00 -4.83249158e-01 -5.06130576e-01 3.89311641e-01 1.53762293e+00 3.05050224e-01 5.44579506e-01 4.58734959e-01 4.62115616e-01 9.80622292e-01 4.99644369e-01 4.87704068e-01 5.37219048e-01 4.68455970e-01 -7.46432543e-02 -9.99957398e-02 2.61107087e-01 -5.14078438e-02 4.68527079e-01 9.70058203e-01 -4.34949249e-01 1.22593850e-01 -1.59601152e+00 3.47885251e-01 -2.32115722e+00 -9.68667090e-01 -7.63789237e-01 1.91465139e+00 7.40742862e-01 -1.58913121e-01 1.59030572e-01 -1.47429574e-02 5.28309762e-01 -2.91635513e-01 -1.11391507e-01 -7.65639782e-01 3.17892611e-01 5.47824875e-02 2.97595352e-01 4.32382643e-01 -5.98599792e-01 9.96105075e-01 5.66315746e+00 3.54752183e-01 -9.41928446e-01 5.83810151e-01 -7.95669034e-02 7.24196285e-02 -4.25424755e-01 4.47075695e-01 -1.05338979e+00 4.68405455e-01 1.41266167e+00 -4.77426648e-01 4.28839475e-01 7.36087739e-01 9.70821977e-01 -8.31191912e-02 -1.06751013e+00 5.31273782e-01 -4.67589855e-01 -1.35821128e+00 -8.54057968e-02 -2.80167431e-01 1.95263639e-01 2.87090629e-01 -9.96759832e-01 1.21313237e-01 7.03613818e-01 -7.45601892e-01 7.85509109e-01 8.33614409e-01 5.42713583e-01 -4.28674698e-01 7.14773178e-01 7.59574115e-01 -1.47634697e+00 7.18926489e-02 1.03045097e-02 -3.59187275e-01 5.54251850e-01 6.00248635e-01 -7.64529169e-01 1.04508281e+00 1.35559225e+00 -1.22905634e-01 -2.52598226e-01 8.37960482e-01 -5.02596349e-02 1.67247817e-01 -4.58165854e-01 -9.89050232e-03 4.22269180e-02 -5.22908509e-01 2.34255701e-01 1.86758745e+00 3.23853761e-01 -9.18931738e-02 8.08436215e-01 6.69115901e-01 5.05995929e-01 7.44352102e-01 -6.60361648e-01 5.74587146e-03 5.74523449e-01 1.96814978e+00 -9.45297778e-01 -6.37198091e-01 -9.18035388e-01 5.17974138e-01 -2.74961721e-02 3.23367774e-01 -4.99377519e-01 -6.23184681e-01 6.07510209e-01 3.44050437e-01 -2.30144277e-01 -3.25495213e-01 -2.97856241e-01 -1.18392932e+00 1.63975447e-01 -1.03440619e+00 5.68304360e-01 -8.95542681e-01 -1.24919999e+00 8.43785286e-01 1.98807031e-01 -8.44456375e-01 -1.53621852e-01 -4.49098021e-01 -7.09430933e-01 1.40301359e+00 -1.50638926e+00 -1.81003833e+00 -6.62846208e-01 6.92728579e-01 4.85009670e-01 -3.01377438e-02 1.23675442e+00 4.91367668e-01 -5.63094378e-01 -1.73785985e-01 -2.58351833e-01 -7.62733147e-02 5.70747316e-01 -1.03669751e+00 2.69295990e-01 1.27026176e+00 -2.78105259e-01 8.98522913e-01 5.72881818e-01 -5.81053913e-01 -1.59004855e+00 -8.53163660e-01 1.38367760e+00 -1.09626211e-01 7.66194224e-01 -6.44515276e-01 -1.05251598e+00 7.42721081e-01 5.57564080e-01 -6.04646157e-05 1.11705041e+00 -1.94687471e-02 -1.45667017e-01 -4.13803756e-01 -1.32776988e+00 6.79921865e-01 9.38151240e-01 -3.04548383e-01 -4.45779592e-01 4.90249932e-01 5.33880889e-01 -1.47052020e-01 -8.96562815e-01 -5.14076389e-02 4.31008816e-01 -8.53368521e-01 2.85303384e-01 -8.06821704e-01 -9.83413011e-02 -1.07724321e+00 -6.36399612e-02 -7.38794446e-01 -1.07578024e-01 -9.91595089e-01 4.21522051e-01 1.83699048e+00 4.29833233e-01 -8.86436164e-01 -7.54372701e-02 1.30744588e+00 -2.77798176e-01 3.69577110e-02 -5.10643303e-01 -5.21608829e-01 -6.00315809e-01 -6.84012771e-01 1.44095182e+00 9.51848924e-01 5.13117075e-01 1.75685436e-01 2.99422354e-01 4.10637259e-01 9.67529863e-02 2.18165413e-01 8.79067838e-01 -1.31576502e+00 -2.20918819e-01 -3.19537461e-01 1.42828345e-01 -3.14838111e-01 -7.43358359e-02 -1.22905505e+00 9.73787159e-03 -2.12871218e+00 -3.45217250e-02 -6.24085009e-01 5.53605519e-02 1.02864254e+00 2.95405149e-01 -3.86487067e-01 2.85606742e-01 5.97509265e-01 -4.56463516e-01 -4.86271083e-01 8.17237020e-01 2.18155846e-01 -5.25036991e-01 -5.29482476e-02 -6.62434936e-01 1.00702608e+00 9.04218197e-01 -4.69495744e-01 -4.31202382e-01 -7.45308220e-01 4.96887594e-01 -1.51478991e-01 -8.91431198e-02 -6.57240748e-01 7.81213641e-01 -7.25863099e-01 -1.74336955e-01 -3.06057572e-01 -3.30382049e-01 -1.14059031e+00 6.81761801e-01 1.57643899e-01 -1.76895514e-01 6.22529566e-01 5.61603233e-02 -5.27253784e-02 -4.03320789e-02 -5.74968636e-01 3.23779047e-01 -6.97908640e-01 -7.95544565e-01 1.81978077e-01 -6.75176322e-01 -3.58057290e-01 1.10477221e+00 2.32441910e-02 -2.29130402e-01 2.78706044e-01 -6.23887062e-01 4.69834566e-01 5.46628594e-01 3.58437181e-01 -2.65503135e-02 -9.11040187e-01 -6.52579188e-01 3.04401577e-01 9.77546498e-02 3.17062587e-01 -1.47308916e-01 7.28764832e-01 -9.42185938e-01 1.70552969e-01 -3.70176613e-01 -3.78766328e-01 -1.25699461e+00 6.57291710e-01 2.67233491e-01 -5.07516503e-01 -6.97388887e-01 1.03600018e-01 -1.28574282e-01 -6.65095687e-01 -7.30150118e-02 -5.06341100e-01 -4.57575709e-01 2.46549156e-02 1.16365266e+00 1.54335827e-01 3.26550990e-01 -4.86877054e-01 -7.47078180e-01 1.20098799e-01 3.68598431e-01 -3.67844254e-01 1.34807432e+00 -2.91654110e-01 -9.38724160e-01 7.27848887e-01 3.13373417e-01 2.41779223e-01 -6.79603696e-01 3.47492332e-03 7.67719805e-01 -4.13547486e-01 -3.48971426e-01 -8.14419270e-01 -6.87704444e-01 5.22923887e-01 3.06324899e-01 7.43822455e-01 1.25941622e+00 -1.84474796e-01 2.37718105e-01 8.43477622e-02 3.13393742e-01 -1.19031489e+00 -1.00670493e+00 3.98878664e-01 8.09040189e-01 -7.60173619e-01 -1.50784805e-01 -6.26274943e-01 -6.00869060e-01 1.46962070e+00 4.19904888e-01 3.75662804e-01 7.39281654e-01 8.88778627e-01 3.71249527e-01 -2.75590360e-01 -9.30240273e-01 -5.08620203e-01 -2.01875046e-01 6.42431259e-01 9.67459202e-01 1.51221052e-01 -1.10834301e+00 1.18943822e+00 2.41220534e-01 4.96070534e-01 3.20169210e-01 1.30304670e+00 -4.04007316e-01 -1.62680972e+00 -5.75076640e-01 3.65344398e-02 -7.30727375e-01 -2.17425637e-02 1.56326313e-02 5.95967948e-01 4.12532121e-01 1.10714555e+00 1.88875929e-01 1.99781820e-01 6.22201562e-01 6.48686111e-01 -1.59111962e-01 -7.36199498e-01 -1.05826283e+00 1.62488922e-01 5.38876772e-01 -5.34484982e-01 -5.29272079e-01 -5.35555959e-01 -1.71182978e+00 -2.60057181e-01 1.51758820e-01 3.98814857e-01 9.92157400e-01 1.02381885e+00 4.88074899e-01 4.47778106e-01 2.25614384e-01 -7.63027191e-01 -2.16535479e-01 -8.71336997e-01 -4.62739915e-01 4.00215715e-01 -3.97693872e-01 5.37522696e-02 2.24546105e-01 5.94960451e-01]
[9.079822540283203, 7.8000640869140625]
ee58f7bd-59ed-46da-abd6-b06771d80c57
a-min-max-cult-algorithm-for-graph
null
null
https://ieeexplore.ieee.org/document/989507
https://ieeexplore.ieee.org/document/989507
A Min-max Cult Algorithm for Graph Partitioning and Data Clustering
An important application of graph partitioning is data clustering using a graph model - the pairwise similarities between all data objects form a weighted graph adjacency matrix that contains all necessary information for clustering. In this paper, we propose a new algorithm for graph partitioning with an objective function that follows the min-max clustering principle. The relaxed version of the optimization of the min-max cut objective function leads to the Fiedler vector in spectral graph partitioning. Theoretical analyses of min-max cut indicate that it leads to balanced partitions, and lower bounds are derived. The min-max cut algorithm is tested on newsgroup data sets and is found to out-perform other current popular partitioning/clustering methods. The linkage-based refinements to the algorithm further improve the quality of clustering substantially. We also demonstrate that a linearized search order based on linkage differential is better than that based on the Fiedler vector, providing another effective partitioning method.
['Horst D. Simon', 'Ming Gu', 'Hongyuan Zhab', 'Xiaofeng He', 'Chris H.Q. Ding']
2002-08-07
null
null
null
proceedings-2001-ieee-international
['graph-partitioning']
['graphs']
[ 1.37864128e-01 2.79274970e-01 -5.26125193e-01 -4.26543415e-01 -3.39104950e-01 -7.52093494e-01 7.83955380e-02 4.92328078e-01 -1.29941404e-01 5.64762473e-01 -4.34471592e-02 -3.24611664e-01 -9.36588466e-01 -8.84025097e-01 -8.57959241e-02 -7.78700948e-01 -5.75620592e-01 9.67333674e-01 3.41747195e-01 -2.29957420e-02 3.70962143e-01 5.88701367e-01 -1.22353816e+00 8.38920549e-02 1.02719963e+00 3.48682970e-01 1.34478405e-01 6.96082056e-01 -1.24574192e-01 2.25424454e-01 -5.36403179e-01 -8.03070515e-02 5.23372173e-01 -7.01417506e-01 -1.16177857e+00 4.12243515e-01 6.71134219e-02 3.87043089e-01 -8.89024809e-02 1.09022629e+00 3.13190132e-01 2.61846542e-01 9.01499212e-01 -1.69504976e+00 -2.81443387e-01 9.36450541e-01 -1.05998504e+00 -5.23167066e-02 3.75002414e-01 -6.67000830e-01 1.25122082e+00 -3.42171997e-01 7.16847599e-01 1.19710624e+00 8.15652490e-01 -6.52601616e-03 -1.83331680e+00 -3.82935405e-01 -5.97729720e-02 -1.56813376e-02 -1.94604492e+00 1.07011714e-04 7.69365370e-01 -5.08185267e-01 7.01884210e-01 7.54428029e-01 7.87931502e-01 -2.58334249e-01 4.81500998e-02 2.51544952e-01 7.68797815e-01 -4.49960053e-01 1.88943282e-01 -7.26128221e-02 6.65842175e-01 7.61645198e-01 8.82289112e-01 -2.95617253e-01 -2.25612074e-02 -5.71982443e-01 4.61299390e-01 -4.42869276e-01 -3.01950216e-01 -1.06983376e+00 -1.03794491e+00 9.50234890e-01 4.80682939e-01 4.22550946e-01 1.29250744e-02 -7.40955584e-03 9.94222611e-02 3.13692838e-01 3.61950099e-01 5.84348083e-01 -1.28393427e-01 3.93090159e-01 -1.30897820e+00 3.41908038e-02 1.03712475e+00 9.76669669e-01 1.09292328e+00 -3.32473189e-01 3.33354175e-01 8.95174921e-01 4.64462847e-01 2.59771585e-01 -1.72689542e-01 -1.05194986e+00 2.64578581e-01 1.04772592e+00 -2.79396325e-01 -1.51333034e+00 -8.56567383e-01 -2.04315245e-01 -8.91181946e-01 6.00201376e-02 3.41721743e-01 3.23291346e-02 -6.92409635e-01 1.65392983e+00 5.97926736e-01 -2.97282308e-01 -1.50332510e-01 5.54140270e-01 5.82422256e-01 5.01962185e-01 -3.10327142e-01 -7.90658176e-01 8.93022180e-01 -5.32877207e-01 -8.01503718e-01 3.30797732e-01 9.39329505e-01 -6.99282348e-01 3.28720003e-01 3.53560418e-01 -9.48027730e-01 -4.27655995e-01 -1.04852033e+00 3.42924505e-01 -1.98013306e-01 -2.06106082e-01 5.56024492e-01 9.57658172e-01 -1.44706035e+00 5.74342489e-01 -7.40853488e-01 -7.31909215e-01 -1.61055371e-01 8.46115291e-01 -6.25168622e-01 4.82145846e-02 -5.31028092e-01 4.12838370e-01 9.52850699e-01 -1.11761220e-01 1.23746209e-01 -4.08946335e-01 -6.89341307e-01 1.05577871e-01 2.75769442e-01 -3.52895498e-01 3.89085680e-01 -6.83989882e-01 -7.83176899e-01 9.64613736e-01 -6.30618185e-02 -1.52112842e-01 2.18725160e-01 5.49286425e-01 -2.92339206e-01 4.94188279e-01 3.18149477e-01 7.45563149e-01 1.65663689e-01 -1.68013179e+00 -3.56400669e-01 -3.79769027e-01 -2.91155934e-01 2.87044704e-01 -3.24922740e-01 -1.88571677e-01 -5.69590032e-01 -4.65404838e-01 6.64181948e-01 -1.06456161e+00 -3.79339784e-01 -6.79812312e-01 -8.19059551e-01 -3.38163763e-01 7.60039210e-01 -3.00293416e-01 1.86490202e+00 -1.99284482e+00 3.09054673e-01 1.23003554e+00 5.84025562e-01 -1.63675025e-01 2.86037195e-02 9.36815262e-01 -5.25315166e-01 3.36888134e-01 -5.76091826e-01 3.38379771e-01 -1.75492540e-01 1.84743091e-01 2.78663069e-01 8.68518174e-01 -3.38914901e-01 3.61878067e-01 -7.75904179e-01 -8.64824712e-01 9.42537859e-02 -1.70981541e-01 -7.10766196e-01 -1.18661925e-01 2.65295267e-01 -2.08907664e-01 3.44152562e-03 2.38204330e-01 9.41049516e-01 -2.48897970e-01 9.52157915e-01 -1.32991120e-01 -1.27240434e-01 -1.77106440e-01 -1.72099531e+00 1.42125010e+00 6.28502250e-01 5.88762939e-01 3.20138395e-01 -1.24110627e+00 1.08149600e+00 4.46130298e-02 1.19253278e+00 2.13411450e-01 7.12866187e-02 -6.82576299e-02 4.44081515e-01 -2.25221112e-01 3.79298925e-01 1.16205104e-01 -1.07895091e-01 6.98203862e-01 -9.15595889e-02 -2.62968183e-01 9.04043198e-01 8.77673090e-01 1.17124403e+00 -3.67666572e-01 4.67300177e-01 -1.25134587e+00 4.08779770e-01 4.30787534e-01 6.26005769e-01 5.99084377e-01 -1.26239527e-02 5.80959141e-01 8.86191487e-01 -4.98007946e-02 -8.71859014e-01 -1.03324616e+00 -2.28612810e-01 7.72195220e-01 4.10307437e-01 -8.92474651e-01 -1.14416170e+00 -4.90172684e-01 1.31024033e-01 2.37846687e-01 -6.03116751e-01 -9.65590924e-02 -3.13947648e-01 -1.20224738e+00 3.02230328e-01 1.79487109e-01 7.28393123e-02 -4.68908817e-01 -1.38670713e-01 1.57633618e-01 -2.09349111e-01 -4.93330419e-01 -5.86886585e-01 2.90138423e-01 -9.63423431e-01 -1.61608899e+00 -1.80397928e-01 -9.99114931e-01 1.05009961e+00 6.67708874e-01 8.81848514e-01 5.02567768e-01 -3.31047982e-01 3.38667572e-01 -4.82443243e-01 1.73069462e-01 -2.82496065e-01 2.70318747e-01 1.33986607e-01 -8.12743455e-02 4.02589887e-01 -4.57727671e-01 -2.72969455e-01 6.50212884e-01 -8.21809530e-01 -1.68260664e-01 1.41792029e-01 7.03227401e-01 5.87378621e-01 7.24802494e-01 4.49984312e-01 -1.11532509e+00 7.46712565e-01 -3.96561086e-01 -5.50263345e-01 3.46806318e-01 -8.93319011e-01 9.23027396e-02 2.01586396e-01 -1.69372305e-01 -7.15894461e-01 4.14120376e-01 2.64746964e-01 -4.52113673e-02 2.26903334e-01 5.32371342e-01 -3.44783336e-01 -2.83737838e-01 6.05033278e-01 -3.75082672e-01 -6.06389046e-02 -2.62860507e-01 5.20846426e-01 6.29852891e-01 4.18137252e-01 -5.46378851e-01 7.04923391e-01 5.66443086e-01 2.64948189e-01 -1.14479315e+00 -1.32845849e-01 -1.02105045e+00 -1.04365313e+00 -2.93953508e-01 8.55087638e-01 -3.22852641e-01 -9.21500146e-01 -4.55745086e-02 -6.60229087e-01 -1.42253831e-01 -8.38608295e-02 6.04240716e-01 -6.47619665e-01 8.00722301e-01 -4.60913569e-01 -7.19610155e-01 1.36814013e-01 -9.00442719e-01 4.91497338e-01 -1.27585381e-01 -5.33489704e-01 -1.05988014e+00 5.25275767e-01 2.37381071e-01 -4.17257309e-01 4.17475045e-01 1.13288951e+00 -8.02244246e-01 -1.79287151e-01 -5.19463560e-03 -2.74043739e-01 -1.51098520e-01 1.91342071e-01 5.47002852e-01 -2.94431925e-01 -6.23930752e-01 -2.42156565e-01 2.65950143e-01 8.68426621e-01 7.10880876e-01 7.57214725e-01 -1.05285220e-01 -8.73049200e-01 5.93678534e-01 1.47281718e+00 5.87175727e-01 3.33651751e-01 -8.41962695e-02 8.72452617e-01 1.00714266e+00 5.30196428e-01 2.35058665e-01 3.07602584e-01 3.89059931e-01 6.83899000e-02 -2.63177216e-01 6.15329109e-02 8.88644829e-02 -2.51343995e-01 1.20780361e+00 8.24053492e-03 -4.19151306e-01 -1.22404158e+00 6.20348096e-01 -2.10714054e+00 -1.08114481e+00 -9.40562785e-01 2.17945933e+00 5.65579116e-01 1.78624759e-03 5.30709922e-01 6.06946051e-01 1.18070424e+00 -1.13644764e-01 -1.36555314e-01 -4.32307333e-01 -1.01216234e-01 -8.35843533e-02 6.66811883e-01 1.01175809e+00 -1.06531024e+00 7.45260298e-01 7.76249695e+00 7.40196586e-01 -2.54254550e-01 -3.77805710e-01 3.76924545e-01 1.53867260e-01 -3.84853482e-01 1.99233681e-01 -4.40257370e-01 1.95760459e-01 6.09702885e-01 -4.83310461e-01 3.72599810e-01 5.45646310e-01 1.55035332e-01 -4.31136578e-01 -9.92962360e-01 8.06871831e-01 -2.43014228e-02 -9.51951921e-01 -1.23117626e-01 5.96842706e-01 9.71998870e-01 -3.95739943e-01 -4.00269270e-01 -4.37694788e-01 8.26859355e-01 -7.78891385e-01 1.16017848e-01 2.51031183e-02 7.25602865e-01 -1.21814549e+00 5.11820912e-01 2.04494283e-01 -1.65589809e+00 -1.07744876e-02 -4.83078390e-01 9.17065218e-02 1.24825828e-01 8.43793690e-01 -9.55677509e-01 8.72524500e-01 4.90762085e-01 5.54275572e-01 -4.20980841e-01 1.23165524e+00 2.23692253e-01 3.50458562e-01 -5.96308827e-01 1.64177135e-01 5.13107404e-02 -8.65836203e-01 5.70539415e-01 1.35213792e+00 4.77674007e-02 4.60848927e-01 5.25098860e-01 5.22369027e-01 9.46243405e-02 3.61165583e-01 -7.48125494e-01 -7.37635866e-02 6.31014049e-01 1.15747082e+00 -1.72199464e+00 -2.44104266e-01 1.12577146e-02 5.66438079e-01 3.43742311e-01 3.33501041e-01 -4.11623120e-01 -7.31651783e-01 5.85382462e-01 1.75077528e-01 1.29308239e-01 -3.82401794e-01 -3.99718523e-01 -7.52507508e-01 -3.87410790e-01 -6.62168324e-01 8.59242499e-01 -3.95499200e-01 -1.11774862e+00 4.75011736e-01 4.65010554e-01 -7.74047375e-01 -1.90768838e-01 -3.57596308e-01 -4.38007027e-01 4.51224148e-01 -3.82947892e-01 -5.66039026e-01 1.05091545e-03 6.52323723e-01 -6.91709742e-02 1.20633312e-01 5.78053176e-01 1.47653058e-01 -4.74372625e-01 3.58740300e-01 3.44347775e-01 1.24861315e-01 5.34077287e-01 -1.45551527e+00 -1.91751316e-01 8.83660674e-01 2.01927274e-01 7.23053157e-01 7.75363684e-01 -1.04789150e+00 -1.10090065e+00 -7.56354809e-01 7.08142817e-01 -2.00715005e-01 5.27836084e-01 -3.32001358e-01 -9.04980540e-01 7.24605203e-01 2.83323169e-01 -6.39847696e-01 1.11781049e+00 4.18655694e-01 -1.91471308e-01 -2.48832226e-01 -1.38256407e+00 2.85253942e-01 1.15035927e+00 -1.12199597e-01 -4.86380041e-01 3.37341368e-01 4.62198287e-01 2.49174654e-01 -1.13768721e+00 3.95071566e-01 4.63417143e-01 -1.08964550e+00 9.82413471e-01 -4.18264419e-01 -2.43281156e-01 -5.98433018e-01 -1.53926149e-01 -1.28789747e+00 -9.18521881e-01 -7.02954769e-01 5.71714878e-01 1.28311968e+00 4.32619840e-01 -5.46390533e-01 8.90426040e-01 2.96244413e-01 1.28768757e-01 -3.81448716e-01 -7.76455522e-01 -7.96091855e-01 -1.92527488e-01 -9.16894339e-03 3.34937662e-01 1.53682923e+00 7.59806752e-01 4.84844297e-01 3.59392203e-02 1.65074170e-01 1.16065228e+00 2.91933596e-01 6.29908860e-01 -1.75094378e+00 -1.35148570e-01 -6.50445044e-01 -6.34334743e-01 -6.08609617e-01 2.04040989e-01 -1.04998195e+00 -1.12215102e-01 -1.86047280e+00 3.92289370e-01 -6.44477963e-01 7.60903582e-03 2.84870386e-01 -1.73165482e-02 3.36526483e-01 9.88859907e-02 2.51185417e-01 -4.95938599e-01 -1.06484473e-01 7.12493539e-01 -9.30941626e-02 -7.22655892e-01 -2.26126984e-01 -5.38181245e-01 5.35251439e-01 8.28160346e-01 -6.05317652e-01 -7.87825525e-01 1.69069782e-01 2.52937049e-01 9.35055912e-02 -4.78689492e-01 -8.40969741e-01 4.18285996e-01 -2.56336123e-01 1.77235216e-01 -9.75049376e-01 7.83238411e-02 -1.01453710e+00 9.00349081e-01 6.30496800e-01 -3.38089764e-01 2.10417166e-01 -2.20608234e-01 5.66233516e-01 -7.55748004e-02 -3.09335500e-01 9.27987039e-01 5.73903695e-02 -2.44538829e-01 -9.18758847e-03 -4.52117413e-01 -1.59814022e-02 1.42548466e+00 -5.57839215e-01 -1.29040629e-01 -1.91776291e-01 -1.03595948e+00 4.78694916e-01 6.58429205e-01 -1.28803477e-01 3.44649881e-01 -1.40463269e+00 -6.81620955e-01 1.92137554e-01 -4.76665534e-02 -1.98255271e-01 -1.67487249e-01 8.75806987e-01 -8.75228465e-01 3.51250947e-01 -1.43230140e-01 -7.43857741e-01 -1.85540009e+00 8.83529484e-01 9.39080194e-02 -1.46959394e-01 -2.32669920e-01 7.26875663e-01 3.54014695e-01 -5.73380113e-01 -8.16183686e-02 1.00750066e-01 3.18147242e-02 3.51659685e-01 1.04813226e-01 8.97349954e-01 -1.67964682e-01 -7.98298240e-01 -5.40104091e-01 7.03209937e-01 2.22515538e-01 -1.72226489e-01 1.16117561e+00 -3.55400741e-01 -8.25544596e-01 2.89443642e-01 1.17718637e+00 1.53488368e-01 -4.96267974e-01 1.68471411e-01 3.07871908e-01 -6.05733097e-01 -1.91052929e-01 -3.37765872e-01 -9.38314080e-01 3.63815308e-01 2.00299453e-02 1.13214624e+00 1.20371783e+00 1.90649733e-01 8.74620602e-02 4.28872973e-01 1.84294239e-01 -1.19839382e+00 -3.58909726e-01 4.20457684e-02 3.84095967e-01 -8.05768073e-01 4.31873322e-01 -1.05729604e+00 -3.18745941e-01 9.19495940e-01 4.15068656e-01 -1.97531536e-01 9.19470727e-01 3.40700358e-01 -1.95842221e-01 -4.07105476e-01 -4.63505775e-01 -3.38521391e-01 2.56783307e-01 7.11783707e-01 4.25758809e-01 4.00569916e-01 -9.27120566e-01 2.65516881e-02 -2.85529464e-01 -6.55694664e-01 4.59031403e-01 5.97513616e-01 -7.82781363e-01 -1.15265274e+00 -6.18800938e-01 5.09016275e-01 -1.22100890e-01 2.44088843e-01 -1.14642501e+00 1.05146122e+00 9.91733074e-02 1.38990140e+00 1.48072451e-01 -5.63710153e-01 -3.06540150e-02 -1.93770617e-01 4.24086124e-01 -4.22832936e-01 -5.01571834e-01 5.35087764e-01 2.77409881e-01 -4.09251958e-01 -7.43550360e-01 -5.44224381e-01 -1.43317831e+00 -8.92628610e-01 -7.55273402e-01 9.96336579e-01 3.58652651e-01 4.15242732e-01 1.26538560e-01 2.42916644e-01 8.49887788e-01 -5.08940995e-01 1.00086853e-01 -6.47155285e-01 -1.08424473e+00 3.22557241e-01 -1.11422934e-01 -5.09195745e-01 -4.20119733e-01 1.69969648e-01]
[7.0391387939453125, 5.220965385437012]
d01b58b4-e14d-4a62-9907-274152cd5cb0
zerotop-zero-shot-task-oriented-semantic
2212.10815
null
https://arxiv.org/abs/2212.10815v1
https://arxiv.org/pdf/2212.10815v1.pdf
ZEROTOP: Zero-Shot Task-Oriented Semantic Parsing using Large Language Models
We explore the use of large language models (LLMs) for zero-shot semantic parsing. Semantic parsing involves mapping natural language utterances to task-specific meaning representations. Language models are generally trained on the publicly available text and code and cannot be expected to directly generalize to domain-specific parsing tasks in a zero-shot setting. In this work, we propose ZEROTOP, a zero-shot task-oriented parsing method that decomposes a semantic parsing problem into a set of abstractive and extractive question-answering (QA) problems, enabling us to leverage the ability of LLMs to zero-shot answer reading comprehension questions. For each utterance, we prompt the LLM with questions corresponding to its top-level intent and a set of slots and use the LLM generations to construct the target meaning representation. We observe that current LLMs fail to detect unanswerable questions; and as a result, cannot handle questions corresponding to missing slots. To address this problem, we fine-tune a language model on public QA datasets using synthetic negative samples. Experimental results show that our QA-based decomposition paired with the fine-tuned LLM can correctly parse ~16% of utterances in the MTOP dataset without requiring any annotated data.
['Subhro Roy', 'Jason Wolfe', 'Dheeraj Mekala']
2022-12-21
null
null
null
null
['semantic-parsing']
['natural-language-processing']
[ 6.59707963e-01 8.37527752e-01 1.37444377e-01 -7.29459643e-01 -1.51722038e+00 -5.82926393e-01 2.86325902e-01 1.56615317e-01 -1.69421718e-01 3.16875279e-01 3.73574644e-01 -5.45538485e-01 3.62763047e-01 -1.05304515e+00 -8.13742459e-01 -1.93772987e-01 3.34022641e-01 7.55775094e-01 4.13604617e-01 -5.11030376e-01 -3.32839675e-02 -4.50256526e-01 -1.51455700e+00 7.18401492e-01 1.03405106e+00 7.31368601e-01 5.91010392e-01 9.24199939e-01 -9.57000256e-01 1.03101790e+00 -7.31391430e-01 -5.75421989e-01 -1.34106338e-01 -6.90298617e-01 -1.35010672e+00 -8.50045010e-02 4.35983002e-01 -3.47065598e-01 1.31632805e-01 1.02125418e+00 2.91541427e-01 3.06214303e-01 3.71597409e-01 -1.08285367e+00 -8.27742398e-01 6.26198947e-01 -1.49202980e-02 2.79226094e-01 5.38630545e-01 1.41030520e-01 1.46204269e+00 -8.98939073e-01 7.69052804e-01 1.89851725e+00 3.52614373e-01 1.11248481e+00 -1.29447675e+00 -2.90075988e-01 1.77866802e-01 4.62988801e-02 -5.06779313e-01 -4.78107870e-01 5.52708447e-01 -3.28580320e-01 1.34836996e+00 7.44155571e-02 1.73050426e-02 1.34231997e+00 2.23292634e-02 9.52254832e-01 7.52373219e-01 -6.64258957e-01 6.71331346e-01 -2.79924214e-01 1.02893043e+00 8.94663870e-01 -1.28438428e-01 -4.51572299e-01 -5.04869163e-01 -3.60266328e-01 6.04684697e-03 -2.24930421e-01 -1.15370927e-02 -1.64040878e-01 -8.20877075e-01 1.36758816e+00 2.54973769e-01 1.26369312e-01 -2.28255555e-01 3.15962024e-02 5.28537273e-01 4.24636334e-01 5.37232161e-01 6.86999202e-01 -7.37523496e-01 -2.34847561e-01 -3.31660986e-01 1.97241470e-01 9.34240520e-01 9.21375036e-01 9.21956241e-01 -1.52776167e-01 -5.38675606e-01 1.03907526e+00 -7.77271483e-03 2.86727160e-01 6.90658808e-01 -1.25702655e+00 8.51458609e-01 7.50660360e-01 -3.98538746e-02 -3.17512631e-01 -4.76318985e-01 1.60248026e-01 -7.89326280e-02 -2.10104331e-01 2.60926157e-01 -2.31503293e-01 -1.16781223e+00 1.97869432e+00 3.56577724e-01 7.25160688e-02 7.51187623e-01 6.97514296e-01 1.15300035e+00 8.65054071e-01 6.21344924e-01 1.14670113e-01 1.99754012e+00 -1.07855558e+00 -6.05315328e-01 -8.33444536e-01 9.27348375e-01 -2.90192217e-01 1.81921506e+00 -1.02416217e-01 -9.34084475e-01 -3.74200702e-01 -7.94479668e-01 -4.22803462e-01 -2.15011016e-01 -3.22663069e-01 5.74917376e-01 4.24556136e-01 -7.79694259e-01 2.84398437e-01 -7.91547000e-01 -5.87994635e-01 3.22626799e-01 -2.52885222e-01 -1.98055990e-02 -4.46973175e-01 -1.40493011e+00 6.91329360e-01 3.13156009e-01 -5.05755901e-01 -1.02960908e+00 -6.43747389e-01 -1.45210969e+00 4.33133811e-01 7.26081312e-01 -7.14292824e-01 1.82965803e+00 -7.11135268e-01 -1.26226032e+00 8.31879377e-01 -7.70038307e-01 -5.58796644e-01 -3.01642388e-01 -2.68462062e-01 -1.05384350e-01 3.93878430e-01 6.81263685e-01 9.10664022e-01 7.18244851e-01 -1.00221694e+00 -5.57577610e-01 -4.64265406e-01 5.51284671e-01 -1.00830570e-02 -1.29996389e-01 5.93696870e-02 -2.01601803e-01 -4.01662886e-01 3.04196328e-01 -5.87468207e-01 -2.47075841e-01 -6.01629198e-01 -3.16487193e-01 -5.31318605e-01 5.93700647e-01 -7.89007366e-01 7.61327446e-01 -2.00541973e+00 1.57553345e-01 -6.22761428e-01 -1.77924987e-02 5.44116572e-02 -6.86299622e-01 2.99041927e-01 1.39411524e-01 9.33407024e-02 -6.08273804e-01 -5.14400601e-01 1.31706178e-01 6.89849854e-01 -7.44485557e-01 -3.24571043e-01 6.60244286e-01 1.32034469e+00 -1.29786909e+00 -3.50788414e-01 1.61423031e-02 1.28474645e-02 -7.47181952e-01 5.98943651e-01 -8.67368400e-01 1.50334537e-01 -7.13711917e-01 7.34536827e-01 3.27756166e-01 -4.43803549e-01 1.31893575e-01 1.82618082e-01 5.82434177e-01 7.97129273e-01 -4.28935319e-01 1.99589992e+00 -6.67022943e-01 3.65208238e-01 8.07712146e-04 -9.86662388e-01 8.22591245e-01 3.75886172e-01 -1.64423749e-01 -9.60835040e-01 -3.98837589e-03 -4.61057536e-02 -2.32911691e-01 -7.85962224e-01 4.16432828e-01 -5.55013835e-01 -7.14884996e-01 7.52357841e-01 6.84234619e-01 -2.01505482e-01 7.76362568e-02 5.48549771e-01 1.57546854e+00 1.67114045e-02 2.02526212e-01 -8.05238336e-02 2.25426614e-01 3.36090922e-01 5.05995274e-01 9.01668310e-01 -2.70109028e-01 4.60989594e-01 8.17533195e-01 -3.83747667e-01 -6.19134009e-01 -1.17609453e+00 2.20826760e-01 1.82001293e+00 -9.53484848e-02 -3.30499649e-01 -1.23530936e+00 -1.01403165e+00 -3.04406434e-01 1.43877840e+00 -5.20544112e-01 -2.90732622e-01 -4.38528806e-01 -5.38007975e-01 5.00683129e-01 4.14634526e-01 1.91748023e-01 -1.41215730e+00 -1.04324114e+00 4.04832840e-01 -5.30741453e-01 -1.34605873e+00 -2.95494441e-02 4.09621984e-01 -6.92025721e-01 -1.26105750e+00 -3.83229226e-01 -8.71237338e-01 5.86990118e-01 2.26898298e-01 1.70235085e+00 -2.18236446e-02 -2.11991847e-01 6.13480687e-01 -8.14985752e-01 -2.97234505e-01 -9.71680462e-01 4.20726389e-02 -4.94452029e-01 -2.72123277e-01 8.26120615e-01 -2.69835174e-01 -1.35196865e-01 -1.64245903e-01 -8.15455675e-01 6.87794983e-02 2.02639863e-01 9.50239837e-01 2.22347781e-01 -6.21024728e-01 9.60537672e-01 -1.23292863e+00 8.83846283e-01 -7.47276545e-01 -4.57489580e-01 6.21713996e-01 -6.32714406e-02 6.04402244e-01 5.87549090e-01 -1.77724332e-01 -1.33553779e+00 -3.58860428e-03 -3.04279655e-01 -2.50616193e-01 -2.93084234e-01 3.29609215e-01 -3.48715454e-01 5.86675107e-01 7.84751236e-01 1.91410959e-01 -2.73360521e-01 -5.84054530e-01 7.84826458e-01 5.83335400e-01 7.00842798e-01 -1.00865662e+00 3.82844359e-01 1.37118921e-01 -4.63501126e-01 -5.86512387e-01 -1.64165294e+00 -3.85205925e-01 -2.42779389e-01 3.02570999e-01 1.48091710e+00 -8.17205310e-01 -3.39373738e-01 -8.73216838e-02 -1.46673977e+00 -4.08622652e-01 -3.86052012e-01 -2.04434618e-01 -6.75985992e-01 4.05724585e-01 -8.04778218e-01 -8.27740312e-01 -6.21440649e-01 -1.16232216e+00 1.55327320e+00 1.67556837e-01 -4.49860901e-01 -8.53000283e-01 -1.13981431e-02 7.77323663e-01 2.21772999e-01 -1.30957022e-01 1.57341051e+00 -9.77650821e-01 -3.72282237e-01 1.90966144e-01 -8.77477303e-02 1.85809702e-01 -4.24978845e-02 -7.73182154e-01 -1.40521073e+00 -1.16155455e-02 3.25519472e-01 -1.00829911e+00 1.01595533e+00 7.82759041e-02 1.04521060e+00 -2.91719794e-01 2.88718976e-02 1.73631608e-01 1.05939245e+00 6.74775895e-03 2.70301312e-01 1.57662496e-01 3.14978212e-01 9.38687444e-01 8.29208612e-01 1.45135835e-01 6.76193118e-01 3.48966360e-01 3.96976650e-01 3.69001925e-01 -8.27118531e-02 -5.77632368e-01 3.78278792e-01 5.67352295e-01 6.83315992e-01 -2.48143673e-01 -1.08114767e+00 7.12192833e-01 -1.74312532e+00 -6.80250049e-01 6.73626959e-02 1.78619635e+00 8.06788027e-01 1.47723019e-01 -5.20895302e-01 -5.21082520e-01 5.67579687e-01 2.42879391e-01 -6.80085480e-01 -6.26260102e-01 2.21689120e-01 6.30203843e-01 -6.53589657e-03 7.21100211e-01 -9.74612057e-01 1.47452986e+00 5.82191801e+00 5.73733866e-01 -5.09997427e-01 6.48996890e-01 5.57271361e-01 7.92270303e-02 -6.78345978e-01 3.15261215e-01 -7.56074786e-01 1.74182191e-01 1.36301398e+00 -1.10248670e-01 4.00325686e-01 1.06749594e+00 -2.60453016e-01 1.87198352e-02 -1.11181200e+00 5.95882416e-01 1.44220978e-01 -1.20666480e+00 2.78932363e-01 -4.61972088e-01 3.23466212e-01 6.05079904e-02 -4.45285559e-01 1.15155923e+00 7.32954383e-01 -9.67978060e-01 5.10078907e-01 4.87710163e-02 5.83495975e-01 -3.23208421e-01 5.03546894e-01 6.84352875e-01 -9.11856532e-01 -2.69572198e-01 -7.25046158e-01 -2.24232614e-01 4.48467284e-01 1.67700708e-01 -1.00892985e+00 2.51154125e-01 5.75968027e-01 2.19462946e-01 -4.71472800e-01 -4.76951227e-02 -6.63823962e-01 8.77345920e-01 1.51625142e-01 -4.34095003e-02 4.15964156e-01 2.11252332e-01 3.90335888e-01 1.02432215e+00 2.71780521e-01 4.36960161e-01 1.84998482e-01 1.18259835e+00 -2.28994280e-01 -1.33348450e-01 -4.79513735e-01 -2.78781742e-01 5.50729096e-01 1.16921794e+00 -5.83016157e-01 -8.65479350e-01 -6.05766892e-01 1.15357625e+00 7.06102490e-01 4.25559133e-01 -5.01091301e-01 -2.23540887e-01 8.58563304e-01 -2.71023601e-01 2.01979980e-01 1.44886106e-01 -2.28814468e-01 -1.48366106e+00 -4.24144603e-02 -8.48370314e-01 7.58674860e-01 -1.08237457e+00 -1.35144889e+00 6.25412166e-01 -1.28280565e-01 -5.07980108e-01 -7.59936571e-01 -6.29932523e-01 -8.65653336e-01 8.41354966e-01 -1.45622754e+00 -9.96498883e-01 -1.42906129e-01 2.78255790e-01 1.45839071e+00 -1.07146896e-01 1.31731498e+00 -4.07119840e-01 -4.08881217e-01 3.21093529e-01 -5.43630421e-01 1.15082964e-01 3.68142277e-01 -1.39054716e+00 1.15448534e+00 8.87192070e-01 2.03616917e-01 5.63727021e-01 8.36895347e-01 -5.81224978e-01 -1.40841722e+00 -1.20595157e+00 1.11155546e+00 -8.93864393e-01 6.30016804e-01 -7.22479284e-01 -1.35136139e+00 8.70873868e-01 6.82692900e-02 -5.07520735e-02 6.95994496e-01 3.04217011e-01 -6.59432292e-01 5.60270369e-01 -1.18337870e+00 3.52387488e-01 9.87961948e-01 -7.31104314e-01 -1.62986386e+00 4.51769859e-01 1.60423744e+00 -2.32730895e-01 -4.23511893e-01 1.64192259e-01 -4.18184996e-02 -7.16489613e-01 9.43222761e-01 -1.12735677e+00 7.43587732e-01 2.52377450e-01 -5.64047933e-01 -1.26582825e+00 1.29203364e-01 -1.32080048e-01 -1.27634570e-01 1.20367265e+00 5.24266958e-01 -4.71485645e-01 6.94921494e-01 9.54288423e-01 -3.73269022e-01 -4.33679521e-01 -1.18193054e+00 -6.07925296e-01 1.48984507e-01 -7.24389434e-01 6.33799016e-01 9.10734653e-01 1.87693104e-01 1.14763570e+00 -9.77806374e-03 2.05150515e-01 5.47011971e-01 3.52455914e-01 6.60829127e-01 -1.12745750e+00 -6.12252831e-01 7.04207271e-02 1.16242670e-01 -1.02777755e+00 7.08938241e-01 -9.64512348e-01 5.08917928e-01 -1.85133398e+00 4.35311645e-01 -1.37852117e-01 -1.06864281e-01 7.90655792e-01 -6.65207624e-01 -3.46245497e-01 1.83260709e-01 -2.16030672e-01 -8.00589740e-01 5.36921978e-01 8.24206471e-01 -2.58893400e-01 1.98199898e-02 -3.65838766e-01 -8.67543817e-01 7.67117977e-01 5.69824100e-01 -6.22093439e-01 -6.69577301e-01 -6.14725292e-01 2.97587272e-02 5.13872504e-01 3.23197901e-01 -5.77983737e-01 -1.51224315e-01 -1.05547532e-01 -2.33748287e-01 -1.94347396e-01 4.68655616e-01 -1.70014083e-01 -6.71286404e-01 3.85751307e-01 -5.85541487e-01 -1.66190729e-01 9.54320878e-02 5.42703807e-01 -2.39218011e-01 -7.10881114e-01 4.98878330e-01 -5.76503634e-01 -1.17316294e+00 -1.88261807e-01 -2.52115279e-01 6.47961140e-01 6.15781128e-01 2.81374007e-01 -7.09955037e-01 -3.39557827e-01 -6.92634463e-01 4.69964117e-01 3.24910849e-01 7.90000916e-01 6.57907665e-01 -7.92101741e-01 -5.50046325e-01 2.29688123e-01 5.84360540e-01 1.65929884e-01 3.37272495e-01 6.63655698e-02 1.63056888e-02 4.87869650e-01 8.88143554e-02 -4.96169895e-01 -9.87952352e-01 6.45003438e-01 2.11979926e-01 -2.60437548e-01 -7.35461652e-01 9.98132706e-01 7.59859979e-01 -1.02155864e+00 6.75708726e-02 -5.90898275e-01 -1.21177323e-01 -1.34318933e-01 7.23723054e-01 -2.00577244e-01 -6.31165924e-03 -3.23163211e-01 -2.34765574e-01 2.04084575e-01 8.40988681e-02 -1.39575705e-01 1.24116623e+00 -6.32309616e-02 1.10697769e-01 4.62091744e-01 1.18873155e+00 -4.64253664e-01 -1.17130446e+00 -2.70200729e-01 4.47882712e-01 -1.49461389e-01 -2.94854432e-01 -8.74454200e-01 -3.94389361e-01 1.26425064e+00 2.35143334e-01 1.27675757e-01 8.83342683e-01 7.22517312e-01 1.23608470e+00 7.42997289e-01 5.21889985e-01 -8.03899229e-01 4.15733188e-01 9.97453868e-01 5.50927937e-01 -1.44291592e+00 -9.15234864e-01 -4.19073969e-01 -7.72097886e-01 8.75928998e-01 8.13136697e-01 2.26582736e-01 -8.27512294e-02 -1.26873953e-02 2.27904737e-01 -5.20775318e-01 -1.25423169e+00 -4.00645047e-01 -4.33218963e-02 5.77539980e-01 3.84385645e-01 1.46787986e-01 -1.63106937e-02 1.21384430e+00 -3.31366003e-01 -2.02271625e-01 4.09469157e-01 1.13496232e+00 -1.12553251e+00 -1.03979886e+00 -2.12917656e-01 3.66975904e-01 -3.14280957e-01 -4.27072674e-01 -3.21485400e-01 2.79669017e-01 -2.18121305e-01 1.22859657e+00 2.08492532e-01 -1.35827765e-01 4.00417268e-01 8.15539539e-01 2.23347157e-01 -1.32808006e+00 -2.39023700e-01 -4.53672171e-01 4.34567422e-01 -7.97178328e-01 1.24745384e-01 -4.07623649e-01 -1.77626336e+00 4.22422588e-01 1.51425466e-01 1.97073430e-01 5.10663152e-01 1.26014566e+00 4.35102105e-01 5.83952546e-01 1.18252113e-01 -3.16056401e-01 -9.26742554e-01 -1.04406011e+00 -2.28330158e-02 7.25085080e-01 3.03385943e-01 -5.52299619e-01 -3.45603824e-01 -9.44828242e-03]
[10.796916007995605, 8.973309516906738]
2d9d0368-6125-43aa-9038-3021a4dbe9bc
video-swin-transformer
2106.13230
null
https://arxiv.org/abs/2106.13230v1
https://arxiv.org/pdf/2106.13230v1.pdf
Video Swin Transformer
The vision community is witnessing a modeling shift from CNNs to Transformers, where pure Transformer architectures have attained top accuracy on the major video recognition benchmarks. These video models are all built on Transformer layers that globally connect patches across the spatial and temporal dimensions. In this paper, we instead advocate an inductive bias of locality in video Transformers, which leads to a better speed-accuracy trade-off compared to previous approaches which compute self-attention globally even with spatial-temporal factorization. The locality of the proposed video architecture is realized by adapting the Swin Transformer designed for the image domain, while continuing to leverage the power of pre-trained image models. Our approach achieves state-of-the-art accuracy on a broad range of video recognition benchmarks, including on action recognition (84.9 top-1 accuracy on Kinetics-400 and 86.1 top-1 accuracy on Kinetics-600 with ~20x less pre-training data and ~3x smaller model size) and temporal modeling (69.6 top-1 accuracy on Something-Something v2). The code and models will be made publicly available at https://github.com/SwinTransformer/Video-Swin-Transformer.
['Han Hu', 'Stephen Lin', 'Zheng Zhang', 'Yixuan Wei', 'Yue Cao', 'Jia Ning', 'Ze Liu']
2021-06-24
null
http://openaccess.thecvf.com//content/CVPR2022/html/Liu_Video_Swin_Transformer_CVPR_2022_paper.html
http://openaccess.thecvf.com//content/CVPR2022/papers/Liu_Video_Swin_Transformer_CVPR_2022_paper.pdf
cvpr-2022-1
['classification']
['methodology']
[-1.00641429e-01 -3.30111861e-01 -4.28591311e-01 -3.20141524e-01 -6.41346276e-01 -5.20365536e-01 8.55282545e-01 -5.00707030e-01 -5.84083378e-01 2.82659978e-01 4.83731031e-01 -3.50195289e-01 1.47634000e-01 -3.68176848e-01 -1.03781962e+00 -4.69646037e-01 -1.73182264e-02 1.42318949e-01 2.77123004e-01 -4.55942675e-02 6.91747591e-02 1.76029250e-01 -1.34463000e+00 9.00777280e-01 1.11251645e-01 1.34717429e+00 -1.50070503e-01 9.33331490e-01 1.83750167e-01 1.32120466e+00 -4.70945090e-02 -4.40401375e-01 2.90159285e-01 -6.96085468e-02 -8.51626277e-01 -2.30328627e-02 9.54655409e-01 -5.41124582e-01 -9.69075382e-01 6.47674799e-01 2.15638503e-01 -1.17888683e-02 3.93858373e-01 -1.35252738e+00 -8.76591742e-01 3.94793063e-01 -4.91277725e-01 7.43176520e-01 1.08956292e-01 5.26372790e-01 1.12374473e+00 -1.12323201e+00 5.14263451e-01 1.10314023e+00 9.21177685e-01 6.39548838e-01 -1.23534834e+00 -6.21901572e-01 5.14451981e-01 5.56912363e-01 -1.37490964e+00 -7.10040331e-01 1.97610125e-01 -6.39338374e-01 1.65300000e+00 8.69597867e-03 9.39853191e-01 1.45911562e+00 3.67710084e-01 9.68366146e-01 8.45171034e-01 5.16728200e-02 2.88138650e-02 -4.16932374e-01 9.75263789e-02 7.07188785e-01 -2.82188773e-01 8.35950524e-02 -1.09797454e+00 2.61558264e-01 9.99197483e-01 3.62244278e-01 -1.51283547e-01 -2.08367780e-01 -1.30068338e+00 5.92700422e-01 5.61581612e-01 2.89953202e-01 -3.50321949e-01 7.72083580e-01 6.92056417e-01 4.10103768e-01 4.95173514e-01 7.55953463e-03 -7.04800665e-01 -6.50258541e-01 -9.97242332e-01 2.88437635e-01 3.15632194e-01 8.99159133e-01 4.58966762e-01 1.82466388e-01 -1.48085088e-01 4.75547522e-01 1.48914158e-01 4.07159626e-01 6.81689858e-01 -1.12819290e+00 3.86576444e-01 5.28501451e-01 -9.75027755e-02 -6.82700217e-01 -2.84323603e-01 -4.95700687e-01 -7.73132563e-01 2.45544557e-02 4.65730637e-01 2.32113108e-01 -1.17720520e+00 1.81295574e+00 1.23397812e-01 6.00505650e-01 -1.88299716e-01 9.00677562e-01 7.47118592e-01 5.97223043e-01 2.28165641e-01 1.61517918e-01 1.41678977e+00 -1.26221097e+00 -4.18004721e-01 -2.73578137e-01 6.97068512e-01 -5.27409732e-01 1.23046553e+00 5.63703060e-01 -1.04117596e+00 -7.34196723e-01 -8.50825429e-01 -5.28457642e-01 -1.80505961e-01 1.76883250e-01 6.80337906e-01 1.86879739e-01 -1.53626764e+00 6.31602347e-01 -1.24280477e+00 -5.51888406e-01 7.44298220e-01 4.19500977e-01 -6.61954463e-01 -1.68932930e-01 -8.61246526e-01 6.41407788e-01 -2.23420113e-01 3.25423921e-03 -1.46866441e+00 -1.10326135e+00 -5.04967451e-01 -7.07469583e-02 1.90282285e-01 -7.49109507e-01 1.56782115e+00 -1.29775620e+00 -1.39500237e+00 8.43747914e-01 -3.24145734e-01 -9.08097744e-01 5.07053435e-01 -6.04972899e-01 -3.33993465e-01 2.15034604e-01 1.01840086e-02 9.33897197e-01 9.64119375e-01 -4.74810392e-01 -6.90974653e-01 -3.20556104e-01 3.40131342e-01 3.82849462e-02 -6.27630770e-01 -6.48513064e-02 -5.91032028e-01 -7.48005211e-01 -2.33304814e-01 -1.00504196e+00 8.93265158e-02 2.53049165e-01 1.81071743e-01 -2.46611401e-01 1.06144249e+00 -6.09850228e-01 1.12625945e+00 -2.18739271e+00 3.02826911e-01 -4.46985751e-01 5.80702722e-01 3.91902715e-01 -4.30751324e-01 4.30106431e-01 -3.29124331e-01 7.41460174e-02 3.39105397e-01 -4.52500373e-01 -1.59235135e-01 -8.07779608e-04 -4.35141474e-01 4.67573375e-01 2.75201738e-01 1.28244221e+00 -6.52168572e-01 -8.68865848e-03 1.98465422e-01 8.16434920e-01 -8.60698044e-01 -5.71405739e-02 -2.09112555e-01 1.55222282e-01 -1.77065626e-01 6.13299787e-01 3.10614944e-01 -6.61695182e-01 1.40971601e-01 -5.19828320e-01 -1.46134913e-01 4.09806728e-01 -4.51615870e-01 2.01283956e+00 -3.18275571e-01 8.99020374e-01 -6.64607137e-02 -1.04545665e+00 3.36081833e-01 4.75348860e-01 8.24897707e-01 -9.10273492e-01 7.33687878e-02 -1.11961141e-01 -7.05990387e-05 -3.01940203e-01 2.21346051e-01 -5.12212664e-02 3.29822958e-01 3.39367390e-01 5.34308672e-01 5.03163874e-01 1.42001882e-01 4.18064296e-01 1.56183970e+00 4.83653754e-01 -1.11656830e-01 -3.84980977e-01 2.41112441e-01 1.22730266e-02 4.77483183e-01 4.04942155e-01 -3.49647760e-01 4.71480072e-01 5.07323384e-01 -9.74196553e-01 -1.05905890e+00 -9.87574279e-01 1.46614864e-01 1.34027219e+00 -3.08480173e-01 -9.65591967e-01 -7.98994660e-01 -6.39948845e-01 3.90611105e-02 1.72422990e-01 -9.85560596e-01 -2.78179079e-01 -5.10533273e-01 -3.73252004e-01 7.66819417e-01 9.93143320e-01 5.15373826e-01 -6.73319161e-01 -6.18075132e-01 1.01349950e-01 -2.69469500e-01 -1.30515075e+00 -5.56674123e-01 9.21087861e-02 -1.06157541e+00 -1.05436921e+00 -5.52672386e-01 -4.06640172e-01 1.98344290e-01 4.28129166e-01 1.14176881e+00 -7.32873678e-02 -1.77440017e-01 5.60438633e-01 -3.89567673e-01 -1.10639170e-01 2.71165222e-01 1.28933877e-01 2.12375253e-01 1.89707071e-01 6.27059639e-01 -8.36006224e-01 -8.33077669e-01 3.94352376e-01 -8.12121511e-01 3.29321742e-01 3.09194803e-01 8.31541121e-01 4.63567406e-01 -4.78544354e-01 2.15787306e-01 -4.02403980e-01 2.05406640e-02 -4.68918055e-01 -3.24043393e-01 -5.50234579e-02 -3.77068669e-01 -5.91854230e-02 6.22952819e-01 -6.43610060e-01 -5.58620155e-01 1.74434402e-03 6.88436478e-02 -1.14356411e+00 9.87542495e-02 3.85328859e-01 2.08741829e-01 1.81858242e-02 7.15956032e-01 3.61308455e-01 1.12500772e-01 -4.33557361e-01 3.33137006e-01 1.76733077e-01 4.64811504e-01 -4.79606479e-01 4.51643109e-01 7.27821410e-01 -2.00375989e-01 -5.85549593e-01 -8.30735624e-01 -3.15445900e-01 -5.69679558e-01 -3.52443427e-01 1.03868544e+00 -1.36845911e+00 -8.15708816e-01 7.61636734e-01 -8.26837063e-01 -7.93374300e-01 -2.12068945e-01 4.45988923e-01 -5.65636098e-01 -8.26378725e-03 -9.53827739e-01 -2.07768708e-01 -3.16249192e-01 -1.14127362e+00 1.05371368e+00 -1.66162476e-02 -1.63406551e-01 -7.32343554e-01 -4.49438766e-02 5.78813314e-01 7.11292088e-01 -2.03810528e-01 4.67416525e-01 -2.33605012e-01 -7.97579527e-01 7.61981457e-02 -2.77192324e-01 4.79923517e-01 -9.00252908e-03 5.90987094e-02 -1.19777346e+00 -4.66943681e-01 -2.16841936e-01 -6.60783827e-01 1.18052542e+00 5.38114667e-01 1.01193309e+00 -3.01922172e-01 -2.34795138e-01 9.30575609e-01 1.16795731e+00 6.51952475e-02 7.08419681e-01 2.41802588e-01 9.29058909e-01 3.40109095e-02 2.36253679e-01 4.42421466e-01 5.52891254e-01 7.68155634e-01 4.36092645e-01 5.73863424e-02 -3.56403887e-01 -3.59803110e-01 8.11078966e-01 5.96579313e-01 -3.50997925e-01 -1.72665119e-01 -1.03963387e+00 6.49193466e-01 -2.11082935e+00 -1.17227221e+00 1.33681864e-01 1.87076771e+00 6.04425490e-01 2.16751218e-01 4.72083867e-01 -1.11550659e-01 2.13980228e-01 3.26599628e-01 -6.53897166e-01 -1.99966386e-01 3.08803432e-02 2.09921807e-01 3.94178510e-01 4.24892873e-01 -1.27828157e+00 1.09122074e+00 6.23885202e+00 8.74471188e-01 -1.57436419e+00 3.90727580e-01 8.21370065e-01 -7.41348565e-01 8.54029581e-02 2.32552178e-02 -7.53730595e-01 3.97160143e-01 1.30660832e+00 -7.16460943e-02 5.12081325e-01 7.37080216e-01 2.53187478e-01 1.01437010e-01 -1.45822001e+00 1.23815107e+00 -6.50729090e-02 -1.70427644e+00 2.10178807e-01 2.09021077e-01 4.94142562e-01 6.49001062e-01 2.88012385e-01 3.45005989e-01 1.46307796e-01 -1.21615100e+00 1.07010925e+00 5.66967130e-01 1.00156379e+00 -3.76570910e-01 2.60150999e-01 7.16669336e-02 -1.46002471e+00 -2.04691231e-01 -1.89171270e-01 -4.76755053e-01 -5.37796430e-02 2.75690138e-01 -3.94268960e-01 1.12792827e-01 1.27177835e+00 1.40697575e+00 -6.02312922e-01 5.22104084e-01 1.55613407e-01 7.55987823e-01 -2.51233399e-01 3.35161597e-01 4.84372050e-01 7.38440603e-02 1.20507956e-01 1.28876746e+00 1.04691640e-01 1.45872846e-01 9.79610533e-02 3.88750792e-01 -2.82592177e-01 -2.58576095e-01 -5.29873967e-01 -3.51124078e-01 1.05012380e-01 1.06722438e+00 -3.85581106e-01 -5.48992872e-01 -7.36687124e-01 9.91975486e-01 5.10831177e-01 4.71621990e-01 -1.20689380e+00 2.18491122e-01 1.20918322e+00 5.37162483e-01 6.69120610e-01 -2.87582755e-01 -2.90939417e-02 -1.41726005e+00 4.47085574e-02 -1.19057286e+00 3.16479713e-01 -8.31821322e-01 -1.11293340e+00 5.01100898e-01 -8.88973698e-02 -1.11151540e+00 -4.36536269e-03 -9.03108060e-01 -3.10148031e-01 3.67628813e-01 -1.11049974e+00 -1.49825716e+00 -2.47971222e-01 9.25382853e-01 7.18596816e-01 -1.53301254e-01 6.46921694e-01 6.70522451e-01 -4.90377486e-01 6.57569468e-01 -5.87021038e-02 2.92608291e-01 6.91814125e-01 -7.58126736e-01 6.89806044e-01 8.88494253e-01 3.21279645e-01 6.72965586e-01 3.32448781e-01 -2.81419337e-01 -1.72218931e+00 -1.08315706e+00 8.05611372e-01 -7.62963891e-01 9.63301480e-01 -4.20579255e-01 -6.17549837e-01 1.21875739e+00 3.34540963e-01 4.57696676e-01 5.20919979e-01 2.48725891e-01 -1.15593565e+00 -3.35539579e-01 -6.55672014e-01 5.36202431e-01 1.52783895e+00 -9.43197608e-01 -1.08169310e-01 3.20238024e-01 6.43319011e-01 -3.01767737e-01 -9.66424942e-01 3.25925618e-01 9.33238328e-01 -1.09576595e+00 9.16033447e-01 -9.25431252e-01 7.20263481e-01 -4.15353060e-01 -3.73605758e-01 -8.13363194e-01 -7.72566378e-01 -5.32508671e-01 -4.24160302e-01 8.08949888e-01 2.55018055e-01 -4.06585336e-01 8.91590178e-01 4.11319196e-01 -1.80485383e-01 -1.00400698e+00 -1.09216213e+00 -7.04890847e-01 2.50583142e-02 -6.92288935e-01 1.39570594e-01 8.75646174e-01 -3.39032747e-02 5.08767068e-01 -5.31680882e-01 -2.15749994e-01 3.46771687e-01 -1.79926500e-01 8.43781888e-01 -6.13133967e-01 -4.51659471e-01 -5.16471505e-01 -7.52717912e-01 -1.40890002e+00 -8.85785818e-02 -6.53604507e-01 -4.57152128e-01 -1.26178551e+00 4.57280099e-01 6.53668866e-02 -5.45702159e-01 9.41733122e-01 1.93209544e-01 6.98278248e-01 4.18754399e-01 1.70642272e-01 -8.90618026e-01 6.29275441e-01 9.33806717e-01 -2.81691045e-01 3.40936601e-01 -5.30856550e-01 -6.61801696e-01 5.80016792e-01 7.40918219e-01 -3.04926902e-01 -7.32573211e-01 -1.07573617e+00 8.28805119e-02 -1.91343516e-01 7.06304252e-01 -1.38144696e+00 1.86603114e-01 -8.41899365e-02 4.81566697e-01 -1.87766507e-01 6.42814994e-01 -5.32413244e-01 3.10618013e-01 4.77083653e-01 -3.22676957e-01 4.11189586e-01 4.96616483e-01 5.84399700e-01 -2.93119401e-01 6.96380258e-01 6.66269064e-01 -1.07162677e-01 -8.99189591e-01 6.68201089e-01 -5.30682623e-01 1.07585765e-01 8.95741582e-01 -2.21864939e-01 -6.03633404e-01 -4.53981787e-01 -7.47905850e-01 -3.40277096e-04 5.42669952e-01 6.49531722e-01 5.46241701e-01 -1.37598050e+00 -5.89591920e-01 1.66300476e-01 4.77526821e-02 -5.22104144e-01 6.32092237e-01 1.17924428e+00 -4.75957304e-01 6.36326492e-01 -4.55538422e-01 -8.92072082e-01 -1.22758794e+00 5.84039092e-01 5.05683362e-01 -3.20360392e-01 -6.94648564e-01 1.25825465e+00 6.03400528e-01 1.11797579e-01 1.84065878e-01 -5.91200531e-01 3.26464266e-01 -9.00728852e-02 7.26208627e-01 1.53790936e-01 1.83527634e-01 -6.51434064e-01 -7.54611909e-01 6.38402045e-01 -3.68057042e-01 -1.96513265e-01 1.39465833e+00 2.07723707e-01 1.49626657e-01 3.69814336e-01 1.25988352e+00 -4.50183988e-01 -1.78303039e+00 -2.31810287e-01 -3.36803466e-01 -4.24098253e-01 1.72218233e-01 -7.45454252e-01 -1.35209620e+00 1.16411638e+00 6.55632079e-01 -2.04727232e-01 1.14848137e+00 -4.36988734e-02 7.24832714e-01 2.19302133e-01 3.44168067e-01 -9.02090549e-01 4.09557521e-01 7.01295376e-01 9.62411821e-01 -9.97749150e-01 -7.12439716e-02 5.72491474e-02 -5.99912345e-01 8.49176586e-01 8.30215394e-01 -4.21387970e-01 8.34849596e-01 5.11780798e-01 -6.82148850e-03 -2.77598679e-01 -1.50208116e+00 1.12080775e-01 1.22247234e-01 4.33420807e-01 6.91334903e-01 -4.08355370e-02 1.22620687e-01 4.20385391e-01 1.73574373e-01 4.83736008e-01 1.24379426e-01 9.17243123e-01 -1.05987683e-01 -9.59973574e-01 1.41856208e-01 4.36053634e-01 -5.64342737e-01 -3.90313625e-01 -3.03672373e-01 6.57734692e-01 -2.35167462e-02 7.39931703e-01 1.68097347e-01 -8.77427995e-01 1.99015200e-01 1.60489127e-01 6.01940572e-01 -2.07801640e-01 -6.94929719e-01 8.68903995e-02 1.56597048e-02 -1.34136546e+00 -5.69999635e-01 -6.26621544e-01 -1.04444695e+00 -7.57972538e-01 1.96815342e-01 -2.02483013e-01 3.02610517e-01 7.73364305e-01 8.71563077e-01 5.02686024e-01 4.74692173e-02 -9.57718909e-01 -3.08326274e-01 -9.33664620e-01 -1.83743492e-01 2.67845899e-01 4.62310225e-01 -6.58795118e-01 -1.30147591e-01 5.14081299e-01]
[8.981473922729492, 0.6798035502433777]
4526222b-d2f7-407f-b89e-6d00fe088338
skg-a-versatile-information-retrieval-and
2306.04758
null
https://arxiv.org/abs/2306.04758v1
https://arxiv.org/pdf/2306.04758v1.pdf
SKG: A Versatile Information Retrieval and Analysis Framework for Academic Papers with Semantic Knowledge Graphs
The number of published research papers has experienced exponential growth in recent years, which makes it crucial to develop new methods for efficient and versatile information extraction and knowledge discovery. To address this need, we propose a Semantic Knowledge Graph (SKG) that integrates semantic concepts from abstracts and other meta-information to represent the corpus. The SKG can support various semantic queries in academic literature thanks to the high diversity and rich information content stored within. To extract knowledge from unstructured text, we develop a Knowledge Extraction Module that includes a semi-supervised pipeline for entity extraction and entity normalization. We also create an ontology to integrate the concepts with other meta information, enabling us to build the SKG. Furthermore, we design and develop a dataflow system that demonstrates how to conduct various semantic queries flexibly and interactively over the SKG. To demonstrate the effectiveness of our approach, we conduct the research based on the visualization literature and provide real-world use cases to show the usefulness of the SKG. The dataset and codes for this work are available at https://osf.io/aqv8p/?view_only=2c26b36e3e3941ce999df47e4616207f.
['Han-Wei Shen', 'Rui Qiu', 'Yamei Tu']
2023-06-07
null
null
null
null
['knowledge-graphs', 'information-retrieval']
['knowledge-base', 'natural-language-processing']
[-4.57014233e-01 4.34032222e-03 -1.41657129e-01 -1.97841838e-01 -2.00019076e-01 -7.61417270e-01 5.21518826e-01 4.69442368e-01 -2.12866321e-01 7.47867942e-01 2.44628206e-01 -3.83464545e-01 -3.77521694e-01 -1.10113931e+00 -2.02128902e-01 -2.05172673e-01 2.47202843e-01 4.95801419e-02 5.16299307e-01 -7.45249689e-02 6.20612621e-01 5.24710894e-01 -1.81000555e+00 1.80015802e-01 1.20962465e+00 7.29854941e-01 6.34353518e-01 1.27882287e-01 -8.05085719e-01 3.59497160e-01 -8.62294316e-01 -4.75624561e-01 -1.12451866e-01 -1.56594649e-01 -1.05097187e+00 -4.12655652e-01 -2.60931045e-01 3.95526499e-01 -5.03957868e-02 1.20240974e+00 6.10096574e-01 5.62354065e-02 3.69279087e-01 -1.44022560e+00 -7.14542031e-01 5.48698008e-01 -3.02472621e-01 8.34027156e-02 6.61269844e-01 -2.39531204e-01 7.39785969e-01 -9.67927217e-01 1.04955971e+00 1.20101738e+00 2.33743027e-01 2.73372263e-01 -5.11588037e-01 -1.02111161e+00 2.29991712e-02 6.17348433e-01 -1.55306208e+00 -2.56681949e-01 5.97879529e-01 -2.39884540e-01 8.28470528e-01 5.02901375e-01 6.54661477e-01 9.18934405e-01 -2.40079358e-01 5.26714146e-01 7.90688097e-01 -5.37054002e-01 4.38852400e-01 4.02668774e-01 4.05122608e-01 7.42993057e-01 7.33536363e-01 -6.39905989e-01 -6.17992759e-01 -1.70830816e-01 7.01180398e-01 6.83806092e-02 -2.26227418e-01 -1.73894118e-03 -9.20568526e-01 3.99594277e-01 2.67856181e-01 5.56662023e-01 -1.67149290e-01 -3.72576684e-01 3.39923233e-01 8.59393328e-02 8.90203640e-02 7.27438271e-01 -2.80381650e-01 -1.98824331e-01 -5.49965799e-01 3.69976848e-01 1.10721159e+00 1.51425445e+00 7.34463692e-01 -5.92228889e-01 -9.60533991e-02 7.92695343e-01 3.45436424e-01 3.58354032e-01 5.60192406e-01 -9.54826713e-01 3.56087893e-01 1.40122175e+00 2.31888220e-01 -1.30497444e+00 -3.80730569e-01 -3.36213827e-01 -4.11003321e-01 -1.13369644e-01 8.96569192e-02 1.43981412e-01 -6.62615061e-01 1.36295986e+00 4.82490361e-01 -4.67104241e-02 1.68067411e-01 7.61427820e-01 1.45683849e+00 5.31721830e-01 5.33411801e-01 -1.21397460e-02 1.56314349e+00 -6.63000166e-01 -1.17873883e+00 2.52256930e-01 8.42438340e-01 -7.74532378e-01 1.18129969e+00 3.79712656e-02 -6.90683782e-01 -2.78766066e-01 -9.89362836e-01 -2.46467233e-01 -1.14827311e+00 6.68966249e-02 4.43307042e-01 3.50342780e-01 -7.16446579e-01 3.54922712e-01 -7.50098765e-01 -6.48292899e-01 4.56783950e-01 -1.37579352e-01 -3.90735567e-01 -8.58647898e-02 -1.47771358e+00 8.19585145e-01 1.01853132e+00 -2.93098152e-01 -2.03751445e-01 -7.40878344e-01 -6.73894703e-01 2.24572808e-01 7.12483048e-01 -6.72019064e-01 6.80833101e-01 -4.56375867e-01 -9.97872174e-01 6.54172778e-01 -9.79527310e-02 8.63290876e-02 8.94840136e-02 -6.02046438e-02 -8.51734936e-01 2.75606155e-01 4.44792926e-01 2.76234627e-01 2.77608912e-02 -1.22437286e+00 -7.53855586e-01 -7.46264517e-01 4.72577214e-02 1.77385762e-01 -7.72181511e-01 5.54084182e-01 -1.01011348e+00 -7.44516671e-01 -1.69600874e-01 -4.81921196e-01 5.48119396e-02 -2.76124477e-01 -4.93193269e-01 -2.99739569e-01 1.10569978e+00 -7.06358731e-01 1.99854815e+00 -1.94447052e+00 -1.72211990e-01 3.69650900e-01 2.79502094e-01 3.65090460e-01 2.47901201e-01 6.63137496e-01 1.20299041e-01 5.85946023e-01 -2.63308436e-01 5.45927100e-02 1.43357605e-01 1.12345509e-01 5.43928184e-02 -2.60688394e-01 -1.18087128e-01 9.96349871e-01 -9.65188265e-01 -8.31710994e-01 -4.64940816e-02 3.68890852e-01 -1.13430262e-01 1.83042988e-01 -2.50387460e-01 1.91257715e-01 -1.09494674e+00 8.73784602e-01 5.68881631e-01 -5.26500404e-01 3.58051389e-01 -1.98343933e-01 -3.92052412e-01 -2.65304781e-02 -1.53219318e+00 1.98583686e+00 -2.30580539e-01 2.34841198e-01 9.51617509e-02 -8.81620944e-01 9.11084652e-01 4.81416672e-01 4.30030614e-01 -6.84922636e-01 1.76211074e-01 3.98604006e-01 -5.39718568e-01 -1.00971150e+00 4.89820153e-01 3.83885741e-01 -1.87566295e-01 3.77322733e-01 7.35243857e-02 -4.39519845e-02 7.89611816e-01 5.46293259e-01 1.18309247e+00 3.27416718e-01 3.74141306e-01 -2.55138367e-01 6.42077029e-01 1.44691288e-01 5.38161159e-01 5.01481414e-01 2.84690261e-01 -3.88159254e-03 4.37384039e-01 -7.34508783e-02 -6.70735717e-01 -6.56007826e-01 -3.13192070e-01 7.22478986e-01 3.60585362e-01 -1.06561506e+00 -7.97759056e-01 -5.97217143e-01 1.72176495e-01 8.40420485e-01 -3.11019808e-01 5.42549565e-02 4.92467545e-04 -5.06357968e-01 3.19696844e-01 4.78327155e-01 6.77933633e-01 -1.22518468e+00 -6.37600660e-01 3.75619791e-02 -1.68295398e-01 -1.03410804e+00 1.07146569e-01 -3.50972801e-01 -5.71566641e-01 -1.22494638e+00 -3.02507967e-01 -7.51168251e-01 6.64961755e-01 1.86350659e-01 1.01619494e+00 7.30320513e-02 -5.14496148e-01 3.86835515e-01 -7.44483113e-01 -7.36334562e-01 -1.90998062e-01 3.87435734e-01 -2.42867664e-01 -4.00305241e-01 7.56657541e-01 -5.65508127e-01 -5.09127259e-01 2.44319215e-01 -1.26032555e+00 2.92660058e-01 3.00074339e-01 3.52954298e-01 4.28427666e-01 2.12081447e-01 7.89959490e-01 -8.35801065e-01 9.02312756e-01 -7.54523396e-01 -8.97405565e-01 5.05541563e-01 -9.24953520e-01 7.68582597e-02 5.51547706e-01 1.44022956e-01 -1.26556146e+00 -1.40394107e-01 5.09250611e-02 -1.84830472e-01 -3.69803309e-01 1.01859272e+00 -6.91762269e-01 2.93829203e-01 3.74831945e-01 -2.02599868e-01 -1.28440142e-01 -9.18141127e-01 6.23766661e-01 1.22369564e+00 5.10286689e-01 -7.21848428e-01 6.81512892e-01 3.00299644e-01 -1.21783607e-01 -7.03037977e-01 -6.35617673e-01 -4.99226511e-01 -4.53068793e-01 -8.08355585e-02 8.64421785e-01 -7.47172654e-01 -8.03752124e-01 1.08529747e-01 -1.08728492e+00 2.32815653e-01 -2.68468529e-01 3.55663717e-01 -1.43056568e-02 3.04356366e-01 -2.09109828e-01 -4.43812877e-01 -5.10743558e-01 -7.66017318e-01 5.53354859e-01 6.08012676e-01 -7.77755082e-02 -8.95950675e-01 -2.64738560e-01 2.13610351e-01 3.48586798e-01 3.75742137e-01 9.55400705e-01 -9.10056293e-01 -5.43862760e-01 -1.39750481e-01 -4.14364398e-01 -8.28902423e-02 3.49121571e-01 2.60104567e-01 -6.41225755e-01 1.19757615e-01 -3.13480407e-01 1.67389095e-01 4.52130139e-01 -2.32093886e-01 1.43560970e+00 -2.33306661e-01 -7.27999866e-01 4.88944769e-01 1.37100780e+00 4.59948689e-01 5.36746740e-01 9.71712291e-01 8.95837724e-01 6.44639015e-01 7.01002598e-01 5.34445703e-01 5.69709241e-01 5.73251665e-01 2.18458310e-01 5.70028499e-02 -4.34094603e-04 -2.02764079e-01 -2.74330616e-01 9.39076543e-01 -2.10143134e-01 -2.41232455e-01 -1.04225910e+00 5.39650619e-01 -1.84567809e+00 -8.48748684e-01 -1.52650937e-01 1.87943125e+00 1.00333881e+00 -4.26771343e-02 -2.33085454e-01 3.57161053e-02 5.58316469e-01 -2.33162344e-01 -1.67802125e-01 -2.20684007e-01 -7.24042952e-02 3.38465452e-01 2.46381044e-01 1.73955604e-01 -7.12946177e-01 1.00853264e+00 4.66423893e+00 8.86529744e-01 -9.49588656e-01 4.65580113e-02 -2.70046405e-02 1.35184571e-01 -4.35771495e-01 2.23965272e-01 -8.63124669e-01 6.39374852e-01 9.58090603e-01 -8.69989812e-01 1.02913044e-01 6.78739965e-01 3.30427885e-01 -9.23634768e-02 -5.58940411e-01 9.07877147e-01 -2.36343205e-01 -1.60249209e+00 2.52725154e-01 -2.19990134e-01 3.74789655e-01 -4.07173693e-01 -4.63466495e-01 7.52172917e-02 3.89253765e-01 -7.22384691e-01 4.17107821e-01 6.12940609e-01 8.63371909e-01 -5.90699852e-01 5.12767196e-01 2.09980205e-01 -1.30297363e+00 -2.21995801e-01 -1.42274067e-01 3.51852059e-01 -6.70899004e-02 5.51306784e-01 -8.07593107e-01 1.30364597e+00 1.08479011e+00 9.10068631e-01 -7.97674298e-01 1.15623391e+00 -5.47351182e-01 2.67045110e-01 -2.48270944e-01 -1.86138347e-01 -3.23991835e-01 -2.66029298e-01 5.09474456e-01 1.37882435e+00 4.92649883e-01 1.68944523e-01 -1.56663239e-01 9.71822619e-01 -2.94716507e-01 6.97739601e-01 -3.79356861e-01 -4.63025659e-01 1.13846850e+00 1.43434119e+00 -8.26710582e-01 -5.00844002e-01 -6.72402203e-01 5.54099560e-01 2.24203199e-01 4.55766439e-01 -5.83929121e-01 -1.28418779e+00 5.35655975e-01 2.37351328e-01 1.22672543e-01 -1.69213101e-01 -1.89706892e-01 -1.29694736e+00 2.31567442e-01 -6.30628467e-01 7.24758208e-01 -1.03346705e+00 -9.68425274e-01 5.26486635e-01 3.07051867e-01 -1.05438006e+00 7.17189983e-02 -4.65516120e-01 -4.09309477e-01 9.71682668e-01 -1.27038777e+00 -9.13420618e-01 -6.10926747e-01 5.83844900e-01 9.40341130e-02 -1.63169503e-01 1.08560717e+00 4.76987392e-01 -8.08755755e-01 5.26286885e-02 8.11297596e-02 1.86854437e-01 7.10945547e-01 -1.17661893e+00 2.96045274e-01 8.44773769e-01 -2.66198754e-01 1.02077317e+00 4.89490390e-01 -1.03055048e+00 -1.39584565e+00 -8.64016652e-01 9.06757295e-01 -4.23248380e-01 8.84898961e-01 -3.34086120e-01 -1.19772100e+00 4.77732629e-01 8.89394581e-02 -1.09036468e-01 7.30518222e-01 -1.22700132e-01 -3.19059044e-01 -2.26979386e-02 -1.04775882e+00 6.34594560e-01 1.25665629e+00 -2.95791149e-01 -9.02458549e-01 1.00770453e-02 7.23039746e-01 -2.90122956e-01 -1.29128993e+00 3.62664968e-01 3.97022694e-01 -4.24522460e-01 8.24893653e-01 -5.16241550e-01 2.66439766e-01 -7.14745879e-01 1.39938056e-01 -9.20254230e-01 -1.45069256e-01 -2.71411628e-01 -1.67467698e-01 1.54161811e+00 4.78287220e-01 -8.10807943e-01 5.52458704e-01 7.89764106e-01 -1.25425622e-01 -5.28503418e-01 -6.62471294e-01 -6.58558190e-01 -2.21866712e-01 -2.51067251e-01 9.96668637e-01 1.26565909e+00 4.63971823e-01 2.12769896e-01 3.65987241e-01 1.38936862e-01 2.88293868e-01 5.19944310e-01 4.88001406e-01 -1.52855468e+00 2.41561115e-01 -5.55487752e-01 -4.44201410e-01 -4.16333526e-01 -5.93636222e-02 -1.13829625e+00 -7.17143059e-01 -2.06450558e+00 3.15184176e-01 -5.03502369e-01 -4.12260979e-01 7.78971314e-01 -2.88318515e-01 -2.71532476e-01 1.17366621e-02 3.22169065e-01 -6.52196586e-01 2.99657017e-01 1.18294382e+00 2.31790677e-01 -2.61393070e-01 -4.64811921e-01 -1.04039633e+00 4.84912634e-01 6.31531775e-01 -5.22131085e-01 -5.50116062e-01 -2.63729304e-01 2.39777416e-01 -3.22355986e-01 1.89122126e-01 -7.57846355e-01 4.65298384e-01 -3.43888134e-01 2.83411145e-01 -4.65219736e-01 -4.96306270e-02 -8.67463171e-01 4.19012785e-01 1.22695938e-02 -7.62842894e-02 2.18671948e-01 3.44034910e-01 2.85556436e-01 -3.53133589e-01 -3.76289696e-01 1.53732495e-02 -2.79600412e-01 -1.07232666e+00 1.15239054e-01 -9.33370963e-02 8.70709568e-02 1.01320338e+00 -9.50864032e-02 -5.79698801e-01 1.13716185e-01 -7.28459418e-01 6.62063301e-01 6.31611049e-01 8.25373352e-01 4.02344584e-01 -1.20318294e+00 -2.91416049e-01 5.13171740e-02 5.56659758e-01 1.04784891e-01 1.19838901e-01 3.99268597e-01 -5.17118096e-01 4.42279041e-01 -4.73375887e-01 -4.73680906e-02 -1.29189956e+00 6.72276378e-01 -7.76370615e-02 -8.83677900e-02 -6.27336144e-01 3.89631957e-01 -3.63143027e-01 -2.22413406e-01 2.18939915e-01 -7.01980442e-02 -8.45250785e-01 2.45025277e-01 5.42999744e-01 4.44090158e-01 2.49925688e-01 -4.01472002e-01 -5.69740057e-01 3.36032718e-01 1.36319682e-01 1.19924182e-02 1.50402761e+00 -1.21741146e-01 -3.00674647e-01 1.29019871e-01 9.15609658e-01 3.35896045e-01 -3.97630960e-01 -1.55056268e-01 5.51962972e-01 -5.15481055e-01 -9.94015113e-02 -9.61315691e-01 -8.74902904e-01 4.00884092e-01 3.72222275e-01 4.16148394e-01 1.20748007e+00 2.09560201e-01 3.24021220e-01 2.82506913e-01 3.04710388e-01 -1.22597706e+00 -4.30436134e-01 2.95700610e-01 8.66866112e-01 -9.80708718e-01 2.02128887e-01 -7.84868479e-01 -3.11372638e-01 1.25506961e+00 4.09924269e-01 6.50821507e-01 8.24057460e-01 3.58866751e-01 1.42518759e-01 -6.67521000e-01 -4.69716072e-01 -2.60253340e-01 2.44386449e-01 3.04829121e-01 4.70829368e-01 -3.99302281e-02 -8.68508875e-01 1.04318988e+00 -2.70815343e-01 4.20447797e-01 2.57985711e-01 1.44598615e+00 -3.42633754e-01 -1.52880955e+00 -2.36695692e-01 4.18303192e-01 -5.72165251e-01 3.62354927e-02 -4.77098584e-01 7.26359129e-01 7.71271661e-02 9.61557269e-01 -1.61846578e-01 -1.11567341e-01 6.35059416e-01 2.69453347e-01 4.24837768e-02 -7.67241120e-01 -4.83222812e-01 -4.03140299e-02 1.15140110e-01 -4.69381779e-01 -6.52998626e-01 -3.16525847e-01 -1.71258569e+00 -1.54209077e-01 -2.26510301e-01 6.05587125e-01 9.81693387e-01 6.58327162e-01 1.03818655e+00 6.07205391e-01 9.51906294e-02 4.15843911e-03 1.50151908e-01 -9.35243666e-01 -3.79956812e-01 6.22638047e-01 -4.29291576e-01 -8.51456106e-01 -2.12587878e-01 2.07674041e-01]
[9.297051429748535, 8.170109748840332]
266ccd50-8b56-4c06-b5a9-c52b89107d15
stacked-convolutional-and-recurrent-neural
1706.02292
null
http://arxiv.org/abs/1706.02292v1
http://arxiv.org/pdf/1706.02292v1.pdf
Stacked Convolutional and Recurrent Neural Networks for Music Emotion Recognition
This paper studies the emotion recognition from musical tracks in the 2-dimensional valence-arousal (V-A) emotional space. We propose a method based on convolutional (CNN) and recurrent neural networks (RNN), having significantly fewer parameters compared with the state-of-the-art method for the same task. We utilize one CNN layer followed by two branches of RNNs trained separately for arousal and valence. The method was evaluated using the 'MediaEval2015 emotion in music' dataset. We achieved an RMSE of 0.202 for arousal and 0.268 for valence, which is the best result reported on this dataset.
['Konstantinos Drossos', 'Roman Jarina', 'Sharath Adavanne', 'Miroslav Malik', 'Tuomas Virtanen', 'Dasa Ticha']
2017-06-07
null
null
null
null
['music-emotion-recognition']
['music']
[-1.35607094e-01 -9.60591733e-02 1.40792251e-01 -4.09924716e-01 -5.46347260e-01 -4.14343834e-01 1.00686394e-01 -6.55326396e-02 -6.46027029e-01 6.77783549e-01 3.35802376e-01 3.04565966e-01 -4.29260172e-03 -5.47177970e-01 -3.55128258e-01 -5.70135474e-01 -3.35905999e-01 -1.43490225e-01 -5.77729940e-01 -4.33920473e-01 1.11325338e-01 3.22346896e-01 -1.79775238e+00 5.09733856e-01 1.82266399e-01 1.78325832e+00 -5.41277349e-01 7.72112370e-01 -7.21881585e-03 9.88031805e-01 -7.33169198e-01 -4.91359860e-01 -1.01875179e-02 -3.42090130e-01 -6.23044550e-01 -5.81584334e-01 5.01247346e-02 3.05707008e-01 -1.10020652e-01 7.46253490e-01 7.54960895e-01 3.88685584e-01 4.73360091e-01 -1.07072866e+00 -5.20600438e-01 6.16282046e-01 -2.99188763e-01 -7.38549605e-02 2.06370458e-01 -3.34553242e-01 1.21025813e+00 -7.88739741e-01 4.80074018e-01 7.82662749e-01 1.09473443e+00 7.18794107e-01 -9.53189194e-01 -9.08704698e-01 -1.54136449e-01 2.64727026e-01 -1.34209871e+00 -1.71042159e-01 9.84764695e-01 -2.19700664e-01 1.25281739e+00 2.51365244e-01 1.18034875e+00 1.60508168e+00 2.32720330e-01 6.35715902e-01 1.11334670e+00 -1.58109486e-01 1.74825788e-01 9.33218934e-03 -3.52449678e-02 1.35559663e-01 -4.02102441e-01 2.11521313e-01 -7.48085260e-01 1.14154771e-01 6.17494643e-01 -3.86033922e-01 5.55155538e-02 1.91781580e-01 -9.66366887e-01 7.88421273e-01 5.64947248e-01 3.09627652e-01 -9.16791916e-01 4.50405926e-01 1.01153195e+00 5.12997448e-01 6.84077442e-01 8.05110633e-01 -5.76372802e-01 -6.77020252e-01 -1.02460980e+00 9.43163782e-02 8.18163097e-01 3.46803695e-01 1.65157467e-01 5.98235309e-01 -5.05305044e-02 1.01431835e+00 1.26569688e-01 8.18646327e-02 6.44443989e-01 -9.12013352e-01 -5.70456795e-02 5.49854159e-01 -1.40609086e-01 -9.72694516e-01 -6.54499888e-01 -8.38214159e-01 -1.17256749e+00 2.33269498e-01 -8.34805593e-02 -4.92841929e-01 -6.10266685e-01 1.85338163e+00 -1.06628872e-01 1.86879203e-01 4.74802583e-01 9.54481423e-01 1.43737113e+00 8.72813821e-01 6.02438450e-02 -3.90126556e-01 1.08483839e+00 -9.98746455e-01 -1.02029395e+00 1.70691058e-01 3.20641279e-01 -7.02664852e-01 1.00072396e+00 1.09388065e+00 -1.20280969e+00 -5.98645687e-01 -1.24330640e+00 4.06391136e-02 -6.27510965e-01 4.68295544e-01 9.75034475e-01 5.21116674e-01 -1.19384456e+00 8.33016992e-01 -3.01738381e-01 3.46547626e-02 1.63811877e-01 4.68898177e-01 -3.05339724e-01 9.28749859e-01 -1.50954545e+00 7.90123701e-01 3.34852546e-01 6.19828105e-01 -8.33179653e-01 -4.51087952e-01 -6.16189480e-01 8.36352706e-02 -1.13600053e-01 -1.53943956e-01 1.18560863e+00 -1.35986662e+00 -2.01358747e+00 8.30269635e-01 2.84217298e-01 -3.47384065e-01 7.30578229e-02 -6.76919580e-01 -8.65462065e-01 -1.34585232e-01 -4.81343657e-01 7.16534436e-01 5.35041630e-01 -9.35568213e-01 -1.08860336e-01 -2.92040646e-01 -2.19282210e-01 9.22398642e-02 -4.23582286e-01 1.81967244e-01 -2.52951026e-01 -6.82833850e-01 7.56169260e-02 -9.53794718e-01 -1.81174740e-01 -5.62795460e-01 -3.59861851e-01 -3.30339551e-01 3.16272259e-01 -5.65557241e-01 1.60246348e+00 -2.23411012e+00 3.18161279e-01 3.37827832e-01 -1.86929554e-01 2.46256236e-02 -1.48554698e-01 2.26152197e-01 -5.81114769e-01 -2.31052097e-02 1.56511590e-01 -4.28202152e-01 1.24269411e-01 -5.96436523e-02 -3.27673614e-01 2.09552243e-01 1.09207392e-01 8.43527257e-01 -4.42851603e-01 -1.86509967e-01 -2.24454522e-01 8.49272430e-01 -4.58459556e-01 3.35766971e-01 5.39326593e-02 3.20566922e-01 -2.56433878e-02 8.17782402e-01 2.47981146e-01 1.65210828e-01 1.55330390e-01 -3.08071733e-01 -4.82579798e-01 3.43630493e-01 -1.01253545e+00 1.97005427e+00 -5.90601921e-01 9.31108534e-01 -3.19925174e-02 -6.53965354e-01 1.55719471e+00 8.02571952e-01 5.62747121e-01 -7.09525108e-01 7.04391599e-01 1.96607798e-01 -1.06733300e-01 -4.03612167e-01 6.53748393e-01 -4.18692887e-01 -6.25196517e-01 3.52588743e-01 3.89506400e-01 1.38793647e-01 -7.02919886e-02 -2.88653970e-01 8.88805807e-01 3.03202182e-01 -2.61309482e-02 -1.04222231e-01 3.18572372e-01 -4.51471895e-01 8.21606696e-01 3.39439452e-01 -2.64720917e-01 3.69729042e-01 6.54633224e-01 -5.87225199e-01 -9.49740291e-01 -8.29423726e-01 -3.68752261e-03 1.16617966e+00 -2.49499246e-01 -7.57601559e-01 -5.61764181e-01 -1.20810382e-01 -4.25902396e-01 5.43706477e-01 -1.06470335e+00 -2.54754633e-01 -1.56581849e-01 -5.55691779e-01 1.08118820e+00 7.74216413e-01 4.49543864e-01 -1.74839818e+00 -8.64563107e-01 2.46395260e-01 -2.69757509e-01 -8.21867168e-01 2.09440351e-01 7.25498915e-01 -8.15892160e-01 -5.69397926e-01 -3.32577735e-01 -5.34046471e-01 -2.05068082e-01 -7.70221174e-01 1.35974896e+00 -4.38012064e-01 -2.42149070e-01 1.82108313e-01 -4.51987565e-01 -6.53154492e-01 1.97716624e-01 2.56399721e-01 1.48618504e-01 -6.47818297e-03 4.46282744e-01 -9.16006207e-01 -6.36562169e-01 -1.87244684e-01 -7.01345801e-01 -2.31272668e-01 4.98380214e-01 8.13776791e-01 6.83220983e-01 -2.56112784e-01 7.92323530e-01 -4.65379715e-01 9.94121850e-01 -3.53381902e-01 -1.26392573e-01 -3.36501785e-02 -6.55480385e-01 -2.45216981e-01 5.42568266e-01 -7.16077149e-01 -8.82461190e-01 2.01428756e-01 -5.37510693e-01 -7.20684290e-01 -5.16821258e-02 6.02684557e-01 1.25067934e-01 2.78797477e-01 5.66596627e-01 -2.45744005e-01 -4.38137829e-01 -3.82337362e-01 1.92746446e-01 6.84513032e-01 9.63854015e-01 -3.41550291e-01 -3.99551034e-04 -3.54283862e-02 4.14245911e-02 -4.11814988e-01 -8.80108058e-01 -3.45584571e-01 -4.95056391e-01 -6.71963036e-01 1.06904459e+00 -1.09896612e+00 -1.36791718e+00 2.44526371e-01 -1.09521544e+00 -1.32580116e-01 -3.66470158e-01 8.71007621e-01 -8.44401181e-01 -3.33573371e-01 -9.54988658e-01 -1.09244204e+00 -9.41068351e-01 -6.63084507e-01 6.98963046e-01 2.65104562e-01 -6.67957187e-01 -6.54173851e-01 5.42411804e-01 -2.12781116e-01 6.12307727e-01 7.16430902e-01 5.52405238e-01 -5.37313282e-01 4.36457157e-01 -3.05745870e-01 1.99325252e-02 6.40312433e-01 -3.65862042e-01 9.94313583e-02 -1.41913986e+00 2.17710167e-01 4.67990711e-02 -8.64665449e-01 9.12692785e-01 3.49253833e-01 1.41369092e+00 -1.47190481e-01 5.13675153e-01 7.12821364e-01 1.30397856e+00 2.75332570e-01 9.61185932e-01 5.63497424e-01 4.10624743e-01 3.76378089e-01 6.47339880e-01 8.22615385e-01 4.84910533e-02 4.01718795e-01 7.22379863e-01 -3.50282103e-01 4.36065912e-01 9.13735926e-02 3.95782739e-01 1.10477054e+00 -5.61385751e-01 6.04718216e-02 -6.43473029e-01 4.77162093e-01 -1.85996068e+00 -1.05908275e+00 -8.67801085e-02 1.76630270e+00 8.23808968e-01 -5.85428951e-03 2.43969172e-01 4.87928003e-01 4.14890945e-01 2.18786731e-01 -4.32942957e-01 -1.41133857e+00 -2.56984770e-01 7.29449928e-01 -5.16923629e-02 -1.38934925e-02 -1.16116297e+00 1.12527204e+00 7.15711069e+00 3.83493096e-01 -1.42256415e+00 -1.64915770e-01 4.64988828e-01 -5.81098914e-01 4.42595445e-02 -5.63210130e-01 -2.51209229e-01 8.87500420e-02 1.53292537e+00 2.88740665e-01 6.22953475e-01 9.68582988e-01 -1.72056053e-02 3.67653310e-01 -7.30306864e-01 1.35268915e+00 2.28788108e-01 -7.64947414e-01 -2.84929305e-01 -3.10345411e-01 5.24184585e-01 -5.13711832e-02 2.43782803e-01 8.48566055e-01 -2.47439981e-01 -1.33027077e+00 5.55523098e-01 9.25337672e-01 8.40033531e-01 -1.46554446e+00 9.01414573e-01 -3.05526555e-01 -1.11954105e+00 -1.77542076e-01 -2.98201621e-01 -3.66014034e-01 -1.38549373e-01 6.08507812e-01 -2.57426977e-01 3.50678235e-01 1.24114192e+00 9.55774426e-01 -2.76795477e-01 5.32462060e-01 -1.34200960e-01 5.91174304e-01 -5.53532988e-02 -2.12265283e-01 3.44546765e-01 -1.73320398e-01 3.64270896e-01 1.45400417e+00 5.55962980e-01 1.27621785e-01 -5.37707031e-01 7.64360487e-01 -4.68418837e-01 4.33048904e-01 -4.71093535e-01 -3.79212141e-01 5.22715077e-02 1.75218761e+00 -5.78128040e-01 -3.35371554e-01 1.01787694e-01 9.45768654e-01 2.65336663e-01 1.32324740e-01 -1.00706148e+00 -7.36615419e-01 7.36222029e-01 -7.48936236e-01 3.71769935e-01 1.31600365e-01 -3.71701062e-01 -9.96274292e-01 -2.05656007e-01 -8.13572586e-01 3.74080956e-01 -1.19721162e+00 -1.27830184e+00 1.26740992e+00 -6.84446871e-01 -1.31014717e+00 -4.06385273e-01 -7.18089104e-01 -6.21193469e-01 6.45254195e-01 -1.19713783e+00 -8.27462435e-01 -2.78831452e-01 7.19167411e-01 1.02209367e-01 -2.43537083e-01 1.49145389e+00 4.27606016e-01 -5.39280593e-01 6.85644984e-01 -1.66909188e-01 9.31104273e-02 9.06061351e-01 -1.16799355e+00 -1.49147749e-01 1.30706280e-02 -2.65123956e-02 4.13480520e-01 6.85038567e-01 -2.28995457e-01 -1.25651169e+00 -9.50572550e-01 1.01574981e+00 -1.26989782e-01 5.97296238e-01 -3.35529268e-01 -6.45215929e-01 5.33784211e-01 6.09066129e-01 -1.99201882e-01 1.28082669e+00 8.23903859e-01 -5.18973112e-01 -8.64387453e-02 -9.16473985e-01 4.37824547e-01 6.62568808e-01 -6.63021088e-01 -4.37752128e-01 -4.35407341e-01 4.42419618e-01 -4.07703787e-01 -1.41980290e+00 1.00666630e+00 1.23513842e+00 -1.05458272e+00 8.50567162e-01 -6.98310494e-01 8.49358439e-01 -2.93334723e-02 -4.26232249e-01 -1.39590073e+00 -2.67290443e-01 -3.18293154e-01 -1.97750971e-01 1.11892343e+00 5.39916635e-01 1.82251371e-02 5.14652193e-01 1.09185159e-01 -2.15673238e-01 -1.20183337e+00 -8.79721522e-01 -4.65932012e-01 -1.67365804e-01 -8.21688354e-01 5.36611378e-01 1.10488844e+00 2.45460123e-01 7.81792998e-01 -7.71913826e-01 -3.97472739e-01 1.48131941e-02 2.00744361e-01 4.79812711e-01 -1.52609134e+00 -3.58558372e-02 -6.79733217e-01 -4.88102287e-01 -1.03383869e-01 3.75667572e-01 -7.06048131e-01 1.46360546e-01 -1.22652102e+00 4.80481014e-02 1.93244219e-01 -1.12290812e+00 6.68407500e-01 3.84292275e-01 7.81391442e-01 9.82891098e-02 -2.18562111e-01 -8.35902274e-01 9.03495014e-01 7.98487127e-01 -7.02571049e-02 -3.57845843e-01 -1.73671320e-01 -6.45192087e-01 7.29814231e-01 1.22920704e+00 -3.35522920e-01 -9.97647420e-02 1.50852501e-01 9.24360037e-01 9.74602774e-02 1.15020216e-01 -1.16918838e+00 2.15876158e-02 4.01006877e-01 9.26587701e-01 -9.54639494e-01 8.88452172e-01 -6.27011120e-01 3.20446789e-01 1.17306575e-01 -7.13658333e-01 3.52034330e-01 6.40244365e-01 5.60585558e-02 -5.00760853e-01 1.25631407e-01 5.09227574e-01 5.58957458e-02 -4.37972665e-01 -3.46855447e-02 -4.14709896e-01 -3.24913472e-01 5.94542325e-01 2.43918914e-02 -6.87805703e-03 -5.93356431e-01 -1.23818493e+00 -1.37882695e-01 -1.15411513e-01 5.36018252e-01 7.56736755e-01 -1.84430218e+00 -6.18342578e-01 9.71882045e-02 7.94924200e-02 -7.14327276e-01 3.85200113e-01 8.93564880e-01 -6.49766847e-02 2.45107174e-01 -7.24206507e-01 -3.86915654e-01 -1.35853899e+00 2.94507563e-01 4.36989814e-01 -3.94694537e-01 -2.97142416e-01 8.12943578e-01 -3.94039333e-01 -6.79899216e-01 5.20331860e-01 -9.52132121e-02 -8.14288318e-01 5.48969567e-01 3.50383639e-01 4.52876627e-01 1.00520991e-01 -6.68913126e-01 -3.51136953e-01 4.52717781e-01 3.04938763e-01 -3.87354583e-01 1.72472227e+00 2.98169315e-01 -2.38291264e-01 1.23819673e+00 1.23591971e+00 -2.70366907e-01 -7.27163076e-01 3.31519932e-01 -1.13407575e-01 7.69223496e-02 3.13418210e-01 -1.18543696e+00 -1.23164809e+00 1.01895237e+00 9.76007879e-01 1.91348508e-01 1.55774260e+00 -3.72101933e-01 6.47174060e-01 5.07881403e-01 -2.83988297e-01 -1.65121579e+00 1.74033567e-01 8.75501454e-01 1.08407605e+00 -8.89175177e-01 -2.47380421e-01 3.29648942e-01 -9.99374390e-01 1.27951097e+00 5.76533616e-01 -3.05480748e-01 7.44326949e-01 2.55481571e-01 4.99269813e-01 -5.12769282e-01 -1.24857342e+00 -1.17236122e-01 4.44820315e-01 -1.26309872e-01 1.12282836e+00 1.47388101e-01 -3.26037288e-01 1.20914209e+00 -7.19753861e-01 1.20095357e-01 2.68272996e-01 6.05478287e-01 4.50714827e-02 -6.77466333e-01 -8.83510858e-02 2.26103157e-01 -1.01885712e+00 -4.13455665e-02 -8.36371422e-01 7.33706892e-01 1.34234190e-01 9.40798581e-01 2.04965949e-01 -9.12146688e-01 4.79437709e-01 3.40258688e-01 4.17345524e-01 -8.27987641e-02 -1.33850288e+00 3.24692994e-01 2.32950523e-01 -8.45916748e-01 -7.10120976e-01 -3.41386110e-01 -1.27474594e+00 -2.67929025e-02 -1.24108739e-01 2.95910925e-01 1.16370940e+00 4.75876778e-01 3.04206669e-01 1.14660215e+00 8.73544395e-01 -8.67520988e-01 9.76276621e-02 -1.25146699e+00 -9.52839077e-01 2.42426649e-01 7.80908316e-02 -4.77734089e-01 -3.58913660e-01 -3.43793482e-01]
[13.516203880310059, 5.050252437591553]
49545c8c-2279-4eef-8447-103cc6968b7e
elepose-unsupervised-3d-human-pose-estimation
2112.07088
null
https://arxiv.org/abs/2112.07088v1
https://arxiv.org/pdf/2112.07088v1.pdf
ElePose: Unsupervised 3D Human Pose Estimation by Predicting Camera Elevation and Learning Normalizing Flows on 2D Poses
Human pose estimation from single images is a challenging problem that is typically solved by supervised learning. Unfortunately, labeled training data does not yet exist for many human activities since 3D annotation requires dedicated motion capture systems. Therefore, we propose an unsupervised approach that learns to predict a 3D human pose from a single image while only being trained with 2D pose data, which can be crowd-sourced and is already widely available. To this end, we estimate the 3D pose that is most likely over random projections, with the likelihood estimated using normalizing flows on 2D poses. While previous work requires strong priors on camera rotations in the training data set, we learn the distribution of camera angles which significantly improves the performance. Another part of our contribution is to stabilize training with normalizing flows on high-dimensional 3D pose data by first projecting the 2D poses to a linear subspace. We outperform the state-of-the-art unsupervised human pose estimation methods on the benchmark datasets Human3.6M and MPI-INF-3DHP in many metrics.
['Helge Rhodin', 'James J. Little', 'Bastian Wandt']
2021-12-14
null
http://openaccess.thecvf.com//content/CVPR2022/html/Wandt_ElePose_Unsupervised_3D_Human_Pose_Estimation_by_Predicting_Camera_Elevation_CVPR_2022_paper.html
http://openaccess.thecvf.com//content/CVPR2022/papers/Wandt_ElePose_Unsupervised_3D_Human_Pose_Estimation_by_Predicting_Camera_Elevation_CVPR_2022_paper.pdf
cvpr-2022-1
['unsupervised-3d-human-pose-estimation']
['computer-vision']
[-1.33709326e-01 6.53824583e-02 -1.56642079e-01 -3.79246622e-01 -6.39857233e-01 -4.61197287e-01 4.02413577e-01 -3.97506356e-01 -7.89988399e-01 6.12680852e-01 5.14988840e-01 3.93399715e-01 4.10478443e-01 -2.96838880e-01 -8.58912051e-01 -3.57336581e-01 1.69439139e-04 1.08200753e+00 1.07925966e-01 -1.73367605e-01 -1.06635384e-01 4.06485587e-01 -1.22291172e+00 -2.06243724e-01 3.85407299e-01 5.87894440e-01 -2.69297920e-02 9.24280822e-01 3.28480482e-01 6.38516724e-01 -3.99486184e-01 -3.74929965e-01 5.89018703e-01 -4.36143309e-01 -7.85435379e-01 6.08980000e-01 7.82893479e-01 -6.60165489e-01 -4.77080375e-01 7.86653221e-01 5.88470757e-01 2.49696493e-01 7.48378515e-01 -1.25625384e+00 1.04847834e-01 -1.36203408e-01 -6.12386823e-01 -1.79955900e-01 7.97214866e-01 2.31552526e-01 6.86857820e-01 -9.87108469e-01 9.30115581e-01 1.25827348e+00 7.46299863e-01 6.38493359e-01 -1.15417016e+00 -2.95146406e-01 3.39602232e-02 -2.53220238e-02 -1.59064353e+00 -1.26126334e-01 8.72363091e-01 -8.04119706e-01 8.23017538e-01 -5.30605030e-04 1.03952110e+00 1.32266867e+00 -8.10768753e-02 1.05103779e+00 8.84958684e-01 -4.00016278e-01 1.11158922e-01 -2.09844798e-01 -2.81220138e-01 8.31160784e-01 1.92624614e-01 -2.09143564e-01 -7.81015813e-01 1.03196584e-01 9.87933099e-01 1.01172008e-01 -1.59448445e-01 -1.15123034e+00 -1.48028684e+00 6.82062924e-01 4.01992232e-01 -2.12809727e-01 -4.58901465e-01 2.27256760e-01 2.70037234e-01 -1.93145066e-01 4.46640909e-01 2.68409908e-01 -5.35257816e-01 -4.16186064e-01 -9.91128087e-01 5.48229516e-01 8.08737099e-01 1.05539238e+00 7.79014468e-01 -3.72211456e-01 1.16401948e-01 4.80277896e-01 4.01626647e-01 7.97249556e-01 1.87619478e-01 -1.16476798e+00 6.21485949e-01 5.68528593e-01 3.93766582e-01 -1.02113390e+00 -6.48012996e-01 -2.78959066e-01 -6.18693650e-01 -4.07831743e-02 8.11653197e-01 -3.42350006e-01 -7.97757864e-01 1.72790098e+00 5.60173869e-01 -9.37614590e-03 -1.56724393e-01 1.35108066e+00 3.78790021e-01 4.47257370e-01 -2.04079896e-01 1.85194761e-01 1.13105059e+00 -1.29331362e+00 -5.61671615e-01 -6.90670252e-01 6.75861657e-01 -6.09139919e-01 8.45163047e-01 4.51378912e-01 -7.94926941e-01 -5.97619236e-01 -8.40662420e-01 -2.45845094e-01 -1.86445303e-02 5.11645317e-01 5.33740640e-01 6.62040412e-01 -6.65494263e-01 3.72891814e-01 -1.25204539e+00 -7.70198464e-01 5.43192104e-02 2.85483271e-01 -8.81552935e-01 -1.90695360e-01 -9.67406631e-01 1.10586584e+00 2.61351526e-01 1.68886364e-01 -8.31260085e-01 -3.60168666e-01 -1.13968456e+00 -4.98049617e-01 6.44914031e-01 -9.11156297e-01 1.17895103e+00 -4.78508651e-01 -1.55674231e+00 1.03797674e+00 -5.08950166e-02 -3.68196279e-01 1.06082404e+00 -1.02281797e+00 3.52183759e-01 2.61896104e-01 1.58073217e-01 1.10922885e+00 7.25657642e-01 -1.17546785e+00 -2.70370275e-01 -6.15499377e-01 -2.82228570e-02 7.11634099e-01 -3.12665761e-01 -4.02224988e-01 -1.11021113e+00 -4.46790516e-01 1.22453414e-01 -1.50111103e+00 -4.05416906e-01 2.09562510e-01 -6.12329185e-01 1.47604257e-01 5.45059919e-01 -9.02815282e-01 9.20548737e-01 -1.71007979e+00 7.69642591e-01 1.12342067e-01 8.09440240e-02 3.54519933e-02 1.75120160e-01 3.01486701e-01 2.63470203e-01 -4.54724729e-01 -2.45621279e-01 -8.94499302e-01 2.33675763e-01 2.69663781e-01 9.19907242e-02 9.05269682e-01 1.05918877e-01 8.25135291e-01 -8.51540446e-01 -7.05655873e-01 5.03013968e-01 5.87605357e-01 -9.41091835e-01 5.36805153e-01 -5.37549667e-02 9.14599895e-01 -2.25888416e-01 4.01643723e-01 4.79334265e-01 -3.94453615e-01 2.15898797e-01 -3.00915688e-01 1.08625807e-01 -1.69149134e-02 -1.44468951e+00 2.41281629e+00 -1.62695214e-01 3.84405166e-01 -2.17563078e-01 -8.67183924e-01 7.21803904e-01 3.17766845e-01 9.54837143e-01 1.18459731e-01 2.07535401e-01 2.81238114e-03 -3.37929249e-01 -5.82803965e-01 4.48799998e-01 3.69484164e-02 -4.08936411e-01 3.15522790e-01 3.75942945e-01 -1.99778914e-01 1.65875599e-01 1.27472684e-01 9.45040941e-01 6.75405562e-01 3.74617100e-01 2.47205403e-02 4.33923036e-01 -9.36606433e-03 6.57507002e-01 3.67589116e-01 -3.44110072e-01 1.02349579e+00 4.88197953e-01 -4.93019968e-01 -1.22881997e+00 -1.09738851e+00 2.26327226e-01 7.76158750e-01 5.79810925e-02 -6.92993462e-01 -1.05496883e+00 -8.41451347e-01 -9.49880108e-02 8.07579309e-02 -5.40468276e-01 2.03917906e-01 -7.10473120e-01 -4.88788396e-01 4.11801934e-01 6.73012137e-01 4.14905041e-01 -5.72086811e-01 -8.65528047e-01 -7.46543854e-02 -6.11320615e-01 -1.50016356e+00 -7.73513138e-01 5.62237427e-02 -6.57727897e-01 -1.17411518e+00 -1.18588424e+00 -5.90100408e-01 9.82063591e-01 -4.44713533e-02 1.09151506e+00 -2.95066595e-01 -3.41988415e-01 7.63261139e-01 -1.65494859e-01 -2.45873779e-01 2.12046131e-01 1.64260894e-01 4.87361729e-01 3.05174291e-02 5.69940150e-01 -2.80722558e-01 -6.35442793e-01 4.74692881e-01 -3.08518082e-01 2.30194092e-01 5.29742122e-01 6.94483936e-01 7.80512929e-01 -2.60117650e-01 -1.00374527e-01 -8.72928917e-01 -6.68466315e-02 -4.43309210e-02 -4.15334553e-01 -2.84606293e-02 2.24107709e-02 9.13383365e-02 2.68913388e-01 -4.41166133e-01 -1.01051271e+00 1.06625485e+00 -2.15650015e-02 -6.77679300e-01 -4.61414665e-01 1.20597139e-01 -3.50566179e-01 2.05199406e-01 8.48549008e-01 -1.46356121e-01 1.72980145e-01 -5.77329874e-01 4.47706223e-01 3.14988792e-01 8.19796085e-01 -6.28033519e-01 1.01430559e+00 6.25245571e-01 1.51904866e-01 -8.16416204e-01 -1.05031955e+00 -9.27622259e-01 -1.45204449e+00 -4.22528833e-01 1.26636624e+00 -1.36356783e+00 -4.52444792e-01 5.50336421e-01 -1.25662518e+00 -3.60540897e-01 -4.95523103e-02 9.04243231e-01 -9.17599857e-01 5.46363354e-01 -4.58925277e-01 -7.59670913e-01 -5.24952374e-02 -1.18854475e+00 1.52382386e+00 -2.36883327e-01 -7.22322345e-01 -8.03806782e-01 4.66243804e-01 5.59841931e-01 -2.09246576e-01 5.10568857e-01 -1.45751256e-02 -3.14712405e-01 -4.15056258e-01 -5.60710490e-01 3.54544252e-01 3.35192233e-01 -1.44006371e-01 -4.67744172e-01 -7.18946755e-01 -3.80778342e-01 -1.80844322e-01 -6.30923986e-01 5.85743546e-01 4.55304056e-01 7.36462057e-01 6.26972541e-02 -3.34625006e-01 5.73687255e-01 9.63423014e-01 -7.47447491e-01 4.31145877e-01 1.94058180e-01 1.11023009e+00 8.91171515e-01 7.20617533e-01 5.93395293e-01 6.21996522e-01 9.13320601e-01 2.49379531e-01 6.10547960e-02 7.26800598e-03 -6.60279155e-01 4.24515307e-01 8.01162720e-01 -3.85297745e-01 4.16908972e-02 -1.01517570e+00 3.95732820e-01 -1.99242604e+00 -6.63271189e-01 -2.21617948e-02 2.24236965e+00 6.55082285e-01 1.43567711e-01 4.99786884e-01 6.55694827e-02 6.77031875e-01 8.96861479e-02 -5.58475733e-01 5.99112570e-01 1.91686600e-01 -1.14807524e-01 6.04212046e-01 5.39581060e-01 -1.44831038e+00 9.35430408e-01 5.97167969e+00 9.85877663e-02 -8.16746116e-01 -5.85727841e-02 2.81045705e-01 -4.51351106e-01 3.65551919e-01 -1.28911240e-02 -9.90269542e-01 2.08791941e-01 4.97734368e-01 3.51843059e-01 1.20008431e-01 1.04541111e+00 1.22040085e-01 -1.98220745e-01 -1.37647867e+00 1.46104455e+00 3.90475541e-01 -6.76989019e-01 -2.99394190e-01 2.83388376e-01 8.44285667e-01 -7.16520995e-02 -2.67063469e-01 2.01983035e-01 1.27708778e-01 -8.57906818e-01 8.03729057e-01 6.56620681e-01 6.08375072e-01 -6.83825374e-01 6.61371112e-01 7.89502561e-01 -1.13200855e+00 3.35158288e-01 -3.34483415e-01 -1.79329246e-01 5.25698662e-01 5.02417207e-01 -7.48192966e-01 2.85991400e-01 7.25695550e-01 9.07871306e-01 -6.86952531e-01 9.41543341e-01 -5.89500964e-01 2.71179557e-01 -6.44435644e-01 2.13123783e-01 -1.20963365e-01 -1.25217855e-01 5.47622383e-01 1.07492375e+00 3.48848313e-01 -4.30757292e-02 7.59893894e-01 4.29094881e-01 -3.99771221e-02 2.42782105e-02 -6.28594398e-01 2.34470874e-01 1.61815584e-01 1.31857133e+00 -7.16145396e-01 -1.12270087e-01 -3.32370102e-01 1.26523399e+00 3.01312506e-01 1.41337723e-01 -8.04472506e-01 8.72695521e-02 4.67431635e-01 2.95289576e-01 2.49205247e-01 -7.38485754e-01 7.74554163e-02 -1.58169842e+00 1.95397645e-01 -7.75056958e-01 3.20317000e-01 -6.41236961e-01 -1.19025195e+00 2.59946436e-01 3.33610147e-01 -1.41414654e+00 -7.30698764e-01 -8.32886159e-01 -1.06883347e-01 6.48491859e-01 -9.39773202e-01 -1.14413893e+00 -4.73316401e-01 6.31150067e-01 4.32003498e-01 1.51494108e-02 7.07800090e-01 2.27303162e-01 -3.94365072e-01 4.40556288e-01 -4.44143176e-01 4.75617170e-01 1.23645973e+00 -1.40109229e+00 4.63979393e-01 8.42177749e-01 3.69985074e-01 5.47229707e-01 8.02117825e-01 -7.03326166e-01 -1.55303407e+00 -8.51277411e-01 8.87487650e-01 -1.06882381e+00 2.29423672e-01 -7.07470536e-01 -4.99966532e-01 8.81790161e-01 -2.15150073e-01 4.76380140e-01 6.70923352e-01 1.05472468e-01 -1.61891341e-01 2.40619376e-01 -8.95530045e-01 5.09951711e-01 1.32575774e+00 -4.95119214e-01 -5.80273449e-01 4.87434417e-01 2.56648779e-01 -8.59518468e-01 -8.87804568e-01 2.41607815e-01 5.09204984e-01 -7.77866781e-01 1.11762547e+00 -5.31132042e-01 2.96330154e-01 -5.25825977e-01 -2.64236510e-01 -1.22299612e+00 5.44037819e-02 -6.51707768e-01 -2.94474214e-01 7.22116709e-01 8.96576643e-02 5.83429523e-02 1.47652733e+00 6.86040044e-01 3.42594802e-01 -4.90522534e-01 -6.83475256e-01 -6.61845684e-01 -1.11227147e-02 -4.94613588e-01 2.11270452e-01 4.94963706e-01 -1.36564255e-01 3.99277955e-01 -9.72232759e-01 1.52270138e-01 9.97594535e-01 -2.56750941e-01 1.64049852e+00 -1.19888401e+00 -4.01728868e-01 2.29109019e-01 -7.43982732e-01 -1.61808944e+00 3.04839075e-01 -5.46535671e-01 3.66218239e-01 -1.28317928e+00 3.13266695e-01 3.11475426e-01 3.48129213e-01 2.85825044e-01 -1.40971869e-01 3.53925169e-01 2.82623917e-01 3.34818065e-01 -9.53358352e-01 6.43585324e-01 1.17893875e+00 9.30422843e-02 1.62062384e-02 -7.50731528e-02 -8.84692594e-02 1.14636970e+00 4.34622914e-01 -3.23653013e-01 -4.53670919e-01 -3.78500879e-01 1.57685176e-01 4.21365164e-02 5.98178089e-01 -1.32489407e+00 4.17951912e-01 -3.14383470e-02 7.43322253e-01 -1.00314939e+00 6.86469615e-01 -7.57691085e-01 1.83397874e-01 4.20828909e-01 -2.87135243e-01 6.14401512e-02 -2.96238184e-01 7.86254466e-01 -4.26977091e-02 3.86854149e-02 5.89762568e-01 -4.71165210e-01 -8.33414555e-01 5.72642207e-01 -1.23404860e-01 3.77702713e-01 8.84392917e-01 -1.04456797e-01 3.40151191e-01 -6.04489625e-01 -7.72179127e-01 2.06044525e-01 8.85111809e-01 4.21762437e-01 4.84029710e-01 -1.52089846e+00 -5.68130612e-01 4.03789468e-02 2.55790293e-01 5.93102932e-01 1.26753956e-01 7.62025774e-01 -8.07592571e-01 4.22884345e-01 -3.18170369e-01 -1.01265931e+00 -1.11148131e+00 3.74690205e-01 2.44314015e-01 -3.89733016e-01 -6.79414093e-01 6.70702100e-01 7.83692449e-02 -9.79426384e-01 3.82555515e-01 1.19587734e-01 1.42492208e-04 -9.86297056e-02 3.79535317e-01 4.12712753e-01 -1.14212610e-01 -1.22905231e+00 -5.23611486e-01 8.32357883e-01 1.86663345e-01 -4.27451670e-01 1.24268937e+00 -2.10201010e-01 1.70143917e-01 3.73559922e-01 1.48471189e+00 -8.85119885e-02 -1.82544577e+00 -2.77662843e-01 -1.21380113e-01 -5.99735916e-01 -3.16843420e-01 -3.01335931e-01 -9.70698297e-01 9.98333454e-01 4.47936803e-01 -6.45287573e-01 5.80078959e-01 -2.34080888e-02 8.06017339e-01 7.23774970e-01 6.10766888e-01 -1.48155487e+00 6.15058780e-01 5.49972594e-01 7.76001930e-01 -1.47847903e+00 3.11891377e-01 -4.82235849e-01 -8.54819477e-01 9.95678723e-01 8.34927857e-01 -2.54631668e-01 5.53603947e-01 7.04590455e-02 1.47310212e-01 -9.86082330e-02 -2.12254554e-01 -2.58027107e-01 5.66336095e-01 6.56116545e-01 4.38088745e-01 2.41682027e-02 -4.09738421e-02 3.18460882e-01 -4.69032913e-01 -2.17860788e-02 3.31300318e-01 1.05183637e+00 -2.26182178e-01 -1.10468423e+00 -7.13335574e-01 -1.38182822e-03 -3.63640130e-01 3.84065747e-01 -4.95682746e-01 7.21284389e-01 7.74049163e-02 6.04481876e-01 -5.06282449e-02 -3.90148222e-01 5.09337604e-01 1.69975385e-01 7.40634620e-01 -7.19854891e-01 2.05668546e-02 2.70742446e-01 -6.35858402e-02 -7.37209678e-01 -5.89702904e-01 -9.67261434e-01 -1.22932088e+00 -7.03604296e-02 -3.48567180e-02 -6.74319118e-02 5.64969778e-01 9.85940874e-01 9.51330438e-02 8.89893323e-02 1.45763129e-01 -1.55109572e+00 -5.12706280e-01 -8.68224800e-01 -4.12606210e-01 8.52979600e-01 1.14565574e-01 -1.09133887e+00 -2.44899988e-01 3.82410228e-01]
[7.018719673156738, -0.9516367316246033]
582e6961-aebc-487d-985b-d0fce2f2150e
neuroninspect-detecting-backdoors-in-neural
1911.07399
null
https://arxiv.org/abs/1911.07399v1
https://arxiv.org/pdf/1911.07399v1.pdf
NeuronInspect: Detecting Backdoors in Neural Networks via Output Explanations
Deep neural networks have achieved state-of-the-art performance on various tasks. However, lack of interpretability and transparency makes it easier for malicious attackers to inject trojan backdoor into the neural networks, which will make the model behave abnormally when a backdoor sample with a specific trigger is input. In this paper, we propose NeuronInspect, a framework to detect trojan backdoors in deep neural networks via output explanation techniques. NeuronInspect first identifies the existence of backdoor attack targets by generating the explanation heatmap of the output layer. We observe that generated heatmaps from clean and backdoored models have different characteristics. Therefore we extract features that measure the attributes of explanations from an attacked model namely: sparse, smooth and persistent. We combine these features and use outlier detection to figure out the outliers, which is the set of attack targets. We demonstrate the effectiveness and efficiency of NeuronInspect on MNIST digit recognition dataset and GTSRB traffic sign recognition dataset. We extensively evaluate NeuronInspect on different attack scenarios and prove better robustness and effectiveness over state-of-the-art trojan backdoor detection techniques Neural Cleanse by a great margin.
['Xijie Huang', 'Moustafa Alzantot', 'Mani Srivastava']
2019-11-18
null
null
null
null
['traffic-sign-recognition']
['computer-vision']
[ 9.13750008e-02 -1.77872241e-01 -1.72271222e-01 -2.26123512e-01 -3.15983772e-01 -8.07111382e-01 6.38121784e-01 -2.20628679e-02 -6.37624487e-02 3.33617955e-01 -1.38380930e-01 -8.55643153e-01 -7.72784203e-02 -5.61658800e-01 -1.12853277e+00 -8.15913200e-01 -1.19670860e-01 -2.25305989e-01 1.19397938e-01 -1.77170262e-02 4.73582566e-01 6.89807713e-01 -1.31600606e+00 6.60179019e-01 6.99047804e-01 8.07661176e-01 -7.11243033e-01 6.94610298e-01 1.89575441e-02 9.27577674e-01 -1.13962626e+00 -4.31878418e-01 4.79175419e-01 -2.86648422e-01 -1.58737436e-01 -5.30732214e-01 9.72613811e-01 -3.72298568e-01 -6.78359210e-01 1.45840240e+00 2.12040067e-01 -4.31891501e-01 5.40473521e-01 -1.88834083e+00 -6.14419460e-01 7.80584574e-01 -4.97426897e-01 5.22878468e-01 -1.16892554e-01 5.52359343e-01 7.17745960e-01 -5.96961558e-01 5.50361931e-01 1.15584719e+00 7.29598880e-01 8.08150589e-01 -1.17316234e+00 -1.33611202e+00 2.53085017e-01 2.06344977e-01 -8.34962487e-01 -5.69120884e-01 7.48549044e-01 -3.47635388e-01 7.07833290e-01 3.61050069e-01 2.63844669e-01 1.81162179e+00 6.47287011e-01 7.31422484e-01 7.26912200e-01 8.10922182e-04 4.98726785e-01 -2.67758053e-02 8.96959066e-01 8.78791749e-01 8.25997889e-01 5.11879146e-01 -6.51435077e-01 -4.53381360e-01 3.21719706e-01 5.26478112e-01 -2.68912584e-01 -4.73836921e-02 -8.24465990e-01 8.21578085e-01 5.91836452e-01 -3.98155712e-02 -2.80756086e-01 5.16753316e-01 6.71359777e-01 5.60700834e-01 -1.76740408e-01 5.54856479e-01 -4.24179643e-01 1.21819181e-02 -8.17185819e-01 2.05615371e-01 6.89779997e-01 6.15788937e-01 3.99553627e-01 6.79560721e-01 -8.55109617e-02 2.39817813e-01 2.98825234e-01 4.18991685e-01 6.17969096e-01 -2.73739636e-01 6.17987037e-01 8.23763907e-01 -7.07101047e-01 -1.07325172e+00 -7.33339563e-02 -5.99190712e-01 -8.12080204e-01 6.08892143e-01 4.27116364e-01 -2.19903529e-01 -1.44874084e+00 1.59275031e+00 2.19414532e-02 5.07194102e-01 1.19511701e-01 6.20965719e-01 7.06275344e-01 4.46036458e-01 -2.46639982e-01 4.81800407e-01 1.20152450e+00 -8.02927554e-01 -5.47499776e-01 -1.83250651e-01 8.21140409e-01 -1.37202129e-01 9.85291660e-01 6.76148593e-01 -3.27476948e-01 -1.81070909e-01 -1.49240243e+00 4.59006310e-01 -7.22680390e-01 -1.42853305e-01 6.75722897e-01 1.21442235e+00 -3.19866538e-01 7.83219516e-01 -1.11348832e+00 8.35289657e-02 8.04720998e-01 3.85875672e-01 -6.17906749e-01 3.31536122e-02 -8.65654647e-01 3.47221732e-01 4.26040500e-01 2.84723770e-02 -1.56592679e+00 -7.79598832e-01 -7.18716383e-01 4.83174026e-02 2.77253538e-01 -2.02507690e-01 8.51818264e-01 -7.00045943e-01 -9.25194144e-01 2.78296411e-01 1.44191012e-01 -8.44157457e-01 5.14414251e-01 -2.38923475e-01 -6.53885543e-01 -2.12290525e-01 -1.63429156e-01 -7.20096333e-03 1.23986447e+00 -1.18202150e+00 -4.55491364e-01 -6.11573160e-01 -3.60929757e-01 -7.00837195e-01 -5.08388638e-01 -6.56583384e-02 -2.03829538e-02 -6.19087934e-01 1.08225040e-01 -7.68677771e-01 4.83358912e-02 -2.72716671e-01 -1.23849082e+00 3.23620379e-01 1.25155103e+00 -4.57533896e-01 1.27956498e+00 -2.30472779e+00 -5.13592780e-01 6.88672602e-01 6.10402882e-01 6.51399076e-01 -2.33029425e-01 1.76605165e-01 -5.24121821e-01 3.51431876e-01 -6.70448244e-02 -1.94282591e-01 5.93177117e-02 2.30086342e-01 -1.15422404e+00 6.91428185e-01 8.55270401e-02 7.48862803e-01 -5.68309069e-01 2.08387360e-01 -7.61051998e-02 2.75412768e-01 -6.05411470e-01 -1.18905582e-01 -1.38169318e-01 1.32290367e-02 -4.51590806e-01 1.01981115e+00 6.49538577e-01 3.52940440e-01 -4.31332350e-01 1.66482646e-02 1.81092918e-01 2.04400808e-01 -9.69491124e-01 8.43720496e-01 5.63631617e-02 1.05805409e+00 -3.73405576e-01 -4.55818057e-01 1.04757237e+00 2.52810512e-02 -2.00311184e-01 -5.30360579e-01 4.00329471e-01 3.04661781e-01 2.95863509e-01 -2.90953696e-01 2.84068853e-01 4.10648316e-01 -5.86395003e-02 5.36351264e-01 3.71069796e-02 7.82238960e-01 -1.34977192e-01 3.25251788e-01 1.75672412e+00 -5.17946959e-01 -8.51489902e-02 3.50115821e-02 1.85922161e-01 -1.10102914e-01 5.69348395e-01 1.25733638e+00 -3.36452067e-01 3.56772453e-01 1.10718858e+00 -7.46232748e-01 -7.96436667e-01 -1.25272560e+00 5.95183410e-02 8.03903401e-01 -2.67464191e-01 -3.83623540e-01 -9.73799884e-01 -1.17996585e+00 3.29194158e-01 8.89385104e-01 -9.62247849e-01 -7.48257518e-01 -4.86374497e-01 -6.52586401e-01 1.45965481e+00 3.47691923e-01 5.51634550e-01 -1.01936650e+00 -7.01968193e-01 -1.49033099e-01 4.84401375e-01 -1.05244613e+00 -2.62740940e-01 4.98963892e-01 -9.38225627e-01 -1.30561471e+00 -8.00846028e-04 -2.76399791e-01 8.78720999e-01 -5.49786352e-03 5.46400309e-01 1.89516425e-01 -5.20277441e-01 -2.44518206e-01 -1.19082101e-01 -7.13164270e-01 -6.66680753e-01 -1.81070194e-02 4.12884414e-01 1.71476677e-01 4.60384905e-01 -5.15970290e-01 -2.93777466e-01 1.70083195e-01 -1.14358747e+00 -5.05551875e-01 6.27756536e-01 6.85483217e-01 2.39905611e-01 2.79433280e-01 1.62165806e-01 -1.07202244e+00 9.59906042e-01 -4.87134069e-01 -6.38053954e-01 2.87648477e-02 -5.13052821e-01 4.68662262e-01 1.07261252e+00 -7.36060977e-01 -3.32988173e-01 -4.40813638e-02 5.03273904e-02 -8.94847512e-01 -4.21603113e-01 1.43550515e-01 -4.19005215e-01 -1.14546984e-01 1.14469206e+00 2.90162623e-01 -7.65421093e-02 -4.14580286e-01 2.22114250e-01 4.06190097e-01 9.05917227e-01 -1.79490492e-01 1.28323662e+00 5.39894700e-01 8.15831050e-02 -7.86905229e-01 -2.94671744e-01 -4.10365686e-02 -6.32063597e-02 -4.58557755e-02 3.68449390e-01 -3.66094470e-01 -8.66074562e-01 7.50144541e-01 -1.17181528e+00 -1.32217795e-01 1.12733699e-01 3.33609506e-02 1.81853652e-01 2.61848301e-01 -4.40980703e-01 -8.15883875e-01 -5.59357345e-01 -1.36502528e+00 5.94814181e-01 7.85823539e-02 -9.47088301e-02 -6.39903545e-01 -1.36601804e-02 -1.42170191e-01 1.69014528e-01 7.97672093e-01 1.01032710e+00 -1.53426337e+00 -7.70199299e-01 -7.08355606e-01 2.42571086e-02 4.07400727e-01 -5.95695153e-02 4.17775422e-01 -1.39503813e+00 -1.92632169e-01 -2.16000620e-02 1.35465249e-01 1.24676323e+00 1.05918035e-01 1.22931492e+00 -6.88145280e-01 -4.76007164e-01 1.06236935e+00 1.19395828e+00 2.55613089e-01 7.96714127e-01 5.72915196e-01 9.93656635e-01 2.16573432e-01 -6.06120676e-02 8.19850639e-02 -4.44038779e-01 3.28921407e-01 1.04079330e+00 1.18194893e-01 3.74211192e-01 -5.27019799e-01 8.94619465e-01 3.86723727e-02 4.39408481e-01 -2.07469344e-01 -9.53315437e-01 3.54364872e-01 -1.72995877e+00 -1.01633751e+00 -2.95706779e-01 2.14387441e+00 3.42661798e-01 6.32283390e-01 -9.66583751e-03 5.22715390e-01 5.69830298e-01 8.10060278e-02 -7.43762672e-01 -7.15550125e-01 -2.49677170e-02 7.92291835e-02 9.16762054e-01 1.77824035e-01 -9.80052531e-01 8.81895840e-01 5.45616865e+00 8.30434978e-01 -1.23937893e+00 -1.81427598e-01 5.71852088e-01 -2.29562372e-01 -7.86398053e-02 -1.02607459e-01 -9.39487398e-01 6.22090638e-01 1.02885973e+00 4.14938927e-01 4.01361585e-01 1.06074059e+00 -1.08992718e-02 5.10660589e-01 -1.35355425e+00 8.01496267e-01 1.67295650e-01 -1.38635767e+00 2.61925966e-01 5.33276618e-01 4.74883735e-01 1.92380697e-01 4.93700624e-01 3.04412097e-01 4.45557892e-01 -1.18619227e+00 7.94045448e-01 3.76232028e-01 3.38783443e-01 -9.63739038e-01 6.70599222e-01 2.32503429e-01 -6.92538559e-01 -4.98596698e-01 -3.26464981e-01 2.46028751e-01 -2.89103270e-01 6.09459758e-01 -9.54552293e-01 2.16624424e-01 6.50575638e-01 3.22410882e-01 -8.74751329e-01 9.91110027e-01 -4.46766675e-01 9.12915289e-01 -3.15471500e-01 -4.28559706e-02 4.76672083e-01 1.23888738e-01 8.00538123e-01 1.00622368e+00 1.60619631e-01 -6.52860105e-01 -4.78325248e-01 1.23431873e+00 -3.00818712e-01 -3.78240764e-01 -1.10447788e+00 -3.17150980e-01 6.89070880e-01 1.06418192e+00 -6.61322117e-01 -1.41374364e-01 1.77855641e-01 8.27082992e-01 1.12935752e-01 3.68367493e-01 -1.05292308e+00 -7.62778640e-01 1.21000254e+00 -2.00911745e-01 1.74309626e-01 -2.38760766e-02 -5.87875605e-01 -9.98015761e-01 1.27603710e-01 -1.17856491e+00 3.49408388e-01 -3.62184376e-01 -1.11095893e+00 7.06387460e-01 -3.64963621e-01 -1.07987010e+00 -7.68635198e-02 -1.08436012e+00 -1.32266498e+00 3.84081572e-01 -9.85409915e-01 -8.06300163e-01 -3.02699834e-01 6.27639651e-01 2.70403802e-01 -8.21316302e-01 4.95602190e-01 9.58682746e-02 -1.22475910e+00 1.18403482e+00 1.40416384e-01 7.37038732e-01 4.59530652e-01 -1.02756965e+00 1.13091683e+00 1.37271571e+00 4.96270418e-01 1.12603283e+00 8.46185446e-01 -8.34973216e-01 -1.50925541e+00 -1.38896596e+00 2.89627850e-01 -7.85043001e-01 8.33481491e-01 -6.46845102e-01 -1.01762950e+00 7.82031178e-01 -2.17379421e-01 4.61208299e-02 4.78332222e-01 -6.14872202e-02 -1.00955653e+00 -1.50490627e-01 -1.17270744e+00 1.02897561e+00 7.56492794e-01 -4.40034777e-01 -4.94511545e-01 -1.20336469e-02 7.35136032e-01 -1.74619570e-01 -1.16205767e-01 8.85980800e-02 6.24324918e-01 -1.17481005e+00 9.08584952e-01 -9.88339663e-01 3.11059237e-01 -3.28874230e-01 -4.06754948e-02 -1.13658631e+00 1.64334983e-01 -8.68397951e-01 -5.55149317e-01 1.05251288e+00 6.04443312e-01 -1.01796484e+00 1.07996941e+00 3.50499868e-01 -7.19682425e-02 -5.85821748e-01 -9.41986442e-01 -1.04922152e+00 -1.52459545e-02 -7.56148100e-01 8.80502522e-01 9.41785991e-01 -1.79033443e-01 -2.40471847e-02 -2.11547822e-01 6.22740209e-01 9.84129190e-01 -4.98070180e-01 1.08991587e+00 -1.10679507e+00 -1.05459109e-01 -5.25665820e-01 -7.79010296e-01 -4.19580638e-01 1.77334130e-01 -8.81371856e-01 -2.13272601e-01 -6.93832517e-01 -1.58532456e-01 6.65908726e-03 -5.41642785e-01 8.63412797e-01 8.80016461e-02 3.03719580e-01 2.07589850e-01 1.77156016e-01 -2.51203746e-01 2.72204965e-01 4.06406552e-01 -2.84155250e-01 -3.50927651e-01 -1.45550231e-02 -6.66222811e-01 7.33464122e-01 9.60823774e-01 -1.01761615e+00 -2.08118603e-01 -3.15041721e-01 -1.19182533e-02 -6.33005798e-01 8.03898215e-01 -1.27908576e+00 2.35703155e-01 2.15717375e-01 3.24502319e-01 -5.58873951e-01 1.35155898e-02 -8.55440915e-01 4.50654663e-02 8.64955187e-01 -4.17505622e-01 3.26559842e-01 4.40531284e-01 7.78494358e-01 1.75307482e-01 -2.54295796e-01 5.48970222e-01 2.94899315e-01 -4.86046374e-01 4.26255196e-01 -4.98508215e-01 -2.23118573e-01 8.81472230e-01 -5.42205930e-01 -8.81371915e-01 -6.41065016e-02 -2.55389333e-01 -3.58709730e-02 2.93019503e-01 5.85690856e-01 1.06942165e+00 -1.14021468e+00 -3.09062600e-01 9.01589155e-01 3.19234341e-01 -4.89830554e-01 -8.17407519e-02 5.90233803e-01 -5.83676338e-01 3.47073555e-01 -3.36450785e-01 -4.79749084e-01 -1.63271666e+00 4.79860783e-01 5.09360969e-01 6.14160821e-02 -7.65749633e-01 5.93065679e-01 3.61524336e-02 -4.74533111e-01 5.27250886e-01 -6.74390197e-01 1.12663716e-01 -3.42985481e-01 7.16088533e-01 4.43945944e-01 1.31749257e-01 -2.42433503e-01 -4.65977758e-01 3.37794609e-02 -4.76038963e-01 6.78769276e-02 1.06130648e+00 7.14506328e-01 3.61731462e-02 7.52248429e-03 1.28902996e+00 2.42811721e-02 -1.32569313e+00 1.07799657e-01 2.01970473e-01 -4.62166548e-01 -1.65024307e-02 -7.25500941e-01 -1.27549481e+00 9.76054966e-01 6.92095160e-01 2.31047601e-01 7.48312831e-01 -4.14904803e-01 8.41139078e-01 8.05675685e-01 -9.46738422e-02 -5.40525973e-01 1.52583420e-01 5.55268109e-01 4.94114578e-01 -7.46626318e-01 -2.66398460e-01 9.40228477e-02 -4.73413020e-01 1.13914430e+00 6.51303828e-01 -3.86488259e-01 4.31656331e-01 3.61990005e-01 2.15616509e-01 -5.06547809e-01 -8.11242998e-01 4.34181660e-01 1.54491901e-01 6.03683233e-01 -3.84186685e-01 -3.95042971e-02 5.72036743e-01 6.49681509e-01 -4.51040119e-01 -5.50324440e-01 6.95338607e-01 7.70209014e-01 -5.57563305e-01 -6.50887966e-01 -6.65246904e-01 5.59066474e-01 -5.58180273e-01 -2.44812340e-01 -8.60868394e-01 7.82352865e-01 -3.11081950e-02 7.81877398e-01 -1.29253492e-01 -9.82268631e-01 4.38656449e-01 2.62888521e-01 -1.14986278e-01 -2.98331708e-01 -8.34068894e-01 -5.08937240e-01 -1.60351902e-01 -7.96683311e-01 7.45395899e-01 -3.41745496e-01 -1.29839385e+00 -4.79688048e-01 -4.00183171e-01 -1.21370755e-01 1.08520532e+00 9.05668974e-01 5.18575311e-01 7.10178077e-01 5.65490067e-01 -4.03180391e-01 -7.97826350e-01 -6.79654837e-01 -4.12335783e-01 4.57522243e-01 8.07283401e-01 -5.00772238e-01 -7.81521738e-01 -2.49765694e-01]
[5.712527275085449, 7.744675159454346]
0b86aa19-fa9a-4093-8b54-8939814d6cad
safe-distributional-reinforcement-learning
2102.13446
null
https://arxiv.org/abs/2102.13446v1
https://arxiv.org/pdf/2102.13446v1.pdf
Safe Distributional Reinforcement Learning
Safety in reinforcement learning (RL) is a key property in both training and execution in many domains such as autonomous driving or finance. In this paper, we formalize it with a constrained RL formulation in the distributional RL setting. Our general model accepts various definitions of safety(e.g., bounds on expected performance, CVaR, variance, or probability of reaching bad states). To ensure safety during learning, we extend a safe policy optimization method to solve our problem. The distributional RL perspective leads to a more efficient algorithm while additionally catering for natural safe constraints. We empirically validate our propositions on artificial and real domains against appropriate state-of-the-art safe RL algorithms.
['Paul Weng', 'Jianyi Zhang']
2021-02-26
null
null
null
null
['distributional-reinforcement-learning']
['methodology']
[-6.21690117e-02 4.24771786e-01 -4.81295854e-01 -2.26854473e-01 -8.11249554e-01 -7.64069617e-01 7.11710632e-01 3.40077765e-02 -7.80415118e-01 1.23524857e+00 -2.78590024e-02 -5.82927704e-01 -3.47406358e-01 -7.37486124e-01 -8.97243679e-01 -7.63304949e-01 -4.14842993e-01 1.87525004e-02 3.05574715e-01 -2.52940178e-01 1.34605899e-01 4.84311998e-01 -1.52727079e+00 -4.32197630e-01 7.35834479e-01 9.35099125e-01 -2.83238113e-01 4.99789685e-01 5.24628222e-01 1.07936895e+00 -5.33098459e-01 -1.27741858e-01 3.60643804e-01 -2.56785601e-01 -7.30538905e-01 -3.27507138e-01 -1.31048337e-01 -4.07290906e-01 -1.55635223e-01 1.10492504e+00 3.52000326e-01 4.40985858e-01 5.93268216e-01 -1.80371583e+00 -9.74595994e-02 8.19773853e-01 -2.70021886e-01 -1.39190406e-01 2.34116390e-01 5.43339670e-01 9.38880205e-01 -3.93443182e-02 2.70013601e-01 1.21449447e+00 2.65597731e-01 9.84943390e-01 -1.21617532e+00 -4.91938084e-01 4.25793201e-01 -1.98297739e-01 -1.08019888e+00 -3.72643799e-01 2.95941740e-01 -5.01935899e-01 8.69606495e-01 2.35974193e-02 3.71220469e-01 1.45238769e+00 6.66892409e-01 9.37325120e-01 1.13581896e+00 -3.41082007e-01 8.25146258e-01 8.42455551e-02 5.87085560e-02 4.77428734e-01 4.61969376e-01 9.48771298e-01 -2.78000593e-01 -2.73640752e-01 4.56574678e-01 -4.89017397e-01 1.52889211e-02 -6.69732869e-01 -8.49644184e-01 1.15885234e+00 -1.60400555e-01 -3.13801587e-01 -1.49536818e-01 5.90030134e-01 7.72362232e-01 4.69348937e-01 1.52233392e-01 5.68948925e-01 -4.35509831e-01 -4.05507386e-01 -4.39728498e-01 8.20759535e-01 7.96191871e-01 1.09006655e+00 4.03180450e-01 3.41728419e-01 -3.45333070e-01 1.97718024e-01 4.03218269e-01 4.18932676e-01 2.14555323e-01 -1.33236170e+00 3.23667288e-01 -7.58867860e-02 6.00704134e-01 -2.37523913e-01 -4.34595674e-01 -2.71485686e-01 -2.87638396e-01 6.51173651e-01 3.18940967e-01 -4.96612877e-01 -3.78547251e-01 2.27853584e+00 2.74005324e-01 3.31339449e-01 4.44283754e-01 5.99399030e-01 -1.34602219e-01 4.57907677e-01 2.99500018e-01 -6.97977722e-01 7.40851879e-01 -6.90448999e-01 -6.50249958e-01 -2.40254849e-01 6.88005209e-01 6.19838350e-02 1.38610780e+00 9.19478834e-01 -9.25981045e-01 -2.04720944e-01 -1.09500468e+00 3.77850711e-01 -2.42512785e-02 -4.84352380e-01 5.07525563e-01 8.39073956e-01 -8.08998346e-01 7.39048243e-01 -1.01255476e+00 -4.06352542e-02 1.68238223e-01 4.21889931e-01 -6.41505048e-02 3.58833998e-01 -1.32805192e+00 1.15571296e+00 5.59082985e-01 -3.48621219e-01 -1.70493186e+00 -3.60724390e-01 -1.05236661e+00 -2.87925541e-01 9.45310652e-01 -3.89878869e-01 1.87792134e+00 -5.90385616e-01 -1.96514189e+00 4.21845138e-01 4.15593743e-01 -9.56195116e-01 8.24003577e-01 -6.43007994e-01 -1.90611258e-01 -2.82212943e-01 -4.09671813e-02 3.97238225e-01 9.57307458e-01 -1.18319690e+00 -7.59857714e-01 -9.51838568e-02 4.82204676e-01 4.51547792e-03 -2.24728342e-02 -1.01073161e-01 2.60337800e-01 -3.84859651e-01 -8.06805968e-01 -1.17983854e+00 -3.43862742e-01 -4.54952151e-01 -1.60266683e-01 -4.57540274e-01 5.75280249e-01 -3.55355442e-02 1.24355292e+00 -2.15698552e+00 2.49923602e-01 2.27329165e-01 -1.81009799e-01 -2.36286074e-02 -1.49971008e-01 3.37799191e-01 6.93499371e-02 -2.87902243e-02 -3.22055697e-01 -1.85473546e-01 6.30653441e-01 6.83955491e-01 -8.31805468e-01 9.12833095e-01 1.24943838e-01 7.35290289e-01 -1.06132758e+00 -4.40933436e-01 8.62959176e-02 -5.31294942e-03 -8.04209948e-01 4.89517540e-01 -5.97463906e-01 5.02290428e-01 -5.58195651e-01 2.34234110e-01 1.51730865e-01 5.30242383e-01 1.20747328e-01 6.20218575e-01 -2.17980728e-01 2.17695311e-01 -1.14583540e+00 1.41454422e+00 -5.22460401e-01 1.15141667e-01 -1.26859555e-02 -8.35742235e-01 5.36742091e-01 1.31490335e-01 3.92779350e-01 -6.21855378e-01 2.68084049e-01 -1.88895389e-01 -1.03575334e-01 -3.67014974e-01 5.35014451e-01 -3.21525723e-01 -6.67797387e-01 5.40272117e-01 -8.91165286e-02 -3.26035529e-01 7.01633990e-02 -1.17512271e-01 9.69028413e-01 6.67149842e-01 4.87327784e-01 -5.86557627e-01 4.38482314e-01 -1.22230500e-01 6.50966644e-01 8.59913290e-01 -5.72424650e-01 -4.03907597e-01 1.05776227e+00 -8.42709169e-02 -7.66135991e-01 -1.06843853e+00 -1.10756829e-01 1.29356873e+00 4.62474860e-03 -3.77684802e-01 -8.62240016e-01 -9.71772313e-01 1.85753256e-01 1.32576787e+00 -6.45623863e-01 -7.50500917e-01 -4.50897545e-01 -1.89607739e-01 9.46648896e-01 5.46633840e-01 2.10600179e-02 -1.11095285e+00 -1.40736210e+00 6.11776859e-02 3.03810567e-01 -8.36624682e-01 -3.35843205e-01 6.03350759e-01 -4.72790241e-01 -9.18923259e-01 -1.05043538e-02 -2.75465697e-01 2.22124934e-01 -3.56523871e-01 1.14483821e+00 -2.69066453e-01 1.55696437e-01 5.86549759e-01 -1.95247918e-01 -5.65392554e-01 -5.97084939e-01 -1.56923011e-01 7.33192086e-01 -4.83144462e-01 -2.08446868e-02 -1.32537469e-01 -2.25147977e-01 2.40273401e-01 -9.96256351e-01 -4.28383410e-01 1.30074903e-01 7.73242474e-01 7.26362824e-01 3.69070649e-01 6.65002108e-01 -6.86806560e-01 1.18445575e+00 -4.83292073e-01 -1.31559336e+00 2.83955485e-01 -8.36828709e-01 6.31581724e-01 8.81694019e-01 -5.78452706e-01 -7.51770437e-01 9.17077661e-02 -3.41426693e-02 -3.95507693e-01 -2.03071147e-01 2.18162507e-01 -3.38963866e-01 1.01720273e-01 6.17006540e-01 1.43472075e-01 1.01913400e-01 1.63940191e-01 4.80621338e-01 3.21745187e-01 3.72883737e-01 -1.33734107e+00 7.21757472e-01 -7.63594508e-02 2.20567599e-01 -3.98829341e-01 -7.84859657e-01 1.77990884e-01 -2.39066750e-01 -3.11253667e-01 5.60362816e-01 -6.81079209e-01 -1.43809831e+00 4.47907560e-02 -5.26407719e-01 -1.09260464e+00 -7.34890103e-01 3.78327787e-01 -1.39671385e+00 1.83091864e-01 -3.25326890e-01 -1.37920356e+00 2.98572262e-03 -1.45905161e+00 9.90406036e-01 6.61424473e-02 -2.54583508e-01 -8.82294238e-01 4.60838050e-01 -4.59651589e-01 1.05837695e-01 4.55803663e-01 8.13371837e-01 -5.57306349e-01 -3.59333865e-02 2.60734975e-01 4.09838349e-01 5.65908074e-01 -2.20941305e-01 3.44708003e-02 -7.72508323e-01 -5.06950736e-01 2.36932874e-01 -8.52860630e-01 5.89293420e-01 3.47406030e-01 1.31990635e+00 -8.00928771e-01 7.51875713e-02 4.36406970e-01 1.43770838e+00 3.18595350e-01 3.84739369e-01 5.75184882e-01 2.59628624e-01 8.02714527e-01 1.33990407e+00 8.35471034e-01 1.51989594e-01 4.74022031e-01 8.09920728e-01 6.66557133e-01 7.77070999e-01 -4.42305297e-01 1.02039790e+00 -7.04875514e-02 3.36576998e-01 -2.12157220e-01 -8.34645867e-01 2.06284016e-01 -2.15047145e+00 -9.50846732e-01 3.90754759e-01 2.52633476e+00 1.08280075e+00 5.44450819e-01 6.31679833e-01 1.17537685e-01 2.16516867e-01 2.82628220e-02 -8.37925196e-01 -9.33069229e-01 3.24002117e-01 1.65566444e-01 8.96952033e-01 8.54480743e-01 -1.04129148e+00 1.06740427e+00 7.26408195e+00 8.86673331e-01 -9.06223595e-01 -1.41549364e-01 4.49371159e-01 -3.02811503e-01 -2.59334058e-01 -1.65941507e-01 -8.07611048e-01 3.95685136e-01 1.21054626e+00 -4.43271488e-01 8.15646708e-01 1.24831140e+00 3.21432203e-01 -2.17945218e-01 -1.43841624e+00 6.43344581e-01 -4.06063259e-01 -7.21993923e-01 -4.05162573e-01 9.51781496e-02 4.77882922e-01 -1.84697747e-01 2.84107655e-01 7.98848748e-01 1.01725042e+00 -1.32423806e+00 1.19183064e+00 2.67675787e-01 9.09752488e-01 -1.46970963e+00 3.95095915e-01 6.68580115e-01 -6.71474218e-01 -3.05972934e-01 -1.62469164e-01 -2.05645636e-01 -1.14649452e-01 9.24567878e-02 -3.83456439e-01 3.62007082e-01 2.56851315e-01 5.22444487e-01 -2.16372341e-01 5.28487384e-01 -6.81716681e-01 5.61186552e-01 -2.72377968e-01 -9.42005310e-03 5.74140310e-01 -1.07403211e-01 3.26568663e-01 9.74939466e-01 -1.88086531e-04 -2.06553981e-01 3.30741435e-01 7.12632537e-01 3.57878506e-01 -2.97093481e-01 -9.72590804e-01 5.68949990e-02 3.85423958e-01 9.76207495e-01 -3.11874926e-01 -6.90623596e-02 -1.25754848e-01 4.12689716e-01 3.71378750e-01 2.71659136e-01 -1.35245609e+00 -1.42968893e-01 1.08063269e+00 -3.18985820e-01 -2.53916327e-02 -4.55079883e-01 -1.87209457e-01 -8.52869630e-01 -3.04341257e-01 -1.12225878e+00 5.13321817e-01 -1.79155886e-01 -1.05999494e+00 4.45561707e-01 4.50454205e-01 -1.12157440e+00 -6.93743587e-01 -6.11716926e-01 -4.34565991e-01 4.88747507e-01 -1.57265067e+00 -5.69580972e-01 3.42934251e-01 8.86565924e-01 3.72271299e-01 -1.68288693e-01 7.08554745e-01 -2.93285221e-01 -8.42199504e-01 6.42914832e-01 -9.37912837e-02 -4.42190439e-01 6.32692933e-01 -1.45089877e+00 1.72886014e-01 9.59284484e-01 -3.55957359e-01 3.97508264e-01 1.12435293e+00 -4.31446701e-01 -1.53822196e+00 -1.29935348e+00 -1.59700774e-02 -5.20166814e-01 9.24096286e-01 -2.24768713e-01 -5.00507712e-01 8.78213346e-01 1.20982669e-01 -1.31755829e-01 3.29592854e-01 -9.06704962e-02 -2.69917339e-01 4.16568900e-03 -1.32888377e+00 8.83926630e-01 9.48628724e-01 -4.03640091e-01 -6.06980264e-01 1.47278383e-01 1.00440085e+00 -4.23683912e-01 -7.23900318e-01 3.68176728e-01 4.14844602e-01 -8.46136570e-01 6.61684573e-01 -8.76012802e-01 5.79789244e-02 -2.39795789e-01 -2.77940035e-01 -1.44336355e+00 -2.41625365e-02 -1.02583575e+00 -4.53707457e-01 8.44309032e-01 9.48757306e-02 -7.76957095e-01 5.38108647e-01 8.04180086e-01 -2.23198876e-01 -7.85860896e-01 -1.08024824e+00 -1.30850005e+00 5.61978221e-01 -8.80819142e-01 5.23347080e-01 5.28738916e-01 3.71479720e-01 -4.84065153e-02 -5.25702715e-01 2.13092372e-01 7.99710810e-01 -1.64265692e-01 6.29865766e-01 -8.29443157e-01 -4.84444171e-01 -5.62591553e-01 5.06012775e-02 -6.88341796e-01 1.07820332e+00 -5.16058624e-01 5.26315987e-01 -8.47278059e-01 -2.09622249e-01 -4.62415725e-01 -3.22211295e-01 7.28528559e-01 1.60271838e-01 -4.62765247e-01 2.38404814e-02 -3.30778360e-01 -9.63866770e-01 7.94648528e-01 9.08550024e-01 1.04129024e-01 -3.05405825e-01 1.13512717e-01 -6.36990130e-01 6.68898404e-01 1.21386600e+00 -5.42855680e-01 -9.08045053e-01 -1.09913927e-02 3.90520275e-01 2.94366837e-01 1.81369781e-01 -9.26347911e-01 -1.87536776e-01 -1.04908478e+00 -3.56428355e-01 -7.45827109e-02 -1.07497863e-01 -8.54594290e-01 -1.70579597e-01 8.85792494e-01 -8.27224731e-01 1.30745679e-01 3.36763799e-01 5.70269048e-01 6.89855218e-02 -2.29237452e-01 9.83114123e-01 1.40056357e-01 -6.01455510e-01 2.71180332e-01 -5.64586341e-01 4.09365386e-01 1.67908597e+00 1.37135252e-01 -2.99671918e-01 -3.31875682e-01 -4.46946412e-01 7.67015874e-01 4.75443780e-01 4.10205185e-01 6.02002203e-01 -1.05855775e+00 -3.63131762e-01 1.85175911e-01 1.88914254e-01 -1.13802768e-01 -2.36127600e-01 7.36591518e-01 -2.21729428e-01 4.84305412e-01 -3.03926200e-01 -2.52538264e-01 -9.12258446e-01 9.68942523e-01 4.11503345e-01 -3.58479112e-01 -4.87642795e-01 4.71750259e-01 2.45786980e-01 -2.74965793e-01 6.76962614e-01 -5.94022751e-01 -4.24003191e-02 -2.43698344e-01 3.84794712e-01 2.98561215e-01 -1.87629923e-01 -2.98555553e-01 -5.63973308e-01 1.50353745e-01 2.08082125e-01 -5.43229401e-01 1.02937281e+00 -1.62636675e-02 3.25511247e-01 4.68173116e-01 6.10769570e-01 1.25744399e-02 -1.77912259e+00 2.17108473e-01 2.55990326e-01 -1.46466613e-01 5.49071841e-02 -6.46003008e-01 -6.10045850e-01 5.89056909e-01 4.56192404e-01 2.34908238e-01 9.69639182e-01 -2.22360924e-01 3.15108746e-01 4.88524735e-01 9.08346057e-01 -1.28377843e+00 -6.59300387e-02 7.47500122e-01 8.20853472e-01 -1.01611304e+00 -8.77087787e-02 3.38914454e-01 -1.13986230e+00 7.32865989e-01 8.92482042e-01 -3.09893698e-01 4.60036606e-01 6.35633588e-01 -2.71880686e-01 3.61414850e-01 -1.16341543e+00 -2.39654809e-01 -2.58764207e-01 6.72330618e-01 8.21152478e-02 2.67214119e-01 -3.51016372e-01 7.31821716e-01 -1.76709160e-01 -5.85422739e-02 6.79108977e-01 1.23142493e+00 -5.36476314e-01 -1.17381632e+00 -2.09910735e-01 -1.50148971e-02 -5.12919784e-01 3.33884031e-01 -1.43084839e-01 9.72014487e-01 -8.27549547e-02 1.01821899e+00 -1.29199848e-01 -4.42572057e-01 3.65917206e-01 -3.86495469e-03 6.16218865e-01 -5.83675802e-01 -4.43772972e-01 -1.60705492e-01 1.51324481e-01 -8.64424646e-01 -3.11682791e-01 -5.90022743e-01 -1.52559423e+00 -3.58142436e-01 -2.43891478e-02 2.55278617e-01 3.53547394e-01 9.75556433e-01 -6.01574406e-02 5.58146596e-01 7.35934794e-01 -4.81371641e-01 -1.35293067e+00 -5.00216484e-01 -8.48429441e-01 2.08570100e-02 6.61161363e-01 -8.80277336e-01 -3.64233017e-01 -3.82940322e-01]
[4.459441184997559, 2.147900104522705]
0d522f4d-5258-4151-8e54-b38d76b3a7bf
localtrans-a-multiscale-local-transformer
2106.04067
null
https://arxiv.org/abs/2106.04067v2
https://arxiv.org/pdf/2106.04067v2.pdf
LocalTrans: A Multiscale Local Transformer Network for Cross-Resolution Homography Estimation
Cross-resolution image alignment is a key problem in multiscale gigapixel photography, which requires to estimate homography matrix using images with large resolution gap. Existing deep homography methods concatenate the input images or features, neglecting the explicit formulation of correspondences between them, which leads to degraded accuracy in cross-resolution challenges. In this paper, we consider the cross-resolution homography estimation as a multimodal problem, and propose a local transformer network embedded within a multiscale structure to explicitly learn correspondences between the multimodal inputs, namely, input images with different resolutions. The proposed local transformer adopts a local attention map specifically for each position in the feature. By combining the local transformer with the multiscale structure, the network is able to capture long-short range correspondences efficiently and accurately. Experiments on both the MS-COCO dataset and the real-captured cross-resolution dataset show that the proposed network outperforms existing state-of-the-art feature-based and deep-learning-based homography estimation methods, and is able to accurately align images under $10\times$ resolution gap.
['Yebin Liu', 'Ying Fu', 'Yuemei Zhou', 'Gaochang Wu', 'Ruizhi Shao']
2021-06-08
null
http://openaccess.thecvf.com//content/ICCV2021/html/Shao_LocalTrans_A_Multiscale_Local_Transformer_Network_for_Cross-Resolution_Homography_Estimation_ICCV_2021_paper.html
http://openaccess.thecvf.com//content/ICCV2021/papers/Shao_LocalTrans_A_Multiscale_Local_Transformer_Network_for_Cross-Resolution_Homography_Estimation_ICCV_2021_paper.pdf
iccv-2021-1
['homography-estimation']
['computer-vision']
[ 2.09018156e-01 -3.19905996e-01 2.32420370e-01 -2.95349449e-01 -1.17212856e+00 -4.31388050e-01 4.72535312e-01 -4.55060571e-01 -2.63230413e-01 3.83831590e-01 3.00048470e-01 3.82633567e-01 -4.42331403e-01 -7.54316747e-01 -1.04216504e+00 -5.95811486e-01 4.28676069e-01 4.33564425e-01 7.42652714e-02 -2.36595318e-01 3.62653375e-01 3.37629199e-01 -1.40729856e+00 1.24958545e-01 8.25100362e-01 1.05353498e+00 3.88111740e-01 3.71509433e-01 2.48772517e-01 3.90588164e-01 -1.45588174e-01 -4.20954466e-01 5.34198046e-01 -3.80815148e-01 -5.73158562e-01 2.46306658e-01 1.38846231e+00 -5.30489624e-01 -6.60157800e-01 1.24432147e+00 4.02292162e-01 -1.52487963e-01 4.48237479e-01 -9.23590124e-01 -8.35687041e-01 2.54322737e-01 -9.46784914e-01 4.68451641e-02 3.97022694e-01 -3.94365331e-03 1.12119997e+00 -1.03670025e+00 7.45269895e-01 1.38965690e+00 5.61537266e-01 1.76131483e-02 -1.40042925e+00 -6.73864067e-01 -8.44051093e-02 2.63406873e-01 -1.55389881e+00 -3.81824464e-01 1.05252016e+00 -4.70724404e-01 8.83225679e-01 -9.74608734e-02 4.78713065e-01 8.79956663e-01 3.40317130e-01 2.52544045e-01 1.35026276e+00 -1.43950179e-01 -3.93068016e-01 -2.55377918e-01 -2.96063513e-01 8.39131296e-01 5.82407527e-02 3.80179614e-01 -8.27535808e-01 1.66178286e-01 1.44402945e+00 1.70480013e-01 -6.46324754e-01 -5.78705668e-01 -1.52118301e+00 6.47448123e-01 7.37731040e-01 2.53082126e-01 -2.78677970e-01 1.54020533e-01 2.56912746e-02 1.29571214e-01 2.58530498e-01 5.42845607e-01 -3.38092744e-02 6.81685135e-02 -9.33329523e-01 -3.69554348e-02 3.25308442e-01 1.09936905e+00 1.07413399e+00 -6.28534555e-02 2.86181003e-01 7.79658377e-01 1.18563198e-01 5.71551442e-01 3.26770902e-01 -1.07431102e+00 8.27034473e-01 5.09028137e-01 1.09347187e-01 -1.60406601e+00 -3.34313750e-01 -4.17684376e-01 -1.14134181e+00 2.24760510e-02 2.33849764e-01 2.04367146e-01 -5.23623765e-01 1.68534791e+00 1.85976848e-01 2.02316672e-01 -9.22525376e-02 1.30882478e+00 5.79667211e-01 4.51845884e-01 -6.80782318e-01 -1.38741285e-01 1.37056708e+00 -1.06539536e+00 -7.19775081e-01 -2.34436393e-01 -1.09713189e-01 -1.16733873e+00 8.42759788e-01 1.70397758e-01 -1.18272984e+00 -8.18228424e-01 -1.26468921e+00 -6.38462603e-01 -2.77786136e-01 3.66169035e-01 1.87669992e-01 -5.38505465e-02 -8.69906247e-01 6.31521106e-01 -5.90648532e-01 -2.06901655e-01 7.52742365e-02 2.55634636e-01 -8.33320439e-01 -3.24110001e-01 -1.03029084e+00 7.66325891e-01 1.21960759e-01 3.16183865e-01 -4.61400896e-01 -7.28654504e-01 -1.03917456e+00 1.81699440e-01 2.66345888e-01 -1.02776921e+00 6.24110222e-01 -6.17013633e-01 -1.37967861e+00 8.84307623e-01 -7.45061040e-02 -9.38592404e-02 7.10111856e-01 -3.99847299e-01 -2.09591463e-01 3.53767365e-01 2.46485174e-01 7.17138886e-01 1.04248619e+00 -1.40488672e+00 -5.83361745e-01 -4.94629294e-01 2.32578963e-02 4.15353298e-01 -6.91884905e-02 -3.59132200e-01 -5.44416487e-01 -4.37241226e-01 5.49093544e-01 -6.84528947e-01 6.69082105e-02 -5.76739945e-02 -4.34409559e-01 3.15850168e-01 6.93973243e-01 -8.04497480e-01 9.22161162e-01 -2.01521945e+00 5.78556955e-01 -1.27660185e-02 3.62626612e-01 -1.48560449e-01 -1.70296282e-01 3.68822217e-01 -2.44452283e-02 -2.89531767e-01 -2.55404729e-02 -4.61183399e-01 2.37371344e-02 6.63754568e-02 -3.32111806e-01 7.95734584e-01 1.51226372e-01 8.62540126e-01 -7.84279227e-01 -4.33089197e-01 5.28435767e-01 8.32406342e-01 -4.01573569e-01 3.92642945e-01 2.56987005e-01 6.11015558e-01 -1.71529740e-01 5.32674015e-01 1.07897747e+00 -4.54610169e-01 -5.67441387e-03 -1.18706071e+00 -4.52443272e-01 -1.26580968e-01 -1.04220796e+00 2.01848769e+00 -6.60925746e-01 7.28744566e-01 7.89424777e-02 -7.98305511e-01 1.10826075e+00 -3.61501910e-02 6.01515174e-01 -1.01004004e+00 2.21550539e-01 3.62706661e-01 -1.40617579e-01 -2.16388434e-01 6.21858239e-01 4.42990512e-02 -1.13118216e-01 1.03708640e-01 1.62771493e-01 -2.19330683e-01 -6.10672589e-03 -7.12334141e-02 6.88619435e-01 1.63052827e-02 2.48774961e-01 -1.65359061e-02 7.98144102e-01 -4.17032629e-01 4.88114387e-01 4.26723242e-01 1.78622723e-01 1.21744013e+00 2.99478590e-01 -6.33575559e-01 -1.48855996e+00 -1.11666417e+00 -1.26439437e-01 3.66779238e-01 6.73243523e-01 -3.55490416e-01 -5.41080713e-01 -2.36454785e-01 -9.06807650e-03 -2.01323718e-01 -5.55464864e-01 1.64215900e-02 -7.86369383e-01 -3.35370332e-01 6.65461645e-02 4.42739069e-01 1.13168669e+00 -4.41310972e-01 -6.09471619e-01 1.41582012e-01 -4.85180616e-01 -1.75292408e+00 -9.40413892e-01 -3.57196450e-01 -5.43510616e-01 -1.17132258e+00 -7.77968824e-01 -7.71384656e-01 6.38843596e-01 5.32998860e-01 1.04212749e+00 -2.13439137e-01 -3.15978408e-01 4.76505697e-01 -3.05149369e-02 6.35605276e-01 9.59046185e-02 7.55285770e-02 -1.70086720e-03 4.06869382e-01 1.92247331e-02 -9.92928922e-01 -8.19035769e-01 6.71901166e-01 -7.63320088e-01 4.30365533e-01 9.12954330e-01 1.23087955e+00 8.72773886e-01 -1.60803556e-01 2.31258813e-02 -5.10242999e-01 2.77670473e-01 -2.43616477e-02 -1.03865111e+00 4.03489977e-01 -4.29891855e-01 6.98326975e-02 7.19847798e-01 -2.92028815e-01 -8.17868352e-01 2.50718534e-01 1.52892247e-01 -9.93184805e-01 1.30104229e-01 2.36431345e-01 -3.90597492e-01 -6.04924440e-01 2.39515185e-01 3.58808309e-01 2.54765451e-02 -6.23086274e-01 2.76886821e-01 2.00660840e-01 1.00846267e+00 -3.92441928e-01 1.06574202e+00 7.13123322e-01 3.95761162e-01 -6.07467353e-01 -7.99885988e-01 -4.14684564e-01 -1.12789083e+00 -8.75610486e-02 1.04205096e+00 -1.11301887e+00 -7.90450752e-01 4.21462178e-01 -1.24343467e+00 1.82626516e-01 2.48957217e-01 5.85409641e-01 -8.08459461e-01 5.73659003e-01 -4.88408804e-01 -1.23769067e-01 -3.03262889e-01 -1.52982843e+00 1.62547600e+00 3.36761475e-01 2.46378496e-01 -8.42584312e-01 5.18001504e-02 5.64238608e-01 3.31377447e-01 3.06422502e-01 7.26577699e-01 2.06279486e-01 -1.23706424e+00 -1.33843556e-01 -7.22469091e-01 2.20801637e-01 2.11739272e-01 -1.41601786e-01 -7.05659986e-01 -3.50614339e-01 -7.57034570e-02 -2.98729748e-01 7.51980960e-01 2.02418476e-01 8.09620917e-01 -1.85495615e-01 -2.68001924e-03 1.22118497e+00 1.68602049e+00 -2.70062536e-01 7.13860691e-01 3.66640151e-01 1.19132257e+00 5.66246092e-01 6.65636361e-01 3.57980609e-01 5.48530757e-01 1.12910008e+00 5.70392191e-01 -3.64073396e-01 -7.92378560e-02 -3.92336726e-01 1.41078383e-01 9.08082724e-01 -1.15665428e-01 2.37043843e-01 -5.14743745e-01 3.75168890e-01 -2.01864386e+00 -9.12616789e-01 2.04230443e-01 2.19577312e+00 6.36908233e-01 -3.73117328e-01 -2.25802094e-01 -3.02309722e-01 7.34336019e-01 4.49505419e-01 -5.92050076e-01 1.48234889e-01 -4.12252814e-01 -4.11529131e-02 6.20093942e-01 7.92059124e-01 -1.16350174e+00 9.34916079e-01 5.54486847e+00 7.57547498e-01 -1.35787892e+00 -5.78726493e-02 2.52883106e-01 -1.52027607e-03 -1.27915040e-01 2.55434364e-02 -7.90881157e-01 3.66343975e-01 2.05691338e-01 -1.08001649e-01 6.29974425e-01 4.99052823e-01 -4.63129133e-02 2.50493474e-02 -1.24392533e+00 1.49325907e+00 3.77003163e-01 -1.56276464e+00 1.92001432e-01 3.01160723e-01 9.12788689e-01 -5.91339171e-02 4.02096808e-01 -2.91968107e-01 -1.82010517e-01 -1.06616127e+00 4.73739237e-01 8.15712631e-01 8.76056612e-01 -6.70611799e-01 7.35278726e-01 9.11600366e-02 -1.56833041e+00 1.14281207e-01 -6.78113937e-01 3.88308972e-01 4.36966985e-01 2.86953330e-01 -1.70060337e-01 9.96400476e-01 8.69732320e-01 1.02291703e+00 -6.83008134e-01 7.73827434e-01 1.62232488e-01 -3.82036626e-01 -2.86063939e-01 7.61295438e-01 1.61121950e-01 -6.63509429e-01 4.21829998e-01 8.53253901e-01 6.65354669e-01 2.49798205e-02 2.23926350e-01 1.16144252e+00 -1.21023387e-01 -1.33337647e-01 -6.95287883e-01 1.53845236e-01 4.06362921e-01 1.46660292e+00 -2.42871493e-01 -1.41486108e-01 -5.69549382e-01 1.18439651e+00 4.51643318e-01 4.10551369e-01 -7.43760884e-01 -2.12084830e-01 7.98652411e-01 5.71480319e-02 4.92326528e-01 -4.27359700e-01 -4.10477147e-02 -1.48741806e+00 2.55689800e-01 -8.68979454e-01 1.42585352e-01 -1.06035972e+00 -1.46536589e+00 7.69272923e-01 3.74112837e-02 -1.65490937e+00 -1.66778669e-01 -6.43692851e-01 -3.67126793e-01 1.16248500e+00 -1.62464535e+00 -1.62100077e+00 -8.22434247e-01 7.58444071e-01 3.44603509e-01 -1.41826600e-01 3.92982155e-01 5.35045922e-01 -2.79296845e-01 5.94818354e-01 1.44539729e-01 2.33306527e-01 9.10049200e-01 -1.05337191e+00 1.38109297e-01 8.07167888e-01 9.66347381e-02 8.39882553e-01 4.89152163e-01 -3.74032468e-01 -1.68555474e+00 -8.89349222e-01 5.48945189e-01 -3.08405876e-01 5.58966219e-01 -3.25519323e-01 -9.83454347e-01 6.10567868e-01 4.26469386e-01 3.03877175e-01 1.08632311e-01 -4.35453027e-01 -6.09391987e-01 -4.31867123e-01 -9.55351353e-01 3.77420038e-01 1.16445792e+00 -1.00922132e+00 -2.35148340e-01 8.83140936e-02 5.22502720e-01 -7.44020522e-01 -1.35018003e+00 4.68775302e-01 7.87040472e-01 -1.28122723e+00 1.27059257e+00 -1.51653420e-02 7.73890138e-01 -4.77360725e-01 -3.82765979e-01 -1.18078566e+00 -4.74122584e-01 -5.89947760e-01 1.02484688e-01 1.18175781e+00 6.67078197e-02 -6.47245347e-01 4.19532120e-01 2.53008842e-01 -1.07643346e-03 -6.29844189e-01 -1.15224648e+00 -5.47200859e-01 -4.15821783e-02 3.36166948e-01 5.28589725e-01 9.85563338e-01 -3.29904974e-01 5.79852104e-01 -8.44234288e-01 6.11926258e-01 1.11681080e+00 6.39225245e-01 9.61245537e-01 -8.85304928e-01 -3.42622787e-01 -5.88177443e-01 -6.95424736e-01 -1.45117867e+00 2.14926735e-01 -4.17648584e-01 -1.08539104e-01 -1.11895978e+00 1.48784727e-01 -4.14248183e-02 1.84068739e-01 -1.05329871e-01 7.22446516e-02 5.48817277e-01 2.30241343e-01 3.15346777e-01 -3.77651572e-01 8.66793454e-01 1.49717581e+00 -1.43066630e-01 1.84003368e-01 -4.67679709e-01 -3.77869099e-01 6.13364339e-01 2.64605224e-01 1.25376182e-02 -2.53877550e-01 -6.87010288e-01 1.18522272e-01 2.46879116e-01 5.24962783e-01 -1.03064883e+00 5.86814821e-01 -4.80893403e-02 5.63355029e-01 -9.07659709e-01 7.65176654e-01 -1.01824450e+00 4.69707072e-01 -7.75032565e-02 -2.29069054e-01 5.12462795e-01 -9.23343003e-02 4.52426255e-01 -5.69400251e-01 2.90894687e-01 8.96511912e-01 -9.68201831e-02 -6.36978984e-01 6.40436053e-01 4.84833747e-01 -3.91819812e-02 7.01379180e-01 -2.20977098e-01 -6.57243133e-01 -4.16522354e-01 -5.02789319e-01 1.94496095e-01 9.60880637e-01 4.62093771e-01 1.02754593e+00 -1.56211269e+00 -4.89438593e-01 5.11919916e-01 2.07621247e-01 1.11335546e-01 5.92859924e-01 1.11197591e+00 -5.74249685e-01 3.93262744e-01 -7.17927992e-01 -8.98982823e-01 -1.28070664e+00 3.99201870e-01 7.38356888e-01 -1.87067643e-01 -8.18720818e-01 4.17710751e-01 6.39166772e-01 -3.96038353e-01 -7.40359798e-02 -1.60647199e-01 -2.27015659e-01 -1.94674402e-01 3.31755549e-01 2.65918165e-01 -1.90687343e-01 -1.22399259e+00 -1.31124184e-01 1.73963320e+00 -1.38590083e-01 -2.57288009e-01 1.23936689e+00 -4.91596282e-01 -2.14793980e-01 1.79341227e-01 1.73985267e+00 -5.83016165e-02 -1.47486448e+00 -6.29784226e-01 -4.78248835e-01 -1.08102274e+00 -1.72070526e-02 -2.83764362e-01 -1.34905016e+00 1.18774879e+00 6.50052965e-01 -2.47196496e-01 1.20563066e+00 -5.74599952e-02 8.15330207e-01 1.59565687e-01 4.50919449e-01 -8.37001979e-01 3.29574794e-01 4.29648757e-01 1.22336864e+00 -1.46096492e+00 2.77296871e-01 -5.10253668e-01 -4.25787449e-01 1.38740480e+00 9.34464693e-01 -3.89999449e-01 3.74225110e-01 -2.06755653e-01 1.93849783e-02 -2.64823794e-01 -6.01375937e-01 -1.06526256e-01 6.28253341e-01 4.62428480e-01 2.14417681e-01 -2.56867021e-01 -3.92339751e-02 1.48593470e-01 -2.84327716e-01 -4.98269558e-01 4.47258800e-01 4.47364837e-01 -5.95894754e-02 -1.04734945e+00 -4.14938658e-01 -5.15939221e-02 -4.37154807e-02 -1.34421075e-02 -3.88909400e-01 9.63460207e-01 9.47204977e-02 5.21081448e-01 3.38482261e-01 -5.74044406e-01 4.36692059e-01 -5.91386378e-01 8.30693901e-01 3.77236381e-02 -2.57619053e-01 3.91352564e-01 -2.14101940e-01 -9.98718262e-01 -5.60634077e-01 -4.99013066e-01 -7.71586597e-01 -4.28942919e-01 -4.71828738e-03 -1.87484145e-01 4.42832232e-01 7.93222427e-01 4.25014377e-01 2.35715941e-01 8.47679317e-01 -1.26410866e+00 -4.65110272e-01 -8.03363502e-01 -5.60908556e-01 5.55337727e-01 7.15557456e-01 -8.77925515e-01 -4.87229824e-01 -2.32966268e-03]
[8.799187660217285, -2.286741256713867]
8d4e7cc7-0e5b-43a9-8890-81c1391b89e4
exemplar-bsaed-pattern-synthesis-with
2204.01671
null
https://arxiv.org/abs/2204.01671v2
https://arxiv.org/pdf/2204.01671v2.pdf
Exemplar-based Pattern Synthesis with Implicit Periodic Field Network
Synthesis of ergodic, stationary visual patterns is widely applicable in texturing, shape modeling, and digital content creation. The wide applicability of this technique thus requires the pattern synthesis approaches to be scalable, diverse, and authentic. In this paper, we propose an exemplar-based visual pattern synthesis framework that aims to model the inner statistics of visual patterns and generate new, versatile patterns that meet the aforementioned requirements. To this end, we propose an implicit network based on generative adversarial network (GAN) and periodic encoding, thus calling our network the Implicit Periodic Field Network (IPFN). The design of IPFN ensures scalability: the implicit formulation directly maps the input coordinates to features, which enables synthesis of arbitrary size and is computationally efficient for 3D shape synthesis. Learning with a periodic encoding scheme encourages diversity: the network is constrained to model the inner statistics of the exemplar based on spatial latent codes in a periodic field. Coupled with continuously designed GAN training procedures, IPFN is shown to synthesize tileable patterns with smooth transitions and local variations. Last but not least, thanks to both the adversarial training technique and the encoded Fourier features, IPFN learns high-frequency functions that produce authentic, high-quality results. To validate our approach, we present novel experimental results on various applications in 2D texture synthesis and 3D shape synthesis.
['Yajie Zhao', 'Shichen Liu', 'Weikai Chen', 'Jiayi Liu', 'Haiwei Chen']
2022-04-04
null
http://openaccess.thecvf.com//content/CVPR2022/html/Chen_Exemplar-Based_Pattern_Synthesis_With_Implicit_Periodic_Field_Network_CVPR_2022_paper.html
http://openaccess.thecvf.com//content/CVPR2022/papers/Chen_Exemplar-Based_Pattern_Synthesis_With_Implicit_Periodic_Field_Network_CVPR_2022_paper.pdf
cvpr-2022-1
['texture-synthesis']
['computer-vision']
[ 4.44726527e-01 2.78379083e-01 3.86488847e-02 3.56371611e-01 -1.98484182e-01 -5.71284652e-01 8.99032414e-01 -6.26550794e-01 4.87767011e-01 7.61758983e-01 1.31907701e-01 9.13763344e-02 -9.68853533e-02 -1.06377625e+00 -9.28482473e-01 -1.03968179e+00 8.08845237e-02 2.60280281e-01 -4.14854549e-02 -3.88529599e-01 1.16721697e-01 1.01443577e+00 -1.67047298e+00 1.90778986e-01 6.88809097e-01 1.18013179e+00 1.55822784e-01 5.30598700e-01 -1.71248063e-01 6.75485075e-01 -4.82048035e-01 -2.53727704e-01 4.05137330e-01 -6.52905166e-01 -3.43661755e-01 4.97884899e-01 2.84921169e-01 -5.43081723e-02 -2.43531182e-01 5.95173180e-01 5.41422188e-01 5.50996065e-02 9.52858627e-01 -1.08151376e+00 -1.04997098e+00 7.08338469e-02 -3.04309547e-01 -5.54755747e-01 3.14017117e-01 3.96515518e-01 6.64225042e-01 -8.76891494e-01 9.01192188e-01 1.07442725e+00 8.60067546e-01 8.07307720e-01 -1.57861769e+00 -2.29514912e-01 -1.85443476e-01 -3.57958943e-01 -1.32309902e+00 -3.75168294e-01 1.26576614e+00 -3.56267810e-01 4.21050519e-01 4.86161411e-01 9.56818759e-01 1.37062621e+00 3.41763854e-01 6.17461860e-01 1.17909515e+00 -7.00148761e-01 4.05806810e-01 -9.60457511e-03 -9.84142482e-01 6.27390802e-01 -1.91472709e-01 3.80645066e-01 -3.99015099e-01 -1.53081650e-02 1.48865294e+00 3.11580561e-02 -4.05240536e-01 -7.76968777e-01 -1.20690942e+00 6.17905557e-01 3.72068942e-01 4.93676811e-01 -4.30366814e-01 4.07964498e-01 -1.65991653e-02 3.68353218e-01 5.77489793e-01 6.93526506e-01 -7.18901353e-03 -3.05828471e-02 -9.30969238e-01 3.87124658e-01 6.67836547e-01 1.08657110e+00 6.42174900e-01 7.27194607e-01 -4.24874336e-01 9.78018463e-01 4.96549383e-02 8.98675859e-01 3.86492789e-01 -1.03896940e+00 -1.59929648e-01 3.81845623e-01 1.27393946e-01 -1.20731211e+00 -5.05463146e-02 -5.11355698e-01 -1.18085647e+00 5.48291564e-01 8.38841423e-02 1.35484993e-01 -9.69788492e-01 1.73265362e+00 3.90350133e-01 1.63506493e-01 -6.30374774e-02 5.02297163e-01 4.76237327e-01 7.13342786e-01 -4.22727495e-01 -1.36186093e-01 8.18520129e-01 -5.31252623e-01 -7.31651664e-01 3.84765804e-01 5.30685857e-02 -9.12091196e-01 1.30234516e+00 1.93510368e-01 -1.33023465e+00 -6.68100953e-01 -9.28597867e-01 2.78168052e-01 -2.80632675e-02 8.13770294e-02 3.32465589e-01 5.22061825e-01 -1.21880770e+00 6.82497144e-01 -4.87341136e-01 -1.40734436e-02 4.75456804e-01 1.03523642e-01 -2.11886346e-01 3.56027111e-02 -8.01112473e-01 5.06249607e-01 1.10873990e-01 -1.19795643e-01 -6.77055776e-01 -9.05819237e-01 -8.41555238e-01 5.26961982e-02 -5.19369617e-02 -9.13693726e-01 8.07518601e-01 -1.15422785e+00 -2.13641667e+00 6.75619900e-01 1.15441736e-02 -1.40287682e-01 6.58307016e-01 3.48731935e-01 -2.55304664e-01 1.27397716e-01 -6.73939809e-02 6.38012648e-01 1.56052232e+00 -1.71099925e+00 -9.23838094e-02 1.41957283e-01 -3.39392304e-01 -4.99708462e-04 -2.24265620e-01 -6.26779199e-01 -5.94620332e-02 -1.26083648e+00 -7.37123191e-02 -8.06124270e-01 -1.42065883e-01 2.18442172e-01 -3.08412343e-01 1.82142377e-01 1.09767330e+00 -4.08017129e-01 8.64289224e-01 -2.42558002e+00 3.70022267e-01 5.64043641e-01 5.50143979e-02 -7.85427466e-02 -1.57291964e-01 5.51302850e-01 2.23986432e-02 -8.34115967e-03 -3.43998402e-01 -3.63416255e-01 1.59300819e-01 3.99325132e-01 -6.19501948e-01 2.93811232e-01 4.62714583e-01 1.22525883e+00 -6.53305411e-01 -1.74478784e-01 3.50579649e-01 6.45471573e-01 -8.07365000e-01 2.22156987e-01 -5.60240209e-01 9.18563545e-01 -4.16757315e-01 6.62639439e-01 5.64839959e-01 -2.09510699e-01 -8.15753490e-02 -6.36574477e-02 -2.20179275e-01 -3.95215452e-01 -9.71962750e-01 1.68467736e+00 -7.00551987e-01 4.52370465e-01 -4.78324778e-02 -8.12591910e-01 1.37320960e+00 3.97203684e-01 5.38271904e-01 -9.16431248e-01 5.66326221e-03 5.17766953e-01 -4.61402267e-01 -2.22876921e-01 3.84549886e-01 -2.48768002e-01 -2.27405061e-03 3.25883776e-01 2.24190783e-02 -7.84193695e-01 -2.13105842e-01 -1.34692669e-01 8.72535050e-01 3.31046492e-01 -5.25540039e-02 -2.64804959e-01 5.47874212e-01 -3.83788705e-01 3.05924535e-01 6.38008773e-01 5.65425098e-01 1.00136495e+00 4.47826952e-01 -5.38093507e-01 -1.72851825e+00 -1.10133362e+00 -1.90189183e-01 4.16260451e-01 -6.32259771e-02 -4.52900343e-02 -6.81871772e-01 -3.99754584e-01 -5.39362803e-02 5.04361033e-01 -8.19258809e-01 -1.29230902e-01 -7.92620063e-01 -5.95678668e-03 3.45836133e-01 2.95328349e-01 4.76340830e-01 -1.36990857e+00 -4.38442290e-01 3.37103754e-01 9.73664373e-02 -9.21509922e-01 -5.07381380e-01 -1.13539554e-01 -6.74821198e-01 -6.83919549e-01 -1.28990781e+00 -9.07206655e-01 7.05212533e-01 -1.26938656e-01 9.82409239e-01 2.87157819e-02 -3.42762977e-01 6.35170281e-01 -3.85485440e-01 -1.69279799e-01 -8.34954798e-01 -1.25942066e-01 -1.23661466e-01 4.52019423e-01 -6.67816639e-01 -1.24391520e+00 -6.17311835e-01 3.37693721e-01 -1.32903397e+00 3.14430028e-01 5.71632266e-01 1.16980541e+00 8.80456090e-01 -2.74594594e-02 5.81884682e-01 -5.76513886e-01 5.15304923e-01 -1.86609417e-01 -4.90930021e-01 2.58469015e-01 -2.06203490e-01 2.37490728e-01 9.98285115e-01 -6.64570510e-01 -9.17421281e-01 -2.82678735e-02 -3.08310419e-01 -7.62004495e-01 -5.90734705e-02 -1.73997469e-02 -1.34546950e-01 -4.37720805e-01 8.31488013e-01 7.88129151e-01 3.38773459e-01 -2.41465151e-01 4.48890835e-01 1.36460587e-01 4.62881446e-01 -7.69963264e-01 1.15399325e+00 6.09251022e-01 3.98569465e-01 -1.05450010e+00 -2.34335631e-01 3.40534031e-01 -4.25524414e-01 -3.93167526e-01 6.16821647e-01 -4.54127043e-01 -5.72487235e-01 6.08949721e-01 -9.36133087e-01 -8.67951035e-01 -1.08647752e+00 -5.14466427e-02 -1.32217944e+00 1.49301425e-01 -2.78393924e-01 -6.67805731e-01 -3.05885613e-01 -9.53271568e-01 1.22398710e+00 -3.32654454e-02 -2.32904926e-01 -1.11146224e+00 8.61229971e-02 -3.18120271e-01 8.42411578e-01 9.12243545e-01 1.09531021e+00 2.07639351e-01 -6.92863226e-01 -9.42738075e-03 1.51827142e-01 3.53072554e-01 4.89092231e-01 2.80261990e-02 -8.24187338e-01 -1.70894623e-01 5.96487336e-02 -2.54554868e-01 5.29225051e-01 4.78360981e-01 1.36685252e+00 -6.39458597e-01 -6.97566420e-02 8.90007675e-01 1.36071086e+00 1.07488111e-01 9.08465207e-01 6.66588843e-02 6.90444112e-01 4.14983481e-01 1.11977972e-01 5.62896132e-01 -2.22993135e-01 9.44510520e-01 2.67041892e-01 -2.65995145e-01 -6.28532648e-01 -5.21666110e-01 1.45059690e-01 6.66982293e-01 -3.49803716e-01 -1.89897701e-01 -4.85846549e-01 3.49056214e-01 -1.53567636e+00 -1.19887650e+00 2.94689924e-01 2.03602481e+00 8.50336730e-01 -2.18264982e-01 -8.61047357e-02 4.15128380e-01 5.58113575e-01 1.99204519e-01 -3.25735122e-01 -5.77940941e-01 -4.11308467e-01 7.59703398e-01 1.53245181e-01 3.25669467e-01 -7.28969991e-01 7.68330276e-01 6.07374763e+00 1.32851624e+00 -1.40312362e+00 -1.25909820e-01 6.57224655e-01 1.07090600e-01 -9.66507077e-01 -4.22605693e-01 -3.55659217e-01 5.71865916e-01 3.40471596e-01 -2.34731982e-04 6.11893475e-01 6.84366226e-01 1.31002828e-01 4.13693577e-01 -7.16453671e-01 9.41277444e-01 1.56036601e-03 -1.86892056e+00 4.64868873e-01 1.27373859e-01 1.28177834e+00 -6.73254430e-01 5.54229558e-01 -1.74637899e-01 3.11481357e-02 -1.19099748e+00 1.05498111e+00 8.31397057e-01 1.41782427e+00 -1.02854526e+00 1.54682323e-01 2.13666067e-01 -1.04003012e+00 4.15189564e-02 -1.89864293e-01 3.68680686e-01 1.38906360e-01 6.53983891e-01 -5.76915145e-01 6.53861284e-01 3.11767071e-01 8.03874552e-01 -1.04039274e-01 6.85915470e-01 -2.22234875e-01 2.87689358e-01 -1.80065036e-01 -6.10143393e-02 2.32023507e-01 -3.13160032e-01 7.79340744e-01 7.93794334e-01 7.62895942e-01 -3.18075530e-02 5.50284162e-02 1.26843727e+00 -2.69853640e-02 1.31689698e-01 -1.01220226e+00 1.07805885e-01 2.36541376e-01 9.30963039e-01 -7.23855317e-01 1.28748238e-01 1.11748569e-01 9.76631761e-01 -1.30725617e-03 5.10229468e-01 -6.55663013e-01 -2.35762745e-01 4.16885167e-01 4.65774924e-01 7.05417454e-01 -3.77058923e-01 -4.16588426e-01 -9.45462644e-01 1.75995633e-01 -1.05755746e+00 -5.32035172e-01 -7.72308648e-01 -1.39808249e+00 7.51872361e-01 -3.44088942e-01 -1.55608869e+00 -3.50099146e-01 -4.04408425e-01 -6.48383737e-01 8.60375583e-01 -1.37529385e+00 -1.52993178e+00 -2.90536433e-01 8.18939567e-01 3.98415416e-01 -3.74389082e-01 9.95819449e-01 1.17006125e-02 1.36981875e-01 7.30986893e-01 1.93909824e-01 -1.67027056e-01 3.56805384e-01 -1.03501999e+00 4.31327432e-01 5.16797781e-01 6.94335476e-02 2.30273262e-01 6.06484294e-01 -5.54429710e-01 -1.37654686e+00 -1.16194665e+00 4.41334009e-01 -1.84484169e-01 3.67508739e-01 -2.90247738e-01 -7.45704889e-01 2.85144418e-01 6.77877516e-02 5.10158353e-02 3.53217334e-01 -5.25899887e-01 -1.52167752e-01 -5.24807833e-02 -1.27278090e+00 8.20371509e-01 1.14560235e+00 -3.30344945e-01 -1.63477138e-01 2.57288814e-01 6.25074387e-01 -5.38494945e-01 -8.69326472e-01 5.66467285e-01 5.35157681e-01 -9.88811672e-01 1.03892887e+00 -1.40526667e-01 8.86828303e-01 -2.44272172e-01 -1.78120986e-01 -1.38993835e+00 -4.55786586e-01 -9.16296005e-01 -2.58933872e-01 1.03594840e+00 2.01160580e-01 -7.31724977e-01 6.96909964e-01 4.63035442e-02 -3.57704878e-01 -9.29151952e-01 -8.94016623e-01 -9.56008255e-01 5.62442392e-02 -1.05947688e-01 9.28363144e-01 9.10187244e-01 -5.40695310e-01 -3.44479561e-01 -7.37078726e-01 -1.32386535e-01 5.26051521e-01 3.31771046e-01 8.84074211e-01 -8.51847410e-01 -5.62640011e-01 -5.15776753e-01 -4.78219658e-01 -1.09305263e+00 1.40783682e-01 -7.08076119e-01 2.24162545e-02 -1.07502031e+00 -3.51698786e-01 -9.67159450e-01 3.11624736e-01 2.73474962e-01 2.79931545e-01 5.56284308e-01 1.12983689e-01 2.28769839e-01 1.65047526e-01 1.09057295e+00 1.91556609e+00 -1.65730789e-01 -2.56297946e-01 1.39040470e-01 -3.50602597e-01 2.71689028e-01 6.48384690e-01 -9.28860828e-02 -4.74052161e-01 -1.40648380e-01 2.72971421e-01 3.38284411e-02 5.67525268e-01 -1.06016421e+00 -1.88559234e-01 -1.48960486e-01 6.44454718e-01 -2.13050753e-01 4.10861194e-01 -8.57834637e-01 6.49859250e-01 3.77844006e-01 -1.84725121e-01 -2.74826825e-01 1.29204243e-01 4.58399773e-01 -2.23472163e-01 4.15647775e-02 8.80670726e-01 -9.24652889e-02 -2.20112041e-01 4.89451617e-01 -1.93894029e-01 -7.46409521e-02 9.84359086e-01 -4.68233794e-01 6.37630448e-02 -4.99091595e-01 -8.36033523e-01 -3.93658608e-01 9.05206501e-01 2.06437036e-01 7.84320891e-01 -1.92149758e+00 -6.17635489e-01 7.37387955e-01 -7.71722868e-02 1.26209795e-01 3.47107857e-01 4.46060449e-01 -7.40637064e-01 1.70638055e-01 -4.28485990e-01 -7.57924139e-01 -6.15083039e-01 5.51019490e-01 3.76435906e-01 -2.02999905e-01 -8.43185365e-01 5.56477368e-01 2.08762586e-01 -3.95262122e-01 -1.01764165e-01 -1.40703544e-01 1.56936254e-02 -3.23518485e-01 2.92942375e-01 6.89669400e-02 -7.36246705e-02 -3.92565757e-01 3.14361125e-01 9.49093521e-01 5.76070428e-01 -1.77662566e-01 1.40153873e+00 1.98322073e-01 -8.46924111e-02 2.78805077e-01 1.01699281e+00 4.55146998e-01 -1.64050424e+00 -8.14862847e-02 -5.63840985e-01 -6.18394375e-01 -3.15388978e-01 -3.41041774e-01 -1.20432198e+00 4.77656871e-01 2.65473306e-01 4.47307914e-01 1.27550101e+00 -1.89911574e-01 7.00976372e-01 -9.49645862e-02 6.27732158e-01 -7.07959175e-01 4.84575361e-01 4.44092870e-01 1.44673109e+00 -5.47410965e-01 -3.50552946e-01 -3.86518538e-01 -4.58350152e-01 1.18469524e+00 1.58270255e-01 -4.87461776e-01 5.37366629e-01 4.09506619e-01 -2.15036169e-01 -5.18153235e-02 -4.26723182e-01 1.55838713e-01 5.29164195e-01 7.19392002e-01 3.45645193e-03 -6.00705370e-02 -8.81667882e-02 -3.43879163e-02 -3.13027978e-01 -1.19293733e-02 2.08462089e-01 7.87183166e-01 -1.26975372e-01 -1.33833289e+00 -4.08439964e-01 1.31471917e-01 -1.98311567e-01 1.47153987e-02 -7.80181065e-02 8.07411313e-01 2.42373630e-01 2.79268861e-01 4.71689850e-02 -2.16575623e-01 3.82024676e-01 2.31566522e-02 8.08354020e-01 -3.18029851e-01 -3.64407331e-01 3.78634594e-02 -3.15494448e-01 -4.71638203e-01 -4.10841167e-01 -4.92230326e-01 -7.89129257e-01 -3.53008538e-01 7.09157661e-02 -6.62443563e-02 4.28168595e-01 6.32571638e-01 4.94684935e-01 6.76664352e-01 1.18623281e+00 -1.14375556e+00 -2.85415709e-01 -6.66703284e-01 -6.76345825e-01 4.27650243e-01 3.51879925e-01 -7.06800938e-01 -3.10234457e-01 1.93396613e-01]
[9.125773429870605, -3.46344256401062]
773755ef-044a-4de3-a853-29e3f17a8a0b
triplet-loss-based-embeddings-for-forensic
2102.12564
null
https://arxiv.org/abs/2102.12564v2
https://arxiv.org/pdf/2102.12564v2.pdf
Triplet loss based embeddings for forensic speaker identification in Spanish
With the advent of digital technology, it is more common that committed crimes or legal disputes involve some form of speech recording where the identity of a speaker is questioned [1]. In face of this situation, the field of forensic speaker identification has been looking to shed light on the problem by quantifying how much a speech recording belongs to a particular person in relation to a population. In this work, we explore the use of speech embeddings obtained by training a CNN using the triplet loss. In particular, we focus on the Spanish language which has not been extensively studies. We propose extracting the embeddings from speech spectrograms samples, then explore several configurations of such spectrograms, and finally, quantify the embeddings quality. We also show some limitations of our data setting which is predominantly composed by male speakers. At the end, we propose two approaches to calculate the Likelihood Radio given out speech embeddings and we show that triplet loss is a good alternative to create speech embeddings for forensic speaker identification.
['Ivan Meza', 'Carlos Mena', 'Javier Alvarez-Jimenez', 'Emmanuel Maqueda']
2021-02-24
null
null
null
null
['speaker-identification']
['speech']
[ 4.82294671e-02 3.05380225e-01 3.19742024e-01 -4.43130463e-01 -9.32079554e-01 -5.05957484e-01 8.46647859e-01 1.65427104e-01 -5.89861095e-01 5.35814464e-01 3.86510462e-01 -3.45686108e-01 -1.08735710e-02 -5.72766006e-01 -5.88576913e-01 -6.90141261e-01 6.69837520e-02 4.14901257e-01 -2.14888513e-01 1.52513608e-01 1.29041195e-01 6.06692731e-01 -1.38593793e+00 -7.64015615e-02 3.85963798e-01 9.38012838e-01 -2.53089517e-01 5.65445662e-01 -2.41278186e-01 3.70318294e-01 -1.02014756e+00 -1.04524124e+00 1.89140245e-01 -1.47668496e-01 -8.88191342e-01 1.00292869e-01 5.06373405e-01 -3.92784536e-01 -4.31836098e-01 1.21238005e+00 7.67288923e-01 6.97993189e-02 6.34646714e-01 -1.15555966e+00 -5.54578364e-01 9.12844598e-01 -2.64404625e-01 5.70408344e-01 4.41805094e-01 -1.13347910e-01 9.13718522e-01 -7.88011730e-01 4.56074029e-01 1.49269462e+00 5.65806985e-01 4.08950716e-01 -1.08294523e+00 -6.28038049e-01 -1.41862750e-01 4.07240808e-01 -1.45580018e+00 -7.32569218e-01 1.05009067e+00 -4.66327131e-01 5.16358376e-01 1.47898749e-01 3.33881736e-01 1.71318734e+00 -4.30334955e-01 7.33464658e-01 8.84390354e-01 -5.39234042e-01 2.05428332e-01 4.77628529e-01 3.38737607e-01 2.05595881e-01 7.21163899e-02 -1.61903590e-01 -6.13556683e-01 -2.99245089e-01 2.03757003e-01 -3.01972270e-01 -1.60184160e-01 2.41106674e-02 -9.14443970e-01 9.35336232e-01 -9.38751251e-02 5.09649873e-01 -3.08996260e-01 -5.89116998e-02 5.67217767e-01 1.32522374e-01 5.45809150e-01 2.33124211e-01 1.18995726e-01 -4.78213668e-01 -1.15135646e+00 2.60932654e-01 8.65507007e-01 4.32839692e-01 4.99493480e-01 -2.67578159e-02 6.74127117e-02 1.10328889e+00 3.96791816e-01 3.03304464e-01 4.37470138e-01 -7.56269395e-01 5.72547019e-01 3.50522071e-01 -1.34205386e-01 -8.98294985e-01 -1.36274353e-01 -1.66532502e-01 -5.58811903e-01 -7.54162520e-02 6.44614458e-01 -2.74657663e-02 -2.91375190e-01 1.75455070e+00 3.92684817e-01 4.00652647e-01 -6.24685585e-02 6.59006298e-01 5.17930925e-01 4.99516636e-01 -2.36368313e-01 5.58191538e-02 1.49534929e+00 -3.61979157e-01 -8.45342577e-01 1.54455423e-01 2.01216087e-01 -5.35239279e-01 9.21590626e-01 4.04534757e-01 -8.24866712e-01 -3.89917344e-01 -9.88997340e-01 -7.09092245e-02 -5.74005246e-01 1.06311604e-01 2.30269685e-01 1.40587640e+00 -9.59451377e-01 4.75649178e-01 -7.91586995e-01 -4.11887139e-01 5.21771491e-01 2.24948272e-01 -5.53391695e-01 1.68825701e-01 -1.36621284e+00 8.78353477e-01 9.08120796e-02 1.33402646e-01 -7.09486604e-01 -6.19956255e-01 -8.60419273e-01 1.70344830e-01 1.52279556e-01 -6.55140355e-02 9.44789946e-01 -6.26362801e-01 -1.36631393e+00 1.09562826e+00 -1.21293709e-01 -7.08683252e-01 6.93460107e-01 -5.67766428e-02 -6.91445291e-01 1.67978317e-01 7.51845613e-02 2.07000345e-01 8.97104144e-01 -9.23996985e-01 -3.33767772e-01 -7.20454752e-01 1.14859104e-01 -4.32150513e-01 -7.58865058e-01 4.73003209e-01 -9.71720666e-02 -4.89224136e-01 -2.57638216e-01 -7.39276767e-01 3.86259735e-01 -2.07856130e-02 -5.58533490e-01 -3.38679910e-01 7.99257576e-01 -9.30668652e-01 1.20705366e+00 -2.55462480e+00 -1.76999435e-01 1.73205644e-01 1.53651953e-01 5.38158536e-01 1.41355038e-01 5.77769041e-01 -7.87882358e-02 4.01437700e-01 -4.46717650e-01 -8.99386227e-01 3.89719844e-01 -6.90973736e-03 -6.38790250e-01 6.34266198e-01 4.95885909e-01 3.08661371e-01 -6.61071241e-01 -5.17595291e-01 8.65464136e-02 9.59479988e-01 -3.35579902e-01 5.70354462e-02 4.90492493e-01 1.71621278e-01 -2.37112120e-01 5.95400989e-01 8.78489971e-01 2.91486889e-01 1.52908266e-01 -3.66734415e-02 -1.09352559e-01 6.75679266e-01 -1.25849605e+00 1.32931972e+00 -4.15047675e-01 1.00840771e+00 3.19813579e-01 -1.03934968e+00 9.28426504e-01 3.79236162e-01 1.66702345e-01 -4.42188948e-01 2.80243933e-01 1.55178025e-01 -1.40338310e-03 -5.69775701e-01 6.40942276e-01 -2.38265485e-01 7.61188045e-02 5.88059783e-01 2.14476243e-01 3.56408268e-01 9.82147232e-02 -4.62606959e-02 1.10471213e+00 -3.07453424e-01 8.80433768e-02 5.59098236e-02 6.20033205e-01 -6.38534725e-01 3.22427064e-01 5.36219537e-01 -5.53057849e-01 6.59024298e-01 7.36545205e-01 -1.00744337e-01 -1.11308968e+00 -1.29507065e+00 -5.57539165e-01 6.46478593e-01 -2.89218485e-01 -2.90569574e-01 -9.73088026e-01 -5.53597808e-01 1.62431076e-02 7.52375841e-01 -6.79902196e-01 -2.10433885e-01 -5.89011192e-01 -6.97321415e-01 1.24375856e+00 3.68895710e-01 2.85232186e-01 -9.47653592e-01 -4.46322292e-01 7.00559393e-02 -1.95213795e-01 -1.18763125e+00 -4.71888036e-01 -1.82650201e-02 -4.17971164e-01 -9.40637767e-01 -7.76843369e-01 -4.18191850e-01 1.61831737e-01 -1.88183829e-01 7.55626202e-01 -1.89927563e-01 -2.25348726e-01 4.81253952e-01 -4.54773843e-01 -4.62429971e-01 -7.01926351e-01 5.74540347e-02 3.53392839e-01 7.39511251e-01 6.54970229e-01 -6.44089520e-01 -2.24662274e-01 -6.89222217e-02 -1.06138396e+00 -6.93209946e-01 1.03929959e-01 5.27140439e-01 -9.02121589e-02 -9.81823653e-02 6.22455001e-01 -5.61186552e-01 8.64911675e-01 -6.99435174e-01 -2.54468083e-01 1.91020131e-01 -4.53983210e-02 -1.02327885e-02 4.33755100e-01 -4.28049475e-01 -8.38863969e-01 -2.35189885e-01 -3.92446041e-01 -4.42655265e-01 -4.47619975e-01 2.72628397e-01 -2.75553316e-01 2.34880418e-01 4.19608235e-01 2.07431972e-01 1.96761772e-01 -6.01183236e-01 1.57845914e-01 1.33975315e+00 4.38493460e-01 -5.81833541e-01 6.24705374e-01 5.77448547e-01 -4.38054413e-01 -1.46034718e+00 -3.09679955e-01 -4.43504304e-01 -6.54443741e-01 -2.76554197e-01 8.87394905e-01 -5.37573874e-01 -9.58488584e-01 3.26431304e-01 -1.39572442e+00 2.72247791e-01 -1.74201265e-01 6.16928399e-01 -3.13096106e-01 7.69814491e-01 -4.60495770e-01 -1.50085700e+00 -4.13387492e-02 -1.33764207e+00 1.17674291e+00 -1.55047148e-01 -2.76462525e-01 -9.96880472e-01 1.89123988e-01 4.55124140e-01 2.76972175e-01 3.14199142e-02 6.92635536e-01 -1.00553763e+00 -1.68187648e-01 -4.04151678e-01 -8.63627419e-02 5.82195640e-01 2.08717421e-01 1.27483219e-01 -1.60250902e+00 -1.03615344e-01 1.99922174e-01 -4.48043570e-02 6.93780601e-01 1.84129640e-01 1.02521145e+00 -1.12833485e-01 8.62120942e-04 4.14596945e-01 8.10020268e-01 1.45009458e-01 7.21427500e-01 3.55787069e-01 3.71346563e-01 1.07641363e+00 3.93214114e-02 5.96104681e-01 3.35070163e-01 7.68774092e-01 2.67691195e-01 4.44625527e-01 -2.31883470e-02 -2.73170203e-01 5.91523767e-01 6.88222885e-01 2.15451092e-01 -2.86755502e-01 -9.60655034e-01 7.56893039e-01 -1.27614498e+00 -1.16010284e+00 2.31305674e-01 2.34826326e+00 6.23610616e-01 -8.12301561e-02 5.06582975e-01 5.56180894e-01 1.02964592e+00 2.29430109e-01 -2.49616668e-01 -3.32392752e-01 -1.94943726e-01 1.99745491e-01 2.58517116e-01 4.36262459e-01 -1.08076513e+00 4.72575575e-01 6.29239082e+00 5.92482448e-01 -1.27640831e+00 8.82121772e-02 4.83348578e-01 -5.59789725e-02 -1.31277129e-01 -2.32105553e-01 -8.24228466e-01 8.32595110e-01 1.28685367e+00 -4.73816656e-02 3.27935487e-01 5.42736292e-01 8.79440606e-02 2.25802928e-01 -1.17797196e+00 1.32251859e+00 3.71735245e-01 -9.13140714e-01 1.90719794e-02 5.09034514e-01 -1.05222262e-01 -3.27056110e-01 3.27822804e-01 1.19355619e-01 -9.49000269e-02 -9.93597627e-01 9.88891780e-01 4.48676646e-01 5.02521515e-01 -6.50534749e-01 6.59925878e-01 2.74311393e-01 -8.57431889e-01 -9.47752893e-02 -2.88557798e-01 2.12255374e-01 3.76322657e-01 6.38795137e-01 -1.18189394e+00 3.88560414e-01 7.94421852e-01 3.11496973e-01 -5.24366915e-01 9.96768296e-01 1.14180602e-01 8.93558919e-01 -3.80059332e-01 2.14567129e-03 4.59263325e-02 -1.80620283e-01 6.03866160e-01 1.24337757e+00 6.20155394e-01 -2.78728664e-01 -4.04307008e-01 1.25522542e+00 -2.79098511e-01 5.06454073e-02 -8.48295152e-01 -4.79103833e-01 5.94425797e-01 1.02379012e+00 -4.59104180e-01 -9.91216302e-02 -2.48016834e-01 8.03308249e-01 3.56006771e-01 1.33549571e-01 -6.49226427e-01 -3.35066587e-01 1.00209892e+00 1.44814402e-01 1.51539281e-01 -1.84165627e-01 7.68841244e-04 -1.07258511e+00 3.99203420e-01 -7.18372703e-01 5.40346913e-02 -3.80934030e-01 -1.40072882e+00 6.05255902e-01 -4.50891741e-02 -1.03068340e+00 -3.56847435e-01 -7.31780291e-01 -6.14963651e-01 8.81069243e-01 -1.29917228e+00 -8.08451951e-01 8.49769264e-02 2.92364299e-01 3.23512554e-01 -3.78559768e-01 6.67925417e-01 7.95240104e-01 -7.41781354e-01 7.63110757e-01 6.41044825e-02 5.00475705e-01 6.36432886e-01 -1.18823564e+00 5.96646607e-01 7.27644086e-01 4.69616383e-01 8.84890854e-01 8.30161452e-01 -2.56476223e-01 -1.12800336e+00 -6.05676651e-01 1.19058990e+00 -5.76005101e-01 8.07534575e-01 -7.87451863e-01 -1.06967950e+00 5.64154208e-01 1.84166074e-01 -1.48432553e-01 9.97439206e-01 1.08480059e-01 -5.56218922e-01 -1.23625986e-01 -1.32989204e+00 3.87510955e-01 7.81199932e-01 -1.15425611e+00 -7.38299310e-01 7.40327537e-02 4.98169780e-01 1.78642198e-01 -7.67297804e-01 -1.76429585e-01 5.14235795e-01 -1.02318180e+00 1.06682742e+00 -5.28207839e-01 9.55127925e-02 -8.07760656e-02 -3.05847168e-01 -1.19947529e+00 2.38139734e-01 -4.14689630e-01 1.75063610e-01 1.78650284e+00 3.00487339e-01 -7.23230302e-01 6.59390032e-01 6.03005230e-01 -3.23230401e-02 -2.04627052e-01 -1.47858155e+00 -8.46410275e-01 3.43612194e-01 -7.60240436e-01 9.29062426e-01 1.00658083e+00 1.75988730e-02 -1.28555521e-02 -3.97301137e-01 3.10063303e-01 6.50139034e-01 -4.14167583e-01 7.26855993e-01 -1.32302594e+00 -8.49395096e-02 -4.83032078e-01 -7.96911955e-01 -6.42729342e-01 5.54338336e-01 -8.71456921e-01 -2.15931728e-01 -1.00193024e+00 -3.55313048e-02 -1.54507577e-01 -1.39839083e-01 -1.42102957e-01 1.69353589e-01 2.60876883e-02 1.95831880e-01 -1.71447515e-01 -2.00291827e-01 5.45559824e-01 4.26450759e-01 -4.53432441e-01 2.27671698e-01 -8.66765901e-02 -5.80626130e-01 4.31754649e-01 6.62261665e-01 -4.20411497e-01 -4.94469292e-02 -2.42161095e-01 5.00734374e-02 -7.57708699e-02 6.12740695e-01 -9.07342792e-01 2.11929902e-01 4.26555574e-01 -8.02220777e-02 -4.12883162e-01 6.66430652e-01 -7.20594347e-01 2.32900959e-03 7.76921436e-02 -5.01892030e-01 -5.04646488e-02 -7.27111846e-02 4.85312521e-01 -5.06517470e-01 -6.12060130e-01 5.79991221e-01 2.22445145e-01 -1.69934511e-01 1.04086868e-01 -3.90045255e-01 5.37294149e-02 7.05913365e-01 -2.87194878e-01 -1.09599248e-01 -3.45191211e-01 -8.25746536e-01 -1.14922985e-01 1.07083701e-01 4.89040971e-01 5.44174850e-01 -1.35143828e+00 -7.22866178e-01 2.52979249e-01 1.47884265e-01 -5.81367314e-01 2.40495697e-01 8.37268710e-01 -4.35713768e-01 3.41606408e-01 -1.88194569e-02 -3.60728651e-01 -1.42780101e+00 5.08049309e-01 3.28703076e-01 2.29147062e-01 -3.93629342e-01 7.33214080e-01 -2.17032302e-02 -7.27774382e-01 4.28104728e-01 -1.95845574e-01 -3.41599464e-01 5.78868389e-01 8.21207345e-01 6.31212234e-01 2.29074255e-01 -1.14690197e+00 -6.05155528e-01 1.82549939e-01 4.60993052e-02 -4.43154365e-01 1.32968497e+00 -9.11648497e-02 -5.84944896e-03 6.78169191e-01 1.56911933e+00 1.73424378e-01 -9.28610861e-01 -1.56151205e-01 2.61905193e-01 -5.76220632e-01 -1.83919072e-01 -1.97649539e-01 -1.00015068e+00 1.38870406e+00 6.72424495e-01 7.08060086e-01 6.03386521e-01 1.23434946e-01 6.73963249e-01 2.47987092e-01 2.66440421e-01 -1.19944108e+00 -1.88845471e-01 3.07211667e-01 7.65307367e-01 -1.25340497e+00 -5.63759148e-01 -1.56640053e-01 -4.04917002e-01 1.18753362e+00 -7.59137645e-02 3.10528055e-02 8.01386595e-01 1.15108140e-01 9.14356112e-02 -8.09384957e-02 -1.86462358e-01 -2.16159716e-01 -3.73815112e-02 8.49207342e-01 4.78245258e-01 1.28392711e-01 -9.57193300e-02 6.89357519e-01 -6.92664921e-01 -3.81410152e-01 6.87851369e-01 5.07300258e-01 -1.60115555e-01 -1.26342833e+00 -7.35261023e-01 2.43002802e-01 -9.23064411e-01 -4.25114818e-02 -7.66954184e-01 4.99410689e-01 4.18148823e-02 1.18499434e+00 1.31389171e-01 -3.28073204e-01 3.33151728e-01 5.10842502e-01 3.12165678e-01 -4.56233948e-01 -6.14832342e-01 -3.75073552e-01 1.19473122e-01 -1.08538091e-01 -4.44373727e-01 -1.03048289e+00 -8.13345671e-01 -4.10779595e-01 -2.93523908e-01 2.28232011e-01 1.00933695e+00 9.31870520e-01 1.70339182e-01 3.33896726e-01 5.47511101e-01 -7.99379170e-01 -8.22145820e-01 -1.01952493e+00 -9.92759645e-01 4.94384855e-01 6.18724585e-01 -8.17430079e-01 -6.77156746e-01 -1.15181223e-01]
[14.321846961975098, 6.046711444854736]
6b1f1145-0c19-461b-958d-be9ca721a8e7
sparse-winograd-convolutional-neural-networks
1810.01973
null
http://arxiv.org/abs/1810.01973v1
http://arxiv.org/pdf/1810.01973v1.pdf
Sparse Winograd Convolutional neural networks on small-scale systolic arrays
The reconfigurability, energy-efficiency, and massive parallelism on FPGAs make them one of the best choices for implementing efficient deep learning accelerators. However, state-of-art implementations seldom consider the balance between high throughput of computation power and the ability of the memory subsystem to support it. In this paper, we implement an accelerator on FPGA by combining the sparse Winograd convolution, clusters of small-scale systolic arrays, and a tailored memory layout design. We also provide an analytical model analysis for the general Winograd convolution algorithm as a design reference. Experimental results on VGG16 show that it achieves very high computational resource utilization, 20x ~ 30x energy efficiency, and more than 5x speedup compared with the dense implementation.
['Song-Chun Zhu', 'Haochen Li', 'Feng Shi', 'Yuhe Gao', 'Benjamin Kuschner']
2018-10-03
null
null
null
null
['layout-design']
['computer-vision']
[-6.26766860e-01 -4.06499594e-01 -2.53311574e-01 -2.62697339e-01 3.32357556e-01 -8.28645751e-02 4.08602089e-01 1.54904678e-01 -5.01733720e-01 4.56718296e-01 1.94817394e-01 -7.06111252e-01 -1.25626743e-01 -1.04136789e+00 -3.69546533e-01 -8.54696929e-01 -3.20480131e-02 -1.63181603e-01 3.23041886e-01 -6.06137700e-02 -1.42339140e-01 6.67786241e-01 -1.70206881e+00 1.96012750e-01 7.45280206e-01 1.20939064e+00 2.83713192e-01 2.78278112e-01 1.40328771e-02 6.04342103e-01 -5.13588071e-01 -2.27131516e-01 3.05850446e-01 -8.41951147e-02 -1.83460757e-01 -2.14179263e-01 2.99748957e-01 -4.62861627e-01 -5.91402948e-01 8.78936946e-01 5.57050645e-01 1.44127775e-02 3.63852054e-01 -1.02592969e+00 6.64898083e-02 7.93021917e-01 -7.47668207e-01 6.01753294e-01 -2.56365061e-01 6.20351508e-02 6.54595792e-01 -6.03447735e-01 1.06193855e-01 8.25083733e-01 6.98314905e-01 1.60984471e-01 -6.84571505e-01 -9.69529092e-01 -2.02564746e-01 3.78302395e-01 -1.26795864e+00 -2.36269996e-01 1.94892853e-01 -1.86539680e-01 1.48158419e+00 3.68906379e-01 1.33188188e+00 7.74261415e-01 8.75536203e-01 4.35140193e-01 8.20807874e-01 -3.71816456e-01 6.63706720e-01 -2.58917585e-02 5.89811802e-01 7.78180718e-01 9.86596942e-01 2.40030482e-01 -5.35902441e-01 4.70125787e-02 6.72969460e-01 1.80911615e-01 -2.93894503e-02 1.06012054e-01 -9.20456231e-01 7.84108520e-01 7.63771832e-01 2.42278978e-01 -3.10194016e-01 4.36253846e-01 8.10495615e-01 -1.64974943e-01 3.35587524e-02 3.54145676e-01 -1.58195838e-01 -1.54195800e-01 -1.08527863e+00 2.21313626e-01 5.16807199e-01 1.05152416e+00 2.26808861e-01 5.83383441e-01 1.59638282e-02 2.20696434e-01 2.88194150e-01 3.61144930e-01 9.17697728e-01 -2.90430635e-01 9.29933339e-02 5.37820339e-01 -4.83306676e-01 -5.51571488e-01 -9.01499450e-01 -9.08403695e-01 -1.28122878e+00 2.99277186e-01 1.59792319e-01 -3.05518150e-01 -6.21175468e-01 1.09279335e+00 3.56912076e-01 5.18756136e-02 2.07366914e-01 8.23553443e-01 1.09541726e+00 8.71665776e-01 3.37690353e-01 1.18599169e-01 1.94070578e+00 -1.26450539e+00 -6.18630409e-01 -2.95736223e-01 8.18892598e-01 -7.82838106e-01 6.80984914e-01 2.52812624e-01 -8.97765100e-01 -8.68555307e-01 -1.72140479e+00 -1.42331764e-01 -1.72229618e-01 8.05504322e-01 1.15347123e+00 9.51133072e-01 -8.25401247e-01 5.83841264e-01 -1.29783785e+00 -1.18696481e-01 3.42477143e-01 4.22061175e-01 1.80013418e-01 2.00141028e-01 -8.32605600e-01 5.75002074e-01 7.27998734e-01 -2.92592227e-01 -1.80785090e-01 -1.17379725e+00 -4.13736612e-01 6.64281905e-01 -1.81887850e-01 -9.24250901e-01 1.12870562e+00 -4.48948145e-01 -1.55240333e+00 3.64679962e-01 4.23470050e-01 -8.11125815e-01 5.69920428e-02 -8.76477733e-02 -5.48452497e-01 -1.15392089e-01 -4.93531674e-01 4.33881342e-01 4.44405347e-01 -4.20679189e-02 -8.86150002e-01 -2.29372442e-01 -7.06964210e-02 7.67740756e-02 -6.95158601e-01 -5.23568541e-02 1.52377784e-01 -5.94432890e-01 4.77219373e-02 -8.68189514e-01 -4.42799151e-01 -4.89420183e-02 -1.23433918e-01 -1.38132989e-01 9.45699513e-01 5.80148138e-02 1.37710345e+00 -2.47636271e+00 -3.53723437e-01 8.55905041e-02 2.53504515e-01 5.63839853e-01 3.87617111e-01 9.72068459e-02 -1.52392890e-02 -5.61610103e-01 6.88075185e-01 2.99989600e-02 -2.50412434e-01 8.11872259e-02 -5.25976241e-01 5.06529331e-01 -2.65624911e-01 6.01884067e-01 -6.15239620e-01 -1.20485395e-01 2.53512502e-01 6.91730618e-01 -7.66333342e-01 -1.20630255e-02 3.42420578e-01 -1.67917088e-01 -6.46303773e-01 4.20758516e-01 8.73299778e-01 -3.41402888e-01 4.95163977e-01 -7.42670476e-01 -6.72174692e-01 3.26007217e-01 -1.18637943e+00 1.35834432e+00 -4.59166110e-01 6.86104596e-01 -3.96870673e-02 -7.52827048e-01 8.12685549e-01 1.07027307e-01 2.49147072e-01 -8.18181574e-01 5.89332044e-01 3.28683823e-01 4.52467144e-01 -5.08070784e-03 8.56873035e-01 1.08139887e-01 -2.56984532e-02 3.72676224e-01 1.90225959e-01 3.17900062e-01 2.80155599e-01 -2.53134537e-02 9.64463711e-01 -3.58273923e-01 4.64254528e-01 -1.08742595e+00 2.26122603e-01 1.08676545e-01 4.04238582e-01 3.57534498e-01 2.12932765e-01 -2.07316831e-01 1.64342225e-01 -9.51795161e-01 -1.16150665e+00 -9.12994564e-01 -6.18470728e-01 8.09523165e-01 -3.18944082e-02 -9.91891205e-01 -7.70062208e-01 1.81824878e-01 -6.98790029e-02 4.28405195e-01 -3.21671553e-02 -1.99757829e-01 -7.77519166e-01 -1.29085410e+00 4.55414832e-01 9.94177401e-01 1.00752509e+00 -3.74503732e-01 -1.70285583e+00 2.97902703e-01 6.95821762e-01 -1.25484586e+00 -9.68854204e-02 5.49337327e-01 -1.10907531e+00 -6.42974377e-01 1.89279318e-01 -6.78519309e-01 5.52604437e-01 4.74045038e-01 9.83555079e-01 2.46517807e-01 -6.05111301e-01 -4.23683614e-01 -1.42042696e-01 -5.53725004e-01 3.34083699e-02 2.38282189e-01 3.73149544e-01 -5.84332287e-01 2.77704924e-01 -6.80769682e-01 -9.39411879e-01 6.81757033e-02 -3.70206654e-01 4.85353678e-01 7.79428542e-01 8.89476836e-01 6.38790965e-01 4.85321850e-01 2.41615802e-01 -5.29305160e-01 2.25734353e-01 -1.72857448e-01 -9.98119652e-01 -1.61175743e-01 -6.42501235e-01 1.32913470e-01 1.11619842e+00 -4.29330170e-01 -9.97193277e-01 4.48472172e-01 -4.69770938e-01 -1.15826800e-01 3.65721911e-01 1.63245142e-01 -2.14108720e-01 -2.68749505e-01 7.05651164e-01 1.69982001e-01 -2.14204174e-02 -1.61074474e-01 1.20849624e-01 5.34964681e-01 5.76837897e-01 -4.58612561e-01 4.29184973e-01 3.78145516e-01 3.91188800e-01 -1.10696614e+00 -4.74358708e-01 -8.96714255e-02 -2.42587164e-01 -1.05116762e-01 7.31176436e-01 -1.52067947e+00 -8.45449507e-01 3.30510169e-01 -6.55209482e-01 -2.65539676e-01 -1.50506839e-01 8.29016864e-01 -4.16511595e-02 -1.66584626e-01 -5.83710432e-01 -2.36907840e-01 -1.04541385e+00 -1.39526689e+00 6.38680995e-01 6.86351717e-01 -9.96872708e-02 -7.55729318e-01 -5.45352936e-01 -2.07882136e-01 7.38795877e-01 1.41770542e-01 8.09797645e-01 -2.04721898e-01 -7.71569967e-01 2.00393684e-02 -3.15900624e-01 -3.76836322e-02 -2.76513457e-01 1.29081875e-01 -8.98118198e-01 -5.02478480e-01 1.12516187e-01 2.91326605e-02 8.05194259e-01 5.22127748e-01 1.45840108e+00 -1.70277953e-01 -7.51669407e-01 1.15685213e+00 1.69280040e+00 2.63371050e-01 5.62318027e-01 2.02077240e-01 4.77234542e-01 -4.19327885e-01 4.93395865e-01 7.17478693e-01 -8.84808674e-02 8.05155098e-01 3.28597873e-01 -7.44426399e-02 -1.56806827e-01 -3.91446054e-03 1.42454030e-02 1.10377872e+00 -1.99041255e-02 -7.33241513e-02 -6.28148139e-01 -7.29982033e-02 -1.39608669e+00 -7.74367690e-01 -4.31131363e-01 1.85232508e+00 5.42308807e-01 4.19261009e-01 1.98665839e-02 3.97608221e-01 7.56046623e-02 5.68340644e-02 -1.38741404e-01 -7.29809999e-01 2.52246439e-01 6.50204062e-01 9.05559421e-01 -8.66827816e-02 -1.05481076e+00 4.86942619e-01 6.72991514e+00 1.28637910e+00 -1.48775613e+00 2.07056284e-01 3.58537912e-01 -4.73989427e-01 2.09247306e-01 -3.64353687e-01 -1.54081476e+00 6.60687745e-01 1.26395845e+00 -4.29946393e-01 -6.34770691e-02 1.29975724e+00 -5.64229162e-03 -1.08783469e-01 -1.07661247e+00 1.07013917e+00 -4.53434616e-01 -1.83170712e+00 -1.12147227e-01 2.77335584e-01 5.30164421e-01 7.62162032e-04 3.80020402e-02 1.63412005e-01 -4.02074531e-02 -8.31355691e-01 7.99900413e-01 -1.28917918e-01 6.79571688e-01 -1.16998994e+00 5.75820744e-01 2.91266412e-01 -1.52178037e+00 -4.32956010e-01 -8.21922123e-01 -4.66673732e-01 -8.95252973e-02 9.51031029e-01 -5.47260821e-01 2.57675976e-01 8.09697092e-01 4.21191543e-01 -3.80263865e-01 1.04568744e+00 -4.67491187e-02 5.10299563e-01 -3.83166492e-01 -5.84773481e-01 3.55974466e-01 -9.54649225e-02 -1.68597866e-02 1.38050103e+00 6.04037702e-01 4.50779498e-01 4.26901737e-03 2.31531411e-01 5.07929735e-02 1.31712826e-02 -1.97181076e-01 2.40179852e-01 6.77908599e-01 1.71296334e+00 -9.87355232e-01 -5.66901386e-01 -6.44230425e-01 2.97306061e-01 1.27527982e-01 -4.68429357e-01 -1.02742016e+00 -4.90890712e-01 1.26351547e+00 1.77923486e-01 3.01372796e-01 -5.39100409e-01 -8.14547062e-01 -9.16753829e-01 -2.68946350e-01 -4.02581990e-01 2.17651695e-01 -4.18258309e-01 -6.14262402e-01 8.48274708e-01 -2.23306775e-01 -1.17539883e+00 6.48607463e-02 -9.07634616e-01 -8.65345955e-01 6.08494461e-01 -1.15772188e+00 -7.41514504e-01 -7.12947607e-01 2.67691493e-01 4.32046562e-01 -5.65127313e-01 9.35048103e-01 5.90481102e-01 -7.95729458e-01 9.36018169e-01 3.50612015e-01 -2.08980322e-01 -1.00874901e-01 -7.47671783e-01 7.20901549e-01 8.69568706e-01 -2.16458961e-01 5.32899857e-01 7.24623442e-01 -2.59104937e-01 -2.01584888e+00 -1.30860996e+00 3.07444990e-01 3.96290898e-01 5.46740413e-01 -3.81056011e-01 -7.53533244e-01 4.29816127e-01 3.88384193e-01 2.40075022e-01 8.69064212e-01 -2.03502644e-02 -9.62807238e-02 -4.83067453e-01 -9.69822764e-01 5.45988977e-01 1.01936638e+00 1.64258718e-01 -6.59170747e-02 4.00737852e-01 3.44810903e-01 -1.03911006e+00 -8.46016765e-01 1.67306557e-01 6.79156899e-01 -9.39476669e-01 1.13445687e+00 -1.40475601e-01 4.39322680e-01 -2.22084343e-01 -5.52866869e-02 -1.22283041e+00 -6.43309414e-01 -4.47466597e-02 -1.64949477e-01 7.38750815e-01 -1.46162659e-01 -5.28128743e-01 6.95606351e-01 3.04143261e-02 -5.78322649e-01 -1.06168759e+00 -9.53476846e-01 -8.56850505e-01 -1.88334361e-01 -7.62950704e-02 8.70734811e-01 4.97438431e-01 3.29085022e-01 5.08430600e-01 -2.87937000e-04 1.81665927e-01 5.66312075e-01 3.21775168e-01 4.54945952e-01 -1.07716429e+00 -3.63876581e-01 -5.93013644e-01 -8.79736364e-01 -9.41276729e-01 -9.39614847e-02 -9.09740090e-01 -4.88430142e-01 -1.14341569e+00 8.89217630e-02 -6.36767149e-01 6.03199527e-02 3.92783105e-01 1.53311595e-01 3.07613850e-01 -6.38581738e-02 -2.19606861e-01 -1.82495594e-01 2.02496931e-01 9.39616919e-01 2.22024143e-01 -1.08925007e-01 -1.27622485e-01 -5.44733524e-01 7.67930925e-01 7.87198544e-01 -1.50005981e-01 -6.92457914e-01 -7.04339623e-01 -2.99207726e-03 -2.53637195e-01 -6.61057830e-02 -1.80973315e+00 5.56420267e-01 6.48820102e-02 8.56900275e-01 -5.94876111e-01 3.45827818e-01 -7.19722092e-01 5.66278100e-01 1.11714494e+00 1.77233994e-01 3.41785908e-01 6.39334083e-01 4.44348454e-02 1.72334816e-02 -6.76790103e-02 1.22670746e+00 2.48832807e-01 -8.10516834e-01 2.66993195e-01 -6.55775607e-01 -4.00167555e-01 1.29122925e+00 -3.21902558e-02 -5.77415943e-01 6.50833607e-01 -1.57903552e-01 -4.56488654e-02 5.54487295e-02 5.70266284e-02 4.98702705e-01 -1.50384331e+00 -3.52541983e-01 7.12914288e-01 -3.16744924e-01 -1.45338714e-01 5.23743570e-01 5.00917614e-01 -1.00910699e+00 8.11007798e-01 -1.05625415e+00 -6.45844221e-01 -1.52886736e+00 4.74629372e-01 3.08598667e-01 -1.33422866e-01 -1.31336367e+00 7.88363099e-01 8.27406794e-02 5.34144640e-01 6.23197928e-02 -5.68420351e-01 -4.16439697e-02 -1.93647668e-02 1.03409231e+00 4.28698808e-01 5.21006882e-01 -1.24953084e-01 -5.02629876e-01 3.87605637e-01 3.11056133e-02 7.45496929e-01 1.15028942e+00 4.37476963e-01 6.23269230e-02 -4.84790921e-01 9.24832404e-01 -2.08810210e-01 -8.53377879e-01 3.80387828e-02 -5.13406694e-01 -4.26258504e-01 6.71281099e-01 -1.91940203e-01 -1.51108801e+00 7.67195880e-01 7.67392337e-01 -1.94541737e-02 1.48865473e+00 -3.34929854e-01 8.65662754e-01 2.47830167e-01 4.17526662e-01 -1.01626670e+00 -9.21614096e-02 5.33752382e-01 2.49852717e-01 -4.64979351e-01 6.17001474e-01 -6.83931470e-01 -9.60581563e-03 1.48870647e+00 9.63208020e-01 -3.21461797e-01 1.00059104e+00 1.09927166e+00 -4.50588495e-01 -6.07237481e-02 -9.44907010e-01 2.91917771e-01 -4.48184200e-02 1.00409083e-01 4.39458072e-01 4.21849132e-01 -8.13633621e-01 8.17866921e-01 -9.24127460e-01 -1.82895195e-02 5.77605069e-01 9.60789859e-01 -6.13862991e-01 -7.86759019e-01 -6.24211170e-02 7.01080143e-01 -3.56850386e-01 -3.82661633e-02 7.66090453e-01 7.32816875e-01 1.33074909e-01 3.86383295e-01 8.08846831e-01 -7.11642802e-01 2.45310754e-01 -3.54083121e-01 5.55793941e-01 -2.16268659e-01 -9.59510863e-01 -2.53630546e-03 3.29108723e-02 -5.47288835e-01 4.29930091e-02 -3.38193387e-01 -1.51787305e+00 -9.19917226e-01 -4.96990234e-02 1.19048871e-01 1.10751653e+00 7.44393647e-01 5.52739024e-01 1.13716888e+00 3.83150667e-01 -6.85700595e-01 -5.55626094e-01 -5.39899290e-01 -8.71093333e-01 -4.64373976e-01 -6.83894008e-02 -7.78611541e-01 5.26230149e-02 -4.66429561e-01]
[8.32797908782959, 2.852698564529419]
d6fcaaea-307e-4c09-b057-5a95cdd9ffd0
handseg-an-automatically-labeled-dataset-for
1711.05944
null
http://arxiv.org/abs/1711.05944v4
http://arxiv.org/pdf/1711.05944v4.pdf
HandSeg: An Automatically Labeled Dataset for Hand Segmentation from Depth Images
We propose an automatic method for generating high-quality annotations for depth-based hand segmentation, and introduce a large-scale hand segmentation dataset. Existing datasets are typically limited to a single hand. By exploiting the visual cues given by an RGBD sensor and a pair of colored gloves, we automatically generate dense annotations for two hand segmentation. This lowers the cost/complexity of creating high quality datasets, and makes it easy to expand the dataset in the future. We further show that existing datasets, even with data augmentation, are not sufficient to train a hand segmentation algorithm that can distinguish two hands. Source and datasets will be made publicly available.
['Sri Raghu Malireddi', 'Andrea Tagliasacchi', 'Vincent Lepetit', 'Kwang Moo Yi', 'Franziska Mueller', 'Christian Theobalt', 'Abhishake Kumar Bojja', 'Markus Oberweger']
2017-11-16
null
null
null
null
['hand-segmentation']
['computer-vision']
[ 2.09061667e-01 7.17273429e-02 -1.94352403e-01 -3.05076122e-01 -6.62361920e-01 -1.06067550e+00 1.13159351e-01 -2.93637246e-01 -2.75647908e-01 5.97416818e-01 4.56217974e-02 -2.01267332e-01 2.75577515e-01 -6.52870595e-01 -4.28546041e-01 -3.88728648e-01 4.15358931e-01 7.69384027e-01 6.37123644e-01 6.25100732e-02 -2.90260725e-02 7.41740644e-01 -1.57790446e+00 4.66848351e-02 5.86551130e-01 8.15004051e-01 4.12231982e-01 1.06719220e+00 -7.91625381e-02 3.69166315e-01 -5.88180542e-01 -4.21519250e-01 6.98072910e-01 -5.27260363e-01 -1.32533479e+00 4.55467373e-01 5.29798031e-01 -8.69686127e-01 -3.87879908e-01 6.70781970e-01 9.25495982e-01 -3.40953730e-02 3.66972208e-01 -1.33778834e+00 -4.54569459e-01 3.85462314e-01 -5.06104708e-01 -1.17989220e-01 6.77191913e-01 5.40608585e-01 8.71017218e-01 -5.04681706e-01 1.06850016e+00 1.04132402e+00 4.82764006e-01 1.03924608e+00 -1.06767416e+00 -2.61530906e-01 3.60658944e-01 -3.36669564e-01 -1.40466070e+00 -1.38722956e-01 8.72160912e-01 -6.71698570e-01 8.95202041e-01 1.92747235e-01 1.04973674e+00 1.22216189e+00 -7.96278596e-01 1.35557258e+00 1.17850494e+00 -5.25831223e-01 1.18799768e-01 -1.97442934e-01 8.16639066e-02 9.06308770e-01 1.32036909e-01 6.28558472e-02 -3.81986350e-01 8.81204978e-02 1.37187731e+00 -6.83877915e-02 -3.46855819e-01 -5.97682595e-01 -1.37650621e+00 3.63079906e-01 4.45163548e-01 2.71110713e-01 -4.75588709e-01 -3.65817472e-02 1.31924152e-01 -1.48297742e-01 4.55560572e-02 2.39001885e-01 -6.47567391e-01 -3.18668216e-01 -1.22955346e+00 4.31477100e-01 7.55728960e-01 1.42743099e+00 5.98089457e-01 -4.90361303e-01 -1.38446808e-01 3.63689244e-01 2.84434944e-01 5.59731781e-01 -5.48221990e-02 -1.24006093e+00 3.81172061e-01 7.90623605e-01 4.05471921e-01 -5.95319510e-01 -5.45041382e-01 1.57692611e-01 -3.40981781e-01 5.27230918e-01 1.10501170e+00 -2.35338986e-01 -1.23184443e+00 1.20234501e+00 3.38611037e-01 -2.24142835e-01 -3.55952233e-01 1.36039352e+00 8.01308095e-01 -1.34688646e-01 -1.29294485e-01 2.68198937e-01 1.23305118e+00 -1.06059349e+00 -7.38088667e-01 -3.23056638e-01 4.16856468e-01 -6.69802547e-01 1.58183539e+00 7.41310537e-01 -1.10382760e+00 -4.38080400e-01 -8.28031182e-01 -3.38883877e-01 -4.27484423e-01 1.01186179e-01 8.97296786e-01 8.59202564e-01 -1.00385249e+00 4.66820121e-01 -1.15163410e+00 -3.89281392e-01 6.68748260e-01 3.04397821e-01 -3.71062398e-01 -6.53462708e-02 -7.08521307e-01 5.37105799e-01 3.85565966e-01 1.34881988e-01 -6.98901057e-01 -2.31500208e-01 -7.01559126e-01 -7.01776206e-01 3.59287590e-01 -6.14411950e-01 1.38375187e+00 -7.46798754e-01 -1.54030657e+00 1.23913872e+00 -1.55866101e-01 1.71657994e-01 8.83018613e-01 -1.50807798e-01 2.39160493e-01 3.45238507e-01 -2.15277560e-02 1.06079984e+00 6.02610707e-01 -1.62290025e+00 -6.51222885e-01 -7.44151831e-01 2.32324421e-01 -9.50762406e-02 -8.79992098e-02 9.93690193e-02 -1.05600917e+00 -6.39620662e-01 7.90652707e-02 -1.08370733e+00 -2.44651645e-01 2.87839565e-02 -7.97822118e-01 -5.88143505e-02 6.98566437e-01 -1.14263344e+00 9.72169399e-01 -1.64466870e+00 4.84243661e-01 2.58986354e-01 3.37577760e-01 2.03844771e-01 -1.50008261e-01 -1.53435707e-01 4.70993847e-01 1.96752831e-01 -4.23596472e-01 -7.67872393e-01 5.88935874e-02 3.75809312e-01 -4.93205637e-02 1.19930431e-01 -7.64139071e-02 1.08480763e+00 -1.09170651e+00 -7.96097159e-01 4.10630345e-01 7.85986364e-01 -2.93594778e-01 4.77549791e-01 -5.85206449e-01 9.33640718e-01 -2.84128636e-01 1.09652841e+00 5.54874957e-01 -1.29752144e-01 -3.90243754e-02 -7.12255910e-02 9.81201530e-02 8.24127793e-02 -1.48158777e+00 2.23881721e+00 -2.83965170e-01 5.29508412e-01 1.73854381e-01 -1.11854404e-01 5.21187663e-01 4.35870975e-01 3.13575208e-01 -2.45124042e-01 4.12861794e-01 1.85283184e-01 -3.09225738e-01 -4.14155751e-01 3.31232339e-01 1.85081244e-01 1.33798465e-01 7.79444158e-01 3.45952585e-02 -4.82424676e-01 2.55574852e-01 5.43512702e-02 8.24018419e-01 5.37482083e-01 -1.93278894e-01 3.22372288e-01 9.11910385e-02 1.86879173e-01 3.74302477e-01 6.70509756e-01 -4.18008626e-01 1.29105091e+00 3.43743831e-01 -3.03113937e-01 -1.16713178e+00 -9.01702642e-01 1.04507357e-01 1.09907925e+00 3.35405022e-02 -2.75678992e-01 -1.23879695e+00 -9.40380633e-01 4.23693918e-02 3.57414447e-02 -6.40777588e-01 7.08054423e-01 -5.48746288e-01 -1.28580287e-01 7.25702941e-01 1.24823868e+00 4.70549226e-01 -1.16122341e+00 -8.51341426e-01 -3.72757502e-02 -2.85374492e-01 -1.42674649e+00 -5.53871095e-01 -1.78793427e-02 -9.21091855e-01 -1.41497672e+00 -1.22918224e+00 -8.11098874e-01 6.73890710e-01 -1.28777847e-02 1.17328167e+00 2.47510076e-01 -5.66919029e-01 7.15888619e-01 -4.06668425e-01 -3.89004260e-01 -9.46467146e-02 5.27081132e-01 -6.45970032e-02 -5.18380702e-01 2.82633990e-01 -3.97071123e-01 -9.19738412e-01 3.29591066e-01 -5.64740300e-01 -7.90081266e-03 4.17419076e-01 3.48324329e-01 8.18848252e-01 -4.35594529e-01 6.85005710e-02 -6.24279499e-01 5.32613933e-01 2.78791159e-01 -6.09875739e-01 1.38555154e-01 -3.51271272e-01 5.98206110e-02 1.44680291e-01 -4.05073553e-01 -1.03677082e+00 8.47788513e-01 -2.93066889e-01 -4.65771437e-01 -6.96762443e-01 -2.62231380e-01 -3.68233562e-01 -1.13907717e-01 6.77074492e-01 -3.50317545e-02 -2.38638103e-01 -8.22330117e-01 7.81578243e-01 8.95460129e-01 7.30588078e-01 -5.87559164e-01 5.54810286e-01 5.71938872e-01 -2.54959375e-01 -5.54410219e-01 -7.22229838e-01 -4.69022125e-01 -1.63165236e+00 -1.37044087e-01 1.14208674e+00 -5.82637370e-01 -7.11903512e-01 7.88159609e-01 -1.28157270e+00 -9.00906444e-01 -3.02246600e-01 1.77175850e-01 -6.97754860e-01 5.42954385e-01 -6.21099055e-01 -8.67090583e-01 -2.34771252e-01 -1.10804522e+00 1.50793362e+00 2.04691783e-01 -3.78556877e-01 -9.22541261e-01 -4.25813571e-02 5.95554650e-01 -1.51314870e-01 3.99185896e-01 3.33752960e-01 -1.87029734e-01 -7.14600086e-01 -2.99561560e-01 -2.44332567e-01 2.60243297e-01 2.80126005e-01 1.59460440e-01 -1.17494118e+00 -2.10118294e-01 -6.38050854e-01 -4.10505056e-01 6.74828053e-01 3.18432301e-01 1.27264404e+00 6.87987357e-02 -5.01401186e-01 4.57199275e-01 9.67037916e-01 -9.31279510e-02 5.40735424e-01 2.30533093e-01 1.18393373e+00 5.58006823e-01 4.53667819e-01 4.14694130e-01 5.22656679e-01 5.77129006e-01 2.63424844e-01 -2.52599716e-01 -5.54894090e-01 -3.56665909e-01 -3.56116086e-01 2.88098186e-01 -8.14434171e-01 -2.60770351e-01 -1.28715408e+00 8.26279819e-01 -1.64051402e+00 -5.81388652e-01 -2.07032964e-01 2.05942035e+00 1.12745667e+00 -2.13779025e-02 9.28862572e-01 6.12651467e-01 5.68569183e-01 -2.36477524e-01 -5.60083210e-01 7.59205073e-02 -1.01001084e-01 4.79589522e-01 4.39878106e-01 6.43714964e-01 -1.18629551e+00 1.26306367e+00 7.32708311e+00 6.46810234e-02 -7.29273438e-01 1.33380860e-01 1.62821844e-01 -2.49138735e-02 -1.48533031e-01 -1.66885033e-01 -7.23913729e-01 2.12385878e-01 2.34596759e-01 6.02681279e-01 6.53544903e-01 9.19581473e-01 -9.42102000e-02 -2.06260428e-01 -1.25528359e+00 1.16408920e+00 -8.51087272e-03 -8.17914844e-01 -2.02516481e-01 1.31597891e-01 7.70678401e-01 -2.18963698e-01 -2.22205848e-01 -3.63298804e-01 4.03464079e-01 -1.10256016e+00 7.58759022e-01 4.28234488e-01 7.73461640e-01 -5.14898360e-01 5.63479960e-01 3.98220867e-01 -1.22255898e+00 2.82619268e-01 6.37156740e-02 -1.56905591e-01 3.38284045e-01 3.08235973e-01 -7.13064909e-01 1.58876002e-01 7.79213607e-01 3.18046600e-01 -7.47155726e-01 9.04136360e-01 -7.69435287e-01 2.84238935e-01 -3.82380962e-01 1.27555400e-01 -2.65244305e-01 1.97465003e-01 3.83758008e-01 1.23597312e+00 -2.33605012e-01 2.18044624e-01 3.63807440e-01 9.12187815e-01 -3.49945538e-02 -2.44082034e-01 -4.95003998e-01 -1.53162241e-01 4.92219210e-01 1.17686296e+00 -1.23689377e+00 -3.12092572e-01 -3.53071392e-01 1.81958699e+00 1.62185013e-01 3.13422382e-01 -3.59032631e-01 -7.55237520e-01 6.31251276e-01 1.78832382e-01 2.69444793e-01 -7.32768297e-01 -6.92844152e-01 -1.10907876e+00 3.16721112e-01 -7.03109622e-01 2.92575806e-01 -7.86301136e-01 -1.20165932e+00 5.41394949e-01 -2.48727545e-01 -7.80993342e-01 -2.77025431e-01 -9.05517459e-01 1.97973102e-02 1.02583325e+00 -1.25113094e+00 -1.55879772e+00 -9.28286374e-01 8.27509046e-01 3.75287980e-01 1.99002296e-01 1.11900795e+00 2.02849716e-01 -4.70040113e-01 7.05962837e-01 -6.21116340e-01 7.19741702e-01 5.43154895e-01 -1.57448936e+00 6.78256989e-01 7.25073457e-01 2.09860161e-01 4.17173505e-01 5.21462917e-01 -7.18754470e-01 -1.30442631e+00 -7.14812517e-01 4.37452704e-01 -1.17045891e+00 1.24848977e-01 -5.86142659e-01 -6.80663168e-01 9.32236135e-01 1.99885443e-02 2.06913054e-01 7.23476887e-01 1.05107829e-01 -4.80474412e-01 4.45959777e-01 -1.49876344e+00 3.04619104e-01 1.64056492e+00 -7.63150573e-01 -5.73034644e-01 3.02587509e-01 4.61400330e-01 -9.09481406e-01 -9.43414509e-01 1.42196372e-01 7.91444063e-01 -8.94488096e-01 1.00616658e+00 -6.03098333e-01 1.41599894e-01 -3.41275603e-01 -2.68728286e-03 -8.88040602e-01 1.11944102e-01 -5.99809766e-01 -4.41071630e-01 1.33648396e+00 1.55630380e-01 -2.93998629e-01 1.27799463e+00 1.05315590e+00 2.83809751e-01 -3.85708869e-01 -5.11922240e-01 -8.53857219e-01 4.51779030e-02 -6.45796597e-01 7.92963445e-01 7.04970002e-01 6.68738410e-03 -3.75082120e-02 -2.71196030e-02 2.39153400e-01 8.76918018e-01 2.02833906e-01 8.87051582e-01 -1.37621260e+00 -1.76781788e-01 -4.75105435e-01 -4.74154681e-01 -1.30180657e+00 1.31743863e-01 -6.04444563e-01 2.21838504e-01 -1.99547231e+00 1.19687811e-01 -5.73852479e-01 1.42049998e-01 9.05263245e-01 -2.48414233e-01 7.91655242e-01 2.16154262e-01 1.62034929e-01 -4.35083836e-01 -8.90045017e-02 1.55304265e+00 -5.95805049e-03 -5.73092937e-01 7.08753914e-02 -4.31715369e-01 8.57127011e-01 8.42629194e-01 -1.33853152e-01 -1.99067742e-01 -5.90795577e-01 -1.68545786e-02 -2.26708606e-01 5.93539059e-01 -8.71040165e-01 -1.09540820e-02 -7.29232207e-02 6.60302043e-01 -6.89324498e-01 1.66431233e-01 -7.61904180e-01 -3.00025702e-01 2.56298304e-01 -1.57868087e-01 -2.87361681e-01 6.54457435e-02 2.64001489e-01 9.98775363e-02 9.09022707e-03 6.22711718e-01 -3.97238374e-01 -7.02568471e-01 3.87146413e-01 -1.69510007e-01 6.65334836e-02 1.02136016e+00 -4.94262338e-01 1.20157346e-01 -2.36546591e-01 -9.89494681e-01 1.63412288e-01 8.14006627e-01 5.26711404e-01 4.74242032e-01 -1.19719160e+00 -2.56940305e-01 2.54212558e-01 -6.05819970e-02 6.07672155e-01 -1.59200981e-01 3.68441850e-01 -6.72692478e-01 4.03127849e-01 -3.75988811e-01 -5.75640976e-01 -1.37313890e+00 5.47987401e-01 3.91896576e-01 2.58969337e-01 -7.03379452e-01 1.03671968e+00 -4.42978024e-01 -6.81626618e-01 6.30857944e-01 -5.80509126e-01 2.06750140e-01 -3.08617414e-03 5.93707800e-01 5.87075949e-01 9.25181061e-02 -6.37823045e-01 -5.08178711e-01 7.51604438e-01 4.65942353e-01 -3.77001196e-01 1.00341845e+00 -4.35037613e-02 1.49239637e-02 3.94571245e-01 8.08445692e-01 2.84570195e-02 -1.63939834e+00 -3.74710336e-02 -2.43048057e-01 -7.39801705e-01 -5.10664135e-02 -1.01298916e+00 -1.23167956e+00 1.24540627e+00 7.80490458e-01 5.90553656e-02 1.21103871e+00 3.70251596e-01 1.12631631e+00 3.36603552e-01 5.52981615e-01 -1.29541337e+00 1.55425727e-01 4.83279228e-01 7.86086738e-01 -1.20628834e+00 -2.54020482e-01 -7.31824100e-01 -6.51609242e-01 1.12099564e+00 7.42503107e-01 2.46662498e-01 3.68207246e-01 6.42449558e-01 3.59136045e-01 -1.14784330e-01 3.14378828e-01 -8.72059643e-01 2.47315541e-01 1.08488977e+00 5.22960603e-01 2.39716083e-01 1.08243778e-01 7.48941243e-01 -2.71127343e-01 3.69615078e-01 1.52586266e-01 1.36763251e+00 -9.12449285e-02 -1.49211717e+00 -4.62705463e-01 2.27453485e-01 -1.17618427e-01 3.03186029e-01 -1.06494236e+00 6.99979305e-01 2.98750967e-01 8.95655930e-01 -5.19265421e-02 -5.44549525e-01 6.69048131e-01 2.58865863e-01 1.21699083e+00 -6.53587222e-01 -3.12105179e-01 -1.35250047e-01 -3.09326351e-01 -5.25072694e-01 -6.02764726e-01 -5.02458394e-01 -1.47345960e+00 -2.18697891e-01 -2.62098104e-01 -5.71878016e-01 4.46129888e-01 9.10345674e-01 3.98127496e-01 3.12514186e-01 1.39151230e-01 -1.38963568e+00 -8.24534371e-02 -8.33530545e-01 -6.32678986e-01 3.95027995e-01 2.68136799e-01 -7.91735530e-01 -2.37055719e-02 3.61529917e-01]
[6.614871025085449, -0.6746015548706055]
61a92636-85c9-4466-93cc-47fb449ffe9d
learning-by-aligning-2d-skeleton-sequences-in
2305.19480
null
https://arxiv.org/abs/2305.19480v2
https://arxiv.org/pdf/2305.19480v2.pdf
Learning by Aligning 2D Skeleton Sequences in Time
This paper presents a novel self-supervised temporal video alignment framework which is useful for several fine-grained human activity understanding applications. In contrast with the state-of-the-art method of CASA, where sequences of 3D skeleton coordinates are taken directly as input, our key idea is to use sequences of 2D skeleton heatmaps as input. Unlike CASA which performs self-attention in the temporal domain only, we feed 2D skeleton heatmaps to a video transformer which performs self-attention both in the spatial and temporal domains for extracting effective spatiotemporal and contextual features. In addition, we introduce simple heatmap augmentation techniques based on 2D skeletons for self-supervised learning. Despite the lack of 3D information, our approach achieves not only higher accuracy but also better robustness against missing and noisy keypoints than CASA. Furthermore, extensive evaluations on three public datasets, i.e., Penn Action, IKEA ASM, and H2O, demonstrate that our approach outperforms previous methods in different fine-grained human activity understanding tasks. Finally, fusing 2D skeleton heatmaps with RGB videos yields the state-of-the-art on all metrics and datasets. To the best of our knowledge, our work is the first to utilize 2D skeleton heatmap inputs and the first to explore multi-modality fusion for temporal video alignment.
['M. Zeeshan Zia', 'Andrey Konin', 'Murad Popattia', 'M. Hassan Ahmed', 'Ahmed Mehmood', 'Muhammad Ahmed', 'Quoc-Huy Tran']
2023-05-31
null
null
null
null
['video-alignment']
['computer-vision']
[ 3.80930752e-01 -2.19832599e-01 -3.70850891e-01 -2.03847975e-01 -6.82049155e-01 -3.43851268e-01 7.24738359e-01 -3.02118585e-02 -5.41669965e-01 3.80115569e-01 4.26188111e-01 8.93470645e-02 -1.22252978e-01 -6.82981551e-01 -8.66694570e-01 -4.75872874e-01 -1.16728634e-01 2.66088784e-01 4.29211497e-01 -3.13694268e-01 1.40530661e-01 3.53085458e-01 -1.69154751e+00 2.53384709e-01 5.59444070e-01 1.20848858e+00 -1.47439808e-01 5.93835056e-01 -2.04341523e-02 8.35793376e-01 -2.57627398e-01 -2.16905668e-01 4.24890637e-01 -5.38038850e-01 -1.00393105e+00 2.35874549e-01 7.06942618e-01 -4.17883247e-01 -5.25937855e-01 6.77256227e-01 3.54693651e-01 3.69010687e-01 3.49779963e-01 -1.43197548e+00 -3.17760110e-01 1.29875436e-01 -8.26724172e-01 3.03241789e-01 9.68307316e-01 3.53613287e-01 8.62957358e-01 -7.36573458e-01 7.00095296e-01 1.10120881e+00 7.40428030e-01 2.87906736e-01 -1.00314164e+00 -6.11703515e-01 2.49422312e-01 5.24235904e-01 -1.19637477e+00 -2.93428093e-01 9.12509263e-01 -3.61608177e-01 1.12834644e+00 7.04904795e-02 1.11944950e+00 1.44559729e+00 -1.63025290e-01 1.10382414e+00 1.14221704e+00 -3.88963789e-01 1.53146893e-01 -7.04070032e-01 2.89312657e-02 9.21322823e-01 -1.75797358e-01 -4.35433984e-02 -1.05723202e+00 2.20419485e-02 1.00353372e+00 1.66796997e-01 -1.97433949e-01 -6.79909706e-01 -1.86478090e+00 4.67863560e-01 2.59516954e-01 3.00484449e-01 -5.71059167e-01 3.99161726e-01 5.08304000e-01 1.30252391e-01 2.89586693e-01 2.07754314e-01 -3.66502613e-01 -8.09318662e-01 -1.10049772e+00 1.81971639e-01 3.62110794e-01 6.92950666e-01 7.08474338e-01 -1.03759468e-01 -1.50065973e-01 5.65183640e-01 5.01731411e-02 4.69013870e-01 7.63641238e-01 -1.27633488e+00 5.94949722e-01 6.68775678e-01 6.55726939e-02 -9.57790434e-01 -4.73071158e-01 6.31200895e-03 -6.13206446e-01 1.90446436e-01 6.32363617e-01 3.05425793e-01 -1.09937429e+00 1.71167505e+00 3.85353953e-01 5.30738831e-01 7.18846626e-04 1.01762950e+00 5.41657567e-01 4.78723913e-01 6.60205856e-02 -7.01573342e-02 1.30428874e+00 -1.38974357e+00 -7.76020050e-01 -1.68072328e-01 4.85155910e-01 -4.46173489e-01 1.30848789e+00 3.04637641e-01 -1.15915167e+00 -8.50883603e-01 -1.11464298e+00 -1.74548909e-01 -4.95560408e-01 -7.05050305e-02 6.51001096e-01 3.76768470e-01 -8.54778171e-01 5.73319852e-01 -1.25536108e+00 -8.88042688e-01 4.96929348e-01 -2.21556015e-02 -9.07337904e-01 -5.87110072e-02 -1.06876481e+00 6.82068884e-01 2.82556862e-01 -1.91704318e-01 -7.34809816e-01 -6.66440427e-01 -1.19477129e+00 -3.39074135e-01 6.08940005e-01 -7.45699823e-01 1.21335828e+00 -7.13453293e-01 -1.49178386e+00 8.52438390e-01 -3.04671735e-01 -6.74444616e-01 7.46901751e-01 -6.40695274e-01 -1.34191707e-01 5.37099540e-01 2.97292829e-01 8.65114510e-01 8.33879948e-01 -7.37415433e-01 -6.23779058e-01 -4.30573732e-01 2.09094331e-01 4.20977801e-01 -3.31100076e-01 -2.07915261e-01 -9.71225142e-01 -1.04507089e+00 2.20814720e-01 -9.78347301e-01 -9.22798142e-02 2.72036850e-01 -1.54745743e-01 -1.15015723e-01 1.08719373e+00 -6.08906507e-01 1.17808008e+00 -2.01873851e+00 4.82606530e-01 1.89408258e-01 -3.44725028e-02 1.95483249e-02 -1.20233282e-01 2.20476344e-01 -8.94370079e-02 -1.28798112e-01 -3.76691252e-01 -4.81965005e-01 -1.85366906e-02 3.60550851e-01 7.19963610e-02 5.42618215e-01 1.39381036e-01 1.17418730e+00 -1.12393105e+00 -6.64640248e-01 6.80398762e-01 5.54328144e-01 -3.79948884e-01 2.00007722e-01 1.05116822e-01 6.16225898e-01 -3.88719290e-01 8.41075778e-01 1.58785164e-01 -3.03183228e-01 -1.89493179e-01 -5.19308269e-01 5.39033748e-02 2.82418206e-02 -1.08331859e+00 2.57812142e+00 -4.01410311e-01 5.47324419e-01 -4.52949792e-01 -9.86056983e-01 5.30130327e-01 2.28465393e-01 1.11398685e+00 -9.64749336e-01 -9.46525410e-02 -1.19833894e-01 -4.60440934e-01 -5.80642104e-01 4.87781286e-01 2.27629244e-01 -2.98606098e-01 5.73367357e-01 2.53740221e-01 -7.80456439e-02 3.21823597e-01 2.55970150e-01 1.25994146e+00 8.53610873e-01 4.86714065e-01 2.18104497e-01 5.02846539e-01 6.64674789e-02 4.79969382e-01 5.99968791e-01 -5.08987486e-01 9.32103872e-01 2.19807431e-01 -5.06267428e-01 -1.03476894e+00 -1.10613644e+00 2.89026409e-01 1.12535918e+00 3.17716986e-01 -7.25168526e-01 -8.27169240e-01 -8.16780508e-01 -1.84119821e-01 1.93699569e-01 -9.36581612e-01 9.30412486e-02 -7.18270659e-01 -3.94845337e-01 7.10244954e-01 1.02456784e+00 1.01724029e+00 -9.57297862e-01 -1.07764030e+00 1.16445549e-01 -6.98661625e-01 -1.50702345e+00 -6.92957401e-01 -5.37579358e-02 -8.60747278e-01 -1.41355884e+00 -7.57558644e-01 -2.90195107e-01 3.47464621e-01 4.48641568e-01 8.94885361e-01 -1.74782768e-01 -3.33139867e-01 1.02477086e+00 -6.48407280e-01 -5.36734946e-02 8.06411877e-02 -1.01550668e-01 7.67122880e-02 1.82672128e-01 3.37375224e-01 -5.77883959e-01 -6.69373572e-01 5.38885474e-01 -7.80360460e-01 2.47998819e-01 4.74423289e-01 7.11902082e-01 8.54043901e-01 -8.58489498e-02 1.30377442e-01 -4.34115410e-01 1.50630474e-01 -3.06374729e-01 -1.47732407e-01 2.85745889e-01 -2.06195921e-01 3.69950831e-02 3.26547563e-01 -3.54611367e-01 -1.01149511e+00 3.83747250e-01 -8.94285459e-03 -6.90589011e-01 -5.31565964e-01 2.76283354e-01 -8.26564357e-02 1.81199256e-02 5.96604347e-01 2.14492664e-01 7.40634352e-02 -4.65987384e-01 3.93627286e-01 2.26801202e-01 9.56059515e-01 -5.62061846e-01 8.94254804e-01 9.61574733e-01 9.28158015e-02 -8.60141993e-01 -7.57289469e-01 -8.93314242e-01 -1.16490507e+00 -3.94856691e-01 1.30092144e+00 -9.92421627e-01 -4.82217610e-01 8.43024731e-01 -8.01948428e-01 -4.40082252e-01 -4.09931839e-01 4.64102954e-01 -1.14369965e+00 7.27009714e-01 -3.51279587e-01 -5.58508694e-01 -1.90276161e-01 -1.02267158e+00 1.49133408e+00 -2.14307755e-03 -4.25126284e-01 -8.73068571e-01 1.32273898e-01 8.26160312e-01 1.28614649e-01 6.89932883e-01 1.73859984e-01 -3.24920028e-01 -5.22204876e-01 -1.02428891e-01 8.98756087e-04 8.94267559e-02 2.91512609e-01 -2.26624414e-01 -8.89820993e-01 -1.42032772e-01 -3.51111740e-01 -5.15175939e-01 9.06773746e-01 3.07791322e-01 1.10742915e+00 -9.66641530e-02 -2.16250539e-01 8.18416119e-01 8.96904588e-01 3.14385444e-03 6.86625302e-01 6.44690752e-01 8.70467186e-01 4.55045998e-01 1.11412382e+00 4.78904635e-01 6.40722513e-01 1.04377770e+00 4.56501365e-01 -2.96260297e-01 -3.32664192e-01 -4.09906358e-01 4.28760082e-01 4.89315033e-01 -6.20612681e-01 9.91088748e-02 -8.77752364e-01 6.15063965e-01 -2.27080631e+00 -1.27446210e+00 1.84330106e-01 2.01688290e+00 6.09992623e-01 6.11478351e-02 5.13897598e-01 4.19146806e-01 3.24437290e-01 4.97472107e-01 -5.75146735e-01 1.88722655e-01 -2.74192989e-01 2.92636603e-01 5.03958702e-01 1.99850112e-01 -1.53050089e+00 7.76517332e-01 5.61031103e+00 6.36410475e-01 -7.43326128e-01 1.65075392e-01 2.54499048e-01 -2.70503193e-01 2.23703712e-01 -1.80697083e-01 -2.74605483e-01 4.39749748e-01 6.57189548e-01 2.62829214e-01 4.02678728e-01 6.41087413e-01 2.51637846e-01 -2.60148972e-01 -1.05937028e+00 1.26057673e+00 2.44793445e-01 -1.26834297e+00 -1.89107731e-01 -1.15667462e-01 7.06483603e-01 9.20842737e-02 -1.93570167e-01 6.92006052e-02 1.27999678e-01 -8.33310783e-01 8.43473434e-01 5.76172650e-01 7.07238853e-01 -5.92107177e-01 5.72394073e-01 8.64355862e-02 -1.57953393e+00 -6.72009066e-02 3.33832711e-01 1.16728218e-02 5.12347043e-01 2.21289545e-01 -3.93887877e-01 7.66104579e-01 1.29572785e+00 1.27150285e+00 -6.99863255e-01 8.63760412e-01 -2.31093645e-01 4.34977621e-01 -5.18007755e-01 5.33936679e-01 3.83726865e-01 -3.85847650e-02 3.68333548e-01 1.33818877e+00 2.71329671e-01 1.23216853e-01 2.91103750e-01 2.59596378e-01 1.55761883e-01 -1.03243701e-02 -6.04551435e-01 5.82710095e-02 2.55821258e-01 9.31070507e-01 -7.01700509e-01 -5.09719372e-01 -7.09825277e-01 1.38336694e+00 1.00170940e-01 3.98569822e-01 -1.05922377e+00 -3.20154130e-01 8.29994142e-01 2.29127645e-01 3.65204930e-01 -5.37666321e-01 -1.00357801e-01 -1.31025600e+00 1.49439961e-01 -9.34591889e-01 7.68237591e-01 -8.83418977e-01 -9.16831791e-01 2.65134603e-01 3.76330823e-01 -1.36703634e+00 -4.42532778e-01 -4.34456289e-01 -4.78100717e-01 2.67257988e-01 -1.37900293e+00 -1.58003962e+00 -7.90847778e-01 1.08638799e+00 8.17957759e-01 -4.20098417e-02 5.81494272e-01 2.01369971e-01 -3.67969543e-01 4.23713088e-01 -3.53217125e-01 3.23606402e-01 8.11914861e-01 -1.17953610e+00 7.33224690e-01 9.95033443e-01 4.49559629e-01 4.20979112e-01 5.07003069e-01 -6.22313142e-01 -1.43309534e+00 -9.32994843e-01 4.01689440e-01 -6.54876351e-01 7.40409732e-01 -2.04068705e-01 -7.66762078e-01 7.01557875e-01 1.78076416e-01 3.78035963e-01 5.91369748e-01 -1.86069235e-01 -4.34693933e-01 8.29990022e-03 -9.64467406e-01 5.02337039e-01 1.69079602e+00 -6.66647017e-01 -6.87625766e-01 1.68979958e-01 5.86793244e-01 -5.48865259e-01 -1.18354464e+00 6.69499576e-01 8.90222847e-01 -1.12857234e+00 1.29995167e+00 -5.58861494e-01 1.69911101e-01 -6.31469369e-01 -3.67076248e-01 -1.10777998e+00 -7.31223747e-02 -5.52947760e-01 -5.53343415e-01 7.39672303e-01 -1.41377181e-01 -1.45708799e-01 9.48936462e-01 1.74548045e-01 -5.50654791e-02 -6.18992209e-01 -1.00881577e+00 -7.95599163e-01 -4.87330914e-01 -8.17647099e-01 4.68505353e-01 8.93058300e-01 2.39214599e-02 -8.97673890e-02 -5.09914339e-01 -9.04935524e-02 8.54942560e-01 9.93431211e-02 1.09917760e+00 -9.68774319e-01 -1.68415770e-01 -3.30006242e-01 -7.73964345e-01 -1.18632436e+00 1.91868827e-01 -4.20117497e-01 -7.36569464e-02 -1.44729221e+00 1.19259447e-01 2.35718451e-02 -4.12174940e-01 9.08293784e-01 -1.21791875e-02 7.23261297e-01 3.79811853e-01 1.60096064e-01 -9.27906573e-01 5.20573795e-01 1.10747433e+00 -2.12713704e-01 -2.31136948e-01 -2.26510510e-01 -2.03789338e-01 8.14433575e-01 7.63575435e-01 -1.81911420e-02 -5.08293569e-01 -2.00006485e-01 -1.07281573e-01 -4.33488563e-02 7.37755418e-01 -1.25910151e+00 3.42317373e-01 -2.62578636e-01 3.84112149e-01 -8.94777238e-01 6.97403848e-01 -8.87218416e-01 1.38440266e-01 2.79182613e-01 -1.27788514e-01 2.22872287e-01 1.88056886e-01 7.14803457e-01 -3.70235115e-01 3.76322627e-01 5.44750154e-01 -2.85328537e-01 -1.14489853e+00 3.75612825e-01 -2.53867894e-01 5.49845509e-02 1.16425622e+00 -7.62817681e-01 -1.36774585e-01 -5.00712097e-01 -7.45015621e-01 1.59026086e-01 7.27047503e-01 7.99564302e-01 6.76687121e-01 -1.52721000e+00 -3.83267909e-01 3.56845945e-01 3.64125013e-01 6.07339153e-03 4.42121714e-01 1.22043717e+00 -3.68353426e-01 4.94891882e-01 -5.94610691e-01 -9.61943686e-01 -1.32493448e+00 4.63496238e-01 1.83010504e-01 -2.56138474e-01 -7.48121679e-01 4.16628629e-01 8.44632089e-02 -2.23095685e-01 3.68317813e-01 -5.26744127e-01 3.98709141e-02 1.43707469e-01 5.16735494e-01 5.55454254e-01 1.45808488e-01 -9.75059748e-01 -5.12781203e-01 1.04591405e+00 2.82277375e-01 -2.82703847e-01 1.23938632e+00 -2.39957899e-01 3.29233855e-01 3.84740382e-01 1.21115327e+00 -2.28721306e-01 -1.64306235e+00 -3.18781406e-01 -1.49371758e-01 -6.91009760e-01 -1.61993086e-01 -6.79899096e-01 -1.11994326e+00 8.40487301e-01 6.68423235e-01 -2.14997992e-01 1.38751733e+00 1.67956688e-02 9.44332957e-01 3.72357100e-01 3.93301368e-01 -1.23128295e+00 6.60713553e-01 3.32320094e-01 7.92268157e-01 -1.36235237e+00 1.14426300e-01 1.94040574e-02 -8.48267376e-01 9.32065785e-01 7.36390352e-01 2.48543635e-01 3.84740055e-01 8.77945051e-02 3.79942246e-02 -1.98533550e-01 -4.53429371e-01 -4.95132655e-01 3.83345187e-01 8.28617275e-01 1.84785247e-01 -3.37738246e-01 6.53742552e-02 1.88954189e-01 -9.37236249e-02 1.32736132e-01 2.51263052e-01 1.25102961e+00 -1.79302648e-01 -9.58937109e-01 -5.37814617e-01 1.31976560e-01 -3.07137221e-01 2.73115903e-01 -3.02277535e-01 9.82888877e-01 1.62826385e-02 7.13153243e-01 6.86513558e-02 -4.46098238e-01 5.06157935e-01 1.63984850e-01 6.39941156e-01 -1.89907938e-01 -3.93328279e-01 -3.51370126e-02 2.32802685e-02 -1.29208946e+00 -1.09535265e+00 -9.91120934e-01 -1.33742356e+00 -9.20322612e-02 1.79359242e-01 -2.45874286e-01 5.69644988e-01 9.00862813e-01 3.07409704e-01 3.96129489e-01 2.64647514e-01 -1.20968378e+00 -6.67268224e-03 -8.70202541e-01 -2.00467318e-01 9.06708658e-01 4.60511893e-01 -1.06091952e+00 -1.92264244e-02 3.70338172e-01]
[7.950120449066162, 0.4408995509147644]
1a53c36e-1577-45df-aea8-8462c5335c98
thompson-sampling-for-combinatorial-semi-2
2005.06725
null
https://arxiv.org/abs/2005.06725v1
https://arxiv.org/pdf/2005.06725v1.pdf
Thompson Sampling for Combinatorial Semi-bandits with Sleeping Arms and Long-Term Fairness Constraints
We study the combinatorial sleeping multi-armed semi-bandit problem with long-term fairness constraints~(CSMAB-F). To address the problem, we adopt Thompson Sampling~(TS) to maximize the total rewards and use virtual queue techniques to handle the fairness constraints, and design an algorithm called \emph{TS with beta priors and Bernoulli likelihoods for CSMAB-F~(TSCSF-B)}. Further, we prove TSCSF-B can satisfy the fairness constraints, and the time-averaged regret is upper bounded by $\frac{N}{2\eta} + O\left(\frac{\sqrt{mNT\ln T}}{T}\right)$, where $N$ is the total number of arms, $m$ is the maximum number of arms that can be pulled simultaneously in each round~(the cardinality constraint) and $\eta$ is the parameter trading off fairness for rewards. By relaxing the fairness constraints (i.e., let $\eta \rightarrow \infty$), the bound boils down to the first problem-independent bound of TS algorithms for combinatorial sleeping multi-armed semi-bandit problems. Finally, we perform numerical experiments and use a high-rating movie recommendation application to show the effectiveness and efficiency of the proposed algorithm.
['QiPeng Wang', 'Yifan Xu', 'Bingshan Hu', 'Zhiming Huang', 'Jianping Pan']
2020-05-14
null
null
null
null
['movie-recommendation']
['miscellaneous']
[ 1.73415914e-02 -5.59541993e-02 -5.42206943e-01 -3.71507585e-01 -9.19930696e-01 -6.62254870e-01 -3.49745750e-01 -3.34699482e-01 -8.46328378e-01 1.23205948e+00 -4.04488295e-01 -7.98564672e-01 -8.54282737e-01 -7.53988862e-01 -6.33561075e-01 -8.95643711e-01 -1.91471264e-01 7.46189475e-01 -7.94368535e-02 -1.74855188e-01 2.82982737e-01 4.59139556e-01 -9.54052687e-01 -2.03605905e-01 1.00887799e+00 1.42996156e+00 1.98270530e-02 6.58872783e-01 -2.84150422e-01 8.40173542e-01 -5.80678046e-01 -6.87573195e-01 4.73875344e-01 -5.17708004e-01 -8.30668807e-01 -1.69606626e-01 -2.08615229e-01 -5.58958054e-01 -4.60991651e-01 1.02585506e+00 3.83000642e-01 5.13069987e-01 3.35363626e-01 -1.39188123e+00 2.45394111e-02 1.04127145e+00 -1.36614966e+00 6.63190722e-01 -2.19203681e-02 -2.08205640e-01 1.05253518e+00 -9.31336805e-02 6.07789084e-02 1.31212437e+00 1.61454082e-01 5.31581044e-01 -1.05075896e+00 -1.25195312e+00 5.04892170e-01 -3.04853797e-01 -1.21556461e+00 -5.16916633e-01 2.41908386e-01 9.02803466e-02 3.63291293e-01 1.02250159e+00 6.63822532e-01 2.55349368e-01 -1.18977882e-01 9.96671438e-01 1.21382308e+00 -6.02446139e-01 4.39976335e-01 -1.27673507e-01 4.61895138e-01 4.84290659e-01 4.50303167e-01 7.71317035e-02 -3.99351656e-01 -2.27015540e-01 8.04078877e-01 6.94111735e-02 -6.69007972e-02 2.20111609e-01 -7.00778365e-01 9.35331345e-01 4.67586815e-02 -7.94689283e-02 -3.38291377e-01 7.51144469e-01 1.12533830e-01 3.10302794e-01 7.51413882e-01 -2.27305532e-01 -1.94138184e-01 -2.63273060e-01 -1.16619277e+00 4.68075186e-01 4.66364264e-01 1.35335374e+00 3.72208923e-01 1.11790756e-02 -6.76708400e-01 8.28867376e-01 3.12649935e-01 1.19515979e+00 -3.21631491e-01 -1.49224365e+00 9.10078466e-01 -1.83902860e-01 8.29988420e-01 -3.31392646e-01 -1.98156953e-01 -6.01208985e-01 -8.26615691e-01 -4.51726355e-02 6.64973497e-01 -5.80176115e-01 -5.12006521e-01 1.93761647e+00 8.39294344e-02 -2.44986683e-01 -6.75953329e-01 8.13313246e-01 -1.23332091e-01 6.87961280e-01 -3.58516216e-01 -1.19166386e+00 9.04568315e-01 -7.13027000e-01 -7.92402327e-01 -1.50428504e-01 2.01537669e-01 -5.39256155e-01 6.60418570e-01 5.07518589e-01 -1.75509214e+00 2.02671409e-01 -5.68249345e-01 6.16373181e-01 3.90490860e-01 -3.95915002e-01 7.40599692e-01 1.36626160e+00 -7.65318215e-01 2.45142817e-01 -6.90028310e-01 1.47027150e-01 5.80043256e-01 6.72456741e-01 4.73357946e-01 -3.49235475e-01 -8.06653440e-01 2.47282490e-01 4.06597592e-02 3.26618791e-01 -8.35616529e-01 -3.15988451e-01 -2.47362107e-01 2.86645383e-01 8.15957069e-01 -4.97010022e-01 1.28947139e+00 -6.80560529e-01 -1.53616774e+00 3.15509260e-01 -3.22225660e-01 -3.13543290e-01 7.77168155e-01 9.01599526e-02 3.67742963e-02 4.10124250e-02 3.50076169e-01 1.67591348e-01 5.43472826e-01 -8.64622891e-01 -1.04381287e+00 -4.38107103e-01 3.69279295e-01 1.44580781e-01 -3.55746746e-01 5.50710559e-01 -4.17022854e-01 -5.50094306e-01 1.42992418e-02 -1.10225546e+00 -6.16708577e-01 -5.64763367e-01 -5.16575575e-01 -1.16662510e-01 -8.75724703e-02 -8.08394700e-02 1.74024916e+00 -2.15297151e+00 -1.37772068e-01 8.05191457e-01 1.26692932e-02 -3.47966067e-02 4.95083891e-02 7.84766525e-02 3.41432720e-01 3.53856295e-01 1.90463409e-01 -2.77070343e-01 1.95573121e-01 1.61327258e-01 -1.95563316e-01 4.52513993e-01 -9.50561762e-01 3.59675646e-01 -6.90239251e-01 -3.47373992e-01 -9.85581279e-02 -5.41559458e-01 -4.58237350e-01 7.08598644e-02 -3.03482145e-01 -3.62023921e-03 -9.15634692e-01 3.80181253e-01 9.91094887e-01 -2.05397740e-01 2.27018416e-01 5.67424953e-01 -2.55326927e-01 -3.68400365e-02 -1.53416252e+00 1.14095700e+00 -3.70584339e-01 -1.62571549e-01 5.40112436e-01 -9.36729848e-01 2.61682153e-01 1.39392108e-01 5.84123433e-01 -6.36526227e-01 5.07833779e-01 2.72493422e-01 -3.92665178e-01 -6.49027526e-02 4.41355467e-01 -6.52488410e-01 -2.87207365e-01 9.74241853e-01 -5.86164892e-01 2.05474481e-01 2.80425906e-01 4.81301308e-01 9.47395205e-01 -4.18810904e-01 4.06540610e-04 -3.57956588e-01 1.00665994e-01 -2.67457813e-01 7.85844445e-01 1.43069887e+00 -4.50646102e-01 -1.76313892e-03 7.35628366e-01 -2.30501845e-01 -5.53704917e-01 -9.52487767e-01 1.45011023e-01 1.78439951e+00 4.73768860e-01 2.24979699e-01 -7.41760314e-01 -5.74326217e-01 1.75249428e-01 1.04848123e+00 -5.94458878e-01 4.16464090e-01 -2.53653169e-01 -9.84516442e-01 3.76755774e-01 1.80174813e-01 3.45097333e-01 -6.63282454e-01 -5.86154878e-01 3.41343522e-01 -4.10699606e-01 -7.28858888e-01 -9.18515086e-01 3.24569702e-01 -6.97923899e-01 -6.23991311e-01 -7.54334569e-01 -1.33436080e-02 5.18944621e-01 7.12705433e-01 7.90549159e-01 -7.81030655e-02 2.23103501e-02 8.55222866e-02 -2.26688281e-01 -6.03809059e-01 2.72921950e-01 -6.68897927e-02 9.81753469e-02 -1.90654010e-01 -2.55858064e-01 -3.71794701e-01 -9.53531086e-01 5.56292534e-01 -6.79595411e-01 -1.70693740e-01 2.31130227e-01 5.78720272e-01 5.98799169e-01 5.43984286e-02 7.30075479e-01 -8.99699628e-01 5.61548471e-01 -2.92645395e-01 -8.41206849e-01 5.46603501e-01 -3.54771227e-01 -1.33799285e-01 3.68625581e-01 -1.90475389e-01 -1.11212194e+00 -4.74978477e-01 1.65455505e-01 -2.88167447e-01 4.91902679e-01 3.74753326e-01 1.23144865e-01 9.55180749e-02 2.09623098e-01 4.73594945e-03 -5.25596589e-02 -1.70193017e-01 5.01081705e-01 7.33637214e-01 9.73079950e-02 -8.72564614e-01 3.10011744e-01 3.57101321e-01 9.80652869e-02 -3.10421705e-01 -1.32339382e+00 -9.85372737e-02 4.14440781e-01 -2.25668222e-01 3.46513480e-01 -6.18924618e-01 -1.64736009e+00 8.19972157e-03 -5.25103092e-01 -3.77634495e-01 -2.72705883e-01 5.04272521e-01 -8.22640836e-01 3.79469573e-01 -5.94456077e-01 -1.90337539e+00 -5.62189758e-01 -7.98473537e-01 2.13943124e-01 2.77526945e-01 2.90867925e-01 -2.38814503e-01 -2.84756780e-01 7.33702660e-01 4.04370308e-01 -1.21500552e-01 7.33991146e-01 -2.85773009e-01 -7.07468331e-01 -1.70288086e-01 -3.86870503e-01 1.66612998e-01 -2.37049967e-01 -3.58487189e-01 -2.21276432e-01 -7.11574495e-01 -4.04669121e-02 -3.44037294e-01 5.37074745e-01 1.02843976e+00 1.40791023e+00 -5.99861562e-01 -3.82719219e-01 2.38720134e-01 1.26056588e+00 5.40951073e-01 3.75534594e-01 1.94658965e-01 -1.99901536e-02 6.99032238e-03 1.08885992e+00 1.29683423e+00 1.77291468e-01 4.95344728e-01 6.07415318e-01 2.67117500e-01 7.62266099e-01 5.27619064e-01 4.64718074e-01 4.40173686e-01 -2.53154337e-01 -7.84852862e-01 -4.68179733e-01 4.70439762e-01 -1.93640995e+00 -9.85220253e-01 -8.71031284e-02 2.72209454e+00 7.49115348e-01 6.40602350e-01 6.31445646e-01 7.13018626e-02 9.44793940e-01 -1.00337058e-01 -4.41431493e-01 -7.92447507e-01 3.06678981e-01 5.15349507e-01 9.90808606e-01 5.22821426e-01 -6.19840145e-01 6.78696334e-01 5.75166416e+00 1.56587517e+00 -5.16224504e-01 2.67486304e-01 1.10469210e+00 -1.18984485e+00 -4.11428958e-01 -1.02575459e-01 -7.04378366e-01 8.63846660e-01 1.03872204e+00 -2.40075931e-01 1.08551455e+00 3.86142969e-01 4.56370175e-01 -6.04459524e-01 -7.31113017e-01 8.08919609e-01 -3.40624809e-01 -1.06116796e+00 -2.55714685e-01 2.82165796e-01 5.51479220e-01 -1.29832983e-01 2.49373212e-01 3.18859160e-01 9.25131142e-01 -7.59423196e-01 8.78028274e-01 4.07203078e-01 1.08759022e+00 -1.44763243e+00 5.58169484e-01 6.99983001e-01 -9.83854294e-01 -5.04151762e-01 -4.19153512e-01 -1.05912596e-01 3.90436292e-01 7.93129265e-01 -6.73690289e-02 7.19036937e-01 7.69373596e-01 -3.33318770e-01 4.91948664e-01 1.09218216e+00 3.25693160e-01 4.31648135e-01 -7.45571196e-01 -3.67652863e-01 4.51191008e-01 -4.36246723e-01 3.32486421e-01 7.08205879e-01 4.61883605e-01 9.01657939e-01 3.03366899e-01 4.17003304e-01 -1.23610266e-01 -1.74703747e-02 1.86083496e-01 1.83320239e-01 8.04927230e-01 9.65867579e-01 -8.24453652e-01 -1.64563626e-01 -3.96084711e-02 5.45718312e-01 1.60713181e-01 3.46155912e-01 -1.32317603e+00 -4.52977359e-01 4.20889676e-01 2.70484090e-02 3.75435948e-01 1.57875806e-01 -4.70600992e-01 -6.46988451e-01 -2.38736138e-01 -5.54003119e-01 5.97101331e-01 -5.15750766e-01 -1.15911579e+00 3.40232193e-01 6.69855177e-02 -7.33432174e-01 3.14447075e-01 -6.77714311e-03 -4.68517154e-01 9.42273200e-01 -1.30381835e+00 -6.79339230e-01 2.79218495e-01 8.21512640e-01 2.38898695e-01 9.10519585e-02 3.75808865e-01 3.85186702e-01 -9.79847431e-01 9.97076750e-01 6.30436540e-01 -4.46382016e-01 3.22665542e-01 -9.50192094e-01 -3.58260810e-01 6.69966400e-01 -5.21720469e-01 6.10926986e-01 6.44353867e-01 -3.83107930e-01 -1.10064566e+00 -7.35168219e-01 4.22517806e-01 3.20294425e-02 4.25941885e-01 -8.63091946e-02 4.39289697e-02 6.10959649e-01 -2.41894200e-02 1.36157960e-01 9.45067823e-01 2.21541762e-01 1.13064937e-01 -5.52803576e-01 -1.42290902e+00 3.57644528e-01 1.15714920e+00 -1.42742125e-02 2.25195229e-01 3.14313173e-01 6.36928797e-01 -3.49936098e-01 -7.31768668e-01 1.53101534e-01 7.77606845e-01 -9.86480057e-01 6.06693625e-01 -5.34873664e-01 -1.35922596e-01 1.63094416e-01 -2.69863069e-01 -8.60305369e-01 -4.12810057e-01 -1.15248811e+00 1.56779557e-01 1.02588725e+00 5.63301921e-01 -4.55766827e-01 9.20076370e-01 9.98824239e-01 7.32631534e-02 -7.06438601e-01 -1.52069950e+00 -7.38695502e-01 1.94538862e-01 -4.39212173e-01 6.48288667e-01 5.48242569e-01 2.32589558e-01 -8.11713412e-02 -8.06728363e-01 -1.30242422e-01 7.93579400e-01 3.22170138e-01 3.87876540e-01 -7.20082939e-01 -5.95552623e-01 -3.70817035e-01 8.26856494e-01 -1.19557166e+00 -1.58587694e-01 -4.28176522e-01 2.68076044e-02 -1.18859625e+00 7.55120277e-01 -1.19769561e+00 -8.63395393e-01 4.33161974e-01 -1.26533732e-01 -9.61445943e-02 4.69890714e-01 2.75066737e-02 -1.29250944e+00 3.33885193e-01 1.23965538e+00 5.44602759e-02 -1.21586569e-01 7.38812864e-01 -7.92304754e-01 3.70742232e-01 4.65486228e-01 -7.20412970e-01 -3.91388118e-01 -4.31056261e-01 5.74984431e-01 1.18664467e+00 -3.81204218e-01 -2.97818094e-01 -2.31086165e-02 -8.88386607e-01 -2.74995081e-02 -8.13967109e-01 1.67142540e-01 -8.54263604e-01 1.88844189e-01 5.23471177e-01 -4.87755418e-01 -1.95445523e-01 -5.94045147e-02 9.73140240e-01 5.86683750e-01 -4.47623014e-01 7.31312633e-01 -2.31356412e-01 5.24913609e-01 4.15399998e-01 -6.66224599e-01 7.62590915e-02 1.04520690e+00 -2.59578358e-02 -3.93253326e-01 -6.80107653e-01 -7.14077592e-01 8.16013396e-01 -1.17381595e-01 -3.56362969e-01 2.41009936e-01 -1.02163041e+00 -3.97894740e-01 -3.19722980e-01 -3.88127983e-01 -5.21164164e-02 6.00052118e-01 9.46956336e-01 -3.37211758e-01 3.68058771e-01 9.61239040e-02 -2.71275520e-01 -9.70872283e-01 5.54139614e-01 1.54258385e-01 -6.33873761e-01 2.15974748e-01 1.21907258e+00 -1.74984187e-01 1.83709264e-01 4.65028584e-01 -4.17469367e-02 3.23397130e-01 -7.38417134e-02 3.58487129e-01 9.75654840e-01 -2.83545375e-01 1.09825069e-02 -4.05155748e-01 -2.07957312e-01 -2.77393043e-01 -6.42204583e-01 1.31591308e+00 -5.68104684e-01 -4.70885903e-01 -2.71068886e-02 4.78068292e-01 1.63349286e-01 -1.12519097e+00 -2.62039870e-01 -4.10162598e-01 -8.55230927e-01 1.10030845e-01 -9.43878829e-01 -1.20678508e+00 4.58934993e-01 5.36621869e-01 5.83852768e-01 1.07368815e+00 -4.31626916e-01 8.37783813e-01 1.59477890e-01 9.87364888e-01 -1.18073988e+00 -2.95268185e-02 4.24675643e-01 2.99043685e-01 -6.21727109e-01 2.18366146e-01 -2.87467718e-01 -5.67132533e-01 7.08518863e-01 3.98293346e-01 7.94479400e-02 5.42352498e-01 5.79030067e-02 -5.25211513e-01 7.12138861e-02 -4.62312400e-01 -9.08124000e-02 -2.95523107e-01 -2.76844978e-01 1.85697094e-01 5.31161427e-01 -8.18803489e-01 9.89544749e-01 -1.61152519e-02 2.63961941e-01 5.30466199e-01 1.07053196e+00 -7.89863288e-01 -1.19257295e+00 -4.69955027e-01 9.77233231e-01 -9.87596452e-01 7.64021929e-03 2.32208654e-01 2.82392800e-01 -5.48315905e-02 1.50243795e+00 -2.77111623e-02 1.31950140e-01 1.01129659e-01 -3.14611048e-01 6.42199039e-01 -3.66709977e-01 -5.95042646e-01 7.29706228e-01 2.69027829e-01 -3.47132683e-01 -3.85229409e-01 -4.84269470e-01 -1.02717483e+00 -1.02751696e+00 -9.04417038e-01 6.94970548e-01 2.90756851e-01 1.03953469e+00 -9.64880586e-02 4.99100477e-01 1.30411756e+00 -3.79134268e-01 -8.16774905e-01 -7.45831490e-01 -1.14499080e+00 -2.03258410e-01 -1.63549464e-02 -6.42526388e-01 -2.95551687e-01 -8.74593973e-01]
[4.556407451629639, 3.3446969985961914]
a957fe64-e92f-4e6f-9cee-ee234fbb72ab
avatarclip-zero-shot-text-driven-generation
2205.08535
null
https://arxiv.org/abs/2205.08535v1
https://arxiv.org/pdf/2205.08535v1.pdf
AvatarCLIP: Zero-Shot Text-Driven Generation and Animation of 3D Avatars
3D avatar creation plays a crucial role in the digital age. However, the whole production process is prohibitively time-consuming and labor-intensive. To democratize this technology to a larger audience, we propose AvatarCLIP, a zero-shot text-driven framework for 3D avatar generation and animation. Unlike professional software that requires expert knowledge, AvatarCLIP empowers layman users to customize a 3D avatar with the desired shape and texture, and drive the avatar with the described motions using solely natural languages. Our key insight is to take advantage of the powerful vision-language model CLIP for supervising neural human generation, in terms of 3D geometry, texture and animation. Specifically, driven by natural language descriptions, we initialize 3D human geometry generation with a shape VAE network. Based on the generated 3D human shapes, a volume rendering model is utilized to further facilitate geometry sculpting and texture generation. Moreover, by leveraging the priors learned in the motion VAE, a CLIP-guided reference-based motion synthesis method is proposed for the animation of the generated 3D avatar. Extensive qualitative and quantitative experiments validate the effectiveness and generalizability of AvatarCLIP on a wide range of avatars. Remarkably, AvatarCLIP can generate unseen 3D avatars with novel animations, achieving superior zero-shot capability.
['Ziwei Liu', 'Lei Yang', 'Zhongang Cai', 'Liang Pan', 'Mingyuan Zhang', 'Fangzhou Hong']
2022-05-17
null
null
null
null
['texture-synthesis']
['computer-vision']
[-1.06867269e-01 1.95246786e-01 2.59190142e-01 1.05807610e-01 -3.41446370e-01 -8.50192130e-01 7.97489524e-01 -6.29765272e-01 1.37845069e-01 1.93032280e-01 2.17162326e-01 -1.59054205e-01 5.36878288e-01 -8.93389344e-01 -6.50401294e-01 -5.42295635e-01 3.43984008e-01 6.08155251e-01 -6.34400696e-02 -6.13998592e-01 7.21756220e-02 6.66316926e-01 -1.33952379e+00 -9.85536650e-02 8.04958165e-01 8.41948211e-01 1.91293523e-01 9.40568686e-01 -1.20493367e-01 6.00235760e-01 -5.88125765e-01 -4.07417893e-01 4.18354779e-01 -6.85083628e-01 -4.44875807e-01 4.46095079e-01 3.15623462e-01 -7.07854211e-01 -4.47344512e-01 8.00024390e-01 5.71597099e-01 3.64344716e-01 7.17013061e-01 -1.32626712e+00 -8.08293939e-01 3.98882687e-01 -7.02324033e-01 -5.12723625e-01 8.11012268e-01 8.91891181e-01 7.71584928e-01 -9.70418274e-01 8.83339584e-01 1.50180495e+00 3.37049276e-01 1.03316474e+00 -1.28490162e+00 -7.28060842e-01 2.82376148e-02 -3.08653921e-01 -1.41810822e+00 -2.84410954e-01 1.16090226e+00 -4.17378902e-01 4.82659042e-01 3.30413282e-01 1.44396472e+00 1.53487539e+00 6.53326437e-02 9.21744764e-01 4.34639335e-01 -2.11972818e-01 2.57979482e-01 -1.67936653e-01 -7.44183600e-01 8.63726377e-01 -3.99349719e-01 2.43638992e-01 -4.36827809e-01 -2.32487097e-02 1.70239317e+00 -2.84259021e-01 -8.36981088e-02 -5.46954989e-01 -1.42041445e+00 8.72158706e-01 1.26313761e-01 -2.10499540e-01 -5.47375202e-01 5.99368691e-01 3.72381747e-01 -9.59881395e-02 4.55179214e-01 4.64798123e-01 1.80984084e-02 -3.76355618e-01 -7.90058792e-01 8.39769006e-01 6.04141057e-01 1.33293104e+00 3.71455818e-01 6.94779336e-01 -2.34210014e-01 5.26277542e-01 4.50373292e-01 1.04558563e+00 4.64594401e-02 -1.33719885e+00 3.27692889e-02 5.86409211e-01 -8.02295282e-03 -1.18110502e+00 -1.73993453e-01 2.19414070e-01 -1.00773501e+00 6.30642176e-01 2.85854518e-01 -2.86514491e-01 -8.89121830e-01 1.56197977e+00 7.30422735e-01 1.22220092e-01 -1.58168092e-01 1.38595557e+00 1.00092578e+00 1.04405177e+00 1.49030447e-01 2.06961244e-01 1.24331713e+00 -9.37398374e-01 -6.05398536e-01 1.41180217e-01 2.60208517e-01 -5.74038506e-01 1.42357242e+00 2.17315108e-01 -1.27423847e+00 -5.69674730e-01 -7.86592722e-01 -1.34506628e-01 3.01645547e-01 -2.86942333e-01 6.93217814e-01 4.31893587e-01 -9.04573381e-01 3.08389038e-01 -7.88291931e-01 -6.90478161e-02 3.83665323e-01 3.82587104e-03 -2.66337872e-01 4.66509998e-01 -1.08728898e+00 5.99240363e-01 8.71032104e-02 -2.36706156e-02 -1.13815880e+00 -9.64311779e-01 -1.14231491e+00 -2.89099813e-01 3.42846364e-01 -1.07987392e+00 1.45371222e+00 -1.00319350e+00 -2.25418806e+00 7.50845730e-01 3.55468035e-01 -1.07486114e-01 9.32045937e-01 -1.08112264e-02 -1.91066898e-02 3.66298795e-01 -7.05943555e-02 1.04914832e+00 1.00544608e+00 -1.39179146e+00 -2.86112368e-01 1.75869651e-02 1.40557513e-01 4.08249766e-01 9.65237394e-02 -1.25776678e-01 -6.90442085e-01 -9.44878817e-01 -1.94815844e-01 -8.91894460e-01 -3.68483871e-01 7.28846431e-01 -4.66136664e-01 -1.32265911e-01 1.07387912e+00 -4.74335134e-01 7.95098901e-01 -2.07659507e+00 4.37799007e-01 8.92938301e-02 5.35449862e-01 2.11054295e-01 -1.95032492e-01 2.86388606e-01 1.57996610e-01 1.02935337e-01 2.69902386e-02 -2.53409117e-01 1.14067309e-01 2.78186090e-02 -4.38799679e-01 1.83870181e-01 1.78350240e-01 1.25223923e+00 -1.03863370e+00 -6.74890161e-01 5.87562680e-01 8.79093409e-01 -8.17026973e-01 4.95487839e-01 -7.09517777e-01 8.55233133e-01 -7.69535303e-01 6.04265690e-01 4.73039180e-01 -1.85407400e-01 -1.28489017e-01 -1.27981678e-01 -1.76985532e-01 -2.97406912e-01 -7.65172184e-01 2.12136436e+00 -5.47175884e-01 5.23267806e-01 4.54123914e-02 -3.75219256e-01 1.12844670e+00 4.72863883e-01 4.56459016e-01 -5.92677593e-01 3.75172943e-01 -1.11862302e-01 -3.59724134e-01 -6.06433630e-01 5.82217336e-01 -2.59189725e-01 -3.88510913e-01 6.27286434e-01 -2.48802826e-01 -9.19780552e-01 -2.48484582e-01 4.48128909e-01 5.21414280e-01 8.46609950e-01 7.95483515e-02 8.36388543e-02 2.50736266e-01 1.22477859e-01 1.82937488e-01 2.36909837e-01 7.70954788e-02 9.52538192e-01 3.87729466e-01 -6.36546910e-01 -1.67904222e+00 -1.24815786e+00 5.25326192e-01 8.99605513e-01 4.15446460e-01 -3.26592565e-01 -8.52427185e-01 -3.58239591e-01 -2.57079422e-01 8.77738416e-01 -6.51097476e-01 -2.26100400e-01 -7.74124920e-01 -7.42005408e-02 5.88513374e-01 3.96654099e-01 3.95992965e-01 -1.24751663e+00 -1.01354957e+00 1.85138747e-01 -9.22129378e-02 -1.00764155e+00 -1.05731106e+00 -8.08772683e-01 -6.03387117e-01 -5.04698992e-01 -1.22923589e+00 -7.39254057e-01 6.50652409e-01 1.91696614e-01 9.31043744e-01 6.95985109e-02 -2.39024028e-01 4.47858512e-01 -4.39324856e-01 -2.55324185e-01 -9.86843407e-01 -1.83569312e-01 4.55361158e-02 -1.02052346e-01 -4.12510216e-01 -8.81022394e-01 -7.55618155e-01 1.31978273e-01 -8.31044972e-01 1.13671994e+00 2.23286659e-01 5.54372132e-01 4.45444196e-01 -4.37061876e-01 3.72608483e-01 -4.95620459e-01 5.29373288e-01 -1.38780460e-01 -5.41298151e-01 -1.35968313e-01 -1.58863679e-01 -3.58832143e-02 9.03381765e-01 -1.01423645e+00 -1.11622465e+00 2.66886115e-01 -2.68537223e-01 -8.91521871e-01 -7.45243952e-02 2.59327516e-02 -3.25494111e-01 1.14973485e-02 6.89514816e-01 3.59139949e-01 3.87604088e-01 -1.27574459e-01 9.98338342e-01 3.77967268e-01 7.46848226e-01 -7.41319180e-01 1.27180707e+00 4.31132436e-01 -1.50804922e-01 -7.76312888e-01 -3.51390958e-01 3.72515321e-01 -5.20412922e-01 -7.73823917e-01 1.08290195e+00 -8.05733502e-01 -1.25338280e+00 7.02377260e-01 -1.45530033e+00 -6.82992160e-01 -4.62401807e-01 1.12698846e-01 -1.01140165e+00 3.38762999e-01 -5.51622391e-01 -7.86703527e-01 -6.52009785e-01 -1.20052361e+00 1.31034708e+00 3.70633125e-01 -5.83648562e-01 -9.16388988e-01 2.42589898e-02 3.11565608e-01 2.11113080e-01 8.60157788e-01 8.41731846e-01 3.32685001e-02 -6.16147637e-01 -2.11364552e-01 -5.01559265e-02 -8.39888155e-02 3.07910424e-02 3.46589833e-01 -5.79416454e-01 1.48872182e-01 -6.28142178e-01 -3.52255404e-01 -8.10530856e-02 1.88125685e-01 9.38645005e-01 -4.88599718e-01 3.77248079e-02 7.60141492e-01 8.88990104e-01 2.94473380e-01 4.78360295e-01 -6.23190105e-02 1.26638401e+00 4.69092339e-01 4.06457394e-01 9.06994641e-01 5.32555401e-01 7.64523983e-01 5.13647258e-01 -1.90543011e-01 -3.09476525e-01 -8.34672868e-01 2.29933947e-01 7.17589140e-01 -5.35006464e-01 -1.94599599e-01 -6.60323143e-01 2.95820594e-01 -1.68360686e+00 -8.64795864e-01 -2.07935199e-02 1.77339244e+00 9.53454494e-01 -1.13438934e-01 4.04778183e-01 -3.51413548e-01 5.84373534e-01 3.28237951e-01 -7.09229827e-01 -3.42074960e-01 5.39535806e-02 9.65822563e-02 -6.08316734e-02 3.44044507e-01 -5.44686735e-01 1.39300442e+00 5.82311249e+00 7.76777506e-01 -1.17006886e+00 -3.05826724e-01 4.92990822e-01 -1.90276504e-01 -7.91408718e-01 1.29880374e-02 -2.06187293e-01 2.67007500e-01 3.50587547e-01 -5.65303862e-01 6.18232608e-01 8.63396823e-01 8.00597608e-01 3.40649247e-01 -9.89631295e-01 1.09654868e+00 4.94483598e-02 -1.74276745e+00 5.85307956e-01 -7.13888183e-02 6.28574491e-01 -7.56049752e-01 1.82705581e-01 1.67426899e-01 4.82279569e-01 -1.06424320e+00 1.32194483e+00 6.01534486e-01 1.34788013e+00 -9.49068964e-01 -9.15467972e-04 3.84145141e-01 -1.27514684e+00 5.21211326e-01 3.54477949e-02 -3.09178536e-03 7.25604057e-01 -2.82356083e-01 -6.40434802e-01 2.09756106e-01 3.05934995e-01 4.86736208e-01 -1.76096018e-02 4.65139180e-01 -3.73593211e-01 2.26568282e-01 -7.78263509e-02 -2.14516282e-01 2.58924425e-01 -3.57005537e-01 8.33542347e-01 9.41793442e-01 3.38466197e-01 4.61625904e-01 1.49051443e-01 1.45489848e+00 -2.25605047e-03 9.33955908e-02 -7.00108230e-01 -3.08540761e-01 4.44921076e-01 1.28054440e+00 -5.62029302e-01 -3.23232323e-01 4.09803726e-02 1.22059584e+00 -1.15960594e-02 3.90519410e-01 -1.22011018e+00 -3.09476435e-01 6.42033517e-01 3.06688547e-01 3.62950228e-02 -5.43074429e-01 -3.10986817e-01 -1.05253351e+00 -5.03250599e-01 -9.62415636e-01 -3.41253042e-01 -1.30981302e+00 -1.06535041e+00 6.20360076e-01 1.57990679e-01 -1.41276944e+00 -5.93026340e-01 -1.71008930e-01 -8.45887542e-01 6.79211080e-01 -5.29887438e-01 -1.60961890e+00 -5.33573270e-01 5.34378648e-01 8.69916201e-01 -1.90402001e-01 6.60460889e-01 -1.66111544e-01 -2.54992634e-01 5.67879975e-01 -7.30712712e-01 2.97684699e-01 3.70057583e-01 -8.81819010e-01 1.10918057e+00 4.45580989e-01 5.46302535e-02 2.02610761e-01 8.77794683e-01 -7.06498444e-01 -1.87868094e+00 -1.03282475e+00 1.47507235e-01 -4.38548625e-01 6.81558132e-01 -5.15193880e-01 -7.12761939e-01 4.42719489e-01 1.72348082e-01 -1.06670581e-01 3.82922858e-01 -6.59617305e-01 -2.57534087e-01 3.45999360e-01 -9.40118492e-01 1.33564758e+00 1.34149814e+00 -3.16497535e-01 -3.45179617e-01 7.08394349e-02 1.12924612e+00 -8.30801904e-01 -7.70327270e-01 2.60989983e-02 8.25860262e-01 -7.04304397e-01 1.02943528e+00 -5.10819793e-01 8.18184674e-01 -2.86323994e-01 1.28986880e-01 -1.24945331e+00 -1.96788594e-01 -1.30460811e+00 -1.16166569e-01 1.02528894e+00 1.68112725e-01 -3.27575430e-02 8.82635951e-01 8.24561894e-01 -1.63331881e-01 -5.64526021e-01 -5.31909585e-01 -5.08458257e-01 1.53917953e-01 -6.29858792e-01 7.85277843e-01 8.01903486e-01 -1.94255523e-02 3.51504415e-01 -8.54756594e-01 -1.38864458e-01 6.69209719e-01 1.89767212e-01 1.41786194e+00 -1.03918147e+00 -3.03478837e-01 -4.79256123e-01 -3.27332497e-01 -1.53295851e+00 1.32116750e-01 -7.75952339e-01 1.22810178e-01 -1.42750347e+00 -9.09567177e-02 -4.33256805e-01 7.73366511e-01 2.42081165e-01 -9.37398523e-03 4.08070803e-01 6.21513844e-01 1.12180836e-01 -3.34904760e-01 1.00024128e+00 2.18377781e+00 2.87384563e-03 -6.32343888e-01 -2.08659291e-01 -5.89667499e-01 8.85065556e-01 4.62745219e-01 -3.58478911e-02 -6.04250073e-01 -4.64181781e-01 2.67536134e-01 6.81670725e-01 6.52022779e-01 -6.50992274e-01 4.23863046e-02 -6.19330585e-01 2.72793472e-01 -4.61772025e-01 5.57674289e-01 -5.23460507e-01 4.34445828e-01 2.87806243e-01 -3.19066465e-01 1.75703075e-02 7.47292712e-02 3.12932968e-01 3.29994380e-01 3.39542150e-01 7.18542576e-01 -2.05256715e-01 -5.69225669e-01 8.63686979e-01 -4.88979310e-01 1.72364846e-01 1.07755959e+00 -4.86505121e-01 1.56730086e-01 -8.76648784e-01 -5.36965311e-01 1.72371924e-01 7.25018620e-01 6.15732372e-01 1.06126201e+00 -1.72568333e+00 -8.09447765e-01 3.38638932e-01 1.17816269e-01 5.84489048e-01 5.41327119e-01 2.55060256e-01 -1.05281210e+00 -3.38059932e-01 -3.20091754e-01 -6.33819699e-01 -1.03317380e+00 5.15667439e-01 1.98573351e-01 3.10203254e-01 -1.29518259e+00 6.45657003e-01 5.94355106e-01 -3.83581907e-01 -3.08793280e-02 9.17313024e-02 1.27315640e-01 -4.02159244e-01 6.00446224e-01 1.64385960e-01 -8.57342780e-01 -7.33102024e-01 7.62332752e-02 7.11298645e-01 2.90344238e-01 -5.38368821e-01 1.22297418e+00 1.07849305e-02 3.51727039e-01 3.17024559e-01 7.75909841e-01 4.45693582e-02 -1.91482019e+00 2.24317163e-01 -7.67966032e-01 -4.60111052e-01 -2.67027408e-01 -3.67795706e-01 -1.23050570e+00 8.53938103e-01 -1.16952714e-02 -1.80847675e-01 8.47205758e-01 8.27307347e-03 1.27821326e+00 -2.43984461e-02 5.47569811e-01 -7.50517309e-01 5.26711881e-01 3.84510815e-01 1.28492117e+00 -8.64201963e-01 -2.55101472e-01 -4.02057499e-01 -1.08280504e+00 1.22532892e+00 7.55043089e-01 -3.54617447e-01 3.01921308e-01 3.90807956e-01 2.88907588e-01 -1.87777430e-01 -6.46360576e-01 4.41499770e-01 2.64606893e-01 9.19651151e-01 1.06484525e-01 8.56941864e-02 1.37837574e-01 6.00242198e-01 -6.14507616e-01 -1.27718419e-01 7.09885955e-01 4.56971616e-01 -2.13676885e-01 -8.21096718e-01 -3.26644719e-01 -3.42096329e-01 2.20483020e-01 7.46537969e-02 -3.72999340e-01 8.93281937e-01 -8.97390023e-02 6.08395219e-01 1.60705775e-01 -6.13960862e-01 3.39194864e-01 -3.69542688e-01 5.30737877e-01 -4.74721283e-01 -5.21690071e-01 2.25457177e-01 -1.63849279e-01 -4.66582239e-01 -4.08318527e-02 -3.33956331e-01 -1.65063274e+00 -7.87042260e-01 1.15920648e-01 -2.16593921e-01 4.80584085e-01 6.11304879e-01 3.29187304e-01 3.69688720e-01 7.93177903e-01 -1.40241921e+00 -1.41069323e-01 -5.68939924e-01 -3.75863612e-01 5.78701496e-01 7.23922327e-02 -5.09881616e-01 -9.30860713e-02 4.59514886e-01]
[11.995943069458008, -0.6641638278961182]
b15f86a0-e3f6-43f8-897e-f48494b658bf
agents-that-listen-high-throughput
2107.02195
null
https://arxiv.org/abs/2107.02195v1
https://arxiv.org/pdf/2107.02195v1.pdf
Agents that Listen: High-Throughput Reinforcement Learning with Multiple Sensory Systems
Humans and other intelligent animals evolved highly sophisticated perception systems that combine multiple sensory modalities. On the other hand, state-of-the-art artificial agents rely mostly on visual inputs or structured low-dimensional observations provided by instrumented environments. Learning to act based on combined visual and auditory inputs is still a new topic of research that has not been explored beyond simple scenarios. To facilitate progress in this area we introduce a new version of VizDoom simulator to create a highly efficient learning environment that provides raw audio observations. We study the performance of different model architectures in a series of tasks that require the agent to recognize sounds and execute instructions given in natural language. Finally, we train our agent to play the full game of Doom and find that it can consistently defeat a traditional vision-based adversary. We are currently in the process of merging the augmented simulator with the main ViZDoom code repository. Video demonstrations and experiment code can be found at https://sites.google.com/view/sound-rl.
['Aleksei Petrenko', 'Anssi Kanervisto', 'Shashank Hegde']
2021-07-05
null
null
null
null
['game-of-doom']
['playing-games']
[-3.89978811e-02 -9.83988792e-02 4.13071781e-01 -1.21497102e-01 -4.64649945e-01 -7.84749925e-01 7.02365279e-01 -1.75476484e-02 -8.74900043e-01 5.46715021e-01 -1.23116754e-01 -2.88886815e-01 2.54819810e-01 -4.97850806e-01 -4.97830451e-01 -5.58749080e-01 -1.60829529e-01 6.23501122e-01 5.25414824e-01 -3.82857770e-01 3.26685965e-01 3.42651308e-01 -1.94245565e+00 1.27829358e-01 2.07171708e-01 5.25629401e-01 4.29628313e-01 1.39443707e+00 5.15605092e-01 7.36539364e-01 -8.30205083e-01 4.63088751e-02 3.26697499e-01 -5.21406174e-01 -4.88440096e-01 -3.42297763e-01 3.06460917e-01 -3.89549106e-01 -2.02461973e-01 9.45833802e-01 9.21927214e-01 2.90794164e-01 4.45184976e-01 -1.45383263e+00 -4.66718376e-01 3.46185654e-01 -5.40824570e-02 3.10907334e-01 6.93611264e-01 8.28632712e-01 8.42876792e-01 -5.72202325e-01 3.41226518e-01 1.28682804e+00 4.15802360e-01 7.69366264e-01 -1.07646036e+00 -6.88726425e-01 -5.13281003e-02 3.16165984e-01 -9.71049130e-01 -5.71513057e-01 6.06547058e-01 -3.60482097e-01 1.32946408e+00 1.13710597e-01 7.71694243e-01 1.54252648e+00 2.40802318e-01 6.02922320e-01 1.15638959e+00 -3.51519197e-01 6.36225224e-01 4.73542958e-02 -3.30723882e-01 9.81212616e-01 7.68146962e-02 9.38564897e-01 -7.40132511e-01 -1.62642792e-01 6.61197841e-01 -3.40820253e-01 -2.52976716e-01 -5.17219722e-01 -1.17157352e+00 9.41609204e-01 4.73060280e-01 1.84335917e-01 -4.94903833e-01 4.26470667e-01 1.80980265e-01 5.36232352e-01 -3.18082482e-01 9.38294649e-01 -3.66312265e-01 -5.31604648e-01 -4.23976302e-01 3.98964614e-01 9.20335531e-01 2.62621850e-01 3.43793392e-01 2.84062713e-01 4.28549141e-01 3.95876348e-01 5.10178447e-01 5.91771543e-01 4.16097373e-01 -1.53094244e+00 -1.69716403e-01 7.89215416e-02 2.07793862e-01 -6.50668502e-01 -6.08784199e-01 -3.82849187e-01 -1.75342798e-01 1.10727584e+00 5.10855138e-01 -4.67388570e-01 -6.13880634e-01 1.75509775e+00 3.42901796e-01 3.49265903e-01 3.63640100e-01 1.11800277e+00 8.23074698e-01 6.27305210e-01 -1.57526180e-01 1.66830704e-01 1.30032861e+00 -9.45891917e-01 -4.05563772e-01 -3.98693353e-01 4.00807500e-01 -6.59438252e-01 1.19538832e+00 8.64926100e-01 -1.01039362e+00 -5.60846686e-01 -1.18932343e+00 2.01339513e-01 -5.13605297e-01 -4.26411211e-01 7.64668941e-01 6.73806012e-01 -1.10607111e+00 1.64816409e-01 -1.19076133e+00 -5.41740358e-01 2.07054436e-01 3.76855463e-01 -3.57006282e-01 2.17414364e-01 -1.09146965e+00 9.44985330e-01 2.13409796e-01 -3.03561002e-01 -1.69137371e+00 -2.55272329e-01 -8.97183061e-01 -2.06963062e-01 5.65990746e-01 -8.78509223e-01 1.86368060e+00 -6.82520330e-01 -1.95102417e+00 6.41327441e-01 3.49765837e-01 -5.67583740e-01 4.03164476e-01 -4.95175481e-01 -2.40622219e-04 4.34328675e-01 -1.41452760e-01 7.69726813e-01 6.57751024e-01 -1.47791278e+00 -6.74346089e-01 -3.02348912e-01 3.70765686e-01 2.73375094e-01 2.35665571e-02 8.92318711e-02 1.10796086e-01 -3.83298039e-01 -5.04785359e-01 -1.10657322e+00 -3.05549473e-01 1.03762291e-01 -1.09517619e-01 8.01454578e-03 8.52201879e-01 -1.09257795e-01 3.15040618e-01 -2.06917262e+00 1.68522879e-01 -1.28484860e-01 -1.98109522e-02 4.18636262e-01 -3.53419423e-01 7.76174664e-01 3.98342252e-01 -2.95595318e-01 -3.75659764e-02 -5.46140313e-01 3.18267375e-01 1.56971216e-01 -2.91010052e-01 4.23259526e-01 -3.12135398e-01 6.66051984e-01 -1.17868841e+00 -1.39652357e-01 3.87606353e-01 5.63963234e-01 -8.77086341e-01 4.72892076e-01 -4.06583399e-01 7.29231417e-01 -3.34570765e-01 3.16692114e-01 1.42649367e-01 1.28628463e-01 -6.98075518e-02 5.03685057e-01 -1.92136720e-01 2.82615006e-01 -1.05889344e+00 1.90090954e+00 -5.75218618e-01 7.58403242e-01 4.42061335e-01 -8.21383297e-01 5.54035842e-01 3.66098672e-01 6.68573901e-02 -4.96460527e-01 4.14266318e-01 5.21596745e-02 2.70148039e-01 -7.21564114e-01 2.65373170e-01 -2.41244987e-01 -9.47381705e-02 6.24174416e-01 1.71135128e-01 -5.98976493e-01 -6.64087906e-02 1.07761107e-01 1.37644863e+00 2.43116558e-01 3.27730417e-01 2.58760601e-01 2.19373971e-01 1.47965655e-01 9.09841210e-02 1.27726114e+00 -4.72725183e-01 2.53341943e-01 3.03804353e-02 -2.96174675e-01 -6.09333456e-01 -1.39126766e+00 2.68800527e-01 1.56651998e+00 1.64197206e-01 -3.53809327e-01 -7.80398011e-01 -3.45324755e-01 -2.38114074e-01 1.12549412e+00 -5.12077153e-01 -2.45602384e-01 -2.57088006e-01 -2.01621383e-01 9.45212185e-01 5.38888633e-01 3.81429702e-01 -1.80603671e+00 -1.71782899e+00 -2.26121899e-02 1.08260863e-01 -8.69720995e-01 -1.18918046e-01 5.27017295e-01 -2.02420771e-01 -9.19063032e-01 -3.29273731e-01 -7.87218094e-01 1.74001336e-01 2.90157825e-01 9.66937661e-01 1.42314613e-01 -3.40292126e-01 9.25872087e-01 -4.33851868e-01 -9.30568814e-01 -5.34253657e-01 -3.51784438e-01 4.56019372e-01 -4.65058476e-01 2.20142230e-01 -7.25489974e-01 -3.14324528e-01 3.25558066e-01 -8.36983502e-01 -2.77085733e-02 4.48060602e-01 8.01605642e-01 5.16743325e-02 -2.11907886e-02 4.18109834e-01 -3.47410858e-01 8.22893977e-01 -2.89816439e-01 -8.54502201e-01 -2.64724463e-01 5.94717562e-02 -1.63045246e-02 6.01635277e-01 -5.47987401e-01 -8.93281043e-01 2.97458678e-01 -3.05031657e-01 -2.93503374e-01 -9.22924221e-01 3.49996269e-01 4.07210365e-02 -7.26812184e-02 9.40003574e-01 2.31171489e-01 -1.89448763e-02 -1.95955068e-01 4.28432703e-01 6.69975758e-01 8.13947320e-01 -4.20848757e-01 8.95223737e-01 4.82764065e-01 -1.39845029e-01 -1.06545079e+00 -4.07681346e-01 -2.78066158e-01 -1.20658122e-01 -3.19479376e-01 9.20142353e-01 -6.16459548e-01 -1.29997861e+00 5.51800489e-01 -1.07228661e+00 -8.23285818e-01 -1.79075763e-01 9.35063958e-01 -9.74118173e-01 1.60538957e-01 -5.77992141e-01 -9.65450644e-01 -4.90872422e-03 -1.24767196e+00 9.33493137e-01 4.33178127e-01 -3.42240334e-01 -6.71227217e-01 6.92374349e-01 4.83026773e-01 4.42726195e-01 -8.07850435e-02 3.94937634e-01 -8.11246634e-01 -4.79952544e-01 -2.47701239e-02 4.68927801e-01 -9.44514666e-03 -1.52191505e-01 -1.70753062e-01 -1.20793498e+00 -5.27945757e-01 1.48510501e-01 -9.65135753e-01 9.31411207e-01 4.13420260e-01 6.13247752e-01 -1.10822693e-01 -1.57066241e-01 5.55267572e-01 9.63878512e-01 4.85361964e-01 3.05515856e-01 4.64228004e-01 1.06880538e-01 6.23159051e-01 5.12943149e-01 5.40904403e-01 2.72816449e-01 6.53668344e-01 1.05903280e+00 2.26838499e-01 1.98592439e-01 -2.12252483e-01 7.60559022e-01 4.03523684e-01 -7.97961876e-02 -4.78731692e-01 -8.34405720e-01 2.86828995e-01 -1.66574550e+00 -1.34864318e+00 4.17102575e-01 1.93711615e+00 4.74063158e-01 4.18889821e-01 3.99998039e-01 1.62227243e-01 2.26984248e-01 -5.38680851e-02 -5.63838601e-01 -6.51326656e-01 2.18724757e-01 2.21866325e-01 -2.58084722e-02 7.56499648e-01 -1.03067362e+00 9.46678877e-01 6.12724304e+00 2.24002898e-01 -1.18361783e+00 -1.28648981e-01 -2.30100587e-01 -3.00288677e-01 4.38553207e-02 -1.09654881e-01 -3.45820338e-01 1.76105216e-01 9.44438279e-01 8.63787085e-02 8.99822235e-01 8.51290047e-01 2.92855799e-01 -3.84063423e-01 -1.11548030e+00 9.79220331e-01 1.95374146e-01 -8.86214256e-01 -4.02449131e-01 1.89202577e-01 1.70630217e-01 4.48618412e-01 3.78026396e-01 5.36209166e-01 1.00944006e+00 -1.25050390e+00 6.92663372e-01 2.27828503e-01 2.10880294e-01 -5.16931355e-01 4.86884296e-01 7.56246209e-01 -8.41473937e-01 -2.95887351e-01 -1.66869983e-01 -5.53572774e-01 1.38670638e-01 -3.18282336e-01 -1.22108877e+00 1.44241482e-01 8.61765325e-01 2.70366371e-01 -4.36438143e-01 1.40804100e+00 -4.37077403e-01 7.21119285e-01 -5.01142621e-01 -3.60221922e-01 3.17007601e-01 1.92571625e-01 8.63888919e-01 1.01909077e+00 1.94722340e-01 1.05462857e-01 4.41502750e-01 5.75407267e-01 2.84003794e-01 -3.67372662e-01 -8.93021464e-01 1.04799740e-01 3.97793084e-01 1.03155065e+00 -6.27925992e-01 -3.72330770e-02 -4.33093429e-01 7.91729152e-01 1.20724156e-01 2.76956350e-01 -8.18806946e-01 -3.87429982e-01 7.65373051e-01 -4.37317401e-01 3.79582167e-01 -6.22978151e-01 1.47304237e-01 -9.51531947e-01 -4.71786469e-01 -1.21490681e+00 2.96059877e-01 -1.06598377e+00 -1.03302026e+00 7.66385853e-01 -1.44543141e-01 -9.74636495e-01 -6.10130012e-01 -4.96243358e-01 -5.87789476e-01 3.09154272e-01 -1.03421068e+00 -7.75629580e-01 -4.26678240e-01 7.00894296e-01 7.57922351e-01 -4.39057142e-01 1.26908398e+00 -1.70563221e-01 -2.64381737e-01 3.12106401e-01 -3.07160199e-01 -1.98899470e-02 6.93980515e-01 -1.27976596e+00 3.82983625e-01 6.58951640e-01 6.08840644e-01 4.38569218e-01 1.27035856e+00 -3.41212690e-01 -1.21202409e+00 -4.48440969e-01 7.07137138e-02 -6.65742457e-01 6.02462351e-01 -3.84342074e-01 -6.19967163e-01 7.06758082e-01 6.83934748e-01 -9.50831324e-02 7.71569133e-01 -1.74841106e-01 -4.01809037e-01 3.65725011e-01 -1.11388457e+00 7.19712913e-01 8.69096518e-01 -5.10353923e-01 -9.74853933e-01 -5.30628450e-02 6.63847327e-01 -2.35211462e-01 -9.13937539e-02 1.23622842e-01 8.14495742e-01 -1.22601914e+00 9.02404785e-01 -6.97087228e-01 5.39399199e-02 -6.77002549e-01 -2.98127681e-01 -1.67017579e+00 -1.50141478e-01 -8.15250695e-01 2.69182981e-03 7.75894403e-01 6.56073168e-02 -9.20035243e-01 6.67798817e-01 -2.24680319e-01 -4.80643511e-02 -2.06500605e-01 -8.87780547e-01 -7.27593839e-01 -1.55509546e-01 -6.16705477e-01 1.48299262e-01 5.14167666e-01 2.30765074e-01 6.06818497e-01 -2.77973890e-01 5.60229242e-01 6.88599885e-01 -9.40068737e-02 1.12721634e+00 -1.16867363e+00 -9.05004323e-01 -4.80770707e-01 -5.44137239e-01 -9.85148311e-01 1.68535680e-01 -6.23447120e-01 4.03228492e-01 -1.35987127e+00 -9.44303051e-02 4.88840938e-02 -2.68726438e-01 5.97000957e-01 2.18610242e-01 3.52222234e-01 3.95614058e-01 -1.11055173e-01 -7.78646111e-01 6.60027385e-01 9.94554996e-01 -7.60615617e-02 -1.71446130e-01 3.19202989e-02 -5.32736659e-01 1.04904437e+00 1.27288747e+00 -5.49925447e-01 -6.18067563e-01 -4.17382628e-01 1.23055568e-02 1.13310508e-01 6.43449664e-01 -1.53245890e+00 4.14460570e-01 -2.84645021e-01 2.61979133e-01 -1.39734522e-01 9.66081977e-01 -9.18544412e-01 -1.90013483e-01 8.38186800e-01 -6.16344810e-01 2.45728314e-01 3.74629140e-01 5.24012387e-01 1.83687851e-01 -1.51831105e-01 7.16654003e-01 -3.83744210e-01 -7.42674947e-01 -2.72228509e-01 -1.18404973e+00 4.21492085e-02 1.13526750e+00 4.13438454e-02 -6.19873941e-01 -8.41108441e-01 -5.97221673e-01 1.58770293e-01 6.38185084e-01 5.52918613e-01 7.54547179e-01 -8.20312798e-01 -5.97993970e-01 2.47866869e-01 -7.70113394e-02 -3.41173172e-01 -6.22990727e-02 4.32927519e-01 -6.31832421e-01 4.15348411e-01 -6.12015486e-01 -5.32132089e-01 -1.50464964e+00 8.29643190e-01 5.59421241e-01 1.63372919e-01 -3.72269034e-01 9.96414423e-01 4.57320184e-01 -6.57504022e-01 4.77724910e-01 -5.33101857e-02 -1.72107130e-01 -4.38478768e-01 6.79113984e-01 1.61242008e-01 -3.15751225e-01 -5.59520543e-01 -4.30926770e-01 2.00407073e-01 2.78051704e-01 -7.87528932e-01 1.31434548e+00 1.42733037e-01 4.75965351e-01 5.19910634e-01 6.53268635e-01 1.23480164e-01 -1.32767606e+00 1.01497462e-02 -3.98569286e-01 -3.25352103e-01 4.13297676e-02 -1.00369525e+00 -6.74126148e-01 1.07703340e+00 8.06969583e-01 2.95915127e-01 1.01440263e+00 1.56408697e-02 5.24681628e-01 9.17008340e-01 7.08338916e-01 -7.39650846e-01 7.33283758e-01 5.95075727e-01 8.78685474e-01 -9.68302429e-01 -2.63931155e-01 9.45763215e-02 -6.52004957e-01 7.80625045e-01 8.51146042e-01 -2.48057082e-01 4.64278042e-01 7.55891383e-01 6.32450759e-01 -2.38129139e-01 -1.19672990e+00 -4.69772458e-01 -3.24074566e-01 1.17159343e+00 -1.13509567e-02 -6.05761223e-02 3.37003559e-01 2.53566653e-01 -6.74930394e-01 -4.41318098e-03 7.68376350e-01 1.23089039e+00 -7.40135312e-01 -8.94997120e-01 -7.13134885e-01 -1.08931981e-01 -4.45573002e-01 1.06411554e-01 -5.63004255e-01 3.97441626e-01 -6.99683651e-02 1.30753994e+00 -2.09125966e-01 -5.37305593e-01 2.04854146e-01 -1.89377636e-01 6.33090973e-01 -6.65015996e-01 -6.05669141e-01 -2.40406498e-01 1.24116223e-02 -5.45715809e-01 -3.38462889e-01 -7.43216336e-01 -1.30592239e+00 -1.05009988e-01 -6.88946843e-02 2.46257290e-01 6.64317131e-01 5.84887326e-01 1.18880376e-01 4.95130837e-01 1.98743850e-01 -1.50676513e+00 -6.64361179e-01 -6.33341193e-01 -3.06738079e-01 5.05785830e-02 7.37243533e-01 -8.38574111e-01 -5.30227244e-01 -5.11723496e-02]
[4.265717029571533, 1.0354903936386108]
b549c1cd-ffc0-4d09-b5b5-b6fd702e7fce
stn-homography-estimate-homography-parameters
1906.02539
null
https://arxiv.org/abs/1906.02539v1
https://arxiv.org/pdf/1906.02539v1.pdf
STN-Homography: estimate homography parameters directly
In this paper, we introduce the STN-Homography model to directly estimate the homography matrix between image pair. Different most CNN-based homography estimation methods which use an alternative 4-point homography parameterization, we use prove that, after coordinate normalization, the variance of elements of coordinate normalized $3\times3$ homography matrix is very small and suitable to be regressed well with CNN. Based on proposed STN-Homography, we use a hierarchical architecture which stacks several STN-Homography models and successively reduce the estimation error. Effectiveness of the proposed method is shown through experiments on MSCOCO dataset, in which it significantly outperforms the state-of-the-art. The average processing time of our hierarchical STN-Homography with 1 stage is only 4.87 ms on the GPU, and the processing time for hierarchical STN-Homography with 3 stages is 17.85 ms. The code will soon be open sourced.
['Qiang Zhou', 'Xin Li']
2019-06-06
null
null
null
null
['homography-estimation']
['computer-vision']
[-2.00287059e-01 -4.19155173e-02 1.54997855e-01 -2.99253035e-02 -3.11068773e-01 -2.65347987e-01 3.82019341e-01 -3.84999156e-01 -3.42910409e-01 3.30017716e-01 -4.08039689e-02 -2.04325780e-01 2.72941202e-01 -9.31103110e-01 -1.06881177e+00 -5.22253990e-01 1.92498595e-01 3.74913007e-01 2.65289128e-01 -2.24997163e-01 5.88708103e-01 4.71464425e-01 -1.08170938e+00 -2.70658225e-01 6.02344155e-01 1.04154956e+00 -2.49461625e-02 8.30224991e-01 3.51487249e-01 3.35066050e-01 -3.78216654e-01 -5.59200644e-01 7.86888599e-01 -5.54939687e-01 -4.90053356e-01 7.17385039e-02 9.23157692e-01 -6.57102346e-01 -7.11223483e-01 1.25588572e+00 4.31907088e-01 3.38148661e-02 2.90988833e-01 -1.04057717e+00 -3.57611805e-01 2.80726284e-01 -7.22784698e-01 -9.80789661e-02 1.02514260e-01 -5.71391732e-02 7.98311174e-01 -7.15951085e-01 9.47143793e-01 1.21749210e+00 7.91419804e-01 -2.19520200e-02 -1.05036473e+00 -8.61398518e-01 -6.13597453e-01 1.94730997e-01 -1.58180761e+00 -2.69359211e-03 8.08387101e-01 -2.57710934e-01 1.24494874e+00 6.87847426e-03 9.75500822e-01 5.37173450e-01 8.52370501e-01 4.80670094e-01 9.66180265e-01 -3.62753272e-01 -1.47863790e-01 -4.20609057e-01 1.30020455e-01 1.01313448e+00 4.71607924e-01 3.24061573e-01 -4.69581306e-01 4.26225439e-02 1.51844382e+00 4.41160239e-02 -1.27067491e-01 -4.45745707e-01 -1.49131989e+00 7.49226809e-01 6.06117308e-01 1.77360654e-01 -1.38728052e-01 6.32934928e-01 2.60433733e-01 1.50303736e-01 9.34919789e-02 3.67376119e-01 7.85832468e-04 -2.82551140e-01 -1.08876538e+00 9.66485292e-02 8.02742422e-01 1.62360930e+00 1.17438519e+00 1.61834285e-01 6.01800919e-01 5.00637591e-01 8.22719000e-03 5.66838324e-01 4.62117463e-01 -1.33504307e+00 4.35591012e-01 5.82784891e-01 -4.32077348e-01 -1.77554548e+00 -3.58697265e-01 -4.68492687e-01 -1.31546652e+00 -9.23247542e-03 2.26730809e-01 -1.08077869e-01 -8.40408802e-01 1.06635988e+00 1.66132390e-01 3.26209664e-01 -2.74393469e-01 8.95312607e-01 6.02440000e-01 5.30223072e-01 -6.22349143e-01 1.94666326e-01 1.09185576e+00 -1.43737078e+00 -7.23008156e-01 -1.48172498e-01 6.77490711e-01 -1.02717662e+00 5.73932290e-01 3.08978021e-01 -1.30392981e+00 -6.29218280e-01 -1.62519908e+00 -7.05303073e-01 -3.10534686e-01 4.02203083e-01 7.13814497e-01 3.52987498e-01 -1.26506317e+00 9.34061110e-01 -9.26542163e-01 -4.28422362e-01 -3.68981250e-02 7.11616457e-01 -5.65429628e-01 3.02472532e-01 -9.21987534e-01 7.63797224e-01 6.05872452e-01 3.13079745e-01 -5.87291121e-01 -5.72707593e-01 -8.72317016e-01 2.32827276e-01 5.21309115e-02 -1.01952350e+00 9.03879881e-01 -4.83563691e-01 -1.75716388e+00 7.85727262e-01 -1.81293428e-01 -5.86205661e-01 7.60413408e-01 -1.15672775e-01 2.41413098e-02 2.14529127e-01 -1.56771038e-02 8.79268229e-01 7.66523480e-01 -1.04475844e+00 -6.75748110e-01 -3.06229264e-01 -2.53006876e-01 2.29282036e-01 -2.28646956e-02 -3.70269924e-01 -9.69632387e-01 -2.69827783e-01 8.05187643e-01 -1.20298517e+00 -3.82611275e-01 -1.15447186e-01 -5.45304477e-01 4.22290087e-01 6.52638137e-01 -7.54576623e-01 1.13545716e+00 -1.87278318e+00 -2.33955048e-02 4.61662263e-01 4.70924109e-01 7.53026977e-02 3.49223882e-01 2.61387050e-01 3.25235464e-02 -1.39393747e-01 -1.53170610e-02 -3.29348028e-01 -1.54329181e-01 4.33208533e-02 -7.25199878e-02 7.99394786e-01 -4.20606315e-01 1.00334382e+00 -5.79589844e-01 -4.46977109e-01 5.75554073e-01 7.50943363e-01 -5.53726077e-01 -3.07196099e-03 4.50319499e-01 2.07826823e-01 -1.06675513e-01 4.47485209e-01 1.04904962e+00 -4.29589748e-01 2.51522481e-01 -7.71881461e-01 -3.14185292e-01 -5.25067262e-02 -1.19498634e+00 1.74591875e+00 -4.23529297e-01 9.96814251e-01 -1.67611182e-01 -5.67281723e-01 1.17088687e+00 4.87606898e-02 4.93489057e-01 -5.55706680e-01 8.06843340e-01 3.39912832e-01 9.16191936e-02 1.12141684e-01 6.91909850e-01 3.48500133e-01 2.15997189e-01 -5.55774523e-03 3.32622558e-01 -3.56914937e-01 2.06611425e-01 9.27916020e-02 8.11150193e-01 2.45105103e-02 6.35667324e-01 -5.08231163e-01 7.04733431e-01 1.21318437e-01 4.19729710e-01 6.31725729e-01 -7.61184022e-02 8.61214280e-01 5.60335934e-01 -6.90737903e-01 -1.58092892e+00 -7.40894616e-01 4.33995388e-02 1.17406219e-01 4.81760144e-01 -5.90187252e-01 -9.97748852e-01 -8.35556984e-02 -1.07455820e-01 1.03898071e-01 -4.61193949e-01 1.44311354e-01 -1.03783083e+00 -4.62222070e-01 5.36156297e-01 5.36983669e-01 1.33301091e+00 -5.61786711e-01 -7.04288125e-01 2.49719575e-01 -9.71994177e-03 -1.41472650e+00 -6.94427013e-01 -1.80835396e-01 -1.36156714e+00 -9.94122684e-01 -6.82664454e-01 -8.40021133e-01 8.72409165e-01 5.33245862e-01 9.76133704e-01 3.84402215e-01 -2.54329164e-02 2.37651896e-02 2.30038632e-02 2.73354620e-01 6.72426075e-02 3.92474413e-01 -2.15271145e-01 -2.33153388e-01 3.01378876e-01 -7.93780148e-01 -8.49626541e-01 4.17046100e-01 -5.37794232e-01 4.20295984e-01 6.19604111e-01 9.01853204e-01 8.17562819e-01 -1.66724846e-01 -4.57972556e-01 -8.25655937e-01 2.12732434e-01 1.27716571e-01 -1.19077361e+00 -2.55198646e-02 -8.01326632e-01 4.27591754e-03 6.84640825e-01 -3.12748879e-01 -7.68543065e-01 5.00380814e-01 -2.11365018e-02 -9.40962255e-01 2.23681897e-01 1.79193795e-01 -5.31672724e-02 -8.08685839e-01 3.23313475e-01 2.18317688e-01 5.09745404e-02 -3.46821964e-01 3.61105472e-01 1.45441413e-01 1.02191150e+00 -1.60346642e-01 1.16450012e+00 8.33091915e-01 6.38989568e-01 -8.77569854e-01 -2.27901533e-01 -5.77810407e-01 -1.05455136e+00 -1.50064424e-01 1.03442061e+00 -9.87891793e-01 -1.17188036e+00 4.59889174e-01 -1.48602200e+00 -3.57620791e-02 5.21995068e-01 6.93886101e-01 -6.62556589e-01 7.78185844e-01 -7.39146054e-01 -2.09088340e-01 -5.83214164e-01 -1.54326737e+00 1.11525786e+00 1.21288396e-01 -6.02854118e-02 -1.10109174e+00 1.49078909e-02 2.30310544e-01 1.69774830e-01 2.19416678e-01 5.26872456e-01 -1.93482727e-01 -1.14234734e+00 -4.35415149e-01 -4.06654924e-01 1.39661893e-01 -4.00671065e-01 9.63896513e-02 -5.44281900e-01 -2.21045420e-01 2.07592353e-01 2.61363536e-01 5.27544916e-01 5.13050854e-01 9.01536286e-01 -1.77125379e-01 -1.47742748e-01 1.44934118e+00 1.75555313e+00 3.62875998e-01 1.13488543e+00 6.70310378e-01 1.25602376e+00 2.61281341e-01 3.61790776e-01 7.17498288e-02 4.36856508e-01 6.83550119e-01 3.42619926e-01 -3.40883434e-01 -9.51849446e-02 -5.48618972e-01 -8.21380988e-02 1.59770834e+00 -3.16361815e-01 1.36789694e-01 -7.17209578e-01 2.47583628e-01 -1.85860550e+00 -5.43477714e-01 -6.77886486e-01 2.04672170e+00 2.31113166e-01 4.34412286e-02 -2.90807217e-01 -1.02280557e-01 6.74657226e-01 3.11763465e-01 -2.14095682e-01 -4.33766484e-01 -1.87982678e-01 8.76004696e-02 1.30241954e+00 1.03409851e+00 -1.09488118e+00 1.39351809e+00 6.60756874e+00 6.19009972e-01 -1.01987493e+00 -1.34425372e-01 3.03772181e-01 3.22188973e-01 2.39622474e-01 4.69113022e-01 -1.03521931e+00 2.09121078e-01 4.70224559e-01 -2.72710860e-01 6.01716101e-01 1.12944043e+00 -7.33987913e-02 -1.82317197e-01 -1.00891018e+00 1.64385176e+00 3.69369864e-01 -1.53092527e+00 1.14643626e-01 3.84048194e-01 1.00981510e+00 -4.11340743e-02 -1.05128177e-01 -1.97236970e-01 3.55180760e-04 -7.04793155e-01 3.68343532e-01 1.67747974e-01 5.96214414e-01 -7.53188074e-01 9.27704632e-01 1.21412024e-01 -1.43846226e+00 5.62547088e-01 -8.21396589e-01 -3.86080444e-02 3.18493515e-01 4.97124135e-01 -6.17368877e-01 7.33935952e-01 6.39840484e-01 8.43249798e-01 -6.45986259e-01 1.08670402e+00 -8.08169693e-02 5.88025041e-02 -4.00439411e-01 1.05014242e-01 2.94276416e-01 -7.36278415e-01 3.42624485e-01 1.10323036e+00 7.36962259e-01 3.35660160e-01 -2.76018947e-01 8.04234743e-01 -1.23019204e-01 1.29714325e-01 -7.46458113e-01 3.14638525e-01 2.16047212e-01 1.31192005e+00 -9.23076212e-01 -6.36663556e-01 -2.65932977e-01 1.22497082e+00 2.06638104e-03 1.78684220e-01 -6.48344934e-01 -6.30516589e-01 3.25810045e-01 -1.01037331e-01 2.56158739e-01 -6.62365139e-01 -7.94579029e-01 -1.46695638e+00 -1.88009813e-01 -7.76751339e-01 -8.59661177e-02 -9.49569166e-01 -7.26893067e-01 6.61024451e-01 6.33434355e-02 -1.45607674e+00 -3.86793494e-01 -8.56003761e-01 -5.06184518e-01 7.41930485e-01 -1.00564945e+00 -9.91348982e-01 -6.69069231e-01 4.23236370e-01 3.78653973e-01 -4.53910828e-02 3.58417481e-01 3.78664672e-01 -3.94638062e-01 5.59735477e-01 1.35189131e-01 3.20756406e-01 4.78674024e-01 -9.46920574e-01 8.35001826e-01 9.90851700e-01 -8.06062073e-02 9.89474058e-01 7.63476372e-01 -9.38759327e-01 -1.70084655e+00 -6.26134098e-01 1.07656193e+00 -2.94616789e-01 6.43622100e-01 -4.50778037e-01 -6.89757526e-01 1.05648243e+00 4.88288760e-01 -1.28140554e-01 1.25594810e-01 -4.15872872e-01 -2.13679090e-01 -1.78182423e-01 -7.97422290e-01 7.85546184e-01 1.10381854e+00 -5.41913748e-01 -7.74058700e-02 1.22434206e-01 5.03421307e-01 -1.06364369e+00 -1.02438414e+00 3.76713246e-01 8.11175048e-01 -1.33040488e+00 1.02240932e+00 1.23656183e-01 5.46923459e-01 -3.77118260e-01 -1.93310156e-01 -9.21128333e-01 -4.12266463e-01 -8.42131615e-01 2.94021368e-02 4.46101546e-01 1.56596482e-01 -6.89822614e-01 1.02922750e+00 3.20255846e-01 -1.71469867e-01 -5.06075263e-01 -8.71354640e-01 -7.73822427e-01 -1.73772976e-01 -9.75709632e-02 4.85498250e-01 9.78620172e-01 -9.85158905e-02 4.32862222e-01 -7.16465294e-01 2.55918443e-01 8.65234673e-01 2.52721403e-02 1.27970457e+00 -8.53241205e-01 -1.93000808e-01 -4.83817697e-01 -9.96318698e-01 -1.57309127e+00 2.01180521e-02 -5.52631140e-01 -9.09066945e-02 -1.14014208e+00 5.35805486e-02 4.10562277e-01 4.69061732e-01 5.92353754e-03 2.81031221e-01 4.31562036e-01 3.91810149e-01 2.46910214e-01 -1.22990489e-01 4.48698550e-01 1.64366949e+00 1.59914959e-02 -1.32195875e-01 -4.38773930e-01 9.71931517e-02 8.46483707e-01 6.27948642e-01 -1.54806599e-01 -3.10379207e-01 -5.58736980e-01 8.86363387e-02 1.72321230e-01 1.55451894e-01 -1.29928648e+00 6.34300828e-01 2.69484848e-01 5.50740421e-01 -1.23315883e+00 4.54432517e-01 -8.66076291e-01 6.03794515e-01 5.52143097e-01 1.54114395e-01 5.03058910e-01 8.48747492e-02 8.72016102e-02 -3.56478512e-01 -9.13514867e-02 7.23325610e-01 -2.62030631e-01 -5.02909482e-01 4.99568403e-01 -5.44936303e-03 -4.33196157e-01 6.78058624e-01 -4.99453902e-01 -3.50761950e-01 -6.05521798e-01 -4.40748721e-01 -1.60535157e-01 7.22419262e-01 -1.16840815e-02 7.17945218e-01 -1.56853044e+00 -9.86096412e-02 3.23184729e-01 -4.10118610e-01 -4.80920821e-02 2.20738605e-01 8.35570157e-01 -1.61053109e+00 9.39966202e-01 -5.51142097e-01 -6.77145839e-01 -1.15296817e+00 3.71288002e-01 4.89366949e-01 -1.61890462e-01 -7.48140037e-01 3.75310063e-01 4.06549126e-01 -3.39148521e-01 1.02051921e-01 -3.59888762e-01 1.12081960e-01 -5.10748327e-01 1.46462604e-01 9.40802872e-01 5.64431697e-02 -9.97880161e-01 -4.93881702e-02 1.23859584e+00 4.28225473e-02 -3.96892130e-01 8.70640159e-01 -8.95980969e-02 -4.87507015e-01 -1.96632072e-02 1.85528028e+00 -3.95387828e-01 -1.11446357e+00 2.59608388e-01 -3.50139856e-01 -6.61751211e-01 1.97795585e-01 4.87020127e-02 -1.40349352e+00 1.21962333e+00 6.11780226e-01 -2.31663421e-01 9.60350990e-01 -5.46612561e-01 1.07202733e+00 5.14262557e-01 3.91889900e-01 -1.03855038e+00 -1.08946368e-01 8.31630170e-01 8.44636023e-01 -9.43287969e-01 4.35520768e-01 -8.42751384e-01 -2.91902483e-01 1.40921783e+00 8.92023027e-01 -8.50625455e-01 8.27794015e-01 4.93778661e-02 8.15789700e-02 -3.28902185e-01 -2.73918837e-01 1.03911459e-01 3.38883460e-01 2.84744114e-01 3.21482241e-01 -1.06643945e-01 -5.31882405e-01 -2.10490733e-01 -8.25229645e-01 -2.94722825e-01 6.39761567e-01 5.86103320e-01 -1.45214736e-01 -1.07773054e+00 -4.34331506e-01 4.42183539e-02 -2.75666248e-02 -2.12470353e-01 -4.45061803e-01 1.14085293e+00 -1.65331885e-01 4.69158709e-01 3.65036011e-01 -8.27551544e-01 3.78510922e-01 -4.48394001e-01 6.06228590e-01 1.18372738e-02 -6.62718356e-01 6.21488452e-01 -1.97401524e-01 -9.74113047e-01 -2.38349661e-01 -2.54683048e-01 -1.15821755e+00 -9.24937010e-01 -1.60701558e-01 -2.97639549e-01 8.03628087e-01 7.64074087e-01 2.36367121e-01 6.28580004e-02 4.91025358e-01 -1.15681517e+00 -1.68695167e-01 -7.88642526e-01 -4.36723381e-01 9.62007567e-02 1.66169375e-01 -5.17952085e-01 -5.12768924e-01 -4.92045470e-03]
[8.586602210998535, -2.2151410579681396]
366bd791-618d-494f-8cfb-070a4c5f75cf
multimodal-multipart-learning-for-action
1507.08761
null
http://arxiv.org/abs/1507.08761v1
http://arxiv.org/pdf/1507.08761v1.pdf
Multimodal Multipart Learning for Action Recognition in Depth Videos
The articulated and complex nature of human actions makes the task of action recognition difficult. One approach to handle this complexity is dividing it to the kinetics of body parts and analyzing the actions based on these partial descriptors. We propose a joint sparse regression based learning method which utilizes the structured sparsity to model each action as a combination of multimodal features from a sparse set of body parts. To represent dynamics and appearance of parts, we employ a heterogeneous set of depth and skeleton based features. The proper structure of multimodal multipart features are formulated into the learning framework via the proposed hierarchical mixed norm, to regularize the structured features of each part and to apply sparsity between them, in favor of a group feature selection. Our experimental results expose the effectiveness of the proposed learning method in which it outperforms other methods in all three tested datasets while saturating one of them by achieving perfect accuracy.
['Tian-Tsong Ng', 'Gang Wang', 'Amir Shahroudy', 'Qingxiong Yang']
2015-07-31
null
null
null
null
['multimodal-activity-recognition']
['computer-vision']
[ 3.30860436e-01 -1.16352409e-01 -4.37159270e-01 -1.55359030e-01 -5.04915893e-01 -2.07372010e-01 7.26554096e-01 -1.29799575e-01 -1.85607970e-01 6.74071014e-01 7.02373505e-01 7.16835439e-01 -4.31160778e-01 -3.35173160e-01 -3.76660645e-01 -9.70444441e-01 -6.23337738e-02 5.09700954e-01 2.48006791e-01 -2.24964842e-01 2.07013234e-01 4.70277101e-01 -1.86176169e+00 4.52707201e-01 7.48401344e-01 1.08249867e+00 -1.03510588e-01 3.39806914e-01 -2.54510101e-02 1.05243814e+00 -1.96005955e-01 -1.47374496e-01 5.49348950e-01 -4.92734820e-01 -5.49995363e-01 7.81035125e-01 4.78227645e-01 -2.08456233e-01 -2.57174611e-01 7.98401952e-01 3.57984126e-01 3.77757251e-01 8.78720880e-01 -1.11074638e+00 4.98811752e-02 2.60659993e-01 -8.72462273e-01 -2.14994252e-02 8.45706105e-01 -2.50653923e-02 9.94169176e-01 -8.20401788e-01 6.21820986e-01 1.32484937e+00 5.13897598e-01 2.01010495e-01 -1.10613179e+00 -1.96819544e-01 2.82597572e-01 2.91220427e-01 -1.29215884e+00 -4.80959028e-01 1.12108099e+00 -6.91111028e-01 5.20541131e-01 2.03311905e-01 8.25877011e-01 8.29341292e-01 2.66597718e-01 9.48213518e-01 1.22816849e+00 -2.24935427e-01 1.06656656e-01 -2.02628717e-01 3.43693405e-01 1.13705671e+00 2.12790832e-01 -2.71595418e-01 -7.13120461e-01 -3.64508510e-01 6.01463079e-01 4.10415947e-01 -1.36198308e-02 -7.16019571e-01 -1.12685597e+00 7.27747500e-01 1.95986405e-03 3.73126775e-01 -9.09867108e-01 -7.54316002e-02 4.03589517e-01 -1.57927781e-01 2.34268174e-01 -2.22319320e-01 -3.63121510e-01 -2.46671494e-02 -8.89514089e-01 4.53608751e-01 5.98860264e-01 5.75345099e-01 8.38148534e-01 -2.58748862e-03 -3.15076202e-01 8.72836947e-01 5.89362860e-01 5.09418309e-01 6.51286244e-01 -9.71482456e-01 4.46971476e-01 1.13388538e+00 -4.00542542e-02 -1.34797859e+00 -4.92138505e-01 -1.05421960e-01 -8.38503957e-01 2.87270218e-01 4.76826578e-01 7.90952966e-02 -9.02941644e-01 1.53001118e+00 6.76938117e-01 -7.01913238e-02 -2.60524601e-01 9.48526561e-01 5.99261642e-01 4.78338778e-01 1.37218773e-01 -3.04074824e-01 1.20265448e+00 -9.40862656e-01 -8.04988325e-01 1.85361072e-01 4.36382830e-01 -7.22894251e-01 4.29088742e-01 6.23559833e-01 -1.04345381e+00 -6.87501788e-01 -7.97364533e-01 8.31833407e-02 -9.19586793e-02 5.77641547e-01 6.27739429e-01 1.69843361e-01 -5.66930532e-01 6.24816954e-01 -1.03628087e+00 -3.71379882e-01 2.57705271e-01 6.41393244e-01 -7.54933000e-01 -3.63077149e-02 -6.86313689e-01 6.91813290e-01 3.25010657e-01 2.43716493e-01 -7.69618988e-01 -1.17310695e-01 -1.08651483e+00 -2.70619512e-01 4.35016870e-01 -7.67979085e-01 6.36395752e-01 -1.20534837e+00 -1.63835788e+00 5.53537130e-01 -4.75994274e-02 -2.12418690e-01 5.06856024e-01 -4.00289804e-01 -7.41279647e-02 5.40225983e-01 3.85742337e-02 4.14418578e-01 1.23158443e+00 -1.27431405e+00 -5.39680839e-01 -7.18044043e-01 -9.61743593e-02 4.36056048e-01 -3.40772927e-01 -6.68565929e-02 -3.41937393e-01 -6.26525998e-01 4.62600231e-01 -9.36154842e-01 -3.76104534e-01 4.97357808e-02 -2.83609420e-01 -2.47800842e-01 9.32860076e-01 -8.19307446e-01 1.18396819e+00 -1.92346370e+00 7.93743134e-01 2.58149862e-01 2.42601126e-01 9.16179344e-02 -1.11426169e-03 4.87131208e-01 -1.24732349e-02 -6.63879633e-01 -3.55416209e-01 -5.37178516e-01 -1.39798671e-01 4.97504801e-01 3.33795220e-01 9.94172752e-01 1.38212815e-01 4.80288953e-01 -7.33742058e-01 -9.68725204e-01 4.78525490e-01 6.19540453e-01 -4.30050880e-01 3.82453114e-01 1.02379769e-01 5.42876422e-01 -1.05536950e+00 8.58552098e-01 5.02485693e-01 8.72947350e-02 2.67685920e-01 -4.27976429e-01 -2.92161442e-02 -2.99922138e-01 -1.66869354e+00 1.82185280e+00 -8.38721395e-02 -1.71416372e-01 3.56560647e-01 -1.30867898e+00 8.21816146e-01 3.19487631e-01 1.24696815e+00 -1.31634310e-01 3.53935331e-01 9.18781608e-02 -9.46228951e-03 -8.15882981e-01 2.75885854e-02 -2.08260775e-01 9.99266049e-04 2.17526689e-01 2.82947779e-01 2.11889029e-01 2.83193827e-01 -4.72894609e-02 1.11466014e+00 4.65917557e-01 5.43979883e-01 -5.33748269e-02 1.01906335e+00 -1.96253732e-01 6.00001931e-01 2.30796829e-01 -1.61630705e-01 3.80119205e-01 2.30195507e-01 -6.00725234e-01 -6.46549404e-01 -7.82337248e-01 1.73752889e-01 9.92832243e-01 5.35241514e-03 -2.83002347e-01 -4.47320342e-01 -8.25638771e-01 1.65824324e-01 -1.45009831e-01 -8.09925616e-01 -3.41727561e-03 -6.21307075e-01 -5.25894523e-01 6.23926856e-02 4.20001566e-01 4.15718794e-01 -9.52544808e-01 -6.45547032e-01 1.41884878e-01 -1.80945143e-01 -8.71731997e-01 -1.90618783e-01 1.69814557e-01 -9.37596142e-01 -1.24232173e+00 -8.63146305e-01 -4.69161361e-01 6.51617229e-01 1.51809707e-01 5.39104044e-01 -8.07530712e-03 -3.42679143e-01 7.58157909e-01 -6.89302921e-01 7.35815987e-02 8.19958095e-03 -1.88060075e-01 2.63024598e-01 8.88362288e-01 -1.38510332e-01 -8.20370078e-01 -4.49963510e-01 1.81086212e-01 -8.40464532e-01 -2.42548302e-01 9.87190485e-01 7.97902524e-01 5.94338477e-01 -2.90749278e-02 7.75106326e-02 -6.00119770e-01 2.05082580e-01 -6.28955126e-01 -1.52127415e-01 1.01458356e-01 -1.94684461e-01 8.10102597e-02 5.13978481e-01 -5.01128078e-01 -1.09013391e+00 7.33769178e-01 2.75429994e-01 -6.50173664e-01 -1.89299703e-01 5.50057411e-01 -2.97103852e-01 -2.87644058e-01 1.92164600e-01 4.53120857e-01 3.38866174e-01 -5.93207419e-01 2.63549954e-01 3.13253403e-01 3.02244991e-01 -6.91667438e-01 8.15452099e-01 5.99596560e-01 4.17557031e-01 -9.25993383e-01 -7.10171580e-01 -7.77264535e-01 -1.01177514e+00 -6.97540343e-01 1.01168132e+00 -7.90175855e-01 -7.13746369e-01 5.32481611e-01 -8.49354386e-01 2.55306393e-01 -2.43873328e-01 7.56833792e-01 -7.44230568e-01 1.00129128e+00 -4.87882137e-01 -1.04106784e+00 -1.44634685e-02 -9.82138336e-01 1.35185516e+00 4.66015423e-03 -1.65392414e-01 -7.53242195e-01 4.37646955e-01 6.91834927e-01 -7.93973282e-02 7.56675780e-01 4.96787488e-01 -5.80644131e-01 -3.33976090e-01 -5.19833088e-01 2.29513288e-01 4.35782641e-01 3.72564405e-01 1.22450814e-01 -4.98093396e-01 -2.78988689e-01 1.67241603e-01 -6.35626197e-01 9.15095448e-01 4.33289379e-01 7.92105854e-01 -1.62383571e-01 -2.71129072e-01 2.74784893e-01 1.39814389e+00 -3.23856249e-02 3.42144340e-01 -1.33052506e-02 9.24819887e-01 7.92194426e-01 8.48576367e-01 1.09736383e+00 2.24805266e-01 6.92612052e-01 5.81705689e-01 -9.95824262e-02 2.50803865e-02 -2.50588544e-03 6.45721078e-01 8.65200758e-01 -7.10600019e-01 1.69275343e-01 -6.27773106e-01 2.97514707e-01 -2.25071836e+00 -1.15435636e+00 -1.33939788e-01 2.12617397e+00 4.47932154e-01 -9.90257040e-02 6.24394417e-01 3.06237936e-01 6.37413144e-01 3.09863180e-01 -2.10151434e-01 -1.04407571e-01 -2.19318848e-02 9.04522613e-02 2.67455935e-01 2.66751498e-01 -1.19615138e+00 5.98822117e-01 6.36781120e+00 7.65156806e-01 -7.55315542e-01 -1.06470540e-01 1.52388185e-01 2.67721899e-02 2.50437763e-02 -4.67079356e-02 -6.79030120e-01 3.74366641e-01 4.40171689e-01 2.44807422e-01 1.08321898e-01 7.40188658e-01 5.05151451e-01 -3.55238229e-01 -8.88562620e-01 9.75527704e-01 3.69033515e-01 -7.26936638e-01 2.76432812e-01 1.12305813e-01 7.37379611e-01 -4.91311431e-01 -1.08208030e-01 1.78655222e-01 5.66458739e-02 -7.91975260e-01 6.26055002e-01 9.31949794e-01 3.39685604e-02 -5.13413906e-01 5.85477710e-01 4.93160874e-01 -1.43715930e+00 -3.65522385e-01 -2.31754437e-01 -3.72309387e-01 1.63620606e-01 3.73947978e-01 -2.39236519e-01 9.23936009e-01 3.97841126e-01 1.14812505e+00 -3.98092002e-01 1.00575244e+00 4.48021404e-02 4.03743684e-01 -4.23974395e-01 2.52605766e-01 2.27798507e-01 -5.52058637e-01 6.85404062e-01 8.15671861e-01 7.71867111e-02 2.57241309e-01 8.34927261e-01 4.34008956e-01 5.11633277e-01 5.67154169e-01 -6.43401146e-01 -5.62002473e-02 -1.98713049e-01 1.56395853e+00 -5.78250349e-01 -2.84477681e-01 -5.63664377e-01 9.14780974e-01 3.04882973e-01 4.01920527e-02 -8.28358710e-01 7.37754926e-02 3.30933332e-01 9.23767984e-02 4.62762743e-01 -4.29813176e-01 3.24067958e-02 -1.22537446e+00 3.26560169e-01 -9.50051188e-01 6.71055615e-01 -2.87772030e-01 -1.15651059e+00 4.55479443e-01 2.26107821e-01 -1.64906251e+00 -1.09123409e-01 -5.34967482e-01 -4.00834113e-01 4.18372393e-01 -1.12150562e+00 -1.66162992e+00 -2.90246695e-01 1.14229786e+00 6.78749502e-01 -1.94748804e-01 4.91417557e-01 3.22332203e-01 -6.62192643e-01 1.43485814e-01 -1.53808013e-01 -1.12764455e-01 5.97928882e-01 -9.96616423e-01 -8.92142415e-01 5.26475787e-01 1.02935441e-01 3.41033667e-01 6.72932208e-01 -8.94054592e-01 -1.62323356e+00 -6.61821842e-01 5.91358364e-01 -1.68873459e-01 5.66616774e-01 -6.58819452e-03 -5.02455235e-01 4.96040970e-01 3.37492265e-02 1.46038741e-01 7.44752407e-01 8.77380744e-02 -3.50717157e-02 -1.28102556e-01 -1.01629245e+00 1.34988979e-01 8.90154541e-01 -5.78118302e-02 -8.61331940e-01 3.78062338e-01 -2.45576873e-02 -6.84881136e-02 -1.02929282e+00 4.87451375e-01 7.44273186e-01 -1.10771322e+00 7.71396875e-01 -6.68092132e-01 5.15531301e-01 -5.83703339e-01 -3.14511567e-01 -8.02331269e-01 -4.91855681e-01 -3.14681292e-01 -2.86510557e-01 1.10539567e+00 -1.30182400e-01 -2.90797502e-01 9.62918520e-01 3.49218071e-01 -4.19419669e-02 -9.65372622e-01 -9.46814239e-01 -6.91407144e-01 -4.82516766e-01 1.03599485e-02 -9.40286741e-02 8.53593349e-01 1.44806933e-02 7.84116834e-02 -8.73098016e-01 -5.59436269e-02 8.44089568e-01 2.91320264e-01 8.48128855e-01 -1.28191149e+00 -4.72100198e-01 -1.56317845e-01 -1.00120282e+00 -7.12363422e-01 3.45029652e-01 -5.99092603e-01 -1.08263314e-01 -1.30356431e+00 5.00612617e-01 5.48055246e-02 -3.36015373e-01 5.80463052e-01 -1.31840676e-01 2.42656842e-01 1.51219964e-01 3.35021168e-01 -8.22601676e-01 9.17837441e-01 1.22647905e+00 -2.31972665e-01 -1.50871888e-01 1.77680075e-01 -2.02607796e-01 1.00494969e+00 2.39493355e-01 -4.06431705e-01 -3.78443539e-01 -2.33837739e-02 -3.50255460e-01 2.87148267e-01 2.50084013e-01 -1.36735272e+00 8.26759040e-02 -3.81418645e-01 4.99117672e-01 -6.01718783e-01 6.67497516e-01 -1.09420049e+00 2.37891093e-01 5.19085228e-01 -2.03992501e-01 -2.35473096e-01 -2.74636865e-01 7.52785563e-01 -5.46858251e-01 -6.17607264e-03 7.51443923e-01 -2.17106864e-01 -7.47441471e-01 3.49044293e-01 -3.33403438e-01 -2.54850000e-01 1.22735453e+00 -5.21333933e-01 5.03305078e-01 -3.50808740e-01 -1.11926222e+00 1.64907172e-01 2.30755031e-01 1.34646982e-01 6.03254378e-01 -1.57449818e+00 -6.40043437e-01 -6.28348589e-02 -3.52824070e-02 -3.69645864e-01 5.16641617e-01 1.34019077e+00 -1.83172286e-01 1.52195364e-01 -7.73112237e-01 -6.82600319e-01 -1.59107220e+00 5.63240826e-01 1.92341343e-01 -4.63484913e-01 -5.08973241e-01 5.31544566e-01 1.11475773e-01 -9.05761272e-02 2.89671272e-01 -1.46375030e-01 -7.57635951e-01 3.02227736e-01 2.65284926e-01 7.40738988e-01 -3.62089336e-01 -1.27344787e+00 -4.28843856e-01 1.13787508e+00 8.94492120e-02 3.43618430e-02 1.43710172e+00 -8.72936919e-02 -2.84980804e-01 4.59340364e-01 1.19902468e+00 1.30536824e-01 -1.14895618e+00 -2.61010438e-01 -1.61647618e-01 -5.91580629e-01 -2.04356238e-01 -3.91569227e-01 -1.10593593e+00 4.60198283e-01 4.50713307e-01 -1.53686807e-01 1.30013418e+00 -7.81867132e-02 6.30562067e-01 3.10888916e-01 3.75297070e-01 -1.09508550e+00 3.44580621e-01 3.41459453e-01 9.11591113e-01 -1.20112705e+00 4.05815333e-01 -6.21822119e-01 -7.85464942e-01 1.33178258e+00 5.10917544e-01 -6.77510142e-01 6.88821673e-01 -9.17046815e-02 -1.61349088e-01 -4.07792687e-01 -5.68762124e-01 -4.88644689e-01 6.30394220e-01 3.21202397e-01 3.29613537e-01 -1.63454190e-01 -9.99720752e-01 5.55549860e-01 3.77323747e-01 3.26204076e-02 7.93045480e-03 1.20274067e+00 -5.65427423e-01 -1.29422426e+00 -5.26188552e-01 3.99143070e-01 -4.33086425e-01 4.95744735e-01 -5.97272754e-01 7.97711849e-01 5.37043333e-01 8.30009878e-01 -4.18770939e-01 -4.83555257e-01 3.83698642e-01 1.13022722e-01 7.40096807e-01 -5.18350124e-01 -5.44683039e-01 4.05883998e-01 6.48706704e-02 -9.46457684e-01 -1.02293599e+00 -1.13935053e+00 -1.01989877e+00 3.17422509e-01 -1.69090778e-01 -3.69380340e-02 3.38663161e-01 1.28391969e+00 -1.97377056e-02 1.48470074e-01 8.97219598e-01 -1.16735578e+00 -8.40987563e-01 -8.33716094e-01 -9.76542652e-01 8.71308088e-01 2.05149308e-01 -1.34711277e+00 -3.90593827e-01 2.21920788e-01]
[7.89186954498291, 0.40472736954689026]
5ce598d7-fdf3-4b2c-be2e-67e6a1e468ff
counterfactual-explanations-for-predictive
2202.12018
null
https://arxiv.org/abs/2202.12018v1
https://arxiv.org/pdf/2202.12018v1.pdf
Counterfactual Explanations for Predictive Business Process Monitoring
Predictive business process monitoring increasingly leverages sophisticated prediction models. Although sophisticated models achieve consistently higher prediction accuracy than simple models, one major drawback is their lack of interpretability, which limits their adoption in practice. We thus see growing interest in explainable predictive business process monitoring, which aims to increase the interpretability of prediction models. Existing solutions focus on giving factual explanations.While factual explanations can be helpful, humans typically do not ask why a particular prediction was made, but rather why it was made instead of another prediction, i.e., humans are interested in counterfactual explanations. While research in explainable AI produced several promising techniques to generate counterfactual explanations, directly applying them to predictive process monitoring may deliver unrealistic explanations, because they ignore the underlying process constraints. We propose LORELEY, a counterfactual explanation technique for predictive process monitoring, which extends LORE, a recent explainable AI technique. We impose control flow constraints to the explanation generation process to ensure realistic counterfactual explanations. Moreover, we extend LORE to enable explaining multi-class classification models. Experimental results using a real, public dataset indicate that LORELEY can approximate the prediction models with an average fidelity of 97.69\% and generate realistic counterfactual explanations.
['Klaus Pohl', 'Andreas Metzger', 'Tsung-Hao Huang']
2022-02-24
null
null
null
null
['counterfactual-explanation', 'predictive-process-monitoring']
['miscellaneous', 'time-series']
[ 5.68670094e-01 1.01922774e+00 -5.82780480e-01 -5.15720844e-01 -2.45808829e-02 -2.95872718e-01 1.03231537e+00 1.57060504e-01 3.61840755e-01 9.44677889e-01 4.42749947e-01 -9.16353047e-01 -4.43656474e-01 -8.90699089e-01 -7.00025260e-01 -9.32364985e-02 1.23950183e-01 7.65777349e-01 -3.10230702e-01 2.99043775e-01 5.21339774e-01 4.54313345e-02 -1.53441834e+00 6.16464853e-01 1.11718869e+00 7.48951554e-01 -9.76397991e-02 6.96524799e-01 -2.52305478e-01 1.40980303e+00 -5.12872577e-01 -6.54285133e-01 2.80557603e-01 -6.59023702e-01 -7.92696357e-01 1.52850255e-01 -9.63426009e-02 -2.68491864e-01 2.26242527e-01 6.05267346e-01 -4.02418017e-01 4.04310562e-02 7.97128618e-01 -2.00511003e+00 -9.53048885e-01 1.00557601e+00 -3.71077061e-01 1.06423080e-01 5.36176741e-01 5.88297963e-01 1.28183270e+00 -2.83318222e-01 3.06714386e-01 1.43489635e+00 5.00139594e-01 8.65168273e-01 -1.50852251e+00 -8.45954478e-01 6.00757957e-01 4.26737331e-02 -5.27470708e-01 -2.86451936e-01 6.05370998e-01 -4.38433588e-01 1.04670465e+00 7.39465714e-01 9.41630661e-01 1.06501651e+00 7.09381461e-01 7.56343603e-01 1.32276237e+00 -3.34380716e-01 5.19004762e-01 1.28380716e-01 2.32416913e-01 2.82741249e-01 9.33563948e-01 5.54501891e-01 -6.11797690e-01 -4.90905851e-01 7.16188371e-01 6.78335249e-01 -3.09371263e-01 -3.74543428e-01 -1.25474465e+00 9.26756382e-01 2.34260529e-01 6.25941083e-02 -9.52237308e-01 4.78370756e-01 -7.43940696e-02 2.96712488e-01 3.69805843e-01 1.14742672e+00 -6.24001205e-01 -3.04650992e-01 -6.49121344e-01 6.36883140e-01 1.15352619e+00 9.15897012e-01 4.34397966e-01 7.85608143e-02 -4.03400660e-01 1.04102055e-02 6.42100930e-01 3.84979665e-01 6.47748351e-01 -1.49484932e+00 3.55129331e-01 9.31469023e-01 4.77797091e-01 -8.01005185e-01 -8.18208382e-02 -2.29925379e-01 -6.39343619e-01 2.30085850e-01 2.70551324e-01 -7.19217025e-03 -7.35730588e-01 1.32775867e+00 -1.96430460e-01 1.50707051e-01 1.92450434e-01 7.92067230e-01 4.56320718e-02 6.41886055e-01 5.66856265e-01 -6.58811033e-01 1.03520322e+00 -1.16340351e+00 -8.83072555e-01 -5.96181691e-01 4.08065617e-01 -2.99610764e-01 1.11682963e+00 5.53556919e-01 -1.00103092e+00 -2.84912914e-01 -7.54094779e-01 5.77787876e-01 6.97421506e-02 -5.78017712e-01 1.46900296e+00 6.53827369e-01 -6.06791496e-01 8.28951478e-01 -8.39322329e-01 -4.78968732e-02 4.77951616e-01 2.98301071e-01 1.72130689e-02 2.75391266e-02 -8.75307918e-01 7.50959575e-01 4.44108397e-01 -3.22202235e-01 -7.23372281e-01 -9.70965326e-01 -7.58678079e-01 6.53430641e-01 6.84949517e-01 -1.12801850e+00 1.71440017e+00 -1.33842659e+00 -1.26038146e+00 4.76578902e-03 -5.37601113e-01 -1.04069722e+00 7.31302261e-01 -2.11103395e-01 -4.51865703e-01 -3.79656374e-01 1.41886309e-01 5.69397211e-01 5.80721259e-01 -1.41478896e+00 -7.83252537e-01 -1.34149984e-01 6.49081692e-02 1.04429357e-01 2.11965144e-01 -2.97560185e-01 4.27183211e-01 -3.39041620e-01 2.86031336e-01 -8.28383386e-01 -7.59159207e-01 -2.58510321e-01 -4.98427868e-01 -1.20175794e-01 6.07163131e-01 -2.50093520e-01 1.15105581e+00 -1.56328762e+00 -6.77722991e-01 2.30316535e-01 5.36878407e-01 -2.13707536e-01 2.65166283e-01 1.95652530e-01 -3.06315005e-01 8.26803744e-01 -2.62706906e-01 -9.64439362e-02 2.99245149e-01 3.71542901e-01 -8.80878985e-01 -1.48170561e-01 2.49597922e-01 1.03186917e+00 -9.30939853e-01 -1.07118241e-01 4.40145731e-01 -1.61989674e-01 -7.91787565e-01 2.69223154e-01 -6.41252816e-01 2.48344019e-01 -4.87745821e-01 4.11947161e-01 3.32355320e-01 -4.98477340e-01 3.90182137e-01 6.90908968e-01 1.34740947e-02 6.28010809e-01 -9.87074137e-01 6.52365565e-01 -4.39157218e-01 5.37715316e-01 -6.39812529e-01 -6.05701566e-01 8.97620261e-01 5.33588648e-01 2.83428907e-01 -3.25657576e-01 -2.62985289e-01 9.95661542e-02 3.67592186e-01 -2.04733923e-01 4.50538546e-01 -7.05496490e-01 1.08934887e-01 9.15241420e-01 -6.95811808e-01 -2.69807905e-01 -1.08814657e-01 -3.30845751e-02 1.14378119e+00 1.16967082e-01 1.11730134e+00 -1.16440140e-01 1.31099805e-01 3.65584075e-01 8.85679245e-01 1.07184148e+00 -2.82725602e-01 4.85326022e-01 6.76088035e-01 -9.11000431e-01 -8.03587794e-01 -8.90094042e-01 3.28221977e-01 5.15385866e-01 3.44444960e-02 -4.10780489e-01 -3.52896959e-01 -1.04546404e+00 2.63541043e-01 1.77673912e+00 -6.49111509e-01 -2.28621304e-01 -1.06280603e-01 -7.00047612e-01 6.13107393e-03 6.44080460e-01 4.75805700e-01 -1.10867536e+00 -8.21004927e-01 5.15844345e-01 -1.79997340e-01 -6.00215077e-01 -7.63625354e-02 -9.47991312e-02 -1.25694621e+00 -1.42596412e+00 1.37473464e-01 3.78999740e-01 4.71863568e-01 3.87246639e-01 1.33215785e+00 1.15789659e-01 4.51306522e-01 2.64653593e-01 5.66836335e-02 -1.20712781e+00 -8.68614554e-01 -3.95323753e-01 4.20342386e-02 -3.06209058e-01 8.96485388e-01 -5.45804024e-01 -5.12795150e-01 4.05077696e-01 -4.86988634e-01 4.41550523e-01 6.64344490e-01 8.51433635e-01 4.28777575e-01 1.14726461e-01 7.16804266e-01 -1.27025485e+00 9.10039485e-01 -6.20483518e-01 -2.06191808e-01 2.96347976e-01 -1.59385800e+00 2.97264248e-01 6.73910558e-01 -5.43451667e-01 -1.51952469e+00 -1.90345079e-01 4.58509266e-01 -1.97121471e-01 -3.48717600e-01 4.02273566e-01 -1.39102638e-01 7.41943240e-01 7.09206522e-01 1.07152119e-01 -8.19687173e-02 6.94365576e-02 3.17201972e-01 3.31598192e-01 2.16417953e-01 -3.57482016e-01 6.85879946e-01 4.81718659e-01 -7.27395266e-02 1.25456616e-01 -7.42342710e-01 -9.14450437e-02 -2.09517386e-02 -7.64259920e-02 5.53307772e-01 -5.72824836e-01 -1.01635718e+00 -4.89343315e-01 -1.08443367e+00 -2.95584470e-01 -6.70713425e-01 5.15894532e-01 -9.68043268e-01 -1.01759575e-01 -2.39182934e-01 -1.27450037e+00 -2.87771195e-01 -8.99148285e-01 6.43916428e-01 2.00927228e-01 -1.22294998e+00 -1.04156137e+00 -2.17682362e-01 6.65291905e-01 3.00714761e-01 2.78526634e-01 9.13059354e-01 -1.08624220e+00 -9.77458239e-01 -1.92950293e-01 -8.70202552e-04 -1.89706251e-01 2.74148971e-01 7.25051761e-02 -8.48183751e-01 3.14215004e-01 1.76906303e-01 4.41430300e-01 3.88779938e-01 4.54259217e-01 1.26787543e+00 -1.10749793e+00 -5.76692343e-01 -1.51948975e-02 1.00765204e+00 6.18917882e-01 5.80848217e-01 5.28412700e-01 2.60955542e-01 7.33792543e-01 8.85649800e-01 6.04596794e-01 5.07695138e-01 3.79950911e-01 5.18213928e-01 2.61560947e-01 2.93708831e-01 -8.29410434e-01 3.41566354e-01 4.55757789e-02 -5.18902659e-01 -1.38993887e-02 -1.04852426e+00 4.95988011e-01 -2.23570156e+00 -1.52253056e+00 -5.67869246e-01 2.00773954e+00 3.69619042e-01 4.25375968e-01 -1.79172456e-01 2.99811989e-01 5.31680048e-01 -1.74868897e-01 -6.90622389e-01 -8.33769679e-01 2.32474953e-01 -2.61699408e-01 3.59369963e-01 6.09134674e-01 -5.72026014e-01 5.61117947e-01 6.47438812e+00 1.37417346e-01 -6.01993084e-01 8.55731815e-02 8.90673935e-01 -1.58400580e-01 -1.03004837e+00 5.59356570e-01 -5.32174766e-01 4.29959893e-01 1.21253860e+00 -8.19955766e-01 2.78379977e-01 1.31528151e+00 6.99353814e-01 -2.80424785e-02 -1.66435671e+00 5.75110435e-01 -3.69755983e-01 -1.56901276e+00 4.45979744e-01 4.12727714e-01 1.00041807e+00 -8.06887746e-01 -1.99069321e-01 4.15614367e-01 9.69526291e-01 -1.27218831e+00 9.05943334e-01 6.75034642e-01 -2.36217789e-02 -6.76674783e-01 8.13405693e-01 7.32717931e-01 -7.60181010e-01 -6.15255237e-01 -3.74328971e-01 -9.49409366e-01 7.08596334e-02 5.96676350e-01 -1.52789903e+00 3.27985376e-01 4.10393566e-01 5.48377872e-01 -2.44885504e-01 6.44471169e-01 -7.36275792e-01 8.83694530e-01 2.03705549e-01 -4.66590524e-02 -1.26605970e-03 -2.40683407e-02 4.74816531e-01 7.87483394e-01 4.81815666e-01 9.25338566e-02 -9.95891467e-02 1.41917360e+00 2.71269143e-01 -2.94124097e-01 -9.55173016e-01 7.22210482e-02 5.82966328e-01 6.62195027e-01 -3.39787722e-01 -5.97709537e-01 -2.79820263e-01 6.26419723e-01 -9.25117508e-02 3.49486202e-01 -7.42824256e-01 4.75594610e-01 8.49253833e-01 5.00410318e-01 -1.97265536e-01 3.07444721e-01 -1.09513235e+00 -1.04464364e+00 -1.56087682e-01 -1.14447105e+00 4.62452233e-01 -1.03561103e+00 -1.19243002e+00 2.11963430e-01 3.77924927e-02 -9.36698556e-01 -8.69868636e-01 -1.94389522e-01 -1.03026462e+00 8.46067667e-01 -1.26010764e+00 -1.03899813e+00 -2.20083922e-01 -1.49644818e-02 8.05987835e-01 1.58971865e-02 7.57399440e-01 -7.68884957e-01 -2.27139756e-01 -1.35520369e-01 -4.65617061e-01 -6.03202283e-01 3.78335446e-01 -1.52734041e+00 6.40571892e-01 8.70565116e-01 -2.50428282e-02 1.09262156e+00 1.12797987e+00 -9.01949167e-01 -1.07722962e+00 -1.23048127e+00 1.35241401e+00 -7.64508963e-01 4.85700339e-01 1.64849609e-01 -1.01595986e+00 1.27017748e+00 1.48847833e-01 -4.95767504e-01 8.92527342e-01 3.28309089e-01 -6.59477413e-02 9.97001082e-02 -1.29839671e+00 9.80272710e-01 1.15671551e+00 -2.10380610e-02 -1.18286896e+00 3.60077098e-02 9.41334307e-01 1.67845473e-01 -6.99072778e-01 2.26392463e-01 5.08648574e-01 -1.17779541e+00 6.42200351e-01 -9.42569256e-01 8.94577742e-01 -3.98941815e-01 2.80962680e-02 -1.37001741e+00 -5.15800834e-01 -7.93644071e-01 -6.63622439e-01 1.05552268e+00 6.75242364e-01 -1.01091647e+00 8.15401673e-01 1.66858053e+00 1.84607450e-02 -6.61207378e-01 -5.52886724e-01 -7.36903191e-01 -3.67802054e-01 -7.21671283e-01 1.48617995e+00 1.14617634e+00 4.28595304e-01 3.05205107e-01 -2.56411970e-01 1.82708085e-01 6.77933395e-01 6.57020867e-01 1.02520287e+00 -1.46860337e+00 -5.78526258e-01 -4.95280474e-01 -9.18380171e-02 -8.44323874e-01 1.01967059e-01 -5.61337411e-01 -1.06413119e-01 -1.76760530e+00 4.36796844e-01 -3.02542299e-01 -5.55363521e-02 7.02498734e-01 -5.94198346e-01 -4.65570360e-01 4.53562498e-01 5.56430876e-01 -2.17547566e-01 4.66213763e-01 1.27351880e+00 -1.14444710e-01 -4.59358603e-01 2.95402676e-01 -1.51316798e+00 8.85386229e-01 9.90805924e-01 -5.83466053e-01 -7.74992108e-01 2.28578359e-01 1.13264799e-01 2.33302876e-01 6.03434861e-01 -7.45549679e-01 1.69588123e-02 -9.06920552e-01 4.03162807e-01 -1.04637653e-01 -2.98391897e-02 -9.62150931e-01 9.24760520e-01 9.42313015e-01 -5.99944770e-01 7.80168027e-02 -4.10188399e-02 9.03116226e-01 -2.16787174e-01 -1.00918427e-01 2.39559278e-01 -2.75521129e-01 -5.77924073e-01 6.99161738e-02 -6.82960033e-01 -5.68997979e-01 1.28469217e+00 -5.37895918e-01 -4.66870248e-01 -7.59350002e-01 -5.46035826e-01 1.70386553e-01 5.06382227e-01 3.17707390e-01 5.82594156e-01 -1.15623665e+00 -4.63181198e-01 -1.48677798e-02 1.54937610e-01 -6.47131801e-02 -9.63968188e-02 6.13472879e-01 1.39396332e-04 6.82765365e-01 -6.03415929e-02 -2.71795541e-01 -9.14119840e-01 7.67660201e-01 2.64898777e-01 -6.43804193e-01 -5.23560703e-01 1.54108807e-01 5.77986479e-01 -4.34185147e-01 -4.42928374e-01 -6.26018584e-01 -1.24740019e-01 -6.06379867e-01 6.01124585e-01 3.75244379e-01 -4.21627015e-01 2.88267612e-01 -1.92400768e-01 -3.24416459e-01 1.36281952e-01 -1.24463670e-01 1.35246027e+00 -9.64059010e-02 1.06598109e-01 5.93833506e-01 -2.78292336e-02 -1.19458251e-01 -1.47894812e+00 1.96942136e-01 5.40814877e-01 -9.15012002e-01 -2.31369838e-01 -1.07210863e+00 -4.99424279e-01 7.38421619e-01 -1.06500097e-01 7.06843019e-01 8.89263690e-01 -1.17313415e-01 2.00152859e-01 4.39730734e-01 4.31037068e-01 -6.20461404e-01 -2.28960499e-01 -1.13220550e-02 1.08612978e+00 -1.31031418e+00 2.15480328e-01 -7.15884328e-01 -1.05933392e+00 8.51490200e-01 8.42512906e-01 1.70725971e-01 2.02531159e-01 -8.70314315e-02 -1.33715361e-01 -1.98972419e-01 -1.49691474e+00 3.91286194e-01 1.18120529e-01 6.62849605e-01 7.33511090e-01 6.25998974e-01 -2.39773259e-01 1.19613457e+00 -6.48039758e-01 3.74961227e-01 8.92642558e-01 6.47740245e-01 -3.59934181e-01 -8.44801486e-01 -5.95310211e-01 8.98953736e-01 -4.25121158e-01 1.22085987e-02 -5.64989388e-01 1.00057590e+00 -1.18569754e-01 1.30885971e+00 1.19307272e-01 -1.81653216e-01 4.00075704e-01 3.28554362e-01 -2.70088334e-02 -7.98925638e-01 -6.76743329e-01 -4.31922793e-01 3.89712393e-01 -6.64449692e-01 -3.42917234e-01 -9.37167287e-01 -1.22674513e+00 -7.85487056e-01 -1.49329618e-01 3.35579365e-01 3.26438844e-01 9.95372832e-01 5.17175317e-01 6.61126137e-01 3.17863256e-01 -3.60825419e-01 -7.70151794e-01 -9.41727817e-01 -2.55516201e-01 5.43942928e-01 7.93717150e-03 -6.14797294e-01 -3.71052951e-01 2.56460279e-01]
[8.716687202453613, 5.800018310546875]
fe7ed137-9a8d-4d20-98e8-8fcd95f8deaf
fast-hand-detection-in-collaborative-learning
2110.07070
null
https://arxiv.org/abs/2110.07070v1
https://arxiv.org/pdf/2110.07070v1.pdf
Fast Hand Detection in Collaborative Learning Environments
Long-term object detection requires the integration of frame-based results over several seconds. For non-deformable objects, long-term detection is often addressed using object detection followed by video tracking. Unfortunately, tracking is inapplicable to objects that undergo dramatic changes in appearance from frame to frame. As a related example, we study hand detection over long video recordings in collaborative learning environments. More specifically, we develop long-term hand detection methods that can deal with partial occlusions and dramatic changes in appearance. Our approach integrates object-detection, followed by time projections, clustering, and small region removal to provide effective hand detection over long videos. The hand detector achieved average precision (AP) of 72% at 0.5 intersection over union (IoU). The detection results were improved to 81% by using our optimized approach for data augmentation. The method runs at 4.7x the real-time with AP of 81% at 0.5 intersection over the union. Our method reduced the number of false-positive hand detections by 80% by improving IoU ratios from 0.2 to 0.5. The overall hand detection system runs at 4x real-time.
['Carlos LopezLeiva', 'Sylvia Celedon Pattichis', 'Marios S. Pattichis', 'Venkatesh Jatla', 'Sravani Teeparthi']
2021-10-13
null
null
null
null
['hand-detection']
['computer-vision']
[-3.58706564e-02 -3.95641804e-01 1.66903540e-01 2.25956410e-01 -8.47813547e-01 -7.71307290e-01 1.74690947e-01 2.49733310e-02 -6.73936963e-01 6.49438083e-01 -2.06492618e-01 -1.75320059e-02 1.14948750e-01 -2.79320598e-01 -8.12249780e-01 -6.41059756e-01 -1.83407709e-01 3.22757810e-01 9.37530935e-01 2.70652294e-01 6.39184564e-02 9.67905879e-01 -1.71155918e+00 2.33301297e-01 4.15833354e-01 6.22567654e-01 4.14466202e-01 1.29843831e+00 -2.46067401e-02 5.64507544e-01 -9.57287192e-01 1.36613110e-02 3.37877810e-01 2.81260222e-01 -5.70281208e-01 3.91603053e-01 1.18585873e+00 -8.17420840e-01 -2.44250819e-01 8.88474703e-01 8.41164589e-01 1.83781222e-01 2.27655962e-01 -1.36557424e+00 -1.14935026e-01 2.84080692e-02 -1.04508758e+00 3.92085612e-01 6.03629529e-01 4.42253053e-01 5.36814928e-01 -1.07055354e+00 7.81820774e-01 1.35704136e+00 7.06273854e-01 6.19990468e-01 -1.30667293e+00 -8.06948364e-01 2.98862338e-01 -5.72711602e-02 -1.53387928e+00 -3.83980155e-01 3.38096470e-01 -7.25878835e-01 8.74733388e-01 4.91927594e-01 7.60510504e-01 6.57865882e-01 -7.38202929e-02 8.92862797e-01 6.94528103e-01 -6.03011250e-01 -9.09239799e-02 -2.14739963e-02 2.88910002e-01 7.34022796e-01 3.44815314e-01 -1.36774004e-01 -2.57787228e-01 -5.89518882e-02 1.33495426e+00 2.34837994e-01 -2.21226230e-01 -3.69813181e-02 -1.12858903e+00 2.44399041e-01 1.88964903e-01 2.98316300e-01 -3.55106741e-01 3.79155666e-01 2.08347052e-01 -5.70731610e-02 2.65696079e-01 1.83998913e-01 -3.91848207e-01 -2.44473100e-01 -1.02537119e+00 2.96307117e-01 5.01006067e-01 1.12768745e+00 2.27294832e-01 3.69138382e-02 -4.90802467e-01 6.53750479e-01 7.83200487e-02 6.34467363e-01 -1.10687194e-02 -1.10160220e+00 3.10111701e-01 4.75667626e-01 4.02769536e-01 -7.23775148e-01 -4.75280493e-01 -4.66645628e-01 -3.67548347e-01 6.96208775e-01 9.91991878e-01 -2.93930769e-01 -1.06178570e+00 1.53361380e+00 7.01845407e-01 7.03661516e-02 -5.02389014e-01 7.43798316e-01 5.28038263e-01 4.59481180e-01 1.71643257e-01 -5.46433508e-01 1.52189171e+00 -9.21643376e-01 -9.00656939e-01 -1.19385840e-02 3.85211110e-01 -1.16674197e+00 1.08203351e+00 5.40008664e-01 -1.02668643e+00 -6.75196767e-01 -7.19461977e-01 2.79559851e-01 8.05765465e-02 2.29399383e-01 2.70345718e-01 6.88624740e-01 -1.08044028e+00 4.67388183e-01 -1.08776903e+00 -4.12526399e-01 4.78125006e-01 7.14481235e-01 -3.08343410e-01 1.53294802e-01 -3.66412878e-01 4.24381584e-01 1.67232230e-01 -2.31311843e-01 -5.48910320e-01 -9.00910318e-01 -3.56724888e-01 -1.74444750e-01 5.69710433e-01 -3.95710200e-01 1.20873165e+00 -8.39633584e-01 -1.18783987e+00 5.67988455e-01 -1.58473954e-01 -7.44508877e-02 9.65980649e-01 -5.60149193e-01 -4.45978343e-02 3.57497990e-01 -7.72898942e-02 5.41999936e-01 9.93954599e-01 -1.17782402e+00 -8.87320280e-01 -4.78592545e-01 7.74820372e-02 1.44663498e-01 -5.23445010e-01 4.41438466e-01 -1.00743926e+00 -8.67512822e-01 4.21899930e-02 -1.15385509e+00 -2.82584969e-03 6.54175758e-01 -5.60080595e-02 -2.62543261e-01 1.39079511e+00 -8.32637012e-01 1.17097712e+00 -2.16196561e+00 -2.87501335e-01 -4.45481241e-02 3.34708571e-01 5.49379885e-01 -7.31573701e-02 -1.36628523e-01 -5.53710898e-03 -2.23751128e-01 3.45920771e-01 -5.57069421e-01 -5.03047168e-01 -8.07007030e-02 8.09743628e-02 4.99446034e-01 -1.16603307e-01 6.72878206e-01 -8.22017133e-01 -8.19505095e-01 5.03899157e-01 7.94775188e-01 -4.38102931e-01 2.70117015e-01 -3.57218571e-02 1.20041378e-01 -9.99424793e-03 9.97616410e-01 6.81988060e-01 -2.15592265e-01 -5.43499626e-02 -2.80216604e-01 -3.48120034e-01 -5.46343029e-01 -1.79327929e+00 1.39586508e+00 1.22657628e-03 9.03635740e-01 3.47505540e-01 -5.94404377e-02 4.38165814e-01 4.75601703e-01 8.36050391e-01 -4.92767617e-02 6.22311607e-02 -6.32239282e-02 -1.20084636e-01 -4.37939733e-01 3.57105017e-01 2.45645240e-01 6.00961626e-01 4.33119208e-01 -2.51690149e-01 3.01961303e-01 3.41186970e-01 2.22172141e-01 1.14405727e+00 3.83721888e-02 1.43431082e-01 -2.44707428e-02 2.48894855e-01 -9.92784575e-02 2.90490001e-01 9.30179894e-01 -4.66207623e-01 8.24244261e-01 -9.58101898e-02 -3.89913082e-01 -8.35915685e-01 -9.45522428e-01 -1.83660239e-01 1.31318617e+00 -4.87758107e-02 -2.95883209e-01 -8.91541183e-01 -5.81803262e-01 1.22970887e-01 1.46702603e-02 -4.51839030e-01 3.62425268e-01 -8.55924189e-01 -5.15735745e-01 3.59374404e-01 1.10102022e+00 2.56346673e-01 -1.04464591e+00 -8.51638079e-01 3.73910785e-01 -1.16831185e-02 -1.15616930e+00 -8.85421932e-01 -1.45656437e-01 -8.91117394e-01 -1.20951581e+00 -1.21544492e+00 -6.22270286e-01 8.09573770e-01 5.68426013e-01 6.09735608e-01 2.41875425e-01 -9.93006110e-01 7.60200322e-01 -2.89529979e-01 -3.49652648e-01 -9.41585302e-02 -1.89102128e-01 3.57176721e-01 -2.90748537e-01 -5.92221022e-02 -1.75352514e-01 -7.13736951e-01 5.57707071e-01 -4.91212130e-01 -2.71751910e-01 2.68213749e-01 5.95396698e-01 6.02053106e-01 -2.52837360e-01 1.92476168e-01 -3.32333267e-01 7.04658329e-02 3.80208641e-01 -8.94255757e-01 4.28371102e-01 -3.57611775e-01 -1.39905915e-01 6.46940395e-02 -1.25939012e+00 -1.03974330e+00 6.59981966e-01 1.02009922e-01 -8.93078923e-01 -2.76208818e-02 -4.38189745e-01 9.79164708e-03 -3.47963721e-01 7.54224479e-01 -3.64183225e-02 9.21443477e-02 -4.16631401e-01 1.94248199e-01 6.53855622e-01 6.23973370e-01 -3.87517840e-01 5.18525481e-01 6.10870242e-01 -2.71722585e-01 -1.07165229e+00 -5.02261281e-01 -6.92006111e-01 -9.11956191e-01 -6.16809428e-01 7.84511030e-01 -8.83961260e-01 -8.94050837e-01 6.35089278e-01 -1.22059178e+00 -4.45299089e-01 -2.62937307e-01 5.71745038e-01 -2.19867811e-01 4.07818258e-01 -5.35555422e-01 -1.28697157e+00 -6.76270843e-01 -9.15169418e-01 1.18501830e+00 1.55583560e-01 -2.96624094e-01 -5.86922109e-01 -2.02398479e-01 2.10468844e-01 4.68149446e-02 1.70532927e-01 4.28633904e-03 -1.38259351e-01 -6.50620341e-01 -5.41269600e-01 -4.18548137e-01 7.17962906e-02 3.33751082e-01 3.80689353e-01 -1.07179010e+00 -7.16797709e-01 -5.20162702e-01 3.37485790e-01 5.37974834e-01 7.54331589e-01 1.02035367e+00 -1.32628288e-02 -6.18220508e-01 2.55186498e-01 1.19870508e+00 5.13096869e-01 3.40701103e-01 3.18272769e-01 9.25153255e-01 2.83996105e-01 7.84875274e-01 5.73110998e-01 -2.21795842e-01 1.01028013e+00 3.53157908e-01 -1.79694727e-01 -9.28001881e-01 3.87789935e-01 2.67160565e-01 2.52170805e-02 -5.19203484e-01 -3.19456607e-02 -9.03978169e-01 5.79699874e-01 -1.73941410e+00 -1.01129186e+00 -4.87035125e-01 2.26621032e+00 6.71122253e-01 1.01866066e-01 7.54056334e-01 3.21221501e-01 1.06236351e+00 -2.55765557e-01 -3.98209542e-01 4.39974874e-01 2.36232743e-01 -5.97753748e-02 4.54010606e-01 5.69573879e-01 -1.22174096e+00 9.38059211e-01 6.41298676e+00 5.90234816e-01 -9.76892948e-01 2.02267319e-01 -3.17349993e-02 -4.71265256e-01 6.56835318e-01 -4.07855421e-01 -1.18234658e+00 4.07486558e-01 2.27536634e-01 2.65297204e-01 2.38287568e-01 9.49469209e-01 1.76048636e-01 -2.63382733e-01 -8.62374663e-01 1.27526116e+00 1.37281910e-01 -9.12437677e-01 -2.88485050e-01 1.08924218e-01 6.60533547e-01 -2.91296721e-01 -1.70897633e-01 -3.76265533e-02 1.03278551e-02 -4.62119192e-01 7.84398675e-01 3.38557273e-01 1.03806126e+00 -5.51000416e-01 3.72272640e-01 1.86540782e-01 -1.70074928e+00 -1.31285682e-01 -1.12868343e-02 -3.93351242e-02 1.93634570e-01 4.01398778e-01 -1.08224320e+00 -1.89066499e-01 9.21023190e-01 2.85588890e-01 -5.69725275e-01 1.15636802e+00 5.96790090e-02 1.41695201e-01 -7.57047594e-01 9.47838649e-02 -3.64984512e-01 3.64182234e-01 8.27535212e-01 1.39879525e+00 9.66306999e-02 4.15701568e-01 4.46397692e-01 2.20174193e-01 1.04799807e-01 -9.53837484e-02 -2.22237557e-01 3.79544079e-01 5.69490433e-01 1.09026182e+00 -9.90749002e-01 -5.84376991e-01 -2.04674527e-01 1.18730748e+00 2.54202983e-03 1.88749120e-01 -8.20291340e-01 -4.92248088e-01 5.96653521e-01 5.53210199e-01 3.00841123e-01 -3.65988463e-01 -1.17176615e-01 -9.72354352e-01 3.32672924e-01 -5.44982374e-01 5.55188835e-01 -6.18509531e-01 -8.60932350e-01 4.29603726e-01 -6.28171712e-02 -1.15449405e+00 -7.59301111e-02 -5.39969385e-01 -4.04075056e-01 4.70786244e-01 -6.98599637e-01 -1.15048563e+00 -7.43781567e-01 8.54814947e-01 9.75965440e-01 1.57850549e-01 6.59072459e-01 7.37274766e-01 -7.04850078e-01 8.11332285e-01 1.63469091e-02 2.63032198e-01 8.45946848e-01 -1.29776013e+00 1.81558177e-01 9.34169352e-01 1.99496709e-02 5.87521374e-01 8.04822862e-01 -8.59209299e-01 -1.21240234e+00 -1.10590088e+00 2.96760559e-01 -6.92650557e-01 3.47362369e-01 -1.80941150e-01 -9.01307642e-01 7.28532732e-01 -2.01938733e-01 5.55038750e-01 2.50152886e-01 4.57061641e-02 -2.43446514e-01 -1.50363326e-01 -1.27693594e+00 4.20311838e-01 1.15131390e+00 -2.45356545e-01 -2.33804420e-01 5.49968779e-01 4.30581808e-01 -6.89683914e-01 -6.78333998e-01 2.24341169e-01 1.31325316e+00 -5.10081053e-01 1.23453307e+00 -2.82133311e-01 -4.06411916e-01 -5.43282866e-01 -5.98137826e-02 -4.48686898e-01 -3.24880183e-01 -6.73502803e-01 -5.58111608e-01 1.18637609e+00 -8.44708383e-02 -9.79606286e-02 9.87980545e-01 7.07124472e-01 3.09451491e-01 -3.28652352e-01 -9.13953185e-01 -1.08748341e+00 -4.21771705e-01 -4.60175127e-01 -1.34210855e-01 5.52991271e-01 -1.68027610e-01 -1.19081296e-01 -2.91105837e-01 4.00292248e-01 9.24289584e-01 3.73587161e-02 1.02485180e+00 -1.23499787e+00 -1.98303983e-01 -2.27385446e-01 -4.45206970e-01 -9.51607525e-01 -3.75119567e-01 -2.23201945e-01 7.81803131e-02 -1.50899792e+00 4.73698199e-01 -2.61535048e-01 -8.01468566e-02 7.31260598e-01 -3.65638256e-01 6.62249744e-01 6.40516996e-01 2.54418701e-01 -5.07339776e-01 -2.66936541e-01 1.17418945e+00 -2.68551223e-02 -7.10262358e-01 1.56731516e-01 -3.81823666e-02 8.21855187e-01 7.01914370e-01 -4.95233566e-01 4.28388268e-02 -3.64107281e-01 -3.40195715e-01 6.07137606e-02 4.83980387e-01 -1.22919631e+00 3.77838135e-01 -8.18227455e-02 6.16083086e-01 -9.74153161e-01 5.43597639e-01 -9.66106713e-01 2.07264975e-01 8.12550664e-01 1.06632389e-01 -2.27362178e-02 5.80458999e-01 4.69772071e-01 3.39712590e-01 -1.83264278e-02 9.58355665e-01 -7.40153715e-02 -6.18501782e-01 1.30464539e-01 -2.70796299e-01 -3.71619463e-01 1.44822264e+00 -5.24540484e-01 -1.30866826e-01 -1.63809195e-01 -1.28004169e+00 1.35186091e-01 2.12931708e-01 3.55904162e-01 5.67408085e-01 -1.06068587e+00 -5.58234930e-01 2.27959290e-01 -2.89806634e-01 -3.93208787e-02 2.49602124e-01 8.45357180e-01 -5.19352436e-01 1.07182078e-01 -1.16125174e-01 -1.02375340e+00 -2.52105689e+00 4.91467744e-01 2.45165452e-01 1.43676862e-01 -8.47044230e-01 1.07702529e+00 3.39951552e-02 2.88919359e-01 7.91185379e-01 -2.94548839e-01 3.16882767e-02 3.47494483e-01 1.05778873e+00 9.40802395e-01 -3.28326896e-02 -3.74862164e-01 -5.48883915e-01 8.97995949e-01 -3.10436189e-01 -1.77059636e-01 1.08879364e+00 -3.23911086e-02 3.09523016e-01 1.61779076e-01 9.42688346e-01 2.12222233e-01 -1.74484968e+00 -3.83357480e-02 -4.07468766e-01 -8.28626931e-01 6.27983212e-02 -8.17431331e-01 -1.11276543e+00 6.82494700e-01 1.32040918e+00 -5.08980937e-02 9.81710374e-01 1.46435589e-01 4.67743665e-01 3.97513419e-01 3.35038245e-01 -9.49790180e-01 3.77109647e-01 2.46148542e-01 9.24453139e-01 -1.16078019e+00 3.51979852e-01 -6.34413779e-01 -2.02924669e-01 1.22228158e+00 7.63128936e-01 6.46293387e-02 4.03486818e-01 6.91663563e-01 2.67959312e-02 -9.12287012e-02 -2.17787072e-01 -2.85840213e-01 3.70169789e-01 5.32353461e-01 1.97229296e-01 8.55365545e-02 -9.26329494e-02 6.89786347e-03 2.93243527e-01 1.17486432e-01 3.59295517e-01 1.29046750e+00 -6.42651021e-01 -7.71342039e-01 -1.07477236e+00 3.61603260e-01 -4.94359493e-01 3.08353513e-01 -9.55784246e-02 8.98745775e-01 7.94747621e-02 9.62598860e-01 2.71191776e-01 -1.58771977e-01 5.30212820e-01 -2.13833265e-02 8.00452173e-01 -6.60486221e-01 -5.01335740e-01 7.49645531e-01 -3.39076877e-01 -5.50647616e-01 -4.44862932e-01 -9.93073225e-01 -1.41852665e+00 -2.44383946e-01 -9.01252925e-01 -4.61244285e-01 5.55306256e-01 5.89006245e-01 1.86663508e-01 7.22366333e-01 1.05342753e-01 -1.08222973e+00 -2.98277289e-01 -9.03884828e-01 -5.67614913e-01 1.22988984e-01 6.27900183e-01 -8.46474767e-01 -2.26603240e-01 4.39967901e-01]
[6.711970329284668, -0.8019370436668396]
b80ff0a5-3fd2-4fcf-a463-f02bf77a4c9a
incorporating-the-rhetoric-of-scientific
null
null
https://aclanthology.org/2022.sdp-1.7
https://aclanthology.org/2022.sdp-1.7.pdf
Incorporating the Rhetoric of Scientific Language into Sentence Embeddings using Phrase-guided Distant Supervision and Metric Learning
Communicative functions are an important rhetorical feature of scientific writing. Sentence embeddings that contain such features are highly valuable for the argumentative analysis of scientific documents, with applications in document alignment, recommendation, and academic writing assistance. Moreover, embeddings can provide a possible solution to the open-set problem, where models need to generalize to new communicative functions unseen at training time. However, existing sentence representation models are not suited for detecting functional similarity since they only consider lexical or semantic similarities. To remedy this, we propose a combined approach of distant supervision and metric learning to make a representation model more aware of the functional part of a sentence. We first leverage an existing academic phrase database to label sentences automatically with their functions. Then, we train an embedding model to capture similarities and dissimilarities from a rhetorical perspective. The experimental results demonstrate that the embeddings obtained from our model are more advantageous than existing models when retrieving functionally similar sentences. We also provide an extensive analysis of the performance differences between five metric learning objectives, revealing that traditional methods (e.g., softmax cross-entropy loss and triplet loss) outperform state-of-the-art techniques.
['Akiko Aizawa', 'Kaito Sugimoto']
null
null
null
null
sdp-coling-2022-10
['sentence-embeddings', 'sentence-embeddings']
['methodology', 'natural-language-processing']
[ 7.83272013e-02 7.27319270e-02 -3.84925008e-01 -4.33920622e-01 -5.74041843e-01 -5.52378595e-01 6.65060163e-01 7.40918517e-01 -5.06105363e-01 6.47949755e-01 5.69413364e-01 -3.96716505e-01 -3.28231424e-01 -6.92175508e-01 -4.46818560e-01 -4.89128798e-01 3.93246889e-01 3.78285021e-01 -2.59046882e-01 -5.01320422e-01 8.07717979e-01 2.39848793e-01 -1.42612505e+00 2.52629191e-01 1.24430501e+00 7.44048297e-01 3.75967175e-01 2.61769921e-01 -6.43105268e-01 7.21358895e-01 -8.49440813e-01 -5.51139116e-01 -1.51465863e-01 -3.49454254e-01 -1.07317841e+00 -3.50433707e-01 3.33346426e-01 1.13572590e-01 -1.62761346e-01 1.01508701e+00 6.92680776e-01 2.21807301e-01 8.01848054e-01 -8.06062400e-01 -1.28685391e+00 6.71346307e-01 -3.59772652e-01 4.55316216e-01 5.56341112e-01 -3.84392202e-01 1.57802045e+00 -1.04803991e+00 6.57861352e-01 1.31562018e+00 5.22674203e-01 5.98348618e-01 -9.24250364e-01 -3.92298251e-01 2.24364758e-01 6.56448483e-01 -9.80655789e-01 -2.80754924e-01 1.10410106e+00 -4.54184860e-01 8.29901993e-01 4.70692217e-01 3.58659565e-01 1.25335348e+00 2.04296354e-02 7.64529586e-01 7.51713216e-01 -5.85223317e-01 4.69865687e-02 2.60190278e-01 4.70477670e-01 5.74464619e-01 1.53487116e-01 -4.28834140e-01 -5.83623528e-01 -2.52602160e-01 1.42707378e-01 1.66689038e-01 -6.11839831e-01 -1.71743631e-01 -1.29136372e+00 1.04542959e+00 2.84913599e-01 6.08936012e-01 7.33586634e-03 -1.66585878e-01 6.11330509e-01 4.63386178e-01 4.86720324e-01 1.14239240e+00 -3.77240926e-01 -1.96588591e-01 -6.11390352e-01 1.62621439e-01 7.32526839e-01 6.70976758e-01 5.17432749e-01 -4.27401751e-01 -4.96207774e-01 1.12321651e+00 2.24071026e-01 7.30998069e-02 8.71306121e-01 -8.24426055e-01 4.49789107e-01 7.06368625e-01 -1.41787350e-01 -1.31969965e+00 -1.87185049e-01 -4.62267578e-01 -5.34937024e-01 -5.24400413e-01 9.77610052e-02 3.27845782e-01 1.03211508e-03 1.63095009e+00 2.00268283e-01 3.24212611e-02 2.34668717e-01 7.26904392e-01 1.22560430e+00 4.01217967e-01 -1.80463359e-01 -1.91079080e-01 1.43036449e+00 -1.07673573e+00 -9.56836641e-01 -3.70983072e-02 9.65781212e-01 -8.85083795e-01 1.32806003e+00 -3.34612243e-02 -7.41296947e-01 -5.01211166e-01 -1.09218621e+00 -4.07143921e-01 -5.00194728e-01 7.53344074e-02 5.17598629e-01 3.72155160e-01 -6.09918773e-01 8.83357465e-01 -3.66050303e-01 -1.09924637e-01 3.57480347e-01 -4.73887958e-02 -2.18580052e-01 8.82576928e-02 -1.37163198e+00 1.08231962e+00 3.08099955e-01 -2.87742801e-02 -1.81221485e-01 -8.24166954e-01 -9.86195862e-01 3.88103515e-01 2.31987342e-01 -6.17967725e-01 9.62962627e-01 -4.88852233e-01 -1.29857934e+00 8.74910176e-01 -2.46795252e-01 -3.42722774e-01 5.18005230e-02 -1.24570869e-01 -2.79829800e-01 1.67772815e-01 3.33346844e-01 2.77375460e-01 5.07248640e-01 -9.58834350e-01 -1.06217735e-01 -3.25096697e-01 2.50445366e-01 2.04167783e-01 -1.21673000e+00 1.12904586e-01 -1.00931339e-02 -8.47807586e-01 -2.13743240e-01 -4.69072640e-01 6.44212142e-02 2.25686640e-01 -3.93181741e-01 -8.61087441e-01 7.05550432e-01 -6.36355579e-01 1.50989962e+00 -2.08390164e+00 4.23539311e-01 -1.98311374e-01 3.55303466e-01 2.17698231e-01 -2.70177335e-01 5.40722430e-01 -6.97974116e-02 2.95726359e-01 -1.51021361e-01 -3.37810576e-01 2.32203100e-02 1.86650351e-01 -3.50444049e-01 1.94099054e-01 1.53791323e-01 7.50588536e-01 -1.19353986e+00 -7.45242655e-01 -7.06723556e-02 1.92097381e-01 -3.22064281e-01 1.67272896e-01 -2.64866620e-01 3.67360055e-01 -6.71030641e-01 5.98255515e-01 2.21852988e-01 -4.34718996e-01 3.07637066e-01 -1.37657434e-01 5.16093001e-02 6.51984692e-01 -5.00252187e-01 1.91242766e+00 -7.27180898e-01 7.59966016e-01 -4.97121662e-01 -1.48245406e+00 1.13429070e+00 2.28609487e-01 4.49768633e-01 -6.94159985e-01 6.88535422e-02 2.91399062e-01 1.38800442e-01 -7.87263155e-01 6.30427897e-01 2.58884430e-02 1.99970286e-02 6.11351967e-01 3.61507088e-02 1.63525641e-01 3.27286482e-01 2.15299770e-01 1.16107655e+00 -7.02536255e-02 3.20418775e-01 -2.72427708e-01 8.52742910e-01 -2.81597376e-01 6.06006503e-01 4.08788830e-01 -6.25475645e-02 4.78165388e-01 6.11536801e-01 -4.41773593e-01 -8.68199348e-01 -6.33596063e-01 -4.83087838e-01 9.85142589e-01 1.77569434e-01 -9.10753250e-01 -4.81215537e-01 -8.50940645e-01 2.01622292e-01 8.61967266e-01 -6.92853510e-01 -4.27177072e-01 -5.22765994e-01 -4.46556985e-01 3.91967058e-01 5.55660367e-01 1.13676608e-01 -8.95711601e-01 -3.63660842e-01 7.02945217e-02 -4.39889938e-01 -9.96185541e-01 -5.13350368e-01 -1.74727570e-02 -7.17508733e-01 -1.27384484e+00 -5.66991806e-01 -9.90454614e-01 6.51086509e-01 4.03760970e-01 1.03791404e+00 2.46704429e-01 -1.89020529e-01 2.84174800e-01 -6.44781768e-01 -3.84878010e-01 -3.76956195e-01 -1.68233924e-02 9.19982940e-02 -1.98708504e-01 5.92581570e-01 -5.33230543e-01 -5.12684524e-01 1.16254337e-01 -7.28249073e-01 -2.38892838e-01 4.02370304e-01 1.30148554e+00 5.02270401e-01 -4.63331223e-01 7.89033592e-01 -9.76443887e-01 1.23786724e+00 -6.73622072e-01 -2.47494951e-02 6.87104404e-01 -8.16836417e-01 3.88767719e-01 1.00576317e+00 -4.06785041e-01 -8.85664761e-01 -7.31540799e-01 -1.10709682e-01 -5.10828137e-01 4.01193738e-01 7.84895837e-01 -4.98365760e-02 1.05644807e-01 6.74543202e-01 3.13677758e-01 1.47569776e-01 -6.96919739e-01 3.39546621e-01 1.00639999e+00 2.21678123e-01 -7.54504204e-01 4.48404819e-01 2.13179737e-01 -5.41666113e-02 -7.65213728e-01 -1.23360026e+00 -6.11583352e-01 -3.13858032e-01 1.12764202e-01 6.87130988e-01 -4.93006647e-01 -7.78742433e-01 -2.92630702e-01 -1.54470682e+00 5.63322246e-01 -3.77882451e-01 4.33775663e-01 -3.96446526e-01 9.04981434e-01 -4.17123288e-01 -5.25908291e-01 -5.07300377e-01 -9.19431090e-01 1.07413912e+00 1.47411332e-01 -4.52605337e-01 -1.25737202e+00 2.55132169e-01 6.83127344e-01 2.26192355e-01 1.08121924e-01 1.29490995e+00 -1.10004473e+00 9.82713103e-02 -7.10824803e-02 -7.69391209e-02 3.97378236e-01 3.43860269e-01 -2.28973553e-01 -8.96662533e-01 -9.74002406e-02 2.65380681e-01 -3.81449610e-01 9.00149584e-01 -1.75758794e-01 1.67218173e+00 -5.04424036e-01 -2.47094706e-01 4.52756286e-01 9.26639557e-01 -7.29497448e-02 3.71393263e-01 7.21738100e-01 5.60269833e-01 7.81669497e-01 6.92378342e-01 4.36884820e-01 2.83503026e-01 7.19859958e-01 2.28928074e-01 3.34904701e-01 -7.65998811e-02 -3.01620424e-01 2.27500975e-01 1.11670268e+00 8.67535993e-02 -1.66618615e-01 -7.49533594e-01 4.99339789e-01 -1.88730431e+00 -1.20464361e+00 8.61481801e-02 2.08448744e+00 1.20371187e+00 8.04628246e-03 -3.48413795e-01 1.35641545e-01 6.35257840e-01 2.97657043e-01 -4.50387210e-01 -7.10504830e-01 -2.48986319e-01 3.03041160e-01 -5.42080589e-02 2.38413960e-01 -8.32152307e-01 7.18889117e-01 5.55728626e+00 8.64068747e-01 -7.77132154e-01 1.51656866e-01 3.53288680e-01 1.72293410e-02 -9.45693374e-01 4.43962552e-02 -5.97426534e-01 7.70291567e-01 7.14625299e-01 -3.89555693e-01 3.55282649e-02 8.45532417e-01 1.70841232e-01 4.83854741e-01 -1.37189913e+00 9.98599887e-01 4.73720521e-01 -1.67068672e+00 2.05517158e-01 -6.13407083e-02 4.71362352e-01 -4.41739768e-01 -9.32691097e-02 3.92530173e-01 -1.73076168e-01 -1.02906203e+00 2.77050436e-01 6.33565664e-01 5.12033761e-01 -4.32554334e-01 8.45845342e-01 3.29352379e-01 -6.96522474e-01 -1.40936067e-02 -6.47844613e-01 -1.75614893e-01 -5.33920303e-02 6.43328905e-01 -8.03691626e-01 6.88080609e-01 4.36476678e-01 1.08531725e+00 -6.18998766e-01 7.54271686e-01 -4.69473958e-01 3.65106225e-01 1.36565655e-01 -6.51514649e-01 -5.58241410e-03 -3.13661695e-01 5.83817363e-01 9.92472887e-01 2.32144773e-01 -3.60877901e-01 -1.05852589e-01 9.58555937e-01 -4.01990652e-01 4.72517699e-01 -8.31986964e-01 -3.80328000e-01 6.24640882e-01 1.17097402e+00 -1.61764979e-01 -1.74931377e-01 -5.58482051e-01 1.04618323e+00 7.31678903e-01 2.09950134e-02 -6.26334965e-01 -6.76124156e-01 8.08003366e-01 -3.01077992e-01 2.42421567e-01 -1.34436309e-01 -2.59781420e-01 -1.41348124e+00 4.11683291e-01 -7.18878448e-01 2.99919486e-01 -4.76476312e-01 -1.58836114e+00 3.02423716e-01 -3.58889192e-01 -1.40798187e+00 -2.59661794e-01 -7.90655911e-01 -8.40489268e-01 7.12059975e-01 -1.74005175e+00 -9.18676913e-01 6.89999908e-02 1.46367192e-01 7.99839139e-01 -3.74800652e-01 1.22270215e+00 2.01442882e-01 -4.88722950e-01 8.07157397e-01 5.21946192e-01 2.01382756e-01 8.64230990e-01 -1.24641061e+00 5.18759200e-03 4.89988327e-01 4.48899776e-01 1.09066069e+00 5.91903865e-01 -3.22525561e-01 -1.49286473e+00 -7.74224579e-01 1.51301110e+00 -5.45596123e-01 8.15209150e-01 -2.28265896e-01 -1.04054856e+00 2.47909486e-01 3.13605368e-01 -1.75300866e-01 1.33354974e+00 5.75079083e-01 -5.41109085e-01 7.97236860e-02 -7.76221991e-01 6.59347296e-01 1.14005136e+00 -6.57472610e-01 -1.09269941e+00 6.82535291e-01 9.46200132e-01 -1.18833236e-01 -1.09001195e+00 1.63466528e-01 4.22320843e-01 -5.71613789e-01 9.57871914e-01 -1.16281390e+00 1.00618732e+00 -5.37290275e-02 -6.71142265e-02 -1.25616693e+00 -2.67167062e-01 -4.78003353e-01 -4.96472299e-01 1.30841708e+00 4.46477890e-01 -4.98815536e-01 4.26335186e-01 2.59454876e-01 -4.38786894e-01 -1.20992339e+00 -9.51158822e-01 -9.57657933e-01 3.11580509e-01 -6.55182032e-03 4.81340349e-01 1.46567833e+00 6.01953387e-01 8.11057866e-01 -9.17299911e-02 -3.79605353e-01 1.85999289e-01 5.75121462e-01 4.43643004e-01 -1.46336818e+00 -3.84652674e-01 -7.60219276e-01 -6.83734536e-01 -1.11669528e+00 8.73714745e-01 -1.35547173e+00 -4.48591053e-01 -1.61012924e+00 4.43360686e-01 -2.38939092e-01 -4.87389445e-01 3.70538294e-01 -5.05485773e-01 -3.36357743e-01 2.47098841e-02 3.88121575e-01 -6.04779899e-01 1.15208638e+00 1.34477782e+00 -5.15258610e-01 2.61215270e-01 -1.76304117e-01 -1.16702712e+00 5.85305154e-01 7.09502339e-01 -3.09913665e-01 -3.17435682e-01 -7.26891994e-01 3.63118619e-01 -2.40794256e-01 8.94892290e-02 -5.42162955e-01 3.38791192e-01 -1.08450286e-01 1.10886432e-01 -1.17078096e-01 2.36232087e-01 -5.83116651e-01 -6.87336206e-01 3.13758522e-01 -7.12760091e-01 1.68550745e-01 -1.74416348e-01 7.38041759e-01 -4.39329505e-01 -7.41359413e-01 2.01295108e-01 9.62224882e-03 -4.52334106e-01 -8.81889835e-02 2.41520870e-02 2.64150351e-01 8.28110576e-01 -1.05219029e-01 -5.39878905e-01 -1.46106198e-01 -2.81853437e-01 4.06427681e-01 3.22672725e-01 7.84114242e-01 8.82461548e-01 -1.39271045e+00 -7.34308124e-01 -9.02658924e-02 5.45626283e-01 -1.31363034e-01 -1.13490611e-01 7.93072522e-01 -2.47352913e-01 4.98691022e-01 -6.53812382e-03 -3.18173438e-01 -1.29188383e+00 6.09867573e-01 -6.01602197e-02 -2.97208637e-01 -4.77691561e-01 8.58771503e-01 3.75557574e-03 -6.76587880e-01 2.18110368e-01 -1.01726994e-01 -6.19260430e-01 1.65507138e-01 4.38649267e-01 2.22762898e-01 1.09731451e-01 -4.16364074e-01 -3.51817220e-01 5.08776844e-01 -3.45194489e-01 3.00258666e-01 1.44123292e+00 1.68467298e-01 -3.71168524e-01 5.66146731e-01 1.47386003e+00 2.61996746e-01 -4.17709142e-01 -5.50957620e-01 2.20953584e-01 -6.13229930e-01 -5.11291921e-02 -4.48888928e-01 -5.75174034e-01 1.02978408e+00 1.14606388e-01 1.40192375e-01 6.87922537e-01 1.23894930e-01 9.04619157e-01 8.16069543e-01 -6.95856214e-02 -1.27957940e+00 4.08612132e-01 5.85313797e-01 9.42817509e-01 -1.26207995e+00 -6.84330892e-03 -3.11657101e-01 -4.96103555e-01 1.53176057e+00 6.23183429e-01 2.08302274e-01 4.08677191e-01 -2.60422051e-01 -1.93972334e-01 -3.13163131e-01 -6.59004569e-01 7.46403560e-02 3.82224441e-01 3.03925246e-01 9.73473370e-01 -2.28105560e-01 -1.12371778e+00 7.03172266e-01 -4.12276626e-01 -4.30995315e-01 3.82889390e-01 9.02029335e-01 -4.09445584e-01 -1.40320241e+00 7.63320923e-02 5.88456333e-01 -2.70003915e-01 -3.25580418e-01 -7.15532184e-01 2.51445442e-01 -1.74786747e-01 8.89249206e-01 -8.66781622e-02 -3.44371885e-01 2.09733337e-01 1.81339145e-01 4.09555852e-01 -8.66080344e-01 -4.38398838e-01 -5.39168239e-01 1.27598122e-01 -1.47983804e-01 -4.76882815e-01 -5.90185106e-01 -1.12296879e+00 -1.82789087e-01 -4.85781640e-01 5.37949026e-01 6.93453729e-01 1.07113838e+00 5.01075923e-01 6.40862405e-01 9.33573186e-01 -2.55119622e-01 -1.18087542e+00 -1.13234162e+00 -3.05558175e-01 7.38549173e-01 6.42021671e-02 -7.88300753e-01 -3.37764412e-01 -2.11700767e-01]
[11.01823616027832, 8.688685417175293]
bd9655b8-2221-413d-9563-f32aac645869
eica-at-semeval-2017-task-4-a-simple
null
null
https://aclanthology.org/S17-2124
https://aclanthology.org/S17-2124.pdf
EICA at SemEval-2017 Task 4: A Simple Convolutional Neural Network for Topic-based Sentiment Classification
This paper describes our approach for SemEval-2017 Task 4 - Sentiment Analysis in Twitter (SAT). Its five subtasks are divided into two categories: (1) sentiment classification, i.e., predicting topic-based tweet sentiment polarity, and (2) sentiment quantification, that is, estimating the sentiment distributions of a set of given tweets. We build a convolutional sentence classification system for the task of SAT. Official results show that the experimental results of our system are comparative.
['Yufei Xie', 'Shiyun Chen', 'Maoquan Wang', 'Lu Zhao']
2017-08-01
null
null
null
semeval-2017-8
['twitter-sentiment-analysis']
['natural-language-processing']
[-1.28870383e-01 -1.32760525e-01 -3.40650916e-01 -1.10083926e+00 -7.18573570e-01 -4.60919559e-01 6.46720529e-01 2.50220865e-01 -4.07731771e-01 5.05738497e-01 6.35467291e-01 -3.31034303e-01 7.64062583e-01 -6.93219602e-01 -4.66342807e-01 -4.18535829e-01 1.58647537e-01 3.67804676e-01 -3.81874926e-02 -6.51666641e-01 3.77776265e-01 -2.35531688e-01 -1.28282166e+00 7.18703330e-01 4.71828163e-01 1.70830548e+00 -5.74774802e-01 7.49913037e-01 -3.64335924e-01 1.36899710e+00 -8.72551799e-01 -7.90430069e-01 -4.35607910e-01 -3.87900732e-02 -1.08177805e+00 -8.19572434e-02 2.00673774e-01 4.40171361e-02 2.61150450e-01 1.08771694e+00 1.59504429e-01 -1.10378057e-01 6.81358337e-01 -1.18267763e+00 -5.26337504e-01 1.01467037e+00 -5.89962542e-01 3.05882514e-01 2.80333430e-01 -3.74690741e-01 1.33826685e+00 -1.03051269e+00 3.41642022e-01 1.03342605e+00 6.91546321e-01 3.79110128e-01 -5.82694530e-01 -6.89333975e-01 3.89946163e-01 -1.05649740e-01 -9.52432096e-01 -3.89504015e-01 8.23904872e-01 -7.86930859e-01 9.44519818e-01 2.04204321e-01 6.32178307e-01 1.13274825e+00 5.90854406e-01 1.15110564e+00 1.36117828e+00 1.54444009e-01 4.23302680e-01 5.74053407e-01 9.01579976e-01 5.38412370e-02 -4.79178429e-02 -8.51051569e-01 -8.10028851e-01 -5.60749881e-03 -5.17574430e-01 -1.91041470e-01 1.72979832e-01 4.53190416e-01 -8.69422853e-01 1.18618846e+00 3.93735826e-01 1.18190259e-01 -4.80614424e-01 -5.91365620e-02 1.03811681e+00 5.28794467e-01 1.70079112e+00 4.83656406e-01 -1.01045036e+00 -1.98741667e-02 -9.10866082e-01 4.60480005e-01 1.10012543e+00 7.96503782e-01 5.51344812e-01 5.26227467e-02 -5.08485921e-02 6.93339646e-01 4.24050391e-01 8.72984469e-01 6.21744871e-01 -9.89328623e-02 7.25695908e-01 7.23485053e-01 7.05664605e-02 -1.08939695e+00 -5.75127542e-01 -1.54433727e-01 -6.30751848e-01 -3.57093543e-01 -1.75368831e-01 -9.12379026e-01 -7.89407194e-01 1.37317216e+00 6.45621493e-02 -2.46949583e-01 2.28493139e-01 5.89670360e-01 1.62840462e+00 8.78070474e-01 4.34952199e-01 -2.78411627e-01 1.41583216e+00 -1.06509662e+00 -8.56040418e-01 -5.37900805e-01 7.47602999e-01 -7.77625859e-01 1.05993724e+00 2.10431889e-01 -9.99731362e-01 -6.99662194e-02 -8.60279500e-01 1.74396764e-02 -1.05228007e+00 3.54933143e-02 8.33102703e-01 6.66327477e-01 -1.08902013e+00 2.69106738e-02 -5.02876103e-01 -6.39395863e-02 6.84498549e-01 3.55762601e-01 -7.60681629e-02 5.77168703e-01 -1.59399748e+00 5.88910937e-01 -2.47358531e-01 -2.65890919e-02 -3.79320264e-01 -6.74206138e-01 -1.21474671e+00 1.95711199e-02 -3.14621508e-01 -1.64006844e-01 1.70848727e+00 -1.39391530e+00 -1.55198216e+00 1.33023977e+00 -6.20285571e-01 -4.93666768e-01 5.00004925e-02 -2.86698371e-01 -4.99427646e-01 -4.31842715e-01 4.61965859e-01 1.34332225e-01 4.75176513e-01 -9.15575206e-01 -7.80910552e-01 -4.16456997e-01 6.30678535e-02 2.81077147e-01 -8.41012120e-01 6.14027441e-01 -2.90533751e-01 -1.78974375e-01 -2.75180519e-01 -8.12650561e-01 -3.70468885e-01 -1.02371156e+00 -8.16531897e-01 -7.46604383e-01 7.81590104e-01 -3.24414313e-01 1.24502468e+00 -2.05668545e+00 -3.89085233e-01 1.93307683e-01 8.60213265e-02 -1.05796708e-02 8.87374021e-03 2.72893399e-01 -3.71881574e-01 2.58668274e-01 2.22405896e-01 -8.85225892e-01 -5.33607490e-02 -6.46957040e-01 -1.04157329e+00 3.25320929e-01 2.14035898e-01 1.03078949e+00 -8.68569374e-01 -2.43764862e-01 -8.54335800e-02 2.06965283e-01 -2.89875478e-01 5.99402636e-02 -3.38391662e-01 2.16504991e-01 -8.61715317e-01 6.17214859e-01 6.59889221e-01 -3.46530765e-01 9.92890541e-03 -3.65694761e-02 -3.76016855e-01 1.17917240e+00 -3.99053037e-01 9.36393440e-01 -5.95304549e-01 9.37970936e-01 -2.32597858e-01 -6.75588906e-01 9.01506007e-01 1.99034065e-01 4.83513474e-01 -7.08303750e-01 6.94319308e-01 1.00087777e-01 -5.26780248e-01 -1.44150436e-01 9.22638416e-01 -9.95768458e-02 -9.23818231e-01 6.91674888e-01 -1.07884891e-01 -4.38709378e-01 3.34476799e-01 3.03147972e-01 7.34770954e-01 -3.29017192e-01 1.96410298e-01 -6.83685541e-01 6.66441679e-01 2.50333101e-01 8.22854936e-02 2.33295321e-01 -1.47611961e-01 3.30682665e-01 1.02534938e+00 -7.42629468e-01 -4.67242718e-01 -6.12490296e-01 -2.25843444e-01 1.46465564e+00 1.12076163e-01 -7.34161437e-01 -6.85713112e-01 -9.77000237e-01 -1.88077867e-01 5.36327124e-01 -1.20356405e+00 3.42086822e-01 -2.86353946e-01 -1.24458778e+00 -8.61830171e-03 6.15598679e-01 3.27012062e-01 -1.33443308e+00 -2.34377742e-01 -2.04129398e-01 -4.48840529e-01 -1.21169329e+00 -4.63372797e-01 3.85346681e-01 -4.73674655e-01 -8.12092900e-01 -2.05013767e-01 -9.53063369e-01 5.52975476e-01 3.22896868e-01 1.54904974e+00 -2.89495677e-01 6.22286558e-01 1.01780728e-01 -4.30198759e-01 -1.26796031e+00 9.04716998e-02 5.35985470e-01 3.12691042e-03 9.29987282e-02 1.00038886e+00 -1.07632555e-01 -5.16953409e-01 -4.76172827e-02 -5.76319754e-01 -2.73446649e-01 6.99910149e-02 4.73264396e-01 5.73912144e-01 -1.93505809e-01 8.20744038e-01 -1.54648006e+00 1.06103837e+00 -8.29110980e-01 -2.14312986e-01 -1.60789534e-01 -4.40296471e-01 -7.61591554e-01 8.79332602e-01 3.69219929e-02 -9.77352500e-01 -1.18217565e-01 -6.28325760e-01 4.97879416e-01 1.45965353e-01 1.03815925e+00 1.69070363e-01 4.24015433e-01 4.57530171e-01 1.97909400e-01 -3.94144654e-01 9.55368802e-02 5.57299219e-02 1.30732012e+00 -4.88503128e-02 -1.51662212e-02 2.30137348e-01 5.96007347e-01 -3.99300992e-01 -7.54499078e-01 -2.00648355e+00 -7.95744181e-01 -2.66876936e-01 -4.11328912e-01 1.00516629e+00 -1.50821018e+00 -8.03292513e-01 8.48313034e-01 -1.01696205e+00 -3.41568232e-01 -2.20072493e-01 8.09116140e-02 -2.97477245e-01 -3.33860427e-01 -8.78202081e-01 -7.10979581e-01 -1.04365385e+00 -1.12729967e+00 1.32200515e+00 3.26193482e-01 -7.11690247e-01 -1.28700638e+00 3.85920823e-01 5.20937741e-01 4.38551575e-01 1.67558685e-01 2.42428198e-01 -1.15098190e+00 3.64999384e-01 -7.62504041e-01 -1.37082174e-01 5.02343178e-01 -1.54066399e-01 2.51911074e-01 -1.31953037e+00 8.65899846e-02 5.31940274e-02 -9.77944553e-01 9.34031785e-01 5.43851554e-01 1.28952968e+00 -1.28687799e-01 -1.99916646e-01 3.86494488e-01 9.68497694e-01 -2.79912025e-01 4.43914801e-01 5.06732762e-01 5.15928864e-01 7.44239271e-01 1.06618631e+00 3.95656407e-01 9.64677036e-01 1.99163496e-01 4.81045246e-01 -3.16975027e-01 3.62084538e-01 -1.53175741e-02 6.99894547e-01 1.20428669e+00 3.63451242e-01 -4.90420908e-01 -8.86106491e-01 7.32100010e-01 -1.82984936e+00 -6.70893371e-01 -6.41895890e-01 1.21537340e+00 8.39841783e-01 4.97583061e-01 1.48454159e-01 -5.61182760e-02 4.17301297e-01 7.35715270e-01 -3.80667508e-01 -9.82511640e-01 -1.71194404e-01 2.78428555e-01 3.22540432e-01 3.13658148e-01 -1.62670672e+00 1.12032616e+00 6.66042900e+00 5.04467547e-01 -1.42162526e+00 1.36552587e-01 1.13523781e+00 -6.57215342e-02 -4.33395982e-01 -4.31275398e-01 -8.99630487e-01 7.43403971e-01 1.08174944e+00 -1.89061627e-01 -3.79507303e-01 1.17049909e+00 1.99535981e-01 -1.57004878e-01 -6.12347305e-01 3.42853099e-01 3.12479526e-01 -1.27813792e+00 -6.63765222e-02 -3.67342651e-01 1.10038853e+00 6.34234130e-01 4.64561313e-01 8.01434040e-01 4.17607605e-01 -8.77485394e-01 1.07308149e+00 2.24073052e-01 6.20145679e-01 -8.43370795e-01 1.48598170e+00 4.44008932e-02 -9.50865388e-01 2.54017621e-01 -6.08814061e-02 -2.66811937e-01 1.53650165e-01 1.10544372e+00 -4.54874486e-01 -1.14164324e-02 1.12867522e+00 1.45702016e+00 -3.46793115e-01 2.93462992e-01 -5.39475679e-01 1.07612538e+00 1.81670994e-01 -7.39533842e-01 4.16961849e-01 -2.14166585e-02 2.39183292e-01 1.63184488e+00 -1.91167966e-01 -3.02483529e-01 -8.63163844e-02 3.17970932e-01 -5.16269088e-01 3.57728034e-01 -2.32903421e-01 -9.07094851e-02 1.39808506e-01 1.74105954e+00 -6.86178386e-01 -4.98400241e-01 -4.47966933e-01 4.26009297e-01 3.36598754e-01 1.48052648e-01 -6.66057169e-01 -5.06436408e-01 8.39497864e-01 -1.40503407e-01 2.76655942e-01 2.39057556e-01 -8.78612101e-01 -1.38410544e+00 -1.31184459e-01 -5.13584912e-01 2.51207799e-01 -7.97909558e-01 -1.48463762e+00 1.03857720e+00 -5.48219025e-01 -9.15846467e-01 -1.11787163e-01 -7.71352708e-01 -1.23274696e+00 8.43719184e-01 -1.94485307e+00 -1.32296240e+00 -2.71550506e-01 5.36595345e-01 7.38426745e-01 -1.16656445e-01 8.96489978e-01 1.57495290e-01 -6.57096505e-01 5.50549567e-01 -3.48860361e-02 2.11094618e-01 8.23237479e-01 -1.55861735e+00 8.03091645e-01 4.24957424e-01 -5.73531926e-01 4.09471542e-01 6.59900546e-01 -4.80212748e-01 -1.10328615e+00 -1.46624815e+00 1.68630922e+00 -7.73396730e-01 1.36209285e+00 -7.46685803e-01 -1.28563464e-01 8.27540457e-01 1.98200911e-01 -4.46154028e-01 1.28955412e+00 7.10502625e-01 -2.90710896e-01 -1.28688753e-01 -7.43922174e-01 6.41517699e-01 2.19625950e-01 -4.73597944e-01 -2.61604220e-01 7.18000054e-01 8.52000594e-01 -4.85838413e-01 -7.66448498e-01 3.04761499e-01 5.65563798e-01 -6.95496202e-01 4.59699184e-01 -8.12172294e-01 1.28910661e+00 2.18021590e-02 -9.33885351e-02 -1.55581045e+00 1.31339058e-01 -7.99934715e-02 7.19683692e-02 1.19688058e+00 1.04096007e+00 -6.30905330e-01 8.45898449e-01 3.18668634e-01 -2.90967345e-01 -1.18521404e+00 -2.76925266e-01 1.84022620e-01 2.41967499e-01 -8.71851623e-01 5.87852359e-01 1.09309995e+00 4.86252487e-01 1.07115555e+00 -1.51649937e-01 -1.95199028e-01 1.13499397e-02 4.03887242e-01 7.86646903e-01 -9.82639909e-01 4.26653117e-01 -6.43307745e-01 -3.73531245e-02 -8.80195975e-01 5.40885687e-01 -6.58767402e-01 1.84341565e-01 -1.50521505e+00 3.57781023e-01 -1.17747828e-01 -4.34681118e-01 2.83728957e-01 -2.84710765e-01 5.97818732e-01 -2.39057869e-01 1.14030205e-01 -1.27498090e+00 4.63527352e-01 1.01307118e+00 -3.20640802e-01 2.85833944e-02 4.00740206e-01 -1.41408694e+00 1.01280570e+00 1.15561628e+00 -4.53871936e-01 -1.70734987e-01 -3.82657677e-01 1.10647619e+00 -4.08572733e-01 -2.60472238e-01 -3.55332404e-01 6.61731930e-04 -1.09460972e-01 1.93769976e-01 -1.10761118e+00 1.84193522e-01 -2.51817614e-01 -9.25732017e-01 6.27814531e-02 -6.28624916e-01 -4.75560501e-03 2.58683473e-01 1.40484974e-01 -7.02100992e-01 7.43803978e-02 4.80096638e-01 1.82462394e-01 -4.08626586e-01 3.42983127e-01 -7.72058308e-01 1.87467441e-01 7.69513547e-01 3.71981472e-01 -5.81338763e-01 -5.92124701e-01 -4.41002637e-01 6.39913797e-01 4.88251410e-02 4.59308058e-01 4.49287415e-01 -1.11710179e+00 -8.21143508e-01 -2.68946681e-02 4.41798419e-01 -3.84018719e-02 1.71939015e-01 1.02368903e+00 -1.17684729e-01 5.33416092e-01 4.14766759e-01 -4.14836317e-01 -1.22152483e+00 -1.22081429e-01 1.47655189e-01 -7.26679742e-01 2.47969374e-01 1.21827519e+00 2.72135258e-01 -7.23712504e-01 -5.07516563e-02 -1.40076160e-01 -1.12913787e+00 5.61403632e-01 8.27192426e-01 -8.23406577e-02 3.60128641e-01 -8.21874440e-01 -5.77133894e-01 4.03124750e-01 -4.47590768e-01 2.56451637e-01 1.56856120e+00 -1.09877802e-01 -5.55597365e-01 8.04920256e-01 1.40564382e+00 9.60014835e-02 -5.51285565e-01 -2.52301581e-02 -6.31508976e-02 8.45835581e-02 3.03631514e-01 -9.05760527e-01 -9.99425471e-01 7.64614463e-01 -1.04206823e-01 7.71869302e-01 1.01431882e+00 1.23574570e-01 1.13853180e+00 3.35279942e-01 -2.43452750e-03 -1.31133449e+00 -5.86010329e-03 1.47917795e+00 6.80057883e-01 -1.56651175e+00 -5.73162325e-02 -2.52131194e-01 -1.22933888e+00 1.01091588e+00 5.36236525e-01 -2.36118123e-01 1.29392552e+00 3.18180740e-01 5.09936094e-01 -8.01379979e-01 -9.93131459e-01 -1.88926861e-01 4.70235318e-01 1.51359186e-01 1.10469520e+00 4.38179135e-01 -3.40924293e-01 1.23070383e+00 -9.40958023e-01 -2.88818210e-01 6.83212876e-01 8.16902816e-01 -4.30906415e-01 -4.32711601e-01 2.00823411e-01 7.42463827e-01 -1.03322792e+00 -3.65890533e-01 -8.61438870e-01 3.23726088e-01 -2.23108083e-01 1.50610745e+00 2.97065139e-01 -8.36330891e-01 5.38512886e-01 -2.44745612e-01 -4.64394748e-01 -9.22041655e-01 -9.97991979e-01 -3.66895318e-01 4.93322074e-01 -5.59749544e-01 -6.13453925e-01 -8.12390924e-01 -9.64020669e-01 -5.45821190e-01 -3.41136247e-01 4.68229324e-01 1.11357033e+00 1.03627479e+00 1.21543132e-01 6.70118809e-01 1.30887961e+00 -6.35168314e-01 -4.06317264e-01 -1.14070427e+00 -6.63994968e-01 4.05936658e-01 5.09277880e-01 -9.13756639e-02 -6.83756888e-01 1.48011744e-02]
[11.183066368103027, 6.8962578773498535]
2e359dfd-14cf-47ca-a71d-b07f646249b1
faster-person-re-identification
2008.06826
null
https://arxiv.org/abs/2008.06826v1
https://arxiv.org/pdf/2008.06826v1.pdf
Faster Person Re-Identification
Fast person re-identification (ReID) aims to search person images quickly and accurately. The main idea of recent fast ReID methods is the hashing algorithm, which learns compact binary codes and performs fast Hamming distance and counting sort. However, a very long code is needed for high accuracy (e.g. 2048), which compromises search speed. In this work, we introduce a new solution for fast ReID by formulating a novel Coarse-to-Fine (CtF) hashing code search strategy, which complementarily uses short and long codes, achieving both faster speed and better accuracy. It uses shorter codes to coarsely rank broad matching similarities and longer codes to refine only a few top candidates for more accurate instance ReID. Specifically, we design an All-in-One (AiO) framework together with a Distance Threshold Optimization (DTO) algorithm. In AiO, we simultaneously learn and enhance multiple codes of different lengths in a single model. It learns multiple codes in a pyramid structure, and encourage shorter codes to mimic longer codes by self-distillation. DTO solves a complex threshold search problem by a simple optimization process, and the balance between accuracy and speed is easily controlled by a single parameter. It formulates the optimization target as a $F_{\beta}$ score that can be optimised by Gaussian cumulative distribution functions. Experimental results on 2 datasets show that our proposed method (CtF) is not only 8% more accurate but also 5x faster than contemporary hashing ReID methods. Compared with non-hashing ReID methods, CtF is $50\times$ faster with comparable accuracy. Code is available at https://github.com/wangguanan/light-reid.
['Zeng-Guang Hou', "Guan'an Wang", 'Shaogang Gong', 'Jian Cheng']
2020-08-16
null
https://www.ecva.net/papers/eccv_2020/papers_ECCV/html/566_ECCV_2020_paper.php
https://www.ecva.net/papers/eccv_2020/papers_ECCV/papers/123530273.pdf
eccv-2020-8
['code-search', 'code-search', '2048']
['computer-code', 'computer-vision', 'playing-games']
[-2.76119351e-01 -5.14926434e-01 -1.43638089e-01 -4.15911615e-01 -7.63717413e-01 -5.29914320e-01 3.08182001e-01 4.39979523e-01 -7.83390522e-01 5.03719032e-01 3.18088919e-01 1.30787149e-01 -1.27458379e-01 -9.48639274e-01 -5.87571323e-01 -7.44241059e-01 -2.29028583e-01 5.95267415e-01 2.44828254e-01 -1.47229508e-01 3.09504300e-01 2.19411179e-01 -1.65097713e+00 -1.39744073e-01 1.07936049e+00 6.44782722e-01 1.63420796e-01 4.53692049e-01 9.17009637e-02 3.39450091e-01 -3.95290613e-01 -9.05324817e-01 4.77423698e-01 -2.16072738e-01 -6.80294693e-01 -6.60765529e-01 5.46646655e-01 -3.92516136e-01 -7.05540895e-01 1.17591226e+00 8.21387947e-01 6.06237128e-02 5.59533298e-01 -1.22863412e+00 -9.42579269e-01 6.13275111e-01 -8.18204165e-01 3.41670692e-01 7.16899872e-01 1.44827981e-02 8.92111123e-01 -9.05873835e-01 1.37861475e-01 1.27294886e+00 9.79072869e-01 3.99817407e-01 -1.16099727e+00 -1.26788855e+00 -1.92776963e-01 5.33001900e-01 -2.31979656e+00 -2.53756583e-01 2.65202820e-01 -2.42215708e-01 5.32284260e-01 4.36085373e-01 6.73745513e-01 4.75422621e-01 7.56784976e-02 7.25723624e-01 9.60123301e-01 -2.26210847e-01 3.69927883e-02 -6.43412545e-02 2.52111584e-01 7.66483068e-01 5.76963425e-01 3.01914424e-01 -5.76435268e-01 -4.19739693e-01 6.07847571e-01 3.02572310e-01 -2.63727427e-01 -1.26901418e-01 -1.18745542e+00 9.03480053e-01 8.61789048e-01 1.18650347e-01 -2.51042470e-02 1.24085173e-01 3.11993927e-01 2.30701074e-01 -1.79462776e-01 1.02243938e-01 1.11095123e-01 -1.02092914e-01 -9.43509877e-01 5.70955813e-01 5.35280108e-01 8.98206770e-01 1.12678182e+00 -6.09192491e-01 -3.33491951e-01 8.66549790e-01 1.60048932e-01 8.61758530e-01 4.51040119e-01 -6.56947613e-01 3.82395804e-01 5.39439797e-01 1.17202826e-01 -1.18189430e+00 -3.06597233e-01 -2.37869680e-01 -1.11128414e+00 -2.53566623e-01 2.95801342e-01 3.30832481e-01 -7.72533655e-01 1.71499622e+00 4.23081756e-01 3.02899152e-01 -3.94309342e-01 1.03172493e+00 6.21370733e-01 6.97870374e-01 7.83009678e-02 1.12328075e-01 1.74256134e+00 -7.67123520e-01 -4.87381816e-01 3.17162648e-02 6.00960791e-01 -7.37517715e-01 8.10725808e-01 5.27395308e-02 -9.86341894e-01 -7.96894968e-01 -1.01173890e+00 -2.29596168e-01 -4.30537969e-01 5.87885566e-02 4.06557441e-01 8.88529062e-01 -1.21496236e+00 3.55100870e-01 -3.91330957e-01 -4.62983042e-01 1.67804942e-01 7.26662993e-01 -4.29334730e-01 -4.27501470e-01 -1.36641145e+00 7.40315616e-01 4.89736110e-01 -6.68508485e-02 -3.20183575e-01 -7.09336758e-01 -1.04519510e+00 8.55332315e-02 -1.79217324e-01 -7.68730819e-01 8.49552512e-01 -1.46755651e-01 -9.76727009e-01 9.68765497e-01 -3.93735886e-01 -3.73676538e-01 4.49787170e-01 -2.44552568e-01 -4.02184963e-01 1.97227001e-01 3.05438817e-01 9.65158880e-01 6.54453337e-01 -1.00371552e+00 -7.30316103e-01 -3.64773452e-01 -2.06580982e-01 1.83099359e-01 -6.04802251e-01 1.38901144e-01 -7.44285882e-01 -6.63456440e-01 -6.74524456e-02 -9.12820518e-01 -7.92005807e-02 7.90408999e-02 -7.77824670e-02 -4.18073326e-01 3.18861336e-01 -6.04214549e-01 1.58374667e+00 -2.12794185e+00 -7.64688700e-02 5.64613104e-01 4.07307953e-01 1.63901731e-01 -6.44040257e-02 4.99534637e-01 9.18933749e-02 -1.78140581e-01 -1.19746342e-01 -5.14421999e-01 1.47557259e-01 3.82949784e-02 9.43884160e-03 8.56771588e-01 -3.52279395e-01 9.29820120e-01 -9.51722562e-01 -9.07700539e-01 8.45997706e-02 6.78046942e-01 -6.44174457e-01 1.57084689e-01 7.59990931e-01 3.19109969e-02 -2.05409378e-01 8.08202922e-01 1.13545203e+00 -1.59105316e-01 -2.24580660e-01 -2.97997743e-01 -4.11356211e-01 -6.37479722e-02 -1.45274258e+00 1.54241216e+00 -8.40517953e-02 9.97571051e-02 -1.49270445e-01 -7.84122050e-01 1.12807822e+00 -6.87966943e-02 2.84252584e-01 -8.50021541e-01 1.94130421e-01 3.41506660e-01 -4.44672495e-01 -2.09199607e-01 6.72812581e-01 1.04622953e-01 -2.55197048e-01 3.28690618e-01 -3.15460771e-01 4.01974827e-01 3.64893019e-01 1.91345349e-01 9.63533401e-01 -4.69092011e-01 1.05159529e-01 -3.97807002e-01 8.18355560e-01 -1.57795504e-01 7.46270657e-01 9.45634961e-01 -3.96257758e-01 6.97177887e-01 -2.29592934e-01 -6.26645923e-01 -1.13449013e+00 -1.15812826e+00 -4.66948867e-01 1.06504762e+00 8.57606530e-01 -7.29530931e-01 -7.28509963e-01 -2.75008410e-01 5.03673077e-01 3.83151323e-03 -6.59297407e-01 -2.47754082e-01 -8.24267030e-01 -8.51044774e-01 6.91360235e-01 5.13596773e-01 9.09431934e-01 -7.00104177e-01 -3.18402082e-01 4.65207882e-02 -3.51525366e-01 -7.67959177e-01 -1.19782484e+00 -3.40191007e-01 -4.81439203e-01 -8.60717714e-01 -1.08747005e+00 -1.19703829e+00 7.83882260e-01 6.98099315e-01 7.09929109e-01 4.96386141e-01 -5.44397473e-01 1.60902604e-01 -4.69178826e-01 3.71981524e-02 1.62748307e-01 1.94848791e-01 5.08516669e-01 3.97779644e-02 7.51995146e-01 -4.55960780e-01 -1.12104142e+00 5.88409662e-01 -6.82407737e-01 -2.50709832e-01 8.49617898e-01 9.14742947e-01 4.58325684e-01 5.29364794e-02 1.93411037e-01 -1.99628383e-01 5.33842623e-01 -3.59744400e-01 -4.99010086e-01 2.58465916e-01 -8.01394403e-01 8.51822793e-02 5.26968896e-01 -5.16538978e-01 -5.62820137e-01 8.14906657e-02 -1.49067581e-01 -2.15074658e-01 1.35896713e-01 1.93337053e-01 2.66726352e-02 -4.57853913e-01 5.52163899e-01 6.60106361e-01 4.36503142e-02 -4.45562273e-01 1.73881426e-01 8.14280212e-01 9.94392276e-01 -7.69393146e-01 1.34119213e+00 5.09016633e-01 -3.90476525e-01 -4.15842295e-01 -3.81983638e-01 -8.62618208e-01 -6.45316184e-01 -7.91819021e-02 7.07809508e-01 -1.18811560e+00 -1.26342726e+00 5.74631512e-01 -7.63364077e-01 -9.31017194e-03 1.61953300e-01 4.91826445e-01 -9.50155780e-02 7.56815016e-01 -8.65166783e-01 -5.63410044e-01 -7.23834038e-01 -8.00446749e-01 1.07465816e+00 6.59765542e-01 -2.35702291e-01 -4.79783863e-01 2.91685760e-01 3.74640763e-01 4.07969743e-01 -1.26725733e-01 4.38124448e-01 -2.92919993e-01 -5.54262161e-01 -4.54223752e-01 -6.08321071e-01 -1.86664298e-01 -8.91767368e-02 -4.29513901e-01 -7.06119835e-01 -8.65947545e-01 -5.27690947e-01 -2.19051987e-01 8.44320655e-01 3.05227600e-02 1.11716712e+00 -3.74404252e-01 -5.18447638e-01 9.31495190e-01 1.41716993e+00 -4.45499867e-02 7.76472688e-01 6.61047757e-01 6.35409176e-01 2.50194222e-01 6.08187795e-01 5.67136705e-01 1.06008697e+00 8.17014158e-01 -4.77369018e-02 -7.30020702e-02 3.30907591e-02 -4.62621450e-01 2.93258458e-01 7.62290061e-01 -9.23319459e-02 2.84948289e-01 -9.10755694e-01 6.34093881e-01 -1.81874204e+00 -1.15614522e+00 -3.90616158e-04 2.45876718e+00 1.07585394e+00 -1.23291938e-02 5.13105929e-01 3.03576708e-01 1.14870620e+00 -1.42135069e-01 -3.28863919e-01 -6.12598807e-02 2.67162081e-02 1.62754729e-01 7.44499505e-01 4.40399736e-01 -1.09951472e+00 6.55484140e-01 5.66056490e+00 9.92601693e-01 -7.38479257e-01 2.35558804e-02 2.73090601e-01 5.54321744e-02 -4.83194068e-02 -9.42161754e-02 -1.23936093e+00 8.66907775e-01 7.07785070e-01 -5.56832373e-01 4.89221483e-01 7.76888013e-01 -2.12378666e-01 4.06855941e-02 -8.37473929e-01 1.67630911e+00 1.31072611e-01 -1.17342019e+00 6.61584586e-02 1.12156332e-01 5.39733529e-01 -2.60566503e-01 1.18484646e-01 3.86025637e-01 1.48800120e-01 -9.06581223e-01 8.66728008e-01 2.00477421e-01 9.35322464e-01 -9.95262623e-01 8.80924404e-01 2.00561613e-01 -1.97405410e+00 -4.02975649e-01 -7.15883195e-01 9.54271182e-02 -3.32099311e-02 3.84762824e-01 -1.95238322e-01 4.14094269e-01 1.24092472e+00 4.81628418e-01 -6.82089746e-01 1.67163956e+00 1.83750004e-01 -2.38608588e-02 -4.47832912e-01 -9.17695835e-02 -4.89119403e-02 -9.92966592e-02 1.47600830e-01 1.61040306e+00 4.71467227e-01 4.99485075e-01 3.80084395e-01 5.39991558e-01 -9.42003503e-02 2.11192280e-01 -1.90986931e-01 6.66956604e-01 1.15655172e+00 1.10625124e+00 -5.95477045e-01 -3.60279858e-01 -2.48177364e-01 1.15629721e+00 2.72292376e-01 7.23191276e-02 -9.26629484e-01 -8.51916671e-01 7.00787127e-01 1.73873112e-01 3.31960589e-01 -2.16137081e-01 -6.96667731e-02 -1.10104096e+00 4.93130423e-02 -6.21383429e-01 9.06195402e-01 -2.55271882e-01 -1.36995935e+00 4.00469810e-01 1.02065280e-01 -1.28228772e+00 2.50148475e-01 -1.64328635e-01 -5.82945406e-01 8.58397365e-01 -1.68771887e+00 -9.80379641e-01 -7.41745412e-01 8.90601873e-01 7.18737468e-02 1.55787319e-01 6.76719904e-01 8.13879073e-01 -6.85592711e-01 1.40682542e+00 2.27203935e-01 3.94621760e-01 1.01487613e+00 -1.12559760e+00 5.93160868e-01 7.70227730e-01 -3.35547000e-01 9.78381515e-01 6.58407032e-01 -6.62567377e-01 -1.56894600e+00 -9.37242270e-01 1.13664782e+00 -1.93727151e-01 3.22566807e-01 -4.54716355e-01 -9.52379823e-01 2.04950958e-01 -2.09150523e-01 4.84512858e-02 7.66063929e-01 -9.85612497e-02 -7.94824243e-01 -5.00434935e-01 -1.39621150e+00 3.20938170e-01 1.22918487e+00 -5.86288512e-01 -5.08731544e-01 5.78306317e-02 3.66142631e-01 -3.88944536e-01 -9.99242485e-01 2.55068719e-01 8.35159898e-01 -1.04458797e+00 1.42967308e+00 2.91327596e-01 -2.09859684e-01 -7.08392143e-01 -4.90299659e-04 -7.91797578e-01 -8.81833434e-01 -6.27241850e-01 -5.44657260e-02 1.26603460e+00 -1.37957975e-01 -8.78047585e-01 7.61038721e-01 5.37196636e-01 2.79865772e-01 -5.66837192e-01 -1.06919193e+00 -1.08158576e+00 -1.98189002e-02 2.18910113e-01 1.07031071e+00 8.82728279e-01 1.02397501e-01 -2.41026267e-01 -4.20999229e-01 4.87219840e-01 1.23091865e+00 2.09057599e-01 7.35512018e-01 -1.30727541e+00 6.93332255e-02 -4.81059819e-01 -7.37247169e-01 -1.26096380e+00 -1.50688961e-01 -8.23219597e-01 -2.13710126e-03 -1.15158772e+00 7.35354185e-01 -8.40916216e-01 -3.63085270e-01 5.36855578e-01 -4.81073171e-01 6.29243135e-01 1.72518447e-01 6.15827918e-01 -7.03254044e-01 6.21120095e-01 6.76622987e-01 -3.11298490e-01 -9.27128419e-02 -3.25567842e-01 -7.38822222e-01 1.60649583e-01 6.69539452e-01 -4.48219597e-01 1.59533117e-02 -3.72835964e-01 9.90187302e-02 -3.43272746e-01 4.18495476e-01 -1.33668053e+00 8.69472325e-01 2.57877469e-01 6.66829467e-01 -6.35542035e-01 2.14684248e-01 -4.55366462e-01 2.28955448e-01 7.90073752e-01 -2.91666575e-02 4.98941660e-01 -1.10690333e-01 4.10029113e-01 -1.70203805e-01 -4.81330231e-02 8.20882559e-01 9.83992144e-02 -7.50679910e-01 5.52543283e-01 1.42908111e-01 -2.22455308e-01 9.34244931e-01 -5.51248074e-01 -3.96930277e-01 -2.08247110e-01 -1.21186718e-01 7.01892972e-01 7.42557883e-01 4.30522144e-01 8.19870234e-01 -1.65271974e+00 -9.12160933e-01 2.88759351e-01 3.96003813e-01 -1.73008919e-01 3.77971351e-01 5.98793149e-01 -5.48912764e-01 3.68043900e-01 -1.90021053e-01 -5.82430243e-01 -1.40127742e+00 6.90212071e-01 7.49020651e-02 2.86073089e-02 -7.45006979e-01 1.02420878e+00 3.60929556e-02 -3.37809980e-01 4.84007478e-01 1.55015066e-01 -1.27684504e-01 -5.44515885e-02 1.21342826e+00 4.52620536e-01 -1.89533979e-01 -7.87140250e-01 -4.98496801e-01 1.13274026e+00 -3.43345761e-01 3.13632280e-01 1.08137298e+00 -3.51800025e-01 -2.40069672e-01 -3.76637489e-01 1.39166725e+00 3.43160927e-02 -9.16249275e-01 -3.45430255e-01 -1.26122862e-01 -9.32064712e-01 -2.82488018e-01 -2.64620066e-01 -8.10937822e-01 3.62409472e-01 8.93087506e-01 -8.05619806e-02 1.13398015e+00 5.62733738e-03 1.45721650e+00 2.85242349e-01 7.15391517e-01 -8.26212525e-01 -1.65728536e-02 2.68335760e-01 4.49055970e-01 -1.10146213e+00 7.16643631e-02 -2.67757595e-01 -2.84821659e-01 9.69689071e-01 6.43060029e-01 -1.41130555e-02 4.94116724e-01 1.55745625e-01 -1.57037586e-01 -4.63862605e-02 -8.11810568e-02 -2.75949627e-01 1.25938609e-01 5.08824706e-01 5.14315208e-04 5.81515059e-02 -4.39734936e-01 4.22388226e-01 -5.29032409e-01 -8.54679868e-02 5.80309369e-02 9.60426390e-01 -6.80990100e-01 -1.25424600e+00 -8.71404707e-01 2.36367747e-01 -8.75106901e-02 -4.47541215e-02 1.70922443e-01 4.54721689e-01 4.09156233e-01 7.89203227e-01 9.16983560e-02 -8.52615416e-01 1.74565971e-01 -2.92337328e-01 2.80287683e-01 -1.35485232e-01 -5.00191927e-01 -4.59454179e-01 -5.09706795e-01 -4.42636371e-01 -1.74190059e-01 -7.80393600e-01 -1.25991285e+00 -1.04774070e+00 -3.14590096e-01 5.12528062e-01 3.22476923e-01 5.30461371e-01 3.73351365e-01 -4.07287389e-01 8.30973506e-01 -8.06024194e-01 -4.84461397e-01 -5.31211138e-01 -5.02570033e-01 4.32561785e-01 3.57977718e-01 -7.37251401e-01 -3.27780187e-01 -1.77840963e-01]
[14.733536720275879, 0.8730013370513916]
a99da683-b8f4-423b-8629-38772f111a50
exploiting-bilateral-symmetry-in-brain-lesion
1907.08196
null
https://arxiv.org/abs/1907.08196v1
https://arxiv.org/pdf/1907.08196v1.pdf
Exploiting bilateral symmetry in brain lesion segmentation
Brain lesions, including stroke and tumours, have a high degree of variability in terms of location, size, intensity and form, making automatic segmentation difficult. We propose an improvement to existing segmentation methods by exploiting the bilateral quasi-symmetry of healthy brains, which breaks down when lesions are present. Specifically, we use nonlinear registration of a neuroimage to a reflected version of itself ("reflective registration") to determine for each voxel its homologous (corresponding) voxel in the other hemisphere. A patch around the homologous voxel is added as a set of new features to the segmentation algorithm. To evaluate this method, we implemented two different CNN-based multimodal MRI stroke lesion segmentation algorithms, and then augmented them by adding extra symmetry features using the reflective registration method described above. For each architecture, we compared the performance with and without symmetry augmentation, on the SISS Training dataset of the Ischemic Stroke Lesion Segmentation Challenge (ISLES) 2015 challenge. Using affine reflective registration improves performance over baseline, but nonlinear reflective registration gives significantly better results: an improvement in Dice coefficient of 13 percentage points over baseline for one architecture and 9 points for the other. We argue for the broad applicability of adding symmetric features to existing segmentation algorithms, specifically using nonlinear, template-free methods.
['Tanya Schmah', 'Uladzimir Yahorau', 'Kevin Raina']
2019-07-18
null
null
null
null
['ischemic-stroke-lesion-segmentation']
['medical']
[ 4.46333319e-01 1.25115365e-01 4.14693505e-02 -4.92518932e-01 -8.74715388e-01 -7.42589355e-01 8.37609649e-01 2.12291047e-01 -7.86356628e-01 4.45251614e-01 6.28542364e-01 -1.31133333e-01 -1.07697077e-01 -5.86362839e-01 -4.60270494e-01 -8.04348350e-01 -2.13966921e-01 7.30095565e-01 5.67336380e-01 -1.47386909e-01 2.08412603e-01 9.04058695e-01 -9.53467190e-01 4.18741256e-01 6.42629743e-01 6.34422898e-01 -3.16116065e-02 3.54809701e-01 1.40903339e-01 1.73567489e-01 -3.06218863e-01 -2.30746970e-01 5.62330484e-01 -4.06506240e-01 -1.21114528e+00 1.36488425e-02 6.69392407e-01 -3.20818305e-01 -3.67665499e-01 7.11513042e-01 5.43293357e-01 8.19187164e-02 1.03057873e+00 -8.32239866e-01 -1.43967375e-01 5.54865956e-01 -7.27499127e-01 5.68073332e-01 -5.42618288e-03 1.38612285e-01 5.54654181e-01 -4.88608569e-01 1.02984059e+00 9.36198711e-01 8.97721946e-01 4.84422535e-01 -1.40003586e+00 -4.48399544e-01 -9.83310565e-02 9.89057496e-02 -1.11755383e+00 -4.23542291e-01 1.62921995e-01 -6.80425406e-01 1.09067369e+00 3.53714526e-01 7.91594803e-01 7.33271837e-01 2.86494315e-01 5.57605982e-01 1.48285747e+00 -1.40812710e-01 1.15081035e-01 -5.15184462e-01 3.17769825e-01 3.72801095e-01 -2.41477285e-02 7.07167089e-02 5.86520173e-02 -2.11214334e-01 8.35712910e-01 -5.77729046e-02 -4.22294289e-01 -6.75078154e-01 -1.69005406e+00 7.88112283e-01 7.55576193e-01 5.70634544e-01 -4.14298832e-01 -1.97293848e-01 4.27690268e-01 1.10646963e-01 4.26214606e-01 5.01362383e-01 -2.93498814e-01 9.39040035e-02 -1.44378340e+00 3.67798448e-01 5.12512267e-01 2.67194390e-01 2.54344493e-01 -5.23296058e-01 -6.84118569e-01 7.94447482e-01 -3.95999514e-02 4.35686447e-02 9.12857711e-01 -9.93579626e-01 3.62828672e-01 6.37708366e-01 -4.58516151e-01 -6.40966058e-01 -1.08450198e+00 -6.96810901e-01 -8.27524662e-01 4.49196577e-01 7.77777016e-01 -2.13238403e-01 -1.30211723e+00 1.67956436e+00 1.74327031e-01 3.83223430e-03 -3.49263966e-01 1.02020204e+00 5.80051661e-01 -2.59464961e-02 -1.76995434e-02 5.77998795e-02 1.28930759e+00 -8.94311130e-01 -1.45322233e-01 -2.15835437e-01 8.87782156e-01 -6.73151791e-01 8.55974913e-01 1.73403874e-01 -1.15590179e+00 2.45661333e-01 -8.54087412e-01 -1.16427183e-01 -2.94779956e-01 -8.41626599e-02 3.47234815e-01 6.59095705e-01 -1.36966383e+00 6.74744725e-01 -1.10408425e+00 -4.69339699e-01 8.63482952e-01 4.64316577e-01 -7.63351858e-01 4.20591384e-02 -6.57079160e-01 1.32738698e+00 1.78851053e-01 -2.03300819e-01 -3.72905880e-01 -1.15089476e+00 -6.06733263e-01 -1.58475235e-01 -7.88164064e-02 -7.15965927e-01 9.84737813e-01 -7.24783182e-01 -1.22254086e+00 1.28998733e+00 -9.95164663e-02 -4.77882802e-01 1.08625531e+00 2.41381079e-01 -2.71776039e-02 3.68155211e-01 7.59033784e-02 9.31791484e-01 4.41290230e-01 -1.08263993e+00 -1.85409546e-01 -8.35901022e-01 -1.12087570e-01 3.51605117e-01 1.89955994e-01 3.34829897e-01 -5.98808527e-02 -6.15959644e-01 5.43995261e-01 -1.00304973e+00 -3.58020455e-01 -3.42902727e-02 -3.13329846e-01 1.43074185e-01 7.06710100e-01 -1.10040581e+00 5.91949821e-01 -1.84138036e+00 1.89825550e-01 6.37149692e-01 5.24567306e-01 1.06175937e-01 -2.65083969e-01 -2.57392943e-01 -5.45087755e-01 1.42798394e-01 -8.53386104e-01 -1.35555193e-01 -3.29028338e-01 -4.63808738e-02 2.09094867e-01 8.00599635e-01 6.60139024e-02 1.09579396e+00 -6.10088110e-01 -2.11423352e-01 2.36331895e-02 6.36583507e-01 -5.00730157e-01 -2.41862953e-01 5.04585683e-01 6.45927668e-01 -2.55751442e-02 1.03672422e-01 6.64466560e-01 3.43078226e-02 -1.79913446e-01 -2.61642665e-01 -1.87641412e-01 1.85048684e-01 -8.54540825e-01 1.76772010e+00 -1.56639308e-01 6.39000595e-01 4.78105173e-02 -1.15520334e+00 6.66944504e-01 2.83805490e-01 9.52480197e-01 -6.91238105e-01 2.79115409e-01 2.48522475e-01 5.88212252e-01 -2.62944847e-01 -1.92547962e-01 -2.11942926e-01 4.42299187e-01 6.23147547e-01 -8.95726159e-02 -2.36610606e-01 2.35261992e-01 2.35742733e-01 1.49142063e+00 -1.30142391e-01 1.15324408e-01 -5.17572284e-01 4.38996226e-01 -3.30633409e-02 1.30876020e-01 6.21120989e-01 -3.53702933e-01 1.28259814e+00 7.50337839e-01 -2.04319462e-01 -1.04784226e+00 -1.34862876e+00 -5.17329574e-01 5.88588357e-01 -3.37683350e-01 -7.17629865e-02 -1.24504519e+00 -8.15611839e-01 -8.27750266e-02 3.82177114e-01 -8.22354138e-01 -7.44261593e-02 -8.68083119e-01 -1.19171333e+00 7.25027323e-01 5.76153934e-01 4.28932846e-01 -1.01048040e+00 -9.72116709e-01 9.56071094e-02 -3.13741833e-01 -9.46253657e-01 -6.03446424e-01 8.36289525e-02 -1.05162108e+00 -1.15149558e+00 -1.31277096e+00 -7.04192996e-01 7.49004781e-01 2.11407263e-02 8.72152090e-01 2.56838948e-01 -5.54321408e-01 2.35373542e-01 -2.12350428e-01 -7.29085729e-02 -1.56434983e-01 3.18767637e-01 -3.47607374e-01 -2.01549694e-01 -1.05189011e-01 -5.88858247e-01 -8.93646955e-01 3.34802598e-01 -9.38116789e-01 1.91991162e-02 7.17341542e-01 6.20989919e-01 4.40642297e-01 -5.73001325e-01 4.39125657e-01 -7.28247225e-01 7.04523802e-01 -3.56975585e-01 -1.86909944e-01 3.33214775e-02 -3.63910019e-01 -7.04046488e-02 9.90880430e-02 -3.31447870e-01 -6.28061295e-01 1.72109231e-01 -5.95440529e-02 3.63096148e-02 -2.88262814e-01 4.97030139e-01 1.31231755e-01 -4.11867201e-01 8.26708734e-01 -1.84546281e-02 5.86556137e-01 -4.26845104e-01 3.49831313e-01 3.98609877e-01 7.62274206e-01 -2.67214209e-01 5.84381998e-01 7.00889647e-01 1.38865665e-01 -3.95090222e-01 -3.69668096e-01 -5.59588075e-01 -1.26040876e+00 -3.72647233e-02 1.00366664e+00 -3.05059880e-01 -2.64166296e-01 5.96425891e-01 -1.04860723e+00 -6.16822720e-01 -4.43567634e-01 6.30979359e-01 -6.30709589e-01 4.73337352e-01 -3.45091313e-01 1.59859240e-01 -5.20523369e-01 -1.54352188e+00 9.82246876e-01 -1.09356187e-01 -3.49164844e-01 -9.38723743e-01 1.62034437e-01 3.86390865e-01 8.90383422e-01 4.69065964e-01 1.17366827e+00 -9.43472922e-01 1.00159705e-01 -3.93120676e-01 -5.58379710e-01 1.43837243e-01 6.02054000e-02 -4.06919509e-01 -5.35004318e-01 -2.50986785e-01 -4.30374816e-02 7.50063732e-02 1.13579750e+00 6.31145775e-01 9.29243207e-01 1.18927881e-01 -4.02235478e-01 5.58867514e-01 1.03642988e+00 -5.09230793e-02 7.93041170e-01 5.11938810e-01 5.35560250e-01 6.65965378e-01 -3.04372966e-01 -7.10481331e-02 3.30654591e-01 8.80793571e-01 2.21325502e-01 -5.22151351e-01 -7.44695663e-01 6.50103271e-01 1.72282264e-01 4.73855197e-01 -3.21642220e-01 3.01318556e-01 -1.27689207e+00 5.68823457e-01 -1.48092997e+00 -8.17246556e-01 -4.48744535e-01 2.40580606e+00 9.77782845e-01 -1.24161161e-01 4.13346767e-01 2.26301607e-02 7.04165161e-01 -1.42144054e-01 -4.50695246e-01 -1.70354575e-01 -2.97482997e-01 5.84707975e-01 7.37249315e-01 4.84528124e-01 -1.16757762e+00 6.40926421e-01 6.93544626e+00 5.00972152e-01 -1.35226512e+00 5.12655258e-01 6.33210063e-01 -1.96329013e-01 2.36254726e-02 9.60814673e-03 -1.69728845e-01 1.65867284e-01 7.18520820e-01 -5.80225326e-02 6.27564311e-01 -2.09705140e-02 2.19498992e-01 -2.05725268e-01 -1.07019317e+00 5.66093028e-01 2.59074241e-01 -1.25146019e+00 -8.00762475e-02 2.14939326e-01 7.26582408e-01 6.55767918e-01 -8.15126374e-02 -8.64566267e-02 4.79264371e-02 -1.29507148e+00 4.62819159e-01 6.04859054e-01 5.25299013e-01 -4.67506707e-01 8.51058483e-01 5.57425208e-02 -7.87959933e-01 4.02202815e-01 1.32627383e-01 3.21639150e-01 1.88516900e-01 3.22013348e-01 -8.55836034e-01 3.86402547e-01 6.10468864e-01 4.91269648e-01 -8.68840933e-01 1.69456148e+00 7.95707674e-05 4.66158509e-01 -5.01702249e-01 6.78693235e-01 3.56638074e-01 -2.90374070e-01 7.73175001e-01 1.30069804e+00 1.61456525e-01 6.01632707e-03 -1.57305021e-02 9.66152728e-01 -1.99394487e-02 4.01352465e-01 -2.51205832e-01 6.32316530e-01 -9.56822634e-02 1.44328761e+00 -1.27287304e+00 -4.21163112e-01 -2.80634701e-01 8.46211314e-01 1.23462141e-01 3.00111353e-01 -4.44483757e-01 -2.30352789e-01 2.48070613e-01 3.38014901e-01 -3.71953733e-02 -2.13245228e-01 -6.74699008e-01 -9.94174898e-01 2.87240800e-02 -7.52757668e-01 3.95871401e-01 -5.73082149e-01 -1.14568460e+00 5.01273096e-01 1.26101643e-01 -8.09315503e-01 -9.06682760e-02 -4.40709561e-01 -8.61163676e-01 1.12964165e+00 -1.34598267e+00 -1.07150102e+00 -4.30440784e-01 7.28936732e-01 1.62491545e-01 1.10297641e-02 7.64405072e-01 2.20753878e-01 -2.01231480e-01 5.44499993e-01 -4.62811030e-02 3.77012342e-01 8.33604515e-01 -1.26476002e+00 3.77096504e-01 6.76853001e-01 2.89861448e-02 4.79400933e-01 2.82554358e-01 -6.48138285e-01 -7.63386905e-01 -1.05854607e+00 6.10911667e-01 -3.67966890e-01 6.98804259e-01 -4.54866923e-02 -9.58471179e-01 8.16367745e-01 1.59525275e-01 8.36130083e-02 4.71501976e-01 -2.26290956e-01 -4.09694165e-01 3.52040619e-01 -1.44122088e+00 6.65882051e-01 1.06909740e+00 -1.39106780e-01 -8.14018250e-01 6.09145045e-01 2.83988923e-01 -5.34033298e-01 -1.08911681e+00 6.06474638e-01 7.73379922e-01 -8.33067894e-01 1.07182658e+00 -6.39079273e-01 2.88650870e-01 1.67833138e-02 2.43167236e-01 -1.59901977e+00 -3.35813195e-01 -2.50776887e-01 6.00550175e-01 6.78205490e-01 4.64283258e-01 -9.82493639e-01 9.09729779e-01 7.05769539e-01 -4.18429494e-01 -8.39997828e-01 -1.30989718e+00 -8.19850326e-01 9.07552242e-01 -1.47367150e-01 4.83747691e-01 9.37903702e-01 1.63628414e-01 -1.63009822e-01 4.98741686e-01 -1.55044496e-01 3.90202701e-01 -1.78090081e-01 3.44771087e-01 -1.32518184e+00 1.30786300e-01 -1.22838545e+00 -7.62326717e-01 -4.47071642e-01 2.02978000e-01 -1.74956453e+00 -1.66370511e-01 -1.82114255e+00 4.56040442e-01 -3.09722304e-01 -3.04994166e-01 9.07649636e-01 1.74638733e-01 6.98792815e-01 3.12675774e-01 2.56812513e-01 -3.40780802e-02 9.77458805e-02 1.40630686e+00 -1.51048675e-01 -2.38706619e-01 -1.03154458e-01 -5.42548656e-01 7.21826851e-01 9.30890858e-01 -3.63895833e-01 6.91142604e-02 -5.07741511e-01 -3.49452168e-01 -1.95121557e-01 6.82446837e-01 -1.04538381e+00 2.16586798e-01 3.44300389e-01 4.48794484e-01 -2.83423156e-01 2.85682492e-02 -6.90062642e-01 -1.30997047e-01 5.84781945e-01 -4.67760801e-01 1.43983424e-01 2.15661988e-01 -2.05585256e-01 1.32048875e-01 -2.59860784e-01 9.79111612e-01 4.48623523e-02 2.05314327e-02 3.10655534e-01 -6.52629673e-01 2.89770454e-01 8.44639003e-01 -2.94005245e-01 -3.11834484e-01 -9.48641673e-02 -1.06228864e+00 -4.14928049e-02 4.29651380e-01 3.53058517e-01 4.40878421e-01 -1.07367301e+00 -1.01602602e+00 1.62356094e-01 -1.94367304e-01 -1.72393665e-01 4.62102890e-03 1.92829633e+00 -5.79206169e-01 3.71281385e-01 -6.07992411e-01 -6.74279332e-01 -1.24270833e+00 -1.38446197e-01 7.94421971e-01 -4.33266729e-01 -1.04702342e+00 5.13336837e-01 8.94093290e-02 -5.98656178e-01 -1.35473879e-02 -5.89295924e-01 -3.79577041e-01 1.67845547e-01 3.43320727e-01 4.22840327e-01 6.74383283e-01 -1.08827341e+00 -5.04637361e-01 7.72929788e-01 -2.61991262e-01 -4.59485769e-01 1.40134525e+00 2.31627926e-01 -4.32622671e-01 -6.99951053e-02 1.38710129e+00 -1.36893004e-01 -9.68847990e-01 -1.93584427e-01 1.78344324e-01 -1.63369387e-01 4.31944698e-01 -1.33227456e+00 -1.61047220e+00 7.20250964e-01 1.08117449e+00 -1.64206579e-01 8.43371689e-01 7.52655938e-02 7.02652931e-01 -2.36775051e-03 1.37065858e-01 -5.71079195e-01 -5.27320623e-01 5.29411495e-01 1.18526983e+00 -9.89458144e-01 -8.26376826e-02 -2.72723317e-01 -5.52357554e-01 1.22033370e+00 1.40127465e-01 -4.26678032e-01 6.94378316e-01 3.76536280e-01 1.49738953e-01 -4.66530740e-01 4.47611697e-02 -1.12235852e-01 6.93674803e-01 6.98922157e-01 4.67005610e-01 6.62002563e-02 -7.13235140e-01 4.35529411e-01 -4.61966485e-01 -1.42675042e-01 3.94823849e-01 7.34945297e-01 -2.35594675e-01 -1.13124192e+00 -2.94111133e-01 9.50938106e-01 -4.04535592e-01 -1.96239635e-01 -6.00733876e-01 8.49921286e-01 1.49303734e-01 5.94792366e-01 3.65177512e-01 1.38045684e-01 4.62079227e-01 1.34259120e-01 7.43287921e-01 -5.45811653e-01 -1.00785303e+00 2.24882811e-02 -1.94587156e-01 -6.97313130e-01 -3.38723481e-01 -1.26262641e+00 -1.72294772e+00 1.33379117e-01 3.85648906e-02 -3.14606190e-01 8.45479429e-01 1.33377290e+00 2.68700361e-01 4.59629238e-01 1.65034279e-01 -1.07136881e+00 -3.82514477e-01 -9.07666922e-01 -3.45555186e-01 6.67689204e-01 9.99839082e-02 -7.62475073e-01 -3.22907537e-01 -1.69951059e-02]
[14.223960876464844, -2.133103370666504]
7a9ed627-f935-4394-8a22-4424b525bd26
dukweb-diachronic-word-representations-from
2107.01076
null
https://arxiv.org/abs/2107.01076v2
https://arxiv.org/pdf/2107.01076v2.pdf
DUKweb: Diachronic word representations from the UK Web Archive corpus
Lexical semantic change (detecting shifts in the meaning and usage of words) is an important task for social and cultural studies as well as for Natural Language Processing applications. Diachronic word embeddings (time-sensitive vector representations of words that preserve their meaning) have become the standard resource for this task. However, given the significant computational resources needed for their generation, very few resources exist that make diachronic word embeddings available to the scientific community. In this paper we present DUKweb, a set of large-scale resources designed for the diachronic analysis of contemporary English. DUKweb was created from the JISC UK Web Domain Dataset (1996-2013), a very large archive which collects resources from the Internet Archive that were hosted on domains ending in `.uk'. DUKweb consists of a series word co-occurrence matrices and two types of word embeddings for each year in the JISC UK Web Domain dataset. We show the reuse potential of DUKweb and its quality standards via a case study on word meaning change detection.
['Barbara McGillivray', 'Mihai Cucuringu', 'Marya Bazzi', 'Pierpaolo Basile', 'Adam Tsakalidis']
2021-07-02
null
null
null
null
['diachronic-word-embeddings']
['natural-language-processing']
[-6.40339404e-02 -3.80859137e-01 -2.66424268e-01 -8.91901031e-02 -4.84185338e-01 -8.84350836e-01 9.21138525e-01 8.07963073e-01 -1.22710371e+00 4.89652395e-01 1.05367863e+00 -1.74329996e-01 -2.54711092e-01 -8.77956033e-01 -1.15172908e-01 -4.70917076e-01 -7.84999579e-02 2.67072558e-01 3.71759593e-01 -5.65783799e-01 4.36215192e-01 1.89423725e-01 -1.66322112e+00 2.00380925e-02 3.33489507e-01 5.97425640e-01 2.35806733e-01 6.24149621e-01 -6.17378712e-01 -1.09845415e-01 -2.82600880e-01 -6.97347105e-01 1.20121613e-01 -1.00789055e-01 -9.74305809e-01 -5.29838860e-01 2.10140660e-01 4.15258467e-01 -4.00194645e-01 1.15501475e+00 7.84363210e-01 3.47987473e-01 5.24717629e-01 -9.37501431e-01 -1.19095552e+00 5.87176800e-01 -1.54675543e-01 9.70840096e-01 3.26546431e-01 -1.60732239e-01 1.60415792e+00 -1.23032451e+00 1.26290202e+00 1.18623424e+00 8.05465698e-01 2.18540728e-01 -1.02382815e+00 -4.94806170e-01 5.93820261e-03 6.89161241e-01 -1.09685564e+00 -3.19804579e-01 6.21483147e-01 -6.09970510e-01 1.35033810e+00 1.37671798e-01 1.00008273e+00 1.47423136e+00 3.74371082e-01 2.62991428e-01 7.14307904e-01 -7.76779056e-01 2.76783913e-01 6.80953497e-03 5.42560577e-01 1.31351024e-01 4.33124244e-01 -2.28441790e-01 -8.92505705e-01 -4.49858963e-01 2.55879968e-01 -5.60627505e-02 -1.68877795e-01 -5.09444714e-01 -1.34656107e+00 1.04901743e+00 4.48959284e-02 8.74652565e-01 -3.89973223e-01 9.22197029e-02 8.73998106e-01 6.80506647e-01 1.04795039e+00 5.56506991e-01 -8.07909548e-01 -8.99650156e-01 -2.59961247e-01 4.07007545e-01 5.24280548e-01 5.77496350e-01 4.55831736e-01 -2.63201982e-01 2.42352277e-01 1.33262479e+00 2.72571146e-01 3.12327623e-01 1.23992550e+00 -3.41812789e-01 1.70025513e-01 6.70723379e-01 -9.08729527e-03 -1.26336253e+00 -2.44122341e-01 2.46385828e-01 -3.25190634e-01 -2.89269477e-01 1.57716572e-01 5.90372793e-02 -5.71437061e-01 1.82609439e+00 5.23038447e-01 -7.38207325e-02 -3.72896828e-02 3.84735644e-01 8.04700196e-01 6.93654418e-01 9.25353244e-02 -3.19797061e-02 1.62590837e+00 -2.46752143e-01 -9.06540632e-01 -1.72849730e-01 8.75685036e-01 -8.47053230e-01 1.27829278e+00 8.54715332e-02 -6.59745574e-01 -2.85352081e-01 -1.03979301e+00 -2.67523527e-01 -1.30424368e+00 -9.67853904e-01 4.78838503e-01 6.33149266e-01 -1.06940293e+00 5.01949012e-01 -4.79595482e-01 -1.14701355e+00 1.40629619e-01 -1.21193312e-01 -7.21572280e-01 -1.28643841e-01 -1.78633142e+00 1.35388041e+00 4.64794666e-01 -4.23455477e-01 -1.04953591e-02 -8.24209094e-01 -1.06579256e+00 -3.49861771e-01 9.25548300e-02 -1.26230150e-01 9.07823980e-01 -7.89792240e-01 -1.04232919e+00 1.28041744e+00 -3.10608242e-02 -3.64549011e-01 -3.74166779e-02 -1.94806367e-01 -1.17977834e+00 -2.43255764e-01 2.06121013e-01 2.14241341e-01 6.72989845e-01 -5.75057626e-01 -8.00718069e-01 -4.24907684e-01 -2.04034403e-01 -1.19514940e-02 -1.09993625e+00 1.62765756e-01 -1.93680793e-01 -1.10326338e+00 -3.49800408e-01 -7.30050862e-01 1.38182402e-01 -1.55701399e-01 4.02349949e-01 -6.93209231e-01 6.98773801e-01 -9.78234828e-01 1.77765942e+00 -2.32378459e+00 2.86537349e-01 -3.53816338e-02 2.97598869e-01 2.09505156e-01 -3.77115399e-01 1.05010772e+00 -3.24930012e-01 4.12503451e-01 -3.34113061e-01 -4.95195910e-02 1.69570908e-01 5.35974920e-01 -2.13305578e-01 8.65615606e-01 -6.78474829e-02 8.16494942e-01 -1.29937398e+00 -1.78608656e-01 2.26216987e-01 3.74859393e-01 -4.68978465e-01 -3.62915844e-01 1.94682285e-01 -5.90434909e-01 -1.74294561e-01 2.27848992e-01 3.10685694e-01 3.43563765e-01 2.13564098e-01 5.31589761e-02 -4.90060002e-01 3.35988343e-01 -8.31596375e-01 1.82668436e+00 -5.44844210e-01 1.16553509e+00 -3.54204983e-01 -7.33154655e-01 6.58588469e-01 1.13209859e-01 5.16816735e-01 -9.67065036e-01 1.34893879e-01 1.55557275e-01 -1.71047188e-02 -4.06651229e-01 1.14112759e+00 -2.88905412e-01 -5.67627370e-01 5.47346354e-01 1.23534679e-01 -8.05021748e-02 2.74615020e-01 9.99392122e-02 1.39913738e+00 -1.98177978e-01 7.62205005e-01 -7.99560487e-01 2.72461504e-01 2.71902353e-01 5.48991323e-01 -1.22963022e-02 -3.95831883e-01 3.83352131e-01 2.69329607e-01 -6.73755050e-01 -1.46739459e+00 -1.01203036e+00 -6.58312082e-01 1.32445538e+00 -1.27578884e-01 -9.01128650e-01 -2.23229840e-01 -2.28786364e-01 3.60547602e-01 9.58724499e-01 -1.10031676e+00 -3.79105777e-01 -4.70957875e-01 -7.11095929e-01 3.50389451e-01 1.86793327e-01 -8.60584825e-02 -1.18309128e+00 -5.93650401e-01 4.61251825e-01 1.35505665e-02 -6.41510606e-01 -6.16503716e-01 1.13630824e-01 -3.30469877e-01 -1.07505035e+00 -5.94455421e-01 -9.54659581e-01 4.80209887e-02 1.39906719e-01 1.26301551e+00 -3.78868282e-01 -6.90184236e-01 6.98244214e-01 -9.51956391e-01 -5.38825214e-01 -3.98541301e-01 6.96143582e-02 5.28078377e-01 -1.56940669e-01 9.25403595e-01 -5.05791485e-01 -3.44287157e-01 -2.77968973e-01 -1.32705712e+00 -6.97310805e-01 -8.10534433e-02 9.22250926e-01 3.69046181e-01 -1.10197015e-01 8.18700492e-01 -8.39027107e-01 1.13715374e+00 -9.09485161e-01 -2.51518756e-01 4.11976613e-02 -8.73992503e-01 3.71024758e-02 2.12303683e-01 -3.75587821e-01 -7.13360846e-01 -7.42037773e-01 -2.61727661e-01 2.50518769e-01 2.17479706e-01 6.33318126e-01 2.83940405e-01 4.55640614e-01 5.97904861e-01 1.36637717e-01 -1.48870811e-01 -8.13127458e-01 8.56871784e-01 1.09747934e+00 3.51277411e-01 -2.92848349e-01 5.14066160e-01 2.77412266e-01 -6.22489512e-01 -1.30262470e+00 -3.23569030e-01 -9.85053241e-01 -7.82988906e-01 -1.91702321e-02 1.01628458e+00 -7.04149902e-01 -3.02967224e-02 2.92615086e-01 -9.03442144e-01 9.28694978e-02 -6.37342513e-01 2.14606553e-01 -1.90488055e-01 5.10938168e-01 -1.62765622e-01 -2.18728170e-01 -3.01891029e-01 -3.77471358e-01 5.58092177e-01 -2.34200567e-01 -9.16540742e-01 -1.59287333e+00 9.26936388e-01 -2.85563797e-01 5.35086930e-01 4.21863914e-01 1.38849938e+00 -8.35648835e-01 7.11500823e-01 -3.07752430e-01 1.55426785e-01 2.61016041e-01 7.08914578e-01 5.36387600e-02 -6.35248125e-01 -4.74491864e-01 -2.51658708e-01 -9.98173505e-02 7.05917656e-01 -2.34301742e-02 5.04949808e-01 -1.79520056e-01 -1.87562257e-01 7.24557638e-02 1.55556965e+00 1.42916486e-01 3.88729036e-01 9.18102145e-01 4.16285902e-01 5.15284181e-01 2.20316216e-01 7.18507528e-01 3.61690879e-01 6.30453587e-01 -4.13874462e-02 4.38121289e-01 -2.42311016e-01 -1.52417883e-01 6.10953033e-01 1.58802676e+00 2.64424980e-01 -1.55871943e-01 -1.16302299e+00 1.34058189e+00 -1.60376620e+00 -9.55640376e-01 -1.47184968e-01 2.22286057e+00 1.04451430e+00 1.01398163e-01 -9.62872803e-02 6.88001513e-02 5.91948271e-01 5.52359521e-01 -3.02227646e-01 -9.32367861e-01 -1.22368649e-01 6.96967363e-01 5.58045983e-01 3.34457904e-01 -7.71228135e-01 9.05018628e-01 5.92113209e+00 8.56035411e-01 -5.10450840e-01 5.28589368e-01 -3.08374554e-01 -6.87677264e-02 -7.47896135e-01 -1.86407298e-01 -3.79648745e-01 6.34035170e-01 1.31686318e+00 -8.85863423e-01 3.98487337e-02 6.30677223e-01 3.41053605e-02 -4.09566686e-02 -9.03762221e-01 1.07742333e+00 2.65827596e-01 -1.34178972e+00 -2.93536372e-02 9.71242562e-02 6.35322094e-01 3.07682455e-01 -5.32070734e-02 3.16867530e-01 5.44728637e-01 -7.29419351e-01 6.93365753e-01 2.63630390e-01 1.11286092e+00 -9.96353030e-01 8.81808460e-01 -2.04033092e-01 -1.17514944e+00 1.58262506e-01 -4.81610239e-01 -5.24636395e-02 2.15131357e-01 7.26526558e-01 -4.02581692e-01 3.97180229e-01 9.09593821e-01 1.14930546e+00 -7.50164866e-01 6.06597245e-01 1.54744774e-01 5.93387306e-01 -3.06250509e-02 -3.31602246e-01 4.31829959e-01 -2.81832546e-01 6.89069331e-01 1.48955035e+00 3.75105649e-01 -2.58906484e-01 -4.09785151e-01 1.45810813e-01 -2.39552736e-01 3.95258665e-01 -9.13671970e-01 -6.56967163e-01 7.22479641e-01 1.03775835e+00 -6.04740381e-01 -1.21218815e-01 -7.18564868e-01 1.25204265e+00 4.14892018e-01 -3.17364378e-04 -4.55165207e-01 -8.93367112e-01 1.62480915e+00 5.45384660e-02 2.91048080e-01 -4.96969193e-01 1.80443957e-01 -1.02124465e+00 -9.31174234e-02 -6.04736388e-01 4.81433213e-01 -4.39233959e-01 -1.67477739e+00 5.18641114e-01 1.71417594e-01 -8.94395530e-01 -1.25636786e-01 -7.91113973e-01 -1.63060158e-01 8.08491051e-01 -1.14324176e+00 -5.89641154e-01 1.97642058e-01 4.25397336e-01 8.27814221e-01 -1.89650849e-01 1.22304404e+00 2.57715762e-01 -4.06759948e-01 4.17725295e-01 7.11706042e-01 -5.28840981e-02 9.17303860e-01 -1.28817260e+00 1.04147780e+00 6.70881331e-01 3.37062150e-01 7.44358420e-01 7.11883962e-01 -6.43836677e-01 -1.37733519e+00 -8.69538784e-01 1.55543637e+00 -8.81288052e-01 1.46827126e+00 -5.55812359e-01 -8.98989856e-01 5.28660119e-01 5.50247610e-01 -3.50552231e-01 1.02107930e+00 3.03589165e-01 -3.94865423e-01 1.47307739e-01 -9.70727086e-01 9.00662899e-01 1.26435387e+00 -7.39798486e-01 -1.36740589e+00 2.50230372e-01 1.04591417e+00 2.78652221e-01 -1.08068871e+00 -3.04778516e-01 6.63681626e-01 -4.50668693e-01 8.32520366e-01 -8.12263370e-01 2.70253599e-01 8.88983160e-02 -4.73915219e-01 -1.86547482e+00 -5.43775320e-01 -5.23046553e-01 2.42819160e-01 1.36743665e+00 2.46477470e-01 -9.46386993e-01 1.59369156e-01 2.91519403e-01 -4.39143330e-02 -3.51953983e-01 -1.42271650e+00 -8.02221060e-01 4.99170840e-01 -7.92030275e-01 6.03978574e-01 1.31932390e+00 2.35737771e-01 3.19292098e-01 4.16377522e-02 -5.24731398e-01 2.00903177e-01 -3.11011970e-01 2.38756642e-01 -1.47770882e+00 3.57531130e-01 -5.74217618e-01 -8.53248119e-01 -2.83294797e-01 2.91209847e-01 -1.27374554e+00 -2.47566119e-01 -1.62050092e+00 6.88307658e-02 1.05340712e-01 -4.16175127e-01 3.00067663e-01 -6.71508610e-02 8.56022015e-02 -1.95499539e-01 9.36407894e-02 -5.66069819e-02 7.01730549e-01 5.62389553e-01 -4.23214287e-02 -8.08572210e-03 -8.48363459e-01 -6.20022595e-01 6.33729517e-01 6.69317305e-01 -6.19695604e-01 -3.15006822e-01 -5.56199133e-01 8.54627490e-01 -1.03498363e+00 -3.07642370e-02 -4.04417664e-01 -2.52288699e-01 -1.21723607e-01 1.10455774e-01 -2.55851418e-01 1.08160295e-01 -7.18782246e-01 2.19616011e-01 5.02467215e-01 -3.09624225e-01 9.08052862e-01 2.89920956e-01 7.94367790e-01 -2.92373121e-01 -2.70276606e-01 6.63637340e-01 -6.99864626e-02 -1.49099684e+00 7.60278180e-02 -8.08188796e-01 4.74616587e-01 1.03775930e+00 -1.75661922e-01 -5.06751835e-02 -1.29350154e-02 -4.47169185e-01 1.12962294e-02 6.04703784e-01 1.14914095e+00 6.47268057e-01 -1.82508075e+00 -6.32243633e-01 1.30515993e-01 7.95373619e-01 -6.85559034e-01 4.06871140e-02 1.58323035e-01 -4.89920765e-01 1.67940751e-01 -3.32012117e-01 2.29958206e-01 -1.14421141e+00 5.75444221e-01 -2.72026151e-01 -3.47272567e-02 -9.47238445e-01 7.99373567e-01 -3.80490065e-01 -5.50482035e-01 -2.92027086e-01 -5.59987724e-01 -4.55219656e-01 1.09757984e+00 5.90914249e-01 5.21728992e-01 2.82875389e-01 -9.27766383e-01 -6.69522583e-01 5.53129077e-01 8.80612284e-02 -3.71404111e-01 1.87586331e+00 -4.65946048e-01 -3.33162874e-01 1.08366108e+00 1.70351267e+00 1.56506579e-02 -3.01486164e-01 -4.07496899e-01 6.30748510e-01 -5.17888188e-01 2.00289384e-01 -3.38703126e-01 -4.74458992e-01 4.67279017e-01 8.17470610e-01 4.50890601e-01 6.10360682e-01 2.82806363e-02 1.15713632e+00 3.11927617e-01 3.02871197e-01 -1.58883274e+00 -1.71571642e-01 8.30677032e-01 1.08247983e+00 -9.22776282e-01 6.75817952e-03 2.53815293e-01 -5.12025535e-01 1.04243219e+00 -1.51997685e-01 -7.97101036e-02 1.14694118e+00 -2.29966342e-01 7.05674365e-02 -4.52751517e-01 -6.70098722e-01 -3.06508273e-01 9.80907828e-02 7.32281148e-01 4.02149707e-01 1.67400897e-01 -9.79856312e-01 5.99522591e-01 -4.73324567e-01 -2.88190007e-01 6.18748069e-01 9.82585967e-01 -4.54092920e-01 -1.42808247e+00 -9.85688791e-02 5.92881024e-01 -4.50894475e-01 -4.74588037e-01 -6.18592381e-01 9.15200412e-01 2.06275180e-01 7.26999700e-01 2.17000276e-01 -4.22338307e-01 5.16382456e-01 5.16186595e-01 1.63759902e-01 -7.03520060e-01 -5.28556466e-01 -5.64013302e-01 3.67515147e-01 -4.20501918e-01 -4.06468779e-01 -8.89322817e-01 -9.76419389e-01 -5.43979168e-01 1.77257031e-01 1.82848983e-02 6.84427917e-01 6.20898247e-01 3.58099043e-01 3.06445748e-01 4.14985508e-01 -6.37821496e-01 -2.32013345e-01 -1.06826413e+00 -9.29489851e-01 9.50454772e-01 3.15774642e-02 -7.53230095e-01 -4.77376550e-01 8.10329691e-02]
[10.154223442077637, 8.931760787963867]
95ff1f23-75dc-4ab8-8e22-f2cf822a42a2
meshdepth-disconnected-mesh-based-deep-depth
1905.01312
null
https://arxiv.org/abs/1905.01312v2
https://arxiv.org/pdf/1905.01312v2.pdf
TriDepth: Triangular Patch-based Deep Depth Prediction
We propose a novel and efficient representation for single-view depth estimation using Convolutional Neural Networks (CNNs). Point-cloud is generally used for CNN-based 3D scene reconstruction; however it has some drawbacks: (1) it is redundant as a representation for planar surfaces, and (2) no spatial relationships between points are available (e.g, texture and surface). As a more efficient representation, we introduce a triangular-patch-cloud, which represents the surface of the 3D structure using a set of triangular patches, and propose a CNN framework for its 3D structure estimation. In our framework, we create it by separating all the faces in a 2D mesh, which are determined adaptively from the input image, and estimate depths and normals of all the faces. Using a common RGBD-dataset, we show that our representation has a better or comparable performance than the existing point-cloud-based methods, although it has much less parameters.
['Kiyoharu Aizawa', 'Ken Sakurada', 'Masaya Kaneko']
2019-05-03
null
null
null
null
['3d-scene-reconstruction']
['computer-vision']
[ 8.61441046e-02 2.87300795e-02 1.44940257e-01 -1.73576429e-01 -3.71446401e-01 -3.03342521e-01 2.52572864e-01 -1.76963180e-01 -1.13250203e-01 1.57859355e-01 -1.88932121e-01 -6.90866038e-02 1.70278281e-01 -1.20519829e+00 -9.29130733e-01 -4.96057421e-01 2.70213604e-01 5.17680228e-01 5.55921495e-01 -2.00044751e-01 4.27466869e-01 1.00796866e+00 -1.92624569e+00 2.51236618e-01 6.61654472e-01 1.65451598e+00 1.10773332e-01 3.18325281e-01 -3.19065601e-01 2.50752240e-01 -2.95889765e-01 -2.56453335e-01 3.49605411e-01 4.42900807e-02 -6.45363092e-01 3.10354561e-01 6.91585302e-01 -5.61650813e-01 -2.29170173e-01 8.87326598e-01 2.30711445e-01 -3.23455751e-01 7.51539648e-01 -9.02261794e-01 -1.20957218e-01 -5.62944710e-01 -7.69925773e-01 -6.28999770e-01 4.82930243e-01 -3.49095941e-01 5.36714733e-01 -1.27880168e+00 4.82041657e-01 1.31218803e+00 9.02079046e-01 4.32726502e-01 -8.43605042e-01 -4.21720773e-01 1.11196674e-01 -3.73073742e-02 -1.49975491e+00 -3.44126254e-01 1.19097865e+00 -3.62293094e-01 9.87306178e-01 2.36994296e-01 1.01864529e+00 6.95653796e-01 2.19700828e-01 7.17882454e-01 8.40662658e-01 -3.58762622e-01 3.74868780e-01 -2.70566642e-01 -3.58943343e-01 8.85479987e-01 -3.73884812e-02 -1.29402161e-01 -4.59519535e-01 -3.32780331e-01 1.58775735e+00 2.82967478e-01 -3.52163970e-01 -8.44598353e-01 -9.89179552e-01 6.35382533e-01 5.99064529e-01 -1.34889618e-01 -2.76189715e-01 2.56930947e-01 2.22002659e-02 -4.88827284e-03 9.19670641e-01 -1.53461620e-01 -5.58060110e-01 1.73181161e-01 -6.16079926e-01 2.62569427e-01 6.04745626e-01 1.03049684e+00 1.35682166e+00 -1.78073481e-01 6.09803677e-01 8.56771111e-01 4.24079657e-01 4.47472394e-01 -1.96293835e-02 -1.33320737e+00 5.14076173e-01 1.06109369e+00 1.63520113e-01 -1.22804201e+00 -3.76987457e-01 6.11311384e-02 -9.33055580e-01 5.21779597e-01 7.03660026e-02 3.15025747e-01 -8.96916628e-01 9.59385633e-01 5.45787752e-01 3.25939238e-01 -2.46325195e-01 8.85061204e-01 1.13720155e+00 4.70763922e-01 -8.02229881e-01 8.76988620e-02 1.20633483e+00 -7.67229497e-01 -2.88409919e-01 -2.38689542e-01 2.64454067e-01 -4.91797030e-01 8.16752195e-01 6.63230181e-01 -1.30612707e+00 -4.15286303e-01 -1.01213014e+00 -5.15166402e-01 -2.04150021e-01 1.07507378e-01 6.38124049e-01 3.44318151e-01 -1.29378796e+00 6.57451093e-01 -7.45826423e-01 -4.15946506e-02 5.30802727e-01 3.70564878e-01 -5.01138806e-01 -2.71563470e-01 -5.89185536e-01 4.48382407e-01 -1.11810848e-01 2.12715447e-01 -7.35733032e-01 -6.04207397e-01 -1.09723210e+00 6.22314736e-02 2.59310156e-01 -9.22640502e-01 1.00497901e+00 -7.80853093e-01 -1.74995685e+00 1.21340847e+00 -4.13173288e-01 2.29013756e-01 4.03080076e-01 -6.62881434e-02 2.64481962e-01 2.49291286e-01 -1.52708188e-01 5.68692088e-01 8.69516492e-01 -1.60773766e+00 -2.91674107e-01 -9.64235127e-01 3.45175922e-01 3.31879973e-01 1.37264254e-02 -2.84424901e-01 -8.59859526e-01 -2.11165816e-01 1.02822018e+00 -6.24707043e-01 -3.08821201e-01 8.35373998e-01 -3.80982071e-01 -1.42672330e-01 7.85257161e-01 -3.12930524e-01 6.32106066e-01 -2.30053592e+00 1.12086810e-01 1.73933342e-01 5.63370585e-01 1.41333025e-02 1.12930007e-01 2.51062185e-01 1.61272153e-01 1.37615070e-01 -3.63021374e-01 -8.03540885e-01 -3.56232494e-01 3.03295732e-01 3.24897729e-02 5.37339687e-01 1.50293499e-01 5.59067369e-01 -6.94656670e-01 -3.48359317e-01 5.09141088e-01 8.98248553e-01 -5.97865641e-01 2.15623856e-01 -4.33235794e-01 3.28791171e-01 -5.65333128e-01 8.05484533e-01 1.26590204e+00 -2.42098674e-01 -1.79923121e-02 -2.20033765e-01 -3.03122938e-01 3.18813562e-01 -1.30384004e+00 1.92653704e+00 -6.01076603e-01 3.17878634e-01 3.11273664e-01 -8.06668341e-01 1.21924305e+00 2.45276079e-01 5.37490606e-01 -5.25697291e-01 1.44521773e-01 2.99731553e-01 -6.88203990e-01 -4.06259894e-01 2.82922775e-01 1.01045124e-01 3.24122548e-01 2.80662566e-01 -1.94851473e-01 -5.12266278e-01 -6.83454394e-01 -2.68426865e-01 7.36890674e-01 2.66852587e-01 1.65409833e-01 4.04464733e-03 4.90348607e-01 -1.67787626e-01 7.21503437e-01 1.33713618e-01 2.98877090e-01 1.24553871e+00 5.60747445e-01 -9.47438478e-01 -8.89737964e-01 -9.22285438e-01 -2.98984945e-01 1.53975978e-01 4.99876380e-01 -3.27342123e-01 -5.75168312e-01 -3.05153131e-01 1.19255342e-01 9.62791033e-03 -5.72075307e-01 2.97467232e-01 -6.49058998e-01 -2.13299900e-01 -5.92772625e-02 6.33928061e-01 6.73783600e-01 -6.50906861e-01 -7.81612217e-01 -5.96890263e-02 -2.09424779e-01 -1.17041802e+00 -1.12566799e-01 3.55171859e-02 -1.53491735e+00 -1.38634384e+00 -6.26856208e-01 -7.48896003e-01 8.86866331e-01 7.29630589e-01 1.38728023e+00 3.53903323e-01 9.77347121e-02 4.41751361e-01 -1.77965343e-01 -5.00619590e-01 2.28130534e-01 -3.06520849e-01 -2.20507249e-01 6.12846315e-02 2.04175994e-01 -9.15780067e-01 -9.35539961e-01 4.52740222e-01 -7.63149559e-01 3.23718935e-01 1.87462091e-01 4.80328709e-01 1.18086791e+00 -2.58462457e-03 -2.74698734e-01 -6.57959282e-01 -9.40717161e-02 -2.72321284e-01 -8.28010976e-01 -1.03009187e-01 -1.59809053e-01 -4.07837063e-01 4.14398491e-01 -1.24930568e-01 -7.35129535e-01 2.71045297e-01 -2.67094284e-01 -9.07000184e-01 -3.36857319e-01 2.37667337e-01 -3.12694758e-01 -4.99355525e-01 3.18277895e-01 2.64757156e-01 6.40547052e-02 -8.80865574e-01 2.19009854e-02 5.12569964e-01 1.56438828e-01 -4.21448052e-01 5.43895662e-01 1.17378294e+00 2.56550997e-01 -1.05204105e+00 -6.38412297e-01 -3.96949977e-01 -1.12417817e+00 -3.01821977e-01 7.92764962e-01 -1.09023380e+00 -9.88202989e-01 7.82971442e-01 -1.81726873e+00 -2.95838684e-01 1.14633463e-01 3.30743581e-01 -7.66806126e-01 3.40023279e-01 -5.90078413e-01 -8.31400096e-01 -2.54426748e-01 -1.19130206e+00 1.66444159e+00 -8.62763673e-02 3.55515420e-01 -9.79319096e-01 -4.05982845e-02 4.31696922e-01 -7.48881255e-04 5.40421128e-01 1.04164696e+00 4.28743750e-01 -9.68258500e-01 -1.63103536e-01 -1.99463338e-01 3.05194944e-01 1.28188264e-02 2.72522539e-01 -1.30579937e+00 -1.33390665e-01 2.61404663e-01 -1.20878495e-01 6.10499859e-01 5.59943378e-01 1.68015385e+00 -6.66638389e-02 -4.22048718e-01 1.27141583e+00 1.74720550e+00 1.04385823e-01 9.89641011e-01 3.79967153e-01 8.92686903e-01 6.43318772e-01 2.69239783e-01 4.66462344e-01 6.92352355e-01 7.57875741e-01 1.12368357e+00 -2.73959488e-01 4.73072678e-02 -2.34284461e-01 5.24123013e-02 9.61170435e-01 -4.41958189e-01 -1.06081143e-01 -7.27684200e-01 2.60405391e-01 -1.52529156e+00 -2.41633549e-01 -5.15054226e-01 2.33021522e+00 2.61355370e-01 -7.07826018e-02 5.49883470e-02 2.51659244e-01 3.53904456e-01 8.56055841e-02 -5.92916012e-01 -2.90759295e-01 1.70362014e-02 3.45633537e-01 2.77123868e-01 4.17952061e-01 -8.08554769e-01 8.44933033e-01 6.82155228e+00 3.86552781e-01 -1.23277211e+00 -1.36440352e-01 4.29242730e-01 3.93737145e-02 -4.72149521e-01 -1.76195651e-01 -5.82150221e-01 1.60170928e-01 1.54587120e-01 3.87126923e-01 2.81757653e-01 8.57222736e-01 1.36733027e-02 -1.54914543e-01 -1.10680985e+00 1.41439664e+00 2.72533685e-01 -1.40327203e+00 1.06589802e-01 2.19738379e-01 6.56539500e-01 2.37059861e-01 -7.16372207e-02 -3.26455265e-01 -2.22726166e-01 -1.01153278e+00 8.53193343e-01 4.50664759e-01 1.14803696e+00 -7.73275435e-01 6.65000618e-01 6.20651245e-01 -1.28998983e+00 3.04605722e-01 -8.85392189e-01 -2.65782952e-01 -1.23836048e-01 7.97454238e-01 -2.13206664e-01 6.27299666e-01 9.20157313e-01 8.16556633e-01 -2.55003959e-01 9.68389571e-01 -9.76106897e-02 -7.23119229e-02 -4.08981860e-01 2.75931627e-01 1.13438377e-02 -5.51106215e-01 1.71306849e-01 5.17540276e-01 5.35386562e-01 2.15436041e-01 -1.07815988e-01 1.14079022e+00 -8.73606950e-02 3.49410735e-02 -9.91342247e-01 7.36120820e-01 3.74185145e-01 1.05061412e+00 -8.00075710e-01 -1.65341631e-01 -7.04599082e-01 9.42780614e-01 5.42714894e-01 2.70857215e-01 -2.87837029e-01 -2.23397955e-01 7.55679369e-01 5.15958607e-01 3.60306203e-01 -4.70442623e-01 -3.30554962e-01 -1.24128890e+00 3.30788404e-01 -4.86637741e-01 -1.09767340e-01 -1.23842025e+00 -1.15065575e+00 5.57580829e-01 -1.81346267e-01 -1.53319049e+00 2.07494050e-01 -1.00518656e+00 -6.06040418e-01 9.43294108e-01 -1.75875819e+00 -1.00625467e+00 -7.69328058e-01 8.26801538e-01 3.46556842e-01 3.68642300e-01 1.01899517e+00 1.54537126e-01 -2.46851817e-01 1.10375114e-01 -3.54216099e-02 1.34551182e-01 6.85832053e-02 -9.57493663e-01 6.82573915e-01 2.27918312e-01 -1.15528002e-01 3.95143241e-01 -1.17305815e-01 -5.28128982e-01 -1.68059933e+00 -8.46145332e-01 4.81737554e-01 -4.26176578e-01 -9.72581431e-02 -5.75721920e-01 -9.37984288e-01 6.62800133e-01 -3.35528791e-01 2.26156428e-01 4.11288500e-01 -2.63054557e-02 -3.92470390e-01 -1.05425544e-01 -1.27565920e+00 3.83418947e-01 1.19163358e+00 -6.13263786e-01 -1.80219501e-01 1.84473306e-01 6.40878856e-01 -8.56303453e-01 -8.40005040e-01 5.71768701e-01 6.46224976e-01 -1.62275052e+00 1.18462360e+00 1.57342255e-01 5.90799809e-01 -3.52409363e-01 -2.06676617e-01 -1.20370793e+00 -2.60389835e-01 -2.26313725e-01 -2.48095021e-01 5.32035589e-01 -9.48003456e-02 -6.81478202e-01 1.24819684e+00 3.22519392e-01 -4.40518737e-01 -1.18712354e+00 -1.12332952e+00 -5.19040763e-01 1.39057236e-02 -6.14715517e-01 8.54916930e-01 8.28623891e-01 -5.45591772e-01 1.27039418e-01 2.60720737e-02 3.88201505e-01 6.49600029e-01 2.80261129e-01 9.33036625e-01 -1.68081403e+00 1.33751288e-01 -2.26725727e-01 -4.62652713e-01 -1.62281680e+00 -4.77258004e-02 -5.67649901e-01 -4.92551848e-02 -1.81539142e+00 -1.69924408e-01 -7.98365235e-01 2.09521785e-01 1.61448076e-01 3.50153327e-01 2.55289704e-01 -2.92925537e-01 3.21807295e-01 -2.17934236e-01 7.33593047e-01 1.65783632e+00 1.38248742e-01 -5.75561151e-02 1.51568234e-01 -2.77972639e-01 1.16725504e+00 3.73939902e-01 -1.76507518e-01 -2.21771628e-01 -9.90262568e-01 4.63698089e-01 3.13994467e-01 5.28185487e-01 -9.15603638e-01 5.33691421e-02 -6.67158738e-02 5.69908559e-01 -1.21725440e+00 9.75335181e-01 -9.78924811e-01 1.95287123e-01 1.74774483e-01 4.32317466e-01 3.75450291e-02 5.43622784e-02 5.01501024e-01 -1.77203119e-01 -2.43042693e-01 6.76743329e-01 -6.99843645e-01 -2.31995791e-01 7.96070337e-01 -2.41046529e-02 -4.20939773e-01 7.11588800e-01 -6.90319002e-01 5.61746806e-02 -2.29515031e-01 -2.27488384e-01 -1.22874841e-01 1.18320465e+00 -7.58837676e-03 1.23609245e+00 -1.49272597e+00 -3.43353629e-01 5.19495964e-01 2.61061013e-01 1.07667041e+00 2.16079265e-01 4.08431858e-01 -1.05016255e+00 1.60751671e-01 -3.76031362e-03 -9.98461902e-01 -8.58036578e-01 3.71872813e-01 6.26310945e-01 3.59488159e-01 -9.81897950e-01 7.89038897e-01 5.47202826e-01 -7.11472631e-01 2.98146933e-01 -6.19290888e-01 -2.02707142e-01 -3.84042531e-01 4.30192292e-01 2.95558214e-01 3.02366555e-01 -6.77326381e-01 -3.16234946e-01 1.42754328e+00 5.06530821e-01 2.19278961e-01 1.58508420e+00 9.32276025e-02 -5.85409701e-01 5.57432055e-01 1.19243109e+00 -1.58066392e-01 -1.29587233e+00 -2.36655056e-01 -4.87145483e-01 -8.42624128e-01 2.31787041e-01 -1.99243966e-02 -1.29779720e+00 1.36037993e+00 2.56689638e-01 8.68765935e-02 9.87536967e-01 3.06920670e-02 9.81872380e-01 2.13904619e-01 8.39575648e-01 -7.32346296e-01 1.56949937e-01 6.91697240e-01 1.01106429e+00 -1.00137031e+00 1.30061522e-01 -1.00376379e+00 4.20486294e-02 1.51230526e+00 6.94475770e-01 -2.60748684e-01 8.85757923e-01 3.48600149e-02 -2.05475348e-03 -7.80485213e-01 -4.25336480e-01 -4.55760062e-02 1.83353484e-01 8.22928071e-01 2.18378618e-01 -1.03913903e-01 3.42847347e-01 1.29161179e-01 -1.42089084e-01 -1.85813397e-01 3.51460516e-01 7.66636729e-01 -3.10965180e-01 -7.46891737e-01 -5.33560097e-01 2.75904059e-01 -9.57711488e-02 4.90892865e-02 -3.84766638e-01 6.74202800e-01 3.45572412e-01 5.41793227e-01 5.66220760e-01 -5.41274190e-01 5.91031075e-01 -3.12351435e-01 7.54871190e-01 -9.07687664e-01 -7.31515884e-02 1.63242385e-01 -2.65801966e-01 -8.90554070e-01 -5.99857032e-01 -3.08465242e-01 -9.87060249e-01 -3.53503793e-01 -3.39232087e-01 -3.40518087e-01 9.65750694e-01 7.40451992e-01 5.15459955e-01 3.57038788e-02 1.01612055e+00 -1.47082472e+00 4.55414094e-02 -6.26175344e-01 -7.91093647e-01 1.06407396e-01 3.22957188e-01 -9.73171949e-01 -4.82039750e-01 -3.20007265e-01]
[8.749921798706055, -3.046511650085449]
fb2a077a-5c43-4405-8631-8f052943f492
co-saliency-detection-for-rgbd-images-based
1710.05172
null
http://arxiv.org/abs/1710.05172v1
http://arxiv.org/pdf/1710.05172v1.pdf
Co-saliency Detection for RGBD Images Based on Multi-constraint Feature Matching and Cross Label Propagation
Co-saliency detection aims at extracting the common salient regions from an image group containing two or more relevant images. It is a newly emerging topic in computer vision community. Different from the most existing co-saliency methods focusing on RGB images, this paper proposes a novel co-saliency detection model for RGBD images, which utilizes the depth information to enhance identification of co-saliency. First, the intra saliency map for each image is generated by the single image saliency model, while the inter saliency map is calculated based on the multi-constraint feature matching, which represents the constraint relationship among multiple images. Then, the optimization scheme, namely Cross Label Propagation (CLP), is used to refine the intra and inter saliency maps in a cross way. Finally, all the original and optimized saliency maps are integrated to generate the final co-saliency result. The proposed method introduces the depth information and multi-constraint feature matching to improve the performance of co-saliency detection. Moreover, the proposed method can effectively exploit any existing single image saliency model to work well in co-saliency scenarios. Experiments on two RGBD co-saliency datasets demonstrate the effectiveness of our proposed model.
['Qingming Huang', 'Huazhu Fu', 'Runmin Cong', 'Xiaochun Cao', 'Jianjun Lei', 'Chunping Hou']
2017-10-14
null
null
null
null
['co-saliency-detection']
['computer-vision']
[ 5.05433202e-01 -1.53068304e-01 -8.20734873e-02 -1.02361232e-01 -5.00595272e-01 6.74022734e-02 2.29021832e-01 3.38854104e-01 -4.05917943e-01 2.86798924e-01 1.10636078e-01 1.60368726e-01 -1.73566222e-01 -6.72008872e-01 -4.64376479e-01 -6.67290807e-01 3.70338172e-01 -3.72508764e-01 1.13806307e+00 -2.05015242e-01 6.55326247e-01 1.12911254e-01 -1.87029362e+00 -1.13623496e-02 1.08093083e+00 1.15374053e+00 9.79170561e-01 1.12547964e-01 -1.64632514e-01 6.41085625e-01 -3.85883778e-01 2.71128297e-01 1.15122765e-01 -3.96825373e-01 -6.46311224e-01 2.47770578e-01 -6.82772622e-02 -3.60079221e-02 2.18673185e-01 1.36081457e+00 5.07746279e-01 1.12892017e-02 5.65721840e-02 -1.43932271e+00 -3.96001399e-01 2.97218084e-01 -9.62107718e-01 4.58665520e-01 3.09113860e-01 -1.17792860e-01 7.22014964e-01 -9.83042061e-01 3.38605195e-01 1.14326942e+00 1.86315775e-01 1.28932223e-01 -6.34906292e-01 -7.46627569e-01 2.90826440e-01 4.92771000e-01 -1.49212372e+00 -1.60459185e-03 1.33731854e+00 3.70221888e-03 3.93351138e-01 3.10643286e-01 9.93381679e-01 -5.71066663e-02 -4.69336696e-02 1.19169700e+00 1.26705217e+00 -5.23439646e-01 3.67321819e-02 2.68897742e-01 -8.49420950e-03 6.74434006e-01 1.18192993e-01 -1.27742305e-01 -6.28082812e-01 1.92641601e-01 8.06978106e-01 2.85008162e-01 -3.54272962e-01 -6.17468178e-01 -1.44427669e+00 4.84563202e-01 9.36876833e-01 4.39943701e-01 -3.92844915e-01 -2.59883821e-01 7.41000548e-02 -3.62976015e-01 2.42040277e-01 2.04625368e-01 -1.26028240e-01 2.96150625e-01 -1.19826853e+00 1.31623447e-01 -1.15448805e-02 8.97733450e-01 1.38642645e+00 -2.46223465e-01 -5.18697649e-02 6.80528522e-01 4.98399377e-01 5.39020419e-01 7.08794057e-01 -6.57876968e-01 3.67037535e-01 1.19624162e+00 1.64585009e-01 -1.48598146e+00 -5.23709297e-01 -4.21333730e-01 -4.31121111e-01 1.54213816e-01 -2.08786473e-01 1.91110894e-01 -8.59544575e-01 1.32338452e+00 7.32081652e-01 3.02102387e-01 8.36555660e-02 1.45505035e+00 1.10857499e+00 4.85323071e-01 7.18655065e-02 -1.16094925e-01 1.19307697e+00 -1.23504496e+00 -6.08152390e-01 -4.20221299e-01 9.88094583e-02 -1.01866317e+00 8.64662766e-01 -4.42011729e-02 -1.01678813e+00 -6.45076632e-01 -1.25241256e+00 -1.47195905e-01 -4.12247449e-01 1.47273317e-01 5.72616100e-01 2.58413255e-01 -9.90200579e-01 6.38569891e-02 -5.17668426e-01 -3.52537215e-01 3.86139780e-01 4.69658703e-01 -6.38145581e-02 -4.67954427e-02 -1.26798463e+00 9.93532002e-01 1.01072240e+00 1.66426733e-01 -6.11447275e-01 -5.41241646e-01 -9.13826108e-01 -5.83746620e-02 4.76533532e-01 -2.07195207e-01 8.85066628e-01 -1.23678601e+00 -1.05863798e+00 7.38506734e-01 -4.44861233e-01 -2.15402711e-02 1.01573423e-01 1.56677216e-02 -3.24888468e-01 4.23025638e-01 4.28522974e-01 9.58144426e-01 7.13347137e-01 -1.39104426e+00 -1.26501966e+00 -3.14626992e-01 5.85878715e-02 8.86236608e-01 -1.76389664e-01 1.63829878e-01 -7.49033630e-01 -3.30432445e-01 7.41680086e-01 -5.98906577e-01 -1.08394116e-01 -3.25346857e-01 -5.35001159e-01 -2.18188941e-01 1.12469625e+00 -4.60592210e-01 1.25458276e+00 -2.09111142e+00 3.27482820e-01 2.60023832e-01 2.49558195e-01 1.21240400e-01 2.19521657e-01 -1.48409307e-01 -6.51733801e-02 -1.86923638e-01 -3.98294210e-01 -1.47776827e-01 -4.13024932e-01 -1.73461676e-01 4.02956195e-02 2.98710227e-01 5.06096601e-01 8.57938290e-01 -1.17534244e+00 -1.19117463e+00 6.96890891e-01 1.97462648e-01 1.08029116e-02 6.45622686e-02 6.27637282e-02 3.69087011e-01 -7.09874570e-01 8.74074757e-01 9.36346233e-01 -1.40862882e-01 -3.13972116e-01 -3.11155289e-01 -4.99443352e-01 -2.08002571e-02 -1.31978393e+00 1.79337919e+00 -1.73307210e-01 2.22510859e-01 -1.88454330e-01 -7.77602732e-01 1.23990834e+00 -1.20099515e-01 6.74804032e-01 -9.84900355e-01 2.12768823e-01 4.88782525e-01 -2.11927190e-01 -3.08755517e-01 7.53012180e-01 8.26728567e-02 -4.66425233e-02 4.51727062e-01 -9.58596244e-02 -1.67398334e-01 1.01281159e-01 1.77721262e-01 3.59153062e-01 2.09752351e-01 1.95775241e-01 -2.85297334e-01 9.39445555e-01 2.78981000e-01 7.25787878e-01 2.78134465e-01 -4.67428833e-01 8.71388972e-01 -1.16451249e-01 -1.20608643e-01 -7.47480929e-01 -9.27153289e-01 5.24750240e-02 7.71599650e-01 1.36230946e+00 -1.54426411e-01 -7.47144520e-01 -5.61349809e-01 -2.16036320e-01 3.02961856e-01 -5.80227911e-01 -3.23144943e-01 -3.35380226e-01 -6.01999581e-01 -1.67041540e-01 1.87903464e-01 1.22514367e+00 -1.37709665e+00 -1.14940596e+00 1.36653781e-01 -4.71175045e-01 -7.44857311e-01 -5.50259888e-01 -2.06709877e-01 -7.12771058e-01 -1.09534514e+00 -1.06759906e+00 -1.32862449e+00 7.64202356e-01 1.10553873e+00 5.86962521e-01 4.96700495e-01 -1.99571997e-01 1.21947326e-01 -4.70269561e-01 -4.94966030e-01 1.93554297e-01 1.26899078e-01 -3.22824329e-01 2.08995268e-01 5.70729494e-01 -3.19157690e-01 -8.02056193e-01 4.12678361e-01 -9.86001968e-01 6.34440362e-01 7.88882613e-01 5.09422898e-01 8.33579004e-01 1.87793702e-01 4.69719619e-01 -2.09407821e-01 3.10040593e-01 -4.32845950e-01 -4.22671497e-01 4.43975359e-01 -5.48245966e-01 -1.30057976e-01 4.03731018e-02 -1.76964596e-01 -1.06235015e+00 2.21361846e-01 3.59861732e-01 -3.94453317e-01 7.36457109e-02 4.52638030e-01 -3.80176365e-01 -2.47011572e-01 -1.73756108e-02 7.82882512e-01 -1.76333785e-01 -2.81438380e-01 2.24987477e-01 6.95049882e-01 7.14388013e-01 9.22219406e-05 6.20182812e-01 3.99880260e-01 -1.31150544e-01 -4.34140563e-01 -9.18725848e-01 -7.78491616e-01 -7.72494614e-01 -4.87279266e-01 1.02705765e+00 -9.10795033e-01 -4.58268195e-01 7.00296879e-01 -1.13679922e+00 2.89939284e-01 7.67355040e-02 4.63388294e-01 -3.47497016e-01 3.57217640e-01 -1.71938296e-02 -7.72049427e-01 -4.11036491e-01 -1.42824757e+00 1.23447955e+00 1.06999004e+00 2.53629684e-01 -7.00451136e-01 -3.10158670e-01 1.08295217e-01 3.25513393e-01 2.53001720e-01 4.66787636e-01 -1.10086851e-01 -7.74132490e-01 9.12879705e-02 -7.26866424e-01 -8.92647915e-03 4.83470351e-01 -2.38138616e-01 -6.31731033e-01 7.90246949e-02 -1.62209738e-02 -4.67633381e-02 8.36972713e-01 2.51514405e-01 9.84648883e-01 6.95372671e-02 -5.41170776e-01 2.55253971e-01 1.63357329e+00 2.53215373e-01 4.96835858e-01 7.91879535e-01 9.01861310e-01 6.12299204e-01 1.29635763e+00 2.54740804e-01 6.51609778e-01 5.83045840e-01 6.55696094e-01 -6.49329543e-01 -1.29540712e-01 -3.23782951e-01 9.36163310e-03 8.41177762e-01 1.91389814e-01 5.62650859e-01 -7.79357374e-01 8.65994036e-01 -2.03324103e+00 -7.15684891e-01 -2.04492554e-01 2.10118079e+00 8.52572441e-01 2.23765403e-01 1.42682716e-01 3.92503858e-01 1.21303701e+00 1.03479475e-01 -7.18760729e-01 -6.30640313e-02 -3.68226469e-01 -4.65711616e-02 4.37623739e-01 3.25764418e-01 -1.21721470e+00 1.02207124e+00 5.41093063e+00 9.59883094e-01 -1.32486117e+00 2.30308503e-01 3.84454101e-01 5.93968108e-02 -4.72889900e-01 2.15317875e-01 -6.93300307e-01 8.35053980e-01 -2.55756453e-02 -4.26650017e-01 2.02963442e-01 8.09633195e-01 1.55691579e-01 -9.68226492e-01 -3.66029531e-01 1.04659486e+00 3.72548699e-01 -1.01294792e+00 -5.69533557e-02 -3.22674483e-01 9.32156980e-01 -2.60096133e-01 2.64813125e-01 -2.29724318e-01 -5.63152954e-02 -6.16303682e-01 1.00352514e+00 4.76551682e-01 3.78197461e-01 -9.67054486e-01 8.22444677e-01 3.33175629e-01 -1.48879886e+00 -7.73790404e-02 -4.49966222e-01 3.75677310e-02 1.46245152e-01 7.39374757e-01 -6.93526566e-01 8.37420881e-01 9.06170428e-01 1.09117472e+00 -1.05177486e+00 1.43566537e+00 -3.68326515e-01 -3.26598585e-02 -3.51232976e-01 -2.65038937e-01 3.65164906e-01 -2.10667938e-01 5.13393462e-01 8.00862551e-01 1.83661267e-01 -5.82794808e-02 2.96807617e-01 9.43985701e-01 4.84018654e-01 2.70108312e-01 -1.85323790e-01 5.69404542e-01 5.88651597e-01 1.36989486e+00 -1.10405076e+00 -4.46626216e-01 -2.32516676e-01 1.04133379e+00 6.03891537e-02 2.26692915e-01 -8.53238463e-01 -7.33545482e-01 -2.10659951e-02 -1.52602881e-01 3.39462876e-01 -4.01403606e-02 -4.38002467e-01 -9.32584822e-01 1.20013148e-01 -2.90588975e-01 2.05036327e-01 -1.17632270e+00 -5.51097572e-01 3.62585962e-01 1.14515610e-01 -1.52485847e+00 2.80143738e-01 -2.92156041e-02 -7.83553779e-01 1.15600896e+00 -2.06122851e+00 -1.38045645e+00 -6.43848002e-01 6.98900580e-01 4.92102653e-01 2.50980377e-01 2.45003894e-01 1.06114320e-01 -6.28613830e-01 1.70857251e-01 -2.90298373e-01 -1.54975131e-01 4.05180186e-01 -1.03694344e+00 -1.24332607e-01 1.11917913e+00 -2.85189122e-01 5.22652566e-01 3.93669337e-01 -7.78410017e-01 -8.53040814e-01 -1.05388021e+00 7.54646122e-01 1.55211419e-01 1.74943149e-01 -1.26842149e-02 -8.20857406e-01 1.65904388e-01 2.39002127e-02 -6.23046122e-02 1.75840914e-01 -4.90974993e-01 1.56599775e-01 -1.86531171e-01 -1.09406996e+00 5.54351389e-01 7.72191286e-01 -3.66225898e-01 -7.48285592e-01 3.87485651e-03 1.10539830e+00 -5.18300772e-01 -5.51351845e-01 6.11980498e-01 3.97640318e-01 -1.10078716e+00 1.01846075e+00 1.82550028e-01 5.76599181e-01 -1.05312729e+00 1.27717867e-01 -9.85119402e-01 -2.43885934e-01 -5.12817055e-02 2.53448904e-01 1.39877141e+00 1.05890311e-01 -3.39829594e-01 5.57038903e-01 3.45896214e-01 -1.94204867e-01 -7.92560756e-01 -8.69395316e-01 -3.88092667e-01 -6.49358869e-01 -4.41798270e-02 8.87187243e-01 7.08965361e-01 5.41159511e-02 4.79106233e-02 -1.03959158e-01 2.89817661e-01 7.46068835e-01 6.55970991e-01 4.34553623e-01 -1.03692794e+00 2.62787640e-01 -6.16029263e-01 -5.01921952e-01 -8.65238547e-01 -1.47130921e-01 -9.02632475e-01 3.69891763e-01 -1.67418778e+00 4.12097722e-01 -6.78468645e-01 -8.00087571e-01 5.10484874e-01 -7.47387171e-01 5.67533672e-01 4.08642173e-01 3.73819619e-01 -1.04996109e+00 6.60828829e-01 1.48942590e+00 -2.36332476e-01 -4.63417828e-01 -2.91784883e-01 -8.66725981e-01 5.93038678e-01 8.36716950e-01 -3.10542524e-01 -3.44368279e-01 -3.08079928e-01 -2.78777689e-01 -1.86479285e-01 4.75386769e-01 -1.29831696e+00 5.29554129e-01 -3.00154716e-01 4.55077589e-01 -9.70205605e-01 1.63397133e-01 -7.48223722e-01 -2.46866569e-01 4.96535659e-01 -5.69257773e-02 -1.18539833e-01 2.02758089e-01 4.19701815e-01 -4.89497721e-01 -1.97279707e-01 6.98532224e-01 -2.70735353e-01 -1.39002335e+00 1.55922219e-01 1.45555243e-01 -1.30507082e-01 1.37395763e+00 -5.71990728e-01 -1.78294763e-01 2.91882753e-02 -3.52420509e-01 5.76991260e-01 6.40530348e-01 6.35977864e-01 1.04836214e+00 -1.55636418e+00 -3.64712447e-01 2.25526422e-01 4.27547872e-01 3.53585094e-01 3.98415297e-01 1.05809689e+00 -3.49524707e-01 2.86692142e-01 -5.21169722e-01 -9.02540088e-01 -1.33582282e+00 7.22147524e-01 1.26331300e-01 6.08085580e-02 -1.29078090e-01 7.06926227e-01 2.38176882e-01 -8.10005590e-02 -1.15279160e-01 -1.58462062e-01 -6.71833873e-01 -1.44474432e-01 5.38688540e-01 2.69023597e-01 -1.17923968e-01 -1.13971233e+00 -7.13508427e-01 1.08925760e+00 8.80462229e-02 -2.14286432e-01 9.42816973e-01 -5.79171896e-01 -4.65540558e-01 5.05725503e-01 1.18867469e+00 -2.07953006e-01 -1.23780787e+00 -4.61542487e-01 -5.60356937e-02 -6.21465981e-01 2.11366087e-01 -6.69581175e-01 -1.21207774e+00 8.95622313e-01 7.44338512e-01 -8.71382207e-02 1.62406456e+00 2.36279517e-02 9.24490511e-01 -4.02267843e-01 6.23063385e-01 -1.04426718e+00 2.44162753e-01 1.23597287e-01 7.42776752e-01 -1.29093182e+00 2.40326196e-01 -6.56736493e-01 -7.93705642e-01 7.49580503e-01 8.95888984e-01 -2.06489533e-01 5.98345041e-01 -3.56537580e-01 -1.12444222e-01 -1.59400880e-01 1.28113717e-01 -6.99291825e-01 3.88423413e-01 4.64263052e-01 -1.77958265e-01 -8.86358321e-02 -5.85654855e-01 5.49671173e-01 -5.61156720e-02 9.79573131e-02 4.90912050e-01 1.15429854e+00 -9.23540652e-01 -8.32363725e-01 -5.76499879e-01 1.90292716e-01 -1.62366256e-02 -4.56664711e-02 -2.94573933e-01 4.53096658e-01 5.49584031e-01 1.24688697e+00 1.45285442e-01 -6.75269544e-01 1.09263822e-01 -3.33891720e-01 1.97687373e-01 -6.45144463e-01 -3.84719312e-01 6.84540495e-02 -7.05330908e-01 -4.41574633e-01 -1.04695988e+00 -5.23352802e-01 -1.69068456e+00 2.88387120e-01 -7.09077775e-01 2.06166714e-01 6.45246387e-01 9.71088648e-01 3.28614563e-01 4.63290095e-01 9.48539555e-01 -1.03976655e+00 3.27145487e-01 -7.47746170e-01 -5.73602259e-01 4.73472178e-01 3.86904150e-01 -1.04051280e+00 -2.00454339e-01 -8.87501091e-02]
[9.783515930175781, -0.5411754250526428]
f7da1c4c-58d2-4d1f-869a-c5d78e67749e
a-regularized-framework-for-sparse-and
1705.07704
null
http://arxiv.org/abs/1705.07704v3
http://arxiv.org/pdf/1705.07704v3.pdf
A Regularized Framework for Sparse and Structured Neural Attention
Modern neural networks are often augmented with an attention mechanism, which tells the network where to focus within the input. We propose in this paper a new framework for sparse and structured attention, building upon a smoothed max operator. We show that the gradient of this operator defines a mapping from real values to probabilities, suitable as an attention mechanism. Our framework includes softmax and a slight generalization of the recently-proposed sparsemax as special cases. However, we also show how our framework can incorporate modern structured penalties, resulting in more interpretable attention mechanisms, that focus on entire segments or groups of an input. We derive efficient algorithms to compute the forward and backward passes of our attention mechanisms, enabling their use in a neural network trained with backpropagation. To showcase their potential as a drop-in replacement for existing ones, we evaluate our attention mechanisms on three large-scale tasks: textual entailment, machine translation, and sentence summarization. Our attention mechanisms improve interpretability without sacrificing performance; notably, on textual entailment and summarization, we outperform the standard attention mechanisms based on softmax and sparsemax.
['Mathieu Blondel', 'Vlad Niculae']
2017-05-22
a-regularized-framework-for-sparse-and-1
http://papers.nips.cc/paper/6926-a-regularized-framework-for-sparse-and-structured-neural-attention
http://papers.nips.cc/paper/6926-a-regularized-framework-for-sparse-and-structured-neural-attention.pdf
neurips-2017-12
['abstractive-sentence-summarization']
['natural-language-processing']
[ 7.12793350e-01 6.76521480e-01 -2.73515344e-01 -6.38099134e-01 -6.69109941e-01 -4.45772737e-01 6.54370964e-01 2.45435297e-01 -6.67016983e-01 7.55451202e-01 6.31551981e-01 -4.89487469e-01 1.32207617e-01 -5.55580080e-01 -1.09911203e+00 -3.77341390e-01 9.08711553e-02 3.77445042e-01 -1.09077245e-01 -3.15147847e-01 3.44220549e-01 3.83707970e-01 -1.16123545e+00 5.43793142e-01 8.96663487e-01 8.60052943e-01 2.33426660e-01 8.90723825e-01 -1.73994705e-01 1.05308735e+00 -7.25980341e-01 -7.09226549e-01 -5.15241250e-02 -3.47719520e-01 -1.20834279e+00 4.63108532e-02 4.61487591e-01 -3.22926193e-01 -3.08383673e-01 8.37994754e-01 2.96813816e-01 2.60143191e-01 6.27816677e-01 -7.97611952e-01 -1.26084483e+00 1.15152788e+00 -4.34745967e-01 2.84124255e-01 3.40687215e-01 -2.53788382e-02 1.54215610e+00 -9.64999497e-01 4.95856375e-01 1.26300299e+00 8.03551614e-01 5.13736963e-01 -1.27951479e+00 2.07047433e-01 4.23834801e-01 2.68592358e-01 -8.29814255e-01 -6.21152997e-01 6.01702154e-01 -1.87196702e-01 1.53896451e+00 6.15918636e-01 2.23093703e-01 1.07475412e+00 4.09855172e-02 1.19664443e+00 3.60446751e-01 -5.33502579e-01 1.41008973e-01 6.38451725e-02 3.25485677e-01 5.63916564e-01 6.41676225e-03 -4.13864911e-01 -3.16789597e-01 -3.06873564e-02 5.54397523e-01 9.32577346e-03 -4.43377227e-01 1.52191892e-01 -1.15387869e+00 9.74624932e-01 7.85628438e-01 1.96603820e-01 -4.91944939e-01 4.13196713e-01 4.36730385e-01 3.37935090e-01 6.74130261e-01 6.20359004e-01 -5.60202241e-01 -2.63006892e-02 -9.95032728e-01 3.56379710e-02 9.47843254e-01 8.75116050e-01 7.61489391e-01 6.76889196e-02 -6.41563118e-01 9.29285169e-01 7.69898146e-02 5.09110726e-02 5.06366432e-01 -1.00663841e+00 6.78120315e-01 4.79516476e-01 -8.82869288e-02 -7.54750907e-01 -3.50872815e-01 -5.11110783e-01 -8.86848867e-01 -1.31275326e-01 1.77834556e-01 -3.74917924e-01 -8.93109500e-01 1.75992060e+00 -1.35735527e-01 -4.73788492e-02 -4.79322113e-02 8.13982368e-01 6.25856519e-01 7.64673352e-01 -3.72884162e-02 -1.60621375e-01 1.09676123e+00 -1.31294215e+00 -6.75728142e-01 -5.15798986e-01 7.21156418e-01 -5.65270245e-01 1.27503455e+00 2.20865265e-01 -1.66767502e+00 -4.51517463e-01 -8.68089080e-01 -5.80287218e-01 -2.45048329e-01 2.19848469e-01 7.56885767e-01 2.25109562e-01 -1.25689912e+00 1.10294771e+00 -8.35975707e-01 -2.65838206e-01 3.80048484e-01 5.06885409e-01 -1.57491788e-01 3.45953435e-01 -1.17933214e+00 1.23457670e+00 3.26779544e-01 4.30414617e-01 -3.12924206e-01 -4.58519846e-01 -1.22625101e+00 5.81465662e-01 2.43700475e-01 -1.10883152e+00 1.64251459e+00 -1.35282505e+00 -1.71181440e+00 9.14004087e-01 -4.09702718e-01 -9.67507422e-01 3.15295547e-01 -4.75708902e-01 1.56760454e-01 6.64653555e-02 -1.46823615e-01 6.12382948e-01 7.89271772e-01 -8.10967267e-01 -2.59604931e-01 -8.12965184e-02 4.39266533e-01 1.83922186e-01 -4.37342405e-01 2.65409410e-01 -3.81924689e-01 -8.04937124e-01 -3.16951603e-01 -6.83539033e-01 -4.20838773e-01 -1.59267396e-01 -8.45966280e-01 -3.27447683e-01 3.45499694e-01 -7.57120848e-01 1.43207347e+00 -1.85208619e+00 7.21200764e-01 -9.43322182e-02 7.91785941e-02 3.49035949e-01 -1.87259138e-01 4.48408484e-01 -1.71715051e-01 2.23191902e-01 -9.51062024e-01 -8.02564740e-01 3.64261836e-01 5.05395532e-01 -2.79591948e-01 3.03137124e-01 8.51421893e-01 1.27238977e+00 -6.60755634e-01 -1.67395249e-01 -2.46831272e-02 3.87225062e-01 -8.18285346e-01 1.13899544e-01 -3.44046950e-01 -5.55909500e-02 -1.36139065e-01 1.90318003e-01 3.05117249e-01 -4.94998723e-01 -9.86785814e-02 4.41839872e-03 -1.49663659e-02 6.67081296e-01 -7.76321232e-01 1.86466646e+00 -5.87648690e-01 8.07197988e-01 3.33708405e-01 -1.33723927e+00 6.58388376e-01 1.76266387e-01 -6.44129589e-02 -2.30378658e-01 3.39291066e-01 3.77073921e-02 -1.27531126e-01 -6.37876511e-01 9.64070261e-01 -4.52641547e-02 -2.02675790e-01 6.42037034e-01 2.66815364e-01 1.03558347e-01 3.69960636e-01 4.42887992e-01 9.88384306e-01 1.33576080e-01 3.22692662e-01 -2.29627773e-01 6.05783522e-01 -4.03015882e-01 7.01037943e-02 1.00496995e+00 1.87721744e-01 8.51203978e-01 9.10292208e-01 -3.17904413e-01 -1.11225092e+00 -7.48487532e-01 5.30357771e-02 1.57445455e+00 -4.29767638e-01 -3.04503977e-01 -8.88478637e-01 -7.72984266e-01 -1.85430441e-02 8.47644031e-01 -7.85108089e-01 -2.28873253e-01 -8.36596727e-01 -7.17054129e-01 5.10661244e-01 8.09834659e-01 -6.53012618e-02 -1.34985709e+00 -4.96210098e-01 1.75814271e-01 -3.87341529e-02 -9.08632815e-01 -7.37421215e-01 5.39160848e-01 -8.17223012e-01 -6.60179377e-01 -8.69289875e-01 -8.09868991e-01 6.25408232e-01 -2.70853728e-01 1.28105187e+00 1.06942952e-01 1.51178256e-01 9.90948305e-02 -1.63561180e-01 -3.46441835e-01 -4.27610785e-01 6.18304074e-01 -3.03060502e-01 7.83617347e-02 4.04720120e-02 -5.95939040e-01 -4.38982278e-01 -3.89135629e-01 -1.07504463e+00 1.68841943e-01 7.43571579e-01 9.22643483e-01 1.55612648e-01 -1.01726973e+00 6.03919089e-01 -1.27137530e+00 9.58243072e-01 -5.98068178e-01 -2.23558143e-01 1.62483215e-01 -3.23686749e-01 5.69829464e-01 7.96971321e-01 -1.68729842e-01 -9.05701935e-01 -6.65659010e-02 -5.45521557e-01 -1.87427387e-01 -1.11199275e-01 8.57573390e-01 8.53398219e-02 2.24335298e-01 5.56534767e-01 9.90006998e-02 -9.92304981e-02 -6.03302538e-01 8.03798378e-01 7.04793930e-01 8.30569506e-01 -4.93041307e-01 4.43634301e-01 2.42320895e-01 -3.43717217e-01 -6.39975190e-01 -1.03869104e+00 -2.40853205e-01 -5.29368997e-01 3.30743670e-01 7.31123030e-01 -5.22447109e-01 -6.45260811e-01 1.56186536e-01 -1.66642046e+00 -3.80935341e-01 -5.50431788e-01 2.73245662e-01 -6.81087613e-01 5.08859336e-01 -1.12761986e+00 -5.88152885e-01 -5.92992783e-01 -1.01283646e+00 1.24577928e+00 5.38324118e-02 -6.23176336e-01 -1.14183879e+00 -1.07328646e-01 4.68384027e-02 6.58615232e-01 -1.34444132e-03 8.93815339e-01 -9.59300578e-01 -3.72024953e-01 -6.07305281e-02 -4.02053177e-01 5.84749520e-01 -2.01583415e-01 9.95997936e-02 -1.07735693e+00 -6.08638823e-02 -2.78229304e-02 -2.76919127e-01 1.48441768e+00 6.19705319e-01 1.53229034e+00 -7.91192830e-01 8.06816891e-02 8.95536125e-01 1.08743119e+00 -4.27512050e-01 6.23608768e-01 2.53540963e-01 7.37807214e-01 5.72170198e-01 -1.50873140e-01 4.45639819e-01 2.95264274e-01 5.25457144e-01 5.88326395e-01 -3.19256425e-01 5.28773107e-02 6.93584159e-02 3.90278250e-01 9.07471657e-01 -2.57924914e-01 -3.48459482e-01 -5.29029846e-01 5.88745654e-01 -2.00626755e+00 -9.38722849e-01 -1.08858518e-01 1.99009073e+00 1.07756078e+00 3.13129634e-01 -7.43107125e-02 5.96548319e-02 6.56375647e-01 4.16495681e-01 -4.62348074e-01 -1.19702184e+00 -2.04611197e-01 6.56494558e-01 4.29876149e-01 9.34725761e-01 -1.10822070e+00 7.57751882e-01 6.95943642e+00 5.84730744e-01 -9.95299101e-01 -3.53383049e-02 6.99560106e-01 -2.92345643e-01 -5.25025189e-01 -3.00071806e-01 -5.45294821e-01 3.75293553e-01 1.12011898e+00 -1.93144068e-01 4.11894143e-01 7.81085074e-01 2.86455661e-01 1.47126824e-01 -1.41926062e+00 4.13989246e-01 1.83792561e-01 -1.46722794e+00 4.46891412e-02 -2.29478359e-01 7.89522469e-01 1.96808815e-01 3.01263444e-02 3.11909020e-01 2.53683478e-01 -1.15820444e+00 6.83685362e-01 3.90972733e-01 5.74940503e-01 -7.26804793e-01 7.08650708e-01 1.52240768e-01 -7.34151125e-01 3.53413518e-03 -3.58713627e-01 -4.38468069e-01 3.49933028e-01 3.97880256e-01 -6.77417040e-01 4.75304782e-01 9.35116932e-02 8.59228432e-01 -3.11526388e-01 8.55835617e-01 -5.70044041e-01 6.42867029e-01 -2.73981750e-01 -1.18569307e-01 5.88222504e-01 2.20127422e-02 6.23758495e-01 1.85133588e+00 6.72162846e-02 -2.74793714e-01 -1.55286029e-01 1.02597237e+00 -5.50111473e-01 -5.57718500e-02 -3.45422864e-01 9.52262208e-02 1.49026755e-02 1.19664347e+00 -3.54487538e-01 -6.39047682e-01 -4.78632241e-01 1.35133207e+00 7.44262993e-01 6.34457052e-01 -1.00566208e+00 -7.61533976e-01 5.95998466e-01 -1.80857763e-01 6.85122669e-01 -1.05622252e-02 -6.01527214e-01 -1.27005255e+00 1.44958138e-01 -6.20813251e-01 2.79557019e-01 -7.26182342e-01 -1.19227803e+00 6.80129349e-01 -1.74717918e-01 -5.12313724e-01 -4.40256834e-01 -6.87124550e-01 -9.50136483e-01 1.06190336e+00 -1.75451195e+00 -8.14599633e-01 9.78528336e-02 3.42335314e-01 7.47712851e-01 2.75922060e-01 7.73734331e-01 2.25992888e-01 -7.59700298e-01 6.36304259e-01 1.18806303e-01 5.80044650e-02 4.71890241e-01 -1.54190922e+00 7.65084326e-01 8.32694948e-01 1.80318967e-01 7.69012511e-01 8.08826506e-01 -1.07169591e-01 -1.08895266e+00 -1.08441341e+00 1.36575985e+00 -3.78309846e-01 8.48670185e-01 -2.67915577e-01 -1.14910913e+00 1.18242884e+00 6.93860590e-01 -2.41665483e-01 3.98941457e-01 3.67228895e-01 -1.33736551e-01 3.05656224e-01 -7.22547591e-01 5.39034784e-01 9.03640270e-01 -5.80325544e-01 -9.45794880e-01 3.87626439e-01 1.06417382e+00 -4.66641486e-01 -5.32239139e-01 1.91913366e-01 3.67081583e-01 -8.57238352e-01 8.68314326e-01 -1.02752256e+00 1.04008913e+00 1.00560948e-01 9.12667140e-02 -1.37166190e+00 -4.74880874e-01 -9.76340652e-01 -4.96172845e-01 9.66210186e-01 8.69314909e-01 -5.59756517e-01 7.58142769e-01 7.13781118e-01 -7.48174369e-01 -9.84920442e-01 -6.33528650e-01 -3.75741184e-01 2.55369693e-01 -4.43538159e-01 4.13813531e-01 7.52110243e-01 2.24993438e-01 5.07866919e-01 -4.49603707e-01 -1.03642121e-01 2.13548332e-01 -1.00230105e-01 2.76825577e-01 -7.87711918e-01 -6.69840157e-01 -8.81362021e-01 -9.69448220e-03 -1.45627034e+00 3.07699561e-01 -1.15814233e+00 8.71338919e-02 -1.75759792e+00 2.60374963e-01 2.41908669e-01 -3.42387438e-01 5.71099401e-01 -4.90181118e-01 2.68361568e-01 3.06487024e-01 -7.15555325e-02 -7.35594690e-01 6.51327372e-01 9.16064620e-01 -2.14403525e-01 -1.80267826e-01 8.78907740e-02 -9.76274729e-01 8.10742974e-01 7.48076439e-01 -3.06699991e-01 3.29123321e-03 -1.01566923e+00 3.47194791e-01 -2.28172511e-01 3.00430298e-01 -6.17686749e-01 3.28052551e-01 7.61794671e-02 8.55747238e-02 -2.87749559e-01 2.79698789e-01 -3.64270210e-01 -3.52755159e-01 2.23623842e-01 -9.30029154e-01 8.48445445e-02 1.99617013e-01 3.26479107e-01 -2.73290247e-01 -4.97446239e-01 4.72758532e-01 -2.21490011e-01 -3.17518920e-01 1.43382311e-01 -4.86813456e-01 9.41353589e-02 4.59332317e-01 -1.12997089e-02 -1.47383064e-01 -4.52743977e-01 -1.05207515e+00 3.51564676e-01 2.19447985e-01 1.57481298e-01 4.78757888e-01 -1.16343987e+00 -9.79373157e-01 1.20930314e-01 -2.79501975e-01 1.39670968e-01 -8.76269415e-02 1.04176342e+00 -3.95690084e-01 6.69488430e-01 5.29531650e-02 -2.94452846e-01 -1.00764966e+00 4.72360849e-01 1.90900534e-01 -4.93168384e-01 -2.88720280e-01 1.03138328e+00 2.39024445e-01 -3.58298749e-01 4.01879430e-01 -8.95850599e-01 -9.60432291e-02 -1.24809518e-02 5.23638129e-01 1.83871344e-01 1.63326070e-01 -3.64707947e-01 -2.22092763e-01 2.17304498e-01 -9.45688635e-02 1.28142536e-01 1.54436767e+00 -1.47954658e-01 -3.38403463e-01 4.66385990e-01 1.25221503e+00 -9.45354030e-02 -1.24847817e+00 -2.88819909e-01 1.56509280e-01 1.36601076e-01 -2.16825992e-01 -6.10373318e-01 -8.47245693e-01 9.80672002e-01 -4.16059494e-01 6.03883088e-01 1.08873665e+00 1.44648582e-01 8.16296101e-01 6.31776154e-01 -4.22266245e-01 -8.20711493e-01 -3.65231127e-01 8.87802541e-01 1.10909748e+00 -1.10729325e+00 -1.97517246e-01 -5.84759042e-02 -6.04723752e-01 1.26520050e+00 2.03469545e-01 -4.20720339e-01 1.61564365e-01 5.02016485e-01 -3.44587237e-01 3.36999223e-02 -9.27321911e-01 -2.53798753e-01 3.45358938e-01 2.72289872e-01 8.08485091e-01 -1.76323369e-01 -3.35580707e-01 6.81779325e-01 -2.51389772e-01 -2.95575913e-02 5.27937651e-01 7.28768945e-01 -6.20002210e-01 -9.55162406e-01 -1.81958839e-01 5.19169748e-01 -8.56977522e-01 -6.00072622e-01 -4.78169113e-01 4.47984785e-01 -2.30079547e-01 6.06294215e-01 2.01816201e-01 -2.46897005e-02 4.27709669e-01 1.52363211e-01 4.62606400e-01 -7.62254298e-01 -9.84893322e-01 -3.36628497e-01 3.94130915e-01 -4.38354224e-01 -3.54484588e-01 -4.02255386e-01 -1.06182384e+00 -3.28950137e-01 -2.69401163e-01 5.75658642e-02 5.05588591e-01 1.10164690e+00 3.88793468e-01 7.97657788e-01 3.51548135e-01 -1.22737432e+00 -9.87986207e-01 -1.22498548e+00 -2.09802777e-01 5.33124149e-01 7.95617759e-01 7.47037604e-02 -3.25179368e-01 2.89282888e-01]
[11.520644187927246, 8.797663688659668]
afffb7d4-c168-41ac-92d8-7f66e9d2d520
nsitnlp4if-2019-propaganda-detection-from
null
null
https://aclanthology.org/D19-5021
https://aclanthology.org/D19-5021.pdf
NSIT@NLP4IF-2019: Propaganda Detection from News Articles using Transfer Learning
In this paper, we describe our approach and system description for NLP4IF 2019 Workshop: Shared Task on Fine-Grained Propaganda Detection. Given a sentence from a news article, the task is to detect whether the sentence contains a propagandistic agenda or not. The main contribution of our work is to evaluate the effectiveness of various transfer learning approaches like ELMo, BERT, and RoBERTa for propaganda detection. We show the use of Document Embeddings on the top of Stacked Embeddings combined with LSTM for identification of propagandistic context in the sentence. We further provide analysis of these models to show the effect of oversampling on the provided dataset. In the final test-set evaluation, our system ranked 21st with F1-score of 0.43 in the SLC Task.
['Anubhav Sadana', 'Kartik Aggarwal']
2019-11-01
null
null
null
ws-2019-11
['propaganda-detection']
['natural-language-processing']
[-7.70180002e-02 1.12728581e-01 -1.55552462e-01 -1.72218695e-01 -1.08832467e+00 -4.62987632e-01 1.24878120e+00 3.96317363e-01 -6.25802994e-01 6.34686470e-01 1.08946478e+00 -6.27492130e-01 1.31030068e-01 -7.61559188e-01 -6.94815576e-01 -5.85078418e-01 -2.02150047e-02 1.56773314e-01 1.33007895e-02 -4.04641896e-01 7.80817211e-01 3.07787806e-01 -5.53001106e-01 1.05718064e+00 5.76123357e-01 5.55149972e-01 -2.09144145e-01 9.71358001e-01 9.57395416e-03 1.60084856e+00 -1.23942792e+00 -3.05467218e-01 -3.70882243e-01 -4.23446238e-01 -1.14336610e+00 -4.63374674e-01 7.24409580e-01 -5.31097054e-01 -3.77191931e-01 9.77860272e-01 6.48889661e-01 -1.91564158e-01 9.45475280e-01 -2.99804181e-01 -8.56480896e-01 1.11594713e+00 -4.87456799e-01 8.86716187e-01 3.27442348e-01 -3.22060436e-01 8.48817170e-01 -1.06679976e+00 9.45033491e-01 1.45580375e+00 6.06955171e-01 6.32746875e-01 -9.87281322e-01 -3.36582094e-01 -1.36754587e-01 2.99291432e-01 -8.62053394e-01 -3.55998814e-01 6.14505649e-01 -7.71134257e-01 1.01340175e+00 3.01206019e-02 4.46524352e-01 1.61530256e+00 6.86798632e-01 8.88016403e-01 1.13334930e+00 -6.96629345e-01 1.52721822e-01 2.08644539e-01 7.13795781e-01 8.36502433e-01 5.09740673e-02 -2.13797018e-01 -5.63228548e-01 -6.03533924e-01 2.86274761e-01 -6.62363529e-01 7.02651888e-02 4.16454554e-01 -9.81226683e-01 1.39907551e+00 6.13062859e-01 8.06591272e-01 -3.87864530e-01 3.64972681e-01 6.21390760e-01 2.11462423e-01 1.14404428e+00 5.59928775e-01 -6.44264892e-02 -8.45441148e-02 -8.80163491e-01 4.78955299e-01 7.51315892e-01 -8.36398825e-03 5.98087795e-02 -1.12043666e-02 -6.99558616e-01 4.89382505e-01 -6.52883425e-02 4.30272609e-01 2.78504819e-01 -2.78031528e-01 6.55037105e-01 2.48313129e-01 2.08205327e-01 -1.14057815e+00 -5.14841437e-01 -4.03374344e-01 -4.40939814e-01 -1.20820731e-01 3.31085533e-01 -6.67574108e-01 -1.02432644e+00 1.33750939e+00 1.35567803e-02 -8.30373392e-02 -6.13033623e-02 6.38668835e-01 7.40864575e-01 1.11450577e+00 2.15754494e-01 -7.23506510e-03 1.35613394e+00 -1.08316362e+00 -7.93558359e-01 -5.14826663e-02 1.17906046e+00 -8.50418270e-01 6.85949266e-01 -5.04417084e-02 -6.92666590e-01 -1.37963414e-01 -9.21833158e-01 -9.96579975e-02 -3.29328030e-01 2.10047126e-01 1.50866121e-01 5.26649356e-01 -6.46997690e-01 5.33310294e-01 -5.48661709e-01 -4.61553693e-01 3.47691238e-01 -3.15874606e-01 -2.84435488e-02 2.24991545e-01 -1.39573514e+00 1.32944095e+00 2.61868328e-01 -9.05021131e-02 -1.05747437e+00 -5.37186265e-01 -6.56730175e-01 2.00218032e-03 -5.15612066e-02 -1.66027427e-01 1.17773509e+00 -6.18278980e-01 -8.41782451e-01 9.69280601e-01 -1.73022468e-02 -9.87479329e-01 3.76983345e-01 -7.47461200e-01 -5.75937033e-01 1.61350623e-01 3.17595810e-01 7.80808041e-03 7.90028155e-01 -7.15633512e-01 -5.13088644e-01 -1.32643729e-01 4.65707965e-02 -1.16660967e-01 -4.31247801e-01 8.42839539e-01 5.31301618e-01 -4.26638931e-01 -5.13343811e-01 -4.96581614e-01 4.35221270e-02 -8.30782115e-01 -5.85568249e-01 -5.81550717e-01 8.46122861e-01 -1.09513927e+00 1.38759589e+00 -2.30971742e+00 -5.31362481e-02 -2.14804456e-01 1.82544723e-01 4.71812636e-01 -2.24611446e-01 8.56047630e-01 1.46477282e-01 6.00456238e-01 4.54017818e-01 1.56489685e-01 -1.79928392e-01 -3.72655690e-01 -9.13394868e-01 7.43917584e-01 3.44402045e-01 6.14281535e-01 -9.32011068e-01 -4.68870670e-01 -8.74517560e-02 5.42293966e-01 -3.02300483e-01 -1.24331415e-01 -1.43377185e-01 1.89819366e-01 -6.34289801e-01 1.68129086e-01 3.69772762e-01 -2.97590345e-01 1.38938632e-02 -8.43539368e-03 -3.70608836e-01 9.47840452e-01 -2.91134864e-01 9.96702433e-01 -3.19208533e-01 1.01668048e+00 -4.25201096e-02 -5.64394891e-01 5.10879338e-01 3.93501192e-01 -5.85468709e-02 -7.60669470e-01 5.34223020e-01 1.61345806e-02 1.44761607e-01 -5.14843225e-01 5.30907571e-01 4.50026914e-02 -3.83483708e-01 7.14448690e-01 -8.51119235e-02 3.90361249e-01 3.88359725e-01 6.90786242e-01 1.42933512e+00 -3.32660198e-01 4.27017599e-01 -4.95623559e-01 4.53166634e-01 3.52467507e-01 1.32280216e-01 9.90790069e-01 -3.65121126e-01 2.13842794e-01 7.43598461e-01 -8.40082228e-01 -9.06738281e-01 -1.05227017e+00 -1.92167878e-01 1.43929529e+00 -4.39735323e-01 -5.75957239e-01 -8.44802380e-01 -1.18209589e+00 -2.99635977e-01 1.05048871e+00 -1.16706598e+00 1.83930501e-01 -1.12022901e+00 -8.41026485e-01 7.40271032e-01 2.62988567e-01 3.02711815e-01 -1.21639276e+00 -7.46626198e-01 2.08790421e-01 -3.57719034e-01 -8.23056161e-01 -1.61781281e-01 1.03561074e-01 -4.67894197e-01 -1.16588342e+00 -4.45157290e-01 -9.53293800e-01 3.10933024e-01 -5.47697619e-02 6.91159010e-01 -1.15156315e-01 -2.10906848e-01 -1.32311314e-01 -4.65387851e-01 -2.92238742e-01 -7.66538441e-01 -3.66801880e-02 1.58671867e-02 -3.57599378e-01 2.95625716e-01 1.86214089e-01 -2.15974644e-01 -4.41632599e-01 -4.01976585e-01 6.01631626e-02 2.71417081e-01 9.77729142e-01 -2.55347818e-01 -3.39262515e-01 7.01108038e-01 -1.17770445e+00 1.22447371e+00 -4.81796771e-01 -2.24274248e-01 1.42364815e-01 -1.43688798e-01 -4.29047756e-02 5.41456461e-01 -1.11709997e-01 -1.05902863e+00 -5.71815908e-01 -2.19033748e-01 3.39855671e-01 4.34961841e-02 7.02379405e-01 6.46359861e-01 3.85395795e-01 1.34633565e+00 1.47167072e-02 -4.16731596e-01 -6.59829795e-01 4.63515759e-01 9.45949435e-01 1.24031894e-01 -9.26263928e-02 4.85979855e-01 3.55732232e-01 -6.27822697e-01 -8.89828444e-01 -1.40383518e+00 -3.74850005e-01 -4.05903995e-01 -6.00279532e-02 9.13281500e-01 -8.65771651e-01 -9.61680859e-02 2.78630346e-01 -1.58999670e+00 -2.29987577e-01 3.98313859e-03 3.96062136e-01 -2.56430745e-01 1.59907237e-01 -1.32596385e+00 -9.62110758e-01 -6.80020452e-01 -6.00201011e-01 7.85181344e-01 -2.17460766e-01 -3.98043603e-01 -1.07159531e+00 5.06239951e-01 5.52248538e-01 3.53737593e-01 3.29606980e-01 1.17845321e+00 -1.13200927e+00 2.51101762e-01 -4.18684870e-01 -4.03913885e-01 2.22989917e-01 5.09650372e-02 -3.79600860e-02 -1.05541182e+00 -2.25044206e-01 -1.67236174e-03 -4.97429371e-01 1.24479187e+00 4.33946788e-01 5.48712313e-01 -8.72709274e-01 -2.26342380e-01 -2.13468432e-01 1.04976046e+00 -1.27180099e-01 5.34413934e-01 6.13058448e-01 3.97374600e-01 6.23679161e-01 5.20115376e-01 2.90025204e-01 -2.22778603e-01 3.15004617e-01 1.73582718e-01 9.06026438e-02 -3.81587654e-01 -2.83760577e-01 8.55565369e-01 6.62640750e-01 2.59660453e-01 -4.35985148e-01 -1.01441288e+00 7.69527972e-01 -1.61169100e+00 -1.50439167e+00 -4.43385243e-01 1.60215700e+00 8.04357052e-01 1.53602481e-01 1.23580709e-01 -5.95513694e-02 8.94223273e-01 6.41473770e-01 4.33284134e-01 -1.10599959e+00 -4.92782667e-02 9.66732875e-02 2.84824073e-01 9.20386434e-01 -1.64711428e+00 1.15635288e+00 7.14580297e+00 1.07081807e+00 -1.07127714e+00 5.64576745e-01 5.58405399e-01 -3.37622881e-01 5.16739823e-02 -2.96825439e-01 -9.90534067e-01 4.34860319e-01 1.38034618e+00 -2.49223068e-01 -8.60282332e-02 7.26191401e-01 4.23967957e-01 -1.14407822e-01 -7.37946987e-01 8.78687203e-02 4.54511344e-01 -1.98174131e+00 1.37838759e-02 1.52229220e-01 8.59339774e-01 4.84128505e-01 8.89195800e-02 5.32943249e-01 4.04238701e-01 -1.05653834e+00 7.77561843e-01 -1.55640736e-01 2.22697884e-01 -5.65352738e-01 9.14728224e-01 6.19860709e-01 -2.83753008e-01 -6.27493113e-02 -2.78683573e-01 -4.51371789e-01 4.68810380e-01 7.88152754e-01 -1.35036266e+00 2.05409437e-01 2.73596406e-01 4.37496424e-01 -5.03252268e-01 6.37608409e-01 -7.85841405e-01 1.22751141e+00 3.02332472e-02 -6.25005066e-01 8.05917799e-01 5.45231342e-01 6.64896548e-01 1.78229606e+00 -8.97223949e-02 -2.67480671e-01 4.18043658e-02 5.54995418e-01 -6.58263713e-02 2.65503377e-01 -5.13344228e-01 -3.27642262e-01 2.23864421e-01 1.20039093e+00 -2.79155254e-01 -5.78806758e-01 1.07367128e-01 7.09613621e-01 6.09723687e-01 1.69271648e-01 -8.99971664e-01 -2.23802894e-01 2.25062340e-01 4.10891354e-01 1.15579523e-01 -1.02496631e-01 -1.02306493e-01 -8.92450571e-01 -2.87862986e-01 -4.25160438e-01 5.12546122e-01 -3.53823572e-01 -1.25170207e+00 7.51197577e-01 -5.96924610e-02 -4.08768266e-01 -3.83696973e-01 -5.92051268e-01 -9.87345576e-01 7.32756078e-01 -1.10827422e+00 -1.12259626e+00 4.15927470e-01 -1.01821862e-01 6.68206990e-01 -9.76969451e-02 8.33744943e-01 9.70295072e-02 -5.03828943e-01 5.46889715e-02 2.58583784e-01 6.58086300e-01 7.48523533e-01 -1.04713857e+00 6.04986966e-01 1.00338948e+00 1.89982504e-01 7.14683354e-01 1.03694761e+00 -8.08366537e-01 -7.73684442e-01 -1.19618058e+00 1.60199130e+00 -6.50870621e-01 1.26302421e+00 -6.38127923e-01 -5.80661237e-01 6.00288332e-01 7.08792627e-01 -4.37735677e-01 5.61120331e-01 5.47063828e-01 -6.93618417e-01 5.37758648e-01 -1.13337052e+00 4.68658060e-01 4.89990085e-01 -5.19406796e-01 -1.03376234e+00 1.03051710e+00 5.83542645e-01 -1.18771657e-01 -3.50669026e-01 2.46720761e-02 3.96777719e-01 -6.44688308e-01 5.50739586e-01 -1.12530386e+00 1.10844576e+00 2.94496000e-01 -1.25245467e-01 -1.19909596e+00 -7.83374012e-01 -8.00133422e-02 -1.74239337e-01 8.59320939e-01 5.00185847e-01 -3.08504492e-01 6.31546915e-01 -3.48968327e-01 1.54596791e-02 -6.68724179e-01 -1.16219985e+00 -6.81774974e-01 5.66649377e-01 -3.05808038e-02 -4.09611642e-01 8.71986866e-01 3.41772139e-01 9.36628938e-01 -7.45252371e-01 7.01447502e-02 4.94587094e-01 5.20849740e-03 4.31698263e-01 -7.22825706e-01 -9.34511945e-02 -2.91657239e-01 -1.96013331e-01 -6.34708524e-01 1.15393177e-01 -8.01850379e-01 3.03819329e-02 -2.00336456e+00 4.68628764e-01 2.49818236e-01 -3.24094236e-01 3.57404202e-01 1.06859386e-01 1.83566466e-01 2.48529643e-01 3.33454134e-03 -5.72161674e-01 2.00727060e-01 6.90727711e-01 -4.24529761e-01 6.55596107e-02 -4.21077907e-01 -5.55247247e-01 7.67222464e-01 1.08656871e+00 -6.87314212e-01 2.45038211e-01 -8.09936047e-01 1.81286111e-01 -2.83634275e-01 5.07096171e-01 -7.68580973e-01 -4.10081726e-03 1.17925438e-03 3.07923853e-01 -9.17505324e-01 2.03894138e-01 1.23596095e-01 -6.64574981e-01 1.01571047e+00 -5.42841434e-01 -1.16883598e-01 2.17400491e-01 5.55590332e-01 -2.59348787e-02 -3.95522982e-01 6.48277402e-01 8.33418444e-02 -4.13424820e-01 -5.09529471e-01 -1.05460465e+00 2.00509489e-01 7.85109818e-01 5.53151548e-01 -1.25243497e+00 -1.53410941e-01 -6.58200562e-01 -1.01014465e-01 -1.14469388e-02 4.56622452e-01 4.46752429e-01 -1.12008512e+00 -1.34196281e+00 -2.92366862e-01 -1.31601438e-01 -9.85968232e-01 1.73434108e-01 9.93704796e-01 -7.57542014e-01 7.31297672e-01 5.63500412e-02 -9.31913499e-03 -1.53531051e+00 3.76750350e-01 -2.69884467e-02 -6.06026530e-01 -7.04088688e-01 9.60085154e-01 1.35852724e-01 -2.14090254e-02 -4.76063462e-03 7.38624558e-02 -3.34585875e-01 2.20857784e-01 1.17470765e+00 4.52684343e-01 8.68487284e-02 -6.11632228e-01 -6.62416637e-01 -7.60825351e-02 -6.08115435e-01 -3.12256932e-01 1.27585399e+00 5.08803964e-01 -1.83519572e-01 4.43760097e-01 1.21845806e+00 5.26577055e-01 -3.05159897e-01 -1.13895036e-01 9.11789536e-02 -1.67070001e-01 3.07289869e-01 -1.29087889e+00 -2.00445056e-01 9.89864230e-01 4.71542537e-01 4.26240325e-01 1.91213861e-01 1.55438840e-01 6.39925361e-01 4.87936795e-01 -8.61013681e-03 -1.24166191e+00 1.05835170e-01 1.05115306e+00 1.13305056e+00 -7.26580799e-01 1.94101825e-01 -1.72490478e-01 -3.78477484e-01 1.01897097e+00 1.88573852e-01 -6.51509106e-01 3.98459822e-01 2.14987338e-01 1.30989656e-01 -6.40291214e-01 -9.02937531e-01 1.98978350e-01 1.75231233e-01 2.28991941e-01 8.47991407e-01 1.57940626e-01 -9.76999164e-01 1.78107411e-01 -5.80988266e-02 -3.84379566e-01 6.27501488e-01 9.31011915e-01 -1.15622258e+00 -4.83576536e-01 -3.64069879e-01 4.56554651e-01 -9.84801948e-01 -3.91706109e-01 -1.06595790e+00 6.16632044e-01 -9.57228616e-02 1.18733060e+00 9.80408564e-02 -3.80436003e-01 1.59296449e-02 7.89826140e-02 5.37692010e-01 -8.99704158e-01 -1.11089051e+00 1.44259647e-01 8.87894809e-01 -2.84514844e-01 -1.71750337e-01 -6.57540321e-01 -1.04385638e+00 -5.32254696e-01 -3.07334512e-01 4.82239991e-01 6.97713137e-01 9.86651957e-01 1.56661481e-01 5.16626894e-01 3.87044996e-01 -4.59460318e-01 -1.08154297e+00 -1.51416922e+00 -3.51733327e-01 1.47039622e-01 4.42981988e-01 -3.23078722e-01 -4.55114096e-01 -1.60652414e-01]
[8.478108406066895, 10.664602279663086]
5370bf12-587b-4f7a-bf09-7795ae2b8a46
indexing-ai-risks-with-incidents-issues-and
2211.10384
null
https://arxiv.org/abs/2211.10384v1
https://arxiv.org/pdf/2211.10384v1.pdf
Indexing AI Risks with Incidents, Issues, and Variants
Two years after publicly launching the AI Incident Database (AIID) as a collection of harms or near harms produced by AI in the world, a backlog of "issues" that do not meet its incident ingestion criteria have accumulated in its review queue. Despite not passing the database's current criteria for incidents, these issues advance human understanding of where AI presents the potential for harm. Similar to databases in aviation and computer security, the AIID proposes to adopt a two-tiered system for indexing AI incidents (i.e., a harm or near harm event) and issues (i.e., a risk of a harm event). Further, as some machine learning-based systems will sometimes produce a large number of incidents, the notion of an incident "variant" is introduced. These proposed changes mark the transition of the AIID to a new version in response to lessons learned from editing 2,000+ incident reports and additional reports that fall under the new category of "issue."
['Khoa Lam', 'Kevin Paeth', 'Sean McGregor']
2022-11-18
null
null
null
null
['computer-security']
['miscellaneous']
[ 9.88370627e-02 2.90241361e-01 3.53116659e-03 -2.45786041e-01 -8.17009866e-01 -5.21093845e-01 8.86284232e-01 8.97896469e-01 -4.92954552e-01 7.32644141e-01 8.58569741e-01 -4.63803560e-01 -4.43181455e-01 -9.11677361e-01 -5.00652850e-01 -3.27250808e-02 -1.30727157e-01 5.24394035e-01 7.28888214e-02 -1.81087554e-01 5.28541565e-01 5.97316980e-01 -1.47275734e+00 2.34232888e-01 3.20756406e-01 6.93791747e-01 -5.26900470e-01 1.64332986e-01 -1.41359657e-01 1.09717822e+00 -1.21569324e+00 -4.37247872e-01 5.37254930e-01 -3.42056274e-01 -1.12694478e+00 -3.45540971e-01 2.97477394e-01 -5.48935711e-01 -4.91990685e-01 7.37039387e-01 2.13706344e-01 1.56056523e-01 8.26652467e-01 -1.76900530e+00 -4.52505678e-01 4.01282966e-01 -1.85287014e-01 7.73661554e-01 7.73523033e-01 3.97164047e-01 7.44939923e-01 -5.66597581e-01 9.16328788e-01 1.06099653e+00 7.42819309e-01 2.00879186e-01 -9.16885018e-01 -8.69939208e-01 2.30229646e-01 5.61478958e-02 -1.28719771e+00 -3.35789174e-01 2.73484826e-01 -7.60762513e-01 1.62837243e+00 4.74881232e-01 5.58279335e-01 9.63885128e-01 5.39109051e-01 1.96835548e-01 5.98255873e-01 -1.94091588e-01 3.08739185e-01 1.07138984e-01 2.48719051e-01 -7.03439936e-02 8.16329300e-01 2.24758893e-01 -4.54972416e-01 -8.65086555e-01 1.78004161e-01 1.34893149e-01 5.55621758e-02 2.63074130e-01 -1.03587151e+00 6.42296076e-01 2.74723709e-01 2.92730123e-01 -7.96839654e-01 -1.17724709e-01 7.85407484e-01 1.99173197e-01 4.99790460e-01 7.51885176e-01 -1.93874002e-01 -3.36987764e-01 -4.81446713e-01 8.16762269e-01 9.37524915e-01 8.36305261e-01 5.18835723e-01 -1.54241249e-01 -4.01572175e-02 5.46145558e-01 8.37456211e-02 1.69395521e-01 -6.51060715e-02 -7.36404836e-01 5.87940335e-01 1.01132822e+00 4.37565506e-01 -1.37763631e+00 -6.40526652e-01 -4.01694439e-02 -4.36667621e-01 -2.28719153e-02 3.20670381e-02 -1.24514125e-01 -6.50476933e-01 1.44247818e+00 1.27700090e-01 -8.23086128e-02 1.83742736e-02 7.76970625e-01 4.48147416e-01 7.93733001e-01 6.14496887e-01 -3.97425324e-01 1.10720694e+00 -1.03308139e-02 -1.06392813e+00 -2.63028800e-01 7.44071007e-01 -6.26822233e-01 5.25488615e-01 5.64702094e-01 -1.10523605e+00 1.13617018e-01 -1.02303445e+00 3.50369722e-01 -8.44145238e-01 -9.65892851e-01 7.87357092e-01 3.08016956e-01 -5.00791550e-01 1.40463844e-01 -2.48618826e-01 -7.67882884e-01 1.72211915e-01 -2.16735601e-01 -2.58221954e-01 1.95550695e-02 -1.47499371e+00 1.46291995e+00 3.22189331e-01 -3.98731232e-01 -8.48835766e-01 -1.03005147e+00 -5.32127798e-01 -3.22058678e-01 5.90426087e-01 -4.96529222e-01 1.00603592e+00 -1.34762391e-01 -1.94422930e-01 8.22732449e-01 2.20962241e-01 -4.76415694e-01 2.39454329e-01 -4.59502578e-01 -1.14817643e+00 6.04096949e-02 5.67547560e-01 2.53610052e-02 7.60282576e-02 -1.04515398e+00 -9.80266809e-01 -4.05609399e-01 5.54733396e-01 2.30853736e-01 -1.41122460e-01 9.26485240e-01 2.49125659e-01 -6.08947456e-01 -3.05079162e-01 -5.80931604e-01 -3.62282693e-01 -2.73194641e-01 -2.09859416e-01 -5.50867200e-01 6.80325031e-01 -3.19198161e-01 1.84836709e+00 -1.92741108e+00 -2.79807478e-01 4.50256877e-02 2.26555452e-01 1.89843744e-01 3.70118231e-01 1.12861514e+00 -1.74547955e-01 6.15979373e-01 -2.05987111e-01 3.31241667e-01 -3.64401974e-02 2.28551120e-01 -7.56279230e-01 4.42808717e-01 3.52186769e-01 1.93893656e-01 -1.10774326e+00 -2.51505047e-01 -1.67962052e-02 -1.87253639e-01 -5.30810475e-01 2.08735436e-01 9.20863450e-02 1.46297574e-01 -4.53810900e-01 6.03812873e-01 4.16308284e-01 3.50281209e-01 -2.52922267e-01 3.43495756e-02 -7.31819272e-01 7.16540575e-01 -8.74500990e-01 1.24836361e+00 -1.40171433e-02 2.10447073e-01 -6.47059679e-02 -6.11303806e-01 9.01312232e-01 7.36033142e-01 8.07954669e-01 -5.62468529e-01 -3.39485146e-02 7.67043382e-02 -2.55866081e-01 -4.41531956e-01 6.70247138e-01 -4.03248191e-01 -8.32075894e-01 9.74251390e-01 -5.08890338e-02 -4.28757608e-01 2.60377407e-01 5.13354838e-01 1.88336051e+00 -4.14503813e-01 4.69671190e-01 -2.89233506e-01 -1.53054399e-02 7.56310523e-01 9.69641149e-01 7.30012000e-01 -4.27659571e-01 3.65957886e-01 3.95024240e-01 -1.09091532e+00 -1.09449148e+00 -1.06146407e+00 -4.29213673e-01 5.03648937e-01 1.64529420e-02 -8.16538274e-01 -5.67171395e-01 -6.02002621e-01 2.82206368e-02 1.27310979e+00 -3.65909696e-01 -4.77213889e-01 -4.06401008e-01 -7.81792521e-01 9.72955287e-01 4.04496640e-01 2.12082341e-01 -1.19147456e+00 -1.06961012e+00 4.03431565e-01 -3.30433518e-01 -8.05165112e-01 -1.05732689e-02 1.90343454e-01 -8.59972462e-02 -1.23396552e+00 -7.29331598e-02 3.89217176e-02 4.14429039e-01 -8.47685635e-02 1.04072487e+00 1.13250822e-01 -7.03598738e-01 6.20165944e-01 -3.87016654e-01 -1.21329510e+00 -4.06627536e-01 -4.44391638e-01 5.75883031e-01 -5.45925081e-01 9.56137419e-01 -2.51812220e-01 -3.15734237e-01 -3.75467762e-02 -1.27768064e+00 -5.10671794e-01 1.62113667e-01 1.42584458e-01 1.65287048e-01 2.45750606e-01 1.18604481e+00 -8.77713859e-01 1.11680269e+00 -1.37305593e+00 -3.19636673e-01 1.48025937e-02 -9.43571806e-01 -6.52908981e-01 3.42270851e-01 7.27142394e-02 -9.83827710e-01 -5.56090593e-01 1.01388082e-01 -5.49955890e-02 -8.21612775e-01 7.35354066e-01 1.24995299e-01 6.27510607e-01 1.00531840e+00 -4.85590011e-01 -2.66121030e-01 -2.99391657e-01 -1.57725930e-01 8.56252432e-01 3.95870030e-01 -3.63743335e-01 8.50413263e-01 2.25511760e-01 -4.45454568e-01 -4.28744793e-01 -8.99896801e-01 -7.65631974e-01 -2.85344690e-01 -6.29720151e-01 8.52968872e-01 -7.24074185e-01 -5.06974518e-01 3.15525770e-01 -1.45831597e+00 5.98084852e-02 -5.35328686e-01 4.70232517e-01 -3.33043486e-01 -1.75809190e-01 -4.58090156e-01 -8.00043941e-01 -1.20020561e-01 -7.69538939e-01 3.81685197e-01 8.28060228e-03 -9.55398023e-01 -4.73771840e-01 4.81970549e-01 2.72336397e-02 4.80791181e-01 5.90503037e-01 1.05732977e+00 -1.08380961e+00 -1.75129458e-01 -7.54329324e-01 -1.34155601e-01 8.49600658e-02 2.02161580e-01 -1.55574471e-01 -4.77151781e-01 1.39422327e-01 9.65298936e-02 -1.32261857e-01 1.14368327e-01 -9.28671956e-02 5.33327878e-01 -7.74191201e-01 -4.82906222e-01 -1.04656421e-01 1.19924366e+00 1.10616219e+00 6.96624398e-01 7.00882494e-01 1.43785506e-01 9.57077980e-01 8.47479582e-01 9.55564320e-01 3.87438953e-01 5.92439055e-01 3.09694231e-01 2.13407010e-01 1.43527821e-01 -1.46270126e-01 2.54111230e-01 3.86021405e-01 1.24027915e-01 -4.06471044e-01 -1.52954173e+00 8.43361497e-01 -1.48385382e+00 -1.14267778e+00 -3.15364569e-01 2.12821317e+00 6.88400209e-01 4.81896490e-01 1.59318909e-01 4.27888632e-02 5.14946282e-01 2.09439605e-01 -7.01173902e-01 -1.10523522e+00 3.05568725e-01 -1.18315965e-01 4.19814140e-01 7.54317418e-02 -7.65273094e-01 5.57861209e-01 7.76172924e+00 1.04639553e-01 -5.24662197e-01 -1.83002323e-01 5.14905930e-01 2.89287586e-02 -4.14912105e-01 1.74823627e-01 -6.80056393e-01 4.17866528e-01 1.38403010e+00 -1.18563366e+00 1.18578143e-01 6.69313967e-01 8.28186870e-02 -3.02747190e-01 -9.17103648e-01 3.89627546e-01 2.51960486e-01 -1.35129905e+00 2.79517651e-01 8.52617770e-02 3.30153823e-01 -1.45618878e-02 -2.72595018e-01 3.69211555e-01 5.14404655e-01 -8.88016045e-01 8.22337985e-01 6.91578388e-01 7.23827422e-01 -9.85175490e-01 7.17294872e-01 3.40252548e-01 -7.27151513e-01 -2.61973977e-01 -2.64786243e-01 -6.27503037e-01 6.20297313e-01 4.77094263e-01 -9.16459024e-01 6.39831424e-01 9.29125428e-01 2.86706746e-01 -2.67233938e-01 1.04849839e+00 2.98048824e-01 5.52600324e-01 -1.93649128e-01 3.31089199e-01 1.90686315e-01 1.15422450e-01 1.13108826e+00 1.12946236e+00 -9.56864469e-03 6.54844582e-01 2.67600685e-01 7.77259886e-01 2.82007098e-01 -2.33181864e-01 -1.33817828e+00 -1.47081867e-01 7.52087355e-01 7.19866216e-01 -3.77226204e-01 -3.63856256e-01 -4.00992990e-01 3.82261008e-01 -2.34950200e-01 2.57357985e-01 -6.60049021e-01 -6.91304088e-01 8.64333808e-01 4.57561195e-01 -5.71468413e-01 3.87867950e-02 -2.61542261e-01 -4.37029243e-01 -1.63567021e-01 -8.68035495e-01 6.56912804e-01 -8.28823864e-01 -1.36545765e+00 9.27653849e-01 7.78392732e-01 -1.36758471e+00 -4.77936596e-01 -2.12600708e-01 -5.94868422e-01 7.63340235e-01 -6.27120376e-01 -5.55819392e-01 3.76419127e-02 1.77287683e-01 3.22268099e-01 -1.06354609e-01 1.06497204e+00 4.00143474e-01 -4.43099082e-01 2.29001716e-01 -7.75047898e-01 -3.65782045e-02 9.39260542e-01 -8.34982872e-01 6.09426558e-01 7.15831816e-01 -4.21164662e-01 7.26671815e-01 1.00207961e+00 -1.19960153e+00 -1.07133806e+00 -1.09429491e+00 1.24977541e+00 -9.36579287e-01 9.70991850e-01 5.12905559e-03 -7.68996775e-01 1.11489916e+00 2.63644665e-01 -6.75681055e-01 6.96925342e-01 9.05225053e-02 -2.09800929e-01 -8.38170201e-03 -1.46853316e+00 6.32895172e-01 1.07236159e+00 -3.94545257e-01 -1.40900636e+00 6.02589786e-01 9.69548941e-01 -1.51864573e-01 -8.34262311e-01 5.57023883e-01 6.49934933e-02 -5.99885762e-01 7.07713127e-01 -9.24737453e-01 4.62523639e-01 -2.95061231e-01 3.66526321e-02 -1.27361882e+00 -4.60343421e-01 -5.03311098e-01 3.89963239e-01 1.29302061e+00 3.88255149e-01 -7.95490324e-01 -1.27220392e-01 1.32431805e+00 -6.80772662e-01 -4.80324507e-01 -9.79092360e-01 -7.21379578e-01 -5.20422347e-02 -8.69260192e-01 8.88181984e-01 1.45376480e+00 4.75931346e-01 -1.35185113e-02 -2.68455058e-01 2.37239450e-01 5.07193089e-01 -5.23474038e-01 6.30609155e-01 -1.27882016e+00 5.27239799e-01 -1.91086560e-01 -7.14895904e-01 -4.36047539e-02 -4.26743954e-01 -6.23860657e-01 1.13614853e-02 -1.68335676e+00 3.11948895e-01 -5.05235016e-01 -4.54564065e-01 7.72330403e-01 1.40512720e-01 -1.45617083e-01 1.00951687e-01 3.22637528e-01 -5.24877965e-01 -5.56599572e-02 2.86211133e-01 1.36602065e-02 -2.31004320e-02 -3.49038720e-01 -8.72299194e-01 1.01429880e+00 7.29438722e-01 -7.35724151e-01 -4.89090011e-02 -2.95594215e-01 7.04093575e-01 5.89532815e-02 3.23347062e-01 -1.24825501e+00 4.72432971e-01 -7.59672582e-01 -1.66982591e-01 -7.91528046e-01 1.01877891e-01 -9.48620081e-01 7.34427154e-01 4.13691908e-01 -6.45355880e-01 4.92188871e-01 6.00982666e-01 3.97387236e-01 -3.36502999e-01 -2.41433039e-01 9.28036571e-02 -1.32689744e-01 -6.52488768e-01 1.25444412e-01 -1.02595627e+00 1.25063673e-01 1.45318115e+00 -1.02997012e-01 -8.75552714e-01 -3.32093656e-01 -3.46405685e-01 3.11926275e-01 3.74465287e-01 8.53119552e-01 6.40181243e-01 -1.16122127e+00 -7.36184716e-01 -1.37450531e-01 6.08851433e-01 -7.69262016e-02 2.49317840e-01 6.18209124e-01 -3.00927073e-01 6.31260872e-01 -3.46132547e-01 2.07658604e-01 -5.55852413e-01 6.46281242e-01 -1.13155380e-01 -2.18594491e-01 -9.18776155e-01 4.96756107e-01 3.39704037e-01 -2.71351725e-01 2.83041835e-01 7.47964755e-02 -1.92052826e-01 2.27880612e-01 1.04965019e+00 5.67115903e-01 1.47616327e-01 -7.90132403e-01 -7.33707488e-01 -1.89467683e-01 -3.18152785e-01 -1.04290761e-01 1.49064326e+00 1.71488971e-02 -2.50260025e-01 8.92997980e-01 7.80831397e-01 -1.17730111e-01 -5.14785051e-01 1.89703420e-01 3.69655967e-01 -3.70676666e-01 -1.07384883e-01 -1.44881272e+00 -4.37168539e-01 2.50806838e-01 2.74801075e-01 5.56199372e-01 1.00385845e+00 2.97338124e-02 9.10559595e-01 1.73818648e-01 6.56776071e-01 -1.24325264e+00 -3.22990090e-01 5.59177041e-01 1.53578901e+00 -5.14457047e-01 1.78969830e-01 4.13844623e-02 -8.90712023e-01 6.37466431e-01 6.94476068e-01 -3.51642165e-03 5.36544859e-01 6.41925275e-01 1.89374700e-01 -7.44872034e-01 -1.09167790e+00 3.36287260e-01 -2.07838595e-01 4.97585326e-01 1.52096018e-01 1.07097052e-01 -7.71889329e-01 8.39870870e-01 7.63429375e-03 3.03957295e-02 8.90595078e-01 1.21459007e+00 -3.02437335e-01 -6.93884552e-01 -4.79598284e-01 8.66375089e-01 -6.85223758e-01 -7.69759491e-02 -7.22688854e-01 8.34039509e-01 1.94664761e-01 1.17448318e+00 4.63458091e-01 -6.13303304e-01 9.29424405e-01 2.33281553e-01 -5.09996533e-01 -8.54331136e-01 -1.22232854e+00 -6.50698781e-01 5.54608524e-01 -8.98327112e-01 -2.42517471e-01 -7.26893127e-01 -1.41697383e+00 -5.78103602e-01 -4.07068580e-02 3.19711328e-01 5.56506038e-01 1.01035595e+00 5.11905134e-01 5.12074351e-01 2.82722175e-01 -2.92777508e-01 -3.41334820e-01 -7.50114799e-01 -4.78701800e-01 4.83765393e-01 8.76773223e-02 -8.74057353e-01 -6.12912476e-01 4.55518533e-03]
[8.803709030151367, 9.244643211364746]
ea4ad3f3-cf9e-456c-934c-9a75e79fb08e
repurposing-knowledge-graph-embeddings-for
2208.10328
null
https://arxiv.org/abs/2208.10328v1
https://arxiv.org/pdf/2208.10328v1.pdf
Repurposing Knowledge Graph Embeddings for Triple Representation via Weak Supervision
The majority of knowledge graph embedding techniques treat entities and predicates as separate embedding matrices, using aggregation functions to build a representation of the input triple. However, these aggregations are lossy, i.e. they do not capture the semantics of the original triples, such as information contained in the predicates. To combat these shortcomings, current methods learn triple embeddings from scratch without utilizing entity and predicate embeddings from pre-trained models. In this paper, we design a novel fine-tuning approach for learning triple embeddings by creating weak supervision signals from pre-trained knowledge graph embeddings. We develop a method for automatically sampling triples from a knowledge graph and estimating their pairwise similarities from pre-trained embedding models. These pairwise similarity scores are then fed to a Siamese-like neural architecture to fine-tune triple representations. We evaluate the proposed method on two widely studied knowledge graphs and show consistent improvement over other state-of-the-art triple embedding methods on triple classification and triple clustering tasks.
['Yuan An', 'Alexander Kalinowski']
2022-08-22
null
null
null
null
['triple-classification', 'knowledge-graph-embedding', 'knowledge-graph-embeddings', 'knowledge-graph-embeddings']
['graphs', 'graphs', 'graphs', 'methodology']
[-1.89940825e-01 2.82487631e-01 -7.13050246e-01 -4.05342728e-01 -6.22394383e-01 -4.66984093e-01 6.54829681e-01 6.44160986e-01 -1.96020558e-01 5.56069672e-01 4.85177964e-01 -2.80034561e-02 -2.12816939e-01 -1.12800038e+00 -8.86239171e-01 -4.84489053e-01 2.04073489e-02 8.52372289e-01 1.77081630e-01 -2.83608913e-01 7.45705664e-02 3.20477970e-02 -1.60962474e+00 3.65245283e-01 1.01148367e+00 8.30318868e-01 -3.32034439e-01 3.33618701e-01 -2.52098590e-01 7.04145551e-01 -2.60089278e-01 -6.02371454e-01 1.08015485e-01 -3.87352258e-01 -7.90235996e-01 -1.20087206e-01 5.19181609e-01 7.42093474e-02 -7.07819283e-01 9.05011237e-01 2.60337859e-01 -8.35402533e-02 8.04438233e-01 -1.35234833e+00 -1.61286247e+00 8.53711426e-01 -4.70527351e-01 3.80573468e-03 4.59526867e-01 -2.22109392e-01 1.80435145e+00 -1.11901629e+00 6.95018053e-01 1.06623328e+00 8.00832629e-01 1.57670900e-01 -1.37289107e+00 -3.63544852e-01 -1.03475824e-01 5.15244842e-01 -1.55703449e+00 -2.30341256e-01 9.38786983e-01 -4.73182172e-01 1.24326360e+00 4.14188355e-02 7.78156459e-01 8.24385047e-01 -1.92603678e-01 7.04090893e-01 4.68468904e-01 -5.34626365e-01 4.35027108e-02 3.21469545e-01 3.62055808e-01 1.05822849e+00 6.07029140e-01 -1.88487604e-01 -6.51102483e-01 -3.92176986e-01 2.47543186e-01 9.71245840e-02 -2.17592701e-01 -1.29337656e+00 -1.41708696e+00 1.09992516e+00 8.52896214e-01 3.40239972e-01 -4.75352377e-01 2.19301537e-01 4.53057110e-01 2.84917861e-01 3.17028195e-01 2.30726689e-01 -4.62556571e-01 1.91493362e-01 -5.67567229e-01 1.18406020e-01 8.41505170e-01 9.35569465e-01 1.31023443e+00 -2.13820085e-01 -2.01563522e-01 7.12443531e-01 6.03732884e-01 3.02049041e-01 4.98322546e-01 -3.65716159e-01 6.56975567e-01 1.13867271e+00 -2.80048866e-02 -1.19942188e+00 -1.70217186e-01 5.90805616e-03 -2.94260234e-01 -3.11058253e-01 3.30626428e-01 2.45900467e-01 -1.01637363e+00 1.62384224e+00 4.87253785e-01 1.44265190e-01 2.09786221e-01 7.40021527e-01 8.08715045e-01 5.68277538e-01 -1.98272765e-02 3.94109368e-01 1.33627033e+00 -1.01474810e+00 -5.19039333e-01 -4.48409095e-02 1.01299787e+00 -3.55268747e-01 9.93027568e-01 -1.94091991e-01 -7.56609857e-01 -3.13938439e-01 -1.19265580e+00 -3.16741556e-01 -9.09959078e-01 2.39534304e-01 7.73200512e-01 6.92253828e-01 -8.81308556e-01 5.60660601e-01 -8.83905351e-01 -5.41038930e-01 4.13910329e-01 4.55960035e-01 -5.42154610e-01 -2.12747976e-01 -1.30893660e+00 8.18740070e-01 7.07947850e-01 5.37275523e-03 -6.69732094e-01 -9.96189892e-01 -1.36821663e+00 4.30090487e-01 3.93639416e-01 -8.77431273e-01 6.40900254e-01 -5.75231135e-01 -1.16847467e+00 9.20759320e-01 -2.04061747e-01 -2.76805729e-01 -2.73719996e-01 -1.86399966e-01 -5.42139471e-01 1.93396974e-02 2.17689782e-01 3.97066146e-01 6.55979216e-01 -1.28668261e+00 -3.55900824e-01 -4.00249779e-01 1.60279706e-01 6.22644275e-02 -7.11034954e-01 -3.29454035e-01 -5.29374301e-01 -3.38836908e-01 7.07910135e-02 -8.15746903e-01 1.30578816e-01 -1.74209446e-01 -6.27347529e-01 -4.35527384e-01 9.67702925e-01 -4.50168014e-01 1.27149189e+00 -2.01761270e+00 4.22286600e-01 6.14257216e-01 5.43826640e-01 3.99441242e-01 -2.73029685e-01 7.54333198e-01 -1.97072163e-01 2.65189499e-01 -6.52862042e-02 -1.84354708e-01 5.43522179e-01 5.50046325e-01 -1.34239376e-01 4.17572111e-01 4.91330832e-01 1.22776508e+00 -1.13987184e+00 -4.60624069e-01 1.52704448e-01 6.60793066e-01 -8.12960148e-01 -1.76282376e-02 -4.21129584e-01 -2.07029805e-01 -4.28844541e-01 6.37789369e-01 5.01755476e-01 -5.96953750e-01 7.57899404e-01 -7.19103038e-01 3.10015589e-01 3.55152369e-01 -1.07983613e+00 1.74820626e+00 -3.50531816e-01 2.42663145e-01 -5.62342763e-01 -1.46578765e+00 7.44024336e-01 2.42065862e-01 6.16266549e-01 -6.60530269e-01 8.91036764e-02 2.71960795e-01 -1.75662220e-01 -4.29846704e-01 5.54631233e-01 -2.39094943e-02 -2.28896126e-01 6.10917985e-01 6.43217206e-01 3.17295939e-01 4.88766640e-01 5.22724509e-01 1.18009603e+00 4.74873185e-02 2.59991318e-01 4.98173982e-02 4.84254062e-01 -2.96490099e-02 8.10549974e-01 3.15356344e-01 4.13113721e-02 4.76331800e-01 6.97495937e-01 -4.57455486e-01 -1.02817631e+00 -1.45840085e+00 1.22741520e-01 9.05241370e-01 1.17066421e-01 -1.02246010e+00 -2.64787138e-01 -1.25176036e+00 5.98943830e-01 3.30606878e-01 -9.04015183e-01 -4.69187945e-01 -3.75659555e-01 -5.04938424e-01 4.29138333e-01 7.38537312e-01 -1.09812301e-02 -6.80016518e-01 1.34622499e-01 2.68847458e-02 -5.21136820e-02 -1.12056303e+00 -5.45621991e-01 9.25356969e-02 -5.90590358e-01 -1.41947162e+00 -1.95300877e-01 -1.13691580e+00 8.35102916e-01 6.38259873e-02 1.24887955e+00 2.05474213e-01 -2.05157265e-01 5.02583563e-01 -4.35999721e-01 1.36081010e-01 -1.80682912e-02 8.87221247e-02 2.37725630e-01 1.33001074e-01 1.05695152e+00 -5.54620147e-01 -2.23355308e-01 -3.16870548e-02 -8.21879029e-01 -4.81521606e-01 5.39097309e-01 1.20126247e+00 7.00400352e-01 -4.04824270e-03 5.56172490e-01 -1.16223407e+00 6.25617564e-01 -5.18112361e-01 -4.51796889e-01 6.67987049e-01 -5.81902266e-01 5.24358869e-01 5.95983207e-01 -1.92478552e-01 -7.42255688e-01 8.10253471e-02 3.45479190e-01 -7.22850502e-01 3.47548425e-01 8.29212904e-01 -2.20382884e-01 -2.54985362e-01 6.21968031e-01 2.44237632e-02 -2.04387486e-01 -3.06308508e-01 1.01564264e+00 3.71063530e-01 2.34974727e-01 -6.22225344e-01 1.12689710e+00 3.52243543e-01 -3.05529714e-01 -6.62954211e-01 -9.91544425e-01 -6.21452630e-01 -8.16889942e-01 4.91618812e-01 8.52889776e-01 -8.84808362e-01 -5.66962838e-01 -2.19281375e-01 -9.35865462e-01 6.95606694e-02 -4.46151823e-01 6.36185229e-01 -4.75914925e-01 4.01168227e-01 -5.10114074e-01 -1.61767557e-01 -3.12633328e-02 -8.78150403e-01 9.96891081e-01 6.70399368e-02 -1.84620336e-01 -1.45038199e+00 5.49436212e-01 4.60242689e-01 1.49195969e-01 1.95561741e-02 1.30044949e+00 -8.24101686e-01 -6.92596197e-01 -2.77818054e-01 -3.63729984e-01 2.42786914e-01 2.95604795e-01 9.30902883e-02 -7.47632146e-01 -5.06110638e-02 -7.38994002e-01 -6.23863697e-01 9.26548541e-01 -4.97181267e-02 7.44124591e-01 -2.76562244e-01 -7.30423748e-01 5.77974439e-01 1.54125297e+00 -4.53951895e-01 4.35819238e-01 1.04439080e-01 1.19692242e+00 2.28290483e-01 3.41314346e-01 3.85809213e-01 8.70721221e-01 5.11826336e-01 2.60266900e-01 2.13418648e-01 -1.06232680e-01 -7.90709972e-01 2.03790233e-01 8.97788525e-01 1.43910289e-01 -1.05575155e-02 -9.84150648e-01 1.05368328e+00 -1.93942761e+00 -1.06123722e+00 1.17068507e-01 2.02854085e+00 1.03171539e+00 -1.11416660e-01 4.62780260e-02 1.32697150e-01 5.41315734e-01 1.25426605e-01 -1.76634923e-01 -2.61234164e-01 -1.76424254e-02 3.41421902e-01 4.04863149e-01 5.48336804e-01 -1.10617793e+00 1.00151300e+00 5.76612568e+00 4.29161578e-01 -8.31403852e-01 1.79171458e-01 -1.70911908e-01 4.02641296e-02 -9.64882612e-01 4.26813930e-01 -6.07101917e-01 1.98807657e-01 8.31856012e-01 -3.13817948e-01 2.76953220e-01 5.85325062e-01 -3.88236225e-01 3.55776459e-01 -1.48353791e+00 9.24446821e-01 4.21907783e-01 -1.60454726e+00 2.67060548e-01 2.30876114e-02 9.36601698e-01 4.99101579e-02 -1.14941776e-01 6.88212395e-01 5.93251884e-01 -9.24748778e-01 2.40687504e-01 4.94798154e-01 3.34871918e-01 -6.78907514e-01 6.57093287e-01 -3.07042360e-01 -1.45331371e+00 -3.84533517e-02 -5.62340438e-01 2.68123269e-01 1.69696689e-01 8.68560553e-01 -8.40954661e-01 9.52494025e-01 5.25987506e-01 1.17231143e+00 -6.73428833e-01 7.43053377e-01 -4.85426217e-01 4.20407444e-01 -4.00327027e-01 3.14398408e-02 2.11459368e-01 -2.55747318e-01 1.74673378e-01 9.52958405e-01 1.56803176e-01 -3.34516943e-01 3.15622121e-01 1.06663465e+00 -5.76510668e-01 7.49232024e-02 -8.89752150e-01 -7.56699562e-01 6.18394434e-01 1.26191628e+00 -3.15611511e-01 -4.29740936e-01 -9.38094616e-01 7.69603491e-01 9.23214972e-01 4.80812669e-01 -8.72602046e-01 -6.96736395e-01 8.14383864e-01 -2.76486147e-02 9.28014159e-01 -1.48972899e-01 -1.01501890e-01 -1.51364553e+00 3.10269326e-01 -5.15277267e-01 6.38648212e-01 -4.64180231e-01 -1.52348542e+00 1.02372736e-01 -2.62438983e-01 -9.66064632e-01 -1.04085147e-01 -5.69226146e-01 -5.75378001e-01 6.43246770e-01 -1.71355140e+00 -1.48889101e+00 -6.36702999e-02 6.27860308e-01 -4.62056458e-01 -1.74870417e-01 9.76950049e-01 6.30703270e-01 -5.49642861e-01 8.64634275e-01 1.93539321e-01 5.11872888e-01 6.85770750e-01 -1.44196677e+00 7.31150880e-02 5.96988201e-01 6.57236516e-01 9.21981394e-01 3.77830505e-01 -5.91113031e-01 -1.84152532e+00 -9.99372602e-01 1.35032654e+00 -7.54285872e-01 1.08188474e+00 -4.71964508e-01 -9.90361571e-01 1.23834062e+00 2.23842829e-01 5.89975476e-01 1.08887279e+00 7.93730915e-01 -9.40696239e-01 -2.87461936e-01 -9.62374389e-01 3.64534110e-01 1.00029635e+00 -1.01044810e+00 -9.02410030e-01 3.50699753e-01 6.59047961e-01 7.66725512e-03 -1.32602406e+00 2.30351076e-01 4.57380801e-01 -6.91031575e-01 1.07865393e+00 -9.44803655e-01 2.94802636e-01 -4.58691537e-01 -2.96248525e-01 -1.51654196e+00 -2.26191372e-01 -2.22375661e-01 -6.50911033e-01 1.22110808e+00 7.00743079e-01 -5.30560791e-01 1.08180594e+00 3.54278684e-01 -1.04593851e-01 -6.64612412e-01 -6.89221263e-01 -7.76239455e-01 -8.79089013e-02 1.31696522e-01 6.57765925e-01 1.35927176e+00 7.45604992e-01 6.56991005e-01 -1.38390422e-01 3.68528098e-01 8.34412038e-01 4.49321449e-01 6.84523284e-01 -1.14904881e+00 -8.74769464e-02 -8.99933428e-02 -1.00044239e+00 -6.09065473e-01 4.89529192e-01 -1.53114712e+00 -2.79921830e-01 -1.77029419e+00 2.07353219e-01 -4.75113690e-01 -5.12065768e-01 7.90555656e-01 -2.48912618e-01 1.69908017e-01 -8.72402489e-02 -2.04324007e-01 -8.40210557e-01 1.13205040e+00 8.38135183e-01 -4.84839469e-01 9.89573821e-03 -7.10331559e-01 -8.44653070e-01 3.08826715e-01 5.75006068e-01 -6.18059337e-01 -6.73349440e-01 -6.62596047e-01 4.25590485e-01 -3.94854397e-01 3.86896551e-01 -7.78386474e-01 4.23492134e-01 4.60266061e-02 1.42410725e-01 -3.36419165e-01 3.58047694e-01 -8.58588994e-01 -1.10354401e-01 1.13239475e-01 -5.22492118e-02 3.80233228e-02 -9.49517265e-02 9.03087199e-01 -5.87042689e-01 4.03576382e-02 3.37769717e-01 2.08304912e-01 -8.46290290e-01 4.53739882e-01 2.58280456e-01 3.77697140e-01 1.20134974e+00 -1.97715998e-01 -4.58398908e-01 1.79468580e-02 -9.50009406e-01 4.71432328e-01 5.75842857e-01 4.93335575e-01 7.60832250e-01 -1.90456688e+00 -3.81311297e-01 3.57514828e-01 6.55223072e-01 -2.49745384e-01 6.60344586e-02 8.99851680e-01 -3.37483615e-01 4.91808146e-01 -1.13037050e-01 -3.76915634e-01 -1.00019979e+00 9.35956776e-01 1.56910896e-01 -4.11698014e-01 -5.30925095e-01 6.87940121e-01 -1.05831422e-01 -9.53698456e-01 1.15629986e-01 -3.98896456e-01 -5.08404672e-02 1.65781230e-01 2.44397998e-01 4.81463224e-02 6.57349005e-02 -6.19091094e-01 -6.87156677e-01 7.24599123e-01 -1.22646153e-01 2.84532785e-01 1.37511802e+00 2.27184057e-01 -1.97549894e-01 3.22213799e-01 1.61538756e+00 2.14677185e-01 -6.06057465e-01 -6.50881827e-01 2.71806747e-01 -5.75494349e-01 -1.44828618e-01 -2.27324426e-01 -1.12842965e+00 7.88624287e-01 1.38734370e-01 2.56007500e-02 5.58715105e-01 2.77100325e-01 7.66006052e-01 7.16057420e-01 1.52081102e-01 -1.12927246e+00 2.82917291e-01 4.63628501e-01 3.44178349e-01 -1.27849364e+00 -2.38693208e-02 -4.84448224e-01 -7.21592963e-01 9.30998981e-01 4.50647771e-01 -2.91914582e-01 1.00910330e+00 -2.09823877e-01 -1.68144584e-01 -7.01119065e-01 -7.48975337e-01 -4.42090571e-01 4.26672190e-01 7.75849342e-01 5.14648497e-01 -1.31663932e-02 -1.99708238e-01 4.32268053e-01 -2.94350311e-02 5.27279042e-02 2.81590670e-01 9.22714412e-01 -3.74808878e-01 -1.35782218e+00 5.44337220e-02 6.63483858e-01 4.40711305e-02 -2.21901432e-01 -3.76193136e-01 6.10965729e-01 5.58656715e-02 6.26129627e-01 -3.73703204e-02 -6.32208288e-01 4.17609692e-01 3.17006916e-01 6.05743945e-01 -9.19633389e-01 -2.80274689e-01 -7.39232659e-01 3.44830722e-01 -6.04473412e-01 -4.82379198e-01 -5.33458173e-01 -1.30459678e+00 -4.07023996e-01 -3.21923256e-01 4.90441382e-01 2.03025296e-01 8.42769682e-01 6.32679045e-01 4.31030393e-01 3.73120010e-01 -6.03403747e-01 -6.26156926e-01 -6.39839768e-01 -6.87289655e-01 8.85799408e-01 1.35658592e-01 -9.96734083e-01 -2.11536214e-01 -8.96569863e-02]
[8.747817993164062, 7.881862163543701]
48a084b2-d0be-4993-92be-b280d3e254c9
dotat-a-domain-oriented-text-annotation-tool-1
null
null
https://aclanthology.org/2022.acl-demo.1
https://aclanthology.org/2022.acl-demo.1.pdf
DoTAT: A Domain-oriented Text Annotation Tool
We propose DoTAT, a domain-oriented text annotation tool. The tool designs and implements functions heavily in need in domain-oriented information extraction. Firstly, the tool supports a multi-person collaborative process with automatically merging and review, which can greatly improve the annotation accuracy. Secondly, the tool provides annotation of events, nested event and nested entity, which are frequently required in domain-related text structuring tasks. Finally, DoTAT provides visual annotation specification definition, automatic batch annotation and iterative annotation to improve annotation efficiency. Experiments on the ACE2005 dataset show that DoTAT can reduce the event annotation time by 19.7% compared with existing annotation tools. The accuracy without review is 84.09%, 1.35% higher than Brat and 2.59% higher than Webanno. The accuracy of DoTAT even reaches 93.76% with review. The demonstration video can be accessed from https://ecust-nlp-docker.oss-cn-shanghai.aliyuncs.com/dotat_demo.mp4. A live demo website is available at https://github.com/FXLP/MarkTool.
['Yi Wang', 'Wen Du', 'Tingting Cai', 'Ming Liang', 'Tong Ruan', 'Yupian Lin']
null
null
null
null
acl-2022-5
['text-annotation']
['natural-language-processing']
[-4.86569136e-01 2.30433494e-01 -1.33269921e-01 -3.60535592e-01 -8.54750693e-01 -7.15242445e-01 6.25675738e-01 3.63666952e-01 -4.99743909e-01 8.25556576e-01 4.47192371e-01 -7.43457377e-02 7.66527876e-02 -4.91701305e-01 -1.87227711e-01 -1.81973100e-01 3.38838071e-01 7.32444942e-01 3.51003289e-01 4.53259163e-02 7.05170184e-02 -1.41673714e-01 -1.24515224e+00 6.48849607e-01 9.07356620e-01 7.89296329e-01 5.05641997e-01 4.84243572e-01 -5.30193627e-01 8.45147729e-01 -7.98321009e-01 -3.29436183e-01 3.65561731e-02 -3.73410791e-01 -1.07800794e+00 -1.64580941e-01 4.81752418e-02 -1.20091170e-01 -5.20807765e-02 9.22034740e-01 6.03375733e-01 1.18581593e-01 4.05325085e-01 -1.46172965e+00 -3.59327257e-01 1.10289836e+00 -4.39512312e-01 1.40561000e-01 6.93043232e-01 -1.39602154e-01 1.00427103e+00 -1.11924088e+00 1.11085677e+00 9.45114911e-01 5.37147343e-01 3.38340819e-01 -5.84293306e-01 -8.43416691e-01 -8.30212235e-02 3.85530025e-01 -1.72526407e+00 -3.28901023e-01 3.75098616e-01 -6.42563105e-01 9.83272851e-01 4.12590146e-01 6.20461226e-01 8.09033275e-01 -1.53594524e-01 9.29945529e-01 7.07140326e-01 -4.78914559e-01 -2.21723337e-02 9.14596766e-02 3.77973944e-01 7.09242463e-01 5.03504500e-02 -6.93494916e-01 -6.28092468e-01 -3.61341499e-02 6.51668608e-01 4.66015525e-02 -1.90777525e-01 1.35874137e-01 -1.36172235e+00 3.83088797e-01 -8.63803029e-02 7.43515670e-01 -2.72610843e-01 -2.80575138e-02 7.59413123e-01 -4.41411063e-02 5.66322923e-01 2.56069154e-01 -7.29802966e-01 -6.48643076e-01 -7.73747504e-01 3.87784153e-01 7.07034647e-01 1.46431029e+00 5.32261729e-01 -1.89311802e-01 -2.05452606e-01 1.06018472e+00 2.40515500e-01 1.61170676e-01 4.70932424e-01 -1.01922274e+00 6.07309997e-01 1.04384995e+00 4.51213449e-01 -6.99726999e-01 -7.80914545e-01 5.34576289e-02 -4.78239894e-01 -1.75581276e-01 7.32480645e-01 -5.28949201e-01 -4.82974648e-01 1.16782725e+00 5.75166583e-01 -3.69049132e-01 -2.57624805e-01 8.20686102e-01 1.34000993e+00 7.13608921e-01 4.53396708e-01 -4.61512089e-01 1.77263904e+00 -8.89315188e-01 -1.53020310e+00 2.95512062e-02 1.14051735e+00 -1.06357741e+00 1.30026889e+00 4.66298252e-01 -1.16320217e+00 -4.67758030e-01 -6.10052824e-01 -2.58172482e-01 -5.62947631e-01 7.09055424e-01 4.43531573e-01 2.35149473e-01 -5.49609780e-01 9.28890929e-02 -7.05106020e-01 -9.75519180e-01 2.94883162e-01 -5.97001985e-02 -2.72664994e-01 2.79069752e-01 -1.25702381e+00 6.36272907e-01 8.48990202e-01 -3.63127202e-01 -3.83909643e-01 -8.05388808e-01 -6.72656476e-01 -8.61124601e-03 7.88421333e-01 -2.02776879e-01 1.90027106e+00 -6.60031557e-01 -1.21486270e+00 7.09157646e-01 -2.70502418e-01 -1.79949746e-01 6.77448750e-01 -7.20514715e-01 -8.08946788e-01 2.41043210e-01 4.22607362e-01 5.09989798e-01 6.40933439e-02 -6.96873307e-01 -9.92902279e-01 -1.42152205e-01 -1.11941174e-01 2.65466899e-01 -3.46299618e-01 7.83344209e-01 -9.31590438e-01 -7.74393022e-01 -5.03326297e-01 -6.46480083e-01 -1.51240220e-02 -4.98157851e-02 -4.50472444e-01 -8.10863078e-01 1.01645362e+00 -1.08493280e+00 1.86872458e+00 -2.09629607e+00 -3.75565976e-01 -1.59511343e-01 1.89174026e-01 2.67622679e-01 3.91608506e-01 9.64887619e-01 -2.31691390e-01 3.59956712e-01 1.65873170e-02 -1.68443784e-01 3.92787576e-01 -1.44565418e-01 1.10751875e-01 1.77084431e-01 -1.77356407e-01 8.89627934e-01 -8.97390842e-01 -1.13002098e+00 1.13604777e-01 2.42469355e-01 -2.59558380e-01 5.00762537e-02 -3.83133620e-01 3.61705929e-01 -6.47178352e-01 7.22096503e-01 3.72433245e-01 -4.54246640e-01 6.10244036e-01 -2.19793111e-01 -6.33319318e-01 3.21170360e-01 -1.57211077e+00 1.88649988e+00 -4.51138802e-02 7.59318531e-01 -9.26919430e-02 -7.41521418e-01 8.56001675e-01 7.89007246e-01 5.90160906e-01 -4.40311074e-01 1.57578900e-01 2.39955280e-02 -3.72121066e-01 -8.52302611e-01 5.86619079e-01 6.24891818e-01 -3.29718530e-01 5.43214977e-01 9.55998451e-02 7.68272430e-02 8.36328030e-01 6.45486116e-01 1.07178211e+00 4.30041045e-01 8.11456144e-01 -1.55555308e-01 4.01384026e-01 4.29747224e-01 8.97255719e-01 3.36312264e-01 -1.94375053e-01 2.19696373e-01 6.26417637e-01 -5.85271418e-01 -9.43772733e-01 -5.78578115e-01 -1.45017192e-01 1.21090925e+00 -1.40753284e-01 -1.35150135e+00 -7.67549872e-01 -8.81075263e-01 -4.72060651e-01 8.54809344e-01 -2.65265763e-01 5.83164096e-01 -4.39547956e-01 -3.06025475e-01 5.29356122e-01 5.03176391e-01 7.31150091e-01 -1.04689384e+00 -4.93993729e-01 1.92040518e-01 -7.31422544e-01 -1.16396034e+00 -7.15216458e-01 -3.72438692e-02 -2.72059113e-01 -1.27172923e+00 -3.70627791e-01 -8.08592916e-01 5.87377727e-01 -2.51041055e-01 9.25702989e-01 3.86237018e-02 -3.33162218e-01 4.25265729e-01 -7.82144487e-01 -6.40910268e-01 -4.39306442e-03 1.30517855e-02 -5.26683480e-02 -3.09662968e-01 7.27979422e-01 -1.93331122e-01 -4.27149951e-01 7.51504004e-01 -7.28488445e-01 4.80793715e-01 -5.72093949e-02 4.35530096e-01 5.56996763e-01 1.87171876e-01 5.61037302e-01 -9.94313836e-01 4.29914355e-01 -4.11025941e-01 -6.59473777e-01 1.31882340e-01 -5.52366138e-01 -5.33862829e-01 3.14330429e-01 -2.96593249e-01 -1.38255823e+00 3.32153648e-01 8.45454540e-03 -3.37372757e-02 -4.32218492e-01 6.88130677e-01 -3.40842187e-01 7.90113986e-01 6.36839032e-01 -1.25711873e-01 -5.77183604e-01 -7.31728554e-01 1.48520157e-01 9.37203169e-01 4.15067613e-01 -3.99501830e-01 4.74734038e-01 1.72070459e-01 -7.87059069e-01 -6.57768548e-01 -8.04095805e-01 -6.16325557e-01 -7.67113924e-01 -6.18738949e-01 9.74981487e-01 -9.75589693e-01 -7.67449796e-01 1.96551934e-01 -1.19214046e+00 -7.07438290e-01 -2.00309664e-01 4.72959191e-01 -2.88590550e-01 3.77391517e-01 -5.39579391e-01 -7.38922119e-01 -5.25956154e-01 -5.26718557e-01 6.83968246e-01 3.64709467e-01 -9.19581175e-01 -7.70000637e-01 -1.96646035e-01 5.48633993e-01 -1.96287677e-01 1.16829410e-01 2.95265377e-01 -1.19192803e+00 -2.08943650e-01 -8.16762522e-02 -9.84880701e-02 -1.26307398e-01 -1.04230464e-01 3.88885856e-01 -4.94123548e-01 1.70649543e-01 -6.50048494e-01 -1.04067907e-01 1.16073295e-01 2.18106017e-01 1.05016458e+00 -4.42476183e-01 -7.27489889e-01 3.38066630e-02 9.68399942e-01 6.37707472e-01 5.09562075e-01 4.63596731e-01 5.68721294e-01 5.43660820e-01 1.40452540e+00 1.03971434e+00 5.57460189e-01 9.81908202e-01 7.11190999e-02 1.89236820e-01 -1.03469692e-01 -3.00929159e-01 3.02200228e-01 9.33472276e-01 -1.40572339e-01 -4.01027918e-01 -1.56241274e+00 7.72113502e-01 -2.32570171e+00 -1.06313360e+00 -8.36111009e-01 1.59640980e+00 1.02407801e+00 -1.66069090e-01 4.35226351e-01 2.06664920e-01 9.29574013e-01 -1.81269035e-01 -9.61002484e-02 -1.18536107e-01 9.44859236e-02 -1.17012531e-01 1.40855983e-01 3.21760118e-01 -1.18558145e+00 1.08037078e+00 4.90114546e+00 1.05080545e+00 -5.25138438e-01 4.96474802e-01 2.12031037e-01 -2.64452189e-01 2.56641895e-01 -3.92399766e-02 -1.13133037e+00 7.11731195e-01 9.25559819e-01 -5.67791879e-01 -4.75231595e-02 9.21073318e-01 7.65358746e-01 -2.62610942e-01 -7.23487616e-01 9.61212516e-01 -1.85025647e-01 -1.50583243e+00 -3.68534982e-01 2.78186705e-03 5.44278502e-01 -8.61345157e-02 -6.86887980e-01 4.33498859e-01 5.53137481e-01 -5.18937290e-01 9.66933310e-01 5.14454365e-01 8.52890313e-01 -7.27213264e-01 6.87323093e-01 3.63452047e-01 -1.54874229e+00 1.39115512e-01 6.16468415e-02 -1.38488039e-03 5.49704552e-01 7.95153260e-01 -1.05531931e+00 6.45073652e-01 1.22997928e+00 8.46262038e-01 -4.65589076e-01 1.10570824e+00 -5.22643626e-01 6.01412296e-01 -3.80107015e-01 -1.68319315e-01 -2.33769253e-01 -1.18197083e-01 5.65653145e-01 1.54541934e+00 3.62100840e-01 3.99206966e-01 4.18105394e-01 3.21232289e-01 -8.49523470e-02 5.27892053e-01 -2.11014077e-01 -3.12238991e-01 8.79418135e-01 1.59531355e+00 -7.71791101e-01 -6.28077447e-01 -3.58510494e-01 7.90555239e-01 6.87985346e-02 2.33718053e-01 -1.20749784e+00 -8.31412554e-01 2.32729882e-01 2.62782991e-01 3.64850134e-01 -3.28521430e-01 -8.76354501e-02 -1.08795869e+00 4.51078489e-02 -7.90737748e-01 7.84087658e-01 -1.38366497e+00 -7.38847733e-01 4.92391974e-01 1.45653680e-01 -1.30784392e+00 -1.78066254e-01 -3.58971208e-01 -4.74306166e-01 4.45305884e-01 -5.66480815e-01 -1.30195475e+00 -5.32105625e-01 7.03522384e-01 8.47655952e-01 9.60652232e-02 7.67380357e-01 8.82975876e-01 -8.91304612e-01 3.08137596e-01 -2.54324347e-01 5.13913751e-01 1.17947412e+00 -1.37087858e+00 3.21752094e-02 8.40156138e-01 3.06330714e-02 3.19939226e-01 5.55536687e-01 -1.14806449e+00 -8.58275712e-01 -1.11140907e+00 1.51568174e+00 -6.36539936e-01 9.61560249e-01 -4.32024717e-01 -7.95026183e-01 1.07346821e+00 5.97932160e-01 -1.71278387e-01 8.88705015e-01 -7.36194178e-02 -2.45027002e-02 1.11832954e-02 -9.35208023e-01 6.90272987e-01 1.06172979e+00 -3.26472193e-01 -6.19760692e-01 7.02290177e-01 6.69837236e-01 -6.04839504e-01 -1.24907517e+00 7.64463544e-02 4.49200898e-01 -4.24056798e-01 4.78610039e-01 -2.40262106e-01 1.77588880e-01 -7.94985831e-01 1.59332380e-02 -8.86469483e-01 -1.17115349e-01 -8.66938293e-01 -1.20409064e-01 1.99700201e+00 5.79003632e-01 -2.78674930e-01 3.46272141e-01 5.50309062e-01 -2.89337277e-01 -1.37941435e-01 -6.89188898e-01 -6.73411667e-01 -6.04529023e-01 -8.34125578e-01 4.70439494e-01 1.40810049e+00 7.42080271e-01 5.02013206e-01 -2.94135213e-01 8.77080411e-02 1.81694701e-01 -3.21370274e-01 9.02274132e-01 -1.31157780e+00 6.99488372e-02 -9.81904194e-02 2.43665397e-01 -7.20829368e-01 -2.33048022e-01 -7.64868736e-01 -3.25700670e-01 -2.01345968e+00 9.22805723e-03 -1.09197222e-01 1.53736651e-01 1.10993361e+00 1.89147189e-01 1.10524058e-01 2.00031027e-01 4.24523234e-01 -1.18950152e+00 1.95854932e-01 1.03874075e+00 1.20547965e-01 -2.40282491e-01 -2.29686096e-01 -4.38875496e-01 9.09202754e-01 1.39071417e+00 -5.17508447e-01 -2.09641410e-03 -3.26861024e-01 3.52053493e-01 -2.16288894e-01 3.83411758e-02 -8.88030767e-01 4.98897910e-01 -2.17281371e-01 3.54532868e-01 -1.00641346e+00 1.53670639e-01 -9.48712349e-01 5.63489199e-01 2.85691828e-01 -2.01740175e-01 1.90699086e-01 2.89685458e-01 1.53175443e-01 -1.57774150e-01 -4.14438754e-01 2.92427629e-01 -1.51742414e-01 -9.99875903e-01 -3.65815051e-02 -8.06737006e-01 1.49292335e-01 1.27149343e+00 -1.00428507e-01 -5.71400404e-01 -2.80869216e-01 -9.48207438e-01 4.73391503e-01 2.00436801e-01 3.72095972e-01 7.76762739e-02 -1.36964786e+00 -5.58504105e-01 -3.89147937e-01 3.60730559e-01 -3.29895802e-02 2.50401497e-01 9.89477575e-01 -6.22094333e-01 4.48909402e-01 2.62327176e-02 -3.69473219e-01 -1.67493570e+00 2.53190249e-01 -1.02220125e-01 -4.04455960e-01 -7.01410115e-01 3.86426330e-01 -9.65796039e-02 -2.82737821e-01 2.33524635e-01 -1.86125606e-01 -3.55084449e-01 4.35000092e-01 7.96486735e-01 6.27443373e-01 -1.21081593e-02 -3.92070472e-01 -5.16608000e-01 3.34669143e-01 -8.33758246e-03 -2.20526487e-01 1.33375061e+00 -3.27114731e-01 -1.86510086e-02 4.76605207e-01 5.30372798e-01 2.68227190e-01 -7.48443842e-01 -1.70531884e-01 4.32459682e-01 -1.97192073e-01 -1.27177328e-01 -1.40024316e+00 -6.42323017e-01 4.43858922e-01 2.38952547e-01 4.57209766e-01 9.92389262e-01 2.67378896e-01 5.10717511e-01 2.64929563e-01 1.03886724e-01 -1.59118450e+00 -1.14856251e-01 6.53942227e-01 1.12497807e+00 -1.08183432e+00 1.48174703e-01 -5.72853327e-01 -8.60642850e-01 9.21522260e-01 8.81469727e-01 4.68794018e-01 5.12585282e-01 3.83563608e-01 1.13682218e-01 -3.94083530e-01 -8.53967607e-01 -2.23983601e-01 2.28071094e-01 4.67595816e-01 6.99848473e-01 3.02266944e-02 -9.24744666e-01 1.06230330e+00 -1.72025383e-01 3.08265299e-01 4.40865308e-01 1.08802056e+00 -3.28659117e-01 -1.05847120e+00 -3.78957778e-01 2.10162640e-01 -7.52852321e-01 -8.54496136e-02 -4.23508883e-01 1.06229496e+00 2.66448200e-01 9.31024849e-01 2.98479378e-01 -7.86142610e-03 4.88452375e-01 4.01963741e-01 -1.63051412e-01 -7.02194452e-01 -8.79152954e-01 5.62775135e-01 8.47579718e-01 -4.84243840e-01 -4.94293630e-01 -8.58319640e-01 -1.67865694e+00 -1.91890344e-01 -1.13492206e-01 6.22101188e-01 6.11758351e-01 6.06892347e-01 5.51426291e-01 5.45409262e-01 -9.29429457e-02 -4.53745037e-01 6.00126386e-01 -1.19642866e+00 -4.62711513e-01 1.54511958e-01 -4.53580618e-01 -5.09149313e-01 -9.23837498e-02 6.29413188e-01]
[9.314022064208984, 9.049840927124023]
2b4fd209-eedc-4f66-84fa-1664666c91da
restorex-ai-a-contrastive-approach-towards
2204.01719
null
https://arxiv.org/abs/2204.01719v1
https://arxiv.org/pdf/2204.01719v1.pdf
RestoreX-AI: A Contrastive Approach towards Guiding Image Restoration via Explainable AI Systems
Modern applications such as self-driving cars and drones rely heavily upon robust object detection techniques. However, weather corruptions can hinder the object detectability and pose a serious threat to their navigation and reliability. Thus, there is a need for efficient denoising, deraining, and restoration techniques. Generative adversarial networks and transformers have been widely adopted for image restoration. However, the training of these methods is often unstable and time-consuming. Furthermore, when used for object detection (OD), the output images generated by these methods may provide unsatisfactory results despite image clarity. In this work, we propose a contrastive approach towards mitigating this problem, by evaluating images generated by restoration models during and post training. This approach leverages OD scores combined with attention maps for predicting the usefulness of restored images for the OD task. We conduct experiments using two novel use-cases of conditional GANs and two transformer methods that probe the robustness of the proposed approach on multi-weather corruptions in the OD task. Our approach achieves an averaged 178 percent increase in mAP between the input and restored images under adverse weather conditions like dust tornadoes and snowfall. We report unique cases where greater denoising does not improve OD performance and conversely where noisy generated images demonstrate good results. We conclude the need for explainability frameworks to bridge the gap between human and machine perception, especially in the context of robust object detection for autonomous vehicles.
['Ketan Kotecha', 'Rahee Walambe', 'Pushkar Jain', 'Aboli Marathe']
2022-04-03
null
null
null
null
['robust-object-detection']
['computer-vision']
[ 4.36215132e-01 1.63514495e-01 4.38286245e-01 -1.37871012e-01 -7.61479318e-01 -5.94064653e-01 8.51069927e-01 -2.84881353e-01 -4.51993763e-01 7.54340351e-01 -1.68968141e-01 -3.77637208e-01 1.75414562e-01 -8.57223451e-01 -1.00321877e+00 -8.94385934e-01 2.29828984e-01 -2.87376996e-03 2.83177793e-01 -4.98520225e-01 1.76902324e-01 5.62657475e-01 -1.95621908e+00 -5.95749952e-02 1.11080635e+00 1.00458217e+00 1.53452486e-01 7.49502659e-01 3.19340706e-01 7.90084541e-01 -9.27569985e-01 -4.83948886e-01 5.67502975e-01 -2.01180711e-01 -1.18819468e-01 1.68078378e-01 7.44107902e-01 -4.35613632e-01 -3.31349701e-01 1.06726766e+00 5.68747640e-01 1.96687728e-01 6.70201242e-01 -1.56938291e+00 -5.45673847e-01 -1.82383150e-01 -3.71064276e-01 4.30558532e-01 1.59086972e-01 6.98335290e-01 5.35866737e-01 -7.17399716e-01 2.58055300e-01 1.21315622e+00 6.44407749e-01 4.66649771e-01 -1.14792573e+00 -8.21577549e-01 -2.87124235e-02 1.35748252e-01 -8.37855041e-01 -7.39593089e-01 5.14609456e-01 -4.56037760e-01 8.29805076e-01 1.09747082e-01 3.32137465e-01 1.15954757e+00 3.74822229e-01 3.78783584e-01 1.28744018e+00 -8.64799917e-02 1.50541499e-01 3.17145348e-01 -4.42619950e-01 4.38889921e-01 3.59884113e-01 7.14004815e-01 -4.82191175e-01 3.57681543e-01 3.25260967e-01 -1.49963781e-01 -1.39198065e-01 -1.66955143e-01 -8.96318078e-01 6.70686960e-01 7.07416475e-01 -2.44578108e-01 -4.93931621e-01 1.81644529e-01 8.57151747e-02 4.90159094e-01 5.26885390e-01 4.25082952e-01 7.61220455e-02 1.12606503e-01 -8.75302136e-01 3.22604269e-01 3.74778211e-01 6.04845524e-01 8.29559326e-01 5.53358912e-01 -5.35146669e-02 5.20562768e-01 2.11940557e-01 8.13598394e-01 1.13322228e-01 -8.42316210e-01 4.60040033e-01 2.50437498e-01 3.58353674e-01 -1.11980951e+00 -7.92959407e-02 -5.53412616e-01 -6.82019949e-01 9.67254400e-01 2.23711878e-01 -8.06704834e-02 -1.29919922e+00 1.69911015e+00 3.75856310e-01 2.62109935e-01 3.43639523e-01 9.78578806e-01 6.95869207e-01 5.93673289e-01 2.88178712e-01 1.97042391e-01 1.07450926e+00 -7.57409155e-01 -7.03505635e-01 -6.95529997e-01 1.22812822e-01 -6.74207449e-01 1.02120352e+00 3.19657117e-01 -1.03961325e+00 -6.61144912e-01 -1.53533161e+00 1.52793646e-01 -4.83198613e-01 -9.46889147e-02 2.76113808e-01 9.05085146e-01 -1.04499114e+00 2.92228848e-01 -5.91882944e-01 -2.94846982e-01 5.55611193e-01 2.58519709e-01 -4.40444529e-01 -3.17836076e-01 -1.08779132e+00 1.46092117e+00 3.34256589e-02 4.02025402e-01 -1.39539278e+00 -5.14781594e-01 -9.40077662e-01 -2.26363569e-01 6.64249584e-02 -5.76961815e-01 9.21539962e-01 -1.03649998e+00 -1.10659730e+00 6.98988795e-01 9.82694328e-02 -8.22325766e-01 7.92036057e-01 -3.49632412e-01 -4.93964463e-01 2.13818792e-02 3.61463994e-01 9.03623998e-01 1.40789962e+00 -1.53063250e+00 -8.32846224e-01 -2.62641430e-01 2.18755022e-01 3.22831124e-01 -2.25537196e-02 -2.98476756e-01 -6.77886382e-02 -5.49045622e-01 4.53522541e-02 -9.32129681e-01 -1.50026366e-01 -3.56950313e-02 -2.84366786e-01 2.85494715e-01 1.11266088e+00 -7.06470966e-01 5.29911876e-01 -2.22764182e+00 -2.60259658e-01 1.80443358e-02 -1.78804956e-02 2.62872249e-01 -2.33669013e-01 7.79444799e-02 9.10338983e-02 1.57552689e-01 -3.92436445e-01 -3.30817908e-01 -1.36855870e-01 3.34975511e-01 -5.66325009e-01 6.58086598e-01 7.36694455e-01 8.17354202e-01 -5.26037991e-01 -1.08602352e-01 4.49140400e-01 7.45577872e-01 -2.99100012e-01 2.87049025e-01 -2.83658132e-02 5.89620590e-01 1.67728290e-02 6.94755256e-01 7.58170784e-01 3.87655526e-01 -3.83922368e-01 -1.92186415e-01 -7.94859156e-02 2.82506704e-01 -8.62847686e-01 1.04169822e+00 -5.88758111e-01 9.50636506e-01 1.05086505e-01 -9.61911619e-01 8.23080242e-01 5.57717308e-02 -8.44279975e-02 -1.30374897e+00 -8.47422611e-03 1.00898005e-01 1.49130067e-02 -5.18106401e-01 7.13355362e-01 -3.65092009e-01 1.97512507e-02 1.00382820e-01 -2.11358994e-01 -4.86719161e-01 9.62681696e-02 1.91335469e-01 1.10596049e+00 2.25212172e-01 -1.84912100e-01 3.83323207e-02 1.94614604e-01 3.19395699e-02 3.97824436e-01 7.97959328e-01 -3.55371773e-01 8.32106829e-01 2.83564717e-01 -2.69596636e-01 -1.18158221e+00 -1.12685251e+00 -8.52820203e-02 7.96913505e-01 3.81910682e-01 3.92058522e-01 -6.33260965e-01 -5.78298926e-01 5.64852245e-02 9.12763357e-01 -7.38327682e-01 -5.47218621e-01 -4.51154739e-01 -9.57410038e-01 8.00285757e-01 4.84786510e-01 6.32314444e-01 -1.13428795e+00 -8.77021730e-01 1.02809884e-01 -2.61635512e-01 -1.29817784e+00 -3.55044603e-02 1.87065288e-01 -6.09079838e-01 -9.53727424e-01 -5.89990735e-01 -5.31850278e-01 8.02854300e-01 6.20619953e-01 1.09056008e+00 2.59628028e-01 -3.28896761e-01 3.31689864e-01 -2.49237761e-01 -5.22317052e-01 -6.32485330e-01 -3.67795318e-01 8.72686803e-02 -3.80244181e-02 -2.11624041e-01 -5.07114828e-01 -8.28395784e-01 5.90926170e-01 -1.12027669e+00 -2.06744939e-01 7.70715058e-01 7.58015156e-01 2.40553305e-01 2.90948629e-01 8.10432732e-01 -4.49042588e-01 5.21582782e-01 -4.79596406e-01 -7.34257102e-01 -1.58476815e-01 -6.78626597e-01 -1.20594334e-02 3.93866003e-01 -2.51551658e-01 -1.14654958e+00 -6.49875924e-02 -1.19960956e-01 -2.24036500e-01 -2.17815295e-01 1.56960890e-01 -1.53051138e-01 -2.76504785e-01 9.00411010e-01 1.85927197e-01 9.49855223e-02 8.45937729e-02 2.45419085e-01 4.33801442e-01 8.37003589e-01 -9.94994715e-02 1.28650355e+00 6.35533512e-01 -3.50721478e-02 -7.88042843e-01 -5.74031174e-01 -8.34078714e-02 9.85927135e-03 -4.66557354e-01 7.59594679e-01 -1.17879653e+00 -4.22627717e-01 4.93873745e-01 -9.53013003e-01 -3.29884201e-01 -1.46761581e-01 8.64726603e-02 -3.14100564e-01 2.57256925e-01 -9.35305431e-02 -1.05299640e+00 -1.83044761e-01 -1.22338569e+00 9.87302780e-01 2.68155277e-01 1.95599645e-01 -5.78912914e-01 -3.23843360e-01 6.92188978e-01 5.91401219e-01 4.34253544e-01 6.15058899e-01 -1.49819598e-01 -8.10286701e-01 -3.22032988e-01 -2.82490641e-01 5.45400083e-01 1.90593358e-02 -1.74154043e-01 -1.32448518e+00 -3.28314215e-01 -1.31834373e-01 -3.84100944e-01 1.06211793e+00 2.25977823e-01 4.89221394e-01 -2.39636213e-01 6.71899170e-02 4.07513618e-01 1.33770430e+00 2.34807894e-01 1.13559997e+00 5.81048965e-01 4.98438329e-01 7.92816877e-01 8.15154314e-01 -1.55460499e-02 3.89598578e-01 4.79392350e-01 9.85501051e-01 -3.14411968e-01 -3.54121625e-01 -2.33311728e-01 7.08938599e-01 -9.27847847e-02 2.95308940e-02 -6.86504066e-01 -7.08719909e-01 7.86051452e-01 -1.49060833e+00 -8.81365359e-01 -2.69104280e-02 2.28753066e+00 2.91661680e-01 4.38200951e-01 -1.10644981e-01 3.16152722e-01 5.58551311e-01 7.26689175e-02 -6.14255905e-01 -1.89271390e-01 -4.14160073e-01 5.96716022e-03 7.07321346e-01 4.53613222e-01 -1.05380404e+00 7.90843427e-01 6.11201906e+00 4.93498206e-01 -1.07444477e+00 2.15113550e-01 7.90728390e-01 -3.20255943e-02 -2.85709918e-01 -7.64129236e-02 -6.33971393e-01 5.60269535e-01 8.99345040e-01 3.13530773e-01 2.27100700e-01 6.22261584e-01 3.68761420e-01 -5.61050117e-01 -6.89601421e-01 8.32714021e-01 2.08745539e-01 -1.04952288e+00 -1.88311502e-01 1.32459486e-02 7.36063242e-01 1.21829100e-01 6.04669034e-01 3.52939546e-01 3.48572522e-01 -1.13365006e+00 9.17613149e-01 3.74751508e-01 6.57605231e-01 -8.07719469e-01 7.78551877e-01 2.07053393e-01 -7.76657403e-01 -9.42095369e-02 -2.83995092e-01 2.12543383e-02 1.76224872e-01 5.62535048e-01 -9.89049375e-01 2.24874452e-01 8.64554703e-01 3.32794040e-01 -6.58182263e-01 8.62993240e-01 -5.07296920e-01 5.37953019e-01 -2.72750586e-01 4.18617100e-01 1.74978554e-01 3.40167768e-02 8.34150195e-01 8.79437566e-01 4.11674500e-01 -2.02854425e-01 -1.90474167e-01 8.43400180e-01 6.64714351e-02 -4.79712248e-01 -9.79112446e-01 2.30812147e-01 2.17621610e-01 1.12483561e+00 -7.79167771e-01 -1.81174278e-01 -2.35710204e-01 8.97115886e-01 -6.21409230e-02 4.78560328e-01 -1.15045273e+00 -9.00745243e-02 8.37683558e-01 3.10858548e-01 4.05564219e-01 -1.09814584e-01 -4.70905989e-01 -7.41403580e-01 2.38174513e-01 -9.48955357e-01 -4.48651835e-02 -1.01282024e+00 -1.06884742e+00 7.26594388e-01 -2.50908852e-01 -1.39769292e+00 -2.34791875e-01 -3.08781862e-01 -5.78549027e-01 6.98493540e-01 -1.94751108e+00 -1.10586071e+00 -5.86078286e-01 2.84017742e-01 5.58103144e-01 -1.20214835e-01 2.72327095e-01 3.63436908e-01 -4.29859012e-01 4.40227330e-01 -1.75148219e-01 -3.21325749e-01 6.11080468e-01 -1.07402861e+00 6.26122177e-01 1.43866611e+00 -3.82157452e-02 1.42287105e-01 1.18921173e+00 -6.68841183e-01 -1.23464179e+00 -1.36221385e+00 3.12907904e-01 -4.50747758e-01 1.87440902e-01 -2.19198287e-01 -6.85598016e-01 4.04773504e-01 3.10423583e-01 -1.96668338e-02 5.61837554e-02 -3.54765177e-01 -2.82223254e-01 -2.63777077e-01 -1.42651129e+00 6.05553746e-01 6.82829618e-01 -4.38845187e-01 -3.54995579e-01 1.02392748e-01 5.87665260e-01 -2.90959716e-01 -3.13257545e-01 5.28756022e-01 4.61037219e-01 -1.37469864e+00 1.10258210e+00 -1.31311089e-01 4.26228315e-01 -6.05813026e-01 -1.65834695e-01 -1.32349575e+00 7.16893673e-02 -4.01534110e-01 2.89519966e-01 1.05601597e+00 5.37172914e-01 -8.10294688e-01 6.29265487e-01 5.20961940e-01 -3.40560377e-01 -1.80340543e-01 -9.91850078e-01 -6.75045788e-01 -1.26020759e-01 -6.40114307e-01 2.58520156e-01 5.03849089e-01 -7.79870689e-01 1.03753075e-01 -5.21219969e-01 6.32988632e-01 7.15211630e-01 -2.53312647e-01 9.97976303e-01 -9.38906252e-01 3.57523747e-02 -1.07347339e-01 -7.58325815e-01 -6.55250907e-01 -6.74802158e-03 -3.72635871e-01 5.11508048e-01 -1.40489829e+00 -2.21366122e-01 -3.94633442e-01 -4.70067896e-02 4.59111005e-01 -1.90642521e-01 8.61257195e-01 2.04797000e-01 1.33404344e-01 -3.12754512e-01 5.74054837e-01 9.57397461e-01 -3.36888283e-01 5.32981753e-02 4.06935588e-02 -6.19189620e-01 5.08166671e-01 8.57076645e-01 -6.21489525e-01 -4.63322759e-01 -4.21584219e-01 4.48576927e-01 -1.96202129e-01 1.01516473e+00 -1.36995840e+00 8.83855447e-02 6.99953660e-02 4.13464338e-01 -4.08057183e-01 4.96751070e-01 -8.05942476e-01 1.47022009e-01 4.84048486e-01 8.21091011e-02 1.09718598e-01 4.85655248e-01 7.92456329e-01 -2.78091490e-01 1.52328283e-01 1.02351642e+00 2.70883162e-02 -8.38224411e-01 -2.72564832e-02 -5.61479628e-01 -1.10599622e-01 9.78861749e-01 -5.58602870e-01 -3.29632461e-01 -7.05906212e-01 -5.32061636e-01 1.30205348e-01 5.92261314e-01 5.76477885e-01 8.00007939e-01 -9.88722444e-01 -7.15720356e-01 4.53814119e-01 7.14207366e-02 -5.73499575e-02 3.34678769e-01 8.25525701e-01 -3.95705253e-01 -1.56888962e-02 -3.53671163e-01 -6.80651486e-01 -1.06032073e+00 3.32051605e-01 4.55588073e-01 1.23028316e-01 -4.65473056e-01 6.96894467e-01 3.57851982e-01 -1.43351808e-01 1.28362551e-01 -2.28371799e-01 -3.52480300e-02 7.47441202e-02 3.45303118e-01 3.31179053e-01 5.31714559e-01 -7.64958262e-01 -1.40133828e-01 1.95413470e-01 -2.54081395e-02 -2.70497233e-01 1.16358614e+00 -4.25619662e-01 2.33989239e-01 -2.11699307e-02 6.93177760e-01 -1.78316146e-01 -1.67007732e+00 1.65849075e-01 -3.31445992e-01 -4.88381028e-01 2.88512319e-01 -9.21405494e-01 -1.22448373e+00 7.60380566e-01 1.01393986e+00 2.51453519e-01 1.24212325e+00 -2.58377194e-01 7.26158023e-01 1.76631153e-01 1.62336960e-01 -8.07090104e-01 1.73281059e-01 2.22801849e-01 8.93790662e-01 -1.63905001e+00 -1.75787300e-01 -1.45597115e-01 -7.54201889e-01 5.57860255e-01 6.63139820e-01 -2.27220342e-01 1.33578867e-01 3.18132937e-01 3.88625413e-01 -9.67315584e-02 -6.02136791e-01 -4.67923433e-01 2.08601400e-01 9.24973726e-01 -9.18510184e-02 -1.90836027e-01 2.20611200e-01 1.04103498e-01 -2.41145119e-01 -4.70913619e-01 6.05621099e-01 9.57772493e-01 -4.11601782e-01 -5.57903767e-01 -6.95533633e-01 2.81667888e-01 -3.51419806e-01 -1.95893601e-01 -2.24647462e-01 7.41160512e-01 2.74755955e-01 1.34471619e+00 1.10005075e-02 -4.27286088e-01 4.50446546e-01 -1.12673074e-01 2.29711130e-01 -2.18267143e-01 -5.06434023e-01 -1.86485857e-01 1.41194656e-01 -4.78958875e-01 -5.48591375e-01 -5.34552336e-01 -9.43536878e-01 -1.79942608e-01 -3.81260782e-01 -3.47507894e-01 9.51757312e-01 9.63224471e-01 3.00142467e-01 6.64176404e-01 5.27446628e-01 -1.09440374e+00 -3.61493826e-01 -8.43779624e-01 -2.76717216e-01 3.98264974e-01 7.68563509e-01 -9.21870708e-01 -6.55677855e-01 8.05269182e-02]
[8.287908554077148, -1.4129247665405273]
746c60b2-5fe1-4509-b47e-8dc527fdcc1a
a-prototypical-semantic-decoupling-method-via
2302.13610
null
https://arxiv.org/abs/2302.13610v2
https://arxiv.org/pdf/2302.13610v2.pdf
A Prototypical Semantic Decoupling Method via Joint Contrastive Learning for Few-Shot Name Entity Recognition
Few-shot named entity recognition (NER) aims at identifying named entities based on only few labeled instances. Most existing prototype-based sequence labeling models tend to memorize entity mentions which would be easily confused by close prototypes. In this paper, we proposed a Prototypical Semantic Decoupling method via joint Contrastive learning (PSDC) for few-shot NER. Specifically, we decouple class-specific prototypes and contextual semantic prototypes by two masking strategies to lead the model to focus on two different semantic information for inference. Besides, we further introduce joint contrastive learning objectives to better integrate two kinds of decoupling information and prevent semantic collapse. Experimental results on two few-shot NER benchmarks demonstrate that PSDC consistently outperforms the previous SOTA methods in terms of overall performance. Extensive analysis further validates the effectiveness and generalization of PSDC.
['Weiran Xu', 'QiXiang Gao', 'Xinyue Cui', 'Keqing He', 'Tingfeng Hui', 'Xuefeng Li', 'Chen Zeng', 'Yuxiang Wu', 'Dayuan Fu', 'Daichi Guo', 'LiWen Wang', 'Zechen Wang', 'Guanting Dong']
2023-02-27
null
null
null
null
['few-shot-ner']
['natural-language-processing']
[-1.59095317e-01 3.16459909e-02 -2.04747766e-01 -4.91137743e-01 -5.88994145e-01 -4.91910577e-01 6.11484349e-01 3.94616187e-01 -7.19572961e-01 7.51765668e-01 2.37176761e-01 -2.80629955e-02 -1.32756401e-02 -7.68719375e-01 -3.13541114e-01 -4.04697150e-01 1.09331645e-01 4.11740154e-01 4.74353939e-01 -2.58111000e-01 2.01900885e-01 2.33331859e-01 -1.35905588e+00 1.49253890e-01 1.08503342e+00 3.65332693e-01 3.66748780e-01 1.56163856e-01 -7.49174595e-01 8.23803306e-01 -6.71995461e-01 -6.82735443e-01 -1.27932698e-01 -3.38408679e-01 -9.88510489e-01 -2.88103819e-01 1.37745500e-01 8.16832185e-02 -4.34368737e-02 1.18647122e+00 6.99510098e-01 7.68884957e-01 4.90295678e-01 -1.09620333e+00 -9.41120505e-01 9.61053967e-01 -3.17658186e-01 2.95222461e-01 1.33742347e-01 5.49458899e-03 1.17670929e+00 -1.15093052e+00 7.15312302e-01 1.17181909e+00 8.98116529e-01 8.40857267e-01 -1.03489709e+00 -7.01823831e-01 4.13567305e-01 5.66619575e-01 -1.57607055e+00 -3.47242504e-01 6.43112361e-01 -1.44024268e-01 1.16229820e+00 3.33828069e-02 1.06754668e-01 1.11402822e+00 -5.69136381e-01 8.36588264e-01 9.47715104e-01 -5.84652781e-01 5.75707316e-01 1.94152012e-01 7.98034847e-01 5.91883719e-01 2.95306474e-01 -1.72533765e-01 -1.51372924e-01 -1.35386765e-01 2.10881323e-01 -4.13520783e-02 -1.00908741e-01 -1.62115753e-01 -1.00926113e+00 7.65874863e-01 4.01503503e-01 6.51472032e-01 -3.40807319e-01 -3.28647107e-01 5.53569615e-01 -1.83801372e-02 3.31797868e-01 5.99852264e-01 -5.48981309e-01 -1.22111253e-02 -7.35052943e-01 -2.39821687e-01 8.76126349e-01 1.11921465e+00 8.28547060e-01 -1.00296680e-02 -5.15838206e-01 1.19141221e+00 1.07186206e-01 2.44488105e-01 7.78852761e-01 -5.72230279e-01 2.74805963e-01 6.12197757e-01 1.46505147e-01 -7.08591998e-01 -4.85953063e-01 -2.27588087e-01 -6.06262922e-01 -3.48424762e-01 1.40262730e-02 -1.83913454e-01 -1.15290320e+00 1.78438246e+00 3.37403923e-01 6.83059990e-01 2.79187441e-01 8.94033074e-01 1.19365025e+00 7.58277774e-01 7.96900630e-01 -2.00811252e-01 1.33116913e+00 -1.04871964e+00 -9.02437389e-01 -2.46827483e-01 8.58801365e-01 -4.71390575e-01 1.15332663e+00 -3.59439224e-01 -5.47519565e-01 -6.36856318e-01 -1.07484949e+00 -1.77258924e-01 -8.39944303e-01 1.62074596e-01 3.57759386e-01 7.72889256e-01 -5.10756612e-01 6.57108188e-01 -8.35454822e-01 -5.56620955e-01 1.56521037e-01 3.33529189e-02 -1.49079114e-01 -1.88520981e-03 -1.78450119e+00 1.09724736e+00 1.11975873e+00 -1.42573133e-01 -4.70366567e-01 -9.30151105e-01 -1.01665330e+00 3.07841271e-01 4.92137730e-01 -5.25823772e-01 1.21898365e+00 -2.85531640e-01 -1.28564250e+00 7.36177921e-01 -2.36405894e-01 -4.33311999e-01 8.59451368e-02 -2.09219247e-01 -8.79785061e-01 5.72012588e-02 4.19042259e-01 7.35653758e-01 2.16215044e-01 -1.37071168e+00 -6.53637767e-01 -8.09218809e-02 1.29059970e-01 2.67014682e-01 -6.09890223e-01 5.72409295e-02 -4.50622290e-01 -6.39209151e-01 -1.64798766e-01 -4.78717864e-01 -2.76259929e-01 -6.37204647e-01 -4.79224712e-01 -5.98783851e-01 8.15065861e-01 -3.60011518e-01 1.50817096e+00 -1.97312641e+00 -7.42860436e-02 -1.11316748e-01 7.77315050e-02 7.28921771e-01 -2.97564298e-01 3.51724803e-01 -1.49347723e-01 3.32720220e-01 -1.54781491e-01 -3.42909068e-01 1.27629310e-01 2.53001660e-01 -3.07755738e-01 1.00112349e-01 2.99553335e-01 1.00224400e+00 -1.35766578e+00 -6.99311197e-01 1.61047712e-01 3.78625900e-01 -1.27293184e-01 1.40374735e-01 -1.88059956e-02 5.01814336e-02 -3.35543573e-01 5.61183631e-01 5.34192622e-01 -3.43679965e-01 4.07666981e-01 -3.25166643e-01 -1.03325114e-01 3.37121636e-01 -1.04381597e+00 1.69963515e+00 -4.10617381e-01 1.61462966e-02 -5.27715802e-01 -9.59428608e-01 9.78654981e-01 4.00797516e-01 1.02226123e-01 -6.22424424e-01 1.19073801e-01 6.74717054e-02 -3.62228096e-01 -3.07471037e-01 5.46036124e-01 -7.15208352e-01 -3.17977756e-01 1.04443938e-01 4.81673092e-01 3.63117844e-01 3.68925840e-01 2.38138512e-01 9.39226449e-01 -2.03720224e-03 6.60211742e-01 -1.10073850e-01 5.14944851e-01 1.34339020e-01 1.19883215e+00 9.30850983e-01 -5.03020883e-01 3.43944788e-01 2.45002769e-02 -6.48686364e-02 -8.53478551e-01 -1.18225038e+00 1.30986730e-02 1.28913641e+00 5.63734412e-01 -6.19043827e-01 -6.31915331e-01 -1.11578918e+00 -3.37050438e-01 1.39651310e+00 -4.44114327e-01 -2.37637594e-01 -5.31960189e-01 -6.39528751e-01 8.96921337e-01 8.21123004e-01 4.83061016e-01 -1.10754287e+00 -3.12879384e-01 2.62826651e-01 -2.64499247e-01 -1.02816141e+00 -5.39136052e-01 3.32153469e-01 -6.33610785e-01 -8.55311632e-01 -9.68260288e-01 -1.05239463e+00 4.60943729e-01 4.51835185e-01 9.96546865e-01 -1.67554691e-01 -2.42723554e-01 2.52208173e-01 -7.73685634e-01 -2.15232283e-01 -2.24927396e-01 2.43615881e-01 1.81356683e-01 -2.67990500e-01 7.97695875e-01 -5.36490142e-01 -1.90927327e-01 2.19815731e-01 -7.22795010e-01 -1.65857241e-01 6.04825020e-01 1.07623124e+00 4.62929547e-01 6.62510991e-02 8.36029351e-01 -1.20279944e+00 5.61171651e-01 -7.85207272e-01 -1.85929552e-01 7.88902640e-01 -7.18421936e-01 1.89282194e-01 8.19887877e-01 -4.64669317e-01 -1.69046235e+00 -4.73253727e-02 -3.37316692e-02 -6.48336291e-01 -4.24555868e-01 5.83577454e-01 -4.59627658e-01 3.96651030e-01 6.59945428e-01 2.86579490e-01 -6.84153438e-01 -6.30997121e-01 8.20316672e-01 6.46733522e-01 6.46459401e-01 -6.83549464e-01 5.41705608e-01 1.18145578e-01 -5.75513482e-01 -8.33855212e-01 -1.26403809e+00 -9.91410911e-01 -7.98481882e-01 -2.80134305e-02 9.44629550e-01 -9.51905012e-01 -4.30653900e-01 2.02871844e-01 -1.20232105e+00 -7.91301206e-03 -3.22936594e-01 5.35448432e-01 -1.87444195e-01 5.94818532e-01 -8.26777339e-01 -8.32244873e-01 -3.53154808e-01 -4.75527585e-01 8.24949741e-01 6.29731476e-01 -2.23328680e-01 -1.12973475e+00 3.71206611e-01 1.65970296e-01 1.18315570e-01 -2.40548819e-01 9.85367119e-01 -1.38186622e+00 8.73564836e-03 -2.88328305e-02 -2.56572545e-01 1.08060561e-01 1.25744551e-01 -3.70564967e-01 -1.01296544e+00 -1.46482095e-01 -2.18140900e-01 -2.73257434e-01 1.03523040e+00 -1.24028593e-01 6.71958804e-01 -2.24453926e-01 -5.85553765e-01 2.98049122e-01 1.50878108e+00 3.21838051e-01 5.25289416e-01 3.00645351e-01 8.79748285e-01 5.20640790e-01 7.87768245e-01 3.32047582e-01 3.56086463e-01 4.35378402e-01 -1.40373439e-01 6.90511093e-02 -2.28333429e-01 -5.15228093e-01 7.37771904e-03 1.17719007e+00 6.66152760e-02 -1.59082860e-01 -1.04844475e+00 8.36565554e-01 -1.95199180e+00 -1.18726265e+00 1.67531878e-01 1.89345372e+00 9.73571062e-01 9.10722241e-02 -5.58498017e-02 -2.97043264e-01 1.39883173e+00 7.33325705e-02 -5.56844413e-01 -2.85711344e-02 -3.46213400e-01 2.85079420e-01 4.69866425e-01 1.53857440e-01 -1.38425195e+00 1.29648757e+00 6.01125383e+00 1.09922028e+00 -6.95810795e-01 4.48197216e-01 2.12014750e-01 1.46579161e-01 -1.82110623e-01 1.55168816e-01 -1.23490250e+00 6.11255944e-01 1.08921218e+00 -4.72094774e-01 -1.95616037e-01 1.02857304e+00 -2.17287108e-01 3.10918123e-01 -7.64956355e-01 8.23714674e-01 -1.08214477e-02 -1.31755340e+00 1.91655606e-01 -5.20719051e-01 8.59407306e-01 -8.15775767e-02 -4.80413884e-01 9.09027755e-01 7.34774292e-01 -5.35696864e-01 5.15281141e-01 5.32431960e-01 4.47756082e-01 -8.33726585e-01 7.91503906e-01 3.88536632e-01 -1.38950181e+00 -1.45448819e-01 -5.06847680e-01 1.47959337e-01 5.13359725e-01 5.20261288e-01 -7.55475283e-01 8.40407252e-01 6.06494486e-01 7.12124527e-01 -4.10535276e-01 1.18644357e+00 -4.23077047e-01 6.85228944e-01 -1.43586993e-02 -2.47522756e-01 3.34944397e-01 7.28570074e-02 4.42051798e-01 1.66758597e+00 3.40403989e-02 5.31193376e-01 3.70895386e-01 9.12325025e-01 -1.69289708e-01 3.32453966e-01 -3.43525171e-01 -7.07605332e-02 1.13865387e+00 1.26078963e+00 -9.56830144e-01 -7.83771992e-01 -6.61749482e-01 1.14496124e+00 7.54549026e-01 2.32715085e-01 -9.50463235e-01 -9.14109111e-01 5.24804413e-01 -4.15062040e-01 6.72471404e-01 -4.27633338e-02 -2.55430162e-01 -1.46398830e+00 -2.30498269e-01 -2.33238950e-01 7.44046807e-01 -5.73557675e-01 -1.87159967e+00 4.72029090e-01 -2.79906504e-02 -1.11162496e+00 2.05189228e-01 -3.33704531e-01 -8.61709416e-01 6.07024789e-01 -1.61276031e+00 -1.05821788e+00 -7.58314282e-02 3.62680018e-01 6.47365093e-01 8.87888223e-02 9.90022957e-01 3.94857049e-01 -9.08044517e-01 7.25464642e-01 8.12824145e-02 5.34750998e-01 8.36064100e-01 -1.30494559e+00 4.26399469e-01 1.03368926e+00 5.62986992e-02 1.02026367e+00 5.35032809e-01 -9.61691380e-01 -6.97551370e-01 -1.35943711e+00 1.18010473e+00 -1.54102638e-01 6.57627642e-01 -1.35188103e-02 -1.25227892e+00 6.01883650e-01 -1.66363809e-02 -1.40549079e-01 1.01112950e+00 3.51320207e-01 -7.79155910e-01 2.24306494e-01 -9.69024479e-01 6.64744616e-01 1.12392974e+00 -6.74175799e-01 -1.46417797e+00 1.63855955e-01 1.01358318e+00 1.27850443e-01 -8.17613602e-01 4.59522784e-01 1.14154361e-01 -6.67917609e-01 1.02699590e+00 -1.01533592e+00 -5.77283651e-02 -5.01225054e-01 -1.34771064e-01 -1.32749438e+00 -6.26883090e-01 -2.19111472e-01 -2.19822913e-01 1.77043533e+00 3.53401959e-01 -3.65745038e-01 4.76955503e-01 5.64864635e-01 -2.66571999e-01 -1.70663327e-01 -6.65985644e-01 -1.47986877e+00 6.81437403e-02 -6.25535175e-02 4.69004571e-01 1.54370332e+00 4.25208956e-01 9.29117322e-01 -3.37126285e-01 3.79658014e-01 5.63595295e-01 1.06384754e-01 5.34363240e-02 -1.32342601e+00 -1.54358195e-02 -4.25778061e-01 -3.91470402e-01 -8.09746861e-01 5.41435122e-01 -1.08815241e+00 4.47820485e-01 -1.55166698e+00 5.49313426e-01 -4.13863331e-01 -7.99243927e-01 8.39566290e-01 -8.06035340e-01 -1.14684537e-01 1.68160230e-01 2.51896113e-01 -1.13548791e+00 7.13625848e-01 5.09842873e-01 -7.63655230e-02 -3.14213067e-01 -5.06550908e-01 -6.62153602e-01 5.94740748e-01 8.03049445e-01 -5.10663271e-01 -3.84028286e-01 -1.95038706e-01 -1.49711907e-01 -2.55122870e-01 2.84125693e-02 -9.63114321e-01 6.45155430e-01 -1.61735609e-01 1.11500673e-01 -5.04658341e-01 1.02120072e-01 -3.89938265e-01 -3.98696661e-02 3.16681683e-01 -5.29404879e-01 -4.72154081e-01 1.34267295e-02 7.96196878e-01 -1.76635221e-01 -5.47898173e-01 9.02202308e-01 -2.73177832e-01 -1.59448206e+00 1.69180967e-02 -1.89893126e-01 4.05203968e-01 1.05439591e+00 -9.20212120e-02 -3.96253854e-01 8.83282274e-02 -9.20874119e-01 4.18107092e-01 2.11406976e-01 5.94135046e-01 5.37786663e-01 -1.42709386e+00 -5.09165168e-01 -3.48874211e-01 5.26170909e-01 -4.18670565e-01 5.26438951e-01 5.29402077e-01 -9.29766614e-03 4.76462513e-01 -5.86222857e-02 -7.63269290e-02 -1.13821805e+00 1.07580161e+00 8.55698958e-02 -1.77024812e-01 -6.28112197e-01 9.29781675e-01 2.36740019e-02 -8.93457234e-01 2.13573724e-01 1.49861112e-01 -5.05445182e-01 3.00086468e-01 7.07833827e-01 6.29229248e-01 1.91543496e-03 -5.56577027e-01 -5.49615204e-01 2.87973344e-01 -3.67651612e-01 8.16665068e-02 1.20824647e+00 -2.93923885e-01 1.43871099e-01 6.73124433e-01 1.13350892e+00 -2.54529834e-01 -8.43648434e-01 -5.43078601e-01 7.00850546e-01 -4.09902856e-02 -1.83253124e-01 -8.72194409e-01 -6.93180501e-01 6.97001219e-01 3.26611489e-01 5.41980341e-02 8.57851684e-01 6.72218353e-02 1.01326716e+00 7.23161340e-01 3.07274967e-01 -1.24790978e+00 -1.88110530e-01 8.12935174e-01 -4.44416814e-02 -1.10796499e+00 -2.90659249e-01 -4.97454792e-01 -8.21003139e-01 9.26712394e-01 8.63977313e-01 1.39363632e-01 5.41035056e-01 8.43433850e-03 1.09712578e-01 -1.51157111e-01 -6.56337082e-01 -6.70844972e-01 2.28951767e-01 5.22426605e-01 3.53315443e-01 -3.31645869e-02 -6.33262157e-01 1.09102273e+00 2.72053242e-01 -1.49631798e-01 2.51039267e-01 1.11882091e+00 -7.33483434e-01 -1.00705171e+00 2.78669372e-02 8.44075605e-02 -2.60028511e-01 -4.52000886e-01 -4.11100924e-01 5.70756435e-01 2.02542618e-01 1.02013052e+00 -6.03390858e-02 -2.69056559e-01 3.83129835e-01 5.48495233e-01 8.68810639e-02 -8.77441287e-01 -4.40551937e-01 -2.16064245e-01 1.54941857e-01 -2.18145326e-01 -5.16993225e-01 -3.53742391e-01 -1.73921239e+00 1.26517683e-01 -6.87608421e-01 5.27419150e-01 3.33003670e-01 1.10192752e+00 5.05043149e-01 4.61743355e-01 5.46336710e-01 -2.98161864e-01 -5.86719036e-01 -8.76204789e-01 -6.69999242e-01 6.38613582e-01 -2.83624977e-01 -9.01188433e-01 -2.56841481e-01 1.23186864e-01]
[9.664918899536133, 9.380733489990234]
92db60b1-82f9-405d-ad52-b8b738b7e07d
separable-structure-modeling-for-semi
null
null
https://ieeexplore.ieee.org/document/9356697
https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9356697
Separable Structure Modeling for Semi-supervised Video Object Segmentation
In this paper, we propose a separable structure modeling approach for semi-supervised video object segmentation. Unlike most existing methods which preclude the semantically structural information of target objects, our method not only captures pixel-level similarity relationships between the reference and target frames but also reveals the separable structure of the specified objects in target frames. Specifically, we first compute a pixel-wise similarity matrix by using representations of reference and target pixels and then select top-rank reference pixels for target pixel classification. According to the prior knowledge from these top-rank reference pixels, we further appoint the representative target pixels for object structure modeling. Particularly, in the structure modeling branch, we extract the shared and individual features that can well represent the whole object and its components, respectively. Moreover, the proposed method is a fast algorithm without online fine-tuning and any post-processing. We conduct extensive experiments and ablation studies on the DAVIS-16, DAVIS-17, and YouTube-VOS datasets, and experimental results on three widely-used datasets demonstrate that our method achieves superior performance, compared with state-of-the-art semi-supervised video object segmentation approaches in terms of speed and accuracy.
['Jie zhou', 'Jiwen Lu', 'Jiahao Li', 'Wencheng Zhu']
2021-02-18
null
null
null
null
['one-shot-visual-object-segmentation']
['computer-vision']
[ 3.56362402e-01 -2.96784133e-01 -4.95837986e-01 -3.98822606e-01 -6.81364000e-01 -3.61773610e-01 1.39092281e-01 -7.43386149e-02 -3.27363759e-01 3.27263236e-01 -2.26413161e-02 1.82670295e-01 -8.15811679e-02 -5.49778402e-01 -5.27045131e-01 -7.84804046e-01 1.18312322e-01 3.20422351e-01 8.67255569e-01 3.25096726e-01 2.40982473e-01 2.99910188e-01 -1.55128801e+00 2.79607624e-01 9.43581462e-01 1.43282986e+00 4.47821498e-01 2.25360706e-01 -1.45224914e-01 8.38532329e-01 -2.80214429e-01 -2.54563782e-02 2.84832537e-01 -4.00407076e-01 -8.87265086e-01 9.39091444e-01 6.88605368e-01 -3.76341850e-01 -5.09964287e-01 1.26317286e+00 1.43315613e-01 1.41139507e-01 6.48025513e-01 -1.18677807e+00 -2.65099347e-01 4.05218840e-01 -8.98084581e-01 3.72550309e-01 2.13770077e-01 2.31547907e-01 9.69001532e-01 -9.09262657e-01 6.44969940e-01 1.19729578e+00 3.60253066e-01 2.14078993e-01 -1.15107036e+00 -7.18605459e-01 5.05650103e-01 3.58771026e-01 -1.55207205e+00 -4.02764618e-01 9.82814789e-01 -5.39652109e-01 2.39310592e-01 2.26030603e-01 6.65798962e-01 6.68639362e-01 -4.34670269e-01 1.41494715e+00 9.17278945e-01 4.14867476e-02 1.13830745e-01 -2.42389943e-02 3.93420547e-01 8.38879108e-01 2.07120582e-01 -1.29131451e-01 -2.36303180e-01 2.82405205e-02 8.44944239e-01 2.00179040e-01 -4.32335764e-01 -7.17785180e-01 -1.44765127e+00 3.55667472e-01 2.67631114e-01 2.32860893e-01 -4.05495673e-01 2.69070268e-02 3.99448574e-01 -2.14831337e-01 3.66576791e-01 -2.83287287e-01 -4.07802373e-01 3.03553157e-02 -1.25323939e+00 -1.35384306e-01 3.94646496e-01 1.34150231e+00 9.36993122e-01 -2.48007141e-02 -3.62881213e-01 7.66708970e-01 5.23104429e-01 2.63156831e-01 3.41890633e-01 -1.08594549e+00 5.98006487e-01 8.52913857e-01 9.32983756e-02 -1.08130956e+00 -7.89439380e-02 -3.09010834e-01 -8.05245399e-01 -2.18123376e-01 4.91382390e-01 1.55075938e-01 -1.05441296e+00 1.17402983e+00 6.64521277e-01 6.56119227e-01 -1.09167257e-02 1.17160511e+00 1.07902849e+00 7.22773850e-01 1.42991856e-01 -5.29378176e-01 1.32620692e+00 -1.34253013e+00 -5.57055235e-01 -2.00874135e-01 3.61376315e-01 -7.41317034e-01 9.75212276e-01 1.66091919e-01 -1.11701941e+00 -9.01980460e-01 -8.70048225e-01 1.04519442e-01 1.12260677e-01 5.52799284e-01 5.64939380e-01 5.40661156e-01 -4.82392371e-01 5.22455573e-01 -7.46921480e-01 -8.34913477e-02 8.56648564e-01 2.21234411e-01 -2.39501581e-01 -1.18238308e-01 -7.50590205e-01 4.55129258e-02 7.20522165e-01 1.46801651e-01 -1.09803712e+00 -7.37201691e-01 -8.28658879e-01 1.95540283e-02 8.22179437e-01 -3.84605289e-01 8.30482244e-01 -1.18950164e+00 -1.19538832e+00 8.38036597e-01 -3.51086736e-01 -1.87682331e-01 4.75919247e-01 -3.90664972e-02 -2.34517053e-01 5.85189819e-01 2.16229603e-01 7.56278157e-01 9.21802938e-01 -1.48855007e+00 -1.10234439e+00 -2.48678043e-01 6.11726865e-02 2.93326080e-01 -3.67128909e-01 1.86960205e-01 -1.24496686e+00 -7.17340350e-01 6.95331335e-01 -6.66809082e-01 -2.37085432e-01 1.70162410e-01 -5.34097672e-01 -2.05204293e-01 1.02164626e+00 -5.00860870e-01 1.32648301e+00 -2.39875364e+00 5.98957725e-02 3.17610741e-01 3.05854112e-01 4.89651501e-01 -1.79460093e-01 -2.85548180e-01 6.62234873e-02 -3.91116776e-02 -3.35248649e-01 -1.12858646e-01 -3.37236762e-01 4.18333039e-02 -4.58076857e-02 5.07159710e-01 5.56328334e-03 7.85830855e-01 -9.39425468e-01 -1.11297560e+00 3.83995920e-01 1.52862340e-01 -2.78710037e-01 2.28763282e-01 -2.42627203e-01 3.01077843e-01 -7.26674855e-01 9.13898230e-01 7.09225535e-01 -3.32911789e-01 -7.59131857e-04 -6.82693064e-01 2.46263906e-01 -3.43827344e-02 -1.45814025e+00 1.57539451e+00 2.16449812e-01 3.86204302e-01 2.35591289e-02 -1.08268285e+00 8.99579287e-01 4.13071215e-02 8.42570841e-01 -4.56337005e-01 1.45476356e-01 1.79433107e-01 -1.48950830e-01 -5.46508670e-01 2.96527028e-01 3.97836685e-01 2.96745270e-01 3.60380232e-01 -3.44168171e-02 2.10144877e-01 6.55101120e-01 2.62065321e-01 4.93363857e-01 4.21205163e-01 6.73291758e-02 -3.22333097e-01 8.52045298e-01 9.12002325e-02 9.48712409e-01 3.65572810e-01 -5.09433031e-01 8.83769691e-01 3.93113941e-01 -2.74006873e-01 -6.71209335e-01 -9.80616093e-01 -6.92189559e-02 8.70291114e-01 9.09258902e-01 -4.07839119e-01 -9.38158810e-01 -9.29450870e-01 -1.36096850e-01 3.28014314e-01 -4.37289178e-01 3.70753109e-02 -5.94861209e-01 -5.81131041e-01 2.58636206e-01 6.71187222e-01 8.69077384e-01 -8.98319185e-01 -4.81874496e-01 1.72804162e-01 -4.33212399e-01 -1.39616358e+00 -1.00590086e+00 -3.47332239e-01 -1.10949945e+00 -1.37852859e+00 -8.34878206e-01 -1.25469100e+00 9.40362692e-01 9.57126260e-01 8.04421306e-01 2.27346927e-01 -1.92947969e-01 3.41623247e-01 -2.78514326e-01 1.50887609e-01 7.43849650e-02 -2.36272275e-01 -1.95855156e-01 5.71924031e-01 2.80751258e-01 -2.31201947e-01 -7.72322476e-01 8.45030844e-01 -7.70791829e-01 2.48297378e-01 6.23533487e-01 5.45895398e-01 1.25979757e+00 2.92784929e-01 1.29073933e-01 -8.08231592e-01 -8.36934745e-02 -2.98189849e-01 -5.39663672e-01 4.79803860e-01 -3.39563757e-01 -3.96373212e-01 2.30635822e-01 -6.16415918e-01 -1.02351248e+00 3.27017874e-01 3.65703374e-01 -7.96979010e-01 -2.89326191e-01 2.84282357e-01 -5.21309555e-01 2.47360989e-02 1.02436505e-01 7.12972701e-01 -4.52476814e-02 -5.06409049e-01 1.28334835e-01 6.13609970e-01 6.26768529e-01 -6.89368725e-01 8.50780606e-01 7.07495451e-01 -1.54665649e-01 -6.85145319e-01 -1.00123179e+00 -7.33373582e-01 -9.00170624e-01 -3.37087363e-01 1.02549589e+00 -1.08886170e+00 -4.25785661e-01 5.02329409e-01 -9.89823401e-01 -1.17098637e-01 -2.11714491e-01 4.66257691e-01 -6.47836924e-01 8.28793764e-01 -4.76701587e-01 -5.85079908e-01 -2.12563887e-01 -1.34235954e+00 1.10898101e+00 4.58985388e-01 5.03106490e-02 -7.90132046e-01 -5.30337036e-01 6.63866818e-01 -1.83216631e-01 8.56998190e-02 6.32249653e-01 -5.41330814e-01 -1.02218902e+00 -1.03641367e-02 -6.80122495e-01 4.24140275e-01 2.48721674e-01 2.77594328e-01 -6.39175355e-01 -1.72531843e-01 -4.23008464e-02 -8.40769187e-02 8.29512537e-01 5.30696273e-01 1.44188130e+00 -1.52981341e-01 -5.06345928e-01 7.62535274e-01 1.24756765e+00 3.67116749e-01 5.91563642e-01 2.11806238e-01 1.04946554e+00 7.40074396e-01 1.21737099e+00 2.59965748e-01 4.26851839e-01 5.95057070e-01 3.17041546e-01 -2.59677798e-01 -2.18730554e-01 -2.71143347e-01 3.01520169e-01 5.44475853e-01 -1.05497822e-01 1.37762576e-01 -6.70562088e-01 6.15056574e-01 -1.96030271e+00 -9.72030759e-01 -3.98089319e-01 2.07240081e+00 7.49772847e-01 2.80480832e-01 3.24536562e-01 1.01955958e-01 1.01258433e+00 2.53103465e-01 -6.38360679e-01 6.39625847e-01 -2.01473162e-01 -2.42428035e-01 3.95428389e-01 5.82624413e-02 -1.42581081e+00 1.06647122e+00 5.92262697e+00 1.31552958e+00 -7.76822984e-01 4.48714709e-03 9.90492880e-01 -7.90668130e-02 2.92642508e-02 5.99922277e-02 -7.87802458e-01 6.33697510e-01 1.86129272e-01 -1.99537858e-01 2.05938876e-01 9.53645170e-01 3.39941323e-01 -2.00751841e-01 -1.11986625e+00 1.21184528e+00 1.04825459e-01 -1.31801820e+00 1.83879420e-01 -2.13958636e-01 9.17376161e-01 -3.12887430e-01 -1.07992098e-01 -8.13291967e-02 -1.78012013e-01 -6.85692191e-01 1.03220308e+00 4.27962899e-01 4.98677552e-01 -6.33368075e-01 4.70100313e-01 2.16028363e-01 -1.62793052e+00 4.02490832e-02 -5.50096571e-01 3.53873491e-01 4.36427370e-02 4.28490460e-01 -1.52134821e-01 5.18165171e-01 7.47853577e-01 1.16777015e+00 -6.18986070e-01 1.19159281e+00 -7.90815428e-02 7.71178782e-01 -2.39842504e-01 1.92687601e-01 3.65532428e-01 -5.73134542e-01 4.04764175e-01 9.92891490e-01 3.66689404e-03 4.16217417e-01 6.35470927e-01 8.25777173e-01 1.45300254e-01 2.14791223e-01 -3.17834946e-03 -2.55016536e-02 3.86168689e-01 1.35456526e+00 -1.13992918e+00 -7.39867270e-01 -4.76025075e-01 8.33730578e-01 -6.91977739e-02 5.54510176e-01 -1.00466192e+00 -2.49278262e-01 5.24765134e-01 2.03640059e-01 6.55529439e-01 -2.52426177e-01 -2.12244660e-01 -1.15683699e+00 2.07042396e-01 -7.62354314e-01 4.29201692e-01 -7.22495139e-01 -1.02427018e+00 4.76884097e-01 5.47895730e-02 -1.66844189e+00 2.81040400e-01 -3.57685715e-01 -6.74013853e-01 5.00496864e-01 -1.45319176e+00 -1.18880975e+00 -5.10023177e-01 7.70073235e-01 8.38491201e-01 -1.47876710e-01 1.06788263e-01 3.87062192e-01 -8.51556480e-01 3.19194406e-01 1.58223182e-01 4.70046282e-01 3.43650967e-01 -8.86158764e-01 1.65792592e-02 9.29202974e-01 3.50838184e-01 4.55476403e-01 2.70678163e-01 -6.52000189e-01 -1.25640845e+00 -1.35588562e+00 3.17565501e-01 -9.48019698e-02 4.76806372e-01 -1.23979874e-01 -9.38542545e-01 5.11258125e-01 -9.61625129e-02 3.97436261e-01 5.56715310e-01 -3.86811137e-01 -1.88810930e-01 -2.64798135e-01 -9.49553430e-01 5.41223705e-01 1.35962284e+00 -2.84749955e-01 -5.30041337e-01 4.62094337e-01 6.65223718e-01 -4.31917787e-01 -7.56071150e-01 5.28563082e-01 3.75953972e-01 -1.04026043e+00 1.08331025e+00 -4.73358452e-01 3.31753224e-01 -8.20444942e-01 -1.83931097e-01 -5.89874268e-01 -3.02698344e-01 -4.23563957e-01 -2.45737866e-01 1.53279245e+00 8.07658732e-02 -2.01617002e-01 1.01899624e+00 5.29207051e-01 -4.06968258e-02 -9.20226157e-01 -6.81317687e-01 -7.36407101e-01 -4.59001064e-01 -4.78209138e-01 3.93815666e-01 6.46919787e-01 -5.05504489e-01 6.36425242e-02 -2.13347167e-01 2.63631821e-01 9.99847651e-01 7.03186333e-01 8.42993140e-01 -1.02690923e+00 -1.16972357e-01 -4.29593086e-01 -5.81025481e-01 -1.61133206e+00 1.43036082e-01 -6.64044142e-01 4.41777371e-02 -1.54695988e+00 5.54456770e-01 -5.51297307e-01 -4.77877945e-01 2.89750755e-01 -4.28343028e-01 4.74266618e-01 1.44080982e-01 6.49065316e-01 -1.21616101e+00 5.61831117e-01 1.45116746e+00 -4.59021389e-01 -2.54910856e-01 6.84727654e-02 -6.22978032e-01 9.81878400e-01 4.07601506e-01 -3.77360046e-01 -5.33030033e-01 -2.67864943e-01 -6.37279212e-01 -1.86939538e-01 3.35331559e-01 -1.06624377e+00 1.14811197e-01 -5.08657396e-01 4.96458918e-01 -9.22679603e-01 1.29305571e-01 -9.59132314e-01 7.00782612e-02 4.27710533e-01 -1.95600361e-01 -6.10877573e-01 -8.58234018e-02 7.50667214e-01 -4.96107548e-01 -3.90648663e-01 9.39350128e-01 -2.66627455e-03 -1.19954181e+00 8.18651140e-01 -5.93301952e-02 1.71675831e-01 1.51597369e+00 -7.13489652e-01 3.77804004e-02 -1.39425263e-01 -6.45044446e-01 5.33350706e-01 4.89527524e-01 3.91486973e-01 8.58957231e-01 -1.31384420e+00 -5.24872601e-01 1.47267222e-01 1.54554784e-01 4.14469033e-01 4.16552007e-01 8.98565054e-01 -4.62862432e-01 1.58753186e-01 -1.11932106e-01 -9.86315370e-01 -1.55272114e+00 7.31079280e-01 1.93432108e-01 1.72625124e-01 -6.19425654e-01 7.70030320e-01 7.27810681e-01 1.81860670e-01 3.46137077e-01 -2.97750354e-01 -3.40365022e-01 1.86898857e-01 4.74764287e-01 4.07676667e-01 -5.45346260e-01 -1.12638259e+00 -3.48257452e-01 9.53182518e-01 -1.44883260e-01 3.39036584e-01 9.98942673e-01 -3.89369875e-01 -8.81399661e-02 2.49515489e-01 1.18168879e+00 -2.53613442e-01 -1.65727758e+00 -6.15155041e-01 2.64606662e-02 -8.25611830e-01 4.68015336e-02 -1.81013778e-01 -1.63771176e+00 8.26610804e-01 5.38958073e-01 -1.44113913e-01 1.25839293e+00 5.20292148e-02 8.05801630e-01 8.31111372e-02 3.85348380e-01 -1.11655951e+00 2.45061055e-01 1.37909785e-01 4.12931979e-01 -1.20634770e+00 1.62273973e-01 -1.02320802e+00 -7.45590270e-01 1.01923621e+00 9.33461010e-01 -1.05687879e-01 5.36004066e-01 -1.53411478e-01 3.84968449e-03 -2.51955241e-02 -3.49160373e-01 -3.75076503e-01 6.04735553e-01 5.82189322e-01 1.26446307e-01 -2.43236199e-01 -2.50820667e-01 5.73843777e-01 4.10259992e-01 -1.11631200e-01 1.90813437e-01 7.49580145e-01 -6.30912602e-01 -7.69228041e-01 -3.80795151e-01 5.37585139e-01 -2.23837122e-01 4.15861234e-02 -1.20350711e-01 6.15152717e-01 1.50878876e-01 8.82580698e-01 1.92450762e-01 -3.93201143e-01 1.86553568e-01 -2.79970139e-01 3.73347610e-01 -5.77599466e-01 -2.11050287e-01 4.70223248e-01 -1.86048254e-01 -6.46948755e-01 -9.32303846e-01 -7.85032213e-01 -1.59609604e+00 1.52268961e-01 -4.61619526e-01 6.48892894e-02 3.16391349e-01 9.14849639e-01 2.97156721e-01 4.72864211e-01 7.56620765e-01 -9.70987141e-01 -3.12388569e-01 -6.86784565e-01 -6.02474153e-01 6.58597112e-01 -9.94355604e-02 -7.66223848e-01 -9.95057374e-02 4.48158354e-01]
[9.34609603881836, -0.10150997340679169]
d1e0cd9f-0d1e-4ce5-b540-acb2b1f053d0
beyond-supervised-vs-unsupervised-1
2206.08347
null
https://arxiv.org/abs/2206.08347v1
https://arxiv.org/pdf/2206.08347v1.pdf
Beyond Supervised vs. Unsupervised: Representative Benchmarking and Analysis of Image Representation Learning
By leveraging contrastive learning, clustering, and other pretext tasks, unsupervised methods for learning image representations have reached impressive results on standard benchmarks. The result has been a crowded field - many methods with substantially different implementations yield results that seem nearly identical on popular benchmarks, such as linear evaluation on ImageNet. However, a single result does not tell the whole story. In this paper, we compare methods using performance-based benchmarks such as linear evaluation, nearest neighbor classification, and clustering for several different datasets, demonstrating the lack of a clear front-runner within the current state-of-the-art. In contrast to prior work that performs only supervised vs. unsupervised comparison, we compare several different unsupervised methods against each other. To enrich this comparison, we analyze embeddings with measurements such as uniformity, tolerance, and centered kernel alignment (CKA), and propose two new metrics of our own: nearest neighbor graph similarity and linear prediction overlap. We reveal through our analysis that in isolation, single popular methods should not be treated as though they represent the field as a whole, and that future work ought to consider how to leverage the complimentary nature of these methods. We also leverage CKA to provide a framework to robustly quantify augmentation invariance, and provide a reminder that certain types of invariance will be undesirable for downstream tasks.
['Abhinav Shrivastava', 'Matthew Gwilliam']
2022-06-16
beyond-supervised-vs-unsupervised
http://openaccess.thecvf.com//content/CVPR2022/html/Gwilliam_Beyond_Supervised_vs._Unsupervised_Representative_Benchmarking_and_Analysis_of_Image_CVPR_2022_paper.html
http://openaccess.thecvf.com//content/CVPR2022/papers/Gwilliam_Beyond_Supervised_vs._Unsupervised_Representative_Benchmarking_and_Analysis_of_Image_CVPR_2022_paper.pdf
cvpr-2022-1
['graph-similarity']
['graphs']
[ 1.08607978e-01 -2.42028572e-02 -3.80331784e-01 -5.78784227e-01 -5.82305312e-01 -7.91174769e-01 9.50013995e-01 4.95038629e-01 -4.32688028e-01 2.08544105e-01 7.90980399e-01 -2.70982295e-01 -4.32792842e-01 -5.15031815e-01 -4.90219921e-01 -6.49095833e-01 -3.48452479e-01 9.57619324e-02 1.89401075e-01 -2.95921594e-01 3.97671491e-01 4.90971953e-01 -1.61868262e+00 2.20670536e-01 4.42866236e-01 6.91858351e-01 -3.78683150e-01 4.63727146e-01 7.07419291e-02 9.98620689e-01 -3.14094007e-01 -5.18874705e-01 3.55611533e-01 -3.75382453e-01 -9.26865697e-01 -8.67722705e-02 8.51917803e-01 -8.77610818e-02 -4.67121691e-01 9.31409776e-01 5.92886984e-01 2.58025199e-01 9.90884662e-01 -1.38525808e+00 -1.05939531e+00 5.98592579e-01 -4.28225309e-01 2.95266658e-01 2.14995474e-01 3.72421652e-01 1.38003230e+00 -7.02310622e-01 8.26410413e-01 1.15781879e+00 1.09770441e+00 2.06192523e-01 -1.41867733e+00 -2.89328814e-01 1.60167933e-01 6.47001490e-02 -1.40863109e+00 -5.37811756e-01 5.21640122e-01 -7.84301937e-01 1.01156783e+00 2.45701760e-01 3.81430000e-01 9.28998530e-01 -2.58979830e-03 5.98988414e-01 1.22100031e+00 -3.87438625e-01 2.27936938e-01 7.06826970e-02 4.23220366e-01 7.06042409e-01 3.93920958e-01 1.03751823e-01 -3.52306336e-01 -1.02036305e-01 3.56381655e-01 2.92518735e-02 -2.10411578e-01 -8.35495830e-01 -1.37649870e+00 1.11027801e+00 6.35039032e-01 4.88511443e-01 1.50284871e-01 1.09363519e-01 6.52781904e-01 3.12857926e-01 5.37391722e-01 7.85534561e-01 -8.46464708e-02 -2.13194296e-01 -9.51850653e-01 2.01304257e-01 8.95361781e-01 7.78772950e-01 9.53880668e-01 -2.11320147e-01 -1.06768794e-01 7.59088337e-01 5.40262870e-02 -5.56059666e-02 5.81580222e-01 -1.15794456e+00 -1.62986647e-02 5.32529473e-01 -3.74484003e-01 -1.44173133e+00 -5.01505792e-01 -4.59932506e-01 -7.00477898e-01 3.81971896e-01 5.28037488e-01 1.09704696e-01 -7.55299389e-01 1.75448799e+00 -1.46523088e-01 7.37934858e-02 -1.38317779e-01 8.00254941e-01 7.02724397e-01 3.65477562e-01 -4.34597321e-02 1.39543265e-01 9.74648058e-01 -1.06765103e+00 -2.50611544e-01 -1.09346598e-01 1.03697252e+00 -8.49398494e-01 1.35800898e+00 2.91416794e-01 -8.38044703e-01 -4.63448554e-01 -1.32228148e+00 -2.22696573e-01 -7.43809342e-01 -3.02777946e-01 8.70497763e-01 5.58660209e-01 -1.40741408e+00 8.75642478e-01 -7.13830471e-01 -1.01450181e+00 3.88961405e-01 1.26686811e-01 -5.48492193e-01 -9.92442891e-02 -6.63480699e-01 1.05390668e+00 2.25041825e-02 -4.55164164e-01 -4.99180436e-01 -7.77760744e-01 -9.23311055e-01 -2.17194080e-01 8.27812850e-02 -6.11115992e-01 8.79108846e-01 -9.19678748e-01 -1.09055400e+00 1.14244974e+00 4.34958301e-02 -4.84250784e-01 3.59089494e-01 7.37501867e-03 -2.73383498e-01 3.91474292e-02 6.64030761e-02 7.96447098e-01 5.70001185e-01 -1.25874507e+00 -2.14228749e-01 -3.82660806e-01 1.83599710e-01 2.37036243e-01 -6.24428332e-01 3.13629732e-02 -3.73921216e-01 -7.24338055e-01 6.32885545e-02 -9.15353894e-01 -2.75679410e-01 1.17550448e-01 -2.46075884e-01 -1.82389006e-01 6.96248353e-01 -1.39317736e-01 1.29022646e+00 -2.36285877e+00 4.45168428e-02 2.25076184e-01 5.78584015e-01 -3.17051113e-02 -3.06127459e-01 5.97823322e-01 -2.95398921e-01 4.95172083e-01 -4.22709376e-01 -4.98108983e-01 2.34217241e-01 1.71262741e-01 -1.75642058e-01 8.59343112e-01 1.76897720e-01 9.82345641e-01 -1.02994823e+00 -5.46630919e-01 2.83762187e-01 5.90960145e-01 -6.82868361e-01 -2.00758249e-01 1.85370669e-01 -7.00700507e-02 8.55994001e-02 4.40226674e-01 3.56536090e-01 -4.21356499e-01 1.08846754e-01 -3.58936548e-01 -8.23503956e-02 1.44801825e-01 -9.93185937e-01 1.83491623e+00 4.29196865e-04 9.21827555e-01 -2.97117680e-01 -1.23962247e+00 6.53202355e-01 -1.82658881e-01 5.18271029e-01 -5.93386292e-01 -6.77493811e-02 5.11168204e-02 9.73200053e-02 -4.76771414e-01 6.03456378e-01 1.90152705e-01 7.02757314e-02 6.04285717e-01 2.06878096e-01 -5.83115742e-02 1.41510740e-01 3.73874307e-01 1.50499213e+00 1.49676101e-02 2.34484851e-01 -6.04781687e-01 -1.96149852e-02 1.46942183e-01 1.14149511e-01 8.06395471e-01 -4.69457328e-01 9.36322451e-01 4.86582935e-01 -5.23482859e-01 -1.22073066e+00 -1.20340836e+00 -2.62333930e-01 1.25177252e+00 7.98848718e-02 -7.74321496e-01 -7.61502266e-01 -8.12026918e-01 1.95833415e-01 3.86964053e-01 -7.66551852e-01 -1.16187118e-01 -1.85905397e-01 -9.39400733e-01 7.50160277e-01 5.23533285e-01 9.51756760e-02 -6.73432171e-01 -3.69065583e-01 -1.91268727e-01 2.18251124e-01 -1.06508648e+00 -3.33876848e-01 2.52928555e-01 -8.55513334e-01 -9.81986582e-01 -6.16652608e-01 -9.58885014e-01 5.39056122e-01 5.26065171e-01 1.42292082e+00 1.47475466e-01 -3.62239420e-01 7.74532139e-01 -4.31528360e-01 3.27399187e-02 -2.22032562e-01 2.07874298e-01 1.56846166e-01 -1.80145994e-01 5.91597557e-01 -8.45089853e-01 -7.84327328e-01 3.09199274e-01 -9.84218538e-01 -4.12534297e-01 5.92793643e-01 7.29844332e-01 3.76003265e-01 -2.13402882e-01 3.46968591e-01 -8.83581936e-01 8.31284046e-01 -7.13644803e-01 -9.52484459e-02 1.15629695e-01 -8.19740951e-01 1.72353104e-01 6.09031618e-01 -3.47933322e-01 -2.99520791e-01 -1.94548458e-01 1.77967235e-01 -3.82738858e-01 -3.77901316e-01 4.90777791e-01 2.02110797e-01 -1.93736002e-01 1.12880421e+00 1.08247712e-01 2.76382953e-01 -3.43945205e-01 7.85550356e-01 6.00058079e-01 5.55812776e-01 -5.30289769e-01 9.09077227e-01 6.60369277e-01 -1.69811234e-01 -9.84069943e-01 -5.90461254e-01 -6.43217206e-01 -7.14097142e-01 3.33874971e-02 9.07010853e-01 -9.07572210e-01 -5.68506300e-01 6.50234520e-02 -8.37601542e-01 -3.38385671e-01 -4.80268329e-01 5.25022209e-01 -7.02544808e-01 7.08120286e-01 -5.59523940e-01 -4.06320333e-01 8.48142207e-02 -1.16650212e+00 8.70878100e-01 -1.09719925e-01 -4.87070650e-01 -1.32302189e+00 2.79574960e-01 2.30307743e-01 7.32472420e-01 3.49455893e-01 9.52572525e-01 -9.88766253e-01 -2.80534893e-01 -1.53266564e-01 -3.81800920e-01 6.12722158e-01 1.25042766e-01 1.41229585e-01 -9.90249813e-01 -4.77169722e-01 -2.62484193e-01 -4.16919112e-01 1.03921187e+00 1.84223741e-01 1.03421783e+00 -2.18078107e-01 -2.63605326e-01 7.86002994e-01 1.59371531e+00 -4.30548251e-01 7.38692462e-01 3.66670609e-01 7.44309485e-01 7.06467986e-01 2.36610606e-01 1.00332767e-01 6.36965692e-01 5.37302852e-01 2.48052031e-01 -1.95506349e-01 -3.87980312e-01 -2.10324556e-01 5.20632923e-01 1.07836545e+00 -1.17503189e-01 -1.57170251e-01 -9.94144738e-01 6.40524030e-01 -1.85338473e+00 -1.15425336e+00 4.56537008e-02 2.30715704e+00 5.20302713e-01 1.96153775e-01 3.94941092e-01 1.27230167e-01 5.43664396e-01 4.58379000e-01 -2.84253091e-01 -3.88399720e-01 -2.07618818e-01 9.67842266e-02 6.34515047e-01 4.85539138e-01 -1.20121610e+00 8.55137229e-01 7.56840181e+00 7.07459271e-01 -1.16214740e+00 1.83019228e-03 7.43044674e-01 5.45433834e-02 -3.60665083e-01 2.65912831e-01 -3.28012258e-01 2.53499836e-01 7.54714489e-01 9.35318097e-02 5.06253362e-01 8.43470633e-01 -4.52109762e-02 9.71482322e-02 -1.46522975e+00 1.09854090e+00 4.19997990e-01 -1.37645376e+00 -3.03204078e-03 9.37664434e-02 7.09102154e-01 5.15098751e-01 2.31300518e-01 1.88556284e-01 5.51916480e-01 -1.30723453e+00 3.74358326e-01 4.68345731e-01 4.73398149e-01 -4.87657726e-01 4.57909793e-01 -4.67142537e-02 -1.03174150e+00 5.46630546e-02 -3.74916434e-01 -2.65530705e-01 -3.02638233e-01 6.36239290e-01 -3.81347954e-01 3.70129526e-01 7.86745846e-01 1.11146379e+00 -1.04690135e+00 1.12474608e+00 1.45624548e-01 6.11582160e-01 -1.99341074e-01 -9.31502283e-02 3.98303539e-01 -2.16139272e-01 4.54921156e-01 1.58523071e+00 1.10091818e-02 -3.06229949e-01 1.88236043e-01 6.55974984e-01 1.66491363e-02 3.67511988e-01 -1.10892344e+00 -2.32555032e-01 4.23182875e-01 1.30820274e+00 -8.33378851e-01 -2.91114539e-01 -6.76866829e-01 8.53133202e-01 6.04514480e-01 3.79865289e-01 -5.92175424e-01 -5.67358971e-01 8.04978430e-01 1.79102346e-01 1.32216647e-01 -5.10833740e-01 -4.54170644e-01 -1.15429282e+00 2.33098748e-03 -9.64983642e-01 4.08346206e-01 -5.88518381e-01 -1.80872810e+00 3.69487256e-01 6.87301978e-02 -1.25547004e+00 -1.78463712e-01 -8.17852199e-01 -5.58048368e-01 2.99323887e-01 -1.36767423e+00 -8.52229476e-01 -3.65605235e-01 5.54487288e-01 1.92248613e-01 -1.69954002e-01 8.51474822e-01 2.27303937e-01 -4.75478649e-01 8.44900072e-01 2.11878464e-01 3.07599068e-01 1.09069753e+00 -1.39210665e+00 4.32290524e-01 7.02964425e-01 6.43168867e-01 8.82973850e-01 6.33529544e-01 -1.60405442e-01 -1.36868024e+00 -9.12417412e-01 5.84172308e-01 -9.61827815e-01 8.78807664e-01 -2.81431764e-01 -7.86697447e-01 7.37168491e-01 3.66868258e-01 2.77937233e-01 7.79977441e-01 4.21147913e-01 -8.98379982e-01 -6.74058869e-02 -8.39801788e-01 7.47658551e-01 1.31469166e+00 -7.59498358e-01 -4.94616657e-01 4.51906085e-01 7.45101988e-01 1.29369721e-01 -1.07422197e+00 4.40924793e-01 6.21081352e-01 -1.31616187e+00 1.06300235e+00 -6.61574781e-01 5.06663859e-01 -2.63569087e-01 -5.13779402e-01 -1.28521848e+00 -5.17168522e-01 -5.48081636e-01 2.28105620e-01 1.33691227e+00 3.91595572e-01 -6.87756419e-01 7.46014535e-01 3.14608693e-01 -2.22480193e-01 -8.84987533e-01 -4.70514059e-01 -1.02372658e+00 4.87033308e-01 -5.12057602e-01 4.37382609e-01 1.40849793e+00 3.70622963e-01 4.51994807e-01 2.02590972e-02 -1.39835939e-01 7.15442419e-01 -2.18402192e-01 9.66860533e-01 -1.13971472e+00 -2.41137877e-01 -8.52465689e-01 -1.12435901e+00 -7.91621983e-01 1.64024770e-01 -1.32810032e+00 -1.17968626e-01 -1.41904891e+00 4.41423327e-01 -4.36277062e-01 -3.80014807e-01 4.42324281e-01 3.05387583e-02 5.72893143e-01 1.45924211e-01 4.08856034e-01 -8.64912510e-01 1.57743379e-01 7.62566805e-01 -2.26863965e-01 3.00968736e-02 -4.81456727e-01 -9.74907875e-01 8.73553336e-01 7.97047555e-01 -2.18862936e-01 -5.52713513e-01 -5.38414001e-01 1.22492351e-01 -7.87723005e-01 5.14784157e-01 -1.35572231e+00 3.86160403e-01 1.10403933e-01 2.18719169e-01 -4.94251214e-03 1.00709118e-01 -4.75578338e-01 -2.83382952e-01 2.30810240e-01 -5.77977359e-01 3.62255692e-01 1.00415051e-02 5.90841651e-01 -1.70045361e-01 -5.72960638e-03 6.60093844e-01 -7.19550177e-02 -9.41876829e-01 2.65354723e-01 -1.21804006e-01 4.75086004e-01 8.73013496e-01 -3.85070354e-01 -5.38624346e-01 -5.24574816e-01 -5.03414750e-01 -5.75035624e-02 9.86352503e-01 4.98582989e-01 3.96158367e-01 -1.31324553e+00 -7.79794812e-01 -6.97413683e-02 5.59137642e-01 -4.39870596e-01 -1.35010764e-01 9.25782740e-01 -5.04701436e-01 1.17304109e-01 -1.18356101e-01 -8.20859790e-01 -9.94056702e-01 7.17068255e-01 2.47379169e-01 -1.01047933e-01 -5.70020139e-01 5.51494718e-01 7.99463391e-02 -5.93677104e-01 4.49800640e-01 -2.27471903e-01 -1.17407991e-02 1.98725551e-01 3.39954078e-01 2.53654271e-01 1.39145300e-01 -6.15262270e-01 -4.53418821e-01 6.01848304e-01 -2.96353437e-02 -1.11237830e-02 1.35682881e+00 -1.16405182e-01 -1.74364224e-01 6.04184806e-01 1.67071259e+00 3.46748307e-02 -9.88711238e-01 -1.23124942e-01 1.71079695e-01 -3.14753026e-01 -1.15916684e-01 -5.26409447e-01 -8.88131082e-01 8.86333585e-01 6.70491755e-01 3.70879114e-01 8.73314142e-01 1.19172275e-01 2.65526474e-01 4.56184655e-01 6.83131143e-02 -1.08221245e+00 2.13916391e-01 4.22147512e-01 6.03563786e-01 -1.29595423e+00 3.88987333e-01 -1.68911129e-01 -5.84353089e-01 9.74184155e-01 3.61382127e-01 -4.09188658e-01 7.46216893e-01 2.55797386e-01 9.62637961e-02 -3.44992697e-01 -5.70885062e-01 -4.39429730e-01 3.30344409e-01 7.44218349e-01 7.99423397e-01 -1.10984817e-01 -1.47925735e-01 1.38361737e-01 -4.47993189e-01 -2.40643248e-01 3.38481575e-01 9.01669681e-01 -3.81890267e-01 -9.79436398e-01 -1.22708604e-01 5.44528782e-01 -2.36767963e-01 -2.19025135e-01 -5.67220688e-01 1.01253211e+00 5.98014295e-02 9.80281115e-01 2.22359091e-01 -8.72224271e-01 2.58334041e-01 6.74084350e-02 5.38474143e-01 -5.50922692e-01 -5.05834341e-01 -5.08310735e-01 -9.87785906e-02 -7.53792584e-01 -6.11442685e-01 -4.09145325e-01 -8.87293279e-01 -5.82096457e-01 -1.03242345e-01 5.07632131e-03 6.61386967e-01 8.21242213e-01 5.35663724e-01 1.05872907e-01 5.03215253e-01 -1.03985119e+00 -6.17139399e-01 -7.77612388e-01 -4.16076243e-01 8.49920392e-01 2.63355792e-01 -4.88556534e-01 -7.24583626e-01 -1.70129165e-01]
[9.152427673339844, 3.0998342037200928]
65e1126b-dd73-4b1d-baa8-4764da2b483c
unified-modeling-of-multi-talker-overlapped
2305.16263
null
https://arxiv.org/abs/2305.16263v1
https://arxiv.org/pdf/2305.16263v1.pdf
Unified Modeling of Multi-Talker Overlapped Speech Recognition and Diarization with a Sidecar Separator
Multi-talker overlapped speech poses a significant challenge for speech recognition and diarization. Recent research indicated that these two tasks are inter-dependent and complementary, motivating us to explore a unified modeling method to address them in the context of overlapped speech. A recent study proposed a cost-effective method to convert a single-talker automatic speech recognition (ASR) system into a multi-talker one, by inserting a Sidecar separator into the frozen well-trained ASR model. Extending on this, we incorporate a diarization branch into the Sidecar, allowing for unified modeling of both ASR and diarization with a negligible overhead of only 768 parameters. The proposed method yields better ASR results compared to the baseline on LibriMix and LibriSpeechMix datasets. Moreover, without sophisticated customization on the diarization task, our method achieves acceptable diarization results on the two-speaker subset of CALLHOME with only a few adaptation steps.
['Helen Meng', 'Xixin Wu', 'Haibin Wu', 'Mingyu Cui', 'Jiawen Kang', 'Lingwei Meng']
2023-05-25
null
null
null
null
['automatic-speech-recognition']
['speech']
[ 2.48483464e-01 1.20878667e-01 7.43886456e-02 -7.55512834e-01 -1.64946043e+00 -6.52490377e-01 3.95423591e-01 -3.20415884e-01 -4.05276954e-01 8.34199935e-02 3.49025071e-01 -5.54517627e-01 3.27252746e-01 1.56512097e-01 -3.64944756e-01 -8.78589034e-01 3.47847015e-01 3.23867887e-01 6.80591315e-02 -1.86852917e-01 -3.73644382e-01 3.52704704e-01 -1.29368639e+00 3.91930968e-01 7.13010728e-01 9.05076981e-01 3.70850563e-01 1.06929290e+00 -1.51477113e-01 4.74753112e-01 -1.05849731e+00 -3.34369123e-01 2.98456162e-01 -4.84663785e-01 -5.37367940e-01 1.59474388e-01 5.78278482e-01 -2.36773372e-01 -4.91610944e-01 5.46006382e-01 1.17525768e+00 2.39747614e-01 2.18274787e-01 -8.32697272e-01 -1.75031319e-01 1.19774735e+00 -4.05466676e-01 5.27439475e-01 3.25290188e-02 -1.41474873e-01 9.88984346e-01 -8.33038688e-01 -1.23707406e-01 1.30206776e+00 6.45572841e-01 5.80581129e-01 -1.42014837e+00 -8.70017231e-01 3.02135289e-01 -2.31109504e-02 -1.76477158e+00 -1.37502384e+00 6.19221866e-01 -2.36351080e-02 1.25557518e+00 7.70644367e-01 2.04388201e-01 1.00183713e+00 -6.00970507e-01 8.67219567e-01 8.65216076e-01 -4.73984838e-01 1.12148188e-01 1.88319817e-01 2.60749996e-01 2.14005094e-02 -3.58782202e-01 -7.23949373e-02 -7.67164946e-01 -1.23560756e-01 2.52189606e-01 -3.69803011e-01 -3.48127127e-01 4.89962071e-01 -9.47038293e-01 5.88440299e-01 -1.75336570e-01 5.83196044e-01 -1.06288157e-01 -1.23485431e-01 3.57057422e-01 3.08741927e-01 7.10778058e-01 2.53051013e-01 -4.38728034e-01 -4.52325523e-01 -1.38480175e+00 -6.86042011e-02 8.44290376e-01 1.09669697e+00 5.12923300e-01 5.46955705e-01 -3.26169133e-01 1.43929636e+00 2.26781011e-01 7.95655370e-01 5.60854733e-01 -5.81735134e-01 7.36143947e-01 -1.33147851e-01 -2.03012139e-01 -3.24252397e-01 -8.30585361e-02 -6.68487012e-01 -5.22156239e-01 -4.32205498e-01 1.46903232e-01 -2.74766415e-01 -1.00871825e+00 1.57910395e+00 2.36289993e-01 1.62937865e-02 6.48815408e-02 8.78327250e-01 5.78997254e-01 9.88684833e-01 -2.38796696e-01 -4.13237303e-01 1.35490322e+00 -1.18238747e+00 -1.06574786e+00 -4.06651914e-01 6.40599012e-01 -1.23313856e+00 9.92996275e-01 4.70723659e-01 -1.18377745e+00 -4.21259373e-01 -1.03962922e+00 -7.22717792e-02 9.22061950e-02 4.54249948e-01 2.94798017e-01 1.28849351e+00 -1.21713972e+00 3.99125963e-02 -8.76004457e-01 -1.43550664e-01 -1.68819085e-01 3.46837997e-01 -1.67677253e-01 7.58998841e-02 -9.95303214e-01 7.41174459e-01 3.51303294e-02 1.48160398e-01 -7.85176635e-01 -6.56267524e-01 -7.19206929e-01 1.27749711e-01 3.26955408e-01 -1.83929667e-01 1.66121829e+00 -9.14934695e-01 -2.02445698e+00 8.11293185e-01 -6.72243416e-01 -5.54784656e-01 3.08582753e-01 -3.19006085e-01 -8.22474360e-01 -7.23676980e-02 -4.96018291e-01 2.53385931e-01 9.21370685e-01 -1.15227711e+00 -5.48743904e-01 -2.81023055e-01 -3.56479824e-01 5.72903991e-01 -2.61281133e-01 5.46508014e-01 -6.54284298e-01 -1.02140152e+00 3.38716656e-01 -9.87892330e-01 1.29275143e-01 -7.26810932e-01 -5.40128946e-01 -4.05509472e-02 8.28750908e-01 -1.13095534e+00 1.64359593e+00 -2.43133593e+00 4.86248843e-02 5.11408560e-02 -2.03355178e-01 6.66590095e-01 -1.63888857e-01 3.44062805e-01 -2.21396580e-01 9.27344859e-02 -2.36731753e-01 -9.56309080e-01 7.16244429e-02 1.55735970e-01 -4.77389753e-01 4.29089576e-01 -1.48524702e-01 3.87908250e-01 -3.68056864e-01 -1.86215445e-01 2.60559112e-01 7.45812237e-01 -5.37446797e-01 6.52146280e-01 2.94809759e-01 2.55951524e-01 1.62060350e-01 5.75679064e-01 7.90080786e-01 5.12191415e-01 2.57581592e-01 1.81442138e-03 -2.47179449e-01 1.10867453e+00 -1.21383560e+00 1.57839251e+00 -8.72888923e-01 5.40056765e-01 7.97024786e-01 -9.02969599e-01 1.05522907e+00 7.60117590e-01 2.24628389e-01 -3.96384060e-01 1.96242675e-01 4.20652926e-01 2.10770935e-01 -1.58514455e-01 3.64929080e-01 -4.36247230e-01 7.90983886e-02 3.57099533e-01 1.14045307e-01 -4.21364605e-01 -1.79008946e-01 -3.55552137e-02 1.08872807e+00 -5.71949244e-01 1.00864388e-01 -1.14927255e-01 4.54568624e-01 -6.67647004e-01 4.72483277e-01 7.06454158e-01 -2.61856675e-01 1.11312687e+00 -1.54694423e-01 2.94513166e-01 -6.15426183e-01 -1.10499263e+00 -2.40681451e-02 1.30194378e+00 -3.98792803e-01 -4.38740313e-01 -1.06560659e+00 -4.05122668e-01 -3.21562201e-01 1.02732265e+00 2.34492049e-02 -3.64331482e-03 -8.93557608e-01 -7.74194181e-01 1.14094758e+00 3.24993908e-01 2.79629469e-01 -3.66572201e-01 2.62225330e-01 1.93394303e-01 -6.47663057e-01 -1.32248425e+00 -1.26276600e+00 5.04596949e-01 -4.92924929e-01 -2.47036546e-01 -9.34016109e-01 -8.10213447e-01 1.26031935e-01 6.67155087e-01 7.64148533e-01 -2.84487963e-01 7.92300999e-02 2.23245487e-01 -4.77036566e-01 -2.03592598e-01 -8.84951353e-01 3.39149177e-01 2.86301255e-01 3.27482611e-01 2.30847493e-01 -6.45721376e-01 -3.55656713e-01 6.40260339e-01 -6.36404335e-01 -1.77377284e-01 5.31915843e-01 5.62326968e-01 3.62441599e-01 -1.55498385e-01 6.60284758e-01 -3.55516344e-01 4.52389061e-01 -1.91141516e-01 -4.15488064e-01 2.46423647e-01 -4.45265353e-01 1.45616233e-02 5.39275229e-01 -6.30530298e-01 -1.35948956e+00 9.21148881e-02 -7.04262257e-01 -3.97886515e-01 -1.64911076e-01 1.22993086e-02 -7.23537445e-01 1.27982184e-01 4.73980814e-01 4.41195458e-01 -1.48619696e-01 -8.62215519e-01 5.46075046e-01 1.59415710e+00 4.53453779e-01 -2.41308749e-01 6.97547436e-01 1.04163229e-01 -9.01185155e-01 -1.32642806e+00 -5.01709461e-01 -9.76218402e-01 -3.15300047e-01 3.66352379e-01 5.59715629e-01 -1.16276348e+00 -4.29286689e-01 5.85114956e-01 -1.17729807e+00 -3.26114893e-01 -7.00359121e-02 7.89575815e-01 -3.58362645e-01 4.41435248e-01 -5.88830352e-01 -1.18132293e+00 -5.08568227e-01 -1.25579619e+00 1.15136206e+00 -2.87345439e-01 -3.67811650e-01 -4.91479278e-01 1.35265261e-01 9.19707477e-01 7.06283391e-01 -9.83639479e-01 4.69159484e-01 -1.13801682e+00 -2.53360629e-01 -1.52900919e-01 1.13149397e-01 7.88977981e-01 4.83607024e-01 -2.84292668e-01 -1.53164709e+00 -3.80340487e-01 2.78268576e-01 -2.06564665e-02 7.26994336e-01 3.57893139e-01 8.64824831e-01 -5.50500512e-01 -9.37216207e-02 6.94774806e-01 6.01089299e-01 5.27351558e-01 5.28136969e-01 -2.63418496e-01 5.61616421e-01 4.80669409e-01 2.98576266e-01 2.86331743e-01 3.71247441e-01 1.10899997e+00 -1.82606503e-01 -1.55130133e-01 -5.13129175e-01 -2.32937708e-02 7.88664699e-01 1.65797710e+00 3.73640150e-01 -5.71201324e-01 -7.78554082e-01 6.28780127e-01 -1.21443462e+00 -8.04913938e-01 1.21392071e-01 2.49022007e+00 1.15454292e+00 -3.06325436e-01 2.27872074e-01 2.58969873e-01 9.91904736e-01 3.64079207e-01 -2.34111130e-01 -5.33113658e-01 -4.09507483e-01 2.04091907e-01 5.14829874e-01 8.59431446e-01 -8.98615599e-01 9.44506884e-01 6.95827103e+00 1.32946992e+00 -1.36879575e+00 4.09327328e-01 6.55992031e-01 -5.41487634e-01 -2.21451223e-01 -1.65333673e-01 -1.09641767e+00 2.46614337e-01 1.51375949e+00 -2.24574581e-02 7.13044643e-01 7.14114547e-01 4.82988060e-01 9.90374535e-02 -1.09095716e+00 1.25316691e+00 3.73998791e-01 -6.76156998e-01 -1.87908351e-01 -2.35587195e-01 2.74331689e-01 1.25093073e-01 4.96408902e-02 3.48678678e-01 7.77164027e-02 -8.38785172e-01 8.39118659e-01 -3.46113563e-01 7.46438920e-01 -5.98404229e-01 4.48826134e-01 1.47966817e-01 -1.23917723e+00 9.52226520e-02 9.93433371e-02 2.44118333e-01 3.59560847e-01 4.83682632e-01 -1.29345417e+00 4.93500978e-01 4.78978008e-01 -1.54656336e-01 -1.12460501e-01 7.51851797e-01 -7.69823045e-02 1.31939542e+00 -4.74511504e-01 3.95913005e-01 -1.28545225e-01 1.42887859e-02 8.89932275e-01 1.56632173e+00 3.77931744e-01 3.32685583e-03 -5.52497357e-02 4.72101033e-01 1.40912803e-02 2.44204178e-01 -1.11407146e-01 -2.15568215e-01 8.10611546e-01 1.16818845e+00 -3.49191129e-01 -3.29424173e-01 -2.56879419e-01 1.14145696e+00 3.36854830e-02 5.13700008e-01 -9.19931412e-01 -4.94985759e-01 8.72125924e-01 1.19150408e-01 4.57372308e-01 -3.85386825e-01 -1.76804155e-01 -1.07291448e+00 1.34930059e-01 -1.29806554e+00 1.47853836e-01 -4.25604790e-01 -9.25060213e-01 1.09433877e+00 -1.44035563e-01 -9.40771461e-01 -3.20990741e-01 -1.16295345e-01 -4.63459611e-01 1.10829997e+00 -1.40133119e+00 -1.15147984e+00 1.78444475e-01 5.40956557e-01 1.00486481e+00 -1.20996535e-01 7.54216552e-01 8.07867587e-01 -8.31141293e-01 1.19166040e+00 8.48270804e-02 -5.98948486e-02 9.95596468e-01 -9.95038211e-01 6.18108749e-01 1.03416204e+00 2.35848501e-01 7.22539663e-01 8.40045333e-01 -2.53928781e-01 -1.25423610e+00 -1.07745695e+00 1.12952852e+00 -2.54455179e-01 4.14640963e-01 -9.35986698e-01 -1.00943005e+00 7.20220864e-01 3.00452262e-01 -2.29830593e-01 9.22912300e-01 3.55210185e-01 -4.67535436e-01 -4.05480385e-01 -8.39511693e-01 4.60238427e-01 9.44924593e-01 -8.16427469e-01 -6.90623641e-01 4.16975692e-02 1.07295215e+00 -3.31921846e-01 -5.36457717e-01 1.45027693e-02 3.21558028e-01 -7.42498457e-01 7.91181862e-01 -2.48890780e-02 -6.13387585e-01 -2.82516479e-01 -6.46072090e-01 -1.47704518e+00 1.20011000e-02 -1.33202732e+00 4.86068353e-02 1.77704346e+00 6.52444899e-01 -6.76581204e-01 2.64176369e-01 6.26375794e-01 -6.70576692e-01 -1.50116697e-01 -1.16637218e+00 -1.18220460e+00 -1.87955901e-01 -5.36277115e-01 3.73218924e-01 6.51797771e-01 -2.62761619e-02 3.71777654e-01 -5.47586262e-01 4.88788843e-01 1.95387766e-01 -2.48612821e-01 6.60249770e-01 -6.17226005e-01 -6.47615194e-01 -2.63294816e-01 2.79990714e-02 -1.48564756e+00 1.70258641e-01 -9.34710085e-01 5.65933287e-01 -1.03524697e+00 -1.67153284e-01 -4.94692236e-01 -1.77709922e-01 4.46286112e-01 -2.26903230e-01 -8.05691034e-02 2.54825145e-01 1.13618039e-02 -3.14376950e-01 7.87135303e-01 4.86371219e-01 -1.85741901e-01 -6.01748407e-01 3.66628617e-01 -6.62884772e-01 4.74481672e-01 7.21958876e-01 -5.65140128e-01 -4.09103155e-01 -5.69762111e-01 -6.39213204e-01 1.83993816e-01 -3.45348775e-01 -9.13169801e-01 1.46671176e-01 1.78356051e-01 -5.12838364e-01 -5.04797280e-01 8.56063545e-01 -5.54510593e-01 1.55117571e-01 -6.80111796e-02 -4.17080581e-01 -1.50141433e-01 4.31166470e-01 1.74437478e-01 -3.78918558e-01 -2.59053353e-02 8.75959456e-01 2.48046920e-01 -5.56325018e-02 4.94544804e-02 -8.12134981e-01 1.16603449e-01 5.00058711e-01 2.28132419e-02 -1.01199627e-01 -5.64376771e-01 -6.63188338e-01 -2.11158872e-01 -8.44974592e-02 5.79853356e-01 2.26450711e-01 -9.23967004e-01 -7.41575718e-01 3.68338734e-01 -9.56572890e-02 -1.53553098e-01 5.19124448e-01 9.34326410e-01 -3.79416198e-02 5.84995091e-01 6.10763609e-01 -5.21327198e-01 -1.72613704e+00 2.87715107e-01 5.93454599e-01 5.04096877e-03 -3.05808306e-01 1.13825130e+00 4.15958017e-01 -5.80521762e-01 5.66117167e-01 -4.23319697e-01 3.49957049e-01 2.23317929e-02 6.38971925e-01 5.06729722e-01 6.95927739e-01 -9.36750174e-01 -5.95089257e-01 2.11140230e-01 -1.83008924e-01 -5.75644851e-01 9.46646452e-01 -7.09431589e-01 2.86004752e-01 7.21381366e-01 1.32759309e+00 5.44086218e-01 -9.15149152e-01 -2.34668612e-01 -2.45265529e-01 -2.61263549e-01 3.91938299e-01 -8.36519420e-01 -8.49251330e-01 1.04504919e+00 5.19204378e-01 1.65481567e-01 1.13437009e+00 2.83377175e-03 9.49661732e-01 3.49124461e-01 -7.94416964e-02 -1.08434033e+00 -1.90578669e-01 5.56372583e-01 1.08328962e+00 -1.09753025e+00 -5.43651223e-01 -5.58940709e-01 -7.83692598e-01 7.29835987e-01 3.43101948e-01 5.42046428e-01 5.69567800e-01 5.61312854e-01 5.86592019e-01 4.53803450e-01 -7.78514862e-01 -1.52130470e-01 2.67917424e-01 4.66635257e-01 7.53322423e-01 1.80341721e-01 4.34114039e-02 8.14105392e-01 -4.94579136e-01 -6.56171262e-01 2.64517665e-01 6.60690784e-01 -3.90882641e-01 -1.19407928e+00 -5.99476159e-01 -8.04945379e-02 -6.89928412e-01 -5.98744988e-01 -5.02393901e-01 2.49226451e-01 -4.18648541e-01 1.61474395e+00 1.03937797e-01 -4.88499224e-01 3.53327990e-01 3.57143164e-01 8.25195536e-02 -6.99464858e-01 -8.36585820e-01 8.17496717e-01 4.51551825e-01 -3.47372204e-01 2.62203980e-02 -5.39671779e-01 -1.04977179e+00 -1.20086215e-01 -8.13968480e-01 2.81090915e-01 9.87803042e-01 9.89329278e-01 6.56219661e-01 6.39785051e-01 9.18204546e-01 -7.42726505e-01 -8.49761367e-01 -1.33262300e+00 -7.75622249e-01 3.32108885e-02 5.70223391e-01 -2.07770392e-01 -7.19232559e-01 3.98653001e-03]
[14.628008842468262, 6.302112102508545]
56cb79b7-b83b-4dac-b216-714f9065706c
multiview-compressive-coding-for-3d
2301.08247
null
https://arxiv.org/abs/2301.08247v1
https://arxiv.org/pdf/2301.08247v1.pdf
Multiview Compressive Coding for 3D Reconstruction
A central goal of visual recognition is to understand objects and scenes from a single image. 2D recognition has witnessed tremendous progress thanks to large-scale learning and general-purpose representations. Comparatively, 3D poses new challenges stemming from occlusions not depicted in the image. Prior works try to overcome these by inferring from multiple views or rely on scarce CAD models and category-specific priors which hinder scaling to novel settings. In this work, we explore single-view 3D reconstruction by learning generalizable representations inspired by advances in self-supervised learning. We introduce a simple framework that operates on 3D points of single objects or whole scenes coupled with category-agnostic large-scale training from diverse RGB-D videos. Our model, Multiview Compressive Coding (MCC), learns to compress the input appearance and geometry to predict the 3D structure by querying a 3D-aware decoder. MCC's generality and efficiency allow it to learn from large-scale and diverse data sources with strong generalization to novel objects imagined by DALL$\cdot$E 2 or captured in-the-wild with an iPhone.
['Georgia Gkioxari', 'Christoph Feichtenhofer', 'Jitendra Malik', 'Justin Johnson', 'Chao-yuan Wu']
2023-01-19
null
http://openaccess.thecvf.com//content/CVPR2023/html/Wu_Multiview_Compressive_Coding_for_3D_Reconstruction_CVPR_2023_paper.html
http://openaccess.thecvf.com//content/CVPR2023/papers/Wu_Multiview_Compressive_Coding_for_3D_Reconstruction_CVPR_2023_paper.pdf
cvpr-2023-1
['single-view-3d-reconstruction']
['computer-vision']
[ 4.66992944e-01 1.11703672e-01 -1.76866487e-01 -4.84029114e-01 -8.82879496e-01 -7.60946631e-01 5.96516371e-01 -4.49979156e-01 2.35207930e-01 2.86739707e-01 1.46998942e-01 -8.38699937e-03 6.13748934e-03 -5.27900755e-01 -1.02420282e+00 -3.60142410e-01 1.94531903e-01 5.18158793e-01 1.33458987e-01 -1.10341264e-02 2.57330269e-01 9.39926445e-01 -2.03529620e+00 4.52582359e-01 2.18633041e-01 1.23047400e+00 5.60818017e-01 6.78671062e-01 -1.46277741e-01 5.69830477e-01 -7.84788877e-02 -9.82060060e-02 7.76545048e-01 -1.00777939e-01 -3.67670923e-01 8.16949308e-01 1.04080272e+00 -6.93953097e-01 -6.57835722e-01 6.33501172e-01 3.85007262e-01 -2.43253157e-01 6.13363624e-01 -1.20188832e+00 -6.28766775e-01 -3.55330348e-01 -4.44008291e-01 -4.97022122e-02 7.27242529e-01 2.73393273e-01 8.21277440e-01 -1.07107735e+00 8.52685571e-01 1.30610800e+00 6.14809275e-01 4.81046021e-01 -1.31095338e+00 -1.64547205e-01 3.95215869e-01 7.63031840e-02 -1.31638217e+00 -5.42021096e-01 1.19036007e+00 -4.95388806e-01 1.12019694e+00 9.97779444e-02 7.01424479e-01 1.33862197e+00 -2.28830814e-01 7.68139780e-01 1.10646737e+00 -3.70543003e-01 3.62278134e-01 6.22722059e-02 -5.11016965e-01 6.13290608e-01 2.06893504e-01 1.97691604e-01 -7.90120900e-01 1.90366004e-02 1.14266396e+00 6.18230700e-01 -3.14672142e-01 -1.15622735e+00 -1.22151792e+00 5.92071831e-01 4.43883449e-01 -1.54389486e-01 -3.26053590e-01 1.81280732e-01 -1.04707986e-01 3.53077263e-01 5.15950143e-01 1.18197300e-01 -7.14752793e-01 8.96129757e-02 -7.42839992e-01 4.19218726e-02 5.40530026e-01 1.39036167e+00 9.64741826e-01 2.00434387e-01 5.87595165e-01 6.15317702e-01 3.65570337e-01 9.32496190e-01 2.29863733e-01 -1.34659612e+00 5.21577656e-01 5.69377959e-01 -5.38380258e-02 -1.09546614e+00 -1.82101220e-01 -4.25737768e-01 -7.62465358e-01 3.72349709e-01 2.37579584e-01 4.06879514e-01 -9.79553044e-01 1.46610475e+00 3.85356694e-01 -5.65442070e-02 -2.00673372e-01 9.93914843e-01 5.74004471e-01 3.57720971e-01 -5.11597633e-01 1.85683779e-02 8.19162905e-01 -5.96754372e-01 2.14083195e-02 -5.00376403e-01 2.02999532e-01 -6.31391406e-01 8.29741418e-01 6.31956518e-01 -1.03529918e+00 -7.67427564e-01 -8.75834346e-01 -2.96579719e-01 -3.89552355e-01 -6.69788495e-02 5.04284263e-01 5.65568566e-01 -9.27292705e-01 2.71183580e-01 -7.74468124e-01 -4.95114356e-01 6.71480596e-01 1.64376825e-01 -7.57236242e-01 -7.67310202e-01 -3.05684060e-01 6.48165822e-01 3.35770589e-03 -7.15551600e-02 -1.25221288e+00 -6.71518624e-01 -1.05294979e+00 -3.07499856e-01 6.50884986e-01 -9.94227171e-01 8.51208329e-01 -8.86198580e-01 -1.32054055e+00 1.22666121e+00 -8.73553008e-02 -2.36759931e-02 4.17948902e-01 -3.21270794e-01 -9.79517102e-02 6.06069148e-01 1.58087239e-01 7.34601915e-01 1.25793374e+00 -1.73913538e+00 -2.89216161e-01 -9.14209545e-01 2.82163560e-01 2.17269823e-01 7.69282058e-02 -5.90229154e-01 -5.36218703e-01 -3.93574715e-01 9.00069594e-01 -8.27753901e-01 -2.13003144e-01 6.25197351e-01 -2.80572206e-01 3.45229417e-01 9.67676163e-01 -4.83981043e-01 2.23827213e-01 -2.09713864e+00 4.15482104e-01 -6.32903948e-02 1.93592414e-01 -1.09518230e-01 -2.21085444e-01 4.55315828e-01 -3.14637460e-02 -2.14661494e-01 4.41994518e-03 -4.65189695e-01 5.01491278e-02 4.60109174e-01 -4.69831228e-01 6.67532861e-01 2.69649506e-01 8.72700095e-01 -8.62380087e-01 -2.75954723e-01 6.17116868e-01 5.56692541e-01 -9.05911684e-01 5.14249265e-01 -4.75013316e-01 5.85266769e-01 -4.65278983e-01 1.14687526e+00 7.78551579e-01 -6.72609746e-01 9.91546810e-02 -4.46155548e-01 6.78090677e-02 -2.36874949e-02 -1.32762575e+00 2.55946732e+00 -5.05029917e-01 3.72846484e-01 2.24905118e-01 -1.49793124e+00 9.81401563e-01 6.60783947e-02 4.11792815e-01 -5.19591749e-01 -1.91528395e-01 3.16278845e-01 -8.03139806e-01 -6.90783262e-01 2.26780236e-01 -1.97253138e-01 -2.15893593e-02 3.02821964e-01 2.96015531e-01 -8.46016884e-01 -5.62397063e-01 2.72018284e-01 9.32711959e-01 5.90844870e-01 3.57391328e-01 2.12282971e-01 1.91735953e-01 -1.53143466e-01 2.61252403e-01 7.90952682e-01 1.51289448e-01 1.05281842e+00 -9.79811139e-03 -6.27591431e-01 -1.16961074e+00 -1.45946181e+00 5.60685471e-02 6.60061300e-01 1.93571046e-01 -2.54588872e-01 1.26663625e-01 -6.96217775e-01 2.83974230e-01 2.93654859e-01 -4.20307875e-01 7.05285883e-03 -3.75542969e-01 -1.08061107e-02 -1.38090029e-01 4.41101909e-01 3.23062450e-01 -4.38901365e-01 -8.59523714e-01 -7.49435797e-02 1.90380402e-02 -1.35560000e+00 -1.12237304e-01 3.10905933e-01 -1.11022937e+00 -1.08203089e+00 -8.65352273e-01 -6.52244568e-01 7.22884595e-01 9.41230178e-01 1.13375151e+00 -3.06283891e-01 -5.93998313e-01 1.24629784e+00 -2.73916215e-01 -7.64937475e-02 -9.83520993e-04 -2.80112654e-01 1.44931167e-01 1.66894183e-01 1.56511828e-01 -1.27808917e+00 -6.97646141e-01 2.29249462e-01 -8.16987276e-01 1.56648770e-01 8.21401894e-01 6.29882991e-01 9.24578190e-01 -4.71877962e-01 3.50301325e-01 -4.97787833e-01 -4.08697873e-01 -5.61151147e-01 -4.71437633e-01 9.86788794e-02 -2.81264246e-01 -4.92416583e-02 6.51625514e-01 -2.79301226e-01 -8.28495920e-01 5.25626183e-01 1.36396140e-01 -1.20549536e+00 -6.51776493e-01 -5.11979423e-02 -5.39512813e-01 -5.53515032e-02 5.97366333e-01 5.22745967e-01 9.39004272e-02 -7.64809787e-01 6.67571604e-01 5.73139966e-01 6.18635952e-01 -5.71665049e-01 8.99027050e-01 9.55837667e-01 1.80999279e-01 -1.01459479e+00 -1.13369882e+00 -4.69369024e-01 -1.11292744e+00 -2.09920391e-01 6.52897954e-01 -1.41185796e+00 -4.39433336e-01 1.87340453e-01 -1.02428269e+00 -9.34824646e-02 -5.26056528e-01 4.07779604e-01 -9.34631288e-01 4.55408633e-01 -1.02231540e-01 -7.08764851e-01 2.16862917e-01 -8.19202244e-01 1.66697681e+00 -9.73201171e-02 2.32956201e-01 -7.25517750e-01 -1.47965744e-01 6.23995900e-01 2.27276236e-01 3.69453490e-01 6.57999754e-01 -1.55897602e-01 -1.34301484e+00 -2.10635528e-01 -2.21243322e-01 4.98041272e-01 1.69174194e-01 -5.03707528e-01 -1.15685451e+00 -2.63589919e-01 1.57961652e-01 -7.81481922e-01 6.52636528e-01 2.82510936e-01 1.43624377e+00 -1.52797565e-01 -2.45286644e-01 1.09541106e+00 1.67897499e+00 -2.34586880e-01 2.41039693e-01 -3.45266424e-02 8.54171574e-01 5.25303602e-01 1.62345007e-01 6.07637823e-01 4.55520391e-01 7.74419844e-01 8.84922147e-01 2.41406664e-01 -3.15375179e-01 -5.91610730e-01 1.13311693e-01 8.80120873e-01 -9.01052877e-02 -1.85643006e-02 -7.87750185e-01 4.33125317e-01 -1.34376001e+00 -8.88420582e-01 3.07448000e-01 2.08330250e+00 3.82433772e-01 1.27176782e-02 -1.75813943e-01 6.60000416e-03 3.13080698e-01 2.56998360e-01 -1.05371594e+00 1.25622123e-01 -4.57799345e-01 1.33270800e-01 3.18550080e-01 2.29615927e-01 -7.11270034e-01 4.62910533e-01 5.83798456e+00 4.83639181e-01 -1.16042650e+00 6.34599058e-03 5.61299324e-01 -4.22806263e-01 -6.15804672e-01 1.49641126e-01 -7.03612089e-01 -3.65491658e-02 4.23994273e-01 1.89366609e-01 4.29231972e-01 1.05924523e+00 -2.02338636e-01 4.28706110e-02 -1.39132524e+00 1.59950650e+00 6.61143780e-01 -1.46324301e+00 3.31510216e-01 3.67383331e-01 8.30705047e-01 2.60441363e-01 1.22767977e-01 9.43903346e-04 8.77172351e-02 -8.54643703e-01 8.13083053e-01 7.64427364e-01 1.18742955e+00 -2.67219514e-01 7.58303776e-02 6.40936732e-01 -1.09510553e+00 -2.14268535e-01 -4.33122665e-01 -2.17726216e-01 3.10526818e-01 5.02899706e-01 -5.19433141e-01 6.81108296e-01 7.32279122e-01 1.31912231e+00 -5.80243289e-01 5.72464645e-01 2.55593918e-02 9.79095548e-02 -6.36967778e-01 4.33318287e-01 2.69444417e-02 -9.75729600e-02 5.17461002e-01 7.54100919e-01 5.47756672e-01 2.95584977e-01 2.95942098e-01 8.14144433e-01 -3.68407853e-02 -4.02979761e-01 -1.11192465e+00 1.79957569e-01 1.97086468e-01 1.01742876e+00 -4.36718643e-01 -1.42120019e-01 -7.60401905e-01 1.19704533e+00 3.03022563e-01 3.02622437e-01 -4.43174601e-01 1.83843389e-01 5.48520029e-01 3.60818267e-01 9.19377983e-01 -6.87388420e-01 -7.04725757e-02 -1.55152190e+00 1.98748261e-01 -6.80424869e-01 1.04065411e-01 -1.23357224e+00 -1.55463779e+00 4.62945849e-01 7.21168444e-02 -1.69193316e+00 -3.13844919e-01 -8.86887074e-01 -1.49818948e-02 5.27454555e-01 -1.64545536e+00 -1.31890655e+00 -5.64391017e-01 9.90669847e-01 7.51055181e-01 -2.09891304e-01 7.87830830e-01 1.22804880e-01 8.66820216e-02 1.89006731e-01 7.36820847e-02 -2.57127225e-01 5.98119438e-01 -9.51891422e-01 1.32376477e-01 5.58924437e-01 4.81574178e-01 1.95083722e-01 3.92051786e-01 -3.14737201e-01 -2.12809587e+00 -8.88095796e-01 5.13421774e-01 -8.30333769e-01 1.74452692e-01 -6.33341134e-01 -5.20912349e-01 7.04367161e-01 -3.22051585e-01 6.25673532e-01 4.78021324e-01 -1.20600320e-01 -8.79748881e-01 -2.10368648e-01 -1.03609526e+00 1.82107955e-01 1.73209727e+00 -9.63115990e-01 -6.26114964e-01 3.50348622e-01 6.36532009e-01 -4.04394269e-01 -7.30104029e-01 2.62112200e-01 6.19770646e-01 -1.23855805e+00 1.56299734e+00 -5.91723204e-01 6.63963497e-01 -2.46370405e-01 -1.08395672e+00 -9.75779355e-01 5.54733686e-02 -4.24383283e-01 -3.97378772e-01 6.71387911e-01 -1.87625512e-01 -3.44471455e-01 9.14844811e-01 4.65199977e-01 -3.06595862e-01 -5.56371450e-01 -9.65617359e-01 -8.04805458e-01 -1.57986045e-01 -7.51440942e-01 4.96384203e-01 7.37750053e-01 -4.88086879e-01 3.17160606e-01 -6.09105349e-01 4.31993306e-01 1.09755886e+00 7.85353422e-01 1.20531845e+00 -1.21382642e+00 -6.45637631e-01 8.31996053e-02 -7.67523050e-01 -1.93579769e+00 1.76233128e-02 -9.50874448e-01 -2.18058378e-01 -1.27254164e+00 6.14715703e-02 -4.46289062e-01 1.55742943e-01 2.18737751e-01 4.46848750e-01 3.71433735e-01 4.11989272e-01 3.97152722e-01 -8.89298677e-01 8.79096329e-01 1.23696041e+00 -1.69196442e-01 1.38205960e-01 -7.26032034e-02 -5.68013430e-01 7.83906698e-01 2.89145589e-01 -2.04126000e-01 -5.13444722e-01 -7.89546609e-01 2.35256657e-01 3.33967507e-01 8.28777254e-01 -9.31215763e-01 1.79765418e-01 -2.00954840e-01 7.19887972e-01 -8.38446736e-01 8.78734767e-01 -1.16009378e+00 3.45587075e-01 -2.13761851e-02 -1.89504221e-01 -2.02134088e-01 -1.26394540e-01 1.12661159e+00 -9.05713588e-02 1.40340000e-01 4.10286754e-01 -6.61248565e-01 -9.17629719e-01 6.38611794e-01 1.15043789e-01 1.05297513e-01 8.06236744e-01 -6.85100436e-01 1.33384010e-02 -5.13999641e-01 -9.19850647e-01 -1.83380082e-01 8.63565922e-01 5.20256102e-01 1.04775476e+00 -1.22130442e+00 -5.70644259e-01 7.98348188e-01 4.33060557e-01 4.38250244e-01 5.38822293e-01 3.90745103e-01 -3.77573699e-01 4.69144285e-01 -3.43306690e-01 -1.20896363e+00 -8.57018232e-01 7.45527089e-01 2.17335999e-01 3.85667890e-01 -9.52546477e-01 7.90927172e-01 4.21862900e-01 -6.02543712e-01 2.65526146e-01 -2.14074224e-01 4.74514931e-01 -2.59825528e-01 3.44067395e-01 -5.11507094e-02 -5.33217527e-02 -6.78016484e-01 -3.33699018e-01 9.92519438e-01 2.21874163e-01 2.81380545e-02 1.65942979e+00 -5.17802060e-01 2.97291845e-01 7.61974573e-01 1.51297736e+00 -1.24902114e-01 -1.88530707e+00 -5.34581661e-01 -4.37910706e-01 -1.00784922e+00 3.31049673e-02 -5.10409355e-01 -1.02789164e+00 1.07636786e+00 5.50662100e-01 -1.71603262e-01 9.88015354e-01 5.59714198e-01 5.00146627e-01 6.57644212e-01 8.99537206e-01 -7.45333374e-01 5.84665895e-01 2.60016054e-01 1.02475905e+00 -1.58367097e+00 1.03077829e-01 -2.14424849e-01 -3.91067177e-01 1.25568783e+00 3.77121419e-01 -2.24918485e-01 9.66292739e-01 -1.21371649e-01 -1.95616543e-01 -2.86796898e-01 -7.57056177e-01 -1.28583923e-01 3.29070181e-01 1.02854741e+00 -1.80987135e-01 -2.31148630e-01 8.01233947e-01 2.97647119e-01 6.87254146e-02 -8.44722986e-02 4.20023352e-01 1.04807627e+00 -3.35239887e-01 -8.86338234e-01 -2.90159583e-01 4.17533815e-01 1.77514032e-01 2.07638159e-01 -2.23095149e-01 6.43253684e-01 1.71955377e-01 7.34934151e-01 9.93080735e-02 -2.84441799e-01 3.08782160e-01 7.39752948e-02 9.92249906e-01 -8.21839988e-01 2.83845067e-01 7.77143463e-02 -2.94930369e-01 -1.01776063e+00 -8.30967724e-01 -9.33429062e-01 -5.66113234e-01 3.67477648e-02 3.10865659e-02 -4.98630166e-01 7.92030275e-01 8.53634715e-01 7.07142353e-01 4.47511896e-02 8.64498317e-01 -1.48505688e+00 -5.41563034e-01 -4.42581683e-01 -8.33381712e-01 5.10467231e-01 7.74106443e-01 -8.18249464e-01 -6.73542082e-01 3.23368430e-01]
[8.403233528137207, -3.029836893081665]
5f195e09-8777-4be4-a49f-39ef59fae576
the-mapillary-vistas-dataset-for-semantic
null
null
http://openaccess.thecvf.com/content_iccv_2017/html/Neuhold_The_Mapillary_Vistas_ICCV_2017_paper.html
http://openaccess.thecvf.com/content_ICCV_2017/papers/Neuhold_The_Mapillary_Vistas_ICCV_2017_paper.pdf
The Mapillary Vistas Dataset for Semantic Understanding of Street Scenes
The Mapillary Vistas Dataset is a novel, large-scale street-level image dataset containing 25,000 high-resolution images annotated into 66 object categories with additional, instance-specific labels for 37 classes. Annotation is performed in a dense and fine-grained style by using polygons for delineating individual objects. Our dataset is 5x larger than the total amount of fine annotations for Cityscapes and contains images from all around the world, captured at various conditions regarding weather, season and daytime. Images come from different imaging devices (mobile phones, tablets, action cameras, professional capturing rigs) and differently experienced photographers. In such a way, our dataset has been designed and compiled to cover diversity, richness of detail and geographic extent. As default benchmark tasks, we define semantic image segmentation and instance-specific image segmentation, aiming to significantly further the development of state-of-the-art methods for visual road-scene understanding.
['Gerhard Neuhold', 'Peter Kontschieder', 'Tobias Ollmann', 'Samuel Rota Bulo']
2017-10-01
null
null
null
iccv-2017-10
['road-scene-understanding']
['computer-vision']
[ 3.88021678e-01 1.77897885e-01 -1.47004709e-01 -4.74245727e-01 -5.71276546e-01 -9.28737819e-01 9.67001975e-01 2.13886902e-01 -3.53289157e-01 6.33585930e-01 2.57357270e-01 -2.98156381e-01 1.78569719e-01 -1.24180734e+00 -6.83601439e-01 -3.23735595e-01 -4.93565612e-02 6.52862012e-01 6.33449495e-01 -2.63480276e-01 2.41201699e-01 7.14893401e-01 -1.96006536e+00 2.89226562e-01 8.86374354e-01 9.39143717e-01 4.82217610e-01 7.56346524e-01 -3.43668878e-01 7.03148782e-01 -3.72903049e-01 -3.27031791e-01 4.24185425e-01 1.94633812e-01 -9.23943102e-01 6.54608428e-01 9.69568491e-01 -1.97437420e-01 -1.35612771e-01 1.01363945e+00 3.33258286e-02 9.83024314e-02 5.56369424e-01 -9.15630758e-01 -5.16959012e-01 2.82625437e-01 -6.76930130e-01 3.18570733e-01 9.48292390e-02 3.40642214e-01 9.12274480e-01 -5.03632307e-01 7.39329755e-01 1.14761937e+00 6.68312371e-01 -1.23357169e-01 -1.17635214e+00 -3.48159343e-01 3.79500955e-01 -1.76460985e-02 -1.67901921e+00 -2.49094084e-01 3.63783896e-01 -6.39972985e-01 8.96284699e-01 4.00243253e-01 7.03172684e-01 8.40250134e-01 -5.57669640e-01 4.07087386e-01 1.48138762e+00 -1.20065168e-01 2.02868760e-01 2.36181155e-01 8.59495029e-02 5.56982100e-01 2.95362920e-01 -2.91736037e-01 1.25808651e-02 2.01189026e-01 8.94093752e-01 8.17094520e-02 -1.08037405e-01 -2.74288893e-01 -1.35109377e+00 4.98327762e-01 6.26023769e-01 1.35016337e-01 -3.97970021e-01 -8.91362652e-02 2.79353887e-01 -1.50906473e-01 6.69184864e-01 1.31841406e-01 -5.28454840e-01 -3.64487730e-02 -1.03947532e+00 1.81279197e-01 5.22556126e-01 1.00800157e+00 1.22772753e+00 -7.37273842e-02 3.41801278e-05 9.70938981e-01 8.33298713e-02 7.10421264e-01 -1.07660592e-01 -1.20514143e+00 8.39778841e-01 7.94634759e-01 3.20214778e-01 -1.10752499e+00 -6.31570697e-01 -1.38243601e-01 -7.11086690e-01 2.47679815e-01 5.08817852e-01 1.00948334e-01 -1.31108236e+00 1.18888748e+00 4.50065494e-01 2.44556263e-01 -1.72884792e-01 9.05378640e-01 1.08817136e+00 6.99428737e-01 3.18752885e-01 3.88270140e-01 1.71069551e+00 -9.69572604e-01 -3.70515108e-01 -5.53638399e-01 2.99112022e-01 -5.68850279e-01 1.41120434e+00 2.02798948e-01 -7.00781107e-01 -6.53101802e-01 -7.27409124e-01 -2.15521365e-01 -1.16867840e+00 2.49081418e-01 7.06847429e-01 6.84777677e-01 -1.00081551e+00 1.10018909e-01 -2.47451335e-01 -6.45030618e-01 8.40466797e-01 -2.59346098e-01 -4.72426087e-01 3.94002609e-02 -1.03880036e+00 7.46327758e-01 5.54432452e-01 -1.80188026e-02 -7.50642300e-01 -8.60999882e-01 -9.15729582e-01 -3.94129418e-02 3.12111527e-01 -4.06654119e-01 7.65465856e-01 -8.16852391e-01 -1.12089968e+00 1.63787758e+00 8.74976516e-02 -5.93761027e-01 5.80887973e-01 3.84462823e-04 -5.51496744e-01 2.96993583e-01 5.00332654e-01 9.19763505e-01 5.55584729e-01 -1.57918668e+00 -1.03132582e+00 -2.96977729e-01 4.02282268e-01 2.54969507e-01 2.29614660e-01 -8.28281417e-02 -7.45743752e-01 -6.11669898e-01 -2.96400487e-01 -8.15079689e-01 -5.09576976e-01 -1.96821913e-01 -5.34454048e-01 2.50130892e-01 9.08031881e-01 -6.56316161e-01 1.07773864e+00 -2.16459060e+00 -3.17851305e-01 2.93621063e-01 1.99428394e-01 3.29965740e-01 -5.38244750e-03 2.70860165e-01 1.08199775e-01 5.11555254e-01 -5.66909850e-01 -2.41107836e-01 -1.07993089e-01 4.71592516e-01 -4.00047958e-01 2.56079823e-01 2.91586127e-02 8.42393994e-01 -8.11644197e-01 -7.94591546e-01 8.87028515e-01 3.12129319e-01 -5.09810038e-02 2.55047493e-02 -2.76410133e-01 5.58217525e-01 -4.80418891e-01 8.35386038e-01 8.98518801e-01 -2.43143886e-01 -1.91880390e-01 -1.87096208e-01 -5.50451815e-01 -1.10257678e-01 -1.25626063e+00 1.50889635e+00 -6.88665271e-01 1.13376915e+00 1.41164452e-01 -6.41658902e-01 7.39162564e-01 -3.14102709e-01 3.11982036e-01 -9.02532220e-01 1.87593512e-02 -3.37650366e-02 -7.76344478e-01 -5.18243790e-01 9.69900012e-01 5.85673392e-01 -2.91253775e-01 5.30554168e-02 -4.10159796e-01 -5.70002079e-01 4.41823304e-01 -1.82424847e-03 5.51145613e-01 1.98333785e-02 2.99931109e-01 -3.05286318e-01 4.42853153e-01 6.43762231e-01 9.32140723e-02 8.30001354e-01 -2.15665400e-01 1.09002686e+00 3.13748986e-01 -6.36021018e-01 -1.36091745e+00 -1.16074586e+00 -6.07844889e-01 1.06153035e+00 2.74786860e-01 -2.05458313e-01 -1.03850019e+00 -3.01452756e-01 -3.65788192e-02 4.59581882e-01 -7.87025869e-01 7.51614392e-01 -4.68032688e-01 -7.27024257e-01 7.73708940e-01 4.08122867e-01 1.27189982e+00 -1.02849090e+00 -8.09805274e-01 -7.78927058e-02 -2.65070468e-01 -1.71622598e+00 -8.43661427e-02 -1.90661907e-01 -4.42228258e-01 -1.27681184e+00 -6.79609358e-01 -4.70358431e-01 6.28301740e-01 5.11979282e-01 1.53685677e+00 -2.74654239e-01 -5.14806271e-01 4.33997214e-01 -3.24111849e-01 -2.03911304e-01 1.21840239e-01 9.94720832e-02 -5.33613324e-01 1.77858129e-01 -1.39960172e-02 -3.34897965e-01 -5.86680830e-01 5.10118365e-01 -8.74879837e-01 2.12308183e-01 4.14746463e-01 2.29432195e-01 8.95879805e-01 2.69816637e-01 -1.17255151e-02 -1.02694941e+00 2.05268916e-02 -6.31671190e-01 -9.72271621e-01 1.51690662e-01 -9.68983173e-02 -4.65923518e-01 4.19238120e-01 9.72855389e-02 -1.32201123e+00 1.30959958e-01 -7.53751621e-02 4.27426025e-02 -1.03067517e+00 8.31935033e-02 -3.30776334e-01 -9.03260484e-02 6.46516979e-01 1.18603528e-01 -6.16571546e-01 -3.57809126e-01 9.11812305e-01 8.24655235e-01 1.02255392e+00 -4.30412382e-01 8.08701277e-01 9.08796608e-01 -1.23712681e-01 -1.38584721e+00 -1.06820941e+00 -7.30750978e-01 -9.31814551e-01 -3.34250897e-01 1.27632427e+00 -1.21252394e+00 -2.61794448e-01 7.26630747e-01 -7.53917992e-01 -8.01807821e-01 -2.41241872e-01 -2.10448466e-02 -5.01745284e-01 1.12427123e-01 -1.01630762e-01 -4.80681300e-01 -7.89607540e-02 -9.57138717e-01 1.48096454e+00 4.76579070e-01 7.83031136e-02 -1.01878202e+00 -8.83878171e-02 6.71754181e-01 3.29882264e-01 7.20819235e-01 4.94900286e-01 9.31428596e-02 -7.66187668e-01 1.36695713e-01 -9.55200255e-01 2.55809069e-01 4.87794764e-02 3.50790322e-01 -1.08609748e+00 4.87313211e-01 -7.55218327e-01 -7.49654397e-02 1.02252042e+00 5.63929975e-01 1.43567204e+00 -2.13314250e-01 -3.86033565e-01 7.71733105e-01 1.67807484e+00 -8.83170515e-02 9.59189475e-01 8.12412441e-01 1.24286270e+00 8.01015377e-01 7.82821834e-01 3.71859044e-01 8.93333733e-01 7.92948604e-01 7.11697519e-01 -6.10647082e-01 -3.05984616e-01 -2.35803053e-02 -2.10775152e-01 -3.65769953e-01 -4.71620500e-01 -2.31339559e-01 -1.20158863e+00 9.20449376e-01 -1.67649257e+00 -1.07120109e+00 -8.82841885e-01 2.23238611e+00 5.42871952e-01 7.07291886e-02 2.66750872e-01 -1.30029336e-01 8.19259942e-01 5.96426010e-01 -2.76219189e-01 -1.68499470e-01 -5.73986828e-01 -3.05972509e-02 1.26929259e+00 4.21913862e-01 -1.51054275e+00 1.34943008e+00 6.71096087e+00 9.26590741e-01 -9.80881512e-01 -4.28600758e-02 8.93830538e-01 2.98651755e-01 -1.18907787e-01 -7.94005916e-02 -8.32548976e-01 4.06459272e-01 8.49619925e-01 4.87727463e-01 3.26084495e-01 9.46144998e-01 4.09392118e-01 -4.27522361e-01 -2.54724503e-01 9.32318687e-01 -2.73178905e-01 -1.70102453e+00 -1.68427378e-01 1.24647379e-01 1.13829696e+00 7.10998297e-01 3.86353023e-02 -4.84530292e-02 4.77110922e-01 -1.05187988e+00 1.01313472e+00 5.15662134e-01 8.90804708e-01 -6.69653893e-01 4.40048933e-01 8.83094370e-02 -1.71932197e+00 -6.99131936e-02 -1.94103599e-01 -3.46809886e-02 3.32393736e-01 6.01724982e-01 -4.33548212e-01 4.29345548e-01 1.23143363e+00 9.14682150e-01 -1.06481075e+00 8.33845675e-01 -1.87989131e-01 5.12380004e-01 -4.09914881e-01 6.09276116e-01 7.66281307e-01 -5.98192453e-01 2.03505412e-01 1.64208579e+00 -6.44594729e-02 2.89696865e-02 3.28452379e-01 5.86132705e-01 1.44623995e-01 -1.54709935e-01 -8.99781585e-01 2.49297231e-01 5.48125923e-01 1.63446093e+00 -1.23537016e+00 -6.45273805e-01 -3.19130480e-01 8.71013522e-01 1.12018123e-01 5.22120237e-01 -9.57309425e-01 -2.04072207e-01 8.63069177e-01 4.93337840e-01 4.28069174e-01 -4.89972651e-01 -5.25554001e-01 -9.45553005e-01 -1.74492508e-01 -5.18304586e-01 2.65534222e-01 -1.08715165e+00 -8.06559980e-01 4.33763444e-01 2.96779424e-01 -1.14078653e+00 1.45372123e-01 -6.41437352e-01 -4.97368574e-01 5.68193614e-01 -1.77614141e+00 -1.51503456e+00 -9.15184975e-01 6.88434243e-01 8.42385113e-01 1.86274156e-01 5.49402654e-01 3.98254156e-01 -6.62618399e-01 -1.85932666e-01 1.00770362e-01 2.71626323e-01 2.09651262e-01 -1.41278338e+00 7.02326715e-01 8.23192239e-01 5.15672378e-02 7.42190778e-02 4.48533684e-01 -3.57962042e-01 -5.96004426e-01 -1.79447353e+00 5.46056569e-01 -7.07026243e-01 7.35277593e-01 -4.25157577e-01 -8.33466053e-01 6.20338023e-01 8.54921117e-02 2.80279875e-01 1.00141667e-01 -1.70628116e-01 -3.74211937e-01 -1.79373443e-01 -1.22096026e+00 6.24859393e-01 1.24467051e+00 -6.23392582e-01 -3.78552794e-01 5.16358316e-01 4.66833502e-01 -5.94626606e-01 -9.60031450e-01 1.88521981e-01 2.74165660e-01 -1.18992960e+00 1.38717782e+00 3.01054846e-02 3.83378208e-01 -6.06659234e-01 -2.48362333e-01 -1.11329365e+00 -1.47284836e-01 -8.72262344e-02 3.88708085e-01 1.41190600e+00 2.51417756e-01 -6.84245288e-01 5.61383963e-01 3.61296415e-01 -4.71866727e-01 -2.28302047e-01 -6.58005476e-01 -5.87899864e-01 -2.23905057e-01 -6.63406551e-01 9.48806107e-01 8.75558555e-01 -7.77458251e-01 -3.93694341e-02 -1.60634920e-01 4.34820473e-01 5.92066824e-01 2.59341806e-01 9.89932656e-01 -1.41010690e+00 3.59536022e-01 -7.96115518e-01 -4.71549422e-01 -7.53183007e-01 1.46678105e-01 -3.96224976e-01 -1.96941912e-01 -1.87908995e+00 -1.37707189e-01 -7.73769915e-01 4.27056491e-01 6.08361363e-01 -1.37257262e-03 8.89219105e-01 -8.23877454e-02 2.46297210e-01 -6.93210304e-01 -1.68942548e-02 1.00425506e+00 -3.06766957e-01 2.01133080e-02 -2.12603703e-01 -4.16239530e-01 9.98227179e-01 9.48791087e-01 -6.23942725e-02 -3.39773387e-01 -4.55314249e-01 2.45565832e-01 -5.09318352e-01 7.59144187e-01 -1.05056691e+00 -2.36679032e-01 -4.67543602e-01 2.18207955e-01 -9.99611974e-01 2.53015786e-01 -9.94436622e-01 4.15598929e-01 4.73802052e-02 -7.36416578e-02 -2.89429486e-01 3.41198504e-01 4.39019084e-01 -2.74757117e-01 1.99179918e-01 8.19314480e-01 -5.43385148e-01 -1.56638420e+00 4.21122670e-01 -4.92128223e-01 3.76930505e-01 1.22531021e+00 -5.68958163e-01 -7.17843175e-01 3.85859758e-02 -5.63013554e-01 3.39357764e-01 9.08770621e-01 4.95981634e-01 1.85879484e-01 -9.90225673e-01 -6.49188280e-01 6.47419766e-02 6.42705798e-01 4.04916018e-01 5.82193971e-01 4.55204397e-01 -9.94325995e-01 5.18806159e-01 -4.33885992e-01 -8.06400836e-01 -1.09313059e+00 2.21199319e-01 3.02577436e-01 -1.49345383e-01 -8.91853571e-01 3.75540882e-01 2.91728824e-01 -6.68507397e-01 -3.15441608e-01 -6.58772409e-01 -6.24605000e-01 3.80195260e-01 5.05659699e-01 5.28870404e-01 1.04587920e-01 -1.12321591e+00 -3.37314874e-01 1.05791950e+00 4.87321079e-01 5.47620505e-02 1.05779994e+00 -6.49817109e-01 -2.47642361e-02 3.60929072e-01 9.05533373e-01 2.94438861e-02 -1.45487761e+00 -1.15984716e-02 -1.71200439e-01 -6.54057801e-01 7.62117505e-02 -6.85338140e-01 -1.19318783e+00 6.55124128e-01 6.16694152e-01 5.03442347e-01 1.07304966e+00 2.62795866e-01 4.36970115e-01 2.54313469e-01 6.58934176e-01 -1.32812476e+00 -4.58108127e-01 6.35249436e-01 6.95736110e-01 -1.67154658e+00 -4.64562774e-02 -5.94979286e-01 -8.40114713e-01 8.98046374e-01 3.93383712e-01 2.13980973e-02 4.18846518e-01 3.10432911e-02 1.87456757e-01 -3.67639124e-01 3.75875533e-02 -1.06588733e+00 2.81832606e-01 1.11398840e+00 -2.55933162e-02 5.01656890e-01 2.25776672e-01 -6.09809272e-02 -3.55032414e-01 -3.10655624e-01 6.13766015e-01 3.88948947e-01 -6.44016147e-01 -5.95357001e-01 -5.94587088e-01 4.13804114e-01 -1.00708060e-01 -1.61628127e-01 -2.58194178e-01 1.07626176e+00 4.69818085e-01 1.02130926e+00 4.82549489e-01 6.86004609e-02 3.95368397e-01 -7.24992931e-01 1.52827054e-01 -5.31849623e-01 -4.00384277e-01 -5.44611573e-01 5.46273828e-01 -7.92292893e-01 -6.41926885e-01 -6.73454940e-01 -1.25057805e+00 -3.03064913e-01 3.47727925e-01 -3.21701944e-01 8.38064015e-01 8.59741271e-01 3.00943077e-01 4.02717054e-01 5.33857822e-01 -1.35230970e+00 7.03755081e-01 -7.36996293e-01 -6.70299113e-01 3.80803317e-01 3.22414309e-01 -7.74543583e-01 -2.15325668e-01 3.52968544e-01]
[8.574701309204102, -1.5598591566085815]
5b3e4597-cab3-4ebb-80e2-28c8c899d4cf
an-inductive-formalization-of-self
1806.08925
null
http://arxiv.org/abs/1806.08925v1
http://arxiv.org/pdf/1806.08925v1.pdf
An Inductive Formalization of Self Reproduction in Dynamical Hierarchies
Formalizing self reproduction in dynamical hierarchies is one of the important problems in Artificial Life (AL) studies. We study, in this paper, an inductively defined algebraic framework for self reproduction on macroscopic organizational levels under dynamical system setting for simulated AL models and explore some existential results. Starting with defining self reproduction for atomic entities we define self reproduction with possible mutations on higher organizational levels in terms of hierarchical sets and the corresponding inductively defined `meta' - reactions. We introduce constraints to distinguish a collection of entities from genuine cases of emergent organizational structures.
['Janardan Misra']
2018-06-23
null
null
null
null
['artificial-life']
['miscellaneous']
[ 1.16443209e-01 6.93672121e-01 4.23499882e-01 1.81894898e-01 7.25168586e-01 -8.07647586e-01 9.81629252e-01 2.96455652e-01 -1.67072207e-01 1.04547763e+00 3.44126582e-01 1.23167671e-01 -4.55827564e-01 -1.10321081e+00 -4.88281935e-01 -9.66054380e-01 -6.21075094e-01 6.15506887e-01 3.76307160e-01 -8.62883151e-01 -2.79671755e-02 6.03082359e-01 -1.86564076e+00 -3.86234336e-02 1.03217924e+00 1.66599005e-01 -2.79874533e-01 9.35496569e-01 1.65302277e-01 1.13443804e+00 -7.45018542e-01 -6.30848557e-02 2.83837885e-01 -8.88462603e-01 -8.51976514e-01 4.38647121e-01 -2.32849300e-01 6.33245468e-01 -2.15890676e-01 8.80180061e-01 3.71052504e-01 1.12383952e-02 1.06519318e+00 -1.33061171e+00 -1.01712573e+00 9.48587060e-01 -1.06144555e-01 2.42311895e-01 3.75247180e-01 2.30220824e-01 7.68726647e-01 -4.04373050e-01 1.02920258e+00 1.20604813e+00 4.98688012e-01 6.33956790e-01 -1.92418718e+00 2.02810556e-01 -3.48719358e-01 -6.17632568e-01 -1.53559899e+00 -6.47861958e-01 2.56871253e-01 -7.87573695e-01 1.20117867e+00 4.60927963e-01 1.17678881e+00 5.95328033e-01 9.89510059e-01 3.62158231e-02 1.25157642e+00 -5.03262579e-01 5.88626623e-01 1.14413776e-01 5.17320395e-01 8.06177497e-01 1.04628158e+00 1.58480704e-01 -2.08691627e-01 -3.27975422e-01 1.15489030e+00 -5.19161582e-01 7.38521442e-02 -7.12041914e-01 -1.11130416e+00 3.23054284e-01 1.40798509e-01 8.13705027e-01 -4.02776867e-01 5.33895969e-01 1.59627050e-01 5.90336025e-01 7.79056400e-02 9.40574527e-01 -3.34131420e-01 2.80416816e-01 1.30645209e-03 1.45630792e-01 1.05683982e+00 9.72606957e-01 7.55455852e-01 2.25506201e-01 -3.92589122e-02 5.15785456e-01 -6.36003315e-02 1.77803114e-01 3.18082184e-01 -9.29482758e-01 -4.63519275e-01 9.20407474e-01 2.00652689e-01 -7.39021003e-01 -6.30686998e-01 -7.34914124e-01 -1.13801014e+00 -1.34541452e-01 5.01990803e-02 -5.01755513e-02 -5.61923862e-01 1.93183303e+00 3.22793394e-01 -1.72563806e-01 5.56649506e-01 1.36501715e-01 2.94694275e-01 6.12630785e-01 -1.14912257e-01 -9.33243632e-01 1.15385759e+00 -5.72617054e-01 -7.68886149e-01 5.65282404e-01 6.74664021e-01 1.09352551e-01 4.87612993e-01 -1.04002863e-01 -1.38272870e+00 -2.29604617e-01 -8.01718295e-01 2.58404106e-01 -5.35586417e-01 -4.56359804e-01 7.23332167e-01 4.87787098e-01 -1.75118160e+00 4.89281714e-01 -7.39643097e-01 -8.80627096e-01 -3.58490020e-01 6.45359039e-01 -2.62657195e-01 1.15079606e+00 -1.44787085e+00 7.41273820e-01 2.92489558e-01 -8.16338956e-02 -1.08327603e+00 -1.18294924e-01 -5.19258022e-01 1.84984282e-02 4.10673350e-01 -1.21826410e+00 8.46973956e-01 -8.78025174e-01 -1.32405794e+00 1.24893475e+00 2.35285088e-01 -5.91829300e-01 4.86836582e-01 5.57014763e-01 -3.46976250e-01 -2.19906628e-01 1.21273868e-01 4.96351004e-01 6.53712213e-01 -1.77409637e+00 -2.16841280e-01 -1.79982558e-01 4.76420075e-01 2.05695555e-01 -2.51172900e-01 -3.38716507e-01 3.77012104e-01 -3.62318039e-01 -9.84285995e-02 -1.17657149e+00 -6.02854848e-01 -8.95085037e-01 -5.13185620e-01 -4.15550262e-01 -9.19171721e-02 3.27614576e-01 1.51549697e+00 -1.85778356e+00 8.68201375e-01 1.43293694e-01 7.98816264e-01 -4.46098775e-01 2.11511895e-01 1.02570689e+00 3.06645390e-02 5.13741910e-01 -3.42368722e-01 2.20545202e-01 2.10291013e-01 2.51607507e-01 -9.83598754e-02 3.01825285e-01 1.63935065e-01 9.20025170e-01 -7.41171658e-01 -6.32030606e-01 -1.43637672e-01 6.36851117e-02 -5.61139941e-01 3.10188101e-04 -1.78922549e-01 4.19423282e-01 -3.27392638e-01 4.33169931e-01 1.35053411e-01 -3.36675793e-01 4.08111781e-01 4.35916185e-01 -7.14436412e-01 -2.39316687e-01 -1.07128310e+00 7.23456502e-01 -1.66255206e-01 1.24147777e-02 -1.73209107e-03 -4.99881417e-01 6.75800025e-01 2.35571995e-01 6.03272200e-01 -6.65723234e-02 1.82309717e-01 1.32072875e-02 7.39286959e-01 -3.37135851e-01 5.55550694e-01 -2.57198036e-01 -7.08083391e-01 6.57917559e-01 2.59246081e-01 -3.53915960e-01 4.75196004e-01 2.62602180e-01 1.54369867e+00 -3.33017856e-01 8.67622375e-01 -1.29285967e+00 9.05583978e-01 1.04181573e-01 3.96695584e-01 1.26479793e+00 -3.15234989e-01 -2.05369845e-01 7.65316069e-01 -4.78791565e-01 -1.34332061e+00 -1.20935011e+00 -4.43432629e-01 9.55961347e-01 4.00144130e-01 -5.29959917e-01 -1.12654483e+00 1.85807794e-01 1.75369456e-02 2.83050418e-01 -1.11475646e+00 -2.43935719e-01 -6.51627004e-01 -8.83808553e-01 5.56617439e-01 -4.22592387e-02 4.28459436e-01 -1.33948410e+00 -9.76019800e-01 3.68673474e-01 3.64175797e-01 -6.23477519e-01 2.88200956e-02 5.77421069e-01 -7.94255495e-01 -6.09575927e-01 -2.18362883e-01 -6.13025188e-01 8.33839715e-01 -7.95101747e-02 1.13721812e+00 4.36869979e-01 -6.82535827e-01 4.43778396e-01 -3.71967331e-02 -1.66501150e-01 -1.10366035e+00 3.18267047e-01 6.21450126e-01 -1.91245496e-01 -3.10370088e-01 -8.00186574e-01 -1.72463790e-01 2.09902629e-01 -1.32648361e+00 1.88510984e-01 1.33685365e-01 6.53182268e-01 3.19035411e-01 4.67831582e-01 5.57866514e-01 -7.28363097e-01 1.13260508e+00 -3.81891221e-01 -4.44127053e-01 5.56633413e-01 -5.42053163e-01 5.26963353e-01 8.15204382e-01 -1.50441766e-01 -7.83933282e-01 -1.20092906e-01 6.15963995e-01 5.65009832e-01 -9.20425579e-02 3.13743323e-01 -1.21355198e-01 6.83382079e-02 7.98133790e-01 4.80096400e-01 -1.35428295e-01 1.99990064e-01 1.89902380e-01 4.49897647e-01 2.65317738e-01 -9.92320240e-01 6.80088699e-01 5.19617200e-01 5.93172550e-01 -1.12479758e+00 -9.64443386e-02 3.50703418e-01 -1.02291846e+00 -4.15862888e-01 7.86932409e-01 -4.38330889e-01 -8.52002203e-01 5.66635013e-01 -6.96882188e-01 -3.33191276e-01 -9.98758793e-01 -3.06097955e-01 -1.11284614e+00 1.39966914e-02 -7.09192216e-01 -1.19825387e+00 1.75763667e-02 -7.78293490e-01 7.24677563e-01 8.43479037e-02 -3.38055849e-01 -1.09300244e+00 8.16060603e-01 -3.16191792e-01 4.73253965e-01 4.69938606e-01 1.10074663e+00 -3.39373142e-01 -6.68390930e-01 2.16580778e-01 4.63329881e-01 -4.43410903e-01 2.71283925e-01 4.53486562e-01 -3.08673412e-01 -2.47337267e-01 1.62013158e-01 1.58198714e-01 6.17097557e-01 1.89323068e-01 1.61326110e-01 -4.60062295e-01 -5.07948041e-01 5.90242296e-02 1.55784583e+00 5.33693194e-01 5.17911971e-01 4.02293533e-01 2.41373628e-01 9.39102113e-01 -4.74615656e-02 4.89886731e-01 2.26439118e-01 4.09958839e-01 -2.55238488e-02 3.09922189e-01 4.24864262e-01 9.31699723e-02 2.22715259e-01 1.10497618e+00 -7.07887530e-01 -4.51188773e-01 -1.07271683e+00 4.74179298e-01 -2.11741304e+00 -1.36102462e+00 -2.90126711e-01 1.95310545e+00 9.40283895e-01 -6.50277408e-03 4.84036863e-01 -6.10426180e-02 1.23173511e+00 -1.45706192e-01 -4.14556563e-01 -5.18044114e-01 -5.68238974e-01 -1.15332164e-01 2.50477940e-01 7.96620190e-01 -7.38360226e-01 1.06532228e+00 7.74550486e+00 2.29347482e-01 -4.46274251e-01 1.46008223e-01 3.13618481e-01 1.29964381e-01 -2.80491292e-01 3.63841414e-01 -6.77724063e-01 2.53444225e-01 9.86548960e-01 -8.17803860e-01 6.36202753e-01 2.80416697e-01 1.43307686e-01 -1.03619590e-01 -9.65407610e-01 1.91762924e-01 -1.29005790e-01 -1.18827081e+00 4.17008787e-01 5.90359151e-01 1.15128708e+00 -6.36243939e-01 -2.07205042e-01 9.27057117e-02 1.05532825e+00 -1.01505041e+00 7.55784452e-01 8.53904068e-01 4.62284327e-01 -6.27624810e-01 3.74476552e-01 3.42655391e-01 -1.15570974e+00 -2.09946916e-01 -3.26364398e-01 -5.27590513e-01 2.18762755e-01 1.94912568e-01 -1.58513740e-01 3.82364124e-01 2.63647050e-01 3.97877008e-01 -6.37806416e-01 5.40859163e-01 1.76604301e-01 1.75278872e-01 -2.70261645e-01 -4.00575429e-01 -1.37275476e-02 -4.65879112e-01 7.27669060e-01 1.00095892e+00 4.46058698e-02 6.27593696e-01 -2.81598717e-01 8.67660105e-01 9.68563259e-02 -1.87100135e-02 -1.15319109e+00 -7.97912106e-02 2.94518113e-01 1.17997921e+00 -1.35856032e+00 -5.40593266e-01 4.52067405e-01 6.43227875e-01 2.70601064e-01 -2.18733586e-02 -8.72616649e-01 -2.60827780e-01 6.38788164e-01 4.22993809e-01 -3.01494777e-01 -5.73395491e-01 -1.61373839e-02 -1.46473241e+00 -6.90881073e-01 -5.18709123e-01 5.92813045e-02 -5.20701170e-01 -1.06774569e+00 7.65948534e-01 3.54564071e-01 -7.58475244e-01 -1.49315923e-01 -1.88465461e-01 -3.43510598e-01 8.48512501e-02 -3.52589190e-01 -7.41900146e-01 -1.36654675e-01 4.45324332e-01 1.18784584e-01 -1.24640696e-01 6.92516327e-01 -5.44976115e-01 -7.49477923e-01 1.65335059e-01 4.42125916e-01 -3.57787162e-01 -1.36635914e-01 -1.58068085e+00 3.54937822e-01 6.16509020e-01 -2.24127635e-01 1.28195131e+00 1.14859164e+00 -9.54073071e-01 -1.37654257e+00 -7.51771986e-01 8.85064304e-01 -6.45400286e-01 9.45865273e-01 -7.89202631e-01 -5.62632263e-01 6.86822832e-01 6.78725362e-01 -5.12932837e-01 4.78535205e-01 -1.52140364e-01 2.08816454e-01 1.94524571e-01 -1.06939638e+00 1.14121139e+00 2.01984668e+00 -2.64625520e-01 -6.53964341e-01 3.22066694e-01 8.28222632e-01 1.99911967e-01 -1.10511672e+00 6.85922861e-01 5.21711051e-01 -1.06907320e+00 8.02346706e-01 -5.64437330e-01 3.81325960e-01 -5.32893240e-01 1.24845788e-01 -1.10183096e+00 -8.67073476e-01 -9.65006351e-01 1.48347586e-01 1.14015687e+00 1.50903836e-01 -1.09658790e+00 3.80995572e-02 5.11193216e-01 1.16182752e-01 -3.41081262e-01 -7.81311512e-01 -1.07438564e+00 1.90832779e-01 5.83179533e-01 6.90478802e-01 1.13070166e+00 6.31060719e-01 7.09806859e-01 -1.04540832e-01 -2.49151722e-01 6.74800634e-01 -1.74976513e-02 5.99723697e-01 -1.39208734e+00 -4.87980962e-01 -6.16800368e-01 -8.24245870e-01 -2.23524764e-01 1.81943014e-01 -8.95623326e-01 -8.43999758e-02 -1.41040421e+00 4.95826155e-01 -3.39768529e-01 -2.10089087e-01 4.46928814e-02 4.18971092e-01 2.17161290e-02 -3.90069149e-02 6.23500824e-01 -8.68759871e-01 5.80965400e-01 9.86125171e-01 3.03038925e-01 -4.68622059e-01 -5.91556609e-01 -5.41490078e-01 4.14982826e-01 7.82521546e-01 -4.46009517e-01 -4.93890941e-01 1.94943026e-01 7.91109502e-01 3.62802565e-01 1.45644903e-01 -1.10516298e+00 3.11146557e-01 -5.61990619e-01 -1.44763425e-01 -5.48784658e-02 -1.58161417e-01 -5.91816962e-01 1.14044023e+00 1.08844209e+00 -7.20679522e-01 4.45683509e-01 -1.89268470e-01 4.65303153e-01 2.58780509e-01 -2.95066953e-01 7.21731484e-01 -4.46335465e-01 1.50545731e-01 -2.79211998e-01 -1.31832635e+00 -1.13931425e-01 1.56328428e+00 -5.69724202e-01 -4.93570894e-01 -4.36441451e-02 -1.26799154e+00 1.04470551e-01 1.25502241e+00 -6.29508123e-02 -5.57354614e-02 -1.05316007e+00 -4.45359349e-01 6.19557500e-03 6.62510544e-02 -4.15229797e-01 -2.66151037e-02 7.51075745e-01 -8.97194624e-01 5.92766166e-01 -6.20846331e-01 -3.57807577e-01 -9.41932559e-01 8.15565169e-01 6.41168296e-01 -3.75301361e-01 -2.11982086e-01 4.47590053e-01 8.17049384e-01 -3.74038696e-01 -5.55389404e-01 -1.41720682e-01 -3.23894083e-01 -4.81402241e-02 -2.01619685e-01 4.87945497e-01 -5.48619509e-01 -8.95233572e-01 -2.49560088e-01 5.29038608e-01 1.83067933e-01 -4.43730563e-01 1.24247468e+00 -4.15982604e-01 -1.04164207e+00 1.04821670e+00 2.79278219e-01 1.75700411e-01 -5.59401214e-01 2.56495655e-01 2.02540755e-01 1.64333954e-01 -8.77461314e-01 -4.22132075e-01 -1.89810917e-01 1.62919432e-01 1.58504751e-02 1.33599854e+00 7.97815740e-01 3.68647784e-01 -1.17156409e-01 6.30617976e-01 9.52997744e-01 -1.06784463e+00 3.68080586e-02 6.66909635e-01 1.12875354e+00 -3.04221958e-01 -1.42744377e-01 -5.65263987e-01 -3.12803417e-01 9.64138746e-01 7.57436812e-01 -6.32557929e-01 6.83604658e-01 7.17087507e-01 -6.29532993e-01 -4.64419156e-01 -1.45304763e+00 -5.22567868e-01 -4.58760291e-01 3.86597455e-01 5.58546066e-01 2.57138014e-01 -1.07190120e+00 1.72749236e-01 -4.01318789e-01 -2.55941421e-01 1.20547688e+00 1.10794950e+00 -7.32702017e-01 -8.97917449e-01 -2.48468086e-01 4.25052851e-01 -2.77106315e-01 7.51316398e-02 -1.17085361e+00 8.86010647e-01 3.93932313e-01 7.16683567e-01 2.13302538e-01 -3.96558821e-01 1.76855579e-01 -8.45152065e-02 7.51259685e-01 -8.16513062e-01 -1.05794990e+00 -1.65869415e-01 5.78387566e-02 2.06418961e-01 -6.00594997e-01 -9.99239624e-01 -1.33548987e+00 -4.87270176e-01 -5.06926775e-01 5.78739107e-01 1.40360678e-02 4.59596306e-01 5.32946467e-01 6.84317887e-01 6.25565827e-01 -4.34391022e-01 -2.89648026e-01 -6.81083918e-01 -9.44004059e-01 2.61054009e-01 8.54369253e-02 -5.94209313e-01 -5.57840705e-01 2.12705493e-01]
[5.578807830810547, 4.15610933303833]
d46b62e5-77fe-43f4-b074-d18b7e5c987f
investigating-generative-adversarial-networks
1803.10132
null
http://arxiv.org/abs/1803.10132v3
http://arxiv.org/pdf/1803.10132v3.pdf
Investigating Generative Adversarial Networks based Speech Dereverberation for Robust Speech Recognition
We investigate the use of generative adversarial networks (GANs) in speech dereverberation for robust speech recognition. GANs have been recently studied for speech enhancement to remove additive noises, but there still lacks of a work to examine their ability in speech dereverberation and the advantages of using GANs have not been fully established. In this paper, we provide deep investigations in the use of GAN-based dereverberation front-end in ASR. First, we study the effectiveness of different dereverberation networks (the generator in GAN) and find that LSTM leads a significant improvement as compared with feed-forward DNN and CNN in our dataset. Second, further adding residual connections in the deep LSTMs can boost the performance as well. Finally, we find that, for the success of GAN, it is important to update the generator and the discriminator using the same mini-batch data during training. Moreover, using reverberant spectrogram as a condition to discriminator, as suggested in previous studies, may degrade the performance. In summary, our GAN-based dereverberation front-end achieves 14%-19% relative CER reduction as compared to the baseline DNN dereverberation network when tested on a strong multi-condition training acoustic model.
['Ke Wang', 'Fei Xiang', 'Sining Sun', 'Yujun Wang', 'Junbo Zhang', 'Lei Xie']
2018-03-27
null
null
null
null
['robust-speech-recognition', 'speech-dereverberation']
['speech', 'speech']
[ 1.02270819e-01 6.51721656e-02 6.40699506e-01 -4.62081134e-02 -9.87108111e-01 -5.04592717e-01 4.29085106e-01 -6.24614537e-01 -2.82319397e-01 7.36171961e-01 5.34532428e-01 -5.92552841e-01 3.87679577e-01 -4.70520765e-01 -7.54624844e-01 -1.01316655e+00 4.02963549e-01 -3.66156809e-02 -1.47750124e-01 -4.92048591e-01 -4.87328947e-01 6.13921404e-01 -1.13713968e+00 9.73170847e-02 8.83205950e-01 7.61520445e-01 3.12377036e-01 1.16218531e+00 2.84353018e-01 7.23864377e-01 -1.60828221e+00 -3.61258954e-01 3.08320224e-01 -9.92927134e-01 -2.50662714e-01 -2.88549721e-01 4.23195660e-01 -4.82796431e-01 -6.67685807e-01 8.24918926e-01 1.28821218e+00 2.71707207e-01 5.71011901e-01 -7.51289606e-01 -7.10135996e-01 7.66982019e-01 -5.41948453e-02 4.27594274e-01 9.29395407e-02 4.18504119e-01 6.75786912e-01 -7.13170588e-01 -1.19448286e-02 1.28510606e+00 6.72598124e-01 8.68551970e-01 -1.03446496e+00 -1.00958693e+00 -2.93664098e-01 2.19113424e-01 -1.00376797e+00 -9.78051722e-01 8.76381755e-01 6.23247474e-02 1.13100672e+00 4.87855852e-01 3.47259521e-01 1.60032701e+00 -8.45853239e-02 4.63083178e-01 1.10769761e+00 -5.58511496e-01 5.11250645e-03 2.24151611e-02 -2.75156230e-01 5.59688099e-02 -9.95192155e-02 7.03731656e-01 -4.19912606e-01 3.03594202e-01 4.86549228e-01 -5.51580846e-01 -4.01347369e-01 7.96031296e-01 -6.77517533e-01 4.84879941e-01 5.12539685e-01 5.37917733e-01 -4.56512392e-01 5.24900794e-01 3.62769634e-01 6.51460469e-01 7.65696764e-01 7.16803610e-01 -2.84193665e-01 -3.00231397e-01 -1.17458069e+00 -1.50450198e-02 6.13376141e-01 5.12444198e-01 1.32670745e-01 1.10536432e+00 -4.54382211e-01 1.16722810e+00 1.56963736e-01 7.82620668e-01 6.40189648e-01 -6.34016871e-01 4.72152948e-01 -3.34658682e-01 -2.52061397e-01 -5.42043269e-01 -1.12160578e-01 -1.02136552e+00 -1.02290070e+00 3.67357194e-01 1.65832743e-01 -6.85204387e-01 -1.32817078e+00 1.74706006e+00 -2.92591713e-02 4.95999247e-01 3.73696148e-01 7.91707575e-01 7.74652600e-01 8.43563914e-01 -5.44940531e-02 -1.04253829e-01 9.97322023e-01 -1.06197608e+00 -1.24438405e+00 -2.15758115e-01 2.09290087e-01 -1.20511317e+00 9.21853304e-01 4.73895818e-01 -1.16459262e+00 -9.35923398e-01 -1.23364580e+00 1.20183259e-01 -3.59669238e-01 2.21546441e-01 9.57934782e-02 1.32025397e+00 -1.26287770e+00 5.33201039e-01 -8.25581014e-01 -2.66922675e-02 -1.06756259e-02 1.56615451e-01 -1.04867943e-01 1.91983178e-01 -1.56315100e+00 1.04668605e+00 8.67483020e-03 3.07441413e-01 -1.26685178e+00 -7.10581243e-01 -6.94120228e-01 1.09841563e-01 -2.56347302e-02 -5.76996565e-01 1.30824161e+00 -1.00586164e+00 -2.07279086e+00 2.30224237e-01 -1.52830124e-01 -9.18130755e-01 5.20623386e-01 -3.93136472e-01 -8.70016456e-01 -1.62353054e-01 -5.32352328e-01 4.69615996e-01 1.09227383e+00 -1.13897789e+00 -1.98256031e-01 1.31217716e-02 -1.37823880e-01 2.51844168e-01 -1.75855473e-01 2.01384306e-01 7.86683336e-02 -1.21743298e+00 -1.71132162e-01 -8.40661466e-01 -1.19645961e-01 -7.46342182e-01 -4.21081483e-01 1.13283336e-01 1.05714655e+00 -1.40952754e+00 1.18030906e+00 -2.33071375e+00 -2.01504230e-01 -3.40338945e-02 -2.58255631e-01 1.02889168e+00 -2.53236264e-01 4.56540942e-01 -3.88639271e-01 3.15526843e-01 -7.11772367e-02 -8.09960663e-01 -1.70902703e-02 1.42435551e-01 -4.99435723e-01 2.78165579e-01 3.37705821e-01 8.33693564e-01 -4.93450165e-01 2.15337440e-01 5.18710315e-01 1.04557049e+00 -4.38790560e-01 5.81560373e-01 1.41258940e-01 6.59902692e-01 3.26973766e-01 2.21682206e-01 7.12937474e-01 5.84000409e-01 -3.05109054e-01 -1.70155913e-01 1.59173936e-01 8.25450122e-01 -8.31150413e-01 1.30433011e+00 -8.26868236e-01 1.00321209e+00 2.34581590e-01 -9.23769891e-01 9.75443661e-01 8.33262682e-01 -8.11558738e-02 -8.57317924e-01 6.95435852e-02 4.75105286e-01 4.72635865e-01 -1.29244059e-01 4.20070916e-01 -4.42887038e-01 4.07339722e-01 6.64332286e-02 3.28109354e-01 -4.68621165e-01 -3.35342944e-01 -1.97768822e-01 1.07587922e+00 -1.63394108e-01 -9.26480442e-02 1.20150797e-01 4.58615899e-01 -7.72388220e-01 3.04128081e-01 7.61798322e-01 -4.12812177e-03 1.05128074e+00 1.06546514e-01 2.55508244e-01 -1.09735668e+00 -9.27935481e-01 1.00266255e-01 8.58154356e-01 -5.69662213e-01 -5.36370575e-02 -1.24360812e+00 -3.66407663e-01 -4.70431238e-01 1.20964408e+00 -2.80238271e-01 -3.59921485e-01 -8.09397638e-01 -6.97879791e-01 1.20313430e+00 5.46599090e-01 6.86163723e-01 -1.05102396e+00 2.57001612e-02 3.38430375e-01 -4.73480597e-02 -1.08764684e+00 -6.61850989e-01 5.52015543e-01 -4.97575313e-01 -3.27908665e-01 -1.03706026e+00 -5.66882133e-01 2.86434680e-01 1.19172014e-01 7.48584569e-01 -9.73919183e-02 2.82245487e-01 1.85934290e-01 -6.14576459e-01 -6.82163835e-01 -1.14511979e+00 1.64063022e-01 8.65863040e-02 -1.42802045e-01 -8.30691680e-02 -9.04256761e-01 -5.90967119e-01 3.94795060e-01 -7.55312979e-01 -2.91482657e-01 5.25080681e-01 7.91184008e-01 1.13252342e-01 5.62031083e-02 1.01093686e+00 -5.88381052e-01 9.80261385e-01 -2.47325763e-01 -3.45358998e-01 -8.92436281e-02 -5.28035104e-01 -5.96507974e-02 9.27363575e-01 -4.70762581e-01 -1.25080550e+00 -5.69580674e-01 -1.09075487e+00 -5.13104081e-01 -2.76860803e-01 1.71463877e-01 -4.24446821e-01 1.05182461e-01 8.21576238e-01 1.01086855e-01 -7.15441406e-02 -6.04831219e-01 3.42446804e-01 1.00222051e+00 5.58333814e-01 -7.44835362e-02 9.31543827e-01 -3.38520892e-02 -2.64385611e-01 -1.03895760e+00 -4.46205139e-01 -1.51620403e-01 -4.04006913e-02 1.94587447e-02 8.08803201e-01 -9.68903124e-01 -3.15782666e-01 8.37504447e-01 -1.30097628e+00 -6.58027768e-01 -4.35552418e-01 5.73205888e-01 -3.31630856e-01 1.40463993e-01 -7.59036005e-01 -1.01782489e+00 -7.52424777e-01 -1.20279610e+00 7.76237488e-01 2.30344817e-01 -7.73971993e-03 -8.63499522e-01 1.81683933e-03 5.20832598e-01 1.13749588e+00 9.73742083e-03 4.83837694e-01 -7.14804411e-01 -2.35099092e-01 -1.69891536e-01 3.31710309e-01 1.37442791e+00 4.32205081e-01 -1.56831637e-01 -1.66293955e+00 -4.38765228e-01 4.59585816e-01 1.24280199e-01 9.20253456e-01 6.19819641e-01 8.98889601e-01 -5.01995623e-01 2.71218479e-01 6.48520291e-01 7.48686492e-01 8.13333929e-01 1.27433658e+00 -8.76859203e-02 6.52189255e-01 1.65367797e-01 7.45541528e-02 -1.16762735e-01 -3.53788316e-01 6.44738078e-01 2.23080426e-01 -5.96160889e-01 -1.15923953e+00 -2.20845640e-01 8.81395221e-01 1.21442187e+00 -9.97767150e-02 -8.11380923e-01 -4.02093053e-01 3.12041193e-01 -9.38548267e-01 -9.89590347e-01 6.92105889e-02 2.20947027e+00 8.91281068e-01 7.84832761e-02 -3.30969412e-03 5.02896845e-01 7.15722620e-01 4.15967405e-01 -3.55962306e-01 -7.28880703e-01 -4.37279612e-01 8.37531686e-01 5.61551452e-01 7.48040140e-01 -7.24564791e-01 9.14836228e-01 6.24491262e+00 1.02819502e+00 -1.54259801e+00 5.76066375e-01 8.21730494e-01 -1.18649006e-01 -2.28053093e-01 -3.78257215e-01 -5.33210576e-01 4.69214678e-01 1.42801583e+00 2.30977222e-01 6.89204872e-01 5.25180161e-01 5.26981890e-01 2.90749282e-01 -7.52315819e-01 7.43756533e-01 2.04794362e-01 -6.45514548e-01 -3.11574250e-01 -2.47108983e-03 6.99354768e-01 9.42849740e-02 4.46982265e-01 4.08024758e-01 2.45662421e-01 -1.37554622e+00 7.91769505e-01 2.81461358e-01 8.81572306e-01 -8.78446639e-01 8.56474757e-01 -1.28141185e-02 -7.59080768e-01 1.72142118e-01 -2.11944580e-01 1.19006798e-01 2.90127486e-01 8.21586907e-01 -1.35179663e+00 5.66103399e-01 3.33033681e-01 -1.36683648e-02 -2.99946934e-01 8.59236360e-01 -6.83429122e-01 1.46059084e+00 -2.11058214e-01 1.94986522e-01 1.54132158e-01 7.72492066e-02 8.34865928e-01 1.31806004e+00 5.05730689e-01 -2.18255356e-01 -6.35563374e-01 7.86239803e-01 -2.11124450e-01 -2.75210202e-01 -3.83517295e-01 -1.50456280e-01 3.81757706e-01 9.07415450e-01 -1.46373644e-01 -1.13006160e-01 -1.04778558e-01 1.08970392e+00 -1.26076058e-01 7.87195623e-01 -9.68359947e-01 -4.45198894e-01 8.52641761e-01 1.25179127e-01 3.30069423e-01 -3.29956532e-01 -1.42130807e-01 -6.32859111e-01 -1.78236514e-02 -1.24410272e+00 -1.54891208e-01 -9.18354809e-01 -1.11072707e+00 1.05400217e+00 -3.94421935e-01 -8.10696900e-01 -4.06322986e-01 -4.59159851e-01 -8.22790623e-01 1.35902786e+00 -1.41950452e+00 -9.78279114e-01 -1.17094502e-01 3.48891944e-01 7.12707281e-01 -1.91464216e-01 7.74616420e-01 5.87444007e-01 -5.63947797e-01 1.00448382e+00 1.17311411e-01 7.24905506e-02 7.99807250e-01 -1.21994150e+00 9.10055041e-01 1.48894548e+00 2.86579370e-01 4.73054230e-01 9.88843679e-01 -4.91854191e-01 -1.00722790e+00 -1.14007008e+00 4.98432070e-01 -1.89535394e-01 1.97453395e-01 -5.66125274e-01 -9.73881721e-01 5.01756430e-01 8.97459865e-01 -1.99655890e-01 4.06906813e-01 -1.45812541e-01 -1.19945027e-01 -4.26020682e-01 -1.10394871e+00 5.56945026e-01 7.66222477e-01 -7.59829342e-01 -4.57983851e-01 -5.57599701e-02 1.15606987e+00 -6.01207197e-01 -4.29750204e-01 3.35756242e-01 1.46967486e-01 -9.08789456e-01 9.05373931e-01 -1.92586020e-01 1.71060741e-01 -2.67887175e-01 -1.85157493e-01 -2.14746952e+00 3.41832042e-02 -1.08812761e+00 -3.41326706e-02 1.75533664e+00 6.86630309e-01 -9.72654283e-01 4.43658203e-01 2.07963943e-01 -7.00255871e-01 -3.00029784e-01 -1.04631853e+00 -1.00257444e+00 2.12670296e-01 -6.59746051e-01 6.50597870e-01 4.85163748e-01 -6.19420707e-01 2.77427882e-01 -7.04838812e-01 3.04868132e-01 8.48675594e-02 -7.90681899e-01 7.22364664e-01 -3.78359258e-01 -4.43939745e-01 -3.97165239e-01 -1.46178558e-01 -9.38667715e-01 1.27132565e-01 -6.26869202e-01 2.67300218e-01 -1.53960717e+00 -6.53147757e-01 -3.82631719e-01 -4.18490857e-01 2.63459921e-01 -4.52023178e-01 3.07231266e-02 2.77753115e-01 -4.03797656e-01 3.30375403e-01 7.15647876e-01 1.12831628e+00 -1.21992685e-01 -2.62347341e-01 3.04764450e-01 -7.00967431e-01 3.34643513e-01 9.31621790e-01 -5.24029911e-01 -3.55546743e-01 -4.17700946e-01 -3.26812267e-01 8.37255921e-03 2.46936813e-01 -1.33256853e+00 9.52980742e-02 4.13431257e-01 1.68502688e-01 -1.93724871e-01 6.33011878e-01 -7.11495757e-01 4.50730056e-01 3.19771320e-01 -2.29431108e-01 -1.40068559e-02 4.92422700e-01 2.35935599e-01 -3.87642980e-01 -1.73334345e-01 8.93710196e-01 5.77313304e-02 -3.69257778e-02 -1.62349895e-01 -4.82387125e-01 -1.10191591e-02 3.97447348e-01 2.07184758e-02 -2.86801249e-01 -7.35132575e-01 -4.20014530e-01 -5.25718391e-01 -5.41986711e-03 3.35024655e-01 4.63666618e-01 -1.14572752e+00 -9.09722507e-01 2.23326012e-01 -7.05944598e-01 -1.82560399e-01 4.01681066e-01 7.22989202e-01 -3.22491825e-01 3.42973888e-01 7.42273927e-02 -1.58005029e-01 -1.15997970e+00 4.39096153e-01 7.92125404e-01 -1.16505444e-01 -3.13791484e-01 1.02962089e+00 1.21035859e-01 -1.32803515e-01 4.23335671e-01 -4.06379312e-01 3.29958908e-02 -1.10467084e-01 3.95731509e-01 5.28587639e-01 6.96040452e-01 -4.36067432e-01 -1.06183082e-01 7.02474313e-03 1.94160149e-01 -4.45925713e-01 1.20918143e+00 -1.56881753e-02 1.94351465e-01 2.65030891e-01 1.14068973e+00 4.47481394e-01 -1.02063775e+00 1.33385345e-01 -5.06215155e-01 -1.82373703e-01 3.50392491e-01 -1.07440102e+00 -1.49428010e+00 1.00603020e+00 7.71448195e-01 4.14740086e-01 1.51097620e+00 -2.92307049e-01 8.37428808e-01 -6.92141578e-02 -1.15637407e-01 -8.92292798e-01 5.09545356e-02 5.86326957e-01 1.28940916e+00 -8.74525189e-01 -5.79975903e-01 -6.91158026e-02 -6.01627588e-01 7.56917953e-01 4.31313813e-01 -1.46889597e-01 4.48753029e-01 7.00048506e-01 3.71112734e-01 2.82449633e-01 -5.15439451e-01 -2.06299379e-01 3.12700212e-01 7.11359024e-01 5.54176271e-01 2.30869591e-01 -1.56700924e-01 4.14696813e-01 -8.04642975e-01 -6.17598712e-01 4.03360039e-01 3.13961059e-01 -2.15154923e-02 -1.21121311e+00 -6.03151262e-01 2.45175317e-01 -8.48617613e-01 -5.69527030e-01 -3.09859931e-01 3.98313940e-01 2.01453585e-02 1.41437161e+00 -8.99397731e-02 -5.42802513e-01 6.36881232e-01 1.81590378e-01 2.49023318e-01 -3.91772419e-01 -1.10824311e+00 2.84014732e-01 3.45420152e-01 -1.36952609e-01 -2.18605757e-01 -3.42635363e-01 -8.42119634e-01 -2.69634217e-01 -7.54848897e-01 4.99924421e-02 1.00527608e+00 9.61760342e-01 2.66802967e-01 1.20328951e+00 7.82056630e-01 -6.08150780e-01 -5.77787340e-01 -1.65606332e+00 -4.61363077e-01 2.13750392e-01 6.68756545e-01 -2.31238648e-01 -8.35921168e-01 -1.31794894e-02]
[15.067460060119629, 6.044388771057129]
a218b965-8230-4cc5-97df-6bb2a8ff869c
probabilistic-polargmm-unsupervised-cluster
2206.12959
null
https://arxiv.org/abs/2206.12959v1
https://arxiv.org/pdf/2206.12959v1.pdf
Probabilistic PolarGMM: Unsupervised Cluster Learning of Very Noisy Projection Images of Unknown Pose
A crucial step in single particle analysis (SPA) of cryogenic electron microscopy (Cryo-EM), 2D classification and alignment takes a collection of noisy particle images to infer orientations and group similar images together. Averaging these aligned and clustered noisy images produces a set of clean images, ready for further analysis such as 3D reconstruction. Fourier-Bessel steerable principal component analysis (FBsPCA) enables an efficient, adaptable, low-rank rotation operator. We extend the FBsPCA to additionally handle translations. In this extended FBsPCA representation, we use a probabilistic polar-coordinate Gaussian mixture model to learn soft clusters in an unsupervised fashion using an expectation maximization (EM) algorithm. The obtained rotational clusters are thus additionally robust to the presence of pairwise alignment imperfections. Multiple benchmarks from simulated Cryo-EM datasets show probabilistic PolarGMM's improved performance in comparisons with standard single-particle Cryo-EM tools, EMAN2 and RELION, in terms of various clustering metrics and alignment errors.
['Chandrajit L. Bajaj', 'Supawit Chockchowwat']
2022-06-26
null
null
null
null
['cryogenic-electron-microscopy-cryo-em']
['computer-vision']
[ 2.23451421e-01 -4.66346771e-01 5.12266278e-01 -3.26260567e-01 -9.37110245e-01 -7.02918112e-01 8.13003242e-01 1.14341609e-01 -7.92845607e-01 8.32455158e-01 -1.53819963e-01 -1.90857694e-01 -3.82964641e-01 -1.50472865e-01 -6.31588995e-01 -1.41763651e+00 -1.04577079e-01 1.28141642e+00 1.29639283e-01 2.77895957e-01 5.17903507e-01 9.61966991e-01 -1.20822179e+00 1.09396351e-03 3.96032065e-01 4.09369916e-01 8.17466319e-01 7.58661807e-01 3.79564047e-01 3.26087773e-01 -3.60518545e-01 1.99826971e-01 -6.18078783e-02 -1.19784489e-01 -9.34643328e-01 2.56662935e-01 -4.16376479e-02 3.46092075e-01 2.53196150e-01 1.07678294e+00 5.39748311e-01 3.33283693e-01 6.63443863e-01 -6.63136899e-01 -1.75180122e-01 -3.67450230e-02 -3.96470249e-01 2.08079726e-01 3.79485786e-01 1.96322083e-01 6.27272367e-01 -1.10829115e+00 1.29581797e+00 1.18395960e+00 4.14943099e-01 1.41749099e-01 -1.89029431e+00 -4.58020307e-02 -2.69020408e-01 3.61493200e-01 -1.07789648e+00 -4.19903338e-01 6.80199742e-01 -4.74031895e-01 1.17665422e+00 4.14112061e-01 3.17610890e-01 9.43485558e-01 3.47678751e-01 1.45057008e-01 1.69175458e+00 -4.26788062e-01 5.81990719e-01 -1.95393622e-01 2.64969319e-01 2.35460460e-01 7.62648806e-02 -2.72010088e-01 -2.32213080e-01 -5.93867660e-01 3.38429540e-01 1.57869086e-01 -2.81492293e-01 -7.09914923e-01 -1.50823486e+00 4.41681921e-01 4.95259576e-02 5.13212085e-02 -6.62786663e-01 -3.52737427e-01 1.55071244e-01 -7.89144114e-02 3.07525545e-01 5.92339873e-01 -4.00872201e-01 -2.45695233e-01 -1.01998472e+00 2.75928855e-01 4.22027260e-01 4.97272402e-01 8.15199673e-01 -3.12741905e-01 4.84910429e-01 6.49334610e-01 5.05219281e-01 7.93676138e-01 4.23774451e-01 -1.10924828e+00 -1.47402406e-01 7.00883418e-02 2.60105640e-01 -9.19065237e-01 -7.74684012e-01 -5.74534805e-03 -9.88780141e-01 4.37861055e-01 1.71621472e-01 3.63588542e-01 -9.11248207e-01 1.33565080e+00 6.20856881e-01 -1.51592314e-01 -1.14702061e-01 1.01519036e+00 3.93483013e-01 4.69833344e-01 -1.77474439e-01 -6.80301487e-01 1.33037221e+00 -4.20844287e-01 -5.83696127e-01 2.23387197e-01 1.74003527e-01 -1.07179499e+00 4.23540264e-01 6.58153772e-01 -9.65558708e-01 -4.41223532e-01 -1.10207760e+00 8.48393217e-02 -1.45515248e-01 2.41412789e-01 3.22687328e-01 4.31898534e-01 -9.97715116e-01 1.04832256e+00 -1.63935781e+00 -2.42552042e-01 1.33365065e-01 5.46357930e-01 -1.16143560e+00 -3.97546729e-03 -2.97566473e-01 1.00307178e+00 4.54861879e-01 6.63188323e-02 -7.47343302e-01 -2.35305205e-01 -6.08462274e-01 -4.81986582e-01 -9.41184312e-02 -4.16369677e-01 4.63059902e-01 -6.14172332e-02 -1.59330034e+00 1.03904438e+00 -7.07578897e-01 -3.92381936e-01 1.03948966e-01 5.47864884e-02 -1.32012293e-01 5.70045888e-01 1.29413426e-01 3.93633634e-01 8.76390517e-01 -1.48323631e+00 4.97136004e-02 -6.87304199e-01 -7.69605041e-01 1.25775024e-01 5.43323278e-01 3.49105269e-01 -6.31464794e-02 -1.71303585e-01 9.66398239e-01 -1.23027778e+00 -4.17967677e-01 -7.67459214e-01 -5.66963077e-01 7.59579465e-02 1.11558211e+00 -8.89334917e-01 5.16562998e-01 -1.83196020e+00 8.83937478e-01 4.47364479e-01 3.51179183e-01 -1.37813361e-02 4.36471581e-01 3.94189715e-01 -3.53660136e-01 -2.74507433e-01 -3.08872610e-01 -8.75272810e-01 -1.18711576e-01 3.61391902e-01 8.23415294e-02 1.19688129e+00 2.38975719e-01 2.56067008e-01 -8.57123733e-01 -3.15962940e-01 6.06629133e-01 6.94155037e-01 -3.47155064e-01 6.02441132e-02 1.82884365e-01 1.13121486e+00 -1.03590183e-01 3.62555772e-01 1.18265533e+00 -4.71632928e-01 6.24119401e-01 -4.51853067e-01 -4.47470069e-01 3.70179653e-01 -1.33401811e+00 1.61620033e+00 1.64996117e-01 2.85992205e-01 5.25060713e-01 -1.00747013e+00 9.50603724e-01 2.49252722e-01 6.76150322e-01 -2.00200062e-02 -1.78664532e-02 7.44077265e-02 -1.24774456e-01 -9.58436206e-02 5.83919764e-01 -2.00166389e-01 2.79863507e-01 6.22070014e-01 4.25780952e-01 -1.09576344e-01 2.07851067e-01 3.60602200e-01 9.96322989e-01 4.06885087e-01 2.31317699e-01 -7.49177217e-01 6.88675284e-01 -2.44822130e-02 5.03257394e-01 4.38546240e-01 -1.70743421e-01 1.10086906e+00 1.46459967e-01 -5.16156137e-01 -1.53292131e+00 -1.13254237e+00 -6.98760211e-01 4.94129300e-01 -1.11917712e-01 -5.56359887e-01 -7.23570406e-01 -2.38225758e-01 -2.94007719e-01 1.33161291e-01 -2.36375868e-01 2.36502275e-01 -5.47088921e-01 -1.63487232e+00 -2.80824807e-02 6.96435664e-03 -3.34997505e-01 -1.02036524e+00 -6.30619943e-01 2.15416834e-01 -5.12257554e-02 -9.91818547e-01 2.09781870e-01 7.47778416e-01 -8.85425448e-01 -1.11172533e+00 -3.80849838e-01 -4.41082180e-01 9.68515873e-01 2.75197268e-01 7.85957456e-01 -2.02151835e-01 -3.76864344e-01 1.27974585e-01 -3.08936059e-01 3.38961005e-01 -4.81032252e-01 -4.92716789e-01 9.59297001e-01 -1.66800275e-01 4.59374189e-01 -1.04626000e+00 -4.45837826e-01 4.40019906e-01 -6.28751636e-01 -1.05511606e-01 4.36038315e-01 1.15437937e+00 1.26985133e+00 3.53358313e-02 -1.21533237e-01 -5.75310230e-01 3.13663781e-01 -3.63233566e-01 -5.75333953e-01 4.10917923e-02 -3.91061395e-01 2.30086595e-01 7.19101727e-01 -1.04876675e-01 -9.52175975e-01 2.19354510e-01 -2.53871322e-01 -4.77301985e-01 -5.86187899e-01 1.25736177e-01 -2.07885623e-01 -2.89872348e-01 4.33647186e-01 5.68908334e-01 3.65483761e-01 -8.88524830e-01 1.51338682e-01 6.04194462e-01 7.74457633e-01 -8.49936247e-01 5.81725895e-01 1.00407863e+00 3.94134969e-01 -1.18436122e+00 1.52878507e-04 -8.40407491e-01 -1.26747012e+00 6.07504286e-02 9.72591400e-01 -5.41834295e-01 -1.02988243e+00 4.12418127e-01 -1.06012905e+00 1.83476120e-01 2.09317684e-01 8.49125326e-01 -7.74118304e-01 1.17220330e+00 -6.98671937e-01 -7.80572534e-01 -2.68324256e-01 -1.87844741e+00 1.16316319e+00 6.72643483e-02 -1.79466590e-01 -7.32708931e-01 3.38421673e-01 5.30786991e-01 1.07840218e-01 2.30234236e-01 6.41453087e-01 -7.17564225e-01 -3.04549783e-01 -3.06930542e-02 2.39153504e-01 3.96331102e-01 3.21816541e-02 3.12903404e-01 -7.66442299e-01 -6.42324865e-01 4.82674241e-01 3.46816480e-02 7.01349080e-01 4.58179295e-01 6.82028413e-01 7.96144009e-02 -4.94834900e-01 6.79952383e-01 1.34022784e+00 7.85005316e-02 6.55393481e-01 5.62343061e-01 6.83569133e-01 3.70426565e-01 5.72210371e-01 4.91273731e-01 8.49924330e-03 6.65301025e-01 3.25910956e-01 3.40068698e-01 4.57497656e-01 3.39850008e-01 2.10937649e-01 1.12010872e+00 -7.07517684e-01 2.68536270e-01 -8.82485867e-01 2.17240438e-01 -1.72475994e+00 -1.19509876e+00 -4.33772534e-01 2.35380650e+00 6.27719164e-01 -4.44642790e-02 -2.53183395e-02 2.74322152e-01 6.75892413e-01 -3.07696667e-02 -2.60681450e-01 1.39529705e-01 -5.71330547e-01 4.57589984e-01 5.64086080e-01 5.20017087e-01 -1.25871682e+00 5.56372166e-01 6.30266190e+00 6.20123625e-01 -8.97344053e-01 1.71236083e-01 3.44170064e-01 5.56610040e-02 1.27776384e-01 3.94261360e-01 -6.99832201e-01 7.81695485e-01 1.01275289e+00 2.73319781e-01 4.65348661e-01 8.39452922e-01 3.56402278e-01 -3.99305493e-01 -6.98102355e-01 9.18453515e-01 -5.42224245e-03 -1.32463837e+00 -2.73309380e-01 3.51292044e-01 5.49965918e-01 5.30490935e-01 -1.97308853e-01 -3.87043357e-01 3.43547761e-01 -8.07228029e-01 4.02287573e-01 7.57821620e-01 3.88866037e-01 -7.36925066e-01 7.84322500e-01 2.37286299e-01 -8.11791062e-01 4.67651129e-01 -7.87402928e-01 6.63301721e-02 7.18487263e-01 9.03081179e-01 -8.02036107e-01 7.14782715e-01 9.22996283e-01 6.25162184e-01 -3.10101151e-01 6.46209419e-01 2.96115112e-02 2.21044034e-01 -6.92944288e-01 4.15691406e-01 -2.46261321e-02 -1.00454867e+00 9.87254798e-01 1.04874897e+00 1.31295547e-01 -5.75737506e-02 1.93862841e-01 7.60664165e-01 3.40325147e-01 -2.15241194e-01 -6.95306212e-02 2.86455095e-01 5.50091624e-01 1.67701030e+00 -1.21013439e+00 -2.36071840e-01 9.03400406e-02 1.00990188e+00 1.63433090e-01 4.70177382e-02 -2.94218749e-01 1.34004921e-01 8.30040216e-01 -4.20206226e-02 5.51663995e-01 -6.02429450e-01 7.51635507e-02 -1.03770185e+00 7.97187909e-02 -9.44264293e-01 3.07017099e-02 -7.98530519e-01 -1.25018644e+00 4.94224697e-01 5.18475519e-03 -9.03007150e-01 -1.22861512e-01 -9.39017177e-01 -4.12919044e-01 9.44056749e-01 -1.05522454e+00 -9.30735946e-01 1.50528178e-01 2.47128859e-01 -6.83143288e-02 -1.03817686e-01 1.11531353e+00 -1.96861196e-02 -5.15141904e-01 -3.03211093e-01 7.49364734e-01 -3.57967019e-01 7.58546948e-01 -1.66325569e+00 1.95401952e-01 8.40782523e-01 -2.30441354e-02 1.18992412e+00 1.14179695e+00 -7.30767727e-01 -1.68590796e+00 -7.36764014e-01 4.85444486e-01 -7.80997455e-01 5.78489602e-01 -1.70572013e-01 -1.04760170e+00 6.53904319e-01 1.12233885e-01 1.50630891e-01 9.21777368e-01 -1.68851256e-01 -2.62570027e-02 5.42464852e-01 -1.37410569e+00 5.11024334e-02 5.17792761e-01 -6.82609677e-01 -7.09119022e-01 7.25133657e-01 1.75209001e-01 -2.54252017e-01 -1.50868237e+00 2.04629749e-01 3.54626983e-01 -1.13395882e+00 1.21816671e+00 -6.52662992e-01 -1.65375710e-01 -1.00193691e+00 -3.02753270e-01 -1.24374175e+00 -6.19683921e-01 -9.47586417e-01 1.36792257e-01 6.58716142e-01 -4.71089929e-02 -5.09662151e-01 7.77164876e-01 2.31256381e-01 -3.79792631e-01 -4.17191833e-01 -1.21100402e+00 -5.39135754e-01 -1.69487372e-01 -1.15668967e-01 1.79668918e-01 1.01070237e+00 2.66308993e-01 1.39042482e-01 -3.69570881e-01 4.82525617e-01 1.25406229e+00 1.47760972e-01 7.59954929e-01 -1.37308967e+00 -3.20808828e-01 1.28758606e-02 -8.65461648e-01 -8.71163607e-01 1.15143657e-01 -7.12345839e-01 1.36641160e-01 -9.99410748e-01 4.48082834e-01 -3.67069215e-01 -1.80428237e-01 1.02507845e-01 -1.27880007e-01 3.28795969e-01 -2.90919673e-02 6.96167350e-01 -9.47939694e-01 3.81630450e-01 7.03681886e-01 2.34546110e-01 9.04032364e-02 -2.77408838e-01 2.01315340e-02 6.03709519e-01 5.55733442e-01 -6.64508641e-01 3.68840009e-01 1.51626930e-01 3.91629450e-02 -1.77928597e-01 3.36445808e-01 -1.12838078e+00 3.47263634e-01 7.04809800e-02 4.91360873e-01 -1.16271245e+00 7.31775701e-01 -6.33470953e-01 6.57927334e-01 2.34554306e-01 5.24522901e-01 3.02538037e-01 -2.90940553e-01 4.93498266e-01 -2.06823707e-01 -1.33726597e-01 9.94235992e-01 -4.41342175e-01 -2.25253031e-01 1.39613412e-02 -5.59683681e-01 -5.34379959e-01 8.29406917e-01 -1.33761168e-01 -3.12373579e-01 1.44380704e-01 -1.26005125e+00 -5.86699024e-02 1.23710060e+00 -3.62094700e-01 6.66401148e-01 -9.88245726e-01 -5.43511629e-01 3.21400493e-01 -2.31192350e-01 3.01715195e-01 5.48764527e-01 1.54052913e+00 -7.85859227e-01 4.76396561e-01 -5.42086899e-01 -1.18796885e+00 -1.50334680e+00 5.79521358e-01 9.64276791e-02 -2.93090403e-01 -5.69801807e-01 5.73618889e-01 -4.15689051e-01 -7.45655596e-01 -4.57289428e-01 -6.87022367e-03 -1.65905252e-01 -2.62725234e-01 7.01666057e-01 4.54523832e-01 4.77354944e-01 -1.30376971e+00 -5.54942429e-01 5.39722979e-01 -2.06960201e-01 6.76878020e-02 1.86234331e+00 -2.76871920e-01 -7.24801183e-01 1.85885370e-01 1.17832530e+00 1.45519637e-02 -1.25228119e+00 9.67684537e-02 1.02260970e-01 -1.95772693e-01 -8.22111741e-02 -8.91729891e-02 -3.60923678e-01 7.20310509e-01 4.50350702e-01 2.12790668e-02 6.38528109e-01 9.10127535e-02 1.65175408e-01 6.18631780e-01 5.58182836e-01 -1.05090511e+00 -5.76816142e-01 4.46171701e-01 3.12578142e-01 -1.14801347e+00 4.65422928e-01 -1.42253369e-01 -3.71432185e-01 1.25698483e+00 -2.15385184e-01 -2.56586969e-01 4.93874937e-01 2.88850456e-01 -6.21057674e-02 -4.15983588e-01 -6.22974396e-01 1.31515875e-01 -1.56189367e-01 8.92073154e-01 1.74740523e-01 1.49196312e-01 -1.38703316e-01 3.26895297e-01 -2.10420713e-01 -5.65141678e-01 4.76599306e-01 1.09133744e+00 -5.36974072e-01 -1.28945851e+00 -9.25644577e-01 1.99102849e-01 -5.33631384e-01 1.42635763e-01 -9.99683738e-02 3.68124276e-01 -2.78243303e-01 6.36522114e-01 1.70292526e-01 -2.76642740e-01 -1.63142383e-01 3.63569379e-01 5.78734517e-01 -3.70874077e-01 -1.08669095e-01 3.11012983e-01 -2.27751255e-01 -5.72683930e-01 -9.33826983e-01 -1.08700299e+00 -1.25013030e+00 -2.65697151e-01 -4.81806993e-01 6.20739818e-01 1.13015783e+00 1.04305959e+00 5.47443748e-01 7.67680034e-02 6.30990922e-01 -1.63728023e+00 -4.90300804e-01 -1.13212085e+00 -6.71426773e-01 7.64851570e-01 1.45974964e-01 -8.98315251e-01 -6.15865588e-01 2.59657532e-01]
[13.295042991638184, -3.065812587738037]
433cdfd9-370e-4923-b812-05edd9d42cea
dds-decoupled-dynamic-scene-graph-generation
2301.07666
null
https://arxiv.org/abs/2301.07666v1
https://arxiv.org/pdf/2301.07666v1.pdf
DDS: Decoupled Dynamic Scene-Graph Generation Network
Scene-graph generation involves creating a structural representation of the relationships between objects in a scene by predicting subject-object-relation triplets from input data. However, existing methods show poor performance in detecting triplets outside of a predefined set, primarily due to their reliance on dependent feature learning. To address this issue we propose DDS -- a decoupled dynamic scene-graph generation network -- that consists of two independent branches that can disentangle extracted features. The key innovation of the current paper is the decoupling of the features representing the relationships from those of the objects, which enables the detection of novel object-relationship combinations. The DDS model is evaluated on three datasets and outperforms previous methods by a significant margin, especially in detecting previously unseen triplets.
['B. S. Manjunath', 'Suya You', 'Satish Kumar', 'Raphael Ruschel', 'A S M Iftekhar']
2023-01-18
null
null
null
null
['scene-graph-generation']
['computer-vision']
[ 3.74599278e-01 -1.15641423e-01 8.92726704e-02 -5.06651461e-01 -3.69418234e-01 -5.72786808e-01 5.99658966e-01 3.13204497e-01 3.09124868e-02 5.37191927e-01 1.32538944e-01 1.13142073e-01 -3.57630879e-01 -6.89214230e-01 -5.16044915e-01 -6.34777367e-01 -1.77961633e-01 4.98931289e-01 5.84317148e-01 -1.56131193e-01 1.80448323e-01 7.48203874e-01 -1.82156372e+00 4.66538668e-01 6.67819500e-01 9.23887372e-01 1.77895814e-01 4.95590419e-01 -2.82853004e-02 7.00933099e-01 -6.02262795e-01 -3.56047630e-01 4.34593111e-01 -5.59825122e-01 -4.51669484e-01 2.77191967e-01 7.85431921e-01 -1.90319017e-01 -4.03190017e-01 7.81547844e-01 1.60437182e-01 7.54743889e-02 6.64700449e-01 -1.70633233e+00 -5.03828585e-01 2.34565631e-01 -5.57939351e-01 3.65099341e-01 5.61198831e-01 7.49438256e-02 1.30973852e+00 -9.27831531e-01 7.98618972e-01 1.32557571e+00 4.34656382e-01 2.56902903e-01 -1.51496339e+00 -6.24653935e-01 1.35585219e-01 5.33745408e-01 -1.47470605e+00 -4.69139159e-01 1.05100632e+00 -5.88118732e-01 9.94053960e-01 3.23639750e-01 7.19338119e-01 8.71446192e-01 8.53930116e-02 7.93693602e-01 9.04723048e-01 -5.12496829e-01 1.19065568e-01 1.42749444e-01 2.81040967e-01 5.98205030e-01 6.06359005e-01 7.07588419e-02 -8.33522141e-01 -1.02150477e-01 6.67039275e-01 -2.55849123e-01 -2.28102244e-02 -9.93974686e-01 -1.03658962e+00 5.77004433e-01 6.15935683e-01 3.03368479e-01 -2.24386021e-01 -2.55888492e-01 6.34165034e-02 2.33899131e-01 1.51258066e-01 6.28291547e-01 -2.96149731e-01 2.51376629e-01 -5.68448484e-01 2.99570471e-01 6.23767257e-01 1.00328279e+00 9.11817849e-01 -1.84270278e-01 -2.74072766e-01 5.11148036e-01 2.33182713e-01 1.92895636e-01 7.23636150e-02 -3.45607191e-01 5.61665058e-01 1.26490474e+00 -1.35905147e-01 -1.39033079e+00 -4.60207880e-01 -3.94397974e-01 -5.37652135e-01 6.31983764e-03 2.06074238e-01 6.64879382e-02 -9.23195422e-01 1.77720952e+00 6.43371463e-01 9.92493406e-02 3.31391245e-02 8.29810381e-01 1.03061903e+00 2.74686038e-01 4.99306507e-02 -1.00179575e-01 1.19851732e+00 -6.91387773e-01 -5.00845790e-01 -4.11581427e-01 4.03643638e-01 -4.25560236e-01 7.42671311e-01 -5.95120192e-02 -7.33936191e-01 -6.83689713e-01 -1.07208467e+00 -2.58491673e-02 -7.10437477e-01 1.08021155e-01 8.17361236e-01 3.51487130e-01 -1.05003846e+00 2.91192234e-01 -4.80465323e-01 -4.75134969e-01 4.87955540e-01 6.91175759e-01 -5.22412956e-01 9.86197684e-03 -8.93820345e-01 8.49767745e-01 7.40295708e-01 2.09007546e-01 -7.15409994e-01 -5.37098169e-01 -8.76017630e-01 2.37472832e-01 6.73112035e-01 -9.29134786e-01 6.78849816e-01 -8.56800258e-01 -7.79864788e-01 7.24631906e-01 -2.46847019e-01 -1.86894014e-01 3.31070900e-01 -3.32835242e-02 -6.16039991e-01 2.69401312e-01 2.64057815e-01 5.45861959e-01 9.06762421e-01 -1.59782779e+00 -6.74314380e-01 -5.50238669e-01 1.12780079e-01 3.77992570e-01 -3.20672691e-01 -2.74516903e-02 -4.71254081e-01 -4.43371207e-01 3.73477012e-01 -8.19716156e-01 -2.10145444e-01 -7.22177923e-02 -7.85827875e-01 -2.58233875e-01 1.12731576e+00 -1.33723989e-01 9.13040638e-01 -2.34477615e+00 2.35252697e-02 2.71902442e-01 6.44115150e-01 3.74506384e-01 -3.00965399e-01 6.20145500e-01 -4.39196169e-01 -3.06930989e-01 1.80564389e-01 -1.96782753e-01 -3.48979771e-01 1.10497393e-01 -2.63041079e-01 2.92531133e-01 5.71265042e-01 9.37718809e-01 -1.08985913e+00 -6.81399167e-01 3.95696104e-01 8.77880901e-02 -3.01511675e-01 2.53461540e-01 -2.36688599e-01 3.94349456e-01 -4.72384989e-01 5.84022760e-01 7.75807500e-01 -4.95658278e-01 3.57079238e-01 -4.26085740e-01 1.16155639e-01 1.55077621e-01 -1.31877077e+00 1.24265265e+00 2.66805053e-01 7.18689084e-01 -4.90300059e-01 -9.03555274e-01 9.27589238e-01 4.18080017e-02 7.41901398e-01 -6.18846714e-01 2.63133142e-02 -5.73373921e-02 2.65511096e-01 -6.01697743e-01 4.79955435e-01 8.03925022e-02 5.19672316e-03 6.94781840e-02 2.07946584e-01 -1.86651535e-02 6.04317665e-01 4.28062081e-01 1.13874614e+00 1.50348023e-02 4.49734360e-01 -3.60708162e-02 4.04942632e-01 1.40462399e-01 5.73228419e-01 9.22901452e-01 -1.75736815e-01 5.45214891e-01 7.48818159e-01 -4.78512526e-01 -5.61943650e-01 -1.26171768e+00 -9.79004055e-02 7.97901750e-01 5.41667163e-01 -6.63710237e-01 -1.84900999e-01 -9.36587751e-01 1.93548456e-01 7.53900170e-01 -7.46425569e-01 -2.89244324e-01 -4.81460959e-01 -6.33983254e-01 7.61072477e-03 4.92654741e-01 2.54968494e-01 -9.77970421e-01 -6.59195125e-01 7.43375570e-02 -1.83236122e-01 -1.60259700e+00 -2.14307457e-02 2.16039494e-01 -7.42802143e-01 -1.45932341e+00 1.06165066e-01 -7.87877202e-01 9.95823681e-01 6.22137845e-01 1.07503629e+00 -1.04218043e-01 -6.46605670e-01 4.04777318e-01 -3.02279055e-01 -3.79495412e-01 -2.61239171e-01 -1.76907152e-01 -1.02575244e-02 4.45847303e-01 4.41675931e-01 -6.08280182e-01 -4.77771133e-01 3.48852634e-01 -6.97693288e-01 2.74903178e-01 6.01594925e-01 5.87426782e-01 3.71822089e-01 2.23634064e-01 3.25327992e-01 -8.37714970e-01 3.26361626e-01 -5.26054561e-01 -6.09758675e-01 6.12086356e-01 -4.74368662e-01 7.81636909e-02 5.13609111e-01 -5.11317790e-01 -9.93583620e-01 4.20853764e-01 6.50407672e-01 -4.76874352e-01 -2.59609848e-01 3.10411096e-01 -2.57856876e-01 -1.20639885e-02 7.06616819e-01 1.32904977e-01 -3.07804346e-01 -2.94315428e-01 3.55484754e-01 4.36888412e-02 5.02021730e-01 -3.88917267e-01 1.15227985e+00 4.85969096e-01 4.75725561e-01 -7.67039299e-01 -8.82693112e-01 -8.07409763e-01 -1.05779266e+00 -4.30441976e-01 6.16332293e-01 -8.43645394e-01 -3.73100579e-01 1.81903109e-01 -1.12896645e+00 2.51629144e-01 -4.69681293e-01 4.01696950e-01 -2.26756081e-01 2.94918269e-01 -2.06476927e-01 -6.27303839e-01 2.61596233e-01 -9.34906602e-01 1.13341868e+00 2.99253970e-01 -2.23231092e-01 -7.52408803e-01 -2.94403844e-02 3.03943396e-01 -4.20716442e-02 6.29749000e-01 1.09382498e+00 -8.06148171e-01 -9.41817462e-01 -3.83251011e-01 -4.97244149e-01 -1.53764850e-02 5.09608030e-01 1.40719041e-01 -9.09212828e-01 -2.66242921e-01 -1.32755503e-01 -1.72040865e-01 7.07549453e-01 1.26919851e-01 9.13582027e-01 -1.33064762e-01 -6.85576975e-01 3.50306928e-01 1.46185756e+00 2.68761069e-01 3.98697585e-01 1.38544813e-01 9.07475293e-01 7.92272329e-01 4.37688440e-01 3.66797894e-01 3.84552956e-01 8.40382278e-01 6.13135755e-01 -1.63647756e-01 -4.82961506e-01 -4.41484749e-01 1.42387655e-02 3.28916311e-01 2.04521969e-01 -4.35410500e-01 -8.58318627e-01 6.54996276e-01 -1.98348379e+00 -7.53977060e-01 -6.20774388e-01 1.99560106e+00 4.59942102e-01 1.82947412e-01 1.75165266e-01 1.76656675e-02 7.50274241e-01 1.26617730e-01 -5.96689343e-01 9.23420414e-02 -3.17234784e-01 -1.02278851e-01 8.89334641e-03 1.07066922e-01 -9.77441907e-01 1.06121242e+00 6.71123266e+00 3.57310981e-01 -7.30615735e-01 -4.07588542e-01 1.81967661e-01 -7.29650483e-02 -2.40031295e-02 3.02213579e-01 -9.37558055e-01 9.29323807e-02 3.19318444e-01 -1.41805887e-01 -5.75820953e-02 7.43501127e-01 -2.13356484e-02 -3.71153802e-01 -1.41428852e+00 9.39047217e-01 4.32083428e-01 -1.10608912e+00 3.90945643e-01 1.51591241e-01 6.39944911e-01 -2.51579940e-01 9.00688767e-02 -7.66971242e-03 4.51297075e-01 -6.00614369e-01 6.10227287e-01 2.88921952e-01 5.47531903e-01 -2.00295061e-01 3.55055392e-01 1.92911699e-01 -1.26114130e+00 -1.43029705e-01 -1.24957390e-01 -6.88482821e-02 -2.99497470e-02 5.73108673e-01 -1.31115127e+00 6.99902117e-01 5.57057500e-01 7.66230464e-01 -1.23662984e+00 1.27866662e+00 -2.37543896e-01 2.40594298e-01 -3.09020191e-01 -4.62937681e-03 5.89702465e-02 -4.56195623e-02 7.86367655e-01 1.05806804e+00 -4.11827303e-02 -2.02775430e-02 1.87845722e-01 9.74725068e-01 2.08362266e-01 -6.61918381e-03 -8.29510272e-01 1.11366548e-01 2.65073746e-01 1.35312128e+00 -1.00908387e+00 -2.37043694e-01 -4.31827724e-01 6.67587161e-01 5.13480663e-01 3.10423672e-01 -6.89574301e-01 -1.78289145e-01 4.21836495e-01 9.29356962e-02 5.40005982e-01 -2.90741861e-01 -2.01197863e-01 -1.26279163e+00 3.24572116e-01 -6.77370012e-01 6.29163325e-01 -8.22335422e-01 -1.30519807e+00 4.69081044e-01 4.27796066e-01 -1.34016573e+00 -2.61121750e-01 -5.46452343e-01 -3.61997515e-01 5.10086834e-01 -1.28538907e+00 -1.51414800e+00 -5.06987691e-01 7.94380665e-01 3.35906297e-01 -1.86426073e-01 6.02064252e-01 1.61545444e-02 -5.95543563e-01 3.99633765e-01 -1.59943402e-01 1.47932544e-01 3.96711439e-01 -1.09412003e+00 3.27524513e-01 1.15999115e+00 6.20753050e-01 5.38542032e-01 7.11214542e-01 -8.75401139e-01 -1.25906849e+00 -9.61087704e-01 1.09919310e+00 -7.35724628e-01 7.04404831e-01 -8.46929193e-01 -8.62052977e-01 5.72891057e-01 -1.74320564e-01 2.51373947e-01 6.30865872e-01 3.26561630e-01 -7.07535565e-01 -2.86233932e-01 -9.34588850e-01 8.17618310e-01 1.44440949e+00 -4.82895106e-01 -6.29538715e-01 4.13792789e-01 4.12025541e-01 -2.99194783e-01 -4.23436731e-01 6.30904675e-01 5.08261859e-01 -1.27219987e+00 1.04704082e+00 -8.22409213e-01 3.82076949e-01 -3.80412519e-01 -1.39947429e-01 -1.06717348e+00 -6.82211876e-01 -3.35143834e-01 -1.70628890e-01 1.18175030e+00 2.76628941e-01 -4.06646043e-01 7.92756200e-01 5.76574862e-01 1.66757524e-01 -3.67222786e-01 -8.55822384e-01 -8.51005614e-01 -7.12656975e-01 -2.08777934e-01 6.36863828e-01 9.17074978e-01 -1.85085550e-01 8.67714107e-01 -2.22681329e-01 3.43676150e-01 7.08055496e-01 4.42256749e-01 9.47328448e-01 -1.39030111e+00 -2.70910382e-01 -2.51028121e-01 -8.85510921e-01 -6.84762359e-01 1.26614571e-01 -6.65386498e-01 4.40558605e-02 -1.51695037e+00 4.23286349e-01 -5.48750043e-01 -3.98145229e-01 6.76059723e-01 -2.64190167e-01 1.01369247e-01 3.42422903e-01 3.31115723e-01 -8.36598098e-01 5.10357797e-01 1.33692753e+00 -7.74375796e-02 -3.10677707e-01 -2.04262305e-02 -6.45274937e-01 6.28724873e-01 5.56984723e-01 -5.58063686e-01 -6.89878404e-01 -3.28096002e-01 1.31496698e-01 3.26155219e-03 6.22340083e-01 -1.11073101e+00 2.36842558e-01 -2.56773412e-01 5.95547557e-01 -8.28173995e-01 4.35429096e-01 -1.08351016e+00 3.17950785e-01 2.41690919e-01 -2.87419677e-01 8.62375125e-02 2.23317862e-01 8.43448400e-01 -1.56086698e-01 7.79059976e-02 4.31644112e-01 -3.48958466e-03 -9.54750240e-01 1.10601537e-01 -2.88629867e-02 -5.75339571e-02 1.27186024e+00 -5.07883847e-01 -4.76775736e-01 -1.63524315e-01 -5.85729599e-01 1.15856703e-03 2.59015411e-01 7.49788046e-01 7.29553103e-01 -1.26027012e+00 -5.96457303e-01 4.96158570e-01 5.56831598e-01 5.94626181e-03 1.00289904e-01 5.01384556e-01 -5.43665774e-02 3.59837562e-01 -2.76317865e-01 -7.51065016e-01 -1.57302070e+00 6.81199670e-01 1.96368694e-01 -2.15744793e-01 -6.49262190e-01 7.24470377e-01 6.12051368e-01 -6.11962825e-02 3.74296866e-02 -9.31353495e-02 -3.11425209e-01 1.33945659e-01 1.24131069e-01 2.14324281e-01 -3.57415304e-02 -9.59854245e-01 -6.03584111e-01 4.23180968e-01 -2.97449470e-01 2.08681718e-01 1.32704532e+00 -6.46907231e-03 -2.11297631e-01 5.26790619e-01 1.18524456e+00 -1.43445343e-01 -1.18871164e+00 -4.75882173e-01 1.41247526e-01 -8.10071528e-01 -3.20105970e-01 -7.62403727e-01 -8.70321810e-01 3.43014002e-01 4.85476732e-01 3.80854934e-01 1.23816693e+00 3.59482408e-01 2.39493102e-01 2.83876270e-01 2.36560091e-01 -6.88615799e-01 4.81189609e-01 2.22162262e-01 8.49347889e-01 -1.34714460e+00 2.67154455e-01 -8.16296637e-01 -4.34603781e-01 1.18947113e+00 9.87213552e-01 -5.63674085e-02 5.23623526e-01 1.13972969e-01 -2.66345739e-01 -5.85966349e-01 -6.95161402e-01 -4.69953746e-01 7.69889653e-01 7.70731032e-01 -2.06386112e-02 -6.05629124e-02 5.62065728e-02 7.97419902e-03 -1.13311678e-01 -1.74603850e-01 2.83511937e-01 8.37044299e-01 -7.51402453e-02 -1.05471742e+00 -2.57560670e-01 5.86795390e-01 6.49128556e-02 -2.15410255e-02 -8.43008280e-01 8.73026848e-01 2.90790528e-01 1.12792861e+00 7.62592256e-02 -5.33596814e-01 4.56498802e-01 -2.25418746e-01 5.04630268e-01 -7.32277870e-01 -4.64561909e-01 -9.23675522e-02 1.05462097e-01 -6.06702745e-01 -6.50118113e-01 -9.07015920e-01 -8.40099037e-01 2.12196082e-01 -6.05937183e-01 -7.09363371e-02 3.50828737e-01 8.99100900e-01 6.00261033e-01 4.55889881e-01 7.67556190e-01 -4.62618470e-01 -1.34656191e-01 -5.05987346e-01 -5.49670756e-01 7.21741021e-01 4.80184108e-01 -9.24201906e-01 -3.16240460e-01 7.96467736e-02]
[10.251051902770996, 1.6911972761154175]
f48c0b1e-b07d-47ab-a5aa-395c344329df
generative-ode-modeling-with-known-unknowns
2003.10775
null
https://arxiv.org/abs/2003.10775v2
https://arxiv.org/pdf/2003.10775v2.pdf
Generative ODE Modeling with Known Unknowns
In several crucial applications, domain knowledge is encoded by a system of ordinary differential equations (ODE), often stemming from underlying physical and biological processes. A motivating example is intensive care unit patients: the dynamics of vital physiological functions, such as the cardiovascular system with its associated variables (heart rate, cardiac contractility and output and vascular resistance) can be approximately described by a known system of ODEs. Typically, some of the ODE variables are directly observed (heart rate and blood pressure for example) while some are unobserved (cardiac contractility, output and vascular resistance), and in addition many other variables are observed but not modeled by the ODE, for example body temperature. Importantly, the unobserved ODE variables are known-unknowns: We know they exist and their functional dynamics, but cannot measure them directly, nor do we know the function tying them to all observed measurements. As is often the case in medicine, and specifically the cardiovascular system, estimating these known-unknowns is highly valuable and they serve as targets for therapeutic manipulations. Under this scenario we wish to learn the parameters of the ODE generating each observed time-series, and extrapolate the future of the ODE variables and the observations. We address this task with a variational autoencoder incorporating the known ODE function, called GOKU-net for Generative ODE modeling with Known Unknowns. We first validate our method on videos of single and double pendulums with unknown length or mass; we then apply it to a model of the cardiovascular system. We show that modeling the known-unknowns allows us to successfully discover clinically meaningful unobserved system parameters, leads to much better extrapolation, and enables learning using much smaller training sets.
['Neta Ravid', 'Danny Eytan', 'Ori Linial', 'Uri Shalit']
2020-03-24
null
https://openreview.net/forum?id=pmvEzAbl7M
https://openreview.net/pdf?id=pmvEzAbl7M
iclr-workshop-deepdiffeq-2019-12
['known-unknowns']
['miscellaneous']
[ 5.91522008e-02 2.12698743e-01 -3.17217886e-01 1.13496691e-01 -7.94504303e-03 -7.57000029e-01 3.40231806e-01 -6.33178428e-02 -5.44194840e-02 1.36927593e+00 3.77086699e-02 -2.45700508e-01 -2.34718114e-01 -5.06176472e-01 -8.34387302e-01 -1.10403514e+00 -3.37691516e-01 6.89843774e-01 -3.72646004e-01 -1.52944565e-01 -3.07844490e-01 4.88051862e-01 -1.04353702e+00 -3.51363361e-01 6.01987004e-01 7.34532297e-01 -2.32129008e-01 1.15462697e+00 4.09810036e-01 1.00293195e+00 -4.37447071e-01 1.58625364e-01 1.53015614e-01 -7.49144852e-01 -5.17036915e-01 5.85073531e-02 -8.31470191e-02 -6.11808479e-01 -6.76622093e-01 5.57270348e-01 2.04163194e-01 2.79206127e-01 1.02517843e+00 -1.19243574e+00 -4.91540551e-01 2.04560220e-01 -5.37841767e-02 8.78254324e-02 -8.07576627e-02 3.33413333e-01 7.55477786e-01 -7.01390579e-02 5.13771951e-01 9.87041533e-01 7.93182015e-01 7.76897907e-01 -1.70056760e+00 -1.73596472e-01 -2.36996711e-04 -4.21534240e-01 -1.23174036e+00 -2.56294966e-01 5.93302011e-01 -8.49044740e-01 4.07819301e-01 2.06084281e-01 8.30112994e-01 1.48518741e+00 7.55661726e-01 3.51089865e-01 7.46424079e-01 4.84693348e-02 5.20733476e-01 2.33760104e-01 8.90136976e-03 7.83532441e-01 1.70383602e-01 4.81614500e-01 -1.15110233e-01 -4.11484540e-01 1.34619367e+00 5.28287470e-01 -6.62571490e-01 -4.93147731e-01 -1.14300454e+00 8.16636682e-01 8.09119418e-02 1.69868320e-01 -6.71417773e-01 4.91027147e-01 1.78011388e-01 3.07260603e-01 9.25177410e-02 6.10747874e-01 -1.02497172e+00 -3.99921723e-02 -5.71109712e-01 3.62950921e-01 1.25935090e+00 6.31266356e-01 6.53208256e-01 2.55699873e-01 9.54175591e-02 3.94887745e-01 1.85394123e-01 6.32233143e-01 6.27161086e-01 -1.34804213e+00 -1.57800630e-01 2.38622159e-01 4.43082631e-01 -5.53184092e-01 -6.69529021e-01 -3.94980013e-01 -1.14882064e+00 9.62968916e-02 7.08891094e-01 -7.47651637e-01 -1.11196935e+00 2.15479779e+00 3.32308173e-01 5.51363230e-01 6.91832528e-02 8.49979818e-01 5.42875886e-01 6.93331242e-01 -4.29300480e-02 -5.55057466e-01 1.22130513e+00 -2.30604559e-01 -8.11537445e-01 2.46937969e-03 5.37704766e-01 -1.80653766e-01 3.91151160e-01 1.51440620e-01 -1.13928151e+00 -4.57629710e-01 -7.25612700e-01 -3.79910469e-02 -2.57149637e-01 1.08161650e-03 5.59824884e-01 9.44243297e-02 -7.04090297e-01 1.15564430e+00 -1.48807025e+00 -2.13878244e-01 7.43575543e-02 4.40527499e-01 -2.66442508e-01 4.52398926e-01 -1.35860455e+00 9.82547998e-01 -3.81863788e-02 2.93576717e-01 -1.35425389e+00 -1.19940090e+00 -7.59241223e-01 2.14062139e-01 2.95249969e-01 -1.34269845e+00 1.23507571e+00 -4.81150657e-01 -1.64215064e+00 4.31616426e-01 -9.56102386e-02 -4.52926487e-01 4.84514207e-01 1.40159503e-01 -1.17554031e-01 4.68924269e-02 -3.89808297e-01 3.18723589e-01 8.20419610e-01 -1.15764058e+00 2.04943419e-01 -3.35778654e-01 1.55385852e-01 7.93835670e-02 1.81940496e-02 -8.05855215e-01 2.37309650e-01 -3.94599646e-01 1.86722115e-01 -1.18085468e+00 -3.89996886e-01 5.16568363e-01 -4.13412333e-01 5.07125914e-01 7.83897996e-01 -7.36472785e-01 1.01177704e+00 -1.79007673e+00 6.21134341e-01 -2.73980834e-02 8.04016650e-01 -4.81014885e-02 2.74897248e-01 4.08491611e-01 -3.05689216e-01 5.55526428e-02 -5.82343221e-01 -6.21162094e-02 -1.26804069e-01 7.46013343e-01 -2.71769583e-01 6.53373301e-01 3.02469134e-01 1.11986935e+00 -9.04977322e-01 -2.70945758e-01 3.65692586e-01 8.27508926e-01 -3.26994568e-01 3.13594431e-01 -3.77000093e-01 1.17178333e+00 -7.39423633e-01 1.87771156e-01 5.02225123e-02 -4.97166574e-01 2.35419601e-01 -1.72058456e-02 1.96357384e-01 -2.54097760e-01 -1.01378322e+00 1.04881120e+00 -5.62945843e-01 5.24374306e-01 1.04438156e-01 -1.47775352e+00 7.18646765e-01 8.91018867e-01 1.17914248e+00 -7.59940371e-02 5.66630960e-01 -8.21307302e-02 3.72231215e-01 -8.71543467e-01 -1.22807942e-01 -8.63711417e-01 8.99066776e-02 3.39966476e-01 2.68379867e-01 -3.69499743e-01 -1.33157641e-01 1.16677195e-01 1.24587476e+00 1.56862229e-01 4.24617350e-01 -2.22057953e-01 4.92562383e-01 -8.95215478e-03 5.92322290e-01 4.48625654e-01 -2.63717592e-01 4.91593927e-01 9.23429072e-01 -5.33785701e-01 -1.23122489e+00 -1.20615721e+00 -3.99761975e-01 1.21330835e-01 -1.91296354e-01 1.76168367e-01 -5.45995831e-01 8.74901935e-03 4.38052207e-01 2.67409712e-01 -1.06234169e+00 -4.70361382e-01 -5.21773577e-01 -7.78677583e-01 2.31877998e-01 5.14719725e-01 -2.78683275e-01 -9.47354496e-01 -7.03273237e-01 6.03906274e-01 -3.39727029e-02 -1.05807054e+00 -1.72491074e-01 3.98369998e-01 -1.16928613e+00 -1.28488767e+00 -8.88769209e-01 -1.81165084e-01 7.70551026e-01 -6.58533275e-01 9.89642084e-01 3.02944728e-03 -7.90340364e-01 4.59568173e-01 3.31620723e-01 -5.55431604e-01 -6.14427149e-01 -4.66897637e-01 3.93865943e-01 1.57608509e-01 -4.44331139e-01 -9.64544296e-01 -7.11956620e-01 2.19318140e-02 -8.39136839e-01 -2.80798793e-01 7.17579722e-02 1.05905092e+00 6.72740877e-01 -1.54287796e-02 3.18792850e-01 -8.78477871e-01 4.89888161e-01 -6.96723878e-01 -6.17197335e-01 -4.69037220e-02 -2.23640874e-01 2.11346626e-01 1.08546758e+00 -8.39069128e-01 -7.75333822e-01 9.31607336e-02 2.34259009e-01 -8.64657938e-01 -7.95200169e-02 4.64611501e-01 1.83220893e-01 3.67038488e-01 7.07606137e-01 2.65803367e-01 4.06398982e-01 -3.43208969e-01 1.88281104e-01 1.79353073e-01 9.05420661e-01 -8.03255379e-01 4.42971498e-01 5.98998666e-01 6.96040332e-01 -8.52297843e-01 -6.47596538e-01 -1.17895454e-01 -7.09102213e-01 1.51733067e-02 7.30541885e-01 -8.98888230e-01 -1.32787371e+00 3.70973170e-01 -8.64903927e-01 -6.80847704e-01 -7.74517715e-01 8.34297359e-01 -9.95240390e-01 2.16180570e-02 -9.68130171e-01 -1.10128355e+00 3.80775593e-02 -9.75657225e-01 1.01559234e+00 1.80357248e-01 -4.49070811e-01 -1.76686060e+00 3.97507310e-01 -2.17000604e-01 4.02865857e-01 8.67791951e-01 1.15711200e+00 -2.31082052e-01 -4.06907797e-01 -3.47392619e-01 3.51207942e-01 4.07237440e-01 2.67636895e-01 2.09727421e-01 -6.48429692e-01 -2.02446669e-01 3.02127212e-01 2.57813670e-02 5.08438230e-01 1.23538625e+00 9.92843509e-01 -3.52597624e-01 -4.49248999e-01 5.81750572e-01 1.21850073e+00 2.70690203e-01 1.54630870e-01 -6.10450923e-01 6.91193700e-01 3.71507823e-01 -3.04545820e-01 7.40693033e-01 1.91731364e-01 3.65816653e-01 3.89706075e-01 -2.15837777e-01 3.04479539e-01 -9.98340026e-02 3.65314811e-01 7.85714030e-01 -2.57730216e-01 -3.31946015e-01 -6.14847124e-01 3.83826643e-01 -1.76886117e+00 -9.84949589e-01 -2.25542322e-01 2.32871723e+00 8.63617599e-01 -1.99803337e-01 1.79006562e-01 -1.04993181e-02 3.68657082e-01 -2.23148793e-01 -1.36189806e+00 -3.22814554e-01 6.10520244e-02 4.05111760e-01 4.08028722e-01 7.50997543e-01 -8.25710237e-01 2.19548613e-01 7.29443502e+00 -3.84169370e-01 -1.17644966e+00 -2.40384147e-01 6.98553085e-01 -8.09946582e-02 6.64533582e-03 1.32539436e-01 -4.37412351e-01 5.24794400e-01 1.34533966e+00 -5.81272185e-01 6.89103723e-01 4.23034400e-01 5.75356722e-01 1.03755966e-01 -1.43391800e+00 6.73158467e-01 -4.20377791e-01 -1.03450739e+00 -4.61490691e-01 2.56397456e-01 6.23137295e-01 -2.28601411e-01 -6.46053925e-02 3.73830408e-01 4.28902805e-01 -1.07104015e+00 1.01067953e-01 9.62491751e-01 6.61248267e-01 -1.02825433e-01 5.54311991e-01 6.32139564e-01 -6.11996770e-01 -5.47305457e-02 -3.02632719e-01 -2.87324578e-01 2.80323535e-01 7.60908961e-01 -4.87242877e-01 6.31469255e-03 7.58355064e-03 8.03868055e-01 2.73194015e-01 7.03601599e-01 -8.81960616e-03 7.14915693e-01 -5.71225762e-01 1.59037054e-01 -1.33887246e-01 -4.80493098e-01 5.74431241e-01 3.98996025e-01 2.80189663e-01 5.05517602e-01 8.79022665e-03 1.28246367e+00 -9.85018089e-02 -2.89543003e-01 -5.32569289e-01 -3.44701290e-01 -2.86935158e-02 1.06842792e+00 -2.94630885e-01 -6.04086220e-01 -6.28672764e-02 5.86402535e-01 -1.59229860e-01 7.93046355e-01 -9.94512916e-01 1.87189072e-01 1.12208557e+00 1.69066533e-01 2.50427388e-02 -9.03474763e-02 1.19328961e-01 -1.48420405e+00 -1.16530888e-01 -6.21053994e-01 1.66602999e-01 -7.61419594e-01 -1.25799048e+00 2.87303001e-01 7.52458796e-02 -1.28425431e+00 -6.97880983e-01 -6.99255526e-01 -4.67968792e-01 1.06785369e+00 -1.22666609e+00 -5.36208391e-01 -1.49727091e-01 5.93660235e-01 4.18837331e-02 2.30495065e-01 1.00494766e+00 1.86720937e-02 -8.06557298e-01 6.22056685e-02 4.24102217e-01 1.70754343e-01 4.05781507e-01 -1.36340189e+00 1.46141723e-01 1.14158995e-01 -6.84652925e-02 6.37842238e-01 1.02432132e+00 -5.38163841e-01 -1.62994647e+00 -8.40912163e-01 3.66373062e-01 -7.60468841e-01 9.03489053e-01 -2.37668499e-01 -1.15682423e+00 8.95774841e-01 -4.30714786e-01 7.35033274e-01 4.55275983e-01 -2.59947956e-01 1.80912212e-01 1.62761673e-01 -1.06574738e+00 3.11374158e-01 5.69042027e-01 -4.51969862e-01 -3.91878217e-01 5.59801936e-01 3.85605454e-01 -9.43792701e-01 -1.27950692e+00 1.24813333e-01 9.12760675e-01 -5.23781478e-01 9.65854406e-01 -1.28250647e+00 6.11955047e-01 -1.54091105e-01 2.40720332e-01 -1.65445983e+00 -7.36894086e-02 -7.06491351e-01 -8.32610965e-01 6.17089152e-01 3.41710061e-01 -7.64401555e-01 8.77958059e-01 1.09874928e+00 1.89176604e-01 -9.09085274e-01 -7.20829785e-01 -8.15952241e-01 4.32236791e-01 -1.00112939e-02 4.42826748e-01 1.22903669e+00 6.54067770e-02 2.67188877e-01 -5.63750267e-01 2.27379516e-01 5.60876310e-01 -3.58205177e-02 6.02770984e-01 -1.38221943e+00 -8.10618162e-01 -7.46146366e-02 -4.65107560e-01 -9.15271938e-01 6.27936274e-02 -5.71231365e-01 9.04966816e-02 -1.10292983e+00 1.62891537e-01 -1.58831507e-01 -3.84736121e-01 2.62935668e-01 -2.38915339e-01 -3.36412281e-01 -2.91960333e-02 7.41816983e-02 2.31317297e-01 5.63485861e-01 1.58910096e+00 5.06227575e-02 -5.48522234e-01 8.34683552e-02 -4.26569492e-01 5.70278764e-01 5.43431878e-01 -5.78104079e-01 -4.96041983e-01 9.55886394e-03 -2.54693210e-01 1.07184577e+00 6.87239707e-01 -9.05387700e-01 3.13686170e-02 -3.77081454e-01 5.92723310e-01 -1.67822856e-02 5.36477208e-01 -8.16641331e-01 6.00028515e-01 6.35417759e-01 -2.87389606e-01 -7.75457323e-02 1.37602106e-01 5.91966510e-01 -1.43694252e-01 1.11659832e-01 7.01474190e-01 -2.81766295e-01 2.16027517e-02 6.98122978e-01 -4.78310615e-01 2.49069422e-01 8.51651192e-01 5.13795167e-02 -2.75987834e-01 -8.46181750e-01 -1.40823054e+00 3.03894907e-01 2.40057036e-01 1.45331129e-01 2.29475126e-01 -1.01411343e+00 -6.20811760e-01 3.22727025e-01 -2.67907172e-01 4.70953621e-03 4.37705696e-01 1.10535944e+00 -3.60929221e-01 2.93938279e-01 -1.31403595e-01 -6.12770200e-01 -7.84970403e-01 5.97204149e-01 9.06776011e-01 -1.09164111e-01 -5.58902204e-01 4.35395867e-01 5.65941215e-01 -3.70649397e-01 -1.93012580e-01 -6.30650282e-01 1.46964386e-01 -2.86701858e-01 2.07410827e-01 2.76325136e-01 -3.80367398e-01 -5.46745479e-01 -3.79843861e-02 6.14312291e-01 4.84495789e-01 2.05447614e-01 1.37812781e+00 -5.10311872e-02 1.13975003e-01 8.68708134e-01 1.31900859e+00 -6.10529125e-01 -1.51897621e+00 -6.76485375e-02 -5.88619351e-01 1.23905942e-01 1.38634741e-01 -6.51709497e-01 -1.33169878e+00 1.07361960e+00 3.41183335e-01 3.70871156e-01 7.75736570e-01 -1.50190949e-01 7.49713302e-01 2.62821287e-01 9.02380273e-02 -5.79042614e-01 -2.78123289e-01 2.37306893e-01 6.76562250e-01 -1.04317987e+00 1.69153214e-01 -1.87339813e-01 -3.75467122e-01 1.19159806e+00 2.50579238e-01 -3.79253566e-01 9.79961276e-01 5.52939713e-01 2.02381611e-01 -1.24444135e-01 -1.05704761e+00 2.19227120e-01 2.29957104e-01 3.45588744e-01 4.10824239e-01 5.84956966e-02 2.21234020e-02 5.20731091e-01 -5.77059127e-02 3.86248797e-01 9.16663408e-01 7.63457060e-01 -1.65384077e-02 -8.69663775e-01 -3.40345323e-01 6.96577072e-01 -3.55979621e-01 2.59241521e-01 -6.18832856e-02 9.37439322e-01 -5.58950529e-02 4.04371053e-01 1.36048883e-01 1.99728161e-01 3.75726968e-01 2.59487242e-01 4.77478802e-01 -4.77812350e-01 -5.54818735e-02 -8.00403431e-02 -4.75626260e-01 -4.28196460e-01 -1.78974628e-01 -8.95312369e-01 -1.42030442e+00 -3.84794772e-01 -2.85904199e-01 -1.45161245e-02 6.33316338e-01 1.00374973e+00 1.78717062e-01 8.53465676e-01 5.72193980e-01 -6.82536960e-01 -7.95139730e-01 -7.70671487e-01 -9.36821163e-01 5.72931707e-01 1.21481180e+00 -7.78743505e-01 -7.56594181e-01 7.26180315e-01]
[6.534353256225586, 3.57250714302063]
632616b7-ac3b-4094-ba5d-517940f99090
dynamic-graph-attention-for-anomaly-detection
2307.03761
null
https://arxiv.org/abs/2307.03761v1
https://arxiv.org/pdf/2307.03761v1.pdf
Dynamic Graph Attention for Anomaly Detection in Heterogeneous Sensor Networks
In the era of digital transformation, systems monitored by the Industrial Internet of Things (IIoTs) generate large amounts of Multivariate Time Series (MTS) data through heterogeneous sensor networks. While this data facilitates condition monitoring and anomaly detection, the increasing complexity and interdependencies within the sensor network pose significant challenges for anomaly detection. Despite progress in this field, much of the focus has been on point anomalies and contextual anomalies, with lesser attention paid to collective anomalies. A less addressed but common variant of collective anomalies is when the abnormal collective behavior is caused by shifts in interrelationships within the system. This can be due to abnormal environmental conditions like overheating, improper operational settings resulting from cyber-physical attacks, or system-level faults. To address these challenges, this paper proposes DyGATAD (Dynamic Graph Attention for Anomaly Detection), a graph-based anomaly detection framework that leverages the attention mechanism to construct a continuous graph representation of multivariate time series by inferring dynamic edges between time series. DyGATAD incorporates an operating condition-aware reconstruction combined with a topology-based anomaly score, thereby enhancing the detection ability of relationship shifts. We evaluate the performance of DyGATAD using both a synthetic dataset with controlled varying fault severity levels and an industrial-scale multiphase flow facility benchmark featuring various fault types with different detection difficulties. Our proposed approach demonstrated superior performance in collective anomaly detection for sensor networks, showing particular strength in early-stage fault detection, even in the case of faults with minimal severity.
['Olga Fink', 'Mengjie Zhao']
2023-07-07
null
null
null
null
['graph-attention', 'anomaly-detection', 'fault-detection']
['graphs', 'methodology', 'miscellaneous']
[ 3.91518831e-01 4.45751697e-02 1.77685112e-01 1.17321447e-01 -5.57114147e-02 -5.61815083e-01 3.51915807e-01 9.26114321e-01 4.61647034e-01 4.79513764e-01 -9.96495187e-02 -3.54702383e-01 -7.49522030e-01 -9.31004107e-01 -5.02201200e-01 -5.51747799e-01 -7.62950361e-01 3.23050797e-01 3.15269738e-01 -2.89125800e-01 2.77026445e-01 7.73410976e-01 -1.48391962e+00 -1.18740603e-01 7.62261569e-01 1.10177946e+00 -2.97908455e-01 5.76177001e-01 1.10522538e-01 5.55480421e-01 -1.09311080e+00 3.37584853e-01 2.85104603e-01 -2.19548240e-01 -5.97888291e-01 5.37675142e-01 -1.71067789e-01 1.08383104e-01 -3.30764562e-01 1.00731552e+00 2.42859527e-01 8.60159844e-02 2.55663991e-01 -1.77343261e+00 -1.31050751e-01 4.13497984e-01 -5.82658291e-01 7.76679158e-01 5.10339677e-01 4.82357085e-01 8.51794004e-01 -5.04974544e-01 3.17098707e-01 1.00212204e+00 5.87447405e-01 -2.08735690e-01 -1.13518596e+00 -3.19030821e-01 4.39894468e-01 3.64579678e-01 -1.07615638e+00 6.10095784e-02 1.06293952e+00 -3.45678717e-01 1.23405695e+00 2.37660795e-01 6.07259393e-01 9.01042283e-01 8.64338398e-01 -5.00242114e-02 4.95756984e-01 -8.30750689e-02 5.26021421e-01 -8.69805932e-01 1.17917761e-01 2.16514632e-01 6.68859839e-01 -1.07511699e-01 -2.81494409e-01 -2.70986944e-01 4.92388099e-01 6.19624853e-01 -1.76645309e-01 -1.02589555e-01 -1.17535186e+00 3.52625787e-01 4.49937224e-01 4.94550258e-01 -8.92203867e-01 6.96232468e-02 7.46001005e-01 9.12742913e-01 3.72748405e-01 6.56875074e-01 -6.91479683e-01 -1.38841093e-01 -1.93366855e-01 -2.19050601e-01 8.10455024e-01 7.44401753e-01 5.51252425e-01 7.42959261e-01 1.27806708e-01 3.50090414e-01 2.54988700e-01 1.74062997e-01 3.31846476e-01 -3.80067915e-01 4.78520244e-01 1.22862792e+00 -3.40703368e-01 -1.36431098e+00 -6.35726511e-01 -6.35945976e-01 -9.20235515e-01 3.21028531e-01 1.89058647e-01 -1.22262977e-01 -8.19231689e-01 1.45414293e+00 4.16606903e-01 4.06968057e-01 -6.67711422e-02 4.89152014e-01 -1.57622561e-01 4.66375113e-01 -1.56960130e-01 -3.33902448e-01 1.07953751e+00 -2.38564536e-01 -9.02750552e-01 -8.46710280e-02 7.20084548e-01 -4.58251357e-01 5.66305876e-01 4.22371835e-01 -5.53265333e-01 -1.78638056e-01 -1.16282487e+00 9.20812190e-01 -6.17716670e-01 -8.14913929e-01 1.96183577e-01 3.51691455e-01 -6.39573038e-01 7.07490921e-01 -8.84581506e-01 -6.17359042e-01 2.75851846e-01 3.54175508e-01 -2.96275318e-01 -2.31521100e-01 -9.99104321e-01 6.33947968e-01 3.58635217e-01 1.89186484e-01 -8.89241159e-01 -7.40034699e-01 -7.80516922e-01 -2.59236917e-02 9.79648530e-01 -2.81880915e-01 6.93506598e-01 -5.48844576e-01 -8.18804026e-01 -1.84949681e-01 3.00908923e-01 -5.66343307e-01 2.46883869e-01 -1.29435789e-02 -1.33122766e+00 -4.61599976e-02 5.33735193e-02 -6.04209721e-01 7.49782383e-01 -9.56770718e-01 -4.07461077e-01 -6.11400306e-01 -8.27048719e-02 -2.96270758e-01 -1.76926345e-01 -2.35793769e-01 3.90100390e-01 -6.59330547e-01 3.90655994e-01 -7.38755703e-01 -4.17743266e-01 -3.49289984e-01 -7.37196267e-01 -2.29985788e-01 1.71621990e+00 -3.99374396e-01 1.42322636e+00 -2.05543160e+00 -6.85312822e-02 7.47366309e-01 2.88903058e-01 6.50693476e-02 4.53475751e-02 1.28103220e+00 -4.91074413e-01 2.32422408e-02 -4.78326410e-01 2.11850122e-01 -2.43261486e-01 5.82620919e-01 -2.80446172e-01 6.27482414e-01 5.44050217e-01 5.14523447e-01 -1.02622604e+00 2.28603363e-01 4.52778935e-01 1.84280500e-02 -2.70164788e-01 5.64701855e-02 -1.69867992e-01 6.09817326e-01 -6.42073333e-01 1.05367184e+00 3.15289140e-01 -3.12663943e-01 1.63287789e-01 -1.76566198e-01 -1.83276627e-02 -1.51861057e-01 -1.38147008e+00 1.19046950e+00 -2.48320356e-01 3.14210266e-01 1.38928548e-01 -1.28138828e+00 1.00204647e+00 5.36751390e-01 1.14620793e+00 -6.23031199e-01 7.25488085e-03 1.89234465e-01 2.71265864e-01 -6.23684585e-01 1.76559001e-01 3.92158568e-01 -2.24279016e-01 6.20521367e-01 -1.06347606e-01 -9.32197925e-03 2.83829302e-01 1.04835443e-01 2.06682920e+00 -5.12945056e-01 4.65324193e-01 -4.00163323e-01 4.74684209e-01 3.74326594e-02 6.96391404e-01 3.72736067e-01 -2.16363981e-01 1.29929528e-01 7.33744025e-01 -6.15867734e-01 -7.90596962e-01 -1.10347712e+00 1.60331890e-01 4.67255652e-01 1.83117181e-01 -5.70427537e-01 -9.68511999e-02 -9.57326233e-01 4.38467503e-01 5.86132228e-01 -4.91712034e-01 -6.47372007e-01 -5.34657061e-01 -7.31398106e-01 2.89754033e-01 4.15278405e-01 1.96331158e-01 -1.10806608e+00 -5.06078243e-01 7.28295386e-01 3.76972765e-01 -1.04457974e+00 -4.93239611e-02 5.41427791e-01 -9.59242344e-01 -1.70436895e+00 3.98309112e-01 -3.11095715e-01 8.43884587e-01 -1.77511722e-02 1.07296824e+00 1.92386582e-01 -6.52359903e-01 7.30783701e-01 -3.22669476e-01 -4.66814071e-01 -5.58433056e-01 -3.75787854e-01 3.69269371e-01 2.15675488e-01 1.14572778e-01 -1.02395320e+00 -5.06984174e-01 6.06038570e-01 -1.41959202e+00 -8.80134046e-01 3.44075620e-01 5.95416129e-01 3.79423738e-01 7.57378101e-01 1.19058239e+00 -7.79341817e-01 7.68627822e-01 -1.18821716e+00 -5.59181035e-01 -1.11767285e-01 -1.03775334e+00 -2.21508667e-01 1.02507913e+00 -3.15636337e-01 -2.60599315e-01 -4.06390399e-01 2.72341102e-01 -5.91998637e-01 -4.77284253e-01 9.05483603e-01 -2.25649789e-01 2.30929423e-02 5.17212808e-01 -1.37784660e-01 1.38168320e-01 -2.36210272e-01 -2.50013858e-01 2.68961251e-01 3.71617377e-01 -3.98823470e-01 1.00545108e+00 2.80007511e-01 4.25132662e-01 -9.11293745e-01 -3.00217599e-01 -3.90278369e-01 -2.83648074e-01 -4.84949142e-01 4.15161550e-01 -5.28308511e-01 -6.65917337e-01 5.05172551e-01 -6.89853728e-01 -5.09353317e-02 -6.44950747e-01 1.30213261e-01 -1.05879912e-02 4.32892144e-01 -5.99630117e-01 -6.60309196e-01 -1.02166153e-01 -9.00566399e-01 8.40326726e-01 -2.60964483e-01 -4.06294256e-01 -1.38771939e+00 -3.31732119e-03 -1.67392328e-01 6.67910218e-01 1.08227074e+00 9.98018265e-01 -1.08714283e+00 -5.93479395e-01 -7.37246811e-01 2.20229313e-01 2.47873694e-01 9.44563270e-01 6.21348061e-02 -4.70316201e-01 -6.45445585e-01 6.18869625e-03 3.30085963e-01 -1.83764398e-02 -5.40120564e-02 1.09793687e+00 -3.30318660e-01 -3.40802759e-01 1.18298419e-01 1.51451218e+00 4.84654844e-01 4.18823034e-01 2.48699337e-01 9.10050154e-01 4.68383074e-01 4.90627944e-01 8.06200862e-01 4.60410975e-02 4.11545575e-01 1.11701286e+00 1.17307991e-01 2.71210223e-01 1.21183686e-01 3.63362372e-01 9.33630466e-01 1.04505710e-01 -5.50343454e-01 -1.14326298e+00 5.55321872e-01 -1.76515770e+00 -8.93918633e-01 -2.77562797e-01 2.03984284e+00 -7.17241392e-02 3.05973828e-01 -1.13787809e-02 9.03145552e-01 9.26965237e-01 2.12143734e-01 -1.04556584e+00 -3.62416089e-01 -9.64784771e-02 -8.87217969e-02 3.55304062e-01 -2.43387930e-02 -6.54636741e-01 5.56025393e-02 5.64063358e+00 1.30472049e-01 -1.03640378e+00 -2.49025732e-01 2.45808616e-01 2.41543725e-01 -4.33560759e-02 -1.42437965e-01 -9.74353682e-03 5.79049408e-01 1.16458797e+00 -3.69894058e-01 2.38141224e-01 6.35642946e-01 4.30468559e-01 2.26686776e-01 -1.01248884e+00 5.00999272e-01 -2.15575814e-01 -8.39435041e-01 -1.78676650e-01 3.84198397e-01 7.06202388e-01 2.24925920e-01 -2.53547311e-01 -3.79953943e-02 2.61085242e-01 -7.26044774e-01 2.22929537e-01 2.39058971e-01 2.76591867e-01 -6.90595210e-01 8.07111323e-01 -3.24568041e-02 -1.56750476e+00 -6.09160066e-01 3.69477719e-01 -2.63863862e-01 4.23273861e-01 1.13763928e+00 -9.99615073e-01 1.08650792e+00 6.44670010e-01 1.07110631e+00 -4.19826776e-01 9.92533386e-01 3.52773778e-02 7.72074759e-01 -4.98423159e-01 4.41721201e-01 -3.61027720e-04 -3.57154831e-02 1.24821281e+00 6.29510581e-01 4.67087805e-01 -3.84812564e-01 5.06243765e-01 5.50073624e-01 2.67398745e-01 -4.21887368e-01 -1.08189690e+00 -2.49072641e-01 6.32843792e-01 1.24523723e+00 -1.04489827e+00 9.24202576e-02 -3.74303967e-01 5.82685232e-01 -2.85230815e-01 5.78423083e-01 -6.15675747e-01 -3.43159020e-01 1.02170372e+00 2.92079479e-01 2.11677760e-01 -4.05275196e-01 -1.85319707e-01 -6.94747806e-01 2.67088950e-01 -8.22568893e-01 8.60174477e-01 -3.93449336e-01 -1.66584468e+00 5.61653435e-01 -1.09971389e-01 -1.66838384e+00 -3.93213511e-01 -3.80419672e-01 -1.08576858e+00 2.78406978e-01 -1.11442816e+00 -5.78768015e-01 -5.15468776e-01 8.75283420e-01 6.20942354e-01 -1.02982573e-01 7.25069940e-01 3.34607005e-01 -8.89983594e-01 6.93786889e-02 -2.22295329e-01 -1.30559579e-01 2.26985320e-01 -1.30973089e+00 5.10996282e-01 1.29019201e+00 -3.57620537e-01 1.98992252e-01 9.49857712e-01 -1.02978826e+00 -1.84859109e+00 -1.46714842e+00 1.02847628e-01 -1.71537772e-01 1.32108891e+00 -1.80962041e-01 -1.19190300e+00 7.35109270e-01 2.22256910e-02 5.67498982e-01 3.42306286e-01 -2.04983413e-01 -8.74119326e-02 -2.57934302e-01 -1.32009721e+00 3.33128661e-01 1.16731989e+00 -3.27661604e-01 -3.73913974e-01 4.00894940e-01 7.33090281e-01 -5.51091656e-02 -1.30911040e+00 6.80597126e-01 -2.65319884e-01 -7.01227963e-01 6.51535332e-01 -4.82496351e-01 7.06425235e-02 -7.35532045e-01 -1.95854291e-01 -1.66106772e+00 -1.90217972e-01 -7.46142685e-01 -4.90454197e-01 1.14866245e+00 1.50019169e-01 -1.40475261e+00 3.91580909e-01 1.86977863e-01 -7.68989742e-01 -6.74749792e-01 -1.16585290e+00 -9.67492700e-01 -6.50243342e-01 -5.50715208e-01 1.02205098e+00 1.23589420e+00 2.21593529e-01 1.98504776e-01 9.34884846e-02 6.34877443e-01 5.26768088e-01 -1.45538792e-01 4.65100557e-01 -1.56186402e+00 -5.96282519e-02 -2.82771796e-01 -1.15491068e+00 6.56434223e-02 -1.81215435e-01 -5.64000845e-01 -1.48608252e-01 -1.37126327e+00 -8.39458108e-01 -2.32184708e-01 -7.02557027e-01 4.89651024e-01 1.54144645e-01 -1.21826708e-01 -1.30007446e-01 -2.65265871e-02 -5.44887364e-01 5.04257739e-01 8.85739326e-01 -1.66098297e-01 -8.65398794e-02 4.68980782e-02 -1.78253502e-01 4.18404847e-01 9.37961221e-01 -4.72497344e-01 -5.51035345e-01 -5.04987389e-02 3.18203181e-01 3.06430846e-01 3.38790148e-01 -1.51588356e+00 2.33818665e-01 -1.29742220e-01 -3.23367119e-02 -4.90491122e-01 -8.92901495e-02 -1.54244769e+00 6.23218775e-01 7.10200429e-01 2.53609210e-01 1.16330993e+00 3.00260633e-01 1.26725602e+00 -4.31968987e-01 5.73816717e-01 1.66220918e-01 3.50669920e-01 -9.07471299e-01 5.36859453e-01 -6.21677577e-01 -1.73112601e-01 1.43766999e+00 -3.74882579e-01 -6.12972677e-01 -2.09819153e-01 -7.27628350e-01 3.43125671e-01 4.27488059e-01 7.93145597e-01 5.87596893e-01 -1.19236243e+00 -4.62495208e-01 5.39067090e-01 4.93389070e-01 9.17446911e-02 2.14624077e-01 1.09798229e+00 -1.87997371e-01 -5.59417345e-02 -2.01834902e-01 -8.10527503e-01 -8.11765254e-01 5.78984976e-01 3.07215333e-01 -2.96826422e-01 -9.82376516e-01 6.83772564e-02 -3.26121360e-01 -1.49968043e-01 -1.59479186e-01 -5.12800753e-01 -2.53480617e-02 -4.24934961e-02 2.73348898e-01 8.12125266e-01 6.28645182e-01 -3.91245633e-01 -5.08257985e-01 2.55245119e-01 2.57134587e-01 5.62566340e-01 1.27884054e+00 -2.02148110e-01 -2.52526343e-01 8.01540375e-01 7.97681272e-01 -2.35628098e-01 -1.06371462e+00 -3.06464601e-02 3.81734967e-01 -3.36372524e-01 -8.35515112e-02 -7.34014869e-01 -1.42677593e+00 1.45476878e-01 4.81753349e-01 1.29325640e+00 1.27861524e+00 -7.37721920e-02 7.38538027e-01 3.07737261e-01 6.37886882e-01 -7.73402154e-01 3.56051207e-01 5.52638531e-01 6.48951948e-01 -8.75449002e-01 -6.63119629e-02 -3.87318194e-01 -2.36102700e-01 1.11265182e+00 6.39066815e-01 -3.33279222e-01 9.36269939e-01 4.38454270e-01 -1.34613693e-01 -7.41515815e-01 -8.30206752e-01 2.16575086e-01 3.31101492e-02 6.27740085e-01 -8.26940015e-02 -1.94300134e-02 2.01818928e-01 3.02003101e-02 1.40002638e-01 -4.37057048e-01 7.12311566e-01 1.42364156e+00 -2.75660247e-01 -7.91590750e-01 -4.86659080e-01 9.72030222e-01 -4.10903722e-01 5.07389247e-01 -1.16557442e-01 7.51667202e-01 -1.50029778e-01 1.34937882e+00 2.96998203e-01 -6.58600867e-01 7.60220826e-01 1.96176302e-02 -2.51023918e-01 -5.90471327e-01 -6.95288002e-01 -4.05964166e-01 5.91182522e-02 -1.04592371e+00 -1.25801802e-01 -6.74974322e-01 -1.29030704e+00 -2.60907859e-01 -3.96764427e-01 1.04944736e-01 6.50076866e-01 1.07363081e+00 8.36289346e-01 1.37207210e+00 8.48945618e-01 -5.57263374e-01 -3.14728558e-01 -8.40133727e-01 -9.14685547e-01 6.99981749e-01 6.60626471e-01 -8.93222570e-01 -7.96053767e-01 -4.01593089e-01]
[7.250041961669922, 2.871300220489502]
617abf86-c6c5-40bf-aab8-b90d713a6d61
unsupervised-cross-lingual-representation-1
1911.02116
null
https://arxiv.org/abs/1911.02116v2
https://arxiv.org/pdf/1911.02116v2.pdf
Unsupervised Cross-lingual Representation Learning at Scale
This paper shows that pretraining multilingual language models at scale leads to significant performance gains for a wide range of cross-lingual transfer tasks. We train a Transformer-based masked language model on one hundred languages, using more than two terabytes of filtered CommonCrawl data. Our model, dubbed XLM-R, significantly outperforms multilingual BERT (mBERT) on a variety of cross-lingual benchmarks, including +14.6% average accuracy on XNLI, +13% average F1 score on MLQA, and +2.4% F1 score on NER. XLM-R performs particularly well on low-resource languages, improving 15.7% in XNLI accuracy for Swahili and 11.4% for Urdu over previous XLM models. We also present a detailed empirical analysis of the key factors that are required to achieve these gains, including the trade-offs between (1) positive transfer and capacity dilution and (2) the performance of high and low resource languages at scale. Finally, we show, for the first time, the possibility of multilingual modeling without sacrificing per-language performance; XLM-R is very competitive with strong monolingual models on the GLUE and XNLI benchmarks. We will make our code, data and models publicly available.
['Francisco Guzmán', 'Vishrav Chaudhary', 'Kartikay Khandelwal', 'Alexis Conneau', 'Myle Ott', 'Luke Zettlemoyer', 'Guillaume Wenzek', 'Edouard Grave', 'Veselin Stoyanov', 'Naman Goyal']
2019-11-05
unsupervised-cross-lingual-representation-2
https://aclanthology.org/2020.acl-main.747
https://aclanthology.org/2020.acl-main.747.pdf
acl-2020-6
['multilingual-nlp']
['natural-language-processing']
[-6.58531010e-01 -2.12673113e-01 -4.63324040e-01 -3.26200753e-01 -1.80716300e+00 -8.27063143e-01 6.85899913e-01 -4.28233622e-03 -7.90341973e-01 9.56023276e-01 2.42874965e-01 -8.24525714e-01 3.67789775e-01 -6.42308116e-01 -1.07303822e+00 -1.85143173e-01 -1.89560309e-01 5.99725008e-01 9.72770900e-02 -5.55073917e-01 -2.56646097e-01 2.59471595e-01 -9.04848874e-01 3.48345131e-01 1.16286385e+00 5.32154143e-01 2.88740247e-01 4.24269885e-01 -2.18025893e-01 8.60337973e-01 -3.89633566e-01 -7.23899662e-01 2.90567517e-01 -5.89342080e-02 -7.68382728e-01 -4.82632786e-01 6.39367759e-01 -8.31483454e-02 -1.21401787e-01 9.53231990e-01 4.39266801e-01 -2.43975043e-01 4.13242370e-01 -9.08436120e-01 -9.15316999e-01 1.24616909e+00 -6.86066687e-01 3.10655326e-01 1.28471538e-01 1.83425561e-01 1.14700782e+00 -1.29826415e+00 5.44105291e-01 1.39811945e+00 9.00271833e-01 2.77747303e-01 -1.21787429e+00 -9.94418442e-01 9.13990960e-02 -1.91411749e-01 -1.64537895e+00 -7.30450928e-01 -1.69593506e-02 -3.79375577e-01 1.47121346e+00 1.21574504e-02 7.37461522e-02 7.28663802e-01 4.15313303e-01 8.26075912e-01 1.54850245e+00 -6.15594566e-01 -3.38429660e-01 5.47099411e-01 7.63556361e-02 7.84435987e-01 8.46031755e-02 -3.21952738e-02 -7.23307192e-01 -5.09129167e-02 4.75573897e-01 -5.35730720e-01 -1.88055709e-01 1.48009002e-01 -1.33243608e+00 8.87809157e-01 3.73230875e-01 7.47903824e-01 -1.38893485e-01 -1.21676400e-02 4.82095718e-01 8.19732010e-01 7.65334547e-01 4.16158408e-01 -9.58373249e-01 -4.85161841e-02 -9.89064753e-01 -1.02982707e-01 5.95337927e-01 1.03570044e+00 9.68306303e-01 2.79821843e-01 3.33232611e-01 1.00622869e+00 1.60796613e-01 1.01328731e+00 4.87875998e-01 -5.75285554e-01 1.02013242e+00 2.03728348e-01 -1.25845835e-01 -3.38205785e-01 -2.59462684e-01 -5.20370424e-01 -6.04744494e-01 -7.70657510e-02 4.83154058e-01 -2.89999664e-01 -7.53775835e-01 2.05597639e+00 -1.54097274e-01 -3.61125112e-01 4.15943176e-01 4.66479570e-01 4.92000312e-01 1.03336191e+00 2.07710221e-01 2.72242329e-03 1.32440317e+00 -9.52958584e-01 -4.49690402e-01 -3.75001818e-01 1.18121696e+00 -1.12988544e+00 1.43929291e+00 1.25990301e-01 -1.35648406e+00 -5.21432877e-01 -7.93683052e-01 -2.16472730e-01 -5.52503526e-01 3.10010374e-01 6.31431401e-01 6.54336333e-01 -1.39898050e+00 1.42865226e-01 -8.11573207e-01 -6.51585817e-01 -1.03940694e-02 3.27132225e-01 -6.32809579e-01 -4.16007966e-01 -1.38025808e+00 1.15586245e+00 2.36028567e-01 -1.47334754e-01 -8.72033954e-01 -1.02974355e+00 -8.48565578e-01 -1.58060759e-01 -5.63818663e-02 -7.85016641e-02 1.21189642e+00 -6.91977441e-01 -1.02116752e+00 1.08339536e+00 -1.72567442e-01 -5.32675505e-01 4.94565040e-01 -4.96023178e-01 -6.89686239e-01 -3.59897017e-01 5.70784807e-01 7.23611832e-01 1.33385491e-02 -9.50670540e-01 -8.80802095e-01 -1.74967349e-01 -8.28220025e-02 2.96218574e-01 -4.56671596e-01 5.04598618e-01 -4.88860339e-01 -4.92507130e-01 -3.62022698e-01 -7.86588609e-01 -6.72759563e-02 -7.04226434e-01 -7.16044754e-02 -2.48603851e-01 4.23625320e-01 -1.14786351e+00 1.05437756e+00 -2.03408957e+00 -3.61992046e-02 -3.05259954e-02 -1.78099573e-01 3.99266928e-01 -5.10163009e-01 5.29060662e-01 1.72137350e-01 3.80656302e-01 5.40916622e-03 -3.21058303e-01 -1.62611064e-02 1.24241978e-01 -3.17575544e-01 3.04100931e-01 1.47739932e-01 9.67986286e-01 -7.13037789e-01 -2.71079808e-01 -2.99810921e-03 5.96626937e-01 -5.42236745e-01 9.37854499e-02 -8.61520972e-03 3.67871583e-01 1.26400411e-01 6.24430597e-01 5.30283391e-01 5.89483380e-02 2.92206138e-01 -5.18528046e-03 -5.32943606e-01 8.04481506e-01 -6.73564255e-01 1.75013077e+00 -1.08878577e+00 5.32820106e-01 2.67654121e-01 -3.86250317e-01 8.33151340e-01 4.85454530e-01 9.16756839e-02 -1.12817264e+00 -2.87121654e-01 9.03280199e-01 6.77091703e-02 1.54202625e-01 6.07861698e-01 -2.37637550e-01 -4.38230604e-01 5.18403232e-01 3.89475316e-01 1.33557588e-01 3.38587761e-01 3.42552692e-01 6.76479757e-01 -6.39476329e-02 1.30070254e-01 -9.45534348e-01 4.63064104e-01 2.15759389e-02 5.65779448e-01 7.10483789e-01 2.10884511e-02 -1.25948787e-02 1.93663672e-01 -2.38755867e-01 -1.07222867e+00 -1.19636691e+00 -2.15249181e-01 1.50526917e+00 -4.85022575e-01 -4.62976843e-01 -5.86775661e-01 -6.08489573e-01 -2.63950992e-02 7.92695880e-01 -1.89118624e-01 2.24986047e-01 -9.39686239e-01 -1.05829442e+00 8.68138552e-01 5.03820002e-01 5.10518491e-01 -7.84984648e-01 2.59966284e-01 2.21345916e-01 -3.23022902e-01 -1.21870661e+00 -6.24841988e-01 3.36412966e-01 -6.46181464e-01 -5.68327725e-01 -5.86144567e-01 -9.85976994e-01 2.64350355e-01 2.26232246e-01 1.54307771e+00 -9.16911513e-02 1.97421134e-01 8.71503577e-02 -6.63474426e-02 -1.93801194e-01 -7.24054635e-01 7.79552221e-01 3.50780845e-01 -3.60354394e-01 4.81641293e-01 -3.55980098e-01 -1.14873489e-02 2.30236754e-01 -3.51728529e-01 3.24953981e-02 7.03715265e-01 6.66071296e-01 4.86179739e-01 -1.98965594e-01 7.06965744e-01 -1.02268779e+00 2.81464249e-01 -5.26084483e-01 -7.86989808e-01 5.08493066e-01 -6.88754141e-01 4.38742638e-02 7.69533515e-01 -1.50023103e-01 -1.01657367e+00 -3.38818073e-01 -4.08015907e-01 5.73948026e-02 2.27277666e-01 6.52093947e-01 -2.05273584e-01 5.56702679e-03 5.94531000e-01 1.14912629e-01 -4.50298160e-01 -7.50366211e-01 6.20528579e-01 8.00697207e-01 4.55085933e-01 -8.83082509e-01 7.28240848e-01 9.72876400e-02 -7.34891951e-01 -6.89439118e-01 -8.45384896e-01 -3.47864538e-01 -7.79537857e-01 2.47107804e-01 8.78296018e-01 -1.63791418e+00 -1.21646382e-01 3.79656494e-01 -1.00226510e+00 -7.54344285e-01 2.60059461e-02 6.53904557e-01 -2.03168765e-01 -6.83237538e-02 -1.30201805e+00 -5.16009569e-01 -5.79457939e-01 -1.26734805e+00 8.92070472e-01 -2.44120434e-01 -4.01649475e-02 -1.26405811e+00 1.19838759e-01 3.60823721e-01 5.51909208e-01 -3.78626317e-01 1.15332115e+00 -5.74831605e-01 -6.02668405e-01 8.95407870e-02 -3.39157164e-01 4.12423581e-01 6.21828772e-02 -3.05637658e-01 -8.46197844e-01 -7.22011626e-01 -3.77572596e-01 -6.53138936e-01 6.56739295e-01 1.16000660e-01 2.53658265e-01 -2.22045615e-01 -1.82596773e-01 6.03835344e-01 1.65380979e+00 1.49703538e-02 2.47369707e-01 4.24734652e-01 8.34897399e-01 3.90842944e-01 4.59166676e-01 -1.91299900e-01 7.80090868e-01 6.15563929e-01 -2.07241341e-01 -3.54519099e-01 -3.06196660e-01 -4.54071492e-01 8.69011998e-01 1.75020397e+00 2.12177500e-01 3.06183528e-02 -1.34851408e+00 8.09202969e-01 -1.55045915e+00 -5.14857292e-01 -1.03426509e-01 2.20839357e+00 1.16463888e+00 1.44827157e-01 1.48912847e-01 -4.05528039e-01 4.75887984e-01 -1.02604955e-01 -1.08835995e-01 -5.52978754e-01 -5.21907151e-01 3.77239317e-01 8.20166230e-01 1.14514256e+00 -8.89674067e-01 1.59595704e+00 6.70564747e+00 7.77633965e-01 -1.24232757e+00 6.19227409e-01 6.72983825e-01 -5.60625270e-02 -3.67369324e-01 6.86124340e-03 -1.40044963e+00 2.65825391e-01 1.54740655e+00 -3.13270330e-01 6.60096049e-01 6.10925555e-01 -1.77803468e-02 2.34263703e-01 -1.01430881e+00 6.15749359e-01 2.31092405e-02 -1.13678825e+00 2.29478344e-01 2.07217395e-01 9.41694856e-01 1.09329891e+00 8.99084061e-02 7.48424590e-01 9.47359741e-01 -1.10285997e+00 8.51956069e-01 -1.38819246e-02 1.18321300e+00 -9.64339137e-01 6.46472037e-01 3.63074273e-01 -1.33283865e+00 1.90390974e-01 -3.58756959e-01 1.24960594e-01 7.99452141e-02 4.21402335e-01 -5.51026642e-01 6.43135965e-01 8.70644867e-01 5.67673862e-01 -5.87282658e-01 3.90201598e-01 -1.68113798e-01 8.07758212e-01 -3.93264532e-01 4.60556209e-01 5.62213182e-01 -3.84078808e-02 5.88843673e-02 1.74846399e+00 4.24027622e-01 -4.63839293e-01 3.04967195e-01 3.61908019e-01 -4.59578782e-01 5.69501996e-01 -6.41650438e-01 7.30898753e-02 5.98732948e-01 1.13367820e+00 -1.86818346e-01 -4.23225313e-01 -9.12359893e-01 7.31084645e-01 7.40232766e-01 2.77167410e-01 -7.15954721e-01 -2.97898680e-01 6.73412204e-01 1.07138686e-01 9.76720229e-02 -4.54774618e-01 -7.53820762e-02 -1.50886977e+00 -1.33256957e-01 -1.34665847e+00 5.12538731e-01 -5.73284626e-01 -1.32168674e+00 9.51085627e-01 -2.95973241e-01 -6.58185661e-01 -3.34324032e-01 -7.30255842e-01 -1.48077533e-01 1.48374557e+00 -1.88574338e+00 -1.53383577e+00 4.55494016e-01 7.06047058e-01 4.72719759e-01 -5.59004843e-01 1.00823927e+00 9.15849388e-01 -4.02794033e-01 9.25773680e-01 2.72918761e-01 4.36193824e-01 9.79795575e-01 -1.27374661e+00 8.50780487e-01 1.11966133e+00 5.08004785e-01 7.40683496e-01 2.69704700e-01 -4.77047831e-01 -1.31333518e+00 -1.33795869e+00 1.70618498e+00 -7.31472969e-01 1.08167505e+00 -7.14902759e-01 -9.28560734e-01 1.36397076e+00 5.53643882e-01 -2.58054435e-01 6.94838166e-01 6.09318078e-01 -7.95194507e-01 -2.82028943e-01 -9.02666032e-01 6.26403213e-01 7.50111222e-01 -9.46183324e-01 -3.15971196e-01 4.87576157e-01 7.98113704e-01 -1.50625259e-01 -1.34676373e+00 3.11014444e-01 3.55693549e-01 -6.93653405e-01 7.64161766e-01 -5.79149961e-01 1.14247128e-01 -5.67664094e-02 -7.24545777e-01 -1.52763963e+00 -4.09677565e-01 -5.82375646e-01 4.60275650e-01 1.48764491e+00 9.86653626e-01 -7.99232721e-01 1.72736764e-01 1.54026449e-01 -8.19019377e-02 -4.84941006e-01 -8.84873450e-01 -1.00071990e+00 9.63459432e-01 -4.88159984e-01 5.04895687e-01 1.29277956e+00 -2.32968941e-01 7.24727929e-01 -4.34968919e-01 2.00306550e-01 5.28307617e-01 -8.98615271e-02 7.52669811e-01 -8.47878575e-01 -2.78409928e-01 -2.80378222e-01 7.67092779e-02 -9.28494453e-01 3.01270604e-01 -1.42003107e+00 -2.06386700e-01 -1.29527748e+00 2.30947256e-01 -8.25975180e-01 -3.70722443e-01 7.48380005e-01 -6.07847199e-02 5.11608005e-01 3.24620426e-01 3.51870775e-01 -4.26185757e-01 1.66368619e-01 6.99842989e-01 8.07209685e-02 -3.93294878e-02 -4.70848441e-01 -8.67263675e-01 5.57064712e-01 7.31878221e-01 -3.98502320e-01 -8.83118734e-02 -1.23209155e+00 1.51703805e-01 1.20641533e-02 -1.88640133e-01 -8.46678674e-01 -3.02138813e-02 2.75925398e-02 2.90130563e-02 -4.50678498e-01 1.38544559e-01 -4.34386253e-01 -1.57832459e-01 5.02092540e-01 -2.15165943e-01 6.46659732e-01 4.58531380e-01 -2.03157440e-01 -3.15084070e-01 2.53941625e-01 8.29887509e-01 -2.40065798e-01 -6.83910012e-01 2.00445756e-01 -3.98712724e-01 3.85176927e-01 4.64882106e-01 5.38402438e-01 -4.01271343e-01 -1.91012889e-01 -4.35160637e-01 2.95477659e-01 4.41476762e-01 6.38707161e-01 -2.15866074e-01 -1.39338362e+00 -1.18126607e+00 2.42755994e-01 1.46771654e-01 -5.43256223e-01 -1.73174828e-01 7.23937511e-01 -4.74560291e-01 8.35605502e-01 -5.89556210e-02 -5.03856182e-01 -9.26906884e-01 3.01772654e-01 4.61380184e-01 -6.65591538e-01 -3.67022783e-01 6.73124075e-01 2.83793926e-01 -1.03285885e+00 -7.12133870e-02 -2.57661223e-01 3.99225414e-01 -1.50693998e-01 4.60448325e-01 4.48446833e-02 3.37555498e-01 -8.78471196e-01 -4.66815948e-01 4.89028543e-01 -3.84152830e-01 -4.36105967e-01 1.18581283e+00 -1.55396074e-01 -3.51895541e-01 7.48913527e-01 1.26105607e+00 5.41789949e-01 -8.08728278e-01 -5.65278769e-01 1.82458475e-01 6.42081536e-03 7.07374066e-02 -1.16499305e+00 -9.75564599e-01 9.93546963e-01 4.20262545e-01 -2.27034360e-01 8.78016949e-01 9.41654220e-02 9.94492114e-01 2.83473760e-01 8.10860395e-01 -9.20042872e-01 -4.05369937e-01 7.84184933e-01 5.57292342e-01 -1.25962591e+00 -2.97637045e-01 -1.87197834e-01 -5.26069641e-01 5.10176837e-01 5.47312617e-01 -1.17900865e-02 5.38029790e-01 7.59735048e-01 6.20651484e-01 2.07110271e-01 -9.31424856e-01 -7.94339553e-02 9.76347029e-02 2.81760335e-01 1.11821592e+00 4.29060072e-01 -2.50535876e-01 3.60333681e-01 -5.12011766e-01 -3.08183253e-01 6.35549724e-02 5.72749317e-01 -3.62209290e-01 -1.35364687e+00 -3.58650982e-01 7.32330158e-02 -9.40882146e-01 -6.53452039e-01 3.91189083e-02 1.16284084e+00 2.44454797e-02 7.95826197e-01 8.06547552e-02 -2.08680242e-01 2.33265951e-01 3.01807940e-01 3.88662428e-01 -6.30619466e-01 -8.58798444e-01 2.03554511e-01 2.65811831e-01 -2.93480843e-01 -1.03541732e-01 -6.99019909e-01 -1.10794580e+00 -6.90039754e-01 -5.06485924e-02 4.93078411e-01 6.57572746e-01 8.45275223e-01 2.94897020e-01 5.13950475e-02 4.00910884e-01 -2.88324863e-01 -4.69242454e-01 -1.04349029e+00 -4.06583279e-01 9.28460583e-02 8.17033127e-02 -1.30927369e-01 -1.88511863e-01 -1.10250145e-01]
[10.965686798095703, 9.960070610046387]
4d6bbe70-45d7-4de6-aa42-30e94844053b
spiking-two-stream-methods-with-unsupervised
2306.13783
null
https://arxiv.org/abs/2306.13783v1
https://arxiv.org/pdf/2306.13783v1.pdf
Spiking Two-Stream Methods with Unsupervised STDP-based Learning for Action Recognition
Video analysis is a computer vision task that is useful for many applications like surveillance, human-machine interaction, and autonomous vehicles. Deep Convolutional Neural Networks (CNNs) are currently the state-of-the-art methods for video analysis. However they have high computational costs, and need a large amount of labeled data for training. In this paper, we use Convolutional Spiking Neural Networks (CSNNs) trained with the unsupervised Spike Timing-Dependent Plasticity (STDP) learning rule for action classification. These networks represent the information using asynchronous low-energy spikes. This allows the network to be more energy efficient and neuromorphic hardware-friendly. However, the behaviour of CSNNs is not studied enough with spatio-temporal computer vision models. Therefore, we explore transposing two-stream neural networks into the spiking domain. Implementing this model with unsupervised STDP-based CSNNs allows us to further study the performance of these networks with video analysis. In this work, we show that two-stream CSNNs can successfully extract spatio-temporal information from videos despite using limited training data, and that the spiking spatial and temporal streams are complementary. We also show that using a spatio-temporal stream within a spiking STDP-based two-stream architecture leads to information redundancy and does not improve the performance.
['Ioan Marius Bilasco', 'Pierre Tirilly', 'Mireille El-Assal']
2023-06-23
null
null
null
null
['action-classification', 'autonomous-vehicles', 'action-recognition-in-videos']
['computer-vision', 'computer-vision', 'computer-vision']
[ 5.92971027e-01 -3.76197428e-01 1.15552358e-01 5.85631542e-02 1.64307043e-01 -3.84760529e-01 6.27538979e-01 3.74389328e-02 -6.78646743e-01 6.97078526e-01 -3.79809767e-01 -3.19059892e-03 8.20849016e-02 -8.31963480e-01 -1.06747842e+00 -1.09259498e+00 -1.92842156e-01 -2.57510036e-01 1.08038509e+00 -1.20963655e-01 2.46812269e-01 5.62758565e-01 -2.03174949e+00 7.40463197e-01 3.67520869e-01 1.30395508e+00 3.96343917e-01 6.56476140e-01 -1.27497926e-01 9.14319277e-01 -3.59721929e-01 2.17908353e-01 1.82195649e-01 -8.70914459e-01 -1.84777096e-01 -3.63194764e-01 -8.28308240e-02 1.20148912e-01 -7.31592357e-01 8.25032890e-01 2.90837467e-01 -1.88836321e-01 4.63839293e-01 -1.39072454e+00 -2.56507158e-01 3.12000394e-01 -4.73644584e-02 4.83575344e-01 -1.73329011e-01 2.68679112e-01 1.45822138e-01 -5.64543605e-01 7.62714744e-01 7.82217383e-01 9.49797511e-01 8.82002831e-01 -1.22659099e+00 -7.58739769e-01 -2.13778317e-01 3.64335775e-01 -1.10244823e+00 -4.96143520e-01 7.02744007e-01 -2.09298953e-01 1.37186563e+00 -1.18559495e-01 1.38759053e+00 1.18325770e+00 7.65163183e-01 7.22793519e-01 1.15320587e+00 -1.82618439e-01 8.02042067e-01 -4.76292521e-01 5.20532839e-02 5.29492736e-01 2.92202085e-01 1.95375159e-01 -1.02074826e+00 2.12062269e-01 9.69905972e-01 3.08177918e-01 6.41935691e-03 -1.28007770e-01 -1.11839044e+00 4.05795604e-01 5.75449288e-01 6.89976811e-01 -4.34725642e-01 9.61226463e-01 3.80839467e-01 2.87347674e-01 -8.60641077e-02 3.12983811e-01 -1.86556205e-01 -3.90248984e-01 -1.11470914e+00 1.68906018e-01 5.97769618e-01 5.30968964e-01 6.32072151e-01 4.26862150e-01 -8.03003237e-02 4.90251899e-01 1.04232557e-01 4.09408510e-01 7.07906187e-01 -1.39155066e+00 -2.73295343e-01 6.82382584e-01 -4.17665482e-01 -7.55688667e-01 -4.90394562e-01 1.53404146e-01 -1.05821407e+00 6.23731732e-01 5.45340300e-01 1.30209133e-01 -1.08975375e+00 1.64594507e+00 -4.92198169e-01 4.75667924e-01 2.64274567e-01 7.11022139e-01 5.83968222e-01 8.66600096e-01 -1.69141125e-02 -4.04606938e-01 1.22840357e+00 -5.42138398e-01 -6.84514821e-01 -3.00124586e-01 4.11554605e-01 3.52043170e-03 3.94479185e-01 4.11310285e-01 -1.13070226e+00 -2.53678948e-01 -1.20209527e+00 9.84648466e-02 -6.35571718e-01 -7.34425411e-02 5.27433217e-01 5.15394807e-01 -1.45755363e+00 8.78590226e-01 -1.37637591e+00 -7.73089945e-01 7.15594411e-01 6.46351755e-01 -2.60514289e-01 2.76652187e-01 -1.01780450e+00 6.23737514e-01 3.38569611e-01 -2.06589296e-01 -8.94118369e-01 -1.55327424e-01 -5.18298507e-01 1.54949069e-01 -2.03998819e-01 -3.99638921e-01 1.01088846e+00 -1.27539599e+00 -1.57748961e+00 6.68763697e-01 -3.59006584e-01 -9.64961350e-01 -1.19294740e-01 6.21848643e-01 7.18354285e-02 4.65890288e-01 -3.82724553e-01 1.08299255e+00 6.81821644e-01 -8.94515336e-01 -2.97747254e-01 -3.29692841e-01 -3.47957671e-01 -3.32710773e-01 -4.33387309e-01 9.01466087e-02 -2.75178403e-01 -7.33955145e-01 2.51627386e-01 -1.10088921e+00 -1.93935204e-02 4.63341415e-01 3.73224139e-01 -4.25443687e-02 1.03796577e+00 -1.35990873e-01 8.42039526e-01 -2.23576522e+00 1.92676678e-01 -1.20696291e-01 3.38859223e-02 4.93570179e-01 -1.97976321e-01 3.83846879e-01 1.32775173e-01 5.22842743e-02 -5.98511279e-01 -3.48829105e-02 -6.21703625e-01 5.59919238e-01 -9.23283473e-02 3.03698063e-01 6.70010746e-01 9.62018549e-01 -7.59272933e-01 -2.38861933e-01 -6.05786070e-02 6.07934892e-01 -3.70745748e-01 -1.61121115e-01 -2.59298742e-01 5.67710340e-01 -4.22411934e-02 6.39599025e-01 3.03554296e-01 -2.80815680e-02 1.53944597e-01 -6.21888191e-02 -4.84241962e-01 4.66896929e-02 -6.25942290e-01 1.75799310e+00 -7.05068856e-02 1.22892153e+00 -2.11480230e-01 -1.37529099e+00 8.98093104e-01 2.86603987e-01 7.23364651e-01 -1.25773406e+00 2.96619713e-01 4.61172193e-01 2.92798817e-01 -3.23849857e-01 -3.11424918e-02 -6.25117496e-02 2.09665850e-01 3.99301827e-01 4.54932630e-01 1.47639230e-01 3.53044271e-01 -7.74947628e-02 1.73823011e+00 1.57860786e-01 -2.19852075e-01 -3.06692034e-01 -2.28770580e-02 -8.57398063e-02 6.85255826e-01 5.71074724e-01 -2.36716613e-01 6.82985604e-01 6.23263776e-01 -5.07681012e-01 -1.27037108e+00 -1.03720915e+00 -1.35998130e-01 6.83510363e-01 2.21615031e-01 -1.99151009e-01 -8.07780147e-01 9.88412276e-02 -4.13503081e-01 1.40466839e-01 -5.09644508e-01 -2.97653943e-01 -7.38250554e-01 -8.61099720e-01 1.00016761e+00 7.76204765e-01 7.51267791e-01 -1.46595418e+00 -1.49911284e+00 5.63876331e-01 1.38767198e-01 -1.19396603e+00 3.92556608e-01 9.13735032e-01 -1.06407106e+00 -8.66631687e-01 -8.38591278e-01 -1.01950896e+00 5.55777550e-01 -1.03396969e-02 5.60147285e-01 5.60855716e-02 -3.94820541e-01 2.58019865e-01 -3.87488365e-01 -5.00861704e-01 -1.46671250e-01 -2.05477640e-01 7.82202035e-02 1.22534586e-02 4.57177043e-01 -1.02185643e+00 -4.68165249e-01 1.04645543e-01 -1.18216932e+00 1.80991337e-01 4.64225888e-01 7.32075274e-01 8.94820452e-01 -3.73473875e-02 5.35705745e-01 -3.16847056e-01 2.49890491e-01 -2.58852035e-01 -5.49396813e-01 -4.50156815e-02 -2.77422637e-01 3.30643475e-01 8.21429014e-01 -5.57119787e-01 -5.78200638e-01 5.37292182e-01 3.00893895e-02 -4.67911631e-01 -1.00567348e-01 3.76484185e-01 3.89663637e-01 -5.14248371e-01 7.26572335e-01 7.91473150e-01 1.19699121e-01 -2.29229070e-02 -5.39922416e-01 2.74896622e-01 4.98270780e-01 -1.45769611e-01 1.57809809e-01 7.96921909e-01 4.13632810e-01 -9.96810138e-01 2.71613915e-02 -1.56838983e-01 -4.01098818e-01 -5.75708687e-01 1.02016771e+00 -6.03309572e-01 -6.75380528e-01 1.12913811e+00 -1.43817043e+00 -7.44765162e-01 -2.72726387e-01 3.91602546e-01 -9.36077178e-01 4.87298146e-03 -9.24625754e-01 -9.66746986e-01 -1.41703533e-02 -9.63292897e-01 6.97455168e-01 3.91425818e-01 1.97989285e-01 -5.74548781e-01 1.48234874e-01 -5.23475766e-01 6.22233331e-01 3.49103570e-01 7.10271358e-01 -3.61913174e-01 -7.89012372e-01 9.87343118e-02 -1.33607537e-01 1.91104606e-01 -1.60011873e-01 1.95830569e-01 -1.14521277e+00 8.94878432e-02 1.03508510e-01 -4.29276645e-01 1.51621675e+00 6.23147488e-01 1.44665587e+00 -6.34918809e-02 -4.34689522e-01 6.29133701e-01 1.48239708e+00 5.13654232e-01 1.15320039e+00 3.05939108e-01 2.42980689e-01 5.73955238e-01 -1.91817611e-01 1.78829521e-01 7.32937129e-03 4.03561324e-01 6.35818541e-01 3.07415634e-01 -2.28983924e-01 1.39824976e-03 6.66706264e-01 9.06527340e-01 -4.30099607e-01 -2.63483793e-01 -1.03628397e+00 5.40023148e-01 -2.15948582e+00 -1.24663889e+00 -2.51213849e-01 2.07732725e+00 7.23489404e-01 1.44753546e-01 7.31452776e-04 5.62686324e-01 6.84336424e-01 -2.48968869e-01 -7.33951926e-01 -5.19733489e-01 -6.16627812e-01 5.27229130e-01 6.31507516e-01 -3.29615712e-01 -8.79686713e-01 6.00107491e-01 6.18818474e+00 4.96948600e-01 -1.48531866e+00 -5.60760610e-02 3.73536468e-01 -1.55323222e-01 -1.53945880e-02 -1.18442073e-01 -4.85936284e-01 7.79904306e-01 1.55351782e+00 6.30669296e-02 7.79881835e-01 2.70963132e-01 2.20220178e-01 -5.69368660e-01 -9.89445210e-01 1.23740900e+00 -1.53883994e-01 -1.78354919e+00 -1.32783413e-01 7.19005913e-02 5.84570229e-01 3.41372758e-01 -7.14665428e-02 4.26187031e-02 -1.87317848e-01 -1.07811332e+00 6.46185696e-01 6.46534860e-01 5.42576134e-01 -5.24101019e-01 7.16179669e-01 4.70998287e-01 -1.28368318e+00 -2.70988196e-01 -4.86863524e-01 -2.26426601e-01 -5.78154214e-02 4.01104480e-01 3.37430425e-02 -7.51637369e-02 1.12973547e+00 1.05677533e+00 -4.79953796e-01 1.32225525e+00 2.97767580e-01 5.38982511e-01 -6.45464182e-01 -3.72803718e-01 2.59825438e-01 9.44708660e-02 3.06474060e-01 1.13317990e+00 7.09816575e-01 3.01955611e-01 -4.01803881e-01 9.34526265e-01 -2.62027651e-01 -5.24518013e-01 -9.39145625e-01 -5.32324731e-01 3.85630816e-01 9.72968996e-01 -1.38846803e+00 -2.43115202e-01 -2.25652650e-01 9.22524929e-01 3.46137621e-02 2.00399518e-01 -5.99757969e-01 -4.07632977e-01 3.05915356e-01 2.47788489e-01 4.13944036e-01 -4.73929644e-01 -4.30943489e-01 -9.96333599e-01 -1.72003768e-02 -2.83182949e-01 -2.12699085e-01 -9.24019039e-01 -8.98761570e-01 4.41292346e-01 -3.06005955e-01 -1.32735407e+00 -2.45241463e-01 -9.18031394e-01 -7.13738918e-01 2.96102017e-01 -1.46551323e+00 -7.92896330e-01 -4.09030646e-01 7.08408356e-01 4.26542222e-01 -2.84328014e-02 8.89082730e-01 4.79258522e-02 -2.74145842e-01 9.36401784e-02 1.55645370e-01 2.76925892e-01 2.84298331e-01 -8.68107498e-01 3.14285010e-01 8.32034707e-01 6.76215813e-02 2.06944197e-01 4.57702130e-01 -3.91377926e-01 -1.66241753e+00 -1.16082203e+00 6.16203070e-01 1.89024862e-03 5.64895570e-01 -4.30745870e-01 -1.08271933e+00 1.34142518e-01 4.15771335e-01 3.13187063e-01 5.71531713e-01 -8.58375430e-01 -2.13945657e-01 -1.62884325e-01 -1.15607226e+00 5.09162009e-01 1.11366165e+00 -4.95955944e-01 -3.05357903e-01 -1.10595077e-01 2.23371521e-01 1.97770953e-01 -4.68939662e-01 3.71274769e-01 7.36596942e-01 -1.14594960e+00 5.23948848e-01 -1.84970036e-01 4.13769633e-01 -3.93463463e-01 -1.90051645e-02 -1.15459728e+00 -7.08637536e-02 -2.75910676e-01 -2.45228812e-01 7.59780347e-01 2.15080291e-01 -6.45033896e-01 7.98330903e-01 1.93762213e-01 -2.28749216e-01 -6.91032231e-01 -1.20540726e+00 -1.20879221e+00 8.21586978e-03 -4.40881819e-01 -2.23338038e-01 4.54307169e-01 1.84323117e-01 -1.44345030e-01 1.12609804e-01 -2.10277066e-01 4.75944757e-01 -2.67176569e-01 2.65384633e-02 -1.33307743e+00 -2.33886912e-02 -5.64683080e-01 -1.04847145e+00 -5.72788835e-01 1.39678076e-01 -7.47703612e-01 4.35974419e-01 -1.57185650e+00 1.06207170e-01 -1.06577031e-01 -5.35384476e-01 8.27215791e-01 6.89507604e-01 6.39900386e-01 1.32453933e-01 3.47900391e-01 -4.37674165e-01 3.82478684e-01 6.54355943e-01 -1.39124319e-01 -9.69964638e-02 -4.13154989e-01 1.55204728e-01 5.52140474e-01 1.02062404e+00 -7.29134083e-01 -2.10121363e-01 -4.07261252e-01 3.29196215e-01 -1.20773472e-01 6.86333239e-01 -1.66469562e+00 8.49709868e-01 4.08528857e-02 5.44633865e-01 -1.20010279e-01 5.59613287e-01 -7.03396738e-01 9.68974605e-02 8.94039929e-01 -2.37450540e-01 5.61774783e-02 4.92459834e-01 6.65176749e-01 -3.11508477e-01 -1.88244343e-01 9.66432333e-01 -3.94239813e-01 -8.44664276e-01 1.47728562e-01 -1.39385056e+00 -2.97603369e-01 1.15423608e+00 -7.69380093e-01 -4.47130382e-01 5.19791525e-03 -5.25906265e-01 -2.97122002e-01 6.50782883e-01 8.98328945e-02 8.26622784e-01 -1.34560561e+00 -2.20369026e-01 4.25968677e-01 7.13441521e-02 -2.70435512e-01 2.49270685e-02 9.41687942e-01 -7.50731945e-01 4.82302219e-01 -1.07720971e+00 -9.78630126e-01 -8.45113635e-01 4.36866045e-01 4.16188240e-01 1.75361902e-01 -3.71450394e-01 5.73156893e-01 -1.17401004e-01 1.53714225e-01 1.21508911e-01 -4.62711930e-01 -1.61792979e-01 7.47580593e-03 4.69022840e-01 1.49380207e-01 1.08659990e-01 -2.68971711e-01 -4.77074653e-01 6.31743312e-01 5.28738141e-01 -2.73685217e-01 1.59717119e+00 2.22363845e-01 -2.80685753e-01 8.21536899e-01 9.61728275e-01 -8.15680504e-01 -1.43739653e+00 2.58383781e-01 7.54075572e-02 2.07958177e-01 1.21502049e-01 -3.44710380e-01 -1.32898438e+00 1.24988115e+00 7.48082817e-01 3.68580222e-01 1.53341317e+00 -2.29250684e-01 8.28699172e-01 6.98810995e-01 6.21193469e-01 -1.21191907e+00 3.71452153e-01 6.58458829e-01 4.82592851e-01 -7.74636567e-01 -6.26150489e-01 -7.46260509e-02 -2.14797512e-01 1.59265602e+00 5.45686185e-01 -5.90742707e-01 7.62677312e-01 7.81294465e-01 -4.12191123e-01 -4.44896780e-02 -1.20281339e+00 -3.95824313e-01 -3.39364946e-01 7.90046155e-01 1.97122365e-01 -2.95290500e-01 -2.90913790e-01 4.53662783e-01 4.44066077e-01 6.08618677e-01 8.66440892e-01 1.36623037e+00 -6.82267785e-01 -9.47762787e-01 2.81019532e-03 5.20562112e-01 -4.30654317e-01 -1.28857315e-01 -5.09414852e-01 2.10098222e-01 1.44062370e-01 6.43320203e-01 4.63947058e-01 -5.62293589e-01 6.36974256e-03 3.24173421e-01 7.59585798e-01 -3.56826305e-01 -7.00901091e-01 -5.46096675e-02 -3.00546378e-01 -6.44614995e-01 -9.54078674e-01 -5.21132231e-01 -1.74912262e+00 -2.33945459e-01 1.51094839e-01 -4.35011953e-01 1.06620252e+00 8.88204932e-01 5.88668108e-01 6.54932737e-01 3.24893624e-01 -1.15893352e+00 2.94353455e-01 -5.46880364e-01 -4.29310858e-01 1.84026454e-02 2.03894913e-01 -5.18811941e-01 -2.14160174e-01 5.72779775e-01]
[8.222579956054688, 2.4170591831207275]
6ecb1ff8-4b8f-4f6d-9537-7f0b1f67bfc8
latency-control-for-keyword-spotting
2206.07261
null
https://arxiv.org/abs/2206.07261v1
https://arxiv.org/pdf/2206.07261v1.pdf
Latency Control for Keyword Spotting
Conversational agents commonly utilize keyword spotting (KWS) to initiate voice interaction with the user. For user experience and privacy considerations, existing approaches to KWS largely focus on accuracy, which can often come at the expense of introduced latency. To address this tradeoff, we propose a novel approach to control KWS model latency and which generalizes to any loss function without explicit knowledge of the keyword endpoint. Through a single, tunable hyperparameter, our approach enables one to balance detection latency and accuracy for the targeted application. Empirically, we show that our approach gives superior performance under latency constraints when compared to existing methods. Namely, we make a substantial 25\% relative false accepts improvement for a fixed latency target when compared to the baseline state-of-the-art. We also show that when our approach is used in conjunction with a max-pooling loss, we are able to improve relative false accepts by 25 % at a fixed latency when compared to cross entropy loss.
['Brian Kulis', 'Yuriy Mishchenko', 'Mohammad Omar Khursheed', 'Grant P. Strimel', 'Joseph Wang', 'Christin Jose']
2022-06-15
null
null
null
null
['keyword-spotting']
['speech']
[ 8.92030448e-02 1.61206126e-01 -2.37833232e-01 -4.10586208e-01 -1.25170231e+00 -7.34746993e-01 7.22363591e-01 3.06723952e-01 -9.03490245e-01 5.45462251e-01 4.79474403e-02 -3.46061289e-01 2.04565346e-01 -3.68699849e-01 -4.09192920e-01 -4.79820102e-01 1.15891192e-02 3.17052037e-01 2.80728728e-01 4.76038828e-02 9.67583060e-02 2.81104922e-01 -1.11268294e+00 2.39913002e-01 7.09926248e-01 1.12638688e+00 -2.18737006e-01 8.09450567e-01 1.58399865e-02 5.43099225e-01 -8.03909838e-01 -5.36325395e-01 3.08489144e-01 -1.31472036e-01 -9.56810534e-01 -2.52025962e-01 2.00963572e-01 -6.27457857e-01 -2.47348681e-01 6.78478539e-01 7.48664558e-01 1.92868724e-01 3.57974708e-01 -1.43437755e+00 6.93881512e-02 6.02317870e-01 -2.65258193e-01 1.05297968e-01 5.06171346e-01 3.24611217e-01 1.30789745e+00 -6.47225320e-01 2.82514870e-01 1.07208884e+00 6.77710056e-01 5.66673458e-01 -1.73997808e+00 -5.21027625e-01 2.62177736e-01 -5.19961677e-03 -1.53350115e+00 -9.47692215e-01 2.69495100e-01 9.37874690e-02 1.29598761e+00 5.61722755e-01 1.62659287e-01 1.01539505e+00 -1.38378471e-01 1.15694475e+00 8.78418982e-01 -3.76319528e-01 5.15701354e-01 5.41647792e-01 2.55528390e-01 4.24875587e-01 -2.67438561e-01 -1.68477416e-01 -7.01696932e-01 -6.77194595e-01 -2.88096326e-03 -3.71714324e-01 -4.82371718e-01 -2.49181420e-01 -8.66207600e-01 8.23121786e-01 8.44428167e-02 9.95099731e-03 -1.17497452e-01 1.62193596e-01 5.56743324e-01 4.15544331e-01 4.90868032e-01 5.56358635e-01 -7.92493522e-01 -7.07645357e-01 -9.24672604e-01 4.66919661e-01 1.25469828e+00 8.84527683e-01 4.03637528e-01 -4.41598654e-01 -4.20987874e-01 7.97683895e-01 5.76465204e-02 2.57662773e-01 2.73477435e-01 -8.62850726e-01 5.30155599e-01 2.89056599e-01 2.03376785e-01 -4.12052363e-01 -3.42548013e-01 -8.25774968e-02 -1.62704349e-01 4.69105355e-02 3.91275525e-01 -4.18308437e-01 -6.33802831e-01 1.99732006e+00 3.13067406e-01 1.56960353e-01 1.45270422e-01 5.93892276e-01 4.29277003e-01 5.41950405e-01 4.42622155e-02 -6.06840491e-01 1.38490713e+00 -9.25681889e-01 -7.25984097e-01 -1.58473134e-01 8.12634587e-01 -6.58060968e-01 1.53875422e+00 2.98253357e-01 -1.10488248e+00 3.05278063e-01 -9.44104016e-01 -1.14450380e-02 -1.92401215e-01 -2.48171642e-01 5.32893360e-01 9.41715360e-01 -1.19133341e+00 3.44446063e-01 -8.43953669e-01 -4.58282858e-01 -1.28418924e-02 6.56403065e-01 -1.28663242e-01 5.02760530e-01 -1.05149627e+00 9.11506355e-01 3.42794657e-02 -2.52371103e-01 -4.40323830e-01 -9.00025487e-01 -5.69335401e-01 2.87004471e-01 7.05152094e-01 -7.29417980e-01 1.95824850e+00 -4.80701625e-01 -2.11413670e+00 4.15891975e-01 -2.12835655e-01 -7.59477437e-01 8.53113592e-01 -4.26935583e-01 -6.59542605e-02 -4.00982425e-02 -3.37402493e-01 4.74269390e-01 6.09114230e-01 -9.92775619e-01 -7.37062097e-01 -8.11916813e-02 5.52317560e-01 3.65252882e-01 -6.33554935e-01 1.06283441e-01 -7.26929486e-01 -4.06195700e-01 -4.44993228e-01 -1.09282815e+00 -1.47904322e-01 -3.70668657e-02 -3.62771988e-01 -2.69305199e-01 7.51465142e-01 -3.85193199e-01 1.39400101e+00 -2.18483758e+00 2.70606317e-02 2.63066798e-01 1.30169556e-01 3.85957837e-01 2.00528861e-03 5.08333027e-01 4.56972867e-01 2.09742695e-01 -1.22596301e-01 -9.16768074e-01 1.44963801e-01 -9.08854753e-02 -1.74129546e-01 3.18008095e-01 -3.51667777e-02 6.89257741e-01 -6.19399905e-01 -2.39825591e-01 3.68918739e-02 4.93422985e-01 -1.01448083e+00 5.22325099e-01 -3.34654152e-01 -8.29810798e-02 -2.64202535e-01 2.20119730e-01 3.75736564e-01 -1.14183493e-01 3.11351329e-01 -4.35328344e-03 8.23239237e-02 6.98036849e-01 -1.09721386e+00 1.45051396e+00 -9.78395045e-01 4.01190937e-01 4.27633911e-01 -3.77857774e-01 4.12128925e-01 4.86346602e-01 8.53817463e-02 -3.44036072e-01 1.88109368e-01 9.02873725e-02 -1.72644317e-01 -8.22074488e-02 4.98202175e-01 1.41330540e-01 -2.31602155e-02 4.83487964e-01 -3.18276227e-01 9.89674106e-02 -3.41672301e-01 1.85057104e-01 1.30399156e+00 -4.90261436e-01 2.15572059e-01 -2.47434705e-01 4.06383365e-01 -5.15082896e-01 2.82391965e-01 8.25379193e-01 -2.96609253e-01 2.39323616e-01 7.66302109e-01 -2.11096108e-01 -8.12787175e-01 -8.22128952e-01 -1.52263880e-01 1.39051080e+00 -9.16257650e-02 -4.93608207e-01 -1.06175423e+00 -9.12019134e-01 7.38258213e-02 9.05517340e-01 -3.27063292e-01 -1.29654497e-01 -3.69498610e-01 -7.83414662e-01 8.04093063e-01 3.06416988e-01 3.94527376e-01 -7.49194086e-01 -5.31696856e-01 1.76856250e-01 -7.40863234e-02 -1.49847174e+00 -9.69931960e-01 1.54682726e-01 -5.04587591e-01 -4.99437451e-01 -5.83765507e-01 -1.85922414e-01 3.59848738e-01 1.44612119e-01 7.77330279e-01 -1.25695258e-01 -2.09622979e-01 4.41231668e-01 -2.15431169e-01 -9.15062726e-02 -2.88126856e-01 6.51496112e-01 2.23723546e-01 4.94089015e-02 3.62215012e-01 -6.14519119e-01 -7.32603490e-01 9.37519372e-02 -6.93741262e-01 -2.74050474e-01 3.83664608e-01 8.41026723e-01 -2.09487230e-01 -3.03617358e-01 5.42121589e-01 -9.82597768e-01 9.77651894e-01 -2.60413319e-01 -6.23533189e-01 1.99661076e-01 -1.07054925e+00 3.76812965e-01 4.59878534e-01 -4.80810165e-01 -8.05000842e-01 -7.31089115e-02 -2.43183509e-01 -1.67755291e-01 5.16001172e-02 1.54378772e-01 -2.75493354e-01 -2.40298182e-01 4.05597597e-01 1.73877105e-01 2.28184044e-01 -3.90780836e-01 5.88074148e-01 1.10501707e+00 1.51076168e-01 -4.99512672e-01 2.34468594e-01 8.25008303e-02 -5.45305908e-01 -7.65912414e-01 -3.58289838e-01 -5.39259553e-01 6.61810860e-02 1.22554563e-01 4.21466619e-01 -8.26203585e-01 -1.57885885e+00 2.99218267e-01 -1.11525321e+00 -3.95338923e-01 1.08413450e-01 3.21571559e-01 -6.46654546e-01 3.11550558e-01 -8.07445645e-01 -1.30589461e+00 -7.12861419e-01 -1.22272778e+00 1.09945464e+00 -2.59116050e-02 -4.95993853e-01 -7.17274070e-01 -1.85438886e-01 3.36633682e-01 7.14780748e-01 -2.21869409e-01 7.33512700e-01 -1.02939510e+00 -4.27567869e-01 -3.11935931e-01 -2.05878228e-01 2.00051218e-01 2.78132223e-02 -2.78760940e-01 -1.21376264e+00 -5.84090769e-01 -2.28135049e-01 -2.93901801e-01 6.45138741e-01 1.30004585e-02 1.04241300e+00 -7.33013809e-01 -2.74394691e-01 4.30547029e-01 1.03985250e+00 3.42468113e-01 2.54344672e-01 2.06242964e-01 3.38028431e-01 4.42658722e-01 2.87763596e-01 7.90742040e-01 4.98873442e-01 1.17437112e+00 -2.35227682e-03 9.66021046e-02 4.40051347e-01 5.30388430e-02 4.59152728e-01 4.39456582e-01 2.96705246e-01 -5.48651814e-01 -6.98267996e-01 5.19878864e-01 -2.03157926e+00 -6.80186629e-01 6.49257421e-01 2.63083172e+00 1.10047424e+00 4.35393661e-01 5.66103876e-01 1.94385946e-01 4.18996006e-01 2.38424763e-01 -5.98500013e-01 -7.08181143e-01 4.79429901e-01 4.16196659e-02 8.18219006e-01 1.05181336e+00 -9.85716105e-01 1.02540779e+00 7.05137587e+00 8.74628305e-01 -1.06764102e+00 1.12389274e-01 6.51901066e-01 -5.49355209e-01 -2.22950339e-01 -2.68632382e-01 -9.20674503e-01 4.23614979e-01 1.31087649e+00 -3.30169350e-01 8.53211403e-01 8.87080491e-01 2.76466906e-01 3.33349407e-02 -1.37213922e+00 9.09811318e-01 -4.37131152e-02 -1.07243371e+00 -3.68058980e-01 2.13058233e-01 1.90787375e-01 -2.18277335e-01 8.26159716e-02 3.27536941e-01 3.27136308e-01 -8.77887964e-01 3.69946510e-01 1.03595831e-01 6.48423970e-01 -1.15056932e+00 7.54433274e-01 4.10432309e-01 -8.93496633e-01 7.70784765e-02 1.70899078e-01 3.21312547e-02 2.22489581e-01 2.84906089e-01 -1.43297708e+00 1.43608540e-01 4.31715578e-01 -2.40524679e-01 2.87505221e-02 8.46884966e-01 2.03541920e-01 6.09537423e-01 -7.13290930e-01 -4.27441746e-01 2.98541576e-01 2.59711057e-01 5.43033898e-01 1.51078331e+00 -1.50572706e-03 8.44834968e-02 3.07539880e-01 4.44726884e-01 -2.02936962e-01 3.61040413e-01 -3.58104050e-01 -5.03889509e-02 9.29863572e-01 1.05173767e+00 -3.42181891e-01 -2.20010146e-01 -1.46330774e-01 1.33215749e+00 5.87340295e-01 2.96181440e-01 -7.58164763e-01 -5.01541853e-01 1.23945403e+00 3.88048999e-02 3.42130452e-01 -2.33371362e-01 -1.43420175e-01 -8.46451759e-01 3.05073589e-01 -8.83469164e-01 2.85524994e-01 2.85983458e-02 -9.84382510e-01 7.61565983e-01 4.74664941e-02 -7.32683599e-01 -5.28875589e-01 -3.03923219e-01 -4.10778612e-01 7.41416574e-01 -1.25267577e+00 -9.82788026e-01 1.90481544e-01 2.93876916e-01 4.76247460e-01 -3.33836128e-04 1.11404407e+00 4.13978845e-01 -4.64137673e-01 1.41311634e+00 5.20888977e-02 -1.27892897e-01 8.13503623e-01 -1.26039863e+00 5.56379855e-01 5.39190233e-01 -8.60150829e-02 7.88363397e-01 1.00791836e+00 -8.03398266e-02 -1.38599086e+00 -7.84969866e-01 1.07410562e+00 -4.93843615e-01 4.65103686e-01 -7.65268564e-01 -7.33826339e-01 6.43832028e-01 2.11866140e-01 -8.14451948e-02 8.19123328e-01 5.23227513e-01 -5.11488199e-01 -1.73923001e-01 -1.41809130e+00 9.07911003e-01 8.15115035e-01 -7.24040985e-01 -2.32774347e-01 2.57406265e-01 1.23969603e+00 -5.50302923e-01 -6.97667658e-01 2.39597425e-01 7.75161088e-01 -7.61648715e-01 8.10063779e-01 -5.79825997e-01 -2.24261165e-01 -9.34072360e-02 -8.49614441e-02 -1.10231388e+00 1.48736671e-01 -1.25903106e+00 -1.28996164e-01 1.32588923e+00 7.48750687e-01 -7.51688063e-01 7.56300867e-01 1.23403382e+00 2.49040112e-01 -7.77103364e-01 -1.14636922e+00 -7.01713502e-01 -2.39689931e-01 -5.94951510e-01 5.96621156e-01 5.61006427e-01 4.48996127e-01 3.94633055e-01 -6.71690464e-01 3.49279732e-01 3.25424790e-01 -1.90372676e-01 7.71668553e-01 -5.93236148e-01 -4.76952821e-01 -5.69566786e-01 -2.51421183e-01 -1.26492500e+00 2.05092132e-01 -4.92277652e-01 2.72918552e-01 -9.41569924e-01 4.42218781e-02 -3.68068516e-01 -2.37210631e-01 5.40481329e-01 -2.14201108e-01 7.72439912e-02 3.89943123e-01 -9.69752073e-02 -6.45686269e-01 6.01555288e-01 4.47762996e-01 1.59646332e-01 -6.78234875e-01 2.97873050e-01 -6.02859914e-01 4.64502066e-01 7.38709331e-01 -2.68752009e-01 -4.69017357e-01 -8.98876227e-03 -1.29545137e-01 8.28581378e-02 8.81964108e-04 -6.43597126e-01 5.13524830e-01 -7.87465125e-02 -3.82634938e-01 3.32294367e-02 6.84222102e-01 -7.27359235e-01 -1.94519058e-01 2.56107867e-01 -7.56851375e-01 -1.51438406e-02 3.25993985e-01 6.16543174e-01 1.54023334e-01 9.16300341e-02 6.84479177e-01 3.20443928e-01 -1.53864250e-01 1.63407072e-01 -5.64763963e-01 -1.41484421e-02 9.47018981e-01 2.57458270e-01 -2.43846640e-01 -8.30922127e-01 -4.68731374e-01 5.42921722e-01 3.15324277e-01 3.99060458e-01 2.83617854e-01 -1.04954875e+00 -3.84236276e-01 9.64766890e-02 2.76218265e-01 -3.85905653e-01 -8.17622468e-02 6.77754879e-01 -3.26263070e-01 4.14058208e-01 4.88342375e-01 -3.40284139e-01 -1.69104052e+00 4.38544571e-01 2.86065221e-01 -2.34272569e-01 -5.39990723e-01 1.00046325e+00 -1.09124832e-01 -4.26524222e-01 9.04725671e-01 -4.95099574e-01 1.90113947e-01 -3.50593552e-02 7.80203581e-01 2.12552354e-01 3.06017905e-01 -1.42365739e-01 -6.40751362e-01 2.45615770e-03 -4.92608726e-01 -5.57478130e-01 9.18461263e-01 -2.09908992e-01 2.42056042e-01 4.90799427e-01 1.31018734e+00 3.00303668e-01 -1.10310876e+00 -4.95650113e-01 1.66986823e-01 -5.48192799e-01 1.19426534e-01 -9.03313756e-01 -7.98719287e-01 4.25823390e-01 5.46596766e-01 3.09463382e-01 9.24464583e-01 -1.27834111e-01 9.95882690e-01 5.81472993e-01 3.48123401e-01 -9.37210619e-01 -2.81875819e-01 3.77849549e-01 5.60043633e-01 -1.26170766e+00 -3.09767276e-01 -5.22893250e-01 -8.48872125e-01 7.30652630e-01 4.01124388e-01 2.76451319e-01 5.57707667e-01 7.78994620e-01 2.02110052e-01 2.71893948e-01 -1.24070430e+00 -5.93428016e-02 5.83580881e-02 1.45270005e-01 4.74698722e-01 1.32280514e-01 -5.10324717e-01 5.13946831e-01 -2.91627824e-01 -9.85711887e-02 1.89366117e-01 8.90344083e-01 -2.66504914e-01 -1.25718904e+00 6.77013919e-02 2.50822246e-01 -7.57122517e-01 -3.57666761e-01 -3.27449441e-01 4.85234112e-01 -4.75714594e-01 1.15211701e+00 6.03816807e-02 -6.15573883e-01 5.19974470e-01 4.11377817e-01 7.25319088e-02 -3.34901690e-01 -9.53372061e-01 5.72390594e-02 5.42108119e-01 -6.97994590e-01 -2.96402387e-02 -5.82537353e-01 -1.05320525e+00 -7.33515263e-01 -4.66656029e-01 3.29557329e-01 4.69534248e-01 9.47566867e-01 5.99643409e-01 2.35767558e-01 7.60174096e-01 -4.84884709e-01 -1.00821066e+00 -8.51453125e-01 -2.92746216e-01 1.92178804e-02 5.51150739e-01 -3.47278386e-01 -6.82165742e-01 -3.81805956e-01]
[14.009051322937012, 6.891024112701416]
4619ab15-6a13-468d-82ad-b5a2f3b1c897
textdefense-adversarial-text-detection-based
2302.05892
null
https://arxiv.org/abs/2302.05892v1
https://arxiv.org/pdf/2302.05892v1.pdf
TextDefense: Adversarial Text Detection based on Word Importance Entropy
Currently, natural language processing (NLP) models are wildly used in various scenarios. However, NLP models, like all deep models, are vulnerable to adversarially generated text. Numerous works have been working on mitigating the vulnerability from adversarial attacks. Nevertheless, there is no comprehensive defense in existing works where each work targets a specific attack category or suffers from the limitation of computation overhead, irresistible to adaptive attack, etc. In this paper, we exhaustively investigate the adversarial attack algorithms in NLP, and our empirical studies have discovered that the attack algorithms mainly disrupt the importance distribution of words in a text. A well-trained model can distinguish subtle importance distribution differences between clean and adversarial texts. Based on this intuition, we propose TextDefense, a new adversarial example detection framework that utilizes the target model's capability to defend against adversarial attacks while requiring no prior knowledge. TextDefense differs from previous approaches, where it utilizes the target model for detection and thus is attack type agnostic. Our extensive experiments show that TextDefense can be applied to different architectures, datasets, and attack methods and outperforms existing methods. We also discover that the leading factor influencing the performance of TextDefense is the target model's generalizability. By analyzing the property of the target model and the property of the adversarial example, we provide our insights into the adversarial attacks in NLP and the principles of our defense method.
['Yanghe Feng', 'Xing Yang', 'Chunpeng Ge', 'Yuwen Pu', 'Shouling Ji', 'Xuhong Zhang', 'Lujia Shen']
2023-02-12
null
null
null
null
['adversarial-text']
['adversarial']
[ 1.32105485e-01 -1.10408925e-01 7.62715340e-02 -1.33392587e-01 -6.54040992e-01 -1.34876204e+00 8.79595697e-01 1.67911932e-01 -5.92300259e-02 3.61345917e-01 4.00394946e-01 -5.32046914e-01 1.91102087e-01 -1.00679302e+00 -6.77973092e-01 -5.44658840e-01 3.63493823e-02 4.44969498e-02 1.65587470e-01 -5.60020506e-01 2.76620775e-01 6.05571389e-01 -8.10431778e-01 4.76196378e-01 7.97572672e-01 6.15564466e-01 -3.37444156e-01 7.90640891e-01 -3.25889021e-01 7.63618827e-01 -1.12787485e+00 -8.97991240e-01 7.53260493e-01 -1.48333341e-01 -7.19140112e-01 -6.98123515e-01 4.93995726e-01 -3.69876653e-01 -8.01903784e-01 1.50950205e+00 6.90277696e-01 -6.58218265e-02 6.54136002e-01 -1.36102629e+00 -1.04109025e+00 1.14232647e+00 -4.44954753e-01 5.08747756e-01 4.27487135e-01 5.65524101e-01 8.34104598e-01 -5.92742085e-01 2.11213931e-01 1.81631613e+00 5.47523737e-01 9.05882359e-01 -7.04233885e-01 -1.09536493e+00 4.72924531e-01 -2.48402636e-02 -9.54892933e-01 -2.59723157e-01 8.16402018e-01 -2.48948127e-01 7.94963419e-01 5.02558589e-01 1.86161492e-02 1.91565108e+00 6.23699486e-01 9.42757726e-01 9.93586302e-01 -3.93823117e-01 2.55031794e-01 3.09571475e-02 3.55773836e-01 3.71963441e-01 2.75892764e-01 4.81193453e-01 -2.19606832e-01 -8.04403186e-01 2.05117881e-01 -1.27762511e-01 -3.77928555e-01 2.65800238e-01 -6.99805617e-01 1.04297698e+00 3.33917588e-01 3.60996693e-01 -1.02028750e-01 9.66577083e-02 7.12534904e-01 3.91187489e-01 2.52838135e-01 7.44825482e-01 -6.01277232e-01 1.18598953e-01 -2.88954496e-01 2.64320046e-01 9.90844190e-01 8.50061834e-01 5.07705212e-02 3.50251377e-01 -2.43380070e-01 4.08832639e-01 3.62278372e-01 8.48997414e-01 5.89606225e-01 -3.36996108e-01 7.09034026e-01 2.98930883e-01 -3.25408787e-01 -1.35380554e+00 -4.99486662e-02 -3.29719305e-01 -7.71845698e-01 9.22646746e-02 4.20088112e-01 -6.25962853e-01 -7.94948220e-01 1.79587770e+00 1.97978720e-01 1.00152269e-01 3.45604241e-01 4.89739358e-01 5.27407765e-01 7.86541581e-01 4.55517530e-01 1.49209984e-02 1.30369961e+00 -6.55170202e-01 -7.06021845e-01 -4.09637064e-01 4.80093598e-01 -9.14386153e-01 1.27945185e+00 4.23934668e-01 -7.27731109e-01 -7.24409372e-02 -1.06849825e+00 3.26644599e-01 -8.08138728e-01 -7.14516997e-01 6.60931587e-01 1.38817096e+00 -4.07700479e-01 4.05947685e-01 -4.28442270e-01 -7.72752017e-02 4.19083357e-01 2.05423552e-04 -1.08972423e-01 1.29708737e-01 -1.96069503e+00 8.89043689e-01 4.29065585e-01 -1.82436243e-01 -1.20181358e+00 -7.86803484e-01 -7.19202697e-01 1.08625986e-01 3.36306930e-01 -5.72937131e-01 1.11204195e+00 -9.82032418e-01 -1.39819455e+00 3.78202438e-01 3.01545978e-01 -7.49347627e-01 5.77663302e-01 -4.18357670e-01 -7.77500153e-01 1.55797988e-01 -3.40666890e-01 -1.49201691e-01 1.20466685e+00 -1.24834502e+00 -2.45996729e-01 -3.33961964e-01 4.21702266e-01 -1.15748957e-01 -7.89442718e-01 4.90892023e-01 1.06814593e-01 -1.18274713e+00 -5.10338962e-01 -6.64213002e-01 -2.77889878e-01 -3.38161379e-01 -7.96496153e-01 -8.66851434e-02 1.03480577e+00 -2.76956379e-01 1.44207537e+00 -2.39776278e+00 -3.81582111e-01 3.36993098e-01 2.30891123e-01 8.13158512e-01 -3.57321352e-01 7.65700459e-01 -2.10418850e-01 7.62958109e-01 -1.66551396e-01 1.79468736e-01 4.60878849e-01 2.67511364e-02 -1.23430872e+00 4.78767753e-01 1.87120557e-01 8.85886371e-01 -1.02396786e+00 -2.72813857e-01 5.72133996e-02 2.19465688e-01 -4.58826482e-01 2.01780081e-01 -2.61174411e-01 -1.41780004e-01 -8.08212757e-01 6.48640871e-01 8.08229744e-01 3.05810362e-01 -5.79635017e-02 -5.98502643e-02 4.19058800e-01 2.05166459e-01 -1.04697978e+00 9.73230243e-01 -2.24189192e-01 5.11022985e-01 2.20635295e-01 -7.36167550e-01 7.14459598e-01 3.51830214e-01 -9.52612609e-02 -3.12035233e-01 3.79016340e-01 9.31326970e-02 2.49849379e-01 -4.22964334e-01 1.97833329e-01 -8.68485346e-02 -3.07402730e-01 6.35529935e-01 -1.97106883e-01 -8.40293616e-02 -1.00373924e-01 5.71910203e-01 1.51930249e+00 -6.18324876e-01 3.53392065e-01 -1.91801399e-01 6.60109043e-01 -2.20437527e-01 4.25787747e-01 1.36966872e+00 -5.97851515e-01 1.61406085e-01 4.31746811e-01 -4.68011081e-01 -6.72370136e-01 -1.31220007e+00 -7.66873881e-02 1.14753258e+00 -2.29350906e-02 -4.39770073e-01 -8.26345265e-01 -1.42141175e+00 9.39396918e-02 9.19277489e-01 -5.90643644e-01 -6.92720711e-01 -6.83511198e-01 -8.82610202e-01 1.46344924e+00 4.57240731e-01 4.79654729e-01 -1.16510379e+00 4.44191881e-02 -4.73201722e-02 7.62388185e-02 -1.07196200e+00 -6.58826113e-01 -1.37231946e-01 -4.08219367e-01 -1.08850551e+00 4.98628654e-02 -5.61707914e-01 5.57805896e-01 1.23603195e-01 9.52585459e-01 -9.56203938e-02 -2.58203268e-01 5.38142323e-01 -6.02850795e-01 -1.08627689e+00 -1.00352085e+00 -9.05710235e-02 3.49296480e-01 -7.15688467e-02 5.89694977e-01 -5.61972380e-01 -2.04693541e-01 5.59049025e-02 -1.35882246e+00 -8.90851498e-01 6.06551349e-01 5.42365551e-01 -5.24237826e-02 4.95133489e-01 7.89688289e-01 -1.11718166e+00 1.27875757e+00 -6.87287390e-01 -3.02121431e-01 1.75281733e-01 -3.10156167e-01 6.91486872e-04 1.30062079e+00 -9.03920472e-01 -9.32513773e-01 -3.13679546e-01 -4.25392181e-01 -4.13609594e-01 -3.75569880e-01 3.04481387e-01 -7.19922960e-01 -2.83214390e-01 9.74384725e-01 2.40920573e-01 -3.22930038e-01 -2.56477565e-01 6.06364429e-01 4.86749262e-01 3.78590941e-01 -7.98221171e-01 1.59523559e+00 3.32537919e-01 -3.10102075e-01 -7.05223978e-01 -8.47490549e-01 3.46183702e-02 -1.41431168e-01 1.56159773e-01 4.37036693e-01 -5.11954725e-01 -6.25112176e-01 6.25533044e-01 -1.34837723e+00 -4.94646840e-02 -4.74674925e-02 2.22066835e-01 -5.28835095e-02 9.65539813e-01 -8.34969282e-01 -8.06822121e-01 -7.40745723e-01 -9.33127880e-01 4.86426324e-01 -8.54214355e-02 -1.98441014e-01 -1.12346375e+00 1.42424166e-01 1.57643571e-01 5.22459984e-01 4.77049738e-01 1.26455879e+00 -1.51201999e+00 -1.84358895e-01 -5.48812389e-01 1.47996679e-01 5.69237351e-01 8.07693601e-02 2.50605315e-01 -1.05730391e+00 -2.03770399e-01 5.83390415e-01 -2.74488658e-01 5.66525221e-01 -1.03947580e-01 1.21070540e+00 -1.01365089e+00 -6.67567551e-02 4.70886588e-01 1.26473391e+00 1.81194052e-01 7.21785545e-01 4.31115568e-01 6.98704660e-01 5.56081831e-01 4.75274950e-01 3.03115577e-01 -2.42341414e-01 1.40292466e-01 8.08031797e-01 1.79327086e-01 2.89952338e-01 -3.75372142e-01 9.82984185e-01 4.75669801e-01 5.72629869e-01 -8.01159382e-01 -1.00475430e+00 3.44964862e-01 -1.48114705e+00 -1.26186478e+00 3.23487557e-02 1.84545958e+00 1.01392925e+00 4.50169563e-01 -2.34725177e-01 2.38849893e-01 7.49166310e-01 4.37409222e-01 -5.44437408e-01 -9.08344090e-01 -2.02553734e-01 2.39789665e-01 5.16621709e-01 4.19755429e-01 -1.31591964e+00 1.09586012e+00 7.01973057e+00 1.17061985e+00 -9.55894947e-01 -7.29501843e-02 3.07750195e-01 -1.54218510e-01 -2.88890123e-01 -2.59451210e-01 -7.49967217e-01 6.58517778e-01 8.81458402e-01 -5.73129475e-01 3.38959694e-01 9.87345815e-01 -7.64867337e-03 8.43487740e-01 -1.12367678e+00 4.38382477e-01 1.71173796e-01 -9.93502259e-01 7.57107496e-01 -4.70526814e-02 4.16754425e-01 -1.11798085e-01 3.89892638e-01 5.54476500e-01 9.80985284e-01 -1.08368766e+00 5.04270613e-01 -2.56307628e-02 7.39746913e-02 -9.84368980e-01 7.15297222e-01 5.92219710e-01 -7.64555454e-01 -3.56898963e-01 -2.65587151e-01 -1.17078447e-03 -2.98547223e-02 5.11081696e-01 -6.45738661e-01 4.83176827e-01 5.45800686e-01 1.90093845e-01 -4.55815077e-01 2.67930001e-01 -4.86657500e-01 9.04361665e-01 -2.03572273e-01 -9.89420563e-02 4.80647087e-01 3.74916404e-01 1.04532623e+00 1.36129928e+00 -1.24388367e-01 1.46277055e-01 3.77793074e-01 7.68019497e-01 -2.22725064e-01 2.04102471e-01 -1.00157213e+00 -4.08204079e-01 8.40963781e-01 1.05335486e+00 -2.80314416e-01 -1.51377112e-01 -2.97666639e-01 6.95181787e-01 1.34834841e-01 3.40553284e-01 -1.04766667e+00 -6.48686528e-01 8.31404686e-01 -1.93718940e-01 -2.15536505e-02 -1.35365814e-01 -2.45558217e-01 -1.17909563e+00 7.78015181e-02 -1.63188636e+00 5.36486983e-01 -2.57630914e-01 -2.05433774e+00 6.82000399e-01 -9.64648798e-02 -9.41305459e-01 -1.02924649e-03 -7.91332126e-01 -1.06850898e+00 7.69319057e-01 -1.24790263e+00 -1.15175557e+00 3.10678244e-01 9.74945903e-01 5.10659695e-01 -4.93491292e-01 8.79375458e-01 5.13802506e-02 -7.57843077e-01 1.15581489e+00 -2.11570673e-02 8.73278558e-01 8.41985583e-01 -1.17460418e+00 8.38186324e-01 1.44985712e+00 5.14622107e-02 1.13946843e+00 8.27147543e-01 -7.45748818e-01 -1.55916786e+00 -1.25247550e+00 5.02983749e-01 -8.12558413e-01 1.28478861e+00 -4.86600459e-01 -9.65231121e-01 6.94123089e-01 3.37047815e-01 -1.26965582e-01 9.19947386e-01 -7.61722773e-02 -9.82420444e-01 1.88056171e-01 -1.55137563e+00 1.00850070e+00 8.35775912e-01 -6.28553808e-01 -9.71572161e-01 4.83546406e-01 1.18986249e+00 -2.96892375e-01 -5.91823637e-01 3.46992582e-01 2.04970345e-01 -6.69684470e-01 1.19627941e+00 -1.06215990e+00 4.64021951e-01 -7.08157569e-02 -2.15118468e-01 -1.21932435e+00 -3.82158667e-01 -9.32602465e-01 -4.75350410e-01 1.45980120e+00 3.44896525e-01 -1.10334599e+00 3.39089125e-01 5.72910726e-01 -4.73609306e-02 -5.17510951e-01 -6.62160337e-01 -9.66963470e-01 6.62292063e-01 -5.59885740e-01 8.61907840e-01 1.28256679e+00 -1.59117103e-01 1.42396078e-01 -2.85746634e-01 8.98585856e-01 5.71546257e-01 -3.45570832e-01 7.40051866e-01 -8.51343751e-01 -4.36348528e-01 -4.76457447e-01 -2.23874882e-01 -5.66226661e-01 4.31620181e-01 -8.11117887e-01 -1.74546793e-01 -8.05590808e-01 -1.53284431e-01 -1.80744812e-01 -3.84587169e-01 5.40331244e-01 -5.50872982e-01 -5.17787738e-03 4.11740810e-01 1.32161334e-01 -2.11280137e-01 3.08376908e-01 8.41790438e-01 -5.41982353e-01 1.07485853e-01 -4.89477441e-02 -1.07314861e+00 1.11336577e+00 1.09084558e+00 -7.74461389e-01 -4.98605222e-01 -5.12507617e-01 3.52014333e-01 -6.83417082e-01 2.19516903e-01 -6.66135371e-01 1.20164998e-01 -4.09657061e-01 3.48996744e-02 -2.53475130e-01 -2.27051541e-01 -8.69010866e-01 -6.19794786e-01 5.98814368e-01 -5.34306824e-01 5.58836721e-02 4.21208292e-01 8.55884790e-01 4.15469185e-02 -4.68831420e-01 8.43935609e-01 -2.56549537e-01 -3.97399992e-01 4.62624490e-01 -6.97019756e-01 5.02654374e-01 1.16249537e+00 2.37280786e-01 -6.73951983e-01 -2.23773986e-01 -3.66132468e-01 1.40913799e-01 1.76589996e-01 7.07416475e-01 5.38590908e-01 -1.13728201e+00 -7.99023092e-01 1.58862621e-01 -1.04194768e-01 -4.31133837e-01 1.29988357e-01 -2.23277863e-02 -3.27693164e-01 8.78657922e-02 6.38271272e-02 6.70338646e-02 -1.19550765e+00 1.21127033e+00 3.92027915e-01 -5.96103132e-01 -4.44093347e-01 7.59431183e-01 5.48938751e-01 -3.73062432e-01 3.95561665e-01 2.02766657e-02 -1.91465646e-01 -3.94810528e-01 9.71797943e-01 2.36692935e-01 -2.46814653e-01 -3.40363860e-01 -3.74275774e-01 1.22501113e-01 -6.10829532e-01 2.41876245e-01 7.98824906e-01 1.45173356e-01 -1.41730011e-01 3.48609425e-02 8.99123192e-01 6.49186909e-01 -5.77985346e-01 -2.30805948e-01 -1.66400328e-01 -5.68541467e-01 -2.66623944e-01 -1.02103007e+00 -9.61297810e-01 8.70543957e-01 2.41127357e-01 6.22357965e-01 1.08922887e+00 -3.91583204e-01 1.08967710e+00 5.01924813e-01 2.31055900e-01 -7.74218738e-01 2.29855433e-01 9.17783141e-01 9.46624577e-01 -9.72070277e-01 -8.47871900e-02 -6.05590403e-01 -5.79027057e-01 1.02729475e+00 7.20581830e-01 -2.97468573e-01 6.65206790e-01 6.30572557e-01 3.28400403e-01 -4.39492166e-02 -7.34272957e-01 4.23751920e-01 2.61378773e-02 9.08954561e-01 6.10866025e-02 -3.89988162e-02 -1.98133215e-01 8.93728673e-01 -5.36261737e-01 -7.82114387e-01 5.06199598e-01 8.96259665e-01 -3.49493027e-01 -1.29693520e+00 -6.76936448e-01 6.83259442e-02 -1.03475082e+00 -4.79341269e-01 -1.05543923e+00 6.75532877e-01 -6.25765398e-02 1.25803614e+00 -5.10358155e-01 -4.92273241e-01 3.86989772e-01 8.94258395e-02 -2.70979069e-02 -6.34136200e-01 -1.24459195e+00 -5.67993999e-01 -8.97110850e-02 -5.29318154e-01 2.17182815e-01 -2.23002017e-01 -1.06001198e+00 -7.32607484e-01 -2.65057206e-01 2.30054423e-01 3.82295340e-01 8.74520838e-01 4.85744148e-01 3.49038243e-01 9.35866296e-01 -2.87246972e-01 -1.18105793e+00 -7.85456419e-01 -4.15406197e-01 6.08378947e-01 2.49153391e-01 -2.35821888e-01 -9.41338062e-01 -2.29638308e-01]
[5.956864356994629, 8.023102760314941]
1a97341f-76f1-49d1-8957-95686c9fdade
toward-multi-target-self-organizing-pursuit
2206.12330
null
https://arxiv.org/abs/2206.12330v3
https://arxiv.org/pdf/2206.12330v3.pdf
Toward multi-target self-organizing pursuit in a partially observable Markov game
The multiple-target self-organizing pursuit (SOP) problem has wide applications and has been considered a challenging self-organization game for distributed systems, in which intelligent agents cooperatively pursue multiple dynamic targets with partial observations. This work proposes a framework for decentralized multi-agent systems to improve the implicit coordination capabilities in search and pursuit. We model a self-organizing system as a partially observable Markov game (POMG) featured by large-scale, decentralization, partial observation, and noncommunication. The proposed distributed algorithm: fuzzy self-organizing cooperative coevolution (FSC2) is then leveraged to resolve the three challenges in multi-target SOP: distributed self-organizing search (SOS), distributed task allocation, and distributed single-target pursuit. FSC2 includes a coordinated multi-agent deep reinforcement learning (MARL) method that enables homogeneous agents to learn natural SOS patterns. Additionally, we propose a fuzzy-based distributed task allocation method, which locally decomposes multi-target SOP into several single-target pursuit problems. The cooperative coevolution principle is employed to coordinate distributed pursuers for each single-target pursuit problem. Therefore, the uncertainties of inherent partial observation and distributed decision-making in the POMG can be alleviated. The experimental results demonstrate that by decomposing the SOP task, FSC2 achieves superior performance compared with other implicit coordination policies fully trained by general MARL algorithms. The scalability of FSC2 is proved that up to 2048 FSC2 agents perform efficient multi-target SOP with almost 100 percent capture rates. Empirical analyses and ablation studies verify the interpretability, rationality, and effectiveness of component algorithms in FSC2.
['Chin-Teng Lin', 'Yuhui Shi', 'Ye Shi', 'Chao Lyu', 'Yu-Cheng Chang', 'Lijun Sun']
2022-06-24
null
null
null
null
['multi-agent-path-finding']
['playing-games']
[-4.65391040e-01 1.35185085e-02 -1.14086419e-01 3.80970359e-01 -5.06710649e-01 -5.16894102e-01 4.44039971e-01 -2.28163928e-01 -2.17267990e-01 7.26469159e-01 -5.77073209e-02 1.64054960e-01 -8.00097346e-01 -5.40894687e-01 -2.89649993e-01 -1.38308370e+00 -5.88132918e-01 1.00664437e+00 2.09204748e-01 -6.18190050e-01 5.63532710e-02 2.24449635e-01 -1.57089925e+00 -3.17921788e-01 1.35896206e+00 9.12546933e-01 6.44995749e-01 5.68361998e-01 4.64263648e-01 1.02037156e+00 -6.50861084e-01 2.77954072e-01 6.27579153e-01 -2.32044652e-01 -2.33180225e-01 2.41048276e-01 -5.63190341e-01 -4.48958613e-02 -2.01245502e-01 1.22491813e+00 5.96054494e-01 3.90291631e-01 7.43994474e-01 -1.83244848e+00 -9.86253679e-01 4.17939097e-01 -6.86432779e-01 1.65149808e-01 -2.04909563e-01 5.07389903e-01 7.82043278e-01 -4.23817754e-01 3.60763550e-01 1.34884524e+00 4.32390451e-01 5.77363968e-01 -9.58200514e-01 -5.46591103e-01 1.26937047e-01 2.98520684e-01 -1.39300931e+00 -1.68491662e-01 7.27874219e-01 -2.05474094e-01 8.27778757e-01 1.86135173e-01 1.05947912e+00 6.86034799e-01 6.80760324e-01 9.86350656e-01 1.16821849e+00 -3.81935090e-01 7.78206825e-01 -2.97697842e-01 -2.57637888e-01 6.77643418e-01 5.66122353e-01 6.58054233e-01 -3.83838773e-01 -4.85245198e-01 9.31118131e-01 1.73193023e-01 1.25000374e-02 -4.51980531e-01 -1.08892548e+00 9.83637094e-01 4.98418242e-01 2.44713619e-01 -1.07115364e+00 2.21729115e-01 3.46243568e-03 8.25676084e-01 2.68477261e-01 7.49757409e-01 -5.83880581e-02 9.32566971e-02 -5.90519965e-01 3.33848655e-01 8.97589743e-01 8.94846022e-01 8.24971378e-01 5.88538766e-01 4.27699499e-02 4.46750134e-01 4.27621484e-01 1.03001642e+00 8.01721931e-01 -1.48872840e+00 -1.77234367e-01 6.76157415e-01 3.75137568e-01 -1.37577713e+00 -5.55507779e-01 -6.04219973e-01 -9.54244137e-01 6.53689861e-01 -2.53927916e-01 -6.14046633e-01 -4.92130667e-01 1.60313630e+00 4.99282509e-01 7.57986978e-02 7.12415516e-01 1.16702640e+00 1.43517509e-01 6.43217802e-01 -3.05254847e-01 -6.63384557e-01 1.01075590e+00 -1.35413730e+00 -5.52391827e-01 -1.01711936e-01 1.46825328e-01 -1.98331609e-01 4.02653486e-01 1.41958773e-01 -8.03060412e-01 -1.42689973e-01 -8.07836652e-01 8.38604867e-01 1.43260986e-01 -2.84689188e-01 7.53194928e-01 3.34367275e-01 -1.42773557e+00 -7.12626614e-03 -9.71504748e-01 -2.75755525e-01 2.18255460e-01 6.52391016e-01 1.28370106e-01 2.24495858e-01 -1.01163661e+00 6.84916198e-01 2.75886416e-01 -1.14989616e-01 -1.55959344e+00 -1.53916150e-01 -4.02163833e-01 2.89854199e-01 8.51431847e-01 -8.60278547e-01 1.05207860e+00 -1.18460393e+00 -1.85412693e+00 2.55604893e-01 1.31302670e-01 -6.11679196e-01 1.74580008e-01 3.53639752e-01 -2.15586022e-01 3.76411378e-01 5.92346787e-01 3.95039052e-01 1.00534940e+00 -1.37675142e+00 -8.50859404e-01 -3.90413076e-01 -3.15763615e-02 1.03710592e+00 -3.93592000e-01 -3.51639800e-02 1.40433297e-01 -6.02319777e-01 -6.67502657e-02 -1.10888064e+00 -5.28033197e-01 -3.82657588e-01 8.51810351e-02 -6.89982593e-01 8.88291180e-01 1.02984518e-01 8.30277085e-01 -2.03470922e+00 5.21281838e-01 8.95562768e-02 5.90641499e-01 -1.10837720e-01 -4.70289826e-01 5.95505059e-01 6.05199695e-01 -3.08734208e-01 5.76677509e-02 -1.92396745e-01 1.65441200e-01 4.09167260e-01 -4.09670472e-02 2.67775744e-01 -1.59299523e-01 1.02972591e+00 -1.06091177e+00 -4.49919134e-01 -3.84609103e-01 -3.00535977e-01 -3.29913169e-01 7.13648051e-02 -4.43690509e-01 2.73954183e-01 -1.04781711e+00 1.19192874e+00 3.80118459e-01 -4.83839810e-01 3.24046224e-01 6.51796401e-01 -4.81500804e-01 -8.55189860e-01 -1.18708551e+00 1.49575949e+00 -1.79023463e-02 3.46486866e-01 7.50617981e-01 -1.21992481e+00 1.17290688e+00 3.76733601e-01 9.88200963e-01 -6.69967055e-01 9.00627077e-02 5.11156082e-01 2.21704081e-01 -4.28234518e-01 4.38620389e-01 -1.02651581e-01 -1.68467730e-01 9.35417175e-01 4.05594707e-02 -2.52386957e-01 -8.32000840e-03 2.29609534e-02 1.35535800e+00 -3.71210933e-01 2.78179705e-01 -8.74690354e-01 1.52823657e-01 7.44043052e-01 1.03276122e+00 1.06577134e+00 -8.49272132e-01 -1.55391723e-01 1.18285255e-03 -6.10464156e-01 -5.20370424e-01 -8.13315094e-01 6.80707633e-01 1.23660696e+00 8.92517805e-01 1.30272076e-01 -3.85303885e-01 -1.90015107e-01 9.47458521e-02 2.83093721e-01 -6.43454254e-01 -1.67770281e-01 -4.07404393e-01 -8.13346267e-01 4.71549571e-01 -2.07303673e-01 7.96649277e-01 -1.49574649e+00 -1.23489773e+00 4.69170004e-01 -1.43376172e-01 -2.91889489e-01 -7.00613439e-01 1.88465774e-01 -5.43412983e-01 -1.28230858e+00 -8.84983003e-01 -1.24401617e+00 3.72582883e-01 8.68699253e-01 6.38880730e-01 -4.95209694e-02 1.37388974e-01 9.32547927e-01 -4.83709753e-01 -7.00193286e-01 -2.38625556e-01 -7.17349425e-02 8.17137420e-01 2.98938066e-01 3.55115414e-01 -8.38197887e-01 -5.91475785e-01 6.48054719e-01 -8.39109004e-01 -9.31364968e-02 9.26751554e-01 1.07843018e+00 5.76655746e-01 5.57972848e-01 8.88771951e-01 8.19573998e-02 1.18132842e+00 -6.20240271e-01 -6.92105293e-01 5.70173860e-01 -7.98347116e-01 3.86995673e-02 7.42585599e-01 -9.04845893e-01 -1.04949212e+00 -1.84563965e-01 1.07606292e+00 -9.55085993e-01 8.33548009e-02 6.22968376e-01 3.87489200e-01 -4.92318958e-01 8.58850479e-01 9.25783038e-01 7.71773994e-01 1.39453456e-01 -1.14250351e-02 6.71291828e-01 2.24785596e-01 -7.46840656e-01 7.79416144e-01 4.59096789e-01 9.67540145e-02 -6.99289382e-01 -3.06404233e-02 -2.75451809e-01 6.30894303e-02 -4.89155531e-01 4.62355226e-01 -9.54484403e-01 -1.33246326e+00 8.54341328e-01 -9.19465899e-01 -5.73554993e-01 -5.37483573e-01 5.87241769e-01 -9.89193916e-01 1.10984042e-01 -3.80288661e-01 -1.18307364e+00 -6.16693199e-01 -1.00138414e+00 5.06538332e-01 7.55041778e-01 4.44201797e-01 -8.74292374e-01 6.23464465e-01 6.91138506e-02 8.74440849e-01 2.15949133e-01 4.00891066e-01 -6.03391290e-01 -7.14258552e-01 3.39424551e-01 3.91676784e-01 -3.00543517e-01 2.35906690e-01 -4.03902709e-01 -6.09450080e-02 -7.30325937e-01 4.79268104e-01 -7.02617824e-01 4.89304543e-01 8.14686358e-01 -1.67186484e-01 -6.13796115e-01 -3.35398793e-01 4.46170986e-01 1.42127979e+00 7.62934268e-01 2.40224749e-02 9.26600635e-01 3.59427221e-02 6.63055599e-01 9.24519002e-01 7.89772570e-01 6.84729695e-01 3.56912434e-01 9.28198695e-01 2.61987180e-01 2.84566402e-01 1.30649492e-01 6.16023898e-01 8.16546679e-01 -4.91998225e-01 -5.24218567e-02 -8.89195800e-01 6.11378431e-01 -2.47428560e+00 -1.28359854e+00 4.15899903e-01 1.62049365e+00 5.18060327e-01 -5.93103230e-01 2.22455308e-01 -6.01602674e-01 1.00318408e+00 6.99955896e-02 -1.19241142e+00 1.41932115e-01 -6.86835706e-01 -4.50497538e-01 2.64036804e-01 3.69692087e-01 -8.55271578e-01 1.08398449e+00 5.65834379e+00 1.04018676e+00 -1.21929288e+00 2.21562847e-01 2.91588873e-01 -1.93282053e-01 -2.57279463e-02 -8.87458473e-02 -3.41234505e-01 4.61997300e-01 4.71386850e-01 -7.53843248e-01 1.01949763e+00 9.50594664e-01 4.01430190e-01 -1.01892771e-02 -2.18951359e-01 1.11618745e+00 3.22129764e-02 -1.30021346e+00 -2.29378000e-01 2.86655098e-01 1.14134216e+00 4.96805727e-01 2.89429814e-01 4.42112178e-01 1.24191630e+00 -5.68590105e-01 7.21466184e-01 4.98693705e-01 3.82378548e-01 -4.54896957e-01 4.14303362e-01 8.34132552e-01 -1.33236933e+00 -9.35188293e-01 -6.46247208e-01 -3.09435755e-01 9.76322293e-02 -2.10630760e-01 -2.75220722e-01 8.03536236e-01 8.35709453e-01 5.19327223e-01 -6.48480728e-02 9.03608143e-01 1.55470923e-01 1.62843972e-01 -6.56793118e-01 -6.33474767e-01 6.21616244e-01 -5.12582123e-01 1.26922512e+00 7.90650696e-02 3.55356842e-01 5.23833573e-01 7.17072725e-01 7.71363735e-01 3.92173797e-01 -1.26951933e-01 -5.32963514e-01 2.92540528e-02 8.40359032e-01 1.24688971e+00 -5.98044336e-01 -2.13491574e-01 1.92098469e-02 5.68933129e-01 3.84611279e-01 4.21059668e-01 -5.18871248e-01 -2.50016510e-01 5.55787683e-01 -4.79715288e-01 2.87454516e-01 -1.18290320e-01 2.02274416e-02 -1.33676505e+00 -2.52683520e-01 -1.13844097e+00 5.31940043e-01 -5.31044543e-01 -1.46068299e+00 9.37294900e-01 -3.33915532e-01 -1.43630731e+00 -4.11474466e-01 -2.61788070e-02 -9.39209878e-01 3.82356435e-01 -1.47612345e+00 -1.29332614e+00 -2.10047707e-01 1.17888081e+00 6.88334465e-01 -1.18053436e+00 8.16742957e-01 -4.56101596e-01 -5.27430654e-01 1.48816839e-01 7.02926695e-01 -4.06208962e-01 3.11585456e-01 -9.76884425e-01 -5.46527028e-01 7.65981019e-01 -1.40786082e-01 2.16837153e-01 5.55131495e-01 -7.86389828e-01 -1.75735497e+00 -9.78847265e-01 3.54405865e-02 2.38571703e-01 1.00694692e+00 2.54640639e-01 -2.92991132e-01 2.73181468e-01 3.66780132e-01 -1.63710207e-01 3.83318245e-01 -1.04231477e-01 1.58534124e-01 -1.62940055e-01 -1.04274917e+00 5.34963012e-01 8.18957269e-01 1.93448648e-01 -6.55492127e-01 3.94686878e-01 7.90883005e-01 8.42701346e-02 -4.13234144e-01 4.84606102e-02 2.61945218e-01 -7.61954129e-01 5.85828066e-01 -5.21111310e-01 -1.34661468e-02 -4.70404625e-01 -2.65197158e-01 -1.68803513e+00 -9.99382854e-01 -1.21475446e+00 -2.27716804e-01 4.98655468e-01 -9.05175284e-02 -1.10026908e+00 8.47977698e-01 2.81030625e-01 -4.24204588e-01 -4.83374149e-01 -1.28906035e+00 -1.13773012e+00 -3.06400619e-02 4.62455601e-01 3.90542746e-01 1.02012408e+00 1.89629674e-01 2.08502024e-01 -5.28161883e-01 3.82939994e-01 1.08996689e+00 6.72433376e-01 7.20682263e-01 -1.29140294e+00 -7.13255167e-01 -4.58509743e-01 -1.11691140e-01 -9.74955559e-01 4.41675037e-02 -5.94994307e-01 2.09014222e-01 -1.17688310e+00 8.45315605e-02 -8.02018940e-01 -4.25079256e-01 6.00160658e-01 7.77655393e-02 -2.31044754e-01 4.57516432e-01 1.38372171e+00 -1.25307631e+00 1.11256361e+00 1.39476335e+00 -5.53170502e-01 -7.77323186e-01 1.19084515e-01 -7.67635405e-01 3.92377168e-01 8.68089259e-01 -5.84285855e-01 -3.55834007e-01 -5.75650990e-01 -3.06801170e-01 6.13099337e-01 1.91168368e-01 -1.03268659e+00 1.13446605e+00 -8.26553643e-01 -5.01680255e-01 -2.27224082e-01 2.56057918e-01 -9.40231740e-01 2.36662790e-01 1.16350377e+00 1.15460761e-01 9.17111859e-02 -2.75775403e-01 1.04897773e+00 -4.25421089e-01 -2.09579505e-02 4.31726098e-01 -4.08511668e-01 -6.97519243e-01 3.35062951e-01 -6.98700011e-01 4.12912630e-02 1.65645909e+00 -1.59952074e-01 -4.30209488e-01 -6.26425624e-01 -7.67351866e-01 1.03593159e+00 3.37899446e-01 1.22541748e-01 6.01929247e-01 -1.27329314e+00 -8.85123849e-01 -2.91974517e-03 -2.65940994e-01 3.57629396e-02 4.25185174e-01 1.00056207e+00 -5.54601729e-01 3.67687523e-01 -5.92202127e-01 -5.05516589e-01 -8.42757881e-01 6.50739789e-01 5.10108590e-01 -5.63249409e-01 -3.47045869e-01 5.47768831e-01 3.02323163e-01 -4.43244249e-01 -1.08742319e-01 5.46888351e-01 -2.52772987e-01 -1.02148622e-01 3.54961067e-01 5.88709950e-01 -6.85621440e-01 -4.19564962e-01 -1.58118024e-01 4.92759466e-01 1.14073500e-01 -1.29121214e-01 1.83772218e+00 -3.45243365e-01 -3.41885477e-01 -1.06301177e-02 2.08709940e-01 -4.58928257e-01 -1.52519143e+00 -4.25335377e-01 -3.31221044e-01 -2.91614141e-02 -1.10392302e-01 -6.47161841e-01 -9.76007998e-01 7.41788819e-02 3.16227555e-01 6.65931940e-01 1.20837426e+00 -1.83094926e-02 3.54270428e-01 7.99076557e-01 9.98279810e-01 -1.24695766e+00 4.82161433e-01 8.06471109e-01 1.03130853e+00 -1.21620858e+00 -4.73957896e-01 6.82991371e-02 -9.41504717e-01 1.07475734e+00 7.72741795e-01 -6.01972461e-01 6.38672769e-01 1.59401968e-01 1.08104579e-01 -5.47882795e-01 -1.19121540e+00 -1.26583308e-01 -3.29840064e-01 9.55249608e-01 -1.02169228e+00 2.89629668e-01 -8.57916698e-02 6.92868471e-01 9.83069837e-02 -1.46708310e-01 5.49023330e-01 1.10802114e+00 -8.37361813e-01 -5.46662211e-01 -6.75896943e-01 1.88928872e-01 1.87803745e-01 7.82353058e-02 -2.86471009e-01 5.82809865e-01 -2.35778689e-01 1.19235969e+00 1.12888649e-01 -3.78271431e-01 -2.28575289e-01 -6.74708664e-01 -6.63307905e-02 -2.60191977e-01 -5.37856936e-01 3.67698818e-01 -6.26933098e-01 -2.41597101e-01 -7.09762335e-01 -5.40331483e-01 -1.09882295e+00 -3.68427098e-01 -3.77858013e-01 7.67085910e-01 7.23007917e-02 7.33254135e-01 9.73677278e-01 2.18279287e-01 1.07382882e+00 -9.27607119e-01 -1.16602588e+00 -9.14902747e-01 -9.93343592e-01 6.61887752e-04 1.45918339e-01 -1.02419639e+00 -6.00362241e-01 -3.64357114e-01]
[3.747713565826416, 1.995650291442871]
23b5aa20-ffe7-4976-b977-880c34b1f75d
hypercon-image-to-video-model-transfer-for
1912.04950
null
https://arxiv.org/abs/1912.04950v2
https://arxiv.org/pdf/1912.04950v2.pdf
HyperCon: Image-To-Video Model Transfer for Video-To-Video Translation Tasks
Video-to-video translation is more difficult than image-to-image translation due to the temporal consistency problem that, if unaddressed, leads to distracting flickering effects. Although video models designed from scratch produce temporally consistent results, training them to match the vast visual knowledge captured by image models requires an intractable number of videos. To combine the benefits of image and video models, we propose an image-to-video model transfer method called Hyperconsistency (HyperCon) that transforms any well-trained image model into a temporally consistent video model without fine-tuning. HyperCon works by translating a temporally interpolated video frame-wise and then aggregating over temporally localized windows on the interpolated video. It handles both masked and unmasked inputs, enabling support for even more video-to-video translation tasks than prior image-to-video model transfer techniques. We demonstrate HyperCon on video style transfer and inpainting, where it performs favorably compared to prior state-of-the-art methods without training on a single stylized or incomplete video. Our project website is available at https://ryanszeto.com/projects/hypercon .
['Mostafa El-Khamy', 'Jungwon Lee', 'Ryan Szeto', 'Jason J. Corso']
2019-12-10
null
null
null
null
['video-style-transfer', 'video-inpainting']
['computer-vision', 'computer-vision']
[ 3.90134454e-01 -4.96944450e-02 -3.53984505e-01 -1.23884499e-01 -9.77502048e-01 -7.14831948e-01 6.53348863e-01 -7.73938060e-01 -1.12269912e-02 7.47541726e-01 2.34109417e-01 -1.45010576e-01 6.43078208e-01 -2.84759641e-01 -1.38954973e+00 -2.85671383e-01 2.61880964e-01 1.42011181e-01 2.02324867e-01 2.01747447e-01 1.10227123e-01 1.27709597e-01 -1.49541163e+00 8.36044729e-01 4.92472768e-01 7.13548362e-01 3.43707472e-01 1.00781488e+00 -5.31797670e-02 1.20004523e+00 -3.83640051e-01 -2.26771504e-01 2.64942646e-01 -9.29260492e-01 -6.38059378e-01 5.02818346e-01 1.07668364e+00 -7.16177821e-01 -5.12418449e-01 7.77798474e-01 1.63983792e-01 7.26697743e-02 4.65379208e-01 -1.52152431e+00 -1.13982332e+00 1.95039913e-01 -6.26998603e-01 1.45811453e-01 5.83404005e-01 3.55597854e-01 4.32416141e-01 -1.23246717e+00 1.21339679e+00 1.29730034e+00 3.94113481e-01 7.47440159e-01 -1.62191606e+00 -7.38272786e-01 9.70168486e-02 2.63959348e-01 -1.26726687e+00 -8.60085666e-01 5.42762339e-01 -5.65998793e-01 8.82092416e-01 1.63758829e-01 7.84067810e-01 1.39317071e+00 1.18630268e-01 7.38021672e-01 1.02045035e+00 -4.97901797e-01 -1.09664403e-01 1.30489141e-01 -6.00568712e-01 6.93649590e-01 -2.08160505e-01 3.37107658e-01 -9.40986276e-01 7.29844943e-02 1.31839573e+00 2.94981450e-02 -4.70479935e-01 -3.26831490e-01 -1.31869543e+00 3.79001707e-01 5.71694970e-02 9.28967819e-02 -2.74894565e-01 6.88956797e-01 1.98553771e-01 6.39341116e-01 5.17543077e-01 1.68950096e-01 -3.10920358e-01 -3.32491308e-01 -1.37347758e+00 2.27691889e-01 3.81556779e-01 1.20436800e+00 9.60642040e-01 3.70226324e-01 -1.92690924e-01 4.50359970e-01 -1.24789830e-02 6.07510090e-01 3.51882994e-01 -1.71074843e+00 3.51921856e-01 -1.53282523e-01 3.79364133e-01 -6.98449194e-01 4.05387819e-01 1.30559564e-01 -3.68285954e-01 3.88077885e-01 2.07434684e-01 -6.87342659e-02 -1.00372744e+00 1.86072958e+00 2.21190482e-01 7.02618480e-01 -2.02212453e-01 8.20510924e-01 5.24632752e-01 9.93590534e-01 -2.97980029e-02 -4.25921768e-01 9.98366594e-01 -1.35416782e+00 -8.92945051e-01 -1.93457529e-01 2.72448689e-01 -1.05707967e+00 1.28728747e+00 2.72272080e-01 -1.69553947e+00 -6.84942126e-01 -7.95510113e-01 -2.97490001e-01 8.11713859e-02 -2.67688096e-01 5.55687360e-02 1.63718343e-01 -1.50619900e+00 5.09288549e-01 -9.67604518e-01 -3.00093591e-01 2.68159240e-01 3.78529392e-02 -7.25973725e-01 -2.63141304e-01 -9.41373765e-01 8.50647926e-01 1.94267914e-01 -1.81932330e-01 -1.12459576e+00 -9.43038225e-01 -1.06639528e+00 -1.90644190e-01 3.91345561e-01 -1.01785469e+00 1.64527929e+00 -1.76461291e+00 -1.64065611e+00 9.18906808e-01 -6.37442052e-01 -3.31267923e-01 7.62342274e-01 -4.83754367e-01 -3.28455508e-01 6.19840443e-01 8.18464458e-02 1.20382833e+00 1.55099046e+00 -1.49754262e+00 -3.51567179e-01 2.44854376e-01 -1.62297428e-01 2.89123267e-01 -1.18953235e-01 2.03779817e-01 -1.01095319e+00 -9.35075164e-01 -4.69315588e-01 -9.27365482e-01 2.13814259e-01 4.48224068e-01 -4.36207131e-02 4.86200094e-01 1.34728932e+00 -9.39815223e-01 1.11424458e+00 -2.17829037e+00 4.18640971e-01 -3.83590162e-01 2.56378744e-02 2.81424075e-01 -5.25612175e-01 6.82426274e-01 -2.34256953e-01 7.43729770e-02 -1.32456392e-01 -5.26904225e-01 -3.61901879e-01 1.24450766e-01 -4.23116535e-01 3.14677864e-01 2.54973590e-01 1.13678145e+00 -9.07787621e-01 -7.48968899e-01 4.38300222e-01 7.46968865e-01 -8.09156001e-01 4.03536528e-01 -4.90804315e-01 7.39856005e-01 -3.25540788e-02 6.30721331e-01 5.91085255e-01 -3.66189539e-01 -1.58561859e-02 -1.96722955e-01 6.40541986e-02 -1.78538427e-01 -6.08881295e-01 2.04833126e+00 -4.89431798e-01 1.09376478e+00 1.82766229e-01 -5.07422626e-01 3.24921280e-01 6.35733485e-01 4.74864751e-01 -6.75867677e-01 -1.72753915e-01 1.28848031e-01 -5.89550316e-01 -7.17095673e-01 5.12314796e-01 -2.23468184e-01 3.38145405e-01 5.14923990e-01 2.34788358e-01 -3.92605186e-01 1.95524424e-01 4.03486490e-01 9.63033915e-01 8.56218457e-01 -1.47855058e-02 2.10348696e-01 4.52735201e-02 7.73366317e-02 4.12750036e-01 4.67048019e-01 -1.75154135e-01 1.09874916e+00 1.69613212e-01 -3.08072269e-01 -1.42619812e+00 -1.22342062e+00 2.28890494e-01 8.82317960e-01 6.73003942e-02 -5.14160395e-01 -1.07770300e+00 -4.98986989e-01 -4.79452237e-02 6.72664285e-01 -5.72366118e-01 -6.18041530e-02 -6.14045024e-01 2.45990321e-01 3.98885310e-01 4.65826005e-01 3.14446718e-01 -1.11820257e+00 -4.31718290e-01 1.60474837e-01 -4.96003270e-01 -1.29022169e+00 -1.23746014e+00 -3.86902571e-01 -8.27257156e-01 -7.28572011e-01 -9.76251960e-01 -9.43897307e-01 6.61909282e-01 6.97709620e-01 1.32554400e+00 1.55199528e-01 -1.26007795e-01 6.75061703e-01 -3.69946003e-01 2.27063112e-02 -7.24210858e-01 -4.76330549e-01 -8.92191473e-03 2.79105511e-02 4.16141413e-02 -6.74354017e-01 -6.34927869e-01 4.51882869e-01 -1.27232301e+00 7.13975728e-01 3.82475257e-01 8.84792209e-01 6.93279505e-01 -5.06983638e-01 2.39082202e-01 -6.72030389e-01 2.31963918e-01 -2.79744297e-01 -3.57397288e-01 2.89634228e-01 -2.62680024e-01 -8.94603804e-02 5.17669499e-01 -7.88752198e-01 -1.08820963e+00 1.20811380e-01 4.28917527e-01 -1.34051692e+00 -5.49721234e-02 1.99628443e-01 1.05406947e-01 1.16985450e-02 5.53368628e-01 2.31673241e-01 3.15273494e-01 -1.84667230e-01 7.56135523e-01 4.91006166e-01 9.54401731e-01 -5.35956144e-01 1.01541460e+00 5.36624789e-01 -4.14504945e-01 -7.91518450e-01 -5.08090615e-01 -4.78384905e-02 -5.04525721e-01 -4.27567273e-01 1.04223275e+00 -1.20544171e+00 -2.09194139e-01 4.15429533e-01 -1.35525274e+00 -8.81986737e-01 -2.80174404e-01 3.32677305e-01 -9.94698703e-01 3.06204498e-01 -8.74947846e-01 -2.64647603e-01 -3.86819802e-02 -1.04546511e+00 1.23724699e+00 6.62683416e-03 -4.14533198e-01 -9.14156079e-01 5.63869067e-02 2.99893409e-01 4.43195760e-01 2.14569509e-01 5.22424877e-01 2.74668038e-01 -9.21713233e-01 -1.31533211e-02 -1.67914882e-01 4.08723921e-01 9.10110474e-02 3.66324663e-01 -8.48463774e-01 -4.16965067e-01 -6.84894919e-02 -6.14674985e-01 7.94990122e-01 5.58299839e-01 9.42882478e-01 -3.79229486e-01 -1.32342190e-01 7.08336294e-01 1.34719586e+00 9.11699235e-02 8.56649697e-01 2.13960484e-01 7.85310626e-01 2.98488855e-01 5.64935684e-01 1.36806339e-01 2.16170818e-01 9.88931060e-01 1.47933960e-01 -3.58311355e-01 -6.24636114e-01 -7.23935664e-01 9.43093896e-01 6.57684684e-01 -4.68152985e-02 -2.57891566e-01 -4.44268703e-01 4.60191548e-01 -1.86236167e+00 -1.39962757e+00 2.70981461e-01 2.12960482e+00 9.43168581e-01 -2.34342009e-01 4.43757363e-02 -4.09354866e-01 9.04482543e-01 1.56452864e-01 -5.19079268e-01 -3.65885645e-01 -8.81078243e-02 7.03241006e-02 2.84114361e-01 7.63303041e-01 -8.27030778e-01 1.19477642e+00 6.59098625e+00 7.68868804e-01 -1.22710693e+00 4.22978014e-01 6.14577711e-01 -5.34711599e-01 -4.58262473e-01 1.96975484e-01 -2.03594312e-01 5.35264552e-01 9.68461573e-01 -3.03835750e-01 7.97436893e-01 5.12717426e-01 6.63828373e-01 -8.13965425e-02 -1.35952413e+00 1.20004082e+00 3.47683966e-01 -1.60962939e+00 2.66186088e-01 8.65619443e-03 1.14442706e+00 -1.50778666e-01 2.47220472e-01 8.95241871e-02 1.34185761e-01 -1.07164264e+00 1.29248595e+00 5.71189225e-01 1.32051682e+00 -3.04693013e-01 -2.38446351e-02 1.39764585e-02 -1.15102851e+00 1.73358038e-01 -1.10839330e-01 3.79101858e-02 6.35388851e-01 1.29570544e-01 -4.02228385e-01 3.22405010e-01 6.41730905e-01 8.68431568e-01 -3.09793115e-01 7.21661568e-01 -2.00368971e-01 5.15207231e-01 1.86040588e-02 6.81270719e-01 1.25098750e-01 -1.60254464e-01 3.73189569e-01 1.15554082e+00 6.18502378e-01 -3.98447458e-03 9.92259309e-02 9.37101483e-01 -2.42209092e-01 -1.73095003e-01 -9.57309484e-01 -1.25497669e-01 5.34392476e-01 8.93220127e-01 -4.27442610e-01 -6.39007509e-01 -6.36585176e-01 1.70221066e+00 1.23515934e-01 8.59912157e-01 -1.02392590e+00 2.86594965e-02 5.70151150e-01 3.64651471e-01 5.38083851e-01 -3.30161601e-01 1.08462930e-01 -1.47382021e+00 1.04277350e-01 -1.10563743e+00 3.92160155e-02 -1.39236927e+00 -8.95876706e-01 6.29689336e-01 1.31682351e-01 -1.54382253e+00 -6.97730780e-01 -2.53189474e-01 -5.73100269e-01 6.94327891e-01 -1.35917556e+00 -1.36896431e+00 -4.00459379e-01 7.34271407e-01 9.74977255e-01 1.53635889e-01 6.59159780e-01 3.22123885e-01 -1.76813990e-01 5.00087857e-01 4.01981473e-02 -1.74931437e-01 1.29188371e+00 -8.23260963e-01 5.51986396e-01 9.99770820e-01 2.16510922e-01 2.96347469e-01 8.58479440e-01 -6.38902009e-01 -1.54483962e+00 -1.12200451e+00 6.97902918e-01 -7.37115860e-01 4.89038825e-01 -1.94582596e-01 -9.30653691e-01 1.00434852e+00 8.27228367e-01 2.25709632e-01 3.10563743e-01 -6.96717322e-01 -5.35726190e-01 1.63170755e-01 -8.57409716e-01 8.48601341e-01 1.05962622e+00 -9.02373374e-01 -3.39098960e-01 1.79120705e-01 9.67438936e-01 -5.37965655e-01 -6.39658391e-01 -7.98389837e-02 5.90057373e-01 -9.44175839e-01 1.00227821e+00 -5.18598199e-01 9.74439621e-01 -5.36294281e-01 -1.99101806e-01 -1.27649140e+00 -2.79673249e-01 -1.19492507e+00 -3.11679989e-01 1.08822107e+00 2.97957033e-01 -9.81439948e-02 5.84705114e-01 6.92218602e-01 -1.90658078e-01 -2.65354842e-01 -6.52242422e-01 -1.01304018e+00 5.65384775e-02 -3.73694271e-01 1.58075199e-01 1.03592575e+00 1.04352478e-02 1.41936392e-01 -9.54960525e-01 -5.76934963e-02 6.14661098e-01 3.13854218e-02 1.00848103e+00 -3.37284565e-01 -6.07281685e-01 -2.22566992e-01 -1.86646819e-01 -1.04321063e+00 1.80516765e-01 -6.85723186e-01 -3.17652300e-02 -1.29051614e+00 2.84579694e-01 2.65657693e-01 3.31193469e-02 4.32499647e-01 -1.32015884e-01 7.87892103e-01 5.35283566e-01 4.67084169e-01 -5.91296315e-01 4.26439553e-01 1.59368682e+00 2.26923451e-02 -2.49199420e-01 -6.13177836e-01 -3.97295952e-01 4.71319854e-01 4.62559849e-01 -3.30476671e-01 -6.23921156e-01 -8.05261254e-01 -2.03642070e-01 6.20649755e-01 5.48037231e-01 -8.46947312e-01 -9.55975894e-03 -4.09240752e-01 4.12275732e-01 -3.10574889e-01 5.77471912e-01 -6.60593688e-01 8.61102462e-01 3.51494372e-01 -3.85041505e-01 3.98331165e-01 3.55104595e-01 5.92139363e-01 -4.48969394e-01 8.43487531e-02 8.49379420e-01 -2.30121672e-01 -7.46620119e-01 3.69858861e-01 -5.02858281e-01 6.07342273e-02 1.11186504e+00 -4.16530550e-01 -4.27467376e-01 -9.43860531e-01 -6.96952641e-01 -2.86267418e-03 1.17899001e+00 5.24892807e-01 7.54942417e-01 -1.58646369e+00 -6.39261663e-01 2.94216663e-01 -6.26688227e-02 -3.89008492e-01 5.27334332e-01 7.43857920e-01 -8.44977260e-01 1.70533508e-01 -3.66228878e-01 -7.16097534e-01 -1.44250429e+00 8.15940619e-01 2.41942942e-01 1.94404781e-01 -8.44114304e-01 6.46098912e-01 5.10013938e-01 2.02798933e-01 1.30983725e-01 -1.01231024e-01 6.57719731e-01 -3.59982938e-01 7.74389803e-01 8.49632844e-02 -2.73511738e-01 -6.67371452e-01 -4.24156934e-02 7.48082459e-01 -1.15513250e-01 -7.36496627e-01 1.00040436e+00 -3.81684840e-01 2.03689873e-01 2.67800778e-01 1.23543918e+00 -9.52892378e-02 -1.92184472e+00 -2.22861186e-01 -5.78800499e-01 -8.97145033e-01 -1.55238405e-01 -6.08483493e-01 -8.96444738e-01 9.27550435e-01 3.74214500e-01 -2.00566560e-01 1.15786242e+00 -1.26333177e-01 9.59979117e-01 6.55086935e-02 2.65637487e-01 -9.13619697e-01 5.28382540e-01 2.50032634e-01 1.06525755e+00 -1.08681035e+00 -2.86461413e-01 -3.05154204e-01 -9.75465715e-01 9.07771409e-01 5.16401350e-01 -5.84533811e-02 4.04541761e-01 5.41584730e-01 4.09767270e-01 2.66419858e-01 -1.11904871e+00 1.51241377e-01 9.67815518e-02 8.47623289e-01 3.47786039e-01 -2.41257474e-01 1.10205345e-01 -6.97711334e-02 1.69340476e-01 6.16880834e-01 6.90931737e-01 1.05394816e+00 -2.11570114e-01 -1.20873415e+00 -5.97631454e-01 -6.39710501e-02 -3.75976503e-01 -3.04446995e-01 -2.63998985e-01 7.49367177e-01 -1.05030939e-01 8.10684144e-01 1.46677390e-01 -3.53425026e-01 -2.92432625e-02 1.82421297e-01 8.61743391e-01 -5.54237127e-01 -2.42298439e-01 5.51702440e-01 3.61162610e-02 -8.75963509e-01 -6.58180833e-01 -6.39836013e-01 -1.01122582e+00 -4.91404057e-01 8.21451619e-02 -1.26184925e-01 3.75706494e-01 6.91151142e-01 6.52407825e-01 3.50781173e-01 4.96472448e-01 -1.61038005e+00 -8.56880248e-02 -7.39503443e-01 -3.49113435e-01 7.58139908e-01 5.10168314e-01 -4.39325392e-01 -1.32596999e-01 9.53415096e-01]
[10.871296882629395, -0.6865136027336121]
de8c6bed-3fcd-4665-b5ac-092ad5303eb7
incremental-deep-learning-for-robust-object
1810.10323
null
http://arxiv.org/abs/1810.10323v1
http://arxiv.org/pdf/1810.10323v1.pdf
Incremental Deep Learning for Robust Object Detection in Unknown Cluttered Environments
Object detection in streaming images is a major step in different detection-based applications, such as object tracking, action recognition, robot navigation, and visual surveillance applications. In mostcases, image quality is noisy and biased, and as a result, the data distributions are disturbed and imbalanced. Most object detection approaches, such as the faster region-based convolutional neural network (Faster RCNN), Single Shot Multibox Detector with 300x300 inputs (SSD300), and You Only Look Once version 2 (YOLOv2), rely on simple sampling without considering distortions and noise under real-world changing environments, despite poor object labeling. In this paper, we propose an Incremental active semi-supervised learning (IASSL) technology for unseen object detection. It combines batch-based active learning (AL) and bin-based semi-supervised learning (SSL) to leverage the strong points of AL's exploration and SSL's exploitation capabilities. A collaborative sampling method is also adopted to measure the uncertainty and diversity of AL and the confidence in SSL. Batch-based AL allows us to select more informative, confident, and representative samples with low cost. Bin-based SSL divides streaming image samples into several bins, and each bin repeatedly transfers the discriminative knowledge of convolutional neural network (CNN) deep learning to the next bin until the performance criterion is reached. IASSL can overcome noisy and biased labels in unknown, cluttered data distributions. We obtain superior performance, compared to state-of-the-art technologies such as Faster RCNN, SSD300, and YOLOv2.
['Rhee Phill Kyu', 'Ahmed Minhaz Uddin', 'Shin Dong Kyun']
2018-10-13
null
null
null
null
['robust-object-detection']
['computer-vision']
[ 1.49596378e-01 -8.73757526e-02 -5.17617464e-01 -3.18794996e-01 -6.77177608e-01 -3.60709280e-01 4.17720824e-01 1.33808091e-01 -5.67473352e-01 6.52561307e-01 -3.55513006e-01 5.13038151e-02 -1.44703507e-01 -6.60019636e-01 -7.80651093e-01 -1.15524578e+00 3.89850475e-02 2.97504067e-01 8.76649916e-01 2.62316912e-01 -1.82102658e-02 6.91282034e-01 -1.75264978e+00 2.55979091e-01 7.98843443e-01 1.65206063e+00 2.45704859e-01 5.86228013e-01 -3.71003091e-01 7.62196362e-01 -7.27462828e-01 -3.06121316e-02 2.60767847e-01 -9.05823857e-02 -1.42552078e-01 -4.57933079e-03 -9.66174249e-03 -6.05844378e-01 -1.53988540e-01 1.29913652e+00 4.80929911e-01 7.93744624e-02 4.59879071e-01 -1.57351732e+00 -3.27235609e-01 6.49657667e-01 -6.99752688e-01 4.83772337e-01 -1.52803540e-01 5.60067773e-01 4.02047932e-01 -9.56122458e-01 2.12479636e-01 1.28094995e+00 6.00056171e-01 4.93578792e-01 -8.71093929e-01 -1.06365991e+00 4.19226587e-01 4.95994359e-01 -1.22807813e+00 -5.34680843e-01 7.15975046e-01 -3.13568443e-01 5.01930475e-01 1.09712988e-01 8.96672845e-01 1.15657842e+00 -5.21450974e-02 1.50297391e+00 7.11403370e-01 -3.21120918e-01 7.82664180e-01 2.82520622e-01 -1.51559543e-02 4.84548032e-01 6.69455945e-01 2.18883649e-01 -7.99839258e-01 -2.27188915e-01 5.18736064e-01 4.48141754e-01 -1.19413100e-01 -5.03709555e-01 -9.54345882e-01 7.02684104e-01 4.13141012e-01 -1.63911253e-01 -2.95655608e-01 -4.78002336e-03 4.06872064e-01 3.82063352e-02 4.77242112e-01 2.10347697e-02 -3.52303863e-01 -2.16693021e-02 -1.20036256e+00 -1.17050394e-01 5.57739615e-01 1.06071055e+00 6.15821302e-01 1.63390011e-01 -2.86071241e-01 5.91969609e-01 6.33181453e-01 5.94489396e-01 5.59404194e-01 -6.70240283e-01 1.39949441e-01 8.49650502e-01 1.26088977e-01 -7.55716503e-01 -1.20124601e-01 -5.04920125e-01 -7.58556843e-01 6.67125702e-01 1.57784030e-01 -1.05895452e-01 -9.41889644e-01 1.12559581e+00 5.83319485e-01 6.27864227e-02 -1.41901821e-01 7.92702317e-01 9.67192948e-01 5.59945226e-01 -1.25025734e-01 -4.93078947e-01 6.99052095e-01 -8.44168663e-01 -5.77496052e-01 -3.05308282e-01 2.53730863e-01 -3.01805347e-01 6.05552197e-01 7.18235850e-01 -8.59816968e-01 -7.69771338e-01 -1.29409003e+00 5.61167240e-01 -2.33559012e-01 2.07081273e-01 4.71940041e-01 7.04713047e-01 -6.87568605e-01 4.07487154e-01 -1.04798913e+00 -1.29248751e-02 1.26818764e+00 1.66138381e-01 -9.13681164e-02 -2.63745964e-01 -9.25792754e-01 2.98219979e-01 6.52127504e-01 2.33925954e-01 -1.29755199e+00 -4.90140080e-01 -6.18854463e-01 -9.66033638e-02 7.80101120e-01 1.10255912e-01 1.14917243e+00 -1.19931114e+00 -1.39012384e+00 4.88631308e-01 2.17023909e-01 -7.23467350e-01 7.48710930e-01 -1.91097289e-01 -2.83693194e-01 2.36776322e-01 -8.69529173e-02 7.70035684e-01 1.09943688e+00 -1.11782813e+00 -8.16290438e-01 -4.52870369e-01 -2.81693935e-01 5.71972392e-02 -3.04520309e-01 2.77167466e-02 -3.24138850e-01 -2.89415449e-01 3.76886219e-01 -4.48159039e-01 -1.37006924e-01 6.78038299e-01 -2.89756238e-01 -3.91407013e-01 1.24023116e+00 2.87221838e-02 8.46415162e-01 -2.31653571e+00 -4.37740654e-01 4.80986834e-02 2.70039558e-01 8.07444692e-01 5.37652187e-02 3.88586633e-02 3.48221719e-01 -5.91228716e-02 -8.53995681e-02 -3.70705634e-01 -1.36620715e-01 2.01554909e-01 -3.56265426e-01 5.91962993e-01 4.79051113e-01 9.54820871e-01 -1.15559804e+00 -8.41187656e-01 2.42408633e-01 1.02580562e-01 -1.30446196e-01 2.76144743e-01 -2.80823678e-01 2.78727114e-01 -3.89908969e-01 1.19277537e+00 8.94501209e-01 -1.97435439e-01 -3.31646651e-01 -1.69877321e-01 -1.62346095e-01 -2.84886539e-01 -1.44334936e+00 1.25997877e+00 9.48921293e-02 7.79262364e-01 2.96905413e-02 -1.15006542e+00 1.25903976e+00 9.27918628e-02 4.14697707e-01 -7.15954304e-01 1.13764428e-01 2.02129498e-01 -9.20935497e-02 -6.25246882e-01 3.38115811e-01 4.86725986e-01 3.88935626e-01 2.54878193e-01 2.59966910e-01 2.85873562e-01 9.64171439e-02 2.62340397e-01 1.02508676e+00 2.38171279e-01 4.01703179e-01 9.89933461e-02 5.78196011e-02 -2.98728049e-02 8.68229151e-01 1.22235727e+00 -8.29587698e-01 5.72637141e-01 3.96526754e-01 -2.48353988e-01 -7.81662762e-01 -1.05643046e+00 -2.45929897e-01 9.31039512e-01 4.57991928e-01 3.52235138e-01 -3.09866846e-01 -8.11127722e-01 2.62699518e-02 4.94114935e-01 -5.04119873e-01 -4.87856448e-01 -2.49803662e-01 -8.27248573e-01 5.50582111e-01 6.76964700e-01 7.59030759e-01 -1.12152278e+00 -9.39954281e-01 1.90405950e-01 3.21133316e-01 -8.69962454e-01 1.59000084e-01 4.24982399e-01 -8.06052685e-01 -1.01672864e+00 -6.65321589e-01 -4.18723375e-01 5.61935484e-01 3.68518323e-01 8.50827813e-01 -8.80293995e-02 -5.58274567e-01 1.68892756e-01 -8.05396318e-01 -1.00895047e+00 -1.42252773e-01 -3.33698839e-01 6.16129003e-02 1.53627738e-01 4.59998965e-01 -1.06128998e-01 -8.49384010e-01 5.22656441e-01 -1.01494849e+00 -4.13526922e-01 6.34813905e-01 7.93577194e-01 6.56556904e-01 1.19094886e-01 7.91656792e-01 -5.99753499e-01 1.30921915e-01 -5.68207443e-01 -8.02887440e-01 2.32762098e-01 -5.93298972e-01 -5.00817358e-01 4.12105590e-01 -9.88533974e-01 -9.36756253e-01 1.17683955e-01 1.53421402e-01 -7.61291802e-01 -2.43402511e-01 2.67357435e-02 -2.96620935e-01 -9.00000632e-02 9.25235927e-01 2.46971011e-01 7.97814876e-02 -2.47288659e-01 -3.67432982e-02 1.05682635e+00 3.89630705e-01 -2.79108733e-02 4.72664654e-01 7.54115880e-01 -3.13039362e-01 -4.78074998e-01 -9.62996066e-01 -7.86601186e-01 -4.35906142e-01 -7.63877869e-01 3.98853213e-01 -9.44212615e-01 -5.75506985e-01 8.93792033e-01 -8.78730714e-01 -3.69212657e-01 -6.66901231e-01 6.46161079e-01 -3.34007382e-01 2.12091997e-01 -3.05506494e-02 -1.62147975e+00 -2.91410089e-01 -1.08869803e+00 9.75768209e-01 7.86351740e-01 3.67925137e-01 -4.41024274e-01 -3.01602691e-01 7.36286268e-02 4.51286942e-01 1.62083387e-01 2.04466313e-01 -1.08617544e+00 -8.29465210e-01 -5.53826988e-01 -3.27905416e-01 6.56961322e-01 -8.87579992e-02 1.64760679e-01 -1.07912731e+00 -4.38244671e-01 1.56115396e-02 -7.44643629e-01 1.08351541e+00 4.69360203e-01 1.20966375e+00 -1.36301070e-01 -3.92792702e-01 5.41214347e-01 1.21385872e+00 6.25818133e-01 5.64033568e-01 2.23891601e-01 3.13809812e-01 3.87516946e-01 1.08983850e+00 6.87830806e-01 -2.42374197e-01 1.87581733e-01 8.85416985e-01 -1.03537083e-01 -3.40730473e-02 2.72515346e-03 4.70070094e-01 2.12889329e-01 4.15087253e-01 -3.53131443e-01 -7.33073711e-01 5.67585588e-01 -1.93034172e+00 -9.46133196e-01 1.91957057e-01 2.40636373e+00 7.00434804e-01 6.72541082e-01 1.30513683e-01 4.51633215e-01 8.50321889e-01 3.15915160e-02 -1.25363910e+00 2.23394006e-01 -2.42439225e-01 -2.56521106e-01 6.99293971e-01 -2.87319273e-01 -1.21878028e+00 4.37910378e-01 5.08267546e+00 1.20089412e+00 -1.19960618e+00 1.83493197e-01 8.30789030e-01 -2.48786777e-01 2.52026618e-01 -2.13340729e-01 -1.14960563e+00 7.29457200e-01 3.55010211e-01 2.73454934e-01 -9.79814902e-02 1.38016880e+00 -8.89307936e-04 -6.09276056e-01 -9.33054268e-01 1.13718927e+00 1.53796196e-01 -1.33321297e+00 -3.42676550e-01 -2.90060282e-01 6.87099099e-01 2.84318298e-01 -3.20909210e-02 3.06296855e-01 1.30587757e-01 -6.67208076e-01 9.48764741e-01 6.57480955e-01 6.16206467e-01 -6.36270523e-01 9.28721309e-01 9.02372479e-01 -8.46556127e-01 -5.19850910e-01 -6.50821447e-01 3.68575037e-01 9.76226181e-02 9.34171617e-01 -7.76997089e-01 1.53173134e-01 1.11478591e+00 5.21684468e-01 -4.69039917e-01 1.62341464e+00 -1.62688573e-03 1.00632560e+00 -4.82538760e-01 -4.48566020e-01 1.25034884e-01 2.93356031e-02 8.72972608e-01 1.11231792e+00 8.25179741e-02 -1.33132964e-01 3.19483578e-01 9.25701141e-01 1.37137681e-01 -2.40640223e-01 -2.23385319e-01 -1.06687054e-01 7.87936509e-01 1.18999469e+00 -1.09239602e+00 -5.40520251e-01 -1.25439972e-01 5.91378987e-01 -1.03615001e-02 1.58384383e-01 -6.47168040e-01 -4.73740339e-01 8.68907571e-02 -2.40536258e-02 5.55028617e-01 2.24558879e-02 -1.71630517e-01 -8.44082594e-01 -8.25134292e-02 -6.29277587e-01 2.58665204e-01 -6.73276842e-01 -1.31104112e+00 3.45864892e-01 2.32325733e-01 -1.54063189e+00 5.51751554e-02 -6.46208227e-01 -7.10305989e-01 4.20398116e-01 -1.45628488e+00 -8.23969960e-01 -5.74098289e-01 3.90003175e-01 8.71016204e-01 -5.68920135e-01 2.05636799e-01 2.46444270e-01 -7.69300461e-01 6.36603415e-01 3.62466961e-01 3.14389408e-01 5.76974690e-01 -9.09898698e-01 1.30801290e-01 9.04830694e-01 -2.78689619e-02 7.15671480e-02 4.13558900e-01 -7.02201307e-01 -1.14336312e+00 -1.27720141e+00 5.78089319e-02 1.09328955e-01 4.41582084e-01 -3.83169889e-01 -9.30103302e-01 7.34685510e-02 -3.85769844e-01 6.75983787e-01 4.39857543e-01 -4.95730102e-01 -2.87275523e-01 -5.90496123e-01 -1.34350979e+00 3.01432330e-02 8.65400672e-01 -1.42937005e-01 -1.37838423e-01 3.75350296e-01 8.05114865e-01 -2.18186989e-01 -2.81205922e-01 6.30907059e-01 5.69844604e-01 -1.26012206e+00 8.57670188e-01 -4.26504225e-01 -8.50546062e-02 -3.42752188e-01 -1.09493442e-01 -8.79407585e-01 -1.98466539e-01 -3.82918239e-01 -5.18555820e-01 1.20609176e+00 2.58122087e-01 -6.19450450e-01 1.00321734e+00 1.55619621e-01 -1.56621933e-01 -8.56150448e-01 -1.04907811e+00 -9.19511378e-01 -5.34607291e-01 -4.88199800e-01 4.58006948e-01 4.97536480e-01 -4.35174912e-01 -3.55297476e-01 -1.33296177e-01 9.55950022e-02 9.75489378e-01 -2.21439190e-02 6.07399464e-01 -1.33555770e+00 -2.84596622e-01 -2.76099533e-01 -7.19934165e-01 -1.08466494e+00 -4.28091198e-01 -2.35364631e-01 4.86807138e-01 -1.22281337e+00 1.85461700e-01 -6.73574030e-01 -4.96636301e-01 3.08205396e-01 1.43261338e-02 3.97468209e-01 9.60426331e-02 4.45493609e-01 -1.21450317e+00 5.31541348e-01 9.17267144e-01 -2.51696616e-01 -2.31424198e-01 4.69256461e-01 -2.40496725e-01 8.38005066e-01 7.59684443e-01 -6.66160941e-01 -4.40637261e-01 -2.25394312e-02 2.63509333e-01 -2.07827702e-01 5.21362603e-01 -1.20158207e+00 7.10720837e-01 -3.09707940e-01 8.27868938e-01 -1.08856750e+00 2.67269731e-01 -5.90955257e-01 -1.92876443e-01 5.31225026e-01 -3.46112043e-01 -5.62284946e-01 -9.33309644e-02 9.03603554e-01 -1.68575674e-01 -4.83077049e-01 1.06248653e+00 -2.46391013e-01 -7.89932489e-01 4.97350633e-01 -2.70727187e-01 -7.68918870e-03 1.09362626e+00 -6.44469619e-01 -3.48720074e-01 -1.88896388e-01 -3.99298370e-01 3.39584619e-01 1.69961732e-02 2.71408021e-01 9.23824906e-01 -1.14507473e+00 -5.55691004e-01 3.77749234e-01 3.46707970e-01 6.50115490e-01 2.86051989e-01 1.00666988e+00 -2.53503054e-01 3.95959802e-02 -2.94227945e-03 -1.20396364e+00 -1.09982955e+00 3.98756623e-01 2.03552574e-01 1.52641460e-01 -4.21704173e-01 1.14040136e+00 4.45805825e-02 -1.15130253e-01 8.08309555e-01 -9.97450501e-02 -2.32626006e-01 3.23369324e-01 8.20825994e-01 6.58824444e-01 1.67398244e-01 -1.63145646e-01 -2.49679118e-01 2.20967969e-03 -1.56613454e-01 1.99741393e-01 1.08598268e+00 -1.22845933e-01 3.06167781e-01 7.40802944e-01 9.02191281e-01 -4.68592346e-01 -1.78618586e+00 -6.04066253e-01 -1.50286198e-01 -5.28078139e-01 3.11375111e-01 -6.31112218e-01 -9.62946534e-01 7.58834958e-01 1.09373951e+00 3.68870646e-01 1.05251276e+00 1.51139483e-01 4.70579535e-01 5.07751584e-01 4.21653152e-01 -1.39880025e+00 4.70946461e-01 1.12960711e-01 5.84518015e-01 -1.59583259e+00 1.33798644e-01 2.40410771e-02 -7.00776577e-01 1.01701069e+00 8.69795740e-01 -3.23521234e-02 6.72991693e-01 5.02287686e-01 -2.57273391e-02 1.16699949e-01 -8.59734774e-01 -2.68714398e-01 5.33098765e-02 7.49502838e-01 -5.40925920e-01 -1.45946756e-01 4.04959470e-01 4.48403835e-01 4.65440691e-01 1.67833805e-01 2.21244976e-01 1.09992456e+00 -1.01781249e+00 -3.25776100e-01 -4.66134548e-01 7.66928732e-01 -1.51291177e-01 2.02374443e-01 -1.44163191e-01 5.20504534e-01 5.72559237e-01 9.41294849e-01 4.62038159e-01 -3.32811177e-01 -2.57432926e-02 -3.58796149e-01 1.93739384e-01 -4.90239143e-01 -1.17616296e-01 1.87239513e-01 -3.81717920e-01 -5.52206516e-01 -6.01389587e-01 -7.08940744e-01 -1.14279008e+00 2.69476265e-01 -1.01428783e+00 -1.77747890e-01 8.53363872e-01 9.40854371e-01 1.71465442e-01 2.67695457e-01 8.74925256e-01 -9.75877523e-01 -8.46017182e-01 -9.60173070e-01 -5.46756506e-01 -6.16308525e-02 4.79873002e-01 -8.25011253e-01 -6.69781744e-01 -1.36208192e-01]
[9.240062713623047, 1.4009829759597778]
b34596f6-12b3-4f77-98a1-ce910ef2dbc8
fast-matrix-multiplication-without-tears-a
2306.01097
null
https://arxiv.org/abs/2306.01097v1
https://arxiv.org/pdf/2306.01097v1.pdf
Fast Matrix Multiplication Without Tears: A Constraint Programming Approach
It is known that the multiplication of an $N \times M$ matrix with an $M \times P$ matrix can be performed using fewer multiplications than what the naive $NMP$ approach suggests. The most famous instance of this is Strassen's algorithm for multiplying two $2\times 2$ matrices in 7 instead of 8 multiplications. This gives rise to the constraint satisfaction problem of fast matrix multiplication, where a set of $R < NMP$ multiplication terms must be chosen and combined such that they satisfy correctness constraints on the output matrix. Despite its highly combinatorial nature, this problem has not been exhaustively examined from that perspective, as evidenced for example by the recent deep reinforcement learning approach of AlphaTensor. In this work, we propose a simple yet novel Constraint Programming approach to find non-commutative algorithms for fast matrix multiplication or provide proof of infeasibility otherwise. We propose a set of symmetry-breaking constraints and valid inequalities that are particularly helpful in proving infeasibility. On the feasible side, we find that exploiting solver performance variability in conjunction with a sparsity-based problem decomposition enables finding solutions for larger (feasible) instances of fast matrix multiplication. Our experimental results using CP Optimizer demonstrate that we can find fast matrix multiplication algorithms for matrices up to $3\times 3$ in a short amount of time.
['Elias B. Khalil', 'Pashootan Vaezipoor', 'Chang Liu', 'Arnaud Deza']
2023-06-01
null
null
null
null
['problem-decomposition']
['miscellaneous']
[ 1.54585168e-01 9.65838209e-02 -8.23723339e-03 -1.37539163e-01 -5.18988252e-01 -6.91892385e-01 3.56207564e-02 1.44369513e-01 -5.30605555e-01 9.16640520e-01 -4.79199618e-01 -7.92192698e-01 -6.34705603e-01 -8.85071695e-01 -1.05066741e+00 -6.97885573e-01 -7.25399554e-01 4.72667187e-01 -3.79552901e-01 -5.72875440e-01 4.63519394e-01 3.76560181e-01 -1.52297413e+00 3.11229140e-01 7.63233006e-01 1.15330029e+00 -6.14160113e-03 5.53278685e-01 -8.01754892e-02 4.07116145e-01 -4.28764492e-01 -4.87521678e-01 9.48491633e-01 -3.31497043e-01 -9.33556497e-01 -8.00703466e-03 4.42578018e-01 -1.18131518e-01 1.18469916e-01 1.03654385e+00 -9.76129156e-03 3.59326117e-02 1.56765997e-01 -1.62248397e+00 3.65132513e-03 9.78621244e-01 -1.06391466e+00 1.05596900e-01 4.36267316e-01 2.80730069e-01 1.49915016e+00 -4.58407313e-01 5.30858219e-01 8.52953911e-01 4.52762574e-01 5.90214692e-02 -1.49756169e+00 -1.01031148e+00 2.19061881e-01 3.66678655e-01 -1.75453889e+00 -2.11028010e-01 4.47838366e-01 -1.31071314e-01 1.57886279e+00 5.81442773e-01 7.87294149e-01 3.56868774e-01 2.74996430e-01 3.37991416e-01 1.03959835e+00 -5.08135498e-01 1.90412521e-01 -6.05708957e-02 -1.88393265e-01 7.87226021e-01 6.23414576e-01 1.01042755e-01 -6.33669257e-01 -3.16306464e-02 5.32375216e-01 -4.98124570e-01 3.33711952e-02 -4.05037850e-01 -1.31278241e+00 1.07245100e+00 4.37962890e-01 2.79064387e-01 -6.92749545e-02 6.74713969e-01 3.80099028e-01 5.58234811e-01 -2.14911461e-01 9.88292933e-01 -6.32377028e-01 -1.10304743e-01 -1.06803381e+00 6.64650321e-01 1.04830182e+00 1.00624585e+00 8.03521216e-01 1.36464462e-01 4.15265560e-01 9.13287550e-02 -9.79431067e-03 5.49813271e-01 -2.21726358e-01 -1.13598514e+00 8.88754785e-01 3.73260200e-01 1.45986408e-01 -1.16084921e+00 -6.35758638e-01 -4.21453387e-01 -8.59201908e-01 3.29175204e-01 7.93274939e-01 -2.28842601e-01 -2.51593381e-01 1.88454711e+00 2.12072596e-01 -1.15532227e-01 -9.90928188e-02 8.20571542e-01 -5.48544303e-02 7.46874392e-01 -3.22439849e-01 -5.98604262e-01 1.26720607e+00 -7.95809627e-01 -5.09535372e-01 -2.05166101e-01 9.22732115e-01 -7.96325982e-01 6.18658364e-01 1.00198150e+00 -1.54437351e+00 -1.64814726e-01 -1.27870154e+00 2.69618481e-01 -1.95445269e-01 -1.96360707e-01 1.34639740e+00 8.88921082e-01 -9.27507341e-01 6.77974463e-01 -6.94804370e-01 1.92138225e-01 1.17369860e-01 1.05120289e+00 -4.42683160e-01 -1.38192549e-01 -9.85392094e-01 1.03876925e+00 3.85711461e-01 4.21519130e-01 -3.11418414e-01 -9.41171229e-01 -7.65665770e-01 2.16504097e-01 9.38409865e-01 -5.30024409e-01 9.69659686e-01 -8.05119157e-01 -1.21332884e+00 6.94039166e-01 -3.73121835e-02 -7.23821998e-01 4.51324761e-01 2.63071984e-01 -4.64235470e-02 7.22996369e-02 -1.34640738e-01 4.07932132e-01 7.74615645e-01 -7.46301413e-01 -6.27792954e-01 -3.23248386e-01 6.40088320e-01 1.35728419e-01 1.88440029e-02 -1.01888254e-02 -6.27413541e-02 -2.27474347e-01 3.56044501e-01 -1.09593034e+00 -7.00753868e-01 -2.95848638e-01 -4.70752239e-01 -4.15075533e-02 -2.61871666e-01 -2.99761146e-01 1.37271035e+00 -1.87031949e+00 6.67101741e-01 8.52146149e-01 4.29726213e-01 -1.65363535e-01 -7.06700608e-02 7.92352676e-01 -4.15078491e-01 2.18050107e-01 -2.39707664e-01 1.01813875e-01 4.31302935e-01 3.27890784e-01 -2.08619341e-01 6.53550982e-01 1.51072070e-01 6.89564466e-01 -6.87402904e-01 -3.29965800e-01 9.80827957e-03 -1.11079458e-02 -1.15281773e+00 -2.07411334e-01 -2.42077932e-01 3.65817510e-02 3.30118351e-02 3.63272011e-01 9.75130200e-01 -2.01053277e-01 7.68492401e-01 -3.38754326e-01 -3.02953631e-01 1.62701488e-01 -1.98251283e+00 1.69786024e+00 -5.89635313e-01 2.12720841e-01 5.23371398e-01 -1.48925221e+00 3.78747672e-01 -6.19580373e-02 6.00272119e-01 -9.65099394e-01 2.23643854e-01 4.69380349e-01 6.31832540e-01 -2.36002058e-01 4.63350415e-01 -3.38629961e-01 -2.46375725e-01 6.72552288e-01 -3.14871192e-01 -3.95594746e-01 1.02439606e+00 3.05710018e-01 1.29806411e+00 -5.57692870e-02 1.40611744e-02 -5.07025301e-01 6.39974475e-01 1.29541084e-01 5.69553375e-01 7.38831699e-01 1.87958047e-01 -6.53850008e-03 9.82861042e-01 -3.37421387e-01 -1.09845829e+00 -9.25166070e-01 -8.27916935e-02 9.01431441e-01 -5.82645014e-02 -7.93371618e-01 -7.02631772e-01 -1.50903046e-01 1.04092404e-01 4.24646080e-01 -4.36475754e-01 1.02046162e-01 -9.81858373e-01 -7.89215326e-01 2.70592451e-01 4.74595755e-01 2.61010826e-01 -5.67691743e-01 -8.18868697e-01 2.45579675e-01 1.54109314e-01 -9.71305728e-01 -2.47891724e-01 7.76733935e-01 -7.23964810e-01 -1.24796534e+00 1.13382861e-02 -6.25077546e-01 9.08549607e-01 1.67735927e-02 1.04473770e+00 2.42467955e-01 -5.02372980e-01 3.74531397e-03 -4.65140268e-02 2.37898659e-02 -1.39280424e-01 -3.01181823e-02 3.36415410e-01 -4.06198889e-01 7.48225003e-02 -7.24619508e-01 -5.05293250e-01 1.10185198e-01 -9.30796802e-01 7.83899873e-02 7.38494217e-01 9.26985145e-01 6.10702336e-01 6.27720654e-01 1.43235579e-01 -7.67123342e-01 4.62697506e-01 -2.26502851e-01 -1.11488974e+00 1.12418823e-01 -6.05819404e-01 4.64606434e-01 1.10007489e+00 -2.48629913e-01 -3.34186226e-01 3.26798826e-01 9.35882032e-02 -2.97144771e-01 4.71887082e-01 7.88571835e-01 6.40455857e-02 -3.63301098e-01 6.98639095e-01 5.31787015e-02 1.25388712e-01 8.64581764e-02 5.52997231e-01 -1.65011927e-01 2.55451143e-01 -1.11102676e+00 1.12863314e+00 3.66592586e-01 6.26078665e-01 -4.73784208e-01 -4.11613315e-01 -2.20634490e-01 -2.73310244e-01 1.09299280e-01 3.59409004e-01 -6.54899836e-01 -1.61435318e+00 -1.76163375e-01 -8.46994877e-01 -1.61638454e-01 -2.92142481e-01 2.95938402e-01 -7.77194500e-01 2.78053582e-01 -4.00522053e-01 -8.56616497e-01 -1.33781517e-02 -1.35019767e+00 5.02380013e-01 -1.97772145e-01 -3.93688291e-01 -3.97104919e-01 -2.02988014e-01 4.51759547e-01 4.62910891e-01 1.14246331e-01 9.32411194e-01 -1.70477003e-01 -1.03114450e+00 -1.24957994e-01 -2.72548169e-01 1.13472082e-01 -3.40645075e-01 -3.02329264e-03 -2.37313867e-01 -4.45469290e-01 2.94784326e-02 -2.28265837e-01 5.99063098e-01 2.64131516e-01 1.15978479e+00 -4.85196054e-01 -5.34414239e-02 6.94110096e-01 1.75335789e+00 5.81830852e-02 5.72111487e-01 3.63682032e-01 4.01415467e-01 4.46572512e-01 4.92253661e-01 6.66758180e-01 8.36434141e-02 4.94823456e-01 5.56970477e-01 2.05829769e-01 5.79533935e-01 2.81054199e-01 3.75732295e-02 7.37052202e-01 -2.94681758e-01 1.41234189e-01 -7.17603803e-01 2.87985086e-01 -1.62005472e+00 -9.85911012e-01 -1.46994472e-01 2.32020450e+00 9.70955908e-01 4.23765481e-01 1.14379667e-01 6.56685352e-01 2.30695695e-01 -9.32423770e-02 -3.00973505e-01 -1.18011272e+00 2.79085279e-01 1.14097571e+00 9.49204445e-01 6.11978114e-01 -7.03251064e-01 6.48536146e-01 6.11747456e+00 7.81734109e-01 -8.67403626e-01 -2.11293906e-01 2.95248121e-01 -5.74195325e-01 -3.65004718e-01 1.36436746e-02 -7.29218662e-01 2.14599133e-01 9.36387122e-01 -3.07789475e-01 1.00723040e+00 7.84222007e-01 8.81878063e-02 -5.07590175e-01 -1.48260856e+00 1.14705086e+00 -2.17544317e-01 -1.46646607e+00 -3.32819402e-01 1.76543549e-01 6.53251886e-01 -5.39833188e-01 2.81699985e-01 3.23737741e-01 2.41447479e-01 -1.24795008e+00 6.43716872e-01 -1.16277210e-01 5.72170973e-01 -1.21463251e+00 3.60406965e-01 1.86954677e-01 -1.29226398e+00 -2.28040695e-01 -1.70945093e-01 -6.67245209e-01 8.26909542e-02 4.76493031e-01 -5.09708107e-01 5.61731994e-01 5.62301695e-01 1.30176350e-01 -1.00029610e-01 8.62956226e-01 -1.19621404e-01 9.73563939e-02 -8.12181652e-01 -1.20589763e-01 4.37690884e-01 -4.37552124e-01 1.44569203e-01 9.99788284e-01 2.78429300e-01 3.34255874e-01 6.14738651e-03 1.14384961e+00 6.67882860e-02 3.12051922e-01 -3.39520097e-01 -2.59461761e-01 1.72300860e-01 1.28207457e+00 -9.61053491e-01 -9.94743314e-03 -3.49869668e-01 7.14796662e-01 3.62374783e-01 1.02513105e-01 -1.11101890e+00 -4.22268301e-01 7.19013870e-01 -4.50139642e-02 6.49627209e-01 -7.89181590e-01 -5.38198650e-01 -1.16149294e+00 3.82716954e-01 -1.17866457e+00 3.48761469e-01 -3.01113635e-01 -1.01345885e+00 2.76736200e-01 4.26859036e-02 -9.34993625e-01 -2.42538735e-01 -8.00086439e-01 -2.14926243e-01 9.33162630e-01 -1.36748970e+00 -5.82544923e-01 3.09749812e-01 5.75926721e-01 9.82118025e-02 2.46684223e-01 8.22187901e-01 3.78295183e-01 -7.48552740e-01 8.03303599e-01 -8.87179449e-02 -3.81009310e-01 2.06277296e-01 -1.19897485e+00 3.61839794e-02 9.79661107e-01 2.66910791e-01 9.69136059e-01 1.01728559e+00 -2.17331260e-01 -2.37631083e+00 -2.70273745e-01 8.23052824e-01 -1.06559254e-01 1.04037225e+00 -3.87317985e-01 -2.99239188e-01 5.26688933e-01 2.27117166e-01 -1.66749254e-01 6.73802316e-01 2.92494088e-01 -5.85273445e-01 -4.11359549e-01 -8.98734987e-01 6.76204622e-01 1.06519878e+00 -1.97388709e-01 -1.27401829e-01 6.15010142e-01 5.29620767e-01 -8.54465663e-01 -9.45816100e-01 2.31035009e-01 4.72293139e-01 -9.00747895e-01 1.13968205e+00 -5.00882506e-01 8.43736708e-01 -4.06584561e-01 -4.67652142e-01 -9.04762089e-01 -7.26214498e-02 -8.90416563e-01 -1.64449289e-01 2.58124024e-01 6.67437315e-01 -4.29185659e-01 7.91164815e-01 8.60097766e-01 3.51622701e-02 -1.06910968e+00 -1.03536391e+00 -1.00802290e+00 2.76377052e-01 -7.22850502e-01 4.93713647e-01 9.66366351e-01 5.65692663e-01 1.05360702e-01 -3.60274583e-01 1.94905773e-01 4.83992457e-01 4.22861934e-01 6.90756559e-01 -5.82265258e-01 -9.17268753e-01 -7.38717735e-01 -2.18124256e-01 -6.92455649e-01 -9.91610344e-03 -1.10824668e+00 -2.00162113e-01 -1.09531736e+00 -1.00204162e-02 -8.29248548e-01 -1.34120136e-01 4.91331100e-01 4.01115537e-01 3.22847545e-01 4.64407861e-01 -2.92205989e-01 -6.04995966e-01 -1.02356553e-01 1.12891066e+00 -2.15740830e-01 -8.60118791e-02 -3.00527304e-01 -9.00225520e-01 3.05263519e-01 5.96383154e-01 -2.44071141e-01 -2.57101148e-01 -3.99681687e-01 1.33126056e+00 2.94743866e-01 -9.02945027e-02 -9.83391106e-01 2.41896823e-01 -3.88106614e-01 -6.30068481e-02 -3.90584528e-01 2.37415388e-01 -9.25013781e-01 2.92627811e-01 7.84180582e-01 -1.35577261e-01 4.42772180e-01 5.07298768e-01 9.02546793e-02 1.40229404e-01 -3.58610779e-01 4.29493695e-01 -2.19586372e-01 -4.60161179e-01 1.21665940e-01 -3.03126812e-01 -4.14362289e-02 1.08579385e+00 -1.27148837e-01 -8.91135409e-02 -6.84672296e-02 -7.50107884e-01 4.58302975e-01 -2.00556964e-02 -1.09961085e-01 2.47049674e-01 -9.57575858e-01 -5.30878603e-01 1.92204207e-01 -3.71323913e-01 1.32124638e-02 3.09231907e-01 1.19676316e+00 -9.26757157e-01 5.70858419e-01 -3.76871377e-01 -3.79153669e-01 -1.08240616e+00 9.10714626e-01 1.71575308e-01 -6.49017215e-01 -3.09268445e-01 1.09398186e+00 -3.46866399e-01 -2.60724174e-03 9.65268761e-02 -7.56715775e-01 4.92428273e-01 1.66373793e-02 5.19166946e-01 2.75041968e-01 2.60627657e-01 -1.56334847e-01 -5.12843847e-01 4.35357869e-01 -2.34441385e-01 -9.80892926e-02 1.36647558e+00 2.09590077e-01 -6.05709493e-01 -1.77554280e-01 1.04872108e+00 1.71618730e-01 -6.20464981e-01 1.15614384e-01 -3.17327678e-01 -4.65008050e-01 -3.08035940e-01 -5.93421936e-01 -1.39087403e+00 6.24100327e-01 1.36884257e-01 3.96730930e-01 1.29493535e+00 -3.18718821e-01 7.20847070e-01 9.60774004e-01 6.70810997e-01 -1.38570368e+00 -2.02074394e-01 4.34628248e-01 7.99528480e-01 -1.03080809e+00 4.49840426e-01 -6.24949992e-01 -1.71206146e-01 1.20853257e+00 6.02262139e-01 -3.69422019e-01 4.02252018e-01 6.00462973e-01 -5.68545282e-01 -1.62625685e-01 -6.84652388e-01 -1.04072191e-01 -2.18764499e-01 1.53508484e-01 2.24479198e-01 1.74202949e-01 -9.25971270e-01 5.09190321e-01 -8.16284835e-01 -2.12579727e-01 5.92894137e-01 9.91669357e-01 -2.25559518e-01 -1.51665282e+00 -4.49006826e-01 4.04097885e-01 -3.15051287e-01 -3.13465506e-01 3.14974338e-01 8.81576121e-01 3.03364873e-01 8.40721130e-01 -8.57334305e-03 -3.06386948e-01 1.16391800e-01 -5.66726178e-02 1.13952827e+00 -3.22987795e-01 -8.76484394e-01 -2.04547420e-01 2.71235853e-01 -9.60723639e-01 -2.88726807e-01 -5.19148290e-01 -1.52092218e+00 -8.83242905e-01 -3.35577637e-01 1.65413573e-01 8.52471471e-01 1.09668040e+00 1.10054865e-01 4.46784258e-01 5.50553441e-01 -7.56234109e-01 -7.54700243e-01 -2.86303967e-01 -6.86503947e-01 1.20754719e-01 -1.71662107e-01 -5.44970691e-01 -4.21498835e-01 -2.28496388e-01]
[6.444575786590576, 4.599170684814453]
b367c441-67de-40aa-add0-dac4f6b46881
deep-sequential-segmentation-of-organs-in
1807.02437
null
http://arxiv.org/abs/1807.02437v2
http://arxiv.org/pdf/1807.02437v2.pdf
Deep Sequential Segmentation of Organs in Volumetric Medical Scans
Segmentation in 3D scans is playing an increasingly important role in current clinical practice supporting diagnosis, tissue quantification, or treatment planning. The current 3D approaches based on convolutional neural networks usually suffer from at least three main issues caused predominantly by implementation constraints - first, they require resizing the volume to the lower-resolutional reference dimensions, second, the capacity of such approaches is very limited due to memory restrictions, and third, all slices of volumes have to be available at any given training or testing time. We address these problems by a U-Net-like architecture consisting of bidirectional convolutional LSTM and convolutional, pooling, upsampling and concatenation layers enclosed into time-distributed wrappers. Our network can either process the full volumes in a sequential manner, or segment slabs of slices on demand. We demonstrate performance of our architecture on vertebrae and liver segmentation tasks in 3D CT scans.
['Katja Bühler', 'Maria Wimmer', 'Dimitrios Lenis', 'Alexey Novikov', 'David Major']
2018-07-06
null
null
null
null
['liver-segmentation']
['medical']
[ 1.74821422e-01 1.35351807e-01 -2.09716499e-01 -4.45778400e-01 -5.87040782e-01 -2.91435510e-01 2.59722292e-01 3.55618387e-01 -4.92588311e-01 5.32434106e-01 1.87609524e-01 -7.27738082e-01 8.09102952e-02 -9.21272635e-01 -6.03107274e-01 -4.74123895e-01 -4.47331041e-01 5.68008184e-01 5.15559077e-01 3.15409034e-01 -1.06461123e-01 9.59061861e-01 -6.70387506e-01 4.05529082e-01 4.45384115e-01 1.36206615e+00 4.69808886e-03 9.05543089e-01 -4.67032403e-01 7.46685922e-01 -2.88528234e-01 2.51435116e-02 4.37518209e-01 -4.02464211e-01 -1.13057327e+00 2.44312450e-01 1.87966619e-02 -7.83292294e-01 -4.46053237e-01 8.04669619e-01 7.26124763e-01 1.38412761e-02 4.77217436e-01 -6.94856703e-01 -2.99337804e-01 6.89741790e-01 -5.45365632e-01 3.29986691e-01 -1.30138904e-01 -1.91462506e-03 2.89405048e-01 -5.52152693e-01 4.81440842e-01 8.59056413e-01 8.85326385e-01 6.55206323e-01 -1.16323030e+00 -3.10227692e-01 -1.50962755e-01 -3.30536216e-01 -9.89433646e-01 -1.46082431e-01 3.62445593e-01 -5.19286215e-01 1.04588509e+00 2.93701947e-01 8.54262054e-01 5.66724360e-01 4.80223626e-01 5.67380250e-01 8.46189380e-01 -1.38683081e-01 1.73906177e-01 -1.85817063e-01 -1.62226427e-02 7.68490076e-01 -8.41162130e-02 -1.88111112e-01 1.00212447e-01 -4.56538163e-02 1.32140410e+00 2.74292499e-01 -4.32058245e-01 -4.04187173e-01 -1.24335551e+00 7.67060876e-01 1.00501907e+00 5.77977419e-01 -4.76930916e-01 3.40865761e-01 7.30175316e-01 1.60524055e-01 6.27727389e-01 2.13496879e-01 -4.85689044e-01 2.61539936e-01 -1.18946779e+00 -1.22746795e-01 5.30900598e-01 7.90158927e-01 2.30073765e-01 -1.20664269e-01 -2.13329718e-01 6.19973123e-01 1.10462531e-01 6.39443174e-02 8.76249611e-01 -4.33937371e-01 2.57309526e-01 5.83849609e-01 -2.48361081e-01 -3.81663054e-01 -1.06356740e+00 -3.89619023e-01 -1.35389924e+00 1.45124048e-01 3.36089313e-01 -3.38440061e-01 -1.40808678e+00 1.29133785e+00 5.36269784e-01 1.17247619e-01 -2.81606466e-01 9.84401405e-01 1.23981297e+00 4.49100167e-01 1.33815423e-01 -8.50751773e-02 1.24548686e+00 -9.77529347e-01 -4.43507582e-01 5.10308929e-02 5.69181383e-01 -5.93108177e-01 5.16542971e-01 3.88814020e-03 -1.57146823e+00 -3.87165785e-01 -9.88676786e-01 -2.31856659e-01 -3.29760164e-01 -1.85564905e-01 5.72859585e-01 7.06893981e-01 -1.21513832e+00 9.11285758e-01 -1.29089665e+00 -1.23709142e-01 6.67603254e-01 5.82986414e-01 -3.51452053e-01 1.12765253e-01 -9.67221737e-01 8.41407657e-01 4.03671175e-01 3.72980565e-01 -8.11039150e-01 -8.91661584e-01 -8.87011111e-01 2.02895477e-01 1.75902978e-01 -8.09722841e-01 1.31943893e+00 -4.78345811e-01 -1.66584766e+00 8.58614504e-01 2.08989367e-01 -7.73720503e-01 9.62114036e-01 1.18132435e-01 1.53417056e-02 2.24960759e-01 -2.44740129e-01 6.87395811e-01 4.99717712e-01 -6.49132490e-01 -2.38589481e-01 -4.46571380e-01 -4.34191227e-02 2.02672079e-01 3.69067341e-02 -1.23340204e-01 -6.62657917e-01 -4.19198960e-01 6.74349546e-01 -7.77067363e-01 -6.64169729e-01 1.67956471e-01 -6.28170013e-01 2.14725897e-01 6.28955722e-01 -6.90991402e-01 8.98191810e-01 -1.77336538e+00 -2.30350662e-02 2.34768748e-01 3.70975524e-01 2.43054077e-01 2.42004231e-01 -8.82002413e-02 -2.15721563e-01 3.31727624e-01 -3.52079421e-01 -1.50508091e-01 -3.57371837e-01 1.94878727e-01 9.93666872e-02 5.69567919e-01 3.44874598e-02 1.06501985e+00 -7.70584285e-01 -7.52577484e-01 4.18031365e-01 5.74353933e-01 -5.35598576e-01 2.19705760e-01 -3.47711258e-02 7.91534066e-01 -4.16179299e-01 4.88760829e-01 7.33379185e-01 -5.17787397e-01 1.32501677e-01 -2.31285885e-01 -2.03517407e-01 3.28784674e-01 -9.15465772e-01 1.98240125e+00 -5.47984123e-01 1.98536307e-01 1.40642047e-01 -9.21289861e-01 4.00484324e-01 6.04873419e-01 8.62317801e-01 -5.38260579e-01 3.55219245e-01 4.26243573e-01 8.35610405e-02 -4.10006464e-01 1.61777586e-01 -3.66270095e-01 1.68543443e-01 7.33039618e-01 7.87176043e-02 -5.21465898e-01 -2.41893902e-02 -2.35113904e-01 1.08521569e+00 -6.01746235e-03 3.12950939e-01 -2.22525477e-01 4.99114126e-01 -8.06185007e-02 4.40160751e-01 5.38987160e-01 -4.37646270e-01 8.09571326e-01 7.48198509e-01 -7.75628209e-01 -1.13624883e+00 -9.78100121e-01 -5.02220511e-01 5.47944248e-01 -2.21199825e-01 2.17209473e-01 -6.96586251e-01 -8.08929443e-01 -1.74287468e-01 1.92153484e-01 -8.21905196e-01 2.82071471e-01 -8.49578202e-01 -8.61247301e-01 4.30051357e-01 7.39095509e-01 5.26308477e-01 -1.05501640e+00 -1.28425777e+00 5.73263586e-01 1.98351935e-01 -9.97592330e-01 -4.62539166e-01 6.27882719e-01 -1.54212689e+00 -1.00590849e+00 -1.10600734e+00 -7.69976199e-01 8.53776276e-01 1.38847187e-01 1.21285999e+00 2.70306975e-01 -3.76891106e-01 -1.63168237e-01 6.95542246e-02 -1.85868755e-01 -3.46915185e-01 2.84675628e-01 -3.81587952e-01 -6.30277097e-01 -1.45765111e-01 -5.90746820e-01 -9.16754544e-01 1.78919405e-01 -1.07616544e+00 3.41469228e-01 5.75901151e-01 9.28808630e-01 8.19484353e-01 -2.36913517e-01 2.95699954e-01 -1.04075885e+00 3.64469439e-01 -3.65290821e-01 -4.47883487e-01 1.92937240e-01 -1.18500717e-01 -7.47903436e-02 4.66496974e-01 -1.00795425e-01 -6.87113106e-01 9.57459509e-02 -3.51371497e-01 -6.28577054e-01 -4.01921980e-02 4.77156669e-01 2.00161815e-01 -2.26268664e-01 4.34638411e-01 1.35501623e-01 1.44757405e-01 -2.80704945e-01 2.31887862e-01 4.27389055e-01 3.63291830e-01 -3.49523902e-01 3.01868409e-01 5.05998254e-01 2.95008540e-01 -7.02290654e-01 -5.99968433e-01 -1.85681686e-01 -1.05938292e+00 -1.67759076e-01 1.17754090e+00 -5.18252909e-01 -6.53581679e-01 3.52451652e-01 -1.09063959e+00 -6.07726395e-01 -3.76148075e-01 6.39893532e-01 -4.00980264e-01 2.43447036e-01 -1.12753069e+00 -2.35923856e-01 -8.36419702e-01 -1.65632999e+00 6.97465122e-01 2.93349236e-01 3.59570794e-02 -8.65332603e-01 -2.40711793e-01 -4.14105803e-02 7.54886806e-01 5.86835742e-01 1.31830430e+00 -6.93487883e-01 -5.53580284e-01 -2.67624468e-01 -5.13433516e-01 2.03828201e-01 9.48104188e-02 -2.72096246e-01 -6.66147709e-01 -2.17211947e-01 3.61120440e-02 -3.74471992e-01 5.50412416e-01 9.53167975e-01 1.75443614e+00 5.25601814e-03 -2.32455403e-01 8.95195842e-01 1.26447654e+00 2.09784523e-01 1.41421527e-01 -1.09219790e-01 4.35671031e-01 2.41833076e-01 -2.40621179e-01 3.51473421e-01 2.27476954e-01 5.45408614e-02 5.58522165e-01 -5.29179394e-01 -2.37824216e-01 1.31879210e-01 -5.35638750e-01 8.45422983e-01 -3.17330542e-03 6.41326085e-02 -1.11110842e+00 6.04480505e-01 -1.48159397e+00 -5.59094965e-01 -7.38534182e-02 2.25272584e+00 6.93865955e-01 2.51358777e-01 -9.16986167e-02 -5.17096296e-02 5.29029906e-01 5.31943217e-02 -7.42347777e-01 -5.83867908e-01 5.45587599e-01 5.11106431e-01 7.98328698e-01 2.22209498e-01 -1.18706858e+00 4.28008765e-01 7.01474476e+00 3.20267588e-01 -1.59204924e+00 2.64926374e-01 1.18005204e+00 -3.38252753e-01 -3.75617817e-02 -5.49067855e-01 -2.60016918e-01 2.15847373e-01 7.23718584e-01 1.11069664e-01 4.55366261e-02 6.96125746e-01 3.01390570e-02 -1.76377535e-01 -1.22310674e+00 7.68085957e-01 -2.18758062e-01 -1.46418107e+00 -2.04814702e-01 -4.86129820e-02 6.62533939e-01 4.48298782e-01 -2.32620221e-02 2.75732666e-01 2.52190530e-01 -1.26793003e+00 3.74713063e-01 8.01420882e-02 9.08947647e-01 -5.76246798e-01 7.66808450e-01 3.31459641e-01 -1.00546074e+00 3.21144819e-01 -3.20743412e-01 4.32316601e-01 3.12313020e-01 5.12519658e-01 -1.12280977e+00 5.60141206e-01 6.87845886e-01 9.10734460e-02 -1.18925363e-01 1.22235513e+00 9.05629620e-03 3.32542241e-01 -4.70214665e-01 1.37620255e-01 6.98645771e-01 -9.91482660e-02 1.09400891e-01 1.13578153e+00 4.25619364e-01 2.74203867e-01 2.85523921e-01 7.70623982e-01 -3.82439435e-01 9.68482047e-02 -2.32229367e-01 2.31807172e-01 1.09208962e-02 1.30229104e+00 -1.35411894e+00 -5.23634791e-01 -4.91270691e-01 7.70400047e-01 9.30898264e-02 4.99028675e-02 -8.25302660e-01 -1.92517325e-01 2.34258726e-01 2.40832821e-01 2.20805272e-01 -1.97088659e-01 -4.71109569e-01 -9.83008802e-01 -2.12765828e-01 -3.23390037e-01 4.83821005e-01 -4.56408113e-01 -1.08371615e+00 9.33493495e-01 -2.18425959e-01 -1.10020196e+00 -2.17124090e-01 -5.56108296e-01 -5.69686830e-01 1.03873289e+00 -1.67988062e+00 -1.02709675e+00 -3.27804893e-01 5.41028380e-01 4.33832854e-01 4.03757960e-01 8.98652971e-01 5.13599873e-01 -4.49027985e-01 1.30938932e-01 -1.33447766e-01 4.87873703e-01 3.21407288e-01 -1.33953369e+00 5.24412632e-01 5.31682253e-01 -3.19011211e-01 2.79423058e-01 2.36052334e-01 -4.35252607e-01 -1.11979973e+00 -1.04924071e+00 7.80056059e-01 1.05402432e-01 4.61058885e-01 -1.41384333e-01 -8.84310901e-01 7.89676070e-01 1.44729659e-01 7.93819487e-01 8.68715525e-01 -1.10107802e-01 2.26407185e-01 2.37343565e-01 -1.48975921e+00 3.97478163e-01 7.28132606e-01 -2.49580324e-01 -4.02465403e-01 5.10875762e-01 6.06507480e-01 -1.30565262e+00 -1.27898562e+00 4.01743084e-01 4.55177337e-01 -9.96374369e-01 9.99486566e-01 -6.16316736e-01 6.93993926e-01 1.80969499e-02 2.11189181e-01 -1.14602649e+00 -9.70030427e-02 -2.21496716e-01 -6.32309094e-02 3.64220828e-01 4.24627036e-01 -4.79138076e-01 1.10648263e+00 9.87805068e-01 -3.95637244e-01 -1.17084074e+00 -1.26636028e+00 -1.45526528e-01 4.61528927e-01 -2.71251023e-01 8.70132864e-01 9.08785522e-01 -1.48839787e-01 -1.57175228e-01 5.42355003e-03 -4.33043763e-02 4.79768455e-01 3.30543101e-01 3.40845704e-01 -9.23438549e-01 -2.09277749e-01 -6.65839016e-01 -1.97649032e-01 -1.11788833e+00 -3.79515797e-01 -1.01799035e+00 1.53897526e-02 -1.86674988e+00 1.50879160e-01 -6.91347837e-01 -3.99806857e-01 5.30652463e-01 2.10609302e-01 3.21704447e-01 9.37882140e-02 1.39354005e-01 -1.91070005e-01 2.06836119e-01 1.77932751e+00 -1.12426735e-01 -3.44172508e-01 1.14214420e-01 -2.21589245e-02 8.15862119e-01 6.63544834e-01 -2.11261451e-01 -2.69016117e-01 -7.28200495e-01 -1.35999769e-01 8.73163640e-01 1.96330726e-01 -1.00323343e+00 2.51954794e-01 7.58940578e-02 1.05997288e+00 -9.53842402e-01 1.21642172e-01 -9.15034652e-01 -2.30231155e-02 9.71907198e-01 -3.38517994e-01 6.62610978e-02 3.68601918e-01 8.00641030e-02 -1.64358750e-01 -2.28651732e-01 1.04011941e+00 -7.02476025e-01 -1.43281057e-01 8.63412082e-01 -2.61626929e-01 -2.12395445e-01 1.03374577e+00 -1.95096701e-01 2.64820367e-01 1.08986393e-01 -9.76922035e-01 3.23083609e-01 4.04259972e-02 -1.73012633e-03 4.17478383e-01 -1.17787158e+00 -5.31517208e-01 2.77551591e-01 -4.71370935e-01 7.73930371e-01 7.11100101e-01 1.11546779e+00 -1.16648257e+00 5.20618260e-01 -3.91796261e-01 -7.08952308e-01 -9.46900785e-01 3.93846959e-01 7.76391149e-01 -7.91884243e-01 -1.03319585e+00 1.02731669e+00 1.05216146e-01 -5.04144311e-01 1.33310184e-01 -7.70526826e-01 -5.74866720e-02 -1.32788524e-01 4.09888297e-01 8.82401690e-02 5.22697747e-01 -3.58697802e-01 -3.08827341e-01 3.16830665e-01 -1.13031395e-01 6.57420903e-02 1.38522983e+00 2.61559159e-01 -1.09362692e-01 9.80581790e-02 1.18040013e+00 -5.90963244e-01 -1.18054998e+00 -3.98632079e-01 -1.23445563e-01 -2.07062110e-01 4.49334502e-01 -7.06431568e-01 -1.49604225e+00 1.10085392e+00 6.43292665e-01 2.35697404e-01 1.03534389e+00 -1.98330656e-01 1.14376187e+00 -9.07628145e-03 1.61789671e-01 -7.53690422e-01 -4.05087441e-01 2.97537178e-01 6.28403366e-01 -1.16509032e+00 1.76866770e-01 -1.90171182e-01 -2.00680882e-01 1.44624305e+00 4.28034365e-01 -1.96915746e-01 9.45943654e-01 5.11825919e-01 1.04152851e-01 -1.91549212e-01 -4.70624268e-01 6.60812780e-02 3.63672256e-01 6.15636408e-02 8.95024121e-01 7.40621984e-02 -1.34533182e-01 3.54951173e-01 5.60173541e-02 1.64747775e-01 3.86691630e-01 9.86818433e-01 -2.15545967e-01 -9.25732911e-01 -2.87646472e-01 6.92095280e-01 -8.62886250e-01 3.77808861e-03 2.24956870e-01 8.43201041e-01 7.07087889e-02 2.07951382e-01 2.77782083e-01 1.40443549e-01 2.69547224e-01 -1.83402165e-03 6.46762192e-01 -6.02490842e-01 -1.07881832e+00 2.44189918e-01 -3.53898495e-01 -6.07608676e-01 -1.99608281e-01 -3.92106563e-01 -1.50979459e+00 -2.64868319e-01 -1.01051524e-01 -1.99784175e-01 9.05213058e-01 7.55710721e-01 1.53262511e-01 9.69286501e-01 3.53413492e-01 -1.02878022e+00 -6.37780070e-01 -1.00001979e+00 -4.32892025e-01 5.28154522e-02 4.16878462e-01 -1.87914118e-01 1.88971296e-01 -2.07976013e-01]
[14.490823745727539, -2.5437774658203125]
86e6fad4-49af-4b0c-9eef-a22b7f119fe6
saltinet-scan-path-prediction-on-360-degree
1707.03123
null
http://arxiv.org/abs/1707.03123v5
http://arxiv.org/pdf/1707.03123v5.pdf
SaltiNet: Scan-path Prediction on 360 Degree Images using Saliency Volumes
We introduce SaltiNet, a deep neural network for scanpath prediction trained on 360-degree images. The model is based on a temporal-aware novel representation of saliency information named the saliency volume. The first part of the network consists of a model trained to generate saliency volumes, whose parameters are fit by back-propagation computed from a binary cross entropy (BCE) loss over downsampled versions of the saliency volumes. Sampling strategies over these volumes are used to generate scanpaths over the 360-degree images. Our experiments show the advantages of using saliency volumes, and how they can be used for related tasks. Our source code and trained models available at https://github.com/massens/saliency-360salient-2017.
["Noel E. O'Connor", 'Xavier Giro-i-Nieto', 'Kevin McGuinness', 'Marc Assens']
2017-07-11
null
null
null
null
['scanpath-prediction']
['computer-vision']
[ 2.06148788e-01 4.45254624e-01 -4.42953378e-01 -6.36453450e-01 -6.54022336e-01 -1.16785049e-01 5.55493534e-01 -1.91351473e-01 -5.67934057e-03 3.46119881e-01 5.50334275e-01 -2.68077791e-01 1.27856480e-02 -9.42948580e-01 -1.08682728e+00 -2.00237259e-01 -2.74086952e-01 2.08670363e-01 6.56713486e-01 -4.05192405e-01 5.70251048e-01 4.97866929e-01 -1.61015880e+00 1.79952309e-01 9.82492149e-01 1.00566661e+00 3.65672320e-01 6.48060560e-01 2.81941146e-01 7.66173661e-01 -2.43632928e-01 -2.86174119e-01 6.75454378e-01 -1.69151828e-01 -8.79082918e-01 -3.73471320e-01 7.51820028e-01 -5.79098761e-01 -6.36594713e-01 8.83477330e-01 3.25420588e-01 2.75267839e-01 5.94286442e-01 -1.14729512e+00 -5.28022587e-01 7.16007590e-01 -6.15462661e-01 5.89867115e-01 1.21450424e-01 2.83520043e-01 1.13792241e+00 -1.01122749e+00 8.18351626e-01 1.14716482e+00 7.24908590e-01 2.61169970e-01 -1.09718502e+00 -5.60780525e-01 -1.81228355e-01 2.27912754e-01 -1.25848246e+00 -8.25712010e-02 6.62442982e-01 -3.90101284e-01 6.68714881e-01 3.27141911e-01 1.03044760e+00 7.14296341e-01 6.82649195e-01 1.20235574e+00 7.09174812e-01 -7.23578185e-02 2.85027251e-02 1.79675054e-02 -2.26320446e-01 7.46017635e-01 -2.35637762e-02 6.55939877e-01 -5.38463831e-01 1.80818647e-01 1.18076074e+00 -2.36277595e-01 -1.28834695e-01 -6.84128881e-01 -1.31104565e+00 1.11115575e+00 1.39493728e+00 -7.75017813e-02 -4.49677795e-01 3.33949208e-01 5.15335500e-02 -3.07751060e-01 7.90983617e-01 8.18862319e-01 -1.37714118e-01 1.41864881e-01 -1.53353453e+00 6.48043990e-01 2.71314055e-01 9.79978263e-01 7.92975307e-01 3.25837404e-01 -4.12365407e-01 5.01053393e-01 2.11628601e-01 5.85799158e-01 2.65832394e-01 -1.02863622e+00 2.43957266e-01 2.90282220e-01 1.07291467e-01 -1.06903565e+00 -5.80437481e-01 -5.43757141e-01 -2.12928772e-01 2.91296571e-01 -9.21526644e-03 -1.14184208e-01 -1.31719935e+00 1.47331750e+00 3.45240027e-01 3.17405671e-01 -5.78997433e-01 1.34935868e+00 7.77873516e-01 6.85318410e-01 4.75673154e-02 7.01287031e-01 8.17238748e-01 -1.29781199e+00 -2.71325350e-01 -2.86884665e-01 1.02729365e-01 -6.10318542e-01 1.12293744e+00 5.93596138e-02 -1.33120036e+00 -4.27371144e-01 -1.16575086e+00 -3.35279495e-01 -2.97107846e-01 7.89123476e-02 6.27320051e-01 3.88532802e-02 -1.49662042e+00 7.28675961e-01 -7.25911796e-01 3.04415007e-03 6.65999770e-01 1.36444733e-01 3.49025220e-01 4.43511963e-01 -1.36461926e+00 9.40970480e-01 2.49738425e-01 -1.68947101e-01 -1.24375463e+00 -1.00790262e+00 -1.16301548e+00 1.13405995e-01 1.77069992e-01 -5.42225420e-01 1.68569922e+00 -1.01720035e+00 -1.17415822e+00 7.48652697e-01 -1.35394499e-01 -8.72566521e-01 7.10043669e-01 -4.10935044e-01 -1.94940791e-01 3.04688513e-01 4.52181637e-01 1.36312723e+00 1.11796784e+00 -1.07710719e+00 -7.64682174e-01 9.81325284e-02 -1.68104619e-01 3.84856194e-01 3.29122931e-01 -3.07180911e-01 -3.99813652e-01 -8.29313934e-01 9.99514572e-03 -1.02441585e+00 -5.93464792e-01 4.01252471e-02 -7.92941749e-01 1.85983032e-01 6.80089653e-01 -9.22559083e-01 1.12870646e+00 -1.77384162e+00 -1.07929505e-01 3.05962235e-01 2.18157709e-01 6.78595006e-02 -3.46811295e-01 1.15030810e-01 -2.16413677e-01 2.31970578e-01 -3.52997333e-01 1.15944996e-01 -2.11686313e-01 -5.50206363e-01 -7.76690185e-01 2.32917532e-01 4.18164700e-01 1.07539082e+00 -1.01273715e+00 -2.20721185e-01 5.57048917e-01 4.48288351e-01 -6.76225960e-01 6.69402257e-02 -5.23135662e-01 3.50181639e-01 -4.47051316e-01 4.87379730e-01 7.79474139e-01 -2.71586418e-01 -4.47712690e-01 -2.43827969e-01 -2.65703619e-01 7.05018997e-01 -5.18644929e-01 1.90810108e+00 -3.55597198e-01 1.00519347e+00 -4.99096423e-01 -4.23730642e-01 8.63234758e-01 -2.45788768e-01 4.56619561e-01 -7.78643310e-01 1.48451537e-01 1.01547770e-01 -3.15918803e-01 -1.73421487e-01 1.11658764e+00 3.22189629e-01 -1.44683495e-01 3.42912674e-01 -7.99273551e-02 -6.90100253e-01 1.99754387e-01 4.72135127e-01 8.48932028e-01 4.59506840e-01 6.59754872e-02 -3.00401330e-01 6.06201850e-02 3.41351390e-01 2.23750517e-01 5.31404555e-01 -1.57246560e-01 9.72186625e-01 3.21456254e-01 -7.27608204e-01 -1.56119895e+00 -1.51552534e+00 -1.91144004e-01 7.96835780e-01 6.18550539e-01 -2.38097087e-01 -6.44160748e-01 -7.32049108e-01 1.99080601e-01 1.09797478e+00 -8.64895523e-01 -1.93865150e-01 -6.44672334e-01 -2.52241343e-01 2.49877587e-01 4.80119586e-01 5.37896156e-01 -1.17754865e+00 -1.18803275e+00 -1.05036527e-01 -2.68100977e-01 -7.76446104e-01 -7.02117205e-01 -1.73632484e-02 -9.43564653e-01 -8.78477216e-01 -8.01449478e-01 -4.87774104e-01 4.70661610e-01 4.04513001e-01 1.48139751e+00 2.34514082e-04 -4.28206652e-01 5.07502779e-02 -1.04333900e-01 -6.22254491e-01 -7.95461088e-02 4.07507598e-01 -2.98597604e-01 -3.75311553e-01 6.55005947e-02 -5.79879582e-01 -1.04841411e+00 1.91752464e-01 -7.79648542e-01 6.08084142e-01 7.46290505e-01 5.38587868e-01 8.57877433e-01 -5.77573419e-01 3.20269287e-01 -4.69616979e-01 3.19198459e-01 -7.14253545e-01 -9.51511919e-01 -1.44078940e-01 -4.68688667e-01 2.74513811e-01 1.09337099e-01 -1.50552824e-01 -9.74257648e-01 1.73833206e-01 -7.98727721e-02 -6.18948579e-01 9.71591994e-02 1.94047481e-01 4.67510909e-01 -3.24207474e-03 6.92471743e-01 2.79479802e-01 -2.49366120e-01 -2.32743006e-02 6.60861731e-01 8.60501826e-02 5.53951144e-01 2.10868362e-02 9.49237466e-01 4.34972703e-01 -4.59352881e-02 -5.47185004e-01 -1.07159102e+00 -6.48650080e-02 -4.63361174e-01 -4.00032401e-01 8.13135684e-01 -9.09532011e-01 -1.68783709e-01 1.12121359e-01 -8.02716792e-01 -8.83683145e-01 -4.80804652e-01 3.04615974e-01 -7.50476956e-01 -9.32213292e-02 -4.07283187e-01 -4.72198188e-01 -6.22608840e-01 -1.00577950e+00 1.22400677e+00 7.01531827e-01 -1.78507701e-01 -6.71312928e-01 8.11322108e-02 1.02538906e-01 7.72372961e-01 2.79797733e-01 4.64345157e-01 -3.22402954e-01 -1.17149210e+00 1.92529503e-02 -3.51132840e-01 2.62779109e-02 -2.31022298e-01 1.88721940e-01 -7.59029508e-01 7.17607811e-02 -5.10976911e-01 -3.03851575e-01 1.05810630e+00 1.12606311e+00 1.55245781e+00 -3.70877355e-01 -5.30476749e-01 9.50983226e-01 1.36614156e+00 -1.10592455e-01 8.37339699e-01 5.39915144e-01 6.72211409e-01 3.41032833e-01 9.26561952e-01 4.38296974e-01 6.08639300e-01 6.79295301e-01 8.94663930e-01 -2.84207523e-01 -3.00274104e-01 -6.75095379e-01 6.61054403e-02 1.82016477e-01 2.63380110e-01 -7.06356615e-02 -9.56164598e-01 1.09599221e+00 -1.45954299e+00 -9.57075834e-01 -9.77725610e-02 2.15301085e+00 7.57406950e-01 1.65463358e-01 2.81480521e-01 -3.74551415e-01 6.27514839e-01 7.15111196e-01 -9.34446454e-01 -5.08846402e-01 1.06673896e-01 2.23752648e-01 8.00341427e-01 6.26040518e-01 -1.29781425e+00 1.28432441e+00 6.95289040e+00 8.60366225e-01 -1.50347877e+00 -5.74903935e-02 8.56577694e-01 -4.65488672e-01 -7.49289751e-01 9.51058343e-02 -6.65666521e-01 5.46406031e-01 9.11794960e-01 -2.38688380e-01 3.83754343e-01 1.01810646e+00 3.55558962e-01 -4.77505475e-01 -6.85503781e-01 3.37671310e-01 4.77473699e-02 -1.64760685e+00 1.49475500e-01 -1.17461614e-01 9.43028033e-01 8.34895730e-01 5.51897049e-01 1.63281962e-01 4.87247318e-01 -1.06156421e+00 1.15013909e+00 4.45110559e-01 8.53590429e-01 -8.10982466e-01 2.08156437e-01 1.84039194e-02 -1.08414888e+00 1.34570664e-02 -3.55062902e-01 3.08950067e-01 4.29409474e-01 1.00964618e+00 -1.36700082e+00 9.44208801e-02 9.55498815e-01 8.22299302e-01 -7.35864103e-01 1.31861746e+00 -5.21604061e-01 4.60206956e-01 -4.04126763e-01 -4.80943024e-02 3.34532857e-01 -6.83047697e-02 8.14656556e-01 1.00929976e+00 5.22377729e-01 -3.05745780e-01 -2.76318163e-01 1.46573699e+00 -7.47564062e-02 -7.65307769e-02 -7.52114534e-01 2.85726607e-01 6.20904624e-01 1.29009628e+00 -6.48878217e-01 -3.02820355e-01 1.28286600e-01 6.86660767e-01 1.62354946e-01 3.15362155e-01 -1.24825668e+00 -4.20999020e-01 4.72522974e-01 5.37964880e-01 4.78279799e-01 1.47032574e-01 -7.86734045e-01 -1.08524001e+00 -1.23436572e-02 -3.59697282e-01 5.92581071e-02 -1.25977051e+00 -7.05299795e-01 8.37984145e-01 2.28175968e-01 -1.58746195e+00 -3.60086948e-01 3.42661217e-02 -8.52127492e-01 7.66977131e-01 -1.65098500e+00 -1.24179816e+00 -4.90164578e-01 2.29580745e-01 7.66835988e-01 7.25303218e-02 1.77950695e-01 -1.84176117e-01 -1.17235057e-01 4.40862477e-01 -3.64525348e-01 -9.63147506e-02 4.86922264e-01 -1.35763109e+00 1.20832705e+00 9.84415531e-01 5.01657836e-02 1.80424377e-01 8.96563351e-01 -8.57619524e-01 -6.52551949e-01 -1.43347931e+00 7.73942947e-01 -4.93611038e-01 3.89344841e-01 -1.53201565e-01 -7.44104981e-01 9.36952949e-01 3.49416167e-01 2.89995298e-02 7.99687281e-02 -3.32336515e-01 -1.54822364e-01 3.23606879e-02 -1.17695534e+00 9.54839706e-01 9.17106926e-01 -1.43349335e-01 -4.21030402e-01 1.60017565e-01 9.66860294e-01 -9.27191794e-01 -6.73716187e-01 5.47568142e-01 5.54842055e-01 -1.31199992e+00 1.17793787e+00 4.52198908e-02 1.11536479e+00 -2.38084391e-01 2.77947396e-01 -1.55157793e+00 -2.66228527e-01 -3.27481151e-01 -8.80771428e-02 3.06086957e-01 6.75920069e-01 -3.28907847e-01 9.57528353e-01 2.99363226e-01 -4.89862174e-01 -1.03713024e+00 -7.07844853e-01 -3.80199790e-01 7.92075917e-02 -3.51164192e-01 8.27018857e-01 5.11661112e-01 -1.45430565e-01 1.27848655e-01 -2.94623643e-01 5.51828593e-02 6.74371302e-01 1.38641164e-01 6.25652373e-01 -1.00152111e+00 1.32460058e-01 -5.38387358e-01 -1.90208063e-01 -1.19766784e+00 -1.44912317e-01 -8.63434553e-01 2.32139096e-01 -1.60484338e+00 -6.21564463e-02 -4.84593540e-01 -2.77657360e-02 4.87373561e-01 -1.45670995e-01 4.25518632e-01 3.88116509e-01 1.58159509e-01 -4.51280326e-01 8.05267751e-01 1.42435074e+00 -2.19277129e-01 -4.09078121e-01 1.59721509e-01 -5.78599751e-01 7.20490634e-01 1.20572746e+00 -4.04427797e-01 -5.80263972e-01 -4.46586698e-01 2.01503545e-01 3.67038883e-02 4.05391872e-01 -1.28595531e+00 -7.60926604e-02 -2.15326637e-01 5.28031111e-01 -1.27125406e+00 4.56689328e-01 -2.83421934e-01 4.74191569e-02 5.02566993e-01 -5.96046448e-01 5.30467331e-02 3.52436930e-01 2.27887183e-01 -6.53675348e-02 8.59041587e-02 8.72424960e-01 -1.56145031e-02 -9.44012225e-01 5.16616523e-01 -1.95470303e-01 -9.27153528e-02 9.94621992e-01 -1.07302532e-01 -2.94101059e-01 -7.62528419e-01 -5.22871733e-01 3.59007895e-01 5.99117160e-01 6.86922669e-01 9.97333765e-01 -1.53978360e+00 -7.28040338e-01 2.25398675e-01 -2.48330571e-02 2.41365209e-01 3.08979034e-01 8.16797256e-01 -9.39811230e-01 4.70012009e-01 -3.91826034e-01 -7.16334045e-01 -6.04607999e-01 4.61872935e-01 4.50595319e-01 -3.69508684e-01 -7.34183788e-01 9.73642409e-01 3.64332825e-01 -4.88097459e-01 -1.02961212e-01 -5.31362176e-01 -7.83933327e-02 -3.94092321e-01 4.17741060e-01 2.87176192e-01 -1.86545670e-01 -6.46321654e-01 -2.47866198e-01 4.62211013e-01 8.03932250e-02 -5.54585755e-01 1.40427744e+00 -2.18087301e-01 2.40787685e-01 1.14109842e-02 8.89050186e-01 -4.75309528e-02 -1.74707067e+00 -1.68546110e-01 -1.20312430e-01 -5.79193234e-01 2.62341797e-01 -1.02654111e+00 -1.46648955e+00 7.72399664e-01 7.24860549e-01 -5.16820848e-02 1.05543602e+00 -8.15716758e-02 1.06804287e+00 -2.46481210e-01 2.97293127e-01 -1.05817521e+00 1.08414948e-01 5.91416121e-01 1.23325729e+00 -1.12473226e+00 1.24265298e-01 -3.39337409e-01 -9.58630621e-01 8.21676075e-01 7.73314238e-01 -6.08012199e-01 7.58037090e-01 8.51287246e-02 1.18454844e-01 -3.44726026e-01 -6.17326081e-01 -7.27159381e-02 6.03614986e-01 5.76286137e-01 2.37728149e-01 1.90493137e-01 -1.31428257e-01 2.19614357e-01 -7.10541666e-01 7.35995620e-02 8.32368016e-01 5.87254286e-01 -3.98718983e-01 -4.32415664e-01 -3.19434553e-01 7.07617283e-01 -2.69413620e-01 -3.23426217e-01 -1.74378410e-01 5.69928050e-01 -7.04398453e-02 4.80697185e-01 2.97135174e-01 -7.42341816e-01 8.32203850e-02 -5.04152596e-01 1.26080006e-01 -3.82222325e-01 -3.55143547e-01 -1.29674166e-01 -6.04307242e-02 -9.73305523e-01 -7.55680352e-02 -5.69791436e-01 -1.25433123e+00 -3.10419410e-01 -1.30705938e-01 -1.20136894e-01 8.13009560e-01 3.68837535e-01 5.51532090e-01 5.29858887e-01 8.85748208e-01 -1.56123221e+00 -2.76534140e-01 -9.47232544e-01 -4.37111735e-01 5.45035154e-02 4.37826782e-01 -6.58840716e-01 -3.60649765e-01 -7.78718516e-02]
[9.795345306396484, -0.23709633946418762]
39adc51d-2ae0-44fa-8205-0f508d57772c
diversity-aware-coherence-loss-for-improving
2305.16199
null
https://arxiv.org/abs/2305.16199v2
https://arxiv.org/pdf/2305.16199v2.pdf
Diversity-Aware Coherence Loss for Improving Neural Topic Models
The standard approach for neural topic modeling uses a variational autoencoder (VAE) framework that jointly minimizes the KL divergence between the estimated posterior and prior, in addition to the reconstruction loss. Since neural topic models are trained by recreating individual input documents, they do not explicitly capture the coherence between topic words on the corpus level. In this work, we propose a novel diversity-aware coherence loss that encourages the model to learn corpus-level coherence scores while maintaining a high diversity between topics. Experimental results on multiple datasets show that our method significantly improves the performance of neural topic models without requiring any pretraining or additional parameters.
['Giuseppe Carenini', 'Gabriel Murray', 'Linzi Xing', 'Felipe González-Pizarro', 'Raymond Li']
2023-05-25
null
null
null
null
['topic-models']
['natural-language-processing']
[-2.66994596e-01 4.42870438e-01 -2.38382325e-01 -5.01734853e-01 -8.62835586e-01 -9.13246274e-02 8.44531596e-01 9.08896048e-03 -1.57537133e-01 6.29622877e-01 4.88564372e-01 6.91035911e-02 1.05711661e-01 -7.72935987e-01 -8.47555339e-01 -7.73512900e-01 1.83962315e-01 6.44966543e-01 8.47531110e-02 9.69182253e-02 -7.52819031e-02 -3.68522108e-01 -1.36780226e+00 -4.89204712e-02 1.13219881e+00 7.40941823e-01 5.80820858e-01 2.82647967e-01 -2.88201362e-01 5.89216352e-01 -7.98240662e-01 -3.34312290e-01 -1.25100806e-01 -4.91710097e-01 -3.76741707e-01 2.66168714e-01 5.60644925e-01 -3.63507122e-01 -3.16125989e-01 1.02965844e+00 1.16466716e-01 5.00344634e-01 8.79548728e-01 -1.05025232e+00 -5.82112491e-01 8.80481064e-01 -3.23799908e-01 -9.85801369e-02 -3.80344659e-01 -2.50002086e-01 1.46921837e+00 -9.93701994e-01 7.12132275e-01 1.30747306e+00 5.94702959e-01 3.73271286e-01 -1.64285254e+00 -6.96264088e-01 5.82173109e-01 -1.21339915e-04 -1.50755537e+00 -9.89909321e-02 1.15129685e+00 -4.50442702e-01 9.15060818e-01 -3.35422784e-01 8.29892576e-01 1.39100087e+00 4.58581120e-01 9.12134230e-01 5.60502172e-01 -1.14670597e-01 5.48610747e-01 4.44594771e-01 5.01109362e-01 4.71778691e-01 3.26076686e-01 -2.18209609e-01 -6.98230207e-01 -4.59542841e-01 8.47137272e-01 -1.50081115e-02 -3.06951016e-01 -6.58297122e-01 -7.30655909e-01 1.43524337e+00 1.79821536e-01 1.86942607e-01 -7.11001098e-01 3.64488840e-01 1.61909610e-01 -9.09349620e-02 9.28168535e-01 3.02079976e-01 -3.24869305e-01 1.02164611e-01 -1.43501043e+00 4.91023958e-01 7.62950599e-01 9.44176376e-01 8.86221170e-01 3.10881317e-01 -1.42508805e-01 1.08434844e+00 8.39356661e-01 1.57561079e-01 6.34393930e-01 -1.00564003e+00 1.44231007e-01 1.81579262e-01 -1.58102646e-01 -1.06219292e+00 9.98855457e-02 -7.10021257e-01 -6.86620295e-01 -1.89072505e-01 -1.64212570e-01 -3.33483666e-01 -8.51047099e-01 2.19706225e+00 2.72236168e-01 5.11887193e-01 1.78896844e-01 5.79313815e-01 7.18989611e-01 1.31200290e+00 2.56268322e-01 -2.98907667e-01 1.19122577e+00 -9.73764360e-01 -1.24351084e+00 -3.48908097e-01 1.58258562e-03 -3.90688926e-01 1.03706408e+00 4.60680097e-01 -1.34881330e+00 -2.76977122e-01 -1.14157784e+00 -1.83675796e-01 -4.98166271e-02 6.90774843e-02 6.72133565e-01 4.12518263e-01 -1.14048839e+00 3.67095411e-01 -1.21796846e+00 -1.03931792e-01 3.14807653e-01 1.56143218e-01 1.95183098e-01 1.84146509e-01 -1.32654476e+00 8.19265783e-01 7.96460807e-01 -4.01623249e-01 -1.11630213e+00 -1.03373551e+00 -7.59396374e-01 5.36092997e-01 3.09299350e-01 -8.16798091e-01 1.30219817e+00 -5.48743367e-01 -1.84439087e+00 1.30519167e-01 -3.69619578e-01 -8.05282831e-01 1.95838794e-01 -4.25481141e-01 -5.11891320e-02 5.65780215e-02 -1.52637348e-01 1.19249666e+00 8.87890220e-01 -1.58724511e+00 -4.33489770e-01 -3.00786123e-02 -3.19334418e-01 3.00126016e-01 -6.84589624e-01 -5.20763218e-01 -7.79474318e-01 -8.73598397e-01 2.00884566e-01 -7.36156940e-01 -8.87275860e-02 -1.83421224e-01 -4.06988591e-01 -8.31696510e-01 9.78542507e-01 -6.53619647e-01 1.18907678e+00 -1.82818627e+00 3.97773147e-01 1.54798970e-01 2.59927630e-01 -2.29151413e-01 -9.59844738e-02 3.47843051e-01 4.38076615e-01 1.05807073e-02 -4.23035532e-01 -9.21837568e-01 1.84186995e-01 3.63668710e-01 -9.45032895e-01 2.07569480e-01 1.97765350e-01 5.39309442e-01 -5.63799798e-01 -6.29088879e-01 5.13351103e-03 1.05131888e+00 -1.08476055e+00 2.49916822e-01 -8.25115919e-01 6.25717938e-02 -2.42658734e-01 -2.86750235e-02 4.51418012e-01 -4.59794194e-01 3.64919305e-01 1.11368984e-01 2.99997777e-01 7.72978067e-01 -9.56090689e-01 2.04038692e+00 -5.08411646e-01 1.02209353e+00 -4.41776663e-02 -8.25930297e-01 9.44402039e-01 6.18620396e-01 4.06619877e-01 3.32464054e-02 2.52319463e-02 -2.93427855e-01 -2.52947897e-01 1.78255528e-01 9.72710669e-01 -2.44290978e-01 1.39704168e-01 6.58250093e-01 5.55975497e-01 -1.65353328e-01 -1.30800786e-03 4.56698149e-01 2.80558705e-01 -6.71968907e-02 6.05683774e-02 -5.11097968e-01 -2.62865752e-01 -1.43198416e-01 6.13774717e-01 7.22619414e-01 3.14988405e-01 5.69708943e-01 5.65378070e-01 9.84431356e-02 -1.16958332e+00 -1.25380456e+00 -4.21068937e-01 1.02975178e+00 -6.96045756e-02 -6.11256301e-01 -8.63941371e-01 -2.95633644e-01 -1.81470752e-01 1.30092633e+00 -7.04841137e-01 -1.19885959e-01 -3.01095426e-01 -6.63092494e-01 2.83368528e-01 3.12584043e-01 2.38577589e-01 -5.88787615e-01 -4.22304183e-01 3.71547222e-01 -5.26143134e-01 -7.86844075e-01 -5.18785894e-01 1.21420667e-01 -1.38390994e+00 -4.19696420e-01 -8.42423379e-01 -6.40673339e-01 3.72020662e-01 5.04775345e-02 1.18611443e+00 -4.12690252e-01 2.68914253e-01 9.06594619e-02 1.59120113e-01 -5.53019404e-01 -1.70831650e-01 2.64948308e-01 -4.55549248e-02 -2.63192624e-01 3.68641049e-01 -8.45039487e-01 -2.94213623e-01 -8.44763741e-02 -9.03365493e-01 2.89406776e-02 3.24422687e-01 1.09849668e+00 7.00003505e-01 2.39476278e-01 6.45215333e-01 -7.12134719e-01 1.00335550e+00 -7.66085923e-01 -6.57121301e-01 6.45503253e-02 -8.43695760e-01 2.48838961e-01 8.42700601e-02 -8.10634196e-01 -1.36777580e+00 -5.34835398e-01 7.62918591e-02 -7.94579268e-01 1.04602583e-01 8.86667013e-01 -2.47636950e-03 8.65074337e-01 3.98913622e-01 4.85219598e-01 -1.90385282e-01 -5.42744994e-01 5.69301128e-01 2.58199751e-01 4.38441694e-01 -5.50391614e-01 2.64624834e-01 3.93648058e-01 -5.83590329e-01 -9.94768679e-01 -9.00002718e-01 -2.86612749e-01 -1.43268704e-01 1.06183007e-01 8.15233290e-01 -1.25546288e+00 -1.80484638e-01 1.04732096e-01 -1.43378747e+00 -2.43700176e-01 -3.45747381e-01 7.64654696e-01 -4.53852713e-01 1.42558381e-01 -4.66102481e-01 -7.53100276e-01 -4.63219583e-01 -9.73154366e-01 9.11768973e-01 1.16211906e-01 -3.36277038e-01 -1.35401678e+00 5.25636733e-01 -7.57926628e-02 5.06529212e-01 7.44945463e-03 1.00113118e+00 -6.99821949e-01 -6.71740949e-01 -3.59542817e-02 1.00172922e-01 2.30801195e-01 -1.61154971e-01 6.31518513e-02 -9.21138644e-01 -2.71143287e-01 2.92649299e-01 -3.09529245e-01 1.31859219e+00 9.93450224e-01 8.51653934e-01 -5.90981841e-01 -3.16552103e-01 3.47137183e-01 1.37851632e+00 6.59212619e-02 4.24648762e-01 2.24920977e-02 4.24502552e-01 5.98945975e-01 3.40929031e-02 5.47218740e-01 5.37053645e-01 5.88705540e-01 1.97873518e-01 3.26566458e-01 1.03001319e-01 -5.09612322e-01 3.58186662e-01 1.14843643e+00 3.82895827e-01 -6.78391635e-01 -8.03270876e-01 8.66568446e-01 -1.79260957e+00 -8.47137511e-01 3.12990308e-01 1.80793571e+00 1.22056043e+00 7.01832101e-02 -1.02476791e-01 -3.71898860e-01 5.40973008e-01 4.21541005e-01 -6.86253190e-01 -1.75323933e-01 1.50405373e-02 6.57012910e-02 -1.10025682e-01 7.91100979e-01 -9.92176652e-01 1.25568938e+00 7.18017101e+00 8.06986988e-01 -8.87033224e-01 3.91130596e-01 2.69207597e-01 -5.60030520e-01 -7.94778109e-01 -1.46288261e-01 -9.59620655e-01 4.67439234e-01 1.10858989e+00 -3.44371498e-01 -2.14224402e-02 1.16083074e+00 4.44055125e-02 2.15648091e-03 -9.20252562e-01 4.87275332e-01 1.14177719e-01 -1.38281929e+00 2.96537220e-01 3.11365694e-01 1.12382066e+00 1.81812882e-01 3.82211894e-01 4.16985184e-01 7.49919951e-01 -7.92218268e-01 8.11809659e-01 6.09904110e-01 6.30822778e-02 -9.51231182e-01 4.65937108e-01 4.40662503e-01 -6.19128764e-01 3.56509984e-01 -8.56550753e-01 3.81763995e-01 3.87694657e-01 8.85194123e-01 -9.05131876e-01 -1.95556972e-03 5.51804245e-01 6.37496114e-01 4.78678346e-02 7.45682955e-01 -2.86525458e-01 1.04423034e+00 -5.16837835e-01 -1.40516892e-01 3.20064545e-01 -3.38609248e-01 8.34828913e-01 1.16995180e+00 3.46498102e-01 -8.19602311e-02 1.87025338e-01 1.45577765e+00 -2.50548005e-01 1.02848656e-01 -3.35631043e-01 -1.16034284e-01 6.28883183e-01 7.81969845e-01 -2.81173408e-01 -6.14681065e-01 -8.04373249e-02 8.04349363e-01 4.47216868e-01 6.12039030e-01 -7.33734727e-01 -1.55246451e-01 1.01904356e+00 -2.36508921e-01 6.09970331e-01 -3.84103030e-01 -5.09595454e-01 -1.26496077e+00 -1.82367593e-01 -4.77048486e-01 7.12323114e-02 -5.46158671e-01 -1.21315050e+00 6.59075856e-01 3.55760098e-01 -6.41487837e-01 -7.12586164e-01 2.45879237e-02 -7.67460287e-01 9.32894826e-01 -1.24056351e+00 -9.30298686e-01 1.68908864e-01 3.86695802e-01 8.07348967e-01 -3.12789828e-01 6.67064130e-01 -2.02203542e-01 -6.13650739e-01 4.64880615e-01 5.13264000e-01 -4.11747783e-01 4.10514057e-01 -1.36475194e+00 1.85737729e-01 5.60438633e-01 3.70001346e-01 1.04200435e+00 1.14071095e+00 -7.20125020e-01 -6.93297625e-01 -1.02086389e+00 9.73627985e-01 -1.76856577e-01 4.83079106e-01 -4.67194766e-01 -1.32457387e+00 8.71107042e-01 7.33765781e-01 -7.50834465e-01 1.00943708e+00 6.49522841e-01 -4.14105296e-01 3.58079284e-01 -7.12175548e-01 5.98752022e-01 1.81790411e-01 -4.94106650e-01 -9.21706080e-01 2.99248844e-01 1.20570934e+00 -1.45267516e-01 -1.00950181e+00 1.27653638e-02 4.85829979e-01 -6.00518644e-01 9.16484296e-01 -4.59732175e-01 7.15851247e-01 8.78841430e-02 -2.45449156e-01 -1.45098221e+00 -1.50719807e-01 -2.21664801e-01 -7.97969878e-01 1.27785289e+00 4.27955508e-01 -3.60940516e-01 9.91948366e-01 6.78163946e-01 -1.50910437e-01 -7.88059056e-01 -8.48449469e-01 -6.50370598e-01 4.83696491e-01 -5.13205349e-01 4.89291728e-01 9.86297190e-01 6.35134205e-02 5.00832081e-01 -6.19127214e-01 2.10515246e-01 7.57941544e-01 -1.08149849e-01 4.99962896e-01 -1.47761893e+00 -4.47769344e-01 -5.76340914e-01 5.59320711e-02 -1.36344349e+00 4.40081894e-01 -7.15182722e-01 3.62767160e-01 -1.71010351e+00 3.28416705e-01 -9.52567682e-02 -2.17919976e-01 1.86692551e-01 -2.55453527e-01 -2.92533875e-01 -4.13404182e-02 5.25929272e-01 -5.00242770e-01 1.40735829e+00 8.30664635e-01 -7.66587183e-02 -4.28134680e-01 -1.79965913e-01 -8.29562724e-01 7.84414709e-01 8.48334849e-01 -7.87598968e-01 -8.58931661e-01 -6.22008562e-01 4.71642129e-02 -1.40096471e-01 1.22146338e-01 -8.43973458e-01 3.97011459e-01 -2.98249032e-02 1.56892806e-01 -1.19477069e+00 7.51841605e-01 -4.47370082e-01 3.97681594e-02 1.66578472e-01 -7.80303955e-01 -3.52507710e-01 2.30058700e-01 1.16727877e+00 -3.52511257e-01 -2.79991627e-01 7.32082069e-01 1.01148449e-01 -2.09675863e-01 2.28504524e-01 -7.16614902e-01 8.79658619e-04 6.36947095e-01 1.92380250e-01 7.54160248e-03 -7.45976031e-01 -5.57068646e-01 2.93399692e-01 3.57689738e-01 3.64683926e-01 5.38107038e-01 -1.31894577e+00 -7.41528928e-01 -7.32795224e-02 -1.55171752e-01 3.18725139e-01 3.12641948e-01 1.67014897e-01 8.31031241e-03 8.70148361e-01 1.13319710e-01 -7.25608945e-01 -8.20397973e-01 2.79337674e-01 2.33238935e-01 -5.48476517e-01 -7.83179641e-01 1.06384468e+00 5.44753015e-01 -3.49756300e-01 6.49557471e-01 -3.63431960e-01 -2.13609487e-01 1.43587261e-01 3.77318561e-01 1.18692338e-01 -2.45040834e-01 -3.32742900e-01 1.38368636e-01 7.02321678e-02 -5.36198318e-01 -8.10772181e-01 1.32728875e+00 -1.31911337e-01 1.88753784e-01 7.31454134e-01 1.28978479e+00 -5.37127495e-01 -1.74297881e+00 -7.06332028e-01 -9.62120574e-03 -6.27215579e-02 7.88406849e-01 -4.58023667e-01 -9.27664459e-01 1.05490303e+00 3.31191838e-01 1.72295257e-01 7.28967011e-01 2.53827304e-01 9.01133478e-01 6.10925972e-01 -1.87747359e-01 -1.24443495e+00 4.13590342e-01 5.92792690e-01 8.90012920e-01 -9.06348050e-01 6.52779788e-02 -1.60885304e-01 -8.23837399e-01 7.09375322e-01 4.65527862e-01 -3.79384130e-01 9.70065653e-01 -6.18690811e-02 -2.07703546e-01 -2.98288196e-01 -1.37920260e+00 2.53878504e-01 5.49652576e-01 4.63618815e-01 5.35404682e-01 -1.10753730e-01 -2.85752654e-01 8.14381540e-01 -4.72343504e-01 -3.37667078e-01 3.50603789e-01 5.38645208e-01 -8.43279541e-01 -8.91462147e-01 -1.14427902e-01 3.71394694e-01 -4.41784084e-01 -3.32405061e-01 -1.90966442e-01 5.91030777e-01 -3.01681727e-01 7.88451314e-01 6.46779299e-01 -1.29081696e-01 -3.04754287e-01 4.00871247e-01 -2.95183174e-02 -6.48435116e-01 -1.71450391e-01 6.97279572e-01 -2.58190095e-01 -1.48023307e-01 -2.36601934e-01 -9.25131321e-01 -1.01980364e+00 -1.41906574e-01 -6.69440150e-01 4.37164128e-01 1.01780987e+00 8.15283775e-01 4.01747555e-01 6.99579597e-01 2.88971782e-01 -6.33571148e-01 -5.09641767e-01 -1.35419583e+00 -6.76885724e-01 -8.93396959e-02 1.93300039e-01 -7.42484212e-01 -1.45586759e-01 1.65176973e-01]
[10.424806594848633, 6.980449676513672]
12f772a1-4f55-4ab1-8250-026850c038d7
improving-speech-translation-by-understanding
2107.05782
null
https://arxiv.org/abs/2107.05782v1
https://arxiv.org/pdf/2107.05782v1.pdf
Improving Speech Translation by Understanding and Learning from the Auxiliary Text Translation Task
Pretraining and multitask learning are widely used to improve the speech to text translation performance. In this study, we are interested in training a speech to text translation model along with an auxiliary text to text translation task. We conduct a detailed analysis to understand the impact of the auxiliary task on the primary task within the multitask learning framework. Our analysis confirms that multitask learning tends to generate similar decoder representations from different modalities and preserve more information from the pretrained text translation modules. We observe minimal negative transfer effect between the two tasks and sharing more parameters is helpful to transfer knowledge from the text task to the speech task. The analysis also reveals that the modality representation difference at the top decoder layers is still not negligible, and those layers are critical for the translation quality. Inspired by these findings, we propose three methods to improve translation quality. First, a parameter sharing and initialization strategy is proposed to enhance information sharing between the tasks. Second, a novel attention-based regularization is proposed for the encoders and pulls the representations from different modalities closer. Third, an online knowledge distillation is proposed to enhance the knowledge transfer from the text to the speech task. Our experiments show that the proposed approach improves translation performance by more than 2 BLEU over a strong baseline and achieves state-of-the-art results on the \textsc{MuST-C} English-German, English-French and English-Spanish language pairs.
['Dmitriy Genzel', 'Changhan Wang', 'Xian Li', 'Juan Pino', 'Yun Tang']
2021-07-12
null
https://aclanthology.org/2021.acl-long.328
https://aclanthology.org/2021.acl-long.328.pdf
acl-2021-5
['speech-to-text-translation']
['natural-language-processing']
[ 2.35947192e-01 9.03306678e-02 -3.14385772e-01 -3.20509821e-01 -1.35760546e+00 -4.67255741e-01 8.15056920e-01 -3.19574505e-01 -5.14572263e-01 8.35479200e-01 6.84280872e-01 -4.06169057e-01 3.38585794e-01 -3.12286973e-01 -1.00552022e+00 -8.04839969e-01 7.67429888e-01 4.78600472e-01 -7.19198808e-02 -4.43364114e-01 -1.15804203e-01 -1.85040027e-01 -8.25614035e-01 8.22468817e-01 1.22787726e+00 6.09392107e-01 6.64609492e-01 2.48393044e-01 -2.77616620e-01 5.37371457e-01 -5.46555579e-01 -7.90624082e-01 1.05202965e-01 -7.37921715e-01 -8.29852402e-01 -8.42111334e-02 1.98791981e-01 -2.53370255e-01 -2.90842235e-01 9.06061172e-01 8.30297410e-01 1.98639452e-01 7.17030942e-01 -8.87892425e-01 -8.69102776e-01 1.01379311e+00 -5.66399455e-01 5.48235625e-02 6.58415258e-02 4.03054468e-02 9.27891552e-01 -1.21635294e+00 3.63986224e-01 1.28903985e+00 4.01366085e-01 5.99793553e-01 -1.05240989e+00 -6.79346085e-01 9.13129747e-02 1.16342708e-01 -1.23393250e+00 -9.07205105e-01 4.63108301e-01 -1.77438617e-01 1.10401690e+00 1.84381660e-02 5.09227775e-02 1.56970930e+00 2.88388431e-01 8.81535530e-01 1.09842551e+00 -5.80858946e-01 -3.11364561e-01 5.36498249e-01 -3.85873139e-01 4.55094069e-01 -9.44769979e-02 5.21719223e-03 -1.02761149e+00 2.36539409e-01 5.30206978e-01 -1.19152136e-01 -3.85639042e-01 1.28843233e-01 -1.63185298e+00 6.53726935e-01 1.94565326e-01 6.21432304e-01 -4.01776552e-01 1.35479391e-01 6.04192853e-01 6.85007513e-01 6.89245582e-01 2.08838135e-01 -6.54938817e-01 -2.81653911e-01 -8.56601179e-01 -3.80384237e-01 5.52467167e-01 1.12153423e+00 8.49994421e-01 1.20079257e-01 -5.93370259e-01 1.16823947e+00 5.41464746e-01 9.65846062e-01 6.54521286e-01 -4.14929122e-01 1.35037816e+00 2.88697213e-01 -1.94616124e-01 -5.29896140e-01 5.90019040e-02 -5.57208419e-01 -9.26716447e-01 -4.38185275e-01 1.77528068e-01 -4.92939174e-01 -6.71154499e-01 1.84688878e+00 -1.29365429e-01 -2.80483872e-01 4.32055265e-01 9.21929359e-01 8.19527566e-01 9.03846443e-01 1.72396511e-01 -2.42828980e-01 1.30701792e+00 -1.39717591e+00 -1.03741693e+00 -5.05340040e-01 8.68643880e-01 -1.20531416e+00 1.26705921e+00 -3.15563679e-01 -1.22869086e+00 -6.99307859e-01 -7.75403798e-01 -1.99654967e-01 -2.35206231e-01 6.59455955e-01 2.28236973e-01 4.07084554e-01 -9.92856562e-01 3.30609590e-01 -7.71360159e-01 -4.62993830e-01 2.93042421e-01 2.86358356e-01 -2.63665646e-01 -1.21110655e-01 -1.55832160e+00 1.31164861e+00 3.62879992e-01 3.97624522e-02 -8.99927199e-01 -3.78974050e-01 -8.37562084e-01 1.72034860e-01 1.62029088e-01 -8.90631497e-01 1.07515287e+00 -1.52168858e+00 -1.89884198e+00 6.90895796e-01 -5.69393277e-01 -1.31897271e-01 3.92827719e-01 -2.20230326e-01 -3.22516888e-01 -2.91915853e-02 1.60871044e-01 6.59980357e-01 9.60402429e-01 -1.18765390e+00 -5.43275535e-01 -1.96693182e-01 -9.77078378e-02 7.85280883e-01 -6.57080114e-01 1.84754416e-01 -6.91548645e-01 -9.50226128e-01 -6.12710677e-02 -1.03019047e+00 4.17915016e-01 -3.94835681e-01 -4.00710702e-01 -1.42064959e-01 4.54852581e-01 -9.43770885e-01 1.14483881e+00 -2.10026455e+00 5.83427072e-01 -1.65673763e-01 -2.69076258e-01 2.24631891e-01 -4.44481403e-01 6.20983243e-01 2.10867319e-02 -4.67215553e-02 -1.50515959e-01 -8.16063643e-01 -1.92712918e-02 1.15019239e-01 -2.93641508e-01 2.19095424e-01 1.28435194e-01 1.29252923e+00 -5.34514964e-01 -3.02242458e-01 -7.60333762e-02 5.56328535e-01 -2.08532587e-01 2.57875711e-01 -6.46268055e-02 7.41285682e-01 -2.91187286e-01 4.23469871e-01 5.33924222e-01 -3.28289777e-01 1.44833922e-01 -2.70223945e-01 2.31396705e-02 7.41548896e-01 -5.23024380e-01 2.15298963e+00 -8.72583568e-01 5.98276913e-01 9.42143202e-02 -9.59162831e-01 6.88002646e-01 8.11798751e-01 8.43176097e-02 -1.01528239e+00 3.17724198e-01 4.41595644e-01 2.11373985e-01 -3.20755273e-01 4.15217161e-01 -4.41697806e-01 1.65127043e-03 7.15412080e-01 3.24422151e-01 4.51401398e-02 -6.71928376e-02 7.02981800e-02 5.02337217e-01 2.45342568e-01 -8.24809521e-02 -3.23387325e-01 5.19385576e-01 -1.86090723e-01 4.35988009e-01 3.06673914e-01 7.93909580e-02 3.60415161e-01 1.63805708e-01 1.53915837e-01 -9.11250234e-01 -8.33310306e-01 3.34223628e-01 1.64209437e+00 6.24114983e-02 -2.25619465e-01 -7.85635889e-01 -8.54680479e-01 -2.77799398e-01 8.54461908e-01 -3.97753358e-01 -5.29566467e-01 -6.00320876e-01 -7.81515896e-01 6.46353662e-01 4.95850950e-01 7.77587891e-01 -7.67685175e-01 1.63902894e-01 7.79241472e-02 -9.07545745e-01 -1.23198998e+00 -1.01957083e+00 1.61138043e-01 -9.55526412e-01 -4.23486322e-01 -1.17992353e+00 -1.01134408e+00 7.04281926e-01 3.95365447e-01 9.46025372e-01 -2.01919615e-01 7.55865097e-01 1.46584928e-01 -4.43114072e-01 -3.06692630e-01 -7.74536073e-01 5.92344701e-01 3.66391689e-02 1.89970762e-01 3.00866306e-01 -2.83542663e-01 -2.90232986e-01 5.66240013e-01 -6.64887547e-01 5.90636134e-01 1.00082839e+00 9.44554865e-01 3.01451981e-01 -4.94745433e-01 6.80001795e-01 -4.79348421e-01 7.07144797e-01 -4.47674423e-01 -7.04078451e-02 5.19466400e-01 -5.11418223e-01 3.60470206e-01 7.37461984e-01 -5.34613550e-01 -1.47646868e+00 -9.12992582e-02 -9.86205637e-02 -2.30468512e-01 1.31951928e-01 5.33959150e-01 -3.36443961e-01 1.44508600e-01 4.35980916e-01 4.75341380e-01 -6.97504804e-02 -5.00198424e-01 4.27064985e-01 9.74214852e-01 -6.89887479e-02 -6.17894709e-01 7.09866405e-01 1.54706821e-01 -4.79499847e-01 -5.11595845e-01 -7.48703361e-01 -9.32151079e-02 -6.15610480e-01 -1.96112989e-04 9.52463210e-01 -1.27876246e+00 -2.60458082e-01 5.01764894e-01 -1.39479923e+00 -6.04007483e-01 -8.65944661e-03 8.35643888e-01 -5.66303611e-01 1.90170199e-01 -6.73479319e-01 -3.30001950e-01 -4.81050193e-01 -1.44477272e+00 1.24311328e+00 -5.74995428e-02 1.37546599e-01 -1.22898257e+00 -3.76948304e-02 7.00714409e-01 6.53166115e-01 -6.63552225e-01 1.01541555e+00 -7.58670270e-01 -4.20068979e-01 4.11238134e-01 -4.06004965e-01 4.36024696e-01 3.92171770e-01 -6.25009835e-01 -9.62565660e-01 -5.07884979e-01 6.34512678e-02 -4.27655339e-01 1.08427405e+00 2.71506280e-01 5.34182131e-01 -3.14434260e-01 -2.79270411e-01 5.73172510e-01 1.04306448e+00 -3.27469618e-03 6.50041938e-01 1.56251252e-01 8.44875693e-01 5.83818376e-01 3.92725438e-01 -1.41431943e-01 8.11982453e-01 8.39199901e-01 -4.04249728e-02 -3.68364364e-01 -5.31678319e-01 -3.31344366e-01 9.95964646e-01 1.48781121e+00 2.22233720e-02 -3.58162522e-01 -6.87292993e-01 5.12464523e-01 -1.85236108e+00 -5.98290741e-01 1.08035848e-01 2.22901344e+00 1.21473765e+00 -1.41087070e-01 -1.59584224e-01 -4.85807359e-01 9.09961343e-01 5.25827296e-02 -2.80350566e-01 -3.90979797e-01 -2.79973179e-01 -8.99071544e-02 3.89401972e-01 6.73827827e-01 -7.67939508e-01 1.45477724e+00 5.95799398e+00 1.03404617e+00 -1.39905226e+00 6.57463551e-01 5.46993971e-01 -1.96075618e-01 -4.58767563e-01 -6.84246346e-02 -7.96362400e-01 5.09809971e-01 1.22856581e+00 -2.10381553e-01 6.04722679e-01 2.45203286e-01 3.83480072e-01 6.06170148e-02 -1.16468990e+00 9.09650505e-01 2.46962115e-01 -1.07439184e+00 3.72644097e-01 3.26839201e-02 9.03681874e-01 3.32994848e-01 1.08171344e-01 5.00963092e-01 1.02617338e-01 -8.55306089e-01 7.56494343e-01 4.68851864e-01 9.75443661e-01 -5.78999937e-01 7.41578639e-01 4.07684654e-01 -1.00280547e+00 1.15622409e-01 -2.67658502e-01 2.26963565e-01 3.20735008e-01 3.64485502e-01 -7.73997426e-01 7.68580079e-01 2.62621045e-01 6.05008662e-01 -3.69000584e-01 3.33488494e-01 -4.47746992e-01 6.46063745e-01 8.23646188e-02 1.04165591e-01 2.81217396e-01 -1.68078303e-01 4.47720140e-01 1.38881910e+00 5.01221418e-01 -4.42566872e-01 -1.64255910e-02 7.87250400e-01 -5.32819390e-01 3.80421698e-01 -6.26213908e-01 -1.79877639e-01 2.61156678e-01 9.46520865e-01 -2.50986993e-01 -5.27527332e-01 -7.27639914e-01 1.54112685e+00 3.59543324e-01 7.18838036e-01 -9.64795053e-01 -3.04660231e-01 5.54115653e-01 -1.74295232e-01 2.10819185e-01 -3.29850197e-01 -3.73343766e-01 -1.44398415e+00 1.49440065e-01 -9.98738170e-01 -3.86738069e-02 -5.98941445e-01 -1.07264650e+00 6.48008347e-01 -2.67064810e-01 -1.13529301e+00 -2.45498106e-01 -2.83945382e-01 -4.11209285e-01 1.39111495e+00 -1.91141570e+00 -1.42583477e+00 2.34279588e-01 7.97937989e-01 8.38649511e-01 -5.02828002e-01 7.86662996e-01 5.77154517e-01 -4.93452430e-01 9.64018285e-01 3.13277692e-01 2.34269783e-01 1.19537401e+00 -7.72597671e-01 3.33199382e-01 9.26629782e-01 7.96916559e-02 7.75103629e-01 2.45257303e-01 -6.38458014e-01 -1.58978117e+00 -1.09139991e+00 1.31622159e+00 -4.01093632e-01 5.21318018e-01 -4.84578997e-01 -8.25881720e-01 7.11477220e-01 8.25140119e-01 -4.72061634e-01 7.64291942e-01 6.14247620e-02 -3.10057849e-01 -2.30180603e-02 -7.19440520e-01 5.99429667e-01 7.75638998e-01 -9.65436339e-01 -6.61451399e-01 3.48792911e-01 9.04482126e-01 -2.84161627e-01 -7.65912235e-01 1.74204439e-01 3.60405803e-01 -4.07979161e-01 6.40112340e-01 -5.14478862e-01 5.43086469e-01 -6.19113147e-02 -3.70547742e-01 -1.80090773e+00 -1.73809290e-01 -7.02518821e-01 2.93918364e-02 1.27103186e+00 9.47649717e-01 -6.45680845e-01 2.45235011e-01 1.47692591e-01 -3.66027623e-01 -4.15645272e-01 -1.12234747e+00 -7.40257204e-01 5.31697035e-01 -8.64036232e-02 3.41359496e-01 1.02722406e+00 9.24815014e-02 9.49241519e-01 -6.09865606e-01 7.57106841e-02 2.57427216e-01 -5.37754633e-02 5.67185879e-01 -8.07019651e-01 -2.98270106e-01 -3.60365331e-01 5.26792109e-01 -1.49236023e+00 3.99265885e-01 -1.35908830e+00 1.34972602e-01 -1.74575388e+00 4.09254134e-01 -1.07828207e-01 -1.80842459e-01 6.55805767e-01 -4.00308788e-01 7.29375407e-02 3.33138704e-01 4.16244060e-01 -4.96574432e-01 1.02321947e+00 1.61215258e+00 -1.91219583e-01 -1.48287460e-01 -6.06349111e-02 -9.64482307e-01 3.27040464e-01 7.22922504e-01 -5.33244908e-01 -3.85524124e-01 -1.27433729e+00 2.25142583e-01 7.62138516e-02 -1.36367381e-01 -4.60226238e-01 1.93704158e-01 3.59859392e-02 1.01280272e-01 -1.44760638e-01 3.98281038e-01 -7.60148764e-01 -3.19285959e-01 2.63095826e-01 -4.57738727e-01 4.57375571e-02 3.52408975e-01 2.14996070e-01 -2.86952227e-01 -1.71552710e-02 8.60150397e-01 -5.03159873e-02 -1.57725230e-01 1.06603920e-01 -4.75781798e-01 1.31853431e-01 4.99830931e-01 2.34678105e-01 -4.12118554e-01 -5.34935117e-01 -5.91870189e-01 2.37931624e-01 1.53030336e-01 7.35469460e-01 3.74165922e-01 -1.72259808e+00 -1.16287625e+00 2.01328427e-01 1.47571370e-01 -5.55426180e-01 1.04187667e-01 1.34722185e+00 7.94843063e-02 6.79746091e-01 -5.09640239e-02 -5.00170469e-01 -1.18142605e+00 7.40114450e-02 3.43885720e-01 -3.27997416e-01 -1.96286544e-01 7.38562286e-01 3.29286605e-01 -6.68646336e-01 2.34787866e-01 -2.66001463e-01 1.10568084e-01 1.91581637e-01 3.82025063e-01 3.13663870e-01 2.01199278e-01 -8.95727694e-01 -2.99875855e-01 5.00953376e-01 -2.76692629e-01 -4.49280322e-01 1.05150163e+00 -6.41401768e-01 -9.88181904e-02 5.10688066e-01 1.29097462e+00 -5.02063259e-02 -1.06352949e+00 -6.04562819e-01 -2.63329983e-01 -2.20532209e-01 8.13992694e-02 -1.22751582e+00 -1.07755566e+00 1.26208627e+00 4.50249553e-01 -3.10653627e-01 1.09418571e+00 1.06483050e-01 9.23564851e-01 4.49112952e-01 2.09127426e-01 -1.35603476e+00 2.07619444e-01 1.05107474e+00 1.02940464e+00 -1.48142385e+00 -4.95800376e-01 -1.97499081e-01 -1.10811281e+00 9.45785701e-01 4.61394727e-01 4.99419093e-01 2.06379116e-01 1.72952235e-01 2.92324185e-01 1.90874904e-01 -7.74510264e-01 -2.73962259e-01 4.94353086e-01 3.67127031e-01 9.01983678e-01 -9.63229388e-02 -3.20911676e-01 5.52055955e-01 1.07097887e-01 -1.97834983e-01 5.43272719e-02 5.20672977e-01 -3.50867569e-01 -1.37792945e+00 -3.02448541e-01 5.92088215e-02 -5.14995396e-01 -5.96418023e-01 -5.03205836e-01 3.61012578e-01 -9.94381011e-02 1.15983891e+00 -2.08642617e-01 -3.65781605e-01 3.23311985e-01 4.91832882e-01 4.44880694e-01 -6.90442383e-01 -8.55062783e-01 4.15553331e-01 1.57047659e-01 -3.97991806e-01 -4.07666296e-01 -4.25303429e-01 -1.20223796e+00 -2.30736330e-01 -5.39219260e-01 2.21871093e-01 9.33521032e-01 1.20315206e+00 6.76500201e-01 7.79481769e-01 7.31764317e-01 -6.17217481e-01 -4.99758840e-01 -1.40267873e+00 -1.47693276e-01 1.93676800e-01 3.18113595e-01 -4.89052385e-01 -2.82555729e-01 5.90145178e-02]
[14.469467163085938, 7.225993633270264]
247a2e33-01ab-4a8f-bee5-cc73dbe4f4bd
ransomai-ai-powered-ransomware-for-stealthy
2306.15559
null
https://arxiv.org/abs/2306.15559v1
https://arxiv.org/pdf/2306.15559v1.pdf
RansomAI: AI-powered Ransomware for Stealthy Encryption
Cybersecurity solutions have shown promising performance when detecting ransomware samples that use fixed algorithms and encryption rates. However, due to the current explosion of Artificial Intelligence (AI), sooner than later, ransomware (and malware in general) will incorporate AI techniques to intelligently and dynamically adapt its encryption behavior to be undetected. It might result in ineffective and obsolete cybersecurity solutions, but the literature lacks AI-powered ransomware to verify it. Thus, this work proposes RansomAI, a Reinforcement Learning-based framework that can be integrated into existing ransomware samples to adapt their encryption behavior and stay stealthy while encrypting files. RansomAI presents an agent that learns the best encryption algorithm, rate, and duration that minimizes its detection (using a reward mechanism and a fingerprinting intelligent detection system) while maximizing its damage function. The proposed framework was validated in a ransomware, Ransomware-PoC, that infected a Raspberry Pi 4, acting as a crowdsensor. A pool of experiments with Deep Q-Learning and Isolation Forest (deployed on the agent and detection system, respectively) has demonstrated that RansomAI evades the detection of Ransomware-PoC affecting the Raspberry Pi 4 in a few minutes with >90% accuracy.
['Burkhard Stiller', 'Gregorio Martínez Pérez', 'Gérôme Bovet', 'Pedro Miguel Sánchez Sánchez', 'Janik Luechinger', 'Alberto Huertas Celdrán', 'Jan von der Assen']
2023-06-27
null
null
null
null
['q-learning']
['methodology']
[-6.43869415e-02 -2.85428911e-01 1.69282164e-02 3.13602000e-01 4.95118767e-01 -1.04431629e+00 5.53611815e-01 -1.00738287e-01 -4.60233837e-01 6.86408162e-01 -4.84950721e-01 -4.64474857e-01 -3.17540079e-01 -7.73684204e-01 -3.53233665e-01 -6.98703706e-01 -1.15193069e-01 7.09754467e-01 1.75714239e-01 -1.76857665e-01 5.23778915e-01 7.57734358e-01 -1.43761432e+00 -3.71569470e-02 6.77279472e-01 1.12363601e+00 -2.50209868e-01 1.01127219e+00 6.63533270e-01 8.94725263e-01 -1.10249794e+00 -2.70449638e-01 8.53020489e-01 -3.03151101e-01 -2.78157651e-01 -3.77197057e-01 -4.31277901e-02 -7.06013024e-01 -3.40665460e-01 8.00071120e-01 4.10168439e-01 -1.28063992e-01 5.79153836e-01 -1.88143575e+00 -8.65830958e-01 4.82686579e-01 -2.69284427e-01 2.63878316e-01 3.56007993e-01 7.79461503e-01 2.48160467e-01 1.00799352e-01 3.63335937e-01 1.03180933e+00 6.39729977e-01 7.68940151e-01 -6.75476074e-01 -1.06523585e+00 -3.51932466e-01 2.16008544e-01 -1.01848137e+00 1.42516658e-01 3.40508938e-01 -4.45629835e-01 1.36651587e+00 2.94752061e-01 9.30988312e-01 1.39384520e+00 1.12269914e+00 3.81875098e-01 1.35035121e+00 4.77781799e-03 6.41722381e-01 3.23418677e-01 -7.92778656e-02 6.55370653e-01 7.43219554e-01 9.94150639e-01 -1.47152543e-01 -5.67387342e-01 3.82369310e-01 4.67409074e-01 1.61985651e-01 -2.09234338e-02 -8.98460388e-01 7.48011470e-01 2.52805412e-01 2.35049278e-01 -6.13146126e-01 -9.35661257e-04 4.96435016e-01 4.85723913e-01 -2.09872678e-01 9.02892590e-01 -5.52920997e-01 -2.46982574e-01 -3.85281324e-01 1.34835005e-01 1.07831407e+00 5.48851013e-01 2.74095565e-01 3.62197459e-01 -1.73096478e-01 -1.21007793e-01 2.11265072e-01 1.09157097e+00 7.75449455e-01 -5.95701098e-01 -2.86272377e-01 9.04011428e-01 4.25024331e-01 -1.11797571e+00 -5.26630282e-01 5.29326610e-02 -4.49569106e-01 4.29934710e-01 8.85496289e-02 -7.42704034e-01 -7.57538438e-01 8.82872880e-01 5.78981161e-01 5.38035631e-01 4.01587114e-02 1.01965427e+00 -7.39444494e-02 7.77336776e-01 -7.43572265e-02 -1.77608028e-01 1.52911508e+00 -5.16014934e-01 -4.95742679e-01 1.51819691e-01 2.13135362e-01 -5.90018213e-01 6.31761074e-01 7.73696840e-01 -2.16619655e-01 -2.61748433e-01 -1.19089568e+00 9.92752731e-01 -8.20694566e-01 -1.50085822e-01 3.52572441e-01 1.32345152e+00 -6.74031198e-01 4.76966053e-01 -6.36485875e-01 -5.00040531e-01 1.95227891e-01 5.81654966e-01 1.34304628e-01 3.85039032e-01 -1.05900311e+00 1.07513225e+00 5.40640652e-01 -3.09984028e-01 -1.57577050e+00 -4.56166059e-01 -4.41894829e-01 -1.36489332e-01 5.75948596e-01 -6.63346171e-01 9.37185049e-01 -1.03670847e+00 -2.05772662e+00 4.14537013e-01 9.06628907e-01 -1.16771054e+00 5.87447464e-01 -2.25829512e-01 -5.77432632e-01 2.75402904e-01 -5.11159956e-01 8.91016498e-02 1.62711406e+00 -1.02478051e+00 -5.05033612e-01 -4.35284972e-01 1.37050420e-01 -2.45445818e-01 -4.39257771e-01 2.08248198e-01 1.04292476e+00 -5.32878578e-01 -7.82908261e-01 -1.28152788e+00 1.21060811e-01 -6.42711520e-01 -2.53610134e-01 -8.08338299e-02 1.67269683e+00 -3.23709369e-01 7.31201530e-01 -1.70953572e+00 -4.59838867e-01 2.60284394e-01 8.81332606e-02 1.06465924e+00 -1.55075625e-01 5.87809563e-01 3.94374132e-01 2.35886946e-01 5.07942736e-02 4.52142984e-01 -6.64349347e-02 2.15882227e-01 -6.92580938e-01 6.60758972e-01 6.24880493e-02 7.30200946e-01 -1.12081337e+00 -2.52934638e-02 5.29582500e-01 4.20713097e-01 -3.41321796e-01 5.11214435e-01 -2.33728498e-01 3.91174644e-01 -5.11702776e-01 9.13256049e-01 6.10875130e-01 3.60185355e-02 -3.09828501e-05 2.01180145e-01 -7.70376995e-02 -5.88762760e-01 -7.28712976e-01 3.12056005e-01 -3.73109490e-01 2.41655037e-01 5.88256791e-02 -6.12464607e-01 1.12441266e+00 2.58437872e-01 8.14415574e-01 -3.73023808e-01 8.35286140e-01 1.33668676e-01 4.97573391e-02 -8.85457397e-01 4.36074167e-01 1.00449890e-01 6.10415041e-02 1.00663733e+00 -1.37704149e-01 1.00588631e-02 -2.23764196e-01 -2.04078197e-01 1.61541271e+00 -6.25911052e-04 3.89762491e-01 -1.10203736e-02 6.19292557e-01 2.45882630e-01 2.81653613e-01 9.21323061e-01 -7.82987535e-01 -3.80962223e-01 -1.50720149e-01 -8.37077796e-01 -8.74591172e-01 -9.58883941e-01 1.68980345e-01 9.99295712e-01 2.59603471e-01 2.21059322e-01 -9.23876405e-01 -1.03123093e+00 3.30727518e-01 6.73328817e-01 -5.69276154e-01 -5.80559015e-01 -5.67206740e-01 -6.88093483e-01 9.73449349e-01 1.68297645e-02 8.62532794e-01 -1.40414941e+00 -1.60848427e+00 2.89063573e-01 5.53940833e-01 -8.69639695e-01 -2.16807693e-01 6.69239312e-02 -5.30580103e-01 -1.53248715e+00 -1.85742557e-01 -8.68001506e-02 4.70194191e-01 3.95317763e-01 3.13795894e-01 3.74264151e-01 -4.72176403e-01 7.79686749e-01 -6.91336215e-01 -8.22111785e-01 -7.37739563e-01 -2.14906529e-01 3.67422700e-01 2.62938648e-01 5.47174931e-01 1.99509207e-02 -6.15230739e-01 5.19835055e-01 -1.09101892e+00 -9.27147746e-01 5.03143668e-01 8.67057562e-01 -1.82599708e-01 4.31167752e-01 5.73829293e-01 -5.40287673e-01 8.65428269e-01 -5.42092323e-01 -1.22915101e+00 4.18342888e-01 -1.14542198e+00 -1.56347990e-01 1.20526695e+00 -8.97669017e-01 -7.41582155e-01 6.76504001e-02 6.74413979e-01 -5.81192076e-01 -2.97722518e-01 -1.85590222e-01 3.94488931e-01 -4.12893027e-01 9.31821227e-01 2.97097951e-01 3.35794806e-01 2.33522192e-01 1.15198009e-02 1.04939353e+00 2.07627326e-01 -2.32733220e-01 1.15433204e+00 5.78584909e-01 -7.76365325e-02 -7.57173359e-01 -1.18006282e-01 -3.12293798e-01 -8.28537717e-02 -4.79024172e-01 8.82007778e-01 -4.45549190e-01 -1.79867661e+00 8.06525707e-01 -9.78396654e-01 -1.86002955e-01 -3.86793800e-02 3.12033951e-01 -3.50014687e-01 2.58060068e-01 -7.52730072e-01 -1.25082266e+00 -7.51149535e-01 -8.51021826e-01 8.28654170e-01 5.20842135e-01 -1.64670572e-01 -6.96422398e-01 5.03180742e-01 6.35091841e-01 8.72731984e-01 1.83993191e-01 4.21907723e-01 -1.22046137e+00 -6.21030390e-01 -6.58148944e-01 1.49031818e-01 3.02423000e-01 4.99066338e-02 4.35507447e-01 -8.43427300e-01 -5.86098731e-01 4.30033058e-01 -2.66248226e-01 2.40191698e-01 7.71846548e-02 7.59532392e-01 -9.85492766e-01 -2.68299282e-01 3.66505235e-01 1.25295174e+00 1.01048911e+00 4.74706382e-01 6.07258499e-01 2.93154716e-01 3.54457051e-01 6.43436491e-01 7.98981965e-01 1.28677249e-01 3.16006988e-01 9.34822321e-01 4.82199043e-01 7.53722072e-01 -3.03957313e-01 9.39315379e-01 2.63983130e-01 -9.22683030e-02 -3.89558405e-01 -8.01111996e-01 -2.39064515e-01 -1.61763859e+00 -1.15304315e+00 4.13405567e-01 2.26541448e+00 1.21694252e-01 -9.51852202e-02 5.73086977e-01 1.69535577e-01 7.42442787e-01 -2.83173829e-01 -8.97828162e-01 -8.35604429e-01 1.97197989e-01 8.38064179e-02 9.66141105e-01 3.10728818e-01 -1.10379446e+00 8.59552503e-01 5.70415735e+00 2.68499911e-01 -1.55535161e+00 4.43959162e-02 2.57373363e-01 2.70253599e-01 2.10701451e-01 -1.14964172e-01 -5.06107628e-01 6.14799380e-01 1.30134523e+00 -2.16860220e-01 1.20119834e+00 9.38362479e-01 3.86690289e-01 -8.92872177e-03 -7.07677543e-01 7.94740796e-01 2.04794750e-01 -9.29081321e-01 1.93527562e-03 1.29022300e-02 4.83779728e-01 -9.69948694e-02 2.47269608e-02 4.14894700e-01 6.79041445e-01 -9.57456112e-01 3.59587193e-01 3.07240397e-01 2.47129574e-02 -8.01065028e-01 1.03616893e+00 5.23625731e-01 -8.55810761e-01 -8.98794472e-01 -2.94617355e-01 -1.04351401e-01 -2.16516435e-01 1.21549211e-01 -1.54466271e+00 2.82862246e-01 6.79322302e-01 1.88076869e-01 -5.00435352e-01 9.23181951e-01 -7.71884471e-02 4.55631286e-01 -2.21042916e-01 -8.98477674e-01 3.17863710e-02 -2.91985184e-01 8.02398026e-01 7.33534634e-01 3.56925786e-01 2.11949036e-01 2.57301569e-01 7.14130938e-01 2.36847565e-01 -1.64905936e-01 -1.00271857e+00 -3.97474289e-01 6.86439753e-01 1.30488539e+00 -7.80058563e-01 -3.43767703e-01 3.80594611e-01 8.64225388e-01 -3.87620717e-01 2.53643185e-01 -1.11592221e+00 -4.86935169e-01 3.74962032e-01 1.09321281e-01 -9.04699415e-03 -4.06116992e-02 -1.03100583e-01 -8.71460319e-01 -4.67055440e-01 -1.22250140e+00 5.34592628e-01 -6.39418781e-01 -1.49673271e+00 7.26891577e-01 -1.82433605e-01 -1.25344074e+00 -3.20795417e-01 -7.50846028e-01 -4.57301855e-01 1.05501050e-02 -8.89536142e-01 -7.56473362e-01 -1.69473082e-01 6.66262150e-01 1.34407431e-01 -7.63650119e-01 6.25122249e-01 -2.91223973e-01 -6.26542926e-01 4.18519139e-01 -1.35617211e-01 6.78374842e-02 3.06025654e-01 -7.86342919e-01 -1.26454029e-02 7.57069468e-01 -2.39074484e-01 5.75181544e-01 6.07811213e-01 -1.02080715e+00 -2.02769685e+00 -9.61518586e-01 7.66818523e-02 -7.07460761e-01 8.87315929e-01 -8.29294100e-02 -4.42692667e-01 3.52850258e-01 4.65468228e-01 -1.59623504e-01 5.04169047e-01 -9.19241607e-01 -3.36375237e-01 -3.70435894e-01 -2.02403617e+00 5.20599365e-01 3.64301026e-01 -4.67470169e-01 -3.37670803e-01 3.80742908e-01 5.38103282e-01 -1.33659244e-01 -6.36706889e-01 1.56003982e-01 6.43903852e-01 -1.00735772e+00 7.25903034e-01 -7.02226460e-01 -4.75079268e-01 -4.54624921e-01 7.85604864e-02 -1.16737676e+00 1.72888890e-01 -1.27206600e+00 -4.33182955e-01 7.27592051e-01 -3.14412087e-01 -1.29927731e+00 7.53408313e-01 3.72434914e-01 4.54675198e-01 -3.91003907e-01 -1.12188005e+00 -1.28571713e+00 -2.99567431e-01 2.22138956e-01 8.79965842e-01 1.03320229e+00 -1.97207136e-03 -3.94195840e-02 -5.65353513e-01 3.79428059e-01 6.08389556e-01 -1.16172224e-01 8.40125382e-01 -1.04700601e+00 -2.58951843e-01 -2.71090716e-01 -4.89284009e-01 -1.95785895e-01 6.56795725e-02 -5.56753874e-01 -6.18730746e-02 -5.22333145e-01 -1.94564879e-01 -3.06101799e-01 -3.18847686e-01 6.79926753e-01 3.13797742e-01 -1.01355098e-01 4.63477582e-01 2.56594978e-02 -3.10635686e-01 3.45616013e-01 8.20227563e-01 -3.64504874e-01 -4.08190399e-01 1.47543311e-01 -7.95584917e-02 4.32154238e-01 1.08491004e+00 -7.39665687e-01 -3.00062507e-01 2.66676217e-01 1.34129047e-01 2.18548283e-01 5.79853117e-01 -1.27826655e+00 1.94345325e-01 -5.98041415e-01 2.94542968e-01 -2.34418273e-01 -1.79588228e-01 -1.57894266e+00 3.04171711e-01 1.46806765e+00 1.17805868e-01 6.26160741e-01 1.31707668e-01 6.37967288e-01 4.88988966e-01 -2.80747801e-01 8.22771609e-01 -3.18718180e-02 -2.11969465e-01 3.76048535e-01 -9.55341041e-01 -4.18695748e-01 1.92825413e+00 -3.06564271e-01 -8.46927702e-01 -1.37611791e-01 1.98199987e-01 1.09831519e-01 6.27338469e-01 6.21914029e-01 9.04116273e-01 -8.32129776e-01 -2.54364878e-01 3.47033143e-01 -2.78810173e-01 -7.61149704e-01 -7.36161843e-02 5.56409240e-01 -8.19225967e-01 2.04184324e-01 -7.63480425e-01 -4.70539749e-01 -1.08680749e+00 1.22495222e+00 7.01200664e-01 -2.25147873e-01 -1.17795572e-01 2.29901150e-01 -8.32692266e-01 -5.33510923e-01 1.32916927e-01 -8.61547887e-02 -7.70093128e-02 -9.09274220e-02 5.45639992e-01 9.09681201e-01 -7.20665604e-02 -4.45280284e-01 -5.21277368e-01 2.70021349e-01 8.35164562e-02 1.90878794e-01 1.04721200e+00 3.05684358e-01 -1.41899943e-01 -1.72966227e-01 4.86632675e-01 -1.75871670e-01 -1.07869780e+00 5.25215805e-01 5.60721792e-02 -5.84711194e-01 -3.46205890e-01 -1.19642103e+00 -8.29457998e-01 3.72776777e-01 1.22359431e+00 7.24177063e-01 1.00417447e+00 -7.14279830e-01 1.06562173e+00 8.76166940e-01 8.41297686e-01 -1.08746243e+00 4.10498261e-01 6.05504632e-01 6.04459941e-01 -1.15814734e+00 -3.30944806e-02 2.57290006e-01 -7.26180375e-01 1.26361430e+00 8.65039170e-01 -4.25296545e-01 6.47369623e-01 5.04767179e-01 2.53125370e-01 -1.92883715e-01 -5.76959550e-01 4.21566844e-01 -4.35486466e-01 1.24428451e+00 -5.47105908e-01 4.05002892e-01 6.88696206e-02 1.55617520e-01 -1.86787784e-01 1.12518914e-01 7.49307096e-01 9.43025410e-01 -5.18853724e-01 -8.14087808e-01 -9.46353078e-01 4.43522841e-01 -4.06745493e-01 4.65331107e-01 -7.81094491e-01 6.12689793e-01 2.13853046e-01 1.24400723e+00 -1.44264564e-01 -9.66842651e-01 4.56722900e-02 -3.49380374e-01 1.54635534e-01 -4.50645126e-02 -1.37897098e+00 -5.45501709e-01 -3.72604609e-01 -4.81045216e-01 -1.33223400e-01 -3.65688592e-01 -1.30461586e+00 -7.11953759e-01 -2.58495808e-01 -1.99363977e-02 7.09945023e-01 8.59606624e-01 4.78410989e-01 -1.64172783e-01 1.16010582e+00 -6.91817105e-01 -1.21778893e+00 -8.03352356e-01 -3.78205419e-01 7.87770003e-02 3.89865547e-01 -5.89878738e-01 -4.41205263e-01 -3.12104344e-01]
[5.409328937530518, 7.384851932525635]
cff32782-e117-43db-b1b7-a87921abb67d
towards-better-characterization-of
null
null
https://openreview.net/forum?id=t2UJIFZVyz4
https://openreview.net/pdf?id=t2UJIFZVyz4
Towards Better Characterization of Paraphrases
To effectively characterize the nature of paraphrase pairs without expert human annotation, we proposes two new metrics: word position deviation (WPD) and lexical deviation (LD). WPD measures the degree of structural alteration, while LD measures the difference in vocabulary used. We apply these metrics to better understand the commonly-used MRPC dataset and study how it differs from PAWS, another paraphrase identification dataset. We also perform a detailed study on MRPC and propose improvements to the dataset, showing that it improves generalizability of models trained on the dataset. Lastly, we apply our metrics to filter the output of a paraphrase generation model and show how it can be used to generate specific forms of paraphrases for data augmentation or robustness testing of NLP models.
['Anonymous']
2021-09-17
null
null
null
acl-arr-september-2021-9
['paraphrase-generation', 'paraphrase-identification', 'paraphrase-generation']
['computer-code', 'natural-language-processing', 'natural-language-processing']
[ 5.21799862e-01 -3.62944640e-02 -2.86716968e-01 -2.30377346e-01 -8.74551177e-01 -1.16411090e+00 6.72434032e-01 5.63707471e-01 -3.94551039e-01 5.58185637e-01 7.11941242e-01 -5.62044442e-01 -8.82438645e-02 -6.12799227e-01 -7.01704264e-01 -6.01618588e-02 6.20232284e-01 3.70892674e-01 2.08988309e-01 -3.11985999e-01 8.73442650e-01 5.25098026e-01 -1.49330211e+00 7.88249135e-01 8.86655986e-01 3.86507183e-01 9.82904881e-02 5.35710752e-01 -4.09546606e-02 4.65767503e-01 -8.03631067e-01 -4.83655006e-01 3.04373831e-01 -3.79322648e-01 -1.03287554e+00 -5.61904371e-01 5.82915485e-01 1.62466958e-01 -3.19820270e-02 7.92204618e-01 4.93835151e-01 -7.95357972e-02 8.14174116e-01 -1.08069646e+00 -9.17247593e-01 8.22185278e-01 -2.18155179e-02 5.23654282e-01 1.07237685e+00 2.59881407e-01 1.18682885e+00 -1.02316439e+00 8.78377914e-01 9.73820210e-01 8.38822782e-01 3.40204626e-01 -1.42029119e+00 -5.27020574e-01 -3.55569154e-01 3.58181894e-01 -1.01311398e+00 -4.29640085e-01 6.95950568e-01 -4.41394687e-01 1.40324557e+00 4.15817261e-01 6.23607695e-01 1.47426021e+00 -5.55473492e-02 6.16741896e-01 1.22030032e+00 -7.38423228e-01 1.96199909e-01 1.47896215e-01 5.54275036e-01 3.78567308e-01 4.47685391e-01 -6.75233686e-03 -5.32244742e-01 -4.49204087e-01 2.55581468e-01 -4.35490787e-01 -3.93670380e-01 -3.92436504e-01 -1.09739459e+00 9.58045304e-01 -1.25264213e-03 3.55690092e-01 -6.74828067e-02 -2.26986825e-01 5.29209614e-01 7.19522536e-01 2.29007870e-01 1.34544599e+00 -5.13122678e-01 -4.93177384e-01 -1.14667654e+00 5.23099303e-01 9.60178137e-01 1.10450447e+00 4.46573973e-01 -4.89095211e-01 -3.02919149e-01 1.16958559e+00 -1.70537218e-01 3.86612147e-01 1.00009072e+00 -9.27727282e-01 7.33690023e-01 9.51286077e-01 6.54555187e-02 -8.95280004e-01 -1.96315706e-01 -1.36012986e-01 -7.78825879e-02 -3.53673905e-01 3.42080384e-01 2.70378888e-01 -4.54985410e-01 1.58666527e+00 -3.66781890e-01 -3.54418725e-01 -6.50551766e-02 3.74351710e-01 7.52135575e-01 3.64583343e-01 -1.57685056e-01 -1.50984414e-02 1.15357554e+00 -7.24449694e-01 -2.37474889e-01 -4.15235043e-01 8.69193792e-01 -1.07432473e+00 1.65786278e+00 1.86873093e-01 -1.15521741e+00 -4.90138441e-01 -1.24026525e+00 -3.11490923e-01 -5.81358790e-01 -1.19095333e-01 2.28372559e-01 6.99612379e-01 -7.79802680e-01 9.26489949e-01 -3.14446360e-01 -6.16993785e-01 5.40760495e-02 3.89915928e-02 -3.23320389e-01 6.91728890e-02 -1.26298523e+00 1.40870070e+00 4.16117728e-01 -6.61487699e-01 -1.70815542e-01 -1.03098345e+00 -8.51474822e-01 8.45164508e-02 -1.73665836e-01 -7.24077344e-01 1.09382188e+00 -5.08507133e-01 -1.27933574e+00 1.26999831e+00 -1.88674808e-01 -5.43729961e-01 3.28754187e-01 -4.96277399e-02 -2.75673836e-01 7.59429112e-02 1.82936504e-01 6.15435779e-01 5.32909930e-01 -7.83392012e-01 -3.13838929e-01 -8.09804723e-02 4.54189107e-02 1.05878912e-01 -2.69933164e-01 8.03393945e-02 -1.74721986e-01 -9.23193455e-01 -1.46873236e-01 -9.73122776e-01 2.72821903e-01 -3.92264545e-01 -4.98512506e-01 -7.28061646e-02 5.24069250e-01 -8.29691529e-01 1.63063002e+00 -1.88150394e+00 1.92772985e-01 2.68383831e-01 -2.88894713e-01 4.32844311e-01 -6.35128438e-01 9.14125085e-01 -4.66948479e-01 3.92384708e-01 -2.41861328e-01 -9.92995948e-02 -4.53392640e-02 5.02864495e-02 -5.12681246e-01 5.09883324e-03 1.97054967e-01 1.19004738e+00 -8.32031429e-01 -2.68941104e-01 1.19707324e-01 -1.25478715e-01 -6.88416421e-01 -3.44323702e-02 -1.24404624e-01 6.61066920e-02 -2.46620309e-02 4.74898368e-01 5.27645946e-01 3.21051180e-02 -1.74985304e-02 2.53628381e-02 1.10418208e-01 8.37469995e-01 -6.07108474e-01 1.50055265e+00 -7.33825803e-01 7.01324463e-01 -5.98410904e-01 -7.46097326e-01 8.94003034e-01 -4.37569357e-02 -1.58278905e-02 -7.90706813e-01 -1.85529843e-01 3.13412130e-01 -1.85321093e-01 -4.47683156e-01 7.51962006e-01 -1.78792439e-02 -2.64037013e-01 6.46006882e-01 -1.88384980e-01 -4.15260553e-01 5.15983164e-01 2.63483942e-01 1.42930651e+00 -2.27582768e-01 7.30656564e-01 -3.05116326e-01 6.74696624e-01 2.61817873e-01 8.13170001e-02 1.02286363e+00 -1.98791847e-02 7.23661065e-01 6.28601253e-01 -5.39369993e-02 -1.41080523e+00 -1.19932389e+00 -1.43791497e-01 8.39176059e-01 -1.59907058e-01 -7.67560422e-01 -7.47263730e-01 -8.63464177e-01 4.12545472e-01 1.21449459e+00 -5.60155272e-01 -6.21828794e-01 -5.98648489e-01 -4.34254676e-01 9.78844345e-01 6.76183939e-01 1.44976722e-02 -1.39441526e+00 -2.82895416e-01 -7.31942058e-02 -5.38976073e-01 -9.12031233e-01 -7.01424956e-01 3.52352411e-02 -8.30339968e-01 -1.34873831e+00 -4.07621503e-01 -9.44762230e-01 3.45492333e-01 2.62866527e-01 1.46405911e+00 1.32711112e-01 -1.93805009e-01 2.94436425e-01 -7.45608270e-01 -1.71182141e-01 -1.10257745e+00 4.95965451e-01 -7.70849809e-02 -8.75455618e-01 8.64272654e-01 -6.45908058e-01 -4.78660971e-01 2.51851261e-01 -9.12922084e-01 -6.15049116e-02 3.85784209e-01 8.36531997e-01 2.02085271e-01 -4.85962510e-01 6.77084804e-01 -1.01120985e+00 1.40724695e+00 -3.76260221e-01 -2.88526803e-01 5.00544965e-01 -8.65327477e-01 3.11788380e-01 6.29150987e-01 -4.63928312e-01 -5.09196997e-01 -4.55873273e-02 -1.83800310e-01 -8.26791078e-02 -7.63541088e-02 5.40486157e-01 2.58455668e-02 1.06279917e-01 8.54305983e-01 2.79942513e-01 5.92439510e-02 -6.26934826e-01 6.14584029e-01 9.81470168e-01 5.36701202e-01 -4.01096404e-01 7.31400907e-01 -1.41774967e-01 -3.72151226e-01 -6.11249030e-01 -7.04600930e-01 -6.59861684e-01 -6.71561003e-01 4.36287075e-01 3.65423113e-01 -7.05885112e-01 -3.88273925e-01 2.00221434e-01 -1.17932844e+00 -2.67540425e-01 -5.10528386e-01 8.06689449e-03 -7.06248224e-01 7.31344342e-01 -5.50861239e-01 -6.14160746e-02 -4.57910419e-01 -8.94148231e-01 1.02237916e+00 -1.72164679e-01 -1.12160337e+00 -1.04174101e+00 4.52501684e-01 6.43932045e-01 3.46955508e-01 -1.37705967e-01 1.59590662e+00 -1.12556505e+00 -1.86846573e-02 -3.30684930e-01 -7.79781118e-02 5.39373517e-01 2.26222575e-01 5.48719913e-02 -5.79207420e-01 -2.04167068e-01 -1.04254171e-01 -2.87725210e-01 6.89115882e-01 1.06036745e-03 1.01271546e+00 -4.23573196e-01 -1.33746982e-01 4.54926759e-01 1.10480797e+00 -4.50475290e-02 7.98075855e-01 6.41767979e-01 2.73407668e-01 5.51160455e-01 5.66244841e-01 1.99658915e-01 -3.32527198e-02 8.96778286e-01 -8.70668218e-02 4.22805309e-01 -2.54584879e-01 -6.99798763e-01 5.22695422e-01 8.84386659e-01 4.24195975e-01 -2.50798881e-01 -7.13056266e-01 5.17588735e-01 -1.61047709e+00 -1.11185026e+00 -1.89336672e-01 2.22028589e+00 1.16027987e+00 1.80191472e-01 1.89087018e-01 2.43674546e-01 5.63893616e-01 9.11426768e-02 -3.35366011e-01 -9.07395959e-01 -2.98744142e-01 6.68167412e-01 4.97669101e-01 5.25514245e-01 -6.86893582e-01 1.08847010e+00 8.09052372e+00 8.96594405e-01 -6.24316871e-01 -2.73024768e-01 -6.78745955e-02 -1.95710853e-01 -5.47981679e-01 1.39429584e-01 -6.44694805e-01 8.22547972e-01 9.41308975e-01 -3.88126850e-01 5.83060324e-01 8.73094678e-01 2.67968923e-01 1.00530656e-02 -1.56226873e+00 7.86661625e-01 2.63553053e-01 -1.37770176e+00 4.23414886e-01 -1.86279535e-01 4.37232643e-01 -2.07967281e-01 -1.17252156e-01 4.69554842e-01 2.92717069e-01 -9.16122556e-01 4.34995919e-01 3.79207432e-01 7.50256658e-01 -5.62029004e-01 6.51030838e-01 3.13539892e-01 -7.32866883e-01 -1.07138082e-01 -5.18584251e-01 -3.76503207e-02 4.67248000e-02 3.40497553e-01 -1.08184659e+00 3.47883433e-01 1.92135736e-01 6.95258856e-01 -1.50380385e+00 9.24253404e-01 -4.75867212e-01 5.67823052e-01 -1.48320079e-01 -2.46133327e-01 -1.41938612e-01 -2.05201581e-01 7.61531949e-01 1.48640609e+00 2.84736037e-01 -5.19850791e-01 -4.06116933e-01 9.91953850e-01 -7.39787072e-02 2.38876492e-01 -7.97475457e-01 -1.73439533e-01 1.03354228e+00 9.91125584e-01 -1.81300297e-01 -4.03215706e-01 -2.37979501e-01 1.28623021e+00 3.69194806e-01 6.49782345e-02 -8.16236734e-01 -6.12541854e-01 6.30582869e-01 1.84974089e-01 1.30017087e-01 1.10613249e-01 -3.91085774e-01 -1.30602849e+00 2.74738431e-01 -1.34544611e+00 3.28185946e-01 -1.15084803e+00 -1.63436329e+00 2.61416048e-01 1.33306056e-01 -1.16956854e+00 -4.69269037e-01 -6.82363153e-01 -8.50505054e-01 1.01751876e+00 -1.03059268e+00 -8.11944485e-01 -2.67734617e-01 2.46021733e-01 7.04716921e-01 -3.80604655e-01 8.66810322e-01 -1.26909595e-02 -4.00478810e-01 9.15908158e-01 1.34705096e-01 1.05333902e-01 1.05896151e+00 -1.43815422e+00 1.04637992e+00 7.76935875e-01 2.10181952e-01 1.27516532e+00 8.22430730e-01 -7.01916814e-01 -8.73874724e-01 -9.07864988e-01 1.19133317e+00 -8.80183637e-01 9.12761807e-01 -4.10655618e-01 -9.93032634e-01 5.57651699e-01 1.65626660e-01 -7.77465522e-01 7.26998508e-01 2.28617698e-01 -7.99995482e-01 3.72821659e-01 -1.11927605e+00 8.41666818e-01 1.21463478e+00 -8.53995740e-01 -1.37810147e+00 4.18711811e-01 8.08117926e-01 -1.01944625e-01 -8.11022460e-01 3.20346445e-01 5.34731746e-01 -9.91240263e-01 1.09277773e+00 -9.25921202e-01 6.99492097e-01 8.39564577e-02 -9.83807892e-02 -1.59054697e+00 -5.48195541e-01 -3.30037564e-01 2.45992448e-02 1.28997564e+00 8.28392863e-01 -5.17287016e-01 8.23186040e-01 4.34962481e-01 -5.98854292e-03 -5.76104820e-01 -7.02074111e-01 -1.13780344e+00 6.40380561e-01 -1.93358794e-01 5.80948293e-01 9.57630992e-01 6.73479974e-01 5.48867702e-01 8.24606195e-02 -3.47516894e-01 -2.80474275e-02 2.14149226e-02 7.42058456e-01 -1.04555798e+00 -4.94819313e-01 -6.07157111e-01 -2.44161710e-01 -9.65671420e-01 2.67649472e-01 -1.39181268e+00 -3.39389384e-01 -1.41783488e+00 3.75705510e-01 -5.24604805e-02 2.09907234e-01 3.60515594e-01 -3.06916684e-01 2.02755317e-01 2.63038188e-01 4.38873291e-01 -5.23374900e-02 3.06781709e-01 5.90969265e-01 -1.10225640e-01 -3.48317981e-01 3.75238955e-02 -8.36642504e-01 6.20419562e-01 9.80882108e-01 -6.11617565e-01 -4.88611072e-01 -2.50127792e-01 5.30848205e-01 -3.57509285e-01 4.04085189e-01 -7.90324271e-01 -2.96682063e-02 -1.84317753e-02 1.67317241e-01 -5.88839471e-01 -3.88120464e-03 -3.79393667e-01 -4.89427559e-02 4.89359945e-01 -8.32689047e-01 5.49928904e-01 1.61199704e-01 2.82853395e-01 -6.44375235e-02 -9.85443294e-01 8.02517653e-01 -9.95008051e-02 -2.99902380e-01 -5.20949841e-01 -6.01284087e-01 4.22946215e-01 7.17660427e-01 -4.10864323e-01 -5.88748753e-01 -2.75114328e-01 -4.23401564e-01 7.08733425e-02 9.67751682e-01 8.20232213e-01 4.89262074e-01 -1.23658979e+00 -6.24652326e-01 2.20547423e-01 6.46905184e-01 -7.39891171e-01 -3.20580870e-01 4.69679147e-01 -6.82846785e-01 5.78581393e-01 -1.63427651e-01 -1.09621353e-01 -1.61694098e+00 7.62108743e-01 6.68613687e-02 -5.62159240e-01 -4.42658454e-01 4.31492358e-01 -3.58987302e-01 -7.32795477e-01 -8.17371681e-02 -5.12982130e-01 -2.15509117e-01 1.87768325e-01 3.14548522e-01 4.27327454e-01 3.56355220e-01 -3.13692719e-01 -4.19072002e-01 4.26237553e-01 -3.67894024e-01 -2.06229940e-01 1.09897649e+00 -8.40150267e-02 -2.25291047e-02 3.03613007e-01 1.37328672e+00 3.92150104e-01 -4.16345000e-01 -7.51309395e-02 4.72017318e-01 -2.87340045e-01 -5.11020064e-01 -1.08657348e+00 -2.25135043e-01 5.77134550e-01 1.78320721e-01 2.33854845e-01 8.16248596e-01 -1.25160605e-01 8.41820598e-01 6.09347403e-01 2.39619121e-01 -1.20211446e+00 1.90611929e-01 8.48328590e-01 1.18648934e+00 -7.38543928e-01 1.85281023e-01 -4.50769991e-01 -7.77384639e-01 1.05001867e+00 2.66957819e-01 -1.10060968e-01 2.79694080e-01 3.96960340e-02 -8.08395520e-02 1.76806170e-02 -7.07611501e-01 1.03367314e-01 3.64338875e-01 7.35307038e-01 4.37224180e-01 -3.86187196e-01 -7.30728567e-01 6.28616750e-01 -9.37931061e-01 2.31252592e-02 7.09626675e-01 7.98306823e-01 -3.38938624e-01 -1.55553865e+00 -4.97011058e-02 5.42836189e-01 -2.25104019e-01 -6.12770736e-01 -1.19684398e+00 9.44084108e-01 -1.90757245e-01 9.73915815e-01 -1.50582455e-02 -5.37797570e-01 5.88763416e-01 5.86825252e-01 5.58822155e-01 -9.07641172e-01 -8.80780339e-01 -6.43624008e-01 3.92404228e-01 -4.76568907e-01 -6.62852302e-02 -6.84748471e-01 -7.59934843e-01 -4.61741060e-01 -2.88663179e-01 1.32580101e-01 4.38802928e-01 7.28556871e-01 6.25533879e-01 -3.05296667e-02 5.12244940e-01 -2.22361520e-01 -9.04640734e-01 -1.26712370e+00 -2.80730814e-01 1.05045116e+00 -1.78063154e-01 -2.72565722e-01 -6.85902297e-01 6.43243566e-02]
[11.33221435546875, 9.224584579467773]
59e15ac2-84e3-474e-ae81-8c6ce383e113
retrieving-to-answer-zero-shot-video-question
2306.11732
null
https://arxiv.org/abs/2306.11732v1
https://arxiv.org/pdf/2306.11732v1.pdf
Retrieving-to-Answer: Zero-Shot Video Question Answering with Frozen Large Language Models
Video Question Answering (VideoQA) has been significantly advanced from the scaling of recent Large Language Models (LLMs). The key idea is to convert the visual information into the language feature space so that the capacity of LLMs can be fully exploited. Existing VideoQA methods typically take two paradigms: (1) learning cross-modal alignment, and (2) using an off-the-shelf captioning model to describe the visual data. However, the first design needs costly training on many extra multi-modal data, whilst the second is further limited by limited domain generalization. To address these limitations, a simple yet effective Retrieving-to-Answer (R2A) framework is proposed.Given an input video, R2A first retrieves a set of semantically similar texts from a generic text corpus using a pre-trained multi-modal model (e.g., CLIP). With both the question and the retrieved texts, a LLM (e.g., DeBERTa) can be directly used to yield a desired answer. Without the need for cross-modal fine-tuning, R2A allows for all the key components (e.g., LLM, retrieval model, and text corpus) to plug-and-play. Extensive experiments on several VideoQA benchmarks show that despite with 1.3B parameters and no fine-tuning, our R2A can outperform the 61 times larger Flamingo-80B model even additionally trained on nearly 2.1B multi-modal data.
['Hongsheng Li', 'Yu Qiao', 'Yi Wang', 'Renrui Zhang', 'Xiatian Zhu', 'Yuying Ge', 'Ziyi Lin', 'Junting Pan']
2023-06-15
null
null
null
null
['video-question-answering', 'domain-generalization', 'question-answering']
['computer-vision', 'methodology', 'natural-language-processing']
[ 8.55148062e-02 -2.24641055e-01 -8.95929337e-02 -2.61809856e-01 -1.58454156e+00 -8.83075476e-01 6.47936404e-01 -6.42782524e-02 -3.78374994e-01 3.12936366e-01 3.15561622e-01 -3.17625433e-01 2.36434475e-01 -5.94970465e-01 -9.23285246e-01 -4.67243582e-01 3.84606630e-01 6.45472467e-01 3.79317194e-01 -3.64145219e-01 1.36696309e-01 -5.12796529e-02 -1.51336324e+00 7.46102512e-01 6.39222860e-01 1.23343110e+00 5.42640388e-01 8.65517795e-01 -7.04152942e-01 8.57216358e-01 -4.87314463e-01 -6.45501018e-01 2.43698880e-01 -5.47499776e-01 -9.37021375e-01 1.72858033e-02 8.30715239e-01 -6.13721311e-01 -4.45191950e-01 7.87700176e-01 4.39409614e-01 3.04272830e-01 6.29187942e-01 -1.31792879e+00 -7.69243479e-01 1.56655118e-01 -5.01542389e-01 1.23362593e-01 7.25535750e-01 3.07095915e-01 1.22404826e+00 -1.18815899e+00 6.84501946e-01 1.44074500e+00 1.78502753e-01 6.71128690e-01 -9.41146731e-01 -6.01467371e-01 9.55577046e-02 3.04795772e-01 -1.38763273e+00 -6.13671064e-01 4.98769611e-01 -1.82916164e-01 9.13507760e-01 2.66230702e-01 4.01755869e-01 1.09046471e+00 -1.02476105e-01 1.12527537e+00 8.11201513e-01 -3.75732213e-01 1.08640946e-01 1.16595842e-01 -1.89446807e-01 7.39477038e-01 -2.67071128e-01 -3.93249124e-01 -6.75163269e-01 -2.18246460e-01 4.67441797e-01 -1.40552729e-01 -3.30738634e-01 -5.53875983e-01 -1.13934684e+00 8.73855233e-01 3.12770605e-01 7.60780498e-02 -1.48143575e-01 1.86939940e-01 5.50863802e-01 4.72863704e-01 1.35538623e-01 3.94298911e-01 -2.10349783e-01 -2.10863188e-01 -1.14842021e+00 2.45300815e-01 6.85407877e-01 1.15773201e+00 9.11839843e-01 -2.63758361e-01 -3.33366096e-01 8.67576241e-01 4.42815304e-01 8.86593103e-01 5.27460515e-01 -1.00908339e+00 9.35218811e-01 4.92617756e-01 4.45414074e-02 -8.43491852e-01 -1.05819888e-01 2.29903191e-01 -4.03666943e-01 -1.79842338e-01 5.69772422e-01 1.15489207e-01 -9.90721762e-01 1.63108885e+00 2.04027951e-01 9.63872299e-03 2.10106492e-01 1.12296605e+00 1.16993594e+00 1.05648220e+00 1.67019174e-01 6.05444722e-02 1.54706025e+00 -1.34658837e+00 -4.61059958e-01 -5.33222079e-01 6.92903161e-01 -9.45938230e-01 1.41349077e+00 1.74688786e-01 -1.31196451e+00 -6.34337306e-01 -8.56890738e-01 -4.80139315e-01 -3.97210270e-01 -8.09816569e-02 2.28785783e-01 3.90253454e-01 -1.19590878e+00 -1.18638113e-01 -3.87983948e-01 -6.18527055e-01 3.14161390e-01 1.11479208e-01 -5.08334935e-01 -6.86177671e-01 -1.15678954e+00 7.83130705e-01 2.63872594e-01 -2.37736136e-01 -1.04135203e+00 -5.28065383e-01 -9.52245414e-01 -1.59912687e-02 5.68453372e-01 -1.01848555e+00 1.22749329e+00 -1.33546185e+00 -1.38003564e+00 9.61735904e-01 -3.92799377e-01 -2.83535659e-01 3.89689803e-01 -4.45457488e-01 -3.42681497e-01 9.22108293e-01 1.90700069e-01 1.12241268e+00 1.18368268e+00 -1.26796019e+00 -6.43232763e-01 -1.53836638e-01 5.93047023e-01 5.35001278e-01 -4.15278554e-01 2.56979018e-01 -1.24071121e+00 -5.87575853e-01 -1.56365156e-01 -8.98272157e-01 2.04603106e-01 2.32165337e-01 5.10918945e-02 -1.36191830e-01 9.14388418e-01 -8.55549812e-01 1.13875151e+00 -2.27554941e+00 3.84260952e-01 1.31628225e-02 1.09849453e-01 2.21404731e-01 -7.39912033e-01 4.40822214e-01 1.36318237e-01 8.25770199e-03 -7.36350613e-03 -4.87200201e-01 -4.20761555e-02 1.63215667e-01 -4.75809634e-01 2.32647941e-01 3.07540536e-01 1.23548162e+00 -9.60714519e-01 -7.83372879e-01 9.83389467e-02 3.67158651e-01 -6.96133435e-01 4.96070594e-01 -5.15679002e-01 1.23730697e-01 -3.80941778e-01 6.26041651e-01 3.93608570e-01 -5.50818503e-01 7.95461144e-03 -4.09968823e-01 2.69065827e-01 2.11777776e-01 -9.63321805e-01 2.14390540e+00 -6.41980529e-01 8.24637353e-01 1.33976385e-01 -8.64206254e-01 6.35772526e-01 5.10701299e-01 3.47920209e-01 -1.06348896e+00 1.30859520e-02 2.40577817e-01 -4.30897623e-01 -8.63309979e-01 6.27404690e-01 -2.71270163e-02 -1.92896381e-01 5.35418093e-01 2.04679459e-01 -1.70179680e-01 3.34204078e-01 6.64125800e-01 9.64926898e-01 3.00181776e-01 2.04949081e-02 1.27218723e-01 6.39921725e-01 2.54285932e-01 5.41800298e-02 7.29129076e-01 -2.40739524e-01 8.11417341e-01 2.96195120e-01 -2.39932209e-01 -1.03171098e+00 -9.45002258e-01 4.27952945e-01 1.43834376e+00 3.22228849e-01 -5.08155882e-01 -7.04269767e-01 -8.02651048e-01 -3.89970504e-02 5.85807025e-01 -4.46584463e-01 -1.08378991e-01 -6.34771824e-01 -1.58156589e-01 6.02415979e-01 4.63020295e-01 4.24814612e-01 -9.95766819e-01 -5.80006897e-01 -6.41715601e-02 -7.59809971e-01 -1.32896197e+00 -7.11301982e-01 -2.84011155e-01 -5.74618518e-01 -8.86249721e-01 -9.44329023e-01 -8.18997502e-01 4.71798956e-01 7.92513549e-01 1.43432510e+00 2.83679724e-01 7.10259378e-02 9.72741961e-01 -6.63183451e-01 8.03545341e-02 -3.28379124e-01 -7.03490060e-03 -2.26960555e-01 8.52034763e-02 5.50862670e-01 -1.82311371e-01 -6.83576345e-01 3.17255348e-01 -1.13297915e+00 2.34743625e-01 6.75957859e-01 8.26355875e-01 5.40561199e-01 -4.44043636e-01 5.78217089e-01 -4.39116746e-01 3.39862764e-01 -5.63609719e-01 -3.63770872e-01 5.82838535e-01 -8.10012594e-02 9.99112055e-02 5.20439208e-01 -6.76703274e-01 -9.23800170e-01 -5.63807935e-02 1.38408467e-02 -8.64526868e-01 2.21017301e-02 5.39764404e-01 -2.60091543e-01 9.34438854e-02 4.31041837e-01 3.00165862e-01 -6.74687102e-02 -2.95178026e-01 7.71503329e-01 7.47874260e-01 5.75437665e-01 -6.57763362e-01 8.33197653e-01 5.40434301e-01 -4.24760342e-01 -7.16878712e-01 -9.05194521e-01 -7.99525976e-01 -4.39781904e-01 -3.36318970e-01 1.11869061e+00 -1.12419486e+00 -4.46600407e-01 1.52742356e-01 -1.17607653e+00 -3.50062937e-01 -1.21059678e-01 2.29891285e-01 -6.84523344e-01 4.03081715e-01 -4.58774984e-01 -5.76038599e-01 -3.70926380e-01 -1.15342391e+00 1.29280305e+00 1.79420799e-01 -1.30551860e-01 -7.35215724e-01 -1.45584285e-01 1.11411238e+00 4.39350426e-01 -3.18778664e-01 9.13961768e-01 -7.08595157e-01 -7.71087646e-01 -1.12627044e-01 -5.63662291e-01 1.65463611e-01 -1.50299937e-01 -2.53432512e-01 -8.81943703e-01 -4.73750055e-01 -6.75543398e-02 -9.66098249e-01 7.39564240e-01 2.78270878e-02 9.66479957e-01 -1.54068574e-01 -8.89170617e-02 5.17775059e-01 1.31999648e+00 1.24930397e-01 6.34329498e-01 4.13633317e-01 7.33478308e-01 5.55526316e-01 8.57390046e-01 7.22552389e-02 7.68747389e-01 7.56486595e-01 4.44803625e-01 6.42800108e-02 -3.05040956e-01 -3.78692687e-01 6.57118142e-01 8.44021618e-01 2.66162157e-01 -3.96969587e-01 -9.20113683e-01 6.55085504e-01 -1.78235078e+00 -9.12589371e-01 2.41234198e-01 2.01963401e+00 7.60522425e-01 -1.06913283e-01 2.13213503e-01 -4.24438387e-01 3.41173708e-01 2.72217005e-01 -5.64831316e-01 -1.48440570e-01 -1.42455876e-01 -4.83696349e-03 2.31736317e-01 3.86386722e-01 -9.29513454e-01 1.03824854e+00 5.60294437e+00 1.03706706e+00 -1.10602152e+00 1.84483841e-01 4.00322080e-01 -4.37830150e-01 -4.82906193e-01 4.47476655e-02 -7.43373811e-01 2.90621132e-01 1.08176780e+00 -3.20354924e-02 6.18026853e-01 4.63016003e-01 -9.04026702e-02 -1.36104256e-01 -1.10096574e+00 1.30404544e+00 6.57143474e-01 -1.28723228e+00 5.51819324e-01 -2.19555452e-01 4.96990025e-01 3.54495905e-02 -4.24271729e-03 7.17125475e-01 -5.63292019e-02 -1.00284457e+00 9.31222856e-01 4.32864487e-01 1.05537260e+00 -5.13650477e-01 5.55385947e-01 2.24928766e-01 -1.31403887e+00 -1.37726992e-01 -2.77762055e-01 2.48784229e-01 4.51539755e-01 -3.87327820e-02 -3.13823134e-01 6.33765757e-01 9.63631332e-01 4.24207777e-01 -7.34635830e-01 7.28178978e-01 3.19702737e-02 4.12163258e-01 -2.58691669e-01 4.96988706e-02 4.27632242e-01 -1.53419614e-01 5.08914888e-01 1.16403699e+00 2.57540196e-01 1.46052346e-01 1.84508234e-01 4.95775849e-01 -2.88039804e-01 3.35273117e-01 -3.77145648e-01 -2.37918198e-01 3.51927429e-01 1.24634242e+00 -3.53715330e-01 -5.89261174e-01 -9.95486677e-01 1.25238073e+00 3.98282290e-01 6.18723810e-01 -9.17950571e-01 -1.97200939e-01 4.29380745e-01 1.16518728e-01 4.88617331e-01 -1.09622918e-01 2.53604889e-01 -1.42171526e+00 1.11461908e-01 -1.48721337e+00 6.54010534e-01 -1.36650491e+00 -1.25382030e+00 6.06046379e-01 2.74409391e-02 -1.16366386e+00 -3.48075479e-01 -6.67906344e-01 -1.50481403e-01 8.26635242e-01 -1.65503943e+00 -1.51162720e+00 -3.88848811e-01 9.69194651e-01 9.50473845e-01 -1.57405019e-01 6.63408518e-01 5.59898019e-01 -2.63664514e-01 6.25104368e-01 -1.92923024e-02 1.72251925e-01 1.20857012e+00 -1.04630005e+00 1.79174557e-01 6.48184359e-01 3.53545934e-01 4.26768959e-01 4.92648065e-01 -4.00744230e-01 -1.67093265e+00 -9.15780842e-01 7.99739957e-01 -6.46211028e-01 8.91345084e-01 -3.97088200e-01 -1.04176557e+00 6.13749683e-01 3.26820105e-01 -5.54284230e-02 6.75850511e-01 -2.50952840e-01 -7.41290033e-01 -1.00574635e-01 -7.74336159e-01 7.53782809e-01 8.03548455e-01 -9.65611756e-01 -7.50496686e-01 2.48836651e-01 8.41180205e-01 -5.02492785e-01 -6.30746484e-01 1.38764635e-01 5.96655786e-01 -6.97987556e-01 1.12550318e+00 -8.28138411e-01 6.51725531e-01 -3.57881129e-01 -6.10949814e-01 -9.00826156e-01 1.53491246e-02 -5.90452015e-01 -2.95937091e-01 1.13696766e+00 3.27618510e-01 -1.36409521e-01 3.69766116e-01 7.68668711e-01 5.55181652e-02 -5.97272038e-01 -8.30067635e-01 -6.15049779e-01 1.11601353e-01 -5.00651240e-01 4.45079178e-01 7.14369714e-01 -1.42629832e-01 6.69627666e-01 -4.87481922e-01 1.12789363e-01 2.23742366e-01 2.94778705e-01 1.05459583e+00 -6.99191093e-01 -3.33392680e-01 -2.99154252e-01 -1.40866965e-01 -1.73069942e+00 6.10357746e-02 -7.94695437e-01 7.13016139e-03 -1.57975543e+00 4.65792358e-01 -2.06705526e-01 -1.38020039e-01 3.22976708e-01 -3.68359894e-01 3.96201015e-01 5.58825850e-01 3.70663732e-01 -1.19442773e+00 5.09577632e-01 1.25157940e+00 -2.77616680e-01 -6.42114878e-02 -4.23139215e-01 -6.14694238e-01 5.77474594e-01 5.19861042e-01 -2.45058671e-01 -5.37055016e-01 -7.83267796e-01 4.05354708e-01 3.94359112e-01 4.22350198e-01 -7.21889615e-01 3.34200263e-01 -3.69325131e-02 1.26770809e-01 -5.43706775e-01 6.60768509e-01 -8.10386300e-01 -2.49553785e-01 -3.99920903e-02 -3.23680431e-01 3.44617128e-01 3.44121397e-01 6.40642405e-01 -4.27113563e-01 -4.03068602e-01 5.16003549e-01 -6.91327453e-02 -9.44059193e-01 2.32421950e-01 -2.35758364e-01 3.90444547e-01 8.38306904e-01 -1.95173472e-01 -6.35907590e-01 -8.53435814e-01 -4.52798188e-01 6.08015299e-01 6.05237484e-01 7.34559715e-01 6.30600393e-01 -1.35388994e+00 -6.57185376e-01 -1.34195611e-01 5.04075587e-01 -4.02336195e-02 4.46895272e-01 6.66117013e-01 -4.45291609e-01 6.11158490e-01 -3.50735672e-02 -6.58854365e-01 -1.32185495e+00 7.75255263e-01 2.68269449e-01 -2.72572428e-01 -4.00031775e-01 7.59537280e-01 4.58442867e-01 -2.31106564e-01 2.24791095e-01 7.62346163e-02 -1.66923106e-01 2.28132382e-01 7.34471619e-01 3.73561382e-02 -1.15616076e-01 -9.73379314e-01 -3.17111552e-01 6.32208288e-01 -2.55337805e-02 -4.23293233e-01 1.06932843e+00 -4.35725868e-01 3.11894026e-02 3.64421278e-01 1.35743082e+00 -2.27497071e-02 -1.21738338e+00 -4.76889253e-01 -1.72681794e-01 -4.64132160e-01 -1.87686961e-02 -6.16896152e-01 -1.08284569e+00 1.04788792e+00 4.08375174e-01 -4.11211699e-03 1.24935663e+00 2.88940191e-01 1.08265126e+00 4.48086202e-01 2.72521615e-01 -1.08413780e+00 5.59781849e-01 4.07289594e-01 9.61570740e-01 -1.46460962e+00 -1.75971925e-01 -1.59147717e-02 -9.24491107e-01 1.05519068e+00 6.85061693e-01 2.10290700e-01 3.00253272e-01 -1.37961760e-01 4.00887907e-01 -3.18386078e-01 -8.92040193e-01 -2.83692926e-01 6.06151760e-01 4.13556397e-01 1.96124807e-01 -2.73409992e-01 1.61223844e-01 4.75153148e-01 7.62584805e-02 -1.75348788e-01 2.00135469e-01 7.97500193e-01 -4.06828284e-01 -8.89652014e-01 -4.37672645e-01 3.06301117e-01 -3.89908850e-01 -2.73078263e-01 -1.48646504e-01 7.71840632e-01 -2.48868659e-01 1.15585315e+00 2.16033265e-01 -1.82475150e-01 3.03147256e-01 2.82548875e-01 3.61491591e-01 -5.07444739e-01 -4.60898787e-01 9.71194059e-02 -1.21503836e-02 -7.92131126e-01 -6.37966752e-01 -5.28279305e-01 -1.05496228e+00 -1.25730604e-01 -3.05078536e-01 4.01189215e-02 5.55223346e-01 1.11681986e+00 4.21201587e-01 1.72943294e-01 1.98658392e-01 -5.77190340e-01 -3.44160467e-01 -7.86575735e-01 -7.10931346e-02 6.10101223e-01 3.32407683e-01 -4.53056276e-01 -2.74030060e-01 3.10830206e-01]
[10.630904197692871, 1.1666724681854248]
fa8f711c-8c88-4e38-aa6a-9ad846c44b6b
learning-depth-guided-convolutions-for
1912.04799
null
https://arxiv.org/abs/1912.04799v2
https://arxiv.org/pdf/1912.04799v2.pdf
Learning Depth-Guided Convolutions for Monocular 3D Object Detection
3D object detection from a single image without LiDAR is a challenging task due to the lack of accurate depth information. Conventional 2D convolutions are unsuitable for this task because they fail to capture local object and its scale information, which are vital for 3D object detection. To better represent 3D structure, prior arts typically transform depth maps estimated from 2D images into a pseudo-LiDAR representation, and then apply existing 3D point-cloud based object detectors. However, their results depend heavily on the accuracy of the estimated depth maps, resulting in suboptimal performance. In this work, instead of using pseudo-LiDAR representation, we improve the fundamental 2D fully convolutions by proposing a new local convolutional network (LCN), termed Depth-guided Dynamic-Depthwise-Dilated LCN (D$^4$LCN), where the filters and their receptive fields can be automatically learned from image-based depth maps, making different pixels of different images have different filters. D$^4$LCN overcomes the limitation of conventional 2D convolutions and narrows the gap between image representation and 3D point cloud representation. Extensive experiments show that D$^4$LCN outperforms existing works by large margins. For example, the relative improvement of D$^4$LCN against the state-of-the-art on KITTI is 9.1\% in the moderate setting. The code is available at https://github.com/dingmyu/D4LCN.
['Yuqi Huo', 'Zhiwu Lu', 'Mingyu Ding', 'Ping Luo', 'Hongwei Yi', 'Zhe Wang', 'Jianping Shi']
2019-12-10
learning-depth-guided-convolutions-for-1
http://openaccess.thecvf.com/content_CVPR_2020/html/Ding_Learning_Depth-Guided_Convolutions_for_Monocular_3D_Object_Detection_CVPR_2020_paper.html
http://openaccess.thecvf.com/content_CVPR_2020/papers/Ding_Learning_Depth-Guided_Convolutions_for_Monocular_3D_Object_Detection_CVPR_2020_paper.pdf
cvpr-2020-6
['vehicle-pose-estimation']
['computer-vision']
[ 6.90925196e-02 -2.06487507e-01 -8.49822536e-02 -2.74065077e-01 -3.59719068e-01 -4.84042585e-01 4.76747930e-01 -1.54725879e-01 -5.25306463e-01 1.68805912e-01 -2.45283231e-01 -4.01697159e-01 2.37735078e-01 -9.41123068e-01 -9.08313036e-01 -4.24398750e-01 8.87181163e-02 2.86533713e-01 7.36108661e-01 2.41292212e-02 2.46637151e-01 8.94353390e-01 -1.61899471e+00 1.02085717e-01 7.18426287e-01 1.43882072e+00 5.27758539e-01 3.75475436e-01 -5.72130144e-01 8.63474458e-02 -2.48877510e-01 -1.27493456e-01 6.78704321e-01 8.76055751e-03 -2.52049714e-01 -4.16738838e-02 8.60933602e-01 -7.73164988e-01 -4.18649852e-01 1.12837672e+00 4.23078269e-01 -9.73284841e-02 6.25981927e-01 -1.09550238e+00 -5.98728120e-01 1.53601263e-02 -9.12967443e-01 2.44475916e-01 -2.76192967e-02 1.87411174e-01 6.47243083e-01 -1.20770168e+00 2.43743554e-01 1.53605235e+00 7.28634298e-01 4.57214803e-01 -1.14210570e+00 -1.13977075e+00 3.97756577e-01 -1.13408692e-01 -1.57894802e+00 -1.05306573e-01 8.15506458e-01 -3.98604512e-01 1.16268182e+00 -2.52899706e-01 6.46835327e-01 7.22957492e-01 -1.03877917e-01 8.15105796e-01 1.01682723e+00 -2.67159253e-01 7.27241337e-02 -1.38768733e-01 -2.02362984e-01 7.35080898e-01 5.23421705e-01 2.69416273e-01 -5.35562873e-01 2.51635343e-01 1.30611432e+00 3.02213818e-01 -1.53323188e-01 -4.90917861e-01 -9.61983621e-01 7.25227535e-01 9.89906728e-01 1.22445561e-02 -1.59819707e-01 3.28508586e-01 -5.57861384e-03 4.68104780e-02 6.22162282e-01 -8.05469137e-03 -4.10807431e-01 2.08560199e-01 -7.58252501e-01 2.66287357e-01 7.14018717e-02 9.79166329e-01 1.17656279e+00 -1.16236240e-01 2.01941252e-01 5.98544419e-01 4.31161374e-01 8.77756894e-01 7.05120489e-02 -9.90269363e-01 5.44000804e-01 1.06194293e+00 9.43053961e-02 -7.16517091e-01 -2.64303386e-01 -4.23690945e-01 -5.66689551e-01 6.96012616e-01 4.71260756e-01 1.34922057e-01 -1.26993763e+00 1.40823650e+00 4.74029541e-01 1.24775946e-01 -1.98093966e-01 9.52378929e-01 9.31908071e-01 4.98023033e-01 -2.33101442e-01 3.15846324e-01 1.15248692e+00 -6.35604262e-01 -8.32746476e-02 -6.89682543e-01 4.14889485e-01 -7.11185992e-01 1.04578722e+00 1.31073564e-01 -1.01202011e+00 -7.22101986e-01 -1.08159781e+00 -3.22756231e-01 -3.48553598e-01 1.77023053e-01 5.76431394e-01 4.44252819e-01 -9.52171445e-01 2.36328170e-01 -8.98102999e-01 -3.01871747e-01 7.83911586e-01 4.13645744e-01 -2.94127762e-01 -4.08930749e-01 -8.22452724e-01 7.65390635e-01 3.02827328e-01 1.95451066e-01 -8.20066750e-01 -7.80283272e-01 -9.67125535e-01 -9.40002352e-02 3.36260498e-01 -6.42981410e-01 1.25183642e+00 -3.39954674e-01 -1.01423085e+00 1.08947110e+00 -1.90042242e-01 -4.62365597e-01 4.54453707e-01 -2.89319843e-01 9.04139206e-02 2.83993721e-01 2.33502284e-01 1.08507812e+00 9.36664939e-01 -1.19440579e+00 -9.11438167e-01 -5.87684214e-01 2.93873876e-01 2.76042104e-01 -1.09754838e-01 -3.81545812e-01 -8.25452805e-01 -2.50883043e-01 5.58106124e-01 -6.97493017e-01 -1.40087798e-01 6.10913038e-01 -2.35638067e-01 -3.10214311e-01 1.00570989e+00 -5.53580336e-02 6.74869597e-01 -2.21807957e+00 -3.17612320e-01 -1.45269722e-01 2.86924899e-01 4.37553197e-01 -1.37333870e-01 2.55882703e-02 2.02354059e-01 1.05492234e-01 -3.17486703e-01 -4.74303961e-01 -2.89596897e-02 1.91195577e-01 -2.24757999e-01 3.81246865e-01 5.04983604e-01 8.93621087e-01 -6.57469511e-01 -4.00665969e-01 5.94838500e-01 7.78107405e-01 -4.78410363e-01 -5.56876324e-02 -3.82838309e-01 2.38237873e-01 -6.26899123e-01 9.52238321e-01 1.21767974e+00 -1.12516269e-01 -3.43503326e-01 -1.85719460e-01 -4.03047144e-01 3.64566207e-01 -1.07377100e+00 1.84410751e+00 -5.08861542e-01 6.02421463e-01 1.60706043e-01 -7.61575341e-01 1.17985535e+00 -1.95674628e-01 3.85449141e-01 -8.49538147e-01 1.27992079e-01 4.12145823e-01 -1.10073164e-01 -1.04942061e-01 1.08479574e-01 -2.02221856e-01 8.02677944e-02 1.56767160e-01 -1.31425232e-01 -5.97148538e-01 -1.61485057e-02 2.31381413e-02 1.03078806e+00 2.20734164e-01 -1.67845618e-02 6.87077716e-02 3.39212865e-01 -1.00932583e-01 5.76962769e-01 7.76093841e-01 -2.61537522e-01 7.66885459e-01 3.48322123e-01 -4.86638546e-01 -8.71572316e-01 -1.21217668e+00 -4.69554961e-01 2.55444974e-01 4.84839499e-01 -1.49021313e-01 -5.03353715e-01 -6.47548318e-01 4.17292356e-01 3.79191071e-01 -4.61136520e-01 -4.48022150e-02 -6.35384619e-01 -3.65714312e-01 4.51321989e-01 7.78681159e-01 9.03165936e-01 -7.50975609e-01 -8.90394747e-01 1.39196247e-01 9.36237574e-02 -1.40934598e+00 -3.20247799e-01 3.14165592e-01 -1.16911912e+00 -9.70290542e-01 -5.90213358e-01 -6.23440504e-01 6.52453601e-01 8.80334914e-01 9.34482515e-01 2.03964245e-02 -4.37178165e-01 5.28713316e-02 -2.61416435e-01 -7.55755901e-01 1.79054543e-01 4.52085771e-02 4.27471995e-02 -3.08060884e-01 7.64443755e-01 -7.85554647e-01 -9.84045327e-01 3.77455622e-01 -9.13114786e-01 4.77052219e-02 9.55778837e-01 5.86764336e-01 8.11237812e-01 3.31659578e-02 1.43170938e-01 -5.95308602e-01 8.98630079e-03 -7.92907458e-03 -9.14373815e-01 -4.16489333e-01 -6.05111480e-01 -1.08830117e-01 2.28344381e-01 -3.72416407e-01 -7.88600385e-01 3.99094880e-01 -2.20974758e-01 -1.00045681e+00 -2.75733441e-01 2.87858788e-02 -1.60451248e-01 -2.38364756e-01 5.29759645e-01 1.54830664e-01 2.58393176e-02 -6.84426546e-01 2.32036561e-01 5.61656237e-01 3.83144647e-01 -2.65773565e-01 1.15496659e+00 1.03696823e+00 3.22279967e-02 -6.39429033e-01 -9.86313879e-01 -5.00233352e-01 -8.32730174e-01 -2.04323120e-02 7.22116828e-01 -1.22948563e+00 -7.12517083e-01 4.24273252e-01 -1.17385852e+00 -4.34354573e-01 -3.34421724e-01 6.25203788e-01 -3.07738453e-01 1.65627077e-01 -4.04777110e-01 -8.63959193e-01 -1.60666689e-01 -1.10853493e+00 1.47568429e+00 2.84954220e-01 3.50512713e-01 -4.99606133e-01 -3.03325504e-01 2.50222385e-01 2.98989415e-01 2.17652559e-01 9.20979857e-01 4.34642360e-02 -1.03651547e+00 -3.22750896e-01 -8.02528918e-01 5.30854106e-01 7.37959845e-03 -1.92577422e-01 -1.14931738e+00 -1.16728745e-01 -1.11938998e-01 -1.56740293e-01 1.19829535e+00 5.54962695e-01 1.02097404e+00 3.51090610e-01 -4.12876427e-01 8.48513007e-01 1.50561821e+00 8.60288441e-02 4.98405069e-01 3.66741419e-01 7.70722806e-01 3.54163408e-01 5.97949266e-01 2.72579134e-01 3.85511279e-01 5.61764419e-01 9.88668740e-01 -2.34014958e-01 -4.89315689e-01 -4.20222700e-01 2.36732379e-01 2.04081640e-01 -9.49176922e-02 1.09129004e-01 -9.53746915e-01 5.21075368e-01 -1.53083301e+00 -6.02512300e-01 -1.07760370e-01 2.12563539e+00 4.41692203e-01 6.68655097e-01 2.38828901e-02 2.79775597e-02 6.23341203e-01 1.34112358e-01 -8.75428259e-01 -1.01373270e-01 -6.24460913e-02 4.53157097e-01 7.50621557e-01 4.13855106e-01 -1.03247809e+00 1.02358103e+00 5.08640766e+00 8.46967399e-01 -1.16522932e+00 4.21245731e-02 1.26139387e-01 -3.17108095e-01 -1.71368644e-01 3.86883281e-02 -1.29925060e+00 3.31284851e-01 3.69679749e-01 2.51659602e-01 2.32631825e-02 9.03776765e-01 1.91129982e-01 -2.71263510e-01 -9.61956739e-01 1.17320216e+00 -1.13688432e-01 -1.22125244e+00 5.01004374e-03 3.24868292e-01 6.16861999e-01 5.14753878e-01 1.05575137e-01 3.14574391e-01 6.31563887e-02 -8.43912482e-01 8.46915722e-01 8.57404396e-02 9.68336344e-01 -6.61917448e-01 5.00440300e-01 5.78410804e-01 -1.49488330e+00 -9.99911204e-02 -7.64379621e-01 -2.16446623e-01 2.71168556e-02 7.87317514e-01 -7.19175637e-01 1.97845057e-01 1.19117093e+00 8.45412612e-01 -4.75982815e-01 1.06817889e+00 -3.80077302e-01 2.21957982e-01 -7.12567508e-01 1.45603761e-01 4.29450154e-01 -3.59358974e-02 5.18792510e-01 1.05185688e+00 4.78696048e-01 5.77099770e-02 2.58321792e-01 1.18092728e+00 -2.72938430e-01 -2.92701781e-01 -7.58732259e-01 1.67440519e-01 5.00252128e-01 1.14994395e+00 -8.45439076e-01 -7.26378709e-02 -5.66657364e-01 7.65549004e-01 2.83859044e-01 2.14996442e-01 -5.86624086e-01 -4.68146592e-01 1.00333261e+00 5.43932617e-01 7.57392108e-01 -5.53515971e-01 -2.62583703e-01 -1.05334365e+00 1.47536814e-01 -3.08444917e-01 5.13124727e-02 -7.59399235e-01 -1.10008717e+00 4.88208830e-01 2.63059977e-02 -1.48355913e+00 1.94502935e-01 -9.79347646e-01 -4.00065899e-01 1.01717937e+00 -2.05084825e+00 -1.11620021e+00 -5.42804003e-01 5.44770300e-01 5.66841006e-01 3.10927600e-01 4.44560260e-01 3.83377075e-01 -2.34558254e-01 3.46390128e-01 -1.15289971e-01 1.44446880e-01 5.62767506e-01 -9.90417063e-01 6.82727396e-01 8.56328666e-01 2.06816066e-02 4.70000476e-01 1.34041682e-01 -5.97634077e-01 -1.35168898e+00 -1.15206909e+00 5.98780155e-01 -5.54388583e-01 2.65004307e-01 -6.90859318e-01 -8.41208518e-01 4.39400822e-01 -3.60173494e-01 4.44816530e-01 1.47436962e-01 -2.48813331e-01 -6.13512576e-01 -2.76197255e-01 -1.11965644e+00 4.00603503e-01 1.47458100e+00 -6.38622344e-01 -2.23244801e-01 1.39687181e-01 8.89897943e-01 -5.92080355e-01 -5.07201672e-01 6.37043357e-01 5.28081715e-01 -1.22941446e+00 1.31631494e+00 1.21582761e-01 2.22418547e-01 -6.58600926e-01 -3.34990352e-01 -7.48992682e-01 -1.86532021e-01 -6.81784600e-02 -1.73629701e-01 7.98023999e-01 1.22863844e-01 -5.73320746e-01 1.07961714e+00 1.11754082e-01 -3.94360542e-01 -7.53664315e-01 -1.10108578e+00 -9.03351605e-01 1.89130008e-01 -8.31016183e-01 5.37961066e-01 5.61042011e-01 -8.06732059e-01 2.94771105e-01 1.46325395e-01 4.92021590e-01 8.18125010e-01 5.37299037e-01 7.75571704e-01 -1.33387220e+00 5.67143224e-02 -5.71442485e-01 -6.72420621e-01 -1.62737978e+00 -1.17123514e-01 -8.75033081e-01 4.24932875e-02 -1.65210843e+00 -7.66260102e-02 -6.42180085e-01 -1.56721860e-01 5.34163713e-01 6.67458400e-02 7.61555493e-01 2.42223710e-01 3.36719662e-01 -1.77916780e-01 5.63701212e-01 1.41791427e+00 -1.00675173e-01 -1.91641197e-01 -3.40105519e-02 -5.35227835e-01 9.96460378e-01 6.83822632e-01 -4.64245647e-01 -3.88828725e-01 -8.40455711e-01 -5.88047057e-02 -2.78235346e-01 7.37619877e-01 -1.08194542e+00 1.45316884e-01 -4.29842398e-02 7.81297684e-01 -1.21603751e+00 6.63398921e-01 -7.42576480e-01 -2.82581538e-01 5.43962538e-01 1.76416934e-01 -1.70389652e-01 3.13995898e-01 6.06336892e-01 -1.39321893e-01 -4.58753221e-02 8.80723417e-01 -3.55068654e-01 -9.22194064e-01 6.93564415e-01 -5.76947583e-03 -2.25607038e-01 7.88226724e-01 -6.88143432e-01 -2.23900691e-01 -3.64537239e-02 -3.32809001e-01 2.11985946e-01 6.63433969e-01 2.72468925e-01 9.64442670e-01 -1.28642070e+00 -4.60027635e-01 5.15689254e-01 2.27954835e-01 8.12710822e-01 2.28778169e-01 7.67096877e-01 -6.45243943e-01 4.30314451e-01 -1.25368759e-01 -1.02681458e+00 -8.77369881e-01 3.34182918e-01 3.98826540e-01 1.62213117e-01 -8.95508111e-01 1.07390392e+00 6.55182779e-01 -4.98823673e-01 3.27541560e-01 -7.42480457e-01 2.40498558e-01 -1.68921560e-01 4.05670285e-01 -7.71483500e-03 2.36284524e-01 -4.86351609e-01 -4.75480855e-01 1.05150676e+00 -1.94986165e-01 9.92797166e-02 1.31426013e+00 -2.12519467e-01 2.05274656e-01 2.10415930e-01 1.11949027e+00 -1.99059620e-01 -1.77169657e+00 -5.73870897e-01 -4.90705132e-01 -8.79117072e-01 3.61629844e-01 -4.05795842e-01 -1.15747833e+00 1.32391918e+00 8.53574455e-01 -1.24485239e-01 1.11673653e+00 2.05873132e-01 8.44899952e-01 3.01346958e-01 4.36706334e-01 -6.77620351e-01 2.48760119e-01 6.53637946e-01 5.83741605e-01 -1.41047704e+00 9.16721150e-02 -5.27297676e-01 -7.92624652e-02 1.12720990e+00 8.71992052e-01 -3.21113527e-01 6.66347146e-01 3.27550530e-01 1.69689283e-01 -3.22111070e-01 -3.60257000e-01 -6.87553406e-01 1.48215994e-01 6.95653021e-01 5.83736114e-02 -1.96475074e-01 2.01432273e-01 1.96344480e-01 6.88961744e-02 -1.42649531e-01 1.31785512e-01 1.03237283e+00 -6.40355408e-01 -1.07299840e+00 -3.89883429e-01 2.71228135e-01 -2.16167383e-02 -6.99199215e-02 -2.38451257e-01 9.92154717e-01 6.02207839e-01 5.92754662e-01 3.73915434e-01 -4.32840824e-01 4.78778183e-01 -2.21268788e-01 6.11862004e-01 -7.91482925e-01 -9.84699801e-02 3.09684157e-01 -4.45400089e-01 -5.78335166e-01 -5.21287680e-01 -5.61269760e-01 -1.53155923e+00 -3.19736391e-01 -3.95356506e-01 -3.84345680e-01 8.77920151e-01 6.02973521e-01 3.65367591e-01 2.07057327e-01 5.06617546e-01 -1.26803172e+00 -3.98635328e-01 -7.62467742e-01 -5.23276091e-01 5.14895990e-02 3.84018391e-01 -9.97555792e-01 -3.21622938e-01 -3.15363586e-01]
[7.954402923583984, -2.720723867416382]
afdec513-f5f4-48c2-b1a2-81aef76a8bc5
updet-universal-multi-agent-rl-via-policy
null
null
https://openreview.net/forum?id=v9c7hr9ADKx
https://openreview.net/pdf?id=v9c7hr9ADKx
UPDeT: Universal Multi-agent RL via Policy Decoupling with Transformers
Recent advances in multi-agent reinforcement learning have been largely limited in training one model from scratch for every new task. The limitation is due to the restricted model architecture related to fixed input and output dimensions. This hinders the experience accumulation and transfer of the learned agent over tasks with diverse levels of difficulty (e.g. 3 vs 3 or 5 vs 6 multi-agent games). In this paper, we make the first attempt to explore a universal multi-agent reinforcement learning pipeline, designing one single architecture to fit tasks with the requirement of different observation and action configurations. Unlike previous RNN-based models, we utilize a transformer-based model to generate a flexible policy by decoupling the policy distribution from the intertwined input observation with an importance weight measured by the merits of the self-attention mechanism. Compared to a standard transformer block, the proposed model, named as Universal Policy Decoupling Transformer (UPDeT), further relaxes the action restriction and makes the multi-agent task's decision process more explainable. UPDeT is general enough to be plugged into any multi-agent reinforcement learning pipeline and equip them with strong generalization abilities that enables the handling of multiple tasks at a time. Extensive experiments on large-scale SMAC multi-agent competitive games demonstrate that the proposed UPDeT-based multi-agent reinforcement learning achieves significant results relative to state-of-the-art approaches, demonstrating advantageous transfer capability in terms of both performance and training speed (10 times faster).
['Xiaodan Liang', 'Xiaojun Chang', 'Fengda Zhu', 'Siyi Hu']
2021-01-01
null
null
null
iclr-2021-1
['smac-1', 'smac']
['playing-games', 'playing-games']
[-1.34242907e-01 1.62419468e-01 -1.54671654e-01 1.44339323e-01 -3.63652349e-01 -4.54277217e-01 6.58258677e-01 6.70209620e-03 -8.85684133e-01 8.56861353e-01 -3.26317586e-02 -2.58946776e-01 -3.48509192e-01 -6.39822066e-01 -6.90965354e-01 -7.65508413e-01 5.74603900e-02 7.88868368e-01 4.61977243e-01 -6.10648036e-01 1.24518849e-01 4.05396372e-02 -1.52934992e+00 1.33277178e-01 1.13427508e+00 6.41987085e-01 8.94311428e-01 5.05435288e-01 -3.24063189e-02 9.21391308e-01 -7.83137619e-01 -2.98039168e-01 3.29434663e-01 -5.20153642e-01 -7.68167973e-01 -3.89246806e-03 -1.46004066e-01 -5.81487417e-01 -1.78856462e-01 7.73360789e-01 6.73628271e-01 4.31979120e-01 4.83930618e-01 -1.44392872e+00 -8.72316718e-01 7.55673110e-01 -6.58358872e-01 1.74056619e-01 -1.88399758e-02 3.83383870e-01 1.08556962e+00 -3.01833719e-01 2.85496682e-01 1.19120395e+00 1.45661294e-01 1.04187179e+00 -1.06039000e+00 -5.06171823e-01 7.21562743e-01 3.29953760e-01 -6.79117799e-01 1.23195000e-01 6.62571073e-01 -1.19435795e-01 1.29934537e+00 -6.09694533e-02 7.97456145e-01 1.31107962e+00 2.62036264e-01 1.05724895e+00 1.25492573e+00 -2.93404579e-01 4.76229519e-01 -3.25911865e-02 -2.21164361e-01 6.10781074e-01 7.86682665e-02 2.56401986e-01 -3.18313390e-01 6.08863905e-02 1.09451222e+00 1.27590775e-01 -6.17757672e-03 -6.18567109e-01 -1.28283954e+00 9.46709991e-01 5.39031982e-01 2.46615887e-01 -6.71626568e-01 3.96511585e-01 5.99874258e-01 6.28360212e-01 2.30854586e-01 7.41082430e-01 -6.46788836e-01 -2.62308240e-01 -3.74202251e-01 4.69319314e-01 5.96946359e-01 7.92755723e-01 6.21147454e-01 4.75944996e-01 -1.37729764e-01 8.73958409e-01 1.77421063e-01 3.98231775e-01 8.85249019e-01 -8.94646585e-01 4.70800787e-01 5.83005846e-01 7.04372451e-02 -3.31673384e-01 -6.83983326e-01 -7.29356110e-01 -6.45973563e-01 7.81413019e-01 2.72572637e-01 -5.61148822e-01 -7.98738956e-01 1.97069132e+00 4.00778115e-01 2.27741361e-01 4.63964045e-01 9.25824344e-01 4.89200443e-01 4.75076318e-01 3.47625464e-01 -1.10960998e-01 1.41691256e+00 -1.47697031e+00 -4.86224532e-01 -6.69099629e-01 3.36574495e-01 -3.20383638e-01 1.14950323e+00 3.05359215e-01 -1.07495975e+00 -6.42074645e-01 -1.08126509e+00 3.16219479e-01 -2.64586121e-01 -2.50341684e-01 7.70532131e-01 2.14200154e-01 -1.02995479e+00 4.06560779e-01 -8.01065862e-01 -8.13926011e-02 1.38634622e-01 6.15729451e-01 -1.44290969e-01 1.73841998e-01 -1.21570277e+00 1.20040596e+00 6.44967973e-01 -1.99102372e-01 -1.05825818e+00 -5.61164200e-01 -7.72210896e-01 3.39304119e-01 7.34517336e-01 -1.03976798e+00 1.56193590e+00 -1.13916266e+00 -2.11538482e+00 9.76452157e-02 3.71004015e-01 -5.58665931e-01 5.58547735e-01 -1.65307239e-01 -8.37315619e-02 4.06067409e-02 -3.85115966e-02 8.38765085e-01 9.77636814e-01 -1.24860346e+00 -1.02337825e+00 -6.09350316e-02 7.32299447e-01 8.23222518e-01 -3.70928675e-01 -1.45278484e-01 -1.87505230e-01 -6.61631167e-01 -5.87822139e-01 -1.03487718e+00 -4.74385649e-01 -6.27966762e-01 3.03345650e-01 -4.27172720e-01 6.14217520e-01 -9.69569534e-02 9.11124647e-01 -1.97688055e+00 6.53229773e-01 -3.40986639e-01 3.49927664e-01 4.10513788e-01 -7.05391884e-01 5.04239917e-01 4.85399179e-02 -2.72831738e-01 1.26132771e-01 -4.92417246e-01 2.88014293e-01 5.13068020e-01 1.92961842e-02 6.29639775e-02 2.27444455e-01 1.07788026e+00 -1.20803618e+00 -1.60432592e-01 3.24723065e-01 1.22044250e-01 -7.67173171e-01 3.86678785e-01 -5.51727295e-01 4.92117941e-01 -6.53854549e-01 1.40931398e-01 4.03533638e-01 -3.23952347e-01 3.10337782e-01 3.96073312e-01 -1.40931169e-02 1.74304977e-01 -1.12539816e+00 1.68037117e+00 -8.12262297e-01 4.20027077e-02 4.04345021e-02 -9.07912850e-01 7.91836917e-01 5.32014191e-01 4.55593139e-01 -1.00090647e+00 5.13934381e-02 1.31769672e-01 6.00028515e-01 -2.31447354e-01 6.30067885e-01 -1.41466722e-01 -1.50065601e-01 7.25809336e-01 4.24966216e-01 -1.31468764e-02 3.78208220e-01 -8.17968026e-02 1.04724157e+00 3.58777225e-01 3.94949526e-01 -1.71172082e-01 3.33568037e-01 -8.22519958e-02 7.01753080e-01 9.17866051e-01 -2.80267447e-01 2.03778580e-01 3.81697327e-01 -4.40655142e-01 -1.06066394e+00 -8.38082194e-01 4.82219696e-01 1.73543119e+00 2.18224287e-01 -2.55806446e-01 -5.50832272e-01 -9.91846085e-01 1.10730939e-01 5.72774649e-01 -6.75067484e-01 -2.78440535e-01 -7.57249892e-01 -8.91521454e-01 1.45826012e-01 6.13556027e-01 5.88965774e-01 -1.75740218e+00 -1.06535411e+00 5.22476256e-01 8.27269852e-02 -9.91793096e-01 -3.86253804e-01 4.78475928e-01 -6.81314111e-01 -8.58719528e-01 -7.56980836e-01 -8.62535775e-01 3.94108981e-01 2.97282964e-01 9.89008248e-01 7.61247724e-02 2.45707840e-01 4.17602390e-01 -5.37898779e-01 -4.45355564e-01 -4.40617085e-01 3.51462513e-01 2.43035153e-01 -2.71775872e-01 7.43266344e-02 -6.64798379e-01 -7.31431425e-01 3.45399112e-01 -9.71618652e-01 2.32377067e-01 9.09365058e-01 1.18961179e+00 1.36299983e-01 -3.17586549e-02 1.11084008e+00 -6.78406835e-01 1.07349813e+00 -5.46178818e-01 -5.58965802e-01 1.62313119e-01 -7.67205119e-01 1.55758038e-01 9.94068265e-01 -8.79780054e-01 -1.07906914e+00 -2.43795946e-01 -5.73102571e-02 -2.28172168e-01 -8.61820355e-02 4.69535291e-01 2.42621943e-01 1.42624751e-01 4.29680496e-01 3.51142406e-01 3.79650712e-01 -1.73781425e-01 3.89598817e-01 3.38623405e-01 1.04869887e-01 -6.71058655e-01 5.48539042e-01 -5.02076410e-02 -2.82802165e-01 -3.06432456e-01 -4.71936584e-01 -1.66947156e-01 -2.86206305e-01 -1.59085274e-01 7.38130093e-01 -1.04362667e+00 -1.10014629e+00 6.47891998e-01 -9.26097929e-01 -9.31219816e-01 -3.50509584e-01 5.59300601e-01 -8.61571431e-01 -7.36254593e-03 -8.10353220e-01 -5.79694211e-01 -3.12042326e-01 -1.43994689e+00 6.57071888e-01 4.97550994e-01 7.04806298e-02 -1.15654743e+00 1.73339352e-01 2.31366485e-01 7.03012228e-01 -2.27360502e-01 9.75040674e-01 -8.37400675e-01 -3.78608167e-01 3.83203030e-01 4.69909012e-02 2.49513000e-01 1.63518742e-01 -4.96991724e-01 -6.16067469e-01 -5.60573578e-01 -8.10657069e-02 -7.36516237e-01 6.76549435e-01 3.09846789e-01 8.14108729e-01 -1.58952713e-01 7.03409538e-02 2.09029719e-01 1.40837681e+00 2.91867286e-01 4.96185094e-01 9.24536169e-01 5.23535490e-01 1.89384863e-01 6.54449582e-01 5.84540129e-01 6.51884735e-01 6.54829979e-01 8.93864751e-01 -2.17773184e-01 -1.21080680e-02 1.43422028e-02 6.37147248e-01 7.51410127e-01 -3.13131213e-01 -3.18462104e-01 -5.68076015e-01 4.03204590e-01 -2.31927872e+00 -9.44411576e-01 5.70620060e-01 1.90756726e+00 7.37929583e-01 2.12409183e-01 4.19289082e-01 -3.62791330e-01 4.45793748e-01 2.91691452e-01 -9.34728742e-01 -5.83575487e-01 1.53553858e-01 2.43838951e-01 2.76693285e-01 4.63226110e-01 -8.96738291e-01 1.22505856e+00 6.17554283e+00 8.73044610e-01 -1.02324140e+00 2.20543399e-01 1.94250017e-01 -2.85614401e-01 -5.61116524e-02 -3.51425618e-01 -7.00669229e-01 2.92103350e-01 7.77264893e-01 -1.92394331e-01 8.66967678e-01 8.08081388e-01 9.91961360e-02 4.23970222e-02 -9.32427168e-01 5.90549409e-01 -2.22698838e-01 -1.06056428e+00 1.34657174e-01 1.00735538e-01 6.78582191e-01 2.92558402e-01 2.50797838e-01 9.83718693e-01 8.90983462e-01 -6.93644166e-01 8.63837302e-01 4.62934598e-02 3.77852201e-01 -8.39677870e-01 7.09286392e-01 6.38756514e-01 -9.76973951e-01 -5.10963142e-01 -4.39030588e-01 -3.95576656e-01 -2.12913398e-02 -3.26420248e-01 -7.90704429e-01 6.92335069e-01 5.17974257e-01 4.11869377e-01 -3.79083842e-01 7.84162462e-01 -2.50911981e-01 2.15166718e-01 5.02310842e-02 -3.12909722e-01 7.22704351e-01 -2.80664057e-01 3.74428779e-01 7.88631678e-01 1.48215383e-01 -1.34126931e-01 5.54406285e-01 5.85888505e-01 6.27054423e-02 2.52254009e-02 -4.37814146e-01 1.89759374e-01 4.75496173e-01 1.41864336e+00 -5.76709330e-01 -1.99639633e-01 -6.31538987e-01 1.06783426e+00 9.88967538e-01 4.10493314e-01 -1.03916919e+00 -4.49255407e-02 8.02260995e-01 -3.32658708e-01 7.30610549e-01 -2.42915228e-01 1.63945198e-01 -1.03637505e+00 -1.57646388e-01 -1.36940408e+00 3.23430598e-01 -4.83723611e-01 -1.36086833e+00 9.59819257e-01 -5.19042239e-02 -1.12834716e+00 -6.89270437e-01 -7.24532008e-01 -6.52245820e-01 7.61021554e-01 -1.71829093e+00 -1.28609419e+00 -7.48052727e-03 6.58696711e-01 8.86794865e-01 -6.17982268e-01 9.85924840e-01 2.91192364e-02 -7.10279822e-01 5.43170214e-01 1.11414909e-01 -2.01244354e-01 6.10966027e-01 -1.66649365e+00 2.95365125e-01 5.25550902e-01 -2.08374247e-01 3.82053196e-01 5.50636470e-01 -4.11925852e-01 -1.32467461e+00 -8.56585979e-01 7.03952760e-02 -1.70132890e-01 8.88656378e-01 -3.16755325e-01 -7.35768020e-01 6.53906643e-01 7.41809011e-01 -2.20298633e-01 5.77591777e-01 3.09130251e-01 -8.90764967e-02 -1.00852214e-01 -8.76227021e-01 8.53029191e-01 8.53531897e-01 -1.24195509e-01 -7.36633480e-01 3.94565091e-02 8.59631002e-01 -5.22006333e-01 -7.70386577e-01 1.80659100e-01 3.66032988e-01 -8.58664393e-01 8.60850036e-01 -8.53357315e-01 5.04198551e-01 -2.33590782e-01 1.76965535e-01 -1.97692668e+00 -8.66915166e-01 -6.22477353e-01 -2.55555451e-01 8.73569429e-01 4.64905590e-01 -9.19578314e-01 5.28528273e-01 3.39171648e-01 -3.84861171e-01 -9.93988335e-01 -8.35420430e-01 -7.77135372e-01 2.44730130e-01 9.18338373e-02 7.39460111e-01 7.28324533e-01 1.84210360e-01 6.66318059e-01 -5.93148649e-01 1.08463973e-01 2.34101459e-01 2.59712994e-01 7.22360492e-01 -1.01684713e+00 -9.28926528e-01 -6.24293447e-01 -7.55386753e-03 -1.28376245e+00 2.82253593e-01 -7.21898437e-01 4.35648952e-03 -1.55862379e+00 2.06242099e-01 -6.47991061e-01 -7.28988290e-01 6.87181354e-01 -4.95750874e-01 -1.33143529e-01 6.30787015e-01 4.31979932e-02 -8.90163541e-01 9.31951225e-01 1.60866833e+00 -1.35099003e-02 -3.88465315e-01 2.24414878e-02 -8.65782082e-01 5.82741916e-01 1.04388905e+00 -3.36407781e-01 -7.62076259e-01 -7.76484907e-01 7.60783330e-02 -3.25266086e-03 1.75023213e-01 -9.96214807e-01 2.04224303e-01 -3.99850339e-01 1.68398604e-01 1.12585470e-01 2.89652497e-01 -8.75903070e-01 -1.74416497e-01 5.84617674e-01 -2.16201887e-01 5.51052213e-01 3.76529455e-01 5.98110795e-01 -5.09704165e-02 -3.59618425e-01 6.24306738e-01 -3.55367184e-01 -8.99201572e-01 2.68265396e-01 -5.41455090e-01 -2.75077559e-02 1.28941441e+00 8.09099749e-02 -6.71532273e-01 -2.32922882e-01 -6.33858562e-01 6.13642812e-01 2.08076656e-01 7.00674415e-01 5.16386509e-01 -1.22581351e+00 -7.67414033e-01 6.04838319e-02 -3.93378325e-02 -8.88170972e-02 4.35541660e-01 5.70302904e-01 -1.07935697e-01 2.16597140e-01 -8.24545383e-01 -3.14715952e-01 -9.88222957e-01 7.90915787e-01 3.87710780e-01 -9.51554418e-01 -7.19163954e-01 3.50961030e-01 3.88625234e-01 -5.44214129e-01 8.00887942e-02 -2.07579643e-01 -4.07506436e-01 -5.48292175e-02 5.09064615e-01 1.20467708e-01 -1.37092978e-01 -7.46608227e-02 -3.18074115e-02 1.92352578e-01 -5.49620509e-01 -1.84244886e-01 1.61736631e+00 -2.03204453e-02 2.28008673e-01 3.06600362e-01 4.63909596e-01 -3.56235981e-01 -1.76941574e+00 -2.34147191e-01 -2.48500094e-01 -1.02708757e-01 -8.20671394e-02 -1.04617393e+00 -9.95120645e-01 4.51458722e-01 3.29222649e-01 3.50657940e-01 1.02502346e+00 -5.09365946e-02 5.55035472e-01 4.41956997e-01 6.12005651e-01 -1.21543872e+00 5.41945875e-01 8.53073776e-01 9.81934130e-01 -1.27593327e+00 -2.34231830e-01 1.58219844e-01 -1.14892423e+00 8.78010333e-01 1.18388617e+00 -3.36226046e-01 1.77558213e-01 1.88052356e-01 1.26660481e-01 -1.02156624e-01 -1.15139532e+00 -4.73134696e-01 -1.22051220e-03 7.27853000e-01 1.55721322e-01 9.23257172e-02 -2.09647462e-01 5.47858715e-01 -2.09260229e-02 -1.26884073e-01 5.07239044e-01 9.21781898e-01 -5.03452420e-01 -1.37722278e+00 3.96223739e-02 2.88852185e-01 -2.22949281e-01 -1.61004364e-02 1.84584811e-01 1.11436033e+00 3.12198866e-02 8.34088504e-01 8.04613158e-02 -2.85089135e-01 3.16486686e-01 -1.05460413e-01 7.58085787e-01 -7.07809150e-01 -9.83222485e-01 8.82021636e-02 -1.29466385e-01 -4.09342289e-01 -3.81325454e-01 -3.60317439e-01 -1.38302886e+00 -1.58375442e-01 -1.88981831e-01 2.01182440e-01 3.67811084e-01 1.00840914e+00 5.38482845e-01 1.09748685e+00 5.57010829e-01 -8.99939418e-01 -1.03285229e+00 -1.14997756e+00 -6.63604021e-01 3.60386580e-01 3.72707665e-01 -9.53937948e-01 9.36839357e-02 -4.09419298e-01]
[3.855250358581543, 1.857831358909607]
d2d1fe6f-6aaf-49cd-afb5-3a344ad660fc
te-yolof-tiny-and-efficient-yolof-for-blood
2108.12313
null
https://arxiv.org/abs/2108.12313v1
https://arxiv.org/pdf/2108.12313v1.pdf
TE-YOLOF: Tiny and efficient YOLOF for blood cell detection
Blood cell detection in microscopic images is an essential branch of medical image processing research. Since disease detection based on manual checking of blood cells is time-consuming and full of errors, testing of blood cells using object detectors with Deep Convolutional Neural Network can be regarded as a feasible solution. In this work, an object detector based on YOLOF has been proposed to detect blood cell objects such as red blood cells, white blood cells and platelets. This object detector is called TE-YOLOF, Tiny and Efficient YOLOF, and it is a One-Stage detector using dilated encoder to extract information from single-level feature maps. For increasing efficiency and flexibility, the EfficientNet Convolutional Neural Network is utilized as the backbone for the proposed object detector. Furthermore, the Depthwise Separable Convolution is applied to enhance the performance and minimize the parameters of the network. In addition, the Mish activation function is employed to increase the precision. Extensive experiments on the BCCD dataset prove the effectiveness of the proposed model, which is more efficient than other existing studies for blood cell detection.
['Wei Xiang', 'Yali Wang', 'Hang Yang', 'Xiangkui Li', 'Fanxin Xu']
2021-08-27
null
null
null
null
['cell-detection', 'blood-cell-detection']
['computer-vision', 'medical']
[-3.60295385e-01 -3.38563353e-01 6.67543262e-02 -7.24265799e-02 2.29182959e-01 7.47607276e-02 8.19996893e-02 2.62456626e-01 -7.81082869e-01 5.02473891e-01 -1.48362190e-01 -3.87757085e-02 2.84167945e-01 -1.01530659e+00 -1.26945123e-01 -9.78206456e-01 1.50794417e-01 9.13021937e-02 5.45365870e-01 1.83648184e-01 2.08167002e-01 9.38409865e-01 -1.05858207e+00 1.89048678e-01 6.72015309e-01 1.24856114e+00 4.86040115e-02 6.94325268e-01 -3.84643346e-01 1.11568916e+00 -6.96145594e-01 -3.05305690e-01 8.95717293e-02 -3.75469416e-01 -2.77978778e-01 2.57727295e-01 -2.91784614e-01 -8.10893238e-01 -4.09655124e-01 1.05738044e+00 8.88087094e-01 -3.41039926e-01 6.37668014e-01 -8.69457781e-01 -4.68517721e-01 1.78796381e-01 -6.48505688e-01 8.07241857e-01 -7.48663917e-02 3.07435423e-01 8.47153738e-02 -7.55743504e-01 2.08949327e-01 1.40809095e+00 5.11658430e-01 3.39622349e-01 -6.90555871e-01 -6.36952460e-01 -4.48618233e-01 2.32757926e-01 -1.54418802e+00 -3.50529104e-01 4.29394722e-01 -3.12846392e-01 8.13634813e-01 9.79789495e-02 7.26186454e-01 2.26727068e-01 4.64096755e-01 8.88639569e-01 7.91703105e-01 -4.10136759e-01 4.32381630e-02 3.60232532e-01 2.62305081e-01 1.17021930e+00 9.18428421e-01 -5.00274226e-02 -8.31378177e-02 1.31227970e-01 1.20653784e+00 5.66864312e-01 -2.54945248e-01 1.46886364e-01 -9.08465385e-01 6.90328836e-01 7.20451236e-01 4.72197354e-01 -1.02704786e-01 1.41744286e-01 5.30729115e-01 -3.95027250e-02 1.50092408e-01 1.51748240e-01 -2.93887611e-02 1.92536473e-01 -5.92833698e-01 -1.04851231e-01 5.63018322e-01 5.50563753e-01 3.85440171e-01 1.82777241e-01 -4.09588546e-01 7.37470984e-01 3.22402090e-01 5.34124017e-01 5.62332034e-01 -4.58998382e-01 -4.13713716e-02 1.32308960e+00 -6.63601160e-02 -9.57760811e-01 -6.68775201e-01 -3.63379776e-01 -1.15303433e+00 2.33022988e-01 4.44950491e-01 -1.78395405e-01 -9.58634257e-01 1.02052903e+00 5.57100534e-01 2.55758822e-01 4.36024219e-02 1.24302804e+00 1.19816065e+00 6.95298553e-01 1.31028101e-01 -1.52968198e-01 1.84315765e+00 -8.06717992e-01 -7.28466630e-01 4.02232289e-01 7.11116910e-01 -6.37409091e-01 5.45894861e-01 1.25105932e-01 -7.04783916e-01 -6.93803012e-01 -1.10917723e+00 -4.33314979e-01 -4.96201873e-01 5.97608805e-01 8.72135103e-01 7.19871879e-01 -5.82519650e-01 1.95187271e-01 -8.79722476e-01 -3.07089061e-01 9.23049092e-01 5.65843582e-01 -2.11852089e-01 5.07653058e-02 -9.71707642e-01 6.51617169e-01 6.12699568e-01 3.99380952e-01 -5.44202805e-01 -4.61222768e-01 -6.80440187e-01 2.91472822e-01 -7.99179729e-03 -4.65318263e-01 7.49078333e-01 -4.81021583e-01 -1.33715618e+00 7.82326877e-01 9.69523415e-02 -4.03044134e-01 5.14004409e-01 1.97764020e-02 -3.57138246e-01 6.44901454e-01 -6.14820123e-02 6.46737516e-01 4.49321985e-01 -3.79393101e-01 -7.52092779e-01 -2.11325780e-01 -1.61056891e-01 -4.29837331e-02 -4.04000968e-01 2.04046190e-01 -4.67960238e-01 -5.70836067e-01 -1.69340163e-01 -4.62176532e-01 -1.34038597e-01 2.85928696e-01 -3.58680040e-01 -5.29795527e-01 6.78158581e-01 -6.08984053e-01 1.17789769e+00 -2.26815796e+00 -5.42395771e-01 2.05556169e-01 3.96068245e-01 5.29930413e-01 1.77519098e-01 -1.29725114e-01 3.09928507e-01 -4.40810248e-02 3.98216862e-03 1.29479393e-01 -3.45887065e-01 -1.73222885e-01 1.39503092e-01 6.36726141e-01 5.82439005e-01 1.01837063e+00 -5.16645193e-01 -1.12648213e+00 4.83440250e-01 7.59783447e-01 -2.28285581e-01 2.02413037e-01 2.91773051e-01 3.16464990e-01 -5.18528044e-01 1.02088904e+00 9.10945952e-01 -6.31618381e-01 -3.09201896e-01 -3.21446568e-01 -1.07090101e-01 -3.56833935e-01 -1.08226430e+00 9.45486009e-01 -4.65879180e-02 3.74787360e-01 -8.14101696e-02 -7.72064447e-01 1.11892796e+00 2.84611315e-01 3.62083793e-01 -8.48992586e-01 8.49024117e-01 2.87565351e-01 2.91194975e-01 -9.66359913e-01 -3.52257900e-02 -1.40617639e-01 3.67785722e-01 8.91887993e-02 -1.80647716e-01 4.59589481e-01 4.84432250e-01 -1.06689736e-01 9.80830610e-01 -2.81941324e-01 4.43230003e-01 -3.92889142e-01 1.01219928e+00 -8.33909735e-02 7.79403389e-01 3.11191082e-01 -3.74323159e-01 1.92722782e-01 5.16885638e-01 -8.92812967e-01 -8.22655201e-01 -9.18964565e-01 -6.12619460e-01 4.61607039e-01 4.29971218e-01 2.74350375e-01 -7.68669188e-01 -4.87087995e-01 2.02899501e-01 -7.43524954e-02 -5.09387732e-01 -8.43015835e-02 -8.09461594e-01 -1.07496190e+00 5.72320759e-01 6.27566040e-01 1.15978909e+00 -1.38273776e+00 -1.14696193e+00 2.29773864e-01 3.01540405e-01 -9.74493623e-01 -3.21140766e-01 -1.00770950e-01 -8.33478332e-01 -1.40512013e+00 -8.61563444e-01 -1.06950915e+00 9.43261504e-01 2.20151246e-01 5.24721146e-01 8.05758536e-01 -9.39798355e-01 -4.01643544e-01 -2.58138806e-01 -5.53659022e-01 -5.64125367e-02 -1.93151385e-01 -1.43677279e-01 2.77681112e-01 9.56249177e-01 6.38816133e-02 -1.09541535e+00 -2.85725831e-03 -9.59059834e-01 -5.29278107e-02 1.00816369e+00 9.11085725e-01 4.88785088e-01 8.88043717e-02 7.04930544e-01 -1.05228436e+00 4.01951164e-01 -1.69948831e-01 -6.69577420e-01 1.93004802e-01 -2.85210490e-01 -1.11167371e-01 9.86588955e-01 -3.14431012e-01 -9.06089842e-01 -5.78460209e-02 -6.70141578e-02 -2.27644950e-01 -5.18834554e-02 1.17092058e-01 6.68225214e-02 -2.39904985e-01 2.84647405e-01 4.27066863e-01 6.05919026e-02 -3.94307137e-01 -3.19188625e-01 9.37031388e-01 4.74081159e-01 7.70188123e-02 1.78824052e-01 7.33027220e-01 3.24865669e-01 -9.11768258e-01 -4.59563911e-01 -4.42629009e-01 -3.12612653e-01 -8.75756294e-02 1.00524557e+00 -9.14104283e-01 -1.26798344e+00 7.56824970e-01 -1.25318241e+00 7.76981637e-02 1.09580465e-01 8.29000890e-01 1.20691545e-01 3.93516004e-01 -1.26919162e+00 -7.79311419e-01 -8.40391755e-01 -1.18292546e+00 8.21837068e-01 9.00151908e-01 2.19848022e-01 -9.55226302e-01 -4.27916914e-01 -7.71172568e-02 4.06315446e-01 3.66475135e-01 8.72877240e-01 -4.89199758e-01 -6.51348233e-01 -4.43577677e-01 -7.08298445e-01 2.86472738e-01 2.15053171e-01 9.93954465e-02 -5.62309921e-01 -3.13319832e-01 2.01246753e-01 -1.69670209e-01 1.03655088e+00 5.99911034e-01 1.23658192e+00 -1.99212328e-01 -5.03615737e-01 9.16219950e-01 1.52882957e+00 5.54754376e-01 7.97907889e-01 3.31515014e-01 5.28430343e-01 3.54511648e-01 3.29641551e-01 4.86130089e-01 1.43175930e-01 -3.71067552e-03 -1.89138781e-02 -6.33501768e-01 -2.34220922e-01 2.96262264e-01 -9.37537327e-02 7.00526595e-01 6.97283596e-02 -1.59996778e-01 -6.33468270e-01 3.87479573e-01 -1.38843966e+00 -7.27834821e-01 -2.18640015e-01 1.69618833e+00 6.19192839e-01 8.80307108e-02 -7.14126043e-04 1.38922319e-01 9.81308281e-01 -4.56516325e-01 -7.63598382e-01 -1.54999942e-01 -1.01735108e-01 4.68361586e-01 4.29717332e-01 1.24330215e-01 -1.15729654e+00 6.11709714e-01 5.67185354e+00 8.71587336e-01 -1.44791186e+00 -7.40357488e-02 1.00226414e+00 -5.76259941e-02 4.59955961e-01 -4.18866962e-01 -1.06757343e+00 1.02869666e+00 2.44849488e-01 2.25967854e-01 -7.52836466e-02 6.85585678e-01 2.51150101e-01 -4.41153795e-01 -9.08000171e-01 1.19166100e+00 -5.75607903e-02 -1.33360159e+00 3.09946202e-02 -9.69498158e-02 2.03580290e-01 -5.69633901e-01 -2.72808224e-01 6.39584586e-02 -1.76850632e-01 -9.86832976e-01 1.03083491e-01 5.90940297e-01 9.41348076e-01 -7.68951714e-01 1.44093919e+00 9.45033208e-02 -1.22910249e+00 -6.93903342e-02 -8.79655778e-01 -7.84842223e-02 1.95448417e-02 7.23823488e-01 -7.91803002e-01 -6.80085048e-02 7.29933560e-01 6.26908481e-01 -7.03726709e-01 1.39052403e+00 9.07107741e-02 3.07005763e-01 -4.12181020e-01 -5.25613368e-01 -4.94411737e-02 -1.52902409e-01 -7.38930553e-02 1.48026359e+00 3.77124906e-01 2.61902243e-01 1.15717553e-01 6.80562556e-01 -2.62335002e-01 1.69644549e-01 -2.58170098e-01 2.11781546e-01 4.71853673e-01 1.54826164e+00 -1.18238294e+00 -7.24307179e-01 -3.02892298e-01 5.44321656e-01 2.07138896e-01 2.04766747e-02 -7.66346991e-01 -1.03145754e+00 1.46005645e-01 4.92910855e-02 3.21530491e-01 2.49340042e-01 -3.37510079e-01 -9.58426476e-01 -2.45470345e-01 -2.89658695e-01 4.89667058e-01 -3.45786542e-01 -1.05227864e+00 4.14053828e-01 -4.77617294e-01 -1.20940626e+00 3.99024427e-01 -9.20417011e-01 -7.35638082e-01 7.97240913e-01 -1.78686869e+00 -8.29377294e-01 -6.72334909e-01 4.55419689e-01 1.74151763e-01 -3.27851772e-01 4.25770789e-01 4.27307904e-01 -1.08400607e+00 6.60374284e-01 -3.90267558e-02 6.42993271e-01 3.58002931e-01 -8.50505531e-01 -2.04900041e-01 8.36233854e-01 -5.43579161e-01 7.14131057e-01 2.02188835e-01 -5.31963587e-01 -1.17420757e+00 -1.04052830e+00 5.86999178e-01 2.64172077e-01 1.08280689e-01 -2.38547996e-01 -9.26536143e-01 2.30288386e-01 8.87077581e-03 4.87213075e-01 6.93110228e-01 -8.12492669e-01 1.52599484e-01 -4.94591147e-01 -1.40453291e+00 4.34720755e-01 3.85222971e-01 8.25149938e-02 -3.20060164e-01 3.56783211e-01 5.04016697e-01 -4.81811732e-01 -7.94977069e-01 4.18453753e-01 6.00150108e-01 -1.02750671e+00 9.23373640e-01 -2.05693647e-01 2.07896292e-01 -5.44863284e-01 3.98740530e-01 -6.10980332e-01 -3.73666555e-01 -9.48428363e-02 -2.16355801e-01 1.11218345e+00 -3.25330794e-01 -8.68359864e-01 1.08339131e+00 4.27195756e-03 -9.85577628e-02 -9.87310171e-01 -9.32381034e-01 -6.84793234e-01 -1.58958137e-01 5.06994903e-01 7.20520318e-01 5.60080528e-01 -7.82445669e-02 3.65753472e-01 -8.08331892e-02 -8.30071345e-02 5.38869798e-01 7.32515529e-02 6.87749982e-01 -1.10444653e+00 -8.46784264e-02 -5.41508377e-01 -6.21125817e-01 -1.04468739e+00 -3.54122132e-01 -6.83505356e-01 -1.35125697e-01 -1.48516023e+00 2.76012182e-01 -5.10031998e-01 -5.81302524e-01 2.25533009e-01 -3.85014087e-01 4.24974412e-01 -1.09033816e-01 2.29405165e-01 -5.14882684e-01 2.15276837e-01 1.51260328e+00 -1.40165955e-01 -7.90949240e-02 -3.30055386e-01 -4.71011937e-01 5.63648641e-01 8.95193636e-01 -3.50273788e-01 1.51363388e-02 -2.95842290e-01 -2.91260123e-01 6.03891797e-02 2.46332362e-01 -1.08794737e+00 4.90762085e-01 3.63987796e-02 1.06022406e+00 -7.18439281e-01 2.66052753e-01 -6.00881338e-01 -1.07998900e-01 1.41865492e+00 3.05448746e-04 -9.82696488e-02 7.22299665e-02 2.26438344e-01 -3.32474917e-01 -3.68642718e-01 1.16524124e+00 -4.24239725e-01 -6.35793805e-01 4.29251134e-01 -3.39752883e-01 -2.97626257e-01 1.31302738e+00 -5.36867976e-01 -6.97765291e-01 4.67603624e-01 -4.36731018e-02 4.92530227e-01 1.97797582e-01 -1.96112737e-01 9.10591424e-01 -1.26294863e+00 -7.20714808e-01 3.25593859e-01 8.32643881e-02 1.88294396e-01 4.37364668e-01 9.98695433e-01 -1.59641457e+00 5.88869572e-01 -3.85822833e-01 -5.70007920e-01 -1.19542813e+00 6.39609814e-01 4.61457074e-01 -3.39057744e-01 -7.97948420e-01 8.67110431e-01 2.83762217e-01 2.25509703e-01 -1.43586118e-02 -3.72938067e-01 -5.47678649e-01 -2.68666118e-01 9.10367906e-01 5.89244723e-01 -3.01010072e-01 -3.62667918e-01 -5.41571438e-01 6.39586270e-01 -2.03278348e-01 7.05077410e-01 1.05636859e+00 6.58762231e-02 -4.85535651e-01 5.06290905e-02 1.21152246e+00 -1.81019321e-01 -1.08609617e+00 -7.96253532e-02 -2.15102151e-01 -4.27552700e-01 -2.40280014e-02 -4.30865109e-01 -1.27491641e+00 1.25869393e+00 8.08073401e-01 1.45779654e-01 1.04236841e+00 -2.59006441e-01 1.15992045e+00 4.17777508e-01 5.65446503e-02 -9.84247923e-01 2.31987610e-01 1.02645285e-01 3.18886936e-01 -1.29858625e+00 3.06306183e-01 -5.35403192e-01 -1.42305270e-01 1.54473174e+00 1.14469755e+00 -4.45322543e-01 6.69676185e-01 4.35240120e-01 8.11245888e-02 -2.34566405e-01 -3.49612027e-01 -2.83674896e-01 -2.42370963e-01 3.84332150e-01 4.92540419e-01 -7.00320378e-02 -5.91176391e-01 4.08347160e-01 1.60870120e-01 4.98742640e-01 4.12840843e-01 7.98251569e-01 -8.05021763e-01 -4.61706251e-01 -4.52277988e-01 8.39179039e-01 -8.25205564e-01 3.22384834e-02 1.51779026e-01 7.25221813e-01 3.93330961e-01 8.13133597e-01 4.42991018e-01 -8.04222301e-02 6.82318732e-02 -4.12807524e-01 3.64672393e-01 -4.32094455e-01 -5.03319263e-01 1.75570518e-01 -6.09684050e-01 -8.80380943e-02 -2.39626259e-01 -1.36866555e-01 -1.73236966e+00 -3.79054189e-01 -4.23684269e-01 -2.41173655e-02 1.70930862e-01 8.22624564e-01 1.21489167e-01 6.35422349e-01 5.93938410e-01 -1.90905526e-01 -2.32400879e-01 -1.00432837e+00 -7.51309216e-01 4.42994565e-01 3.42463493e-01 -7.23278999e-01 -1.07972614e-01 6.53144196e-02]
[14.803963661193848, -3.0896670818328857]
cee113b7-dc02-438f-9197-c38b377b7342
unleash-the-potential-of-3d-point-cloud
2306.00552
null
https://arxiv.org/abs/2306.00552v1
https://arxiv.org/pdf/2306.00552v1.pdf
Unleash the Potential of 3D Point Cloud Modeling with A Calibrated Local Geometry-driven Distance Metric
Quantifying the dissimilarity between two unstructured 3D point clouds is a challenging task, with existing metrics often relying on measuring the distance between corresponding points that can be either inefficient or ineffective. In this paper, we propose a novel distance metric called Calibrated Local Geometry Distance (CLGD), which computes the difference between the underlying 3D surfaces calibrated and induced by a set of reference points. By associating each reference point with two given point clouds through computing its directional distances to them, the difference in directional distances of an identical reference point characterizes the geometric difference between a typical local region of the two point clouds. Finally, CLGD is obtained by averaging the directional distance differences of all reference points. We evaluate CLGD on various optimization and unsupervised learning-based tasks, including shape reconstruction, rigid registration, scene flow estimation, and feature representation. Extensive experiments show that CLGD achieves significantly higher accuracy under all tasks in a memory and computationally efficient manner, compared with existing metrics. As a generic metric, CLGD has the potential to advance 3D point cloud modeling. The source code is publicly available at https://github.com/rsy6318/CLGD.
['Junhui Hou', 'Siyu Ren']
2023-06-01
null
null
null
null
['scene-flow-estimation']
['computer-vision']
[-1.52824089e-01 -5.52357197e-01 -3.67733352e-02 -1.95777193e-01 -8.31810653e-01 -5.93693018e-01 6.91360474e-01 4.95929569e-01 -1.74249977e-01 2.09231615e-01 2.47220173e-02 -5.24173826e-02 -1.26634285e-01 -8.35723758e-01 -4.76385236e-01 -5.05876422e-01 1.45202219e-01 6.22114837e-01 1.40979737e-01 7.48032853e-02 5.67678034e-01 1.05535483e+00 -1.42298865e+00 -2.99088687e-01 8.94818425e-01 1.00787282e+00 2.00499550e-01 4.46992397e-01 -2.48263016e-01 -2.66948268e-02 -2.37560198e-01 -4.87760454e-03 4.60647911e-01 -6.17693141e-02 -7.24317491e-01 -7.07703531e-02 6.19779527e-01 -3.69301662e-02 -1.46928743e-01 1.00668347e+00 4.03633028e-01 2.61449546e-01 6.26841903e-01 -1.32717741e+00 -4.64034915e-01 -1.89747974e-01 -5.92199564e-01 9.61646512e-02 5.66674709e-01 9.75592900e-03 8.86630177e-01 -1.27266443e+00 5.98976135e-01 1.12597477e+00 7.26157367e-01 2.16817915e-01 -1.36385298e+00 -5.07195532e-01 -4.28470969e-02 -1.95476785e-01 -1.80168509e+00 -3.93664330e-01 1.04817545e+00 -7.54000247e-01 6.83762014e-01 2.81132847e-01 4.52909708e-01 3.24645013e-01 3.26329377e-03 2.53157079e-01 7.93649375e-01 -2.24890009e-01 1.76520914e-01 -3.10620248e-01 -1.28796985e-02 6.55909956e-01 3.11962098e-01 6.78324476e-02 -3.20108294e-01 -3.69407713e-01 9.42210674e-01 2.66716331e-01 -4.46026176e-01 -9.25332785e-01 -1.43076634e+00 6.49145484e-01 5.03432512e-01 3.04389894e-01 -2.89125949e-01 -3.96047495e-02 3.60046290e-02 5.69624752e-02 6.88544989e-01 1.69563681e-01 -1.85060844e-01 -1.45661458e-01 -5.88202417e-01 2.45004579e-01 6.05747700e-01 1.00490367e+00 1.21007776e+00 -2.94050068e-01 2.48597160e-01 7.37338185e-01 4.80778635e-01 6.56905472e-01 7.87401721e-02 -1.17375004e+00 4.53130960e-01 9.24727678e-01 1.67866811e-01 -1.70516419e+00 -2.64266491e-01 -8.73157457e-02 -7.70662487e-01 2.97183275e-01 3.90197098e-01 3.05003375e-01 -3.21156919e-01 1.39933431e+00 7.45634556e-01 7.11311996e-01 -2.11955816e-01 8.97377253e-01 8.63205075e-01 5.88771105e-01 -2.73761660e-01 -4.01522294e-02 6.92060709e-01 -6.00603521e-01 -1.65009230e-01 8.27978924e-02 7.04082966e-01 -1.00680149e+00 1.03467321e+00 -1.31275252e-01 -1.19386029e+00 -4.06682611e-01 -8.45984757e-01 -1.87258646e-01 -2.44423643e-01 -2.04302520e-01 1.95170671e-01 3.01815480e-01 -1.04832935e+00 7.32549906e-01 -9.59279180e-01 -2.78092563e-01 3.05761367e-01 2.05308527e-01 -3.06546479e-01 -5.98274544e-02 -4.78798807e-01 6.20819211e-01 -2.30279192e-01 -1.01595461e-01 -3.57864439e-01 -1.08563221e+00 -9.25392270e-01 -3.07680428e-01 6.40912578e-02 -7.48917699e-01 8.66546392e-01 -2.34098375e-01 -1.26525354e+00 1.25246525e+00 -4.60952491e-01 4.30192612e-02 5.19860625e-01 -8.79632309e-03 -2.39596799e-01 -8.68823305e-02 2.82852948e-01 3.56140077e-01 5.92608571e-01 -1.36500311e+00 -4.06814694e-01 -6.12544060e-01 -3.19011241e-01 3.82829756e-01 1.87743917e-01 -1.89439386e-01 -5.94980538e-01 -3.94902319e-01 6.69521391e-01 -1.00330651e+00 -1.79592967e-01 4.42876726e-01 -3.64617825e-01 -3.85733515e-01 1.03471494e+00 -2.04967350e-01 9.67191935e-01 -2.29246473e+00 1.12248600e-01 4.68322814e-01 3.69291753e-01 1.29286632e-01 -6.18639328e-02 2.44769111e-01 5.49912453e-02 8.02905485e-02 -6.03954911e-01 -3.36630076e-01 -1.08345859e-01 5.74049316e-02 -1.45213723e-01 9.16131318e-01 -3.66224758e-02 7.96418726e-01 -1.16681302e+00 -3.15968305e-01 5.79227269e-01 7.46334016e-01 -3.61648053e-01 1.32962123e-01 2.06005543e-01 8.53456378e-01 -6.14359856e-01 6.15017116e-01 9.67141449e-01 -2.28181988e-01 -3.32323164e-01 -2.82357484e-01 -3.24708313e-01 2.25314841e-01 -1.37191463e+00 2.01248360e+00 -4.87217814e-01 5.06688833e-01 -1.32834807e-01 -7.24101543e-01 1.26488769e+00 1.83171462e-02 8.60138059e-01 -4.51061159e-01 7.89368674e-02 4.22757655e-01 -5.73701024e-01 -9.89595950e-02 4.87348467e-01 1.29049048e-01 1.06589291e-02 4.62999284e-01 -5.63713729e-01 -5.71989655e-01 -2.25526214e-01 -1.08553842e-02 8.65843952e-01 3.16992253e-02 3.01288217e-01 -3.74028742e-01 7.88195074e-01 -1.27788112e-01 4.54142630e-01 3.80030870e-01 -2.29963496e-01 8.04494023e-01 5.17800711e-02 -4.57586378e-01 -1.06632543e+00 -1.41497409e+00 -4.36940312e-01 2.29562566e-01 8.06165397e-01 -5.55043101e-01 -5.19219160e-01 -3.50257933e-01 5.13491929e-01 4.42653447e-01 -2.97662824e-01 -3.11036594e-02 -7.35144138e-01 -8.95269811e-02 1.95002019e-01 3.72215867e-01 3.71671706e-01 -4.54153329e-01 -5.69733202e-01 -1.13760479e-01 -1.46203533e-01 -1.14626372e+00 -9.32825506e-01 -5.41780531e-01 -1.28587544e+00 -1.12477601e+00 -5.97658813e-01 -6.94826543e-01 8.75294983e-01 8.30283463e-01 1.18905520e+00 2.24348202e-01 -1.15865543e-01 5.52194893e-01 -7.35831484e-02 -6.43748231e-03 -1.31444499e-01 -1.41039148e-01 1.33018836e-01 7.62991831e-02 4.43049669e-01 -6.23297989e-01 -7.05177307e-01 6.64654732e-01 -5.95717609e-01 -6.01683445e-02 -4.95927185e-02 2.67604053e-01 1.21851814e+00 -4.24911559e-01 -1.67161882e-01 -4.14965928e-01 4.04574305e-01 -5.03650129e-01 -7.88657248e-01 -5.38064353e-02 -3.18575829e-01 5.68966791e-02 2.26545095e-01 -2.47075967e-02 -5.16192734e-01 2.77084485e-02 1.65476769e-01 -6.92188084e-01 -2.27071911e-01 2.75726080e-01 -1.85709774e-01 -4.04164791e-01 4.48823124e-01 7.73397386e-02 9.58465859e-02 -6.94151759e-01 3.48771811e-01 4.57196832e-01 4.59882647e-01 -7.13468373e-01 1.03426373e+00 9.22710121e-01 3.15984756e-01 -8.85681927e-01 -6.14050388e-01 -8.18023443e-01 -1.07897770e+00 -1.41243577e-01 7.04656184e-01 -6.49388373e-01 -7.72870660e-01 3.48676056e-01 -1.17932296e+00 -1.99528083e-01 -2.88157225e-01 5.32090724e-01 -6.17009580e-01 5.87639511e-01 -3.46606933e-02 -4.79046106e-01 -2.06142426e-01 -1.20848322e+00 1.40107334e+00 3.30927782e-02 -2.34849796e-01 -1.30308664e+00 5.67981064e-01 1.54661104e-01 2.22113162e-01 7.51426697e-01 7.60554314e-01 -6.87224790e-02 -7.53605187e-01 -3.40840191e-01 -1.99691638e-01 3.35216485e-02 5.66648841e-01 2.35429645e-01 -5.55214107e-01 -3.05507004e-01 -2.73751412e-02 3.38917911e-01 3.37231010e-01 4.34580326e-01 1.21938491e+00 1.90844536e-02 -4.81676966e-01 1.07351196e+00 1.56692326e+00 9.42837372e-02 4.11809623e-01 2.75626808e-01 9.53501821e-01 2.99639285e-01 6.76934719e-01 4.61504281e-01 4.45276707e-01 8.71162832e-01 4.70401764e-01 6.96129575e-02 -5.99371456e-02 -3.40534747e-01 -7.24220052e-02 1.16944623e+00 -3.65927815e-01 1.68267503e-01 -1.18981314e+00 5.54122388e-01 -1.72574818e+00 -7.48164892e-01 -4.01928842e-01 2.76339197e+00 3.58824223e-01 -2.43701160e-01 1.64776556e-02 -7.00172223e-03 9.66570079e-01 2.21230492e-01 -6.17586672e-01 -2.00319186e-01 -8.20717663e-02 1.41763225e-01 4.70609039e-01 7.64332294e-01 -9.86474872e-01 7.30176866e-01 5.49867678e+00 4.76062894e-01 -1.15649652e+00 2.74582189e-02 2.70502925e-01 -2.49841735e-02 -3.23535204e-01 4.31206264e-02 -6.21055841e-01 5.60820878e-01 6.80709064e-01 -5.36267698e-01 2.74165124e-01 7.70452857e-01 2.81410426e-01 1.42323300e-01 -1.03802359e+00 1.35359037e+00 7.00406507e-02 -1.38338971e+00 1.49049610e-01 1.41532570e-01 7.87402391e-01 3.12815219e-01 1.34699523e-01 -4.80573535e-01 7.21951649e-02 -8.58319521e-01 5.43760538e-01 7.25163460e-01 9.37743127e-01 -7.33472288e-01 3.33372533e-01 1.97631180e-01 -1.65315986e+00 5.40336788e-01 -4.79722172e-01 1.02436826e-01 2.50522465e-01 5.96108913e-01 -5.92576861e-01 7.00425625e-01 7.31860280e-01 1.11148024e+00 -2.73527890e-01 1.26663625e+00 7.88889602e-02 7.29432181e-02 -4.37807381e-01 3.95221800e-01 9.54332873e-02 -8.36520135e-01 9.96240497e-01 8.17243755e-01 6.51445329e-01 8.82139951e-02 1.92514434e-01 9.17879224e-01 -5.55634238e-02 2.73010492e-01 -9.34702754e-01 4.34584320e-01 8.63661408e-01 1.14229155e+00 -5.99398494e-01 -4.05114591e-02 -4.80487317e-01 7.67190099e-01 2.40491763e-01 1.87260225e-01 -7.34573901e-01 -3.83752733e-01 1.29844725e+00 2.75469840e-01 -8.12254474e-02 -7.40795970e-01 -4.44494933e-01 -1.15800142e+00 1.88882709e-01 -1.22963324e-01 2.65105069e-01 -8.40750456e-01 -1.19048119e+00 4.77502406e-01 -5.24300709e-02 -1.82312107e+00 2.00793341e-01 -4.15907621e-01 -6.95886493e-01 8.98223996e-01 -1.50759399e+00 -5.95521569e-01 -8.66418004e-01 8.61049831e-01 1.53870195e-01 8.40500668e-02 6.13857687e-01 3.46694589e-01 -3.18612576e-01 3.11714053e-01 2.61080563e-01 3.63136269e-02 5.11824548e-01 -1.02878785e+00 6.26978993e-01 6.75461411e-01 1.09870568e-01 5.76286495e-01 2.38100246e-01 -5.73419929e-01 -1.44510460e+00 -1.27988017e+00 8.95122707e-01 -7.10064888e-01 4.80964005e-01 -4.25791405e-02 -1.15181208e+00 6.07291281e-01 -4.70411837e-01 3.29660833e-01 5.61012089e-01 -3.35562885e-01 -3.28060389e-01 -1.93086460e-01 -1.19941449e+00 5.62901258e-01 1.41508567e+00 -6.86303198e-01 -4.25939858e-01 2.90196151e-01 5.82137167e-01 -6.45679772e-01 -1.15594196e+00 4.82157052e-01 1.88098356e-01 -9.19681370e-01 1.34384060e+00 -2.43962660e-01 8.87873471e-02 -6.62784338e-01 -4.20853883e-01 -1.27067280e+00 -2.62430042e-01 -6.62329793e-01 -1.09015919e-01 9.88797188e-01 2.26092599e-02 -8.23267817e-01 7.98572719e-01 5.63759863e-01 -2.57101506e-01 -7.15339184e-01 -1.07032990e+00 -9.84637260e-01 1.92611590e-01 -4.90605831e-01 1.09203362e+00 1.09950018e+00 -2.54401833e-01 -1.05349258e-01 1.59213990e-01 4.44532782e-01 9.34692621e-01 4.50955689e-01 9.43928003e-01 -1.49466825e+00 3.96687627e-01 -6.42025173e-01 -8.07463348e-01 -1.13278866e+00 1.39465779e-01 -1.35803211e+00 -1.75756425e-01 -1.49137557e+00 -1.84135854e-01 -9.72599983e-01 -3.98711599e-02 5.97562157e-02 -7.80181447e-03 2.24001259e-01 1.39686972e-01 7.26196885e-01 -3.67025286e-01 6.52234614e-01 1.32464886e+00 2.19228268e-02 -4.41796839e-01 -3.49929519e-02 -4.07363236e-01 8.59937727e-01 7.69222260e-01 -3.33429039e-01 -9.84187797e-02 -7.10780799e-01 -1.38365373e-01 -5.88868819e-02 5.17883658e-01 -9.79491353e-01 3.89300019e-01 -2.85091609e-01 1.75162945e-02 -8.46707225e-01 3.70293438e-01 -9.69986916e-01 5.00035822e-01 2.31711239e-01 2.55891029e-02 3.25860947e-01 4.75362316e-02 5.11468649e-01 -2.76367635e-01 2.71054432e-02 9.10490215e-01 3.67914401e-02 -5.88671148e-01 8.02449107e-01 4.04242426e-01 1.33077398e-01 1.18763888e+00 -5.42838991e-01 -1.28184214e-01 -7.05868006e-02 -3.58616263e-01 3.13986987e-02 1.03130662e+00 3.18778992e-01 9.88748908e-01 -1.70242202e+00 -7.06208706e-01 3.42226774e-01 3.20979297e-01 4.28176492e-01 3.49340141e-02 9.71186280e-01 -7.79326200e-01 4.36379582e-01 4.70134281e-02 -1.15552795e+00 -1.31380486e+00 3.35006177e-01 4.28346395e-01 2.55786687e-01 -8.05105269e-01 5.75074852e-01 1.68166116e-01 -7.43330657e-01 -8.48449022e-02 -4.16321427e-01 1.20899782e-01 -2.54844189e-01 2.65623510e-01 7.33215034e-01 1.73060179e-01 -1.22378600e+00 -6.33190274e-01 1.55175960e+00 3.59673709e-01 2.75921315e-01 1.18884361e+00 -1.85419634e-01 -1.83842227e-01 4.47921097e-01 1.67112267e+00 1.20400816e-01 -1.26039231e+00 -4.49764311e-01 5.78043312e-02 -1.09266317e+00 3.89898941e-02 6.15114532e-02 -1.30754089e+00 7.17827737e-01 5.74489772e-01 -2.01631198e-03 8.64906907e-01 2.99967825e-01 8.58349323e-01 9.27188322e-02 6.20636225e-01 -4.91861224e-01 -1.34433016e-01 6.01303756e-01 8.64452958e-01 -1.13850224e+00 1.09984137e-01 -6.82897151e-01 -2.74530530e-01 9.67288792e-01 2.16307715e-01 -4.24279749e-01 8.76872182e-01 -5.44148311e-02 -4.32729200e-02 -3.17353040e-01 -1.22183233e-01 -4.02929746e-02 5.23281276e-01 5.11682272e-01 2.79884666e-01 8.61736462e-02 -1.02758165e-02 -2.13757291e-01 -2.78170794e-01 -1.57248899e-01 1.75683782e-01 9.97433424e-01 -3.20044160e-01 -1.04235494e+00 -4.22411889e-01 2.86307067e-01 8.07643011e-02 1.91448405e-01 -1.84319913e-01 5.98883748e-01 -2.06269339e-01 6.92182720e-01 6.88106954e-01 -3.00775230e-01 5.23130715e-01 -2.68878579e-01 3.65278125e-01 -5.67322195e-01 -6.61958382e-02 -2.43895605e-01 -4.95529264e-01 -8.93283546e-01 -6.35390997e-01 -8.97408545e-01 -1.32023597e+00 -5.60917139e-01 -9.39320922e-02 6.18084893e-02 7.00638890e-01 5.92782736e-01 7.49287605e-01 -2.56114274e-01 1.01766455e+00 -1.16417682e+00 -1.99078977e-01 -2.10717246e-01 -4.93918329e-01 6.94888532e-01 3.20994556e-01 -9.27761614e-01 -6.04344189e-01 -4.54926975e-02]
[7.712401866912842, -2.8573415279388428]
0a66b1eb-18c7-4cd6-95e0-21fdfb9f4173
inspecting-spoken-language-understanding-from
2306.00482
null
https://arxiv.org/abs/2306.00482v1
https://arxiv.org/pdf/2306.00482v1.pdf
Inspecting Spoken Language Understanding from Kids for Basic Math Learning at Home
Enriching the quality of early childhood education with interactive math learning at home systems, empowered by recent advances in conversational AI technologies, is slowly becoming a reality. With this motivation, we implement a multimodal dialogue system to support play-based learning experiences at home, guiding kids to master basic math concepts. This work explores Spoken Language Understanding (SLU) pipeline within a task-oriented dialogue system developed for Kid Space, with cascading Automatic Speech Recognition (ASR) and Natural Language Understanding (NLU) components evaluated on our home deployment data with kids going through gamified math learning activities. We validate the advantages of a multi-task architecture for NLU and experiment with a diverse set of pretrained language representations for Intent Recognition and Entity Extraction tasks in the math learning domain. To recognize kids' speech in realistic home environments, we investigate several ASR systems, including the commercial Google Cloud and the latest open-source Whisper solutions with varying model sizes. We evaluate the SLU pipeline by testing our best-performing NLU models on noisy ASR output to inspect the challenges of understanding children for math learning in authentic homes.
['Lama Nachman', 'Saurav Sahay', 'Roddy Fuentes Alba', 'Eda Okur']
2023-06-01
null
null
null
null
['intent-recognition', 'spoken-language-understanding', 'spoken-language-understanding', 'automatic-speech-recognition']
['natural-language-processing', 'natural-language-processing', 'speech', 'speech']
[-2.04298645e-02 3.56235564e-01 3.03736061e-01 -6.25088751e-01 -8.48617733e-01 -6.67779028e-01 4.69761461e-01 1.78436458e-01 -2.04928711e-01 2.76914299e-01 8.98483217e-01 -5.15801609e-01 1.73915431e-01 -9.28380370e-01 -3.96457464e-01 3.58639732e-02 2.04464287e-01 6.37924135e-01 3.30071926e-01 -9.09637213e-01 -2.20908627e-01 -5.97024485e-02 -1.85027933e+00 9.43237066e-01 8.87272418e-01 4.01803792e-01 1.05898969e-01 1.43878889e+00 -4.82118130e-01 1.85368621e+00 -8.96402597e-01 -3.77900362e-01 -1.56880721e-01 -1.11051798e-01 -1.33535945e+00 -3.87421191e-01 5.21604836e-01 -8.58727992e-01 -4.24731016e-01 2.90651768e-01 8.42766881e-01 6.55889571e-01 1.64053202e-01 -1.18710244e+00 -8.27901959e-01 1.16614544e+00 4.20751601e-01 5.31988814e-02 1.19893193e+00 2.76237577e-01 9.02902305e-01 -3.58811408e-01 4.17687774e-01 1.17597389e+00 5.22705972e-01 1.14958870e+00 -8.52780700e-01 -6.36538923e-01 -1.01236835e-01 1.81021869e-01 -1.15405715e+00 -7.88003922e-01 4.67894971e-01 -3.04907858e-01 1.67682827e+00 4.00557727e-01 7.50062525e-01 1.45244002e+00 -5.11857212e-01 1.56848621e+00 8.16735804e-01 -5.16906440e-01 1.45673305e-01 3.18969339e-01 6.98165596e-01 8.41853321e-01 -7.99991786e-01 -5.06742954e-01 -1.07295036e+00 1.46191344e-01 2.50783414e-01 -2.14792058e-01 -1.54743925e-01 2.51233876e-01 -9.26058471e-01 6.80848718e-01 -3.05509210e-01 2.62151331e-01 7.41054583e-03 9.06378925e-02 5.77317536e-01 6.85920656e-01 7.51807988e-01 5.50419271e-01 -6.82446659e-01 -1.26924849e+00 -6.98584437e-01 3.75471830e-01 1.43717277e+00 1.22853279e+00 1.84522212e-01 -1.44039273e-01 -3.86102796e-01 1.37193739e+00 5.33771574e-01 3.58374804e-01 5.94546735e-01 -8.74227107e-01 6.20662987e-01 8.23755801e-01 -3.51711571e-01 -1.01408288e-01 -3.17581087e-01 2.96476692e-01 -3.00808370e-01 -2.07928196e-01 4.50973511e-01 -5.80895543e-01 -5.94584644e-01 1.56933153e+00 5.58483481e-01 6.15257382e-01 6.50920987e-01 3.83923054e-01 2.14654803e+00 9.70976174e-01 3.51277769e-01 5.10944188e-01 1.38341701e+00 -1.23197389e+00 -7.85331666e-01 3.75107490e-03 1.21130192e+00 -6.11715734e-01 1.18846619e+00 4.21721071e-01 -1.11334968e+00 -2.98014373e-01 -7.06644654e-01 -6.95350289e-01 -6.80344582e-01 9.18965787e-02 7.51944363e-01 1.14113271e+00 -1.46105063e+00 1.93886608e-01 -1.07213390e+00 -5.71855903e-01 1.17470622e-01 1.53064445e-01 -4.29759115e-01 9.42518041e-02 -1.13111579e+00 6.06535017e-01 -6.11375843e-04 -5.77571690e-01 -7.88453043e-01 -1.36351037e+00 -1.18515682e+00 3.91227067e-01 9.08777937e-02 -2.30229691e-01 1.99207664e+00 -2.43076995e-01 -2.39127994e+00 9.25188422e-01 1.62600800e-01 -3.96093607e-01 2.75635689e-01 -6.04333341e-01 -1.45882349e-02 -5.54184243e-03 -8.78007933e-02 7.95088589e-01 -1.42992005e-01 -3.53590488e-01 -5.94718754e-01 -1.61965623e-01 2.43337318e-01 4.82916117e-01 -5.33000231e-01 3.57917488e-01 4.50940989e-02 -1.82277605e-01 -1.66230172e-01 -3.64703119e-01 -1.09365303e-02 -6.80444837e-01 -6.04051873e-02 -8.97956014e-01 6.77129269e-01 -9.84191775e-01 1.25930119e+00 -1.85061085e+00 -2.62012035e-01 -1.57458723e-01 -1.02512809e-02 4.17824149e-01 -3.53939474e-01 8.66258323e-01 7.29364455e-02 -1.22970603e-01 5.99916160e-01 -6.93804562e-01 3.37521762e-01 6.22090995e-02 -2.73993760e-01 -3.36821258e-01 2.08937353e-03 8.16180706e-01 -1.16593313e+00 -1.09130092e-01 5.53305745e-01 5.74595571e-01 -7.25183070e-01 1.13048303e+00 -3.43609869e-01 1.35491386e-01 -5.11451185e-01 5.57760894e-01 7.48810992e-02 8.13171342e-02 -1.48682758e-01 6.28720939e-01 -2.76787102e-01 1.16330838e+00 -1.19961488e+00 2.11754775e+00 -9.36333716e-01 7.73934603e-01 3.31905395e-01 -6.34530663e-01 9.63703811e-01 8.77619922e-01 3.46730888e-01 -4.52034891e-01 -3.56398523e-02 -1.01579413e-01 -3.56177658e-01 -9.50984836e-01 6.16956711e-01 4.81316090e-01 -8.12944397e-02 8.46569121e-01 7.98149407e-01 -5.02288043e-01 -8.58183354e-02 5.02896667e-01 1.87163615e+00 1.10780552e-01 9.01058838e-02 -1.00133836e-01 3.50108385e-01 -2.04784676e-01 -2.51252174e-01 9.11955714e-01 -2.77011096e-01 5.15344203e-01 1.08631231e-01 -5.22680879e-01 -4.72864091e-01 -9.37720776e-01 2.63141543e-01 2.09473395e+00 -7.25472927e-01 -7.77702093e-01 -1.08167386e+00 -6.11450076e-01 -4.20762181e-01 1.01129842e+00 -9.24405754e-02 6.05449826e-02 -2.46162027e-01 -1.22028105e-01 9.99165952e-01 4.16144669e-01 6.75180674e-01 -1.08646941e+00 -8.24734449e-01 4.38840568e-01 -1.08891264e-01 -1.24490201e+00 -1.05339766e-01 -4.67301272e-02 -1.42689999e-02 -8.11779082e-01 -6.57548368e-01 -6.74145103e-01 -7.08276778e-02 2.27739289e-01 1.42231727e+00 8.38405043e-02 -2.84064293e-01 1.49213099e+00 -8.20209026e-01 -6.41868949e-01 -9.12278473e-01 4.56797659e-01 -3.81786525e-02 -6.76837683e-01 9.63110626e-01 -8.19708705e-01 -1.43956438e-01 -1.54327855e-01 -4.58608329e-01 6.10301614e-01 -1.75283715e-01 4.24050122e-01 -5.13138354e-01 -3.17511052e-01 5.32336473e-01 -7.52656758e-01 9.17533815e-01 -5.89161634e-01 -2.24184453e-01 4.92830098e-01 1.29149750e-01 -2.13491216e-01 3.03108633e-01 -4.04210418e-01 -1.40414953e+00 4.64412123e-02 -1.07464099e+00 -9.21655595e-02 -1.01699448e+00 1.39201179e-01 -1.19702660e-01 1.88716024e-01 6.21919453e-01 3.41939569e-01 -3.07934552e-01 -6.48419023e-01 7.22342610e-01 1.38911998e+00 4.35749501e-01 -1.08362305e+00 9.76032205e-03 -3.95291686e-01 -8.88292611e-01 -1.48625302e+00 -7.30576158e-01 -7.87506044e-01 -5.39914489e-01 -4.92874682e-01 1.06788671e+00 -1.32403982e+00 -1.45010734e+00 6.62456453e-01 -1.42741239e+00 -1.06479442e+00 -3.41759384e-01 4.20357883e-01 -6.03398800e-01 -6.73072711e-02 -1.01062143e+00 -1.25562882e+00 -6.53580606e-01 -1.23253584e+00 1.11688530e+00 5.90477288e-01 -6.88515604e-01 -1.02221513e+00 4.03288841e-01 1.28670764e+00 3.89354140e-01 -4.00051594e-01 6.98149800e-01 -1.69543862e+00 -4.16565716e-01 1.93001062e-01 1.36128291e-01 1.45997480e-01 -2.76356548e-01 7.67767876e-02 -1.44261062e+00 2.11941108e-01 -3.25612009e-01 -1.10053098e+00 1.71458542e-01 -1.03779681e-01 7.47978926e-01 -5.42281210e-01 1.43118799e-01 2.45043844e-01 6.64182484e-01 2.73667186e-01 2.34647557e-01 1.50168419e-01 7.21468329e-01 8.09208155e-01 1.63839951e-01 4.71540332e-01 1.16070163e+00 4.17119235e-01 -2.36868799e-01 3.77568781e-01 -8.32538083e-02 -6.42789364e-01 5.74261427e-01 1.06156540e+00 9.00978893e-02 -3.53115678e-01 -1.52468705e+00 8.52023661e-01 -1.84322190e+00 -8.21525335e-01 -5.20448498e-02 1.67194724e+00 1.28372288e+00 -5.20089388e-01 1.62688851e-01 -3.26682270e-01 3.59543078e-02 1.58153981e-01 6.64450750e-02 -8.24352980e-01 5.62826157e-01 7.35684514e-01 -1.77695110e-01 6.85257852e-01 -9.35708284e-01 1.36891198e+00 5.82835197e+00 9.08537447e-01 -7.77125537e-01 3.33327204e-01 7.30588198e-01 -1.28448293e-01 -2.28413656e-01 -6.77143157e-01 -9.10958230e-01 6.69924021e-02 1.44789588e+00 4.35237847e-02 6.09850705e-01 1.12178922e+00 9.78104100e-02 -6.07642811e-03 -1.48659587e+00 8.18451703e-01 -1.77917965e-02 -1.53003192e+00 -2.89379805e-01 -6.23927712e-01 4.71009284e-01 4.86621171e-01 -1.99476093e-01 9.77000475e-01 1.18581748e+00 -1.20844030e+00 3.91635716e-01 1.36677399e-01 3.77723038e-01 -6.05565429e-01 4.15368885e-01 6.50208056e-01 -1.02294838e+00 1.19061008e-01 1.76495358e-01 -5.29049456e-01 -1.79629490e-01 -2.75916576e-01 -1.51370382e+00 -1.56436563e-02 1.08477104e+00 4.98872489e-01 -4.00376230e-01 4.32335436e-01 -3.64519626e-01 1.00449896e+00 -7.83593714e-01 -6.56551123e-01 2.04616144e-01 -1.48010403e-01 3.62230897e-01 1.54604650e+00 2.38273412e-01 1.02239871e+00 1.11800559e-01 8.22727561e-01 -9.40790623e-02 4.61609155e-01 -8.74507785e-01 -2.32277811e-01 5.98431885e-01 1.26119447e+00 -4.00273472e-01 -2.21168876e-01 -6.96813703e-01 8.67126048e-01 3.96494329e-01 1.78405553e-01 -3.00745666e-01 -1.67733103e-01 1.09914994e+00 -3.14391144e-02 -3.99780333e-01 -1.93391711e-01 -6.36802092e-02 -1.19477272e+00 -4.50264394e-01 -1.24880922e+00 3.19011807e-01 -9.20224011e-01 -8.70363891e-01 2.33597636e-01 -4.59440053e-02 -3.62606645e-01 -5.88382423e-01 -4.29517090e-01 -1.24447119e+00 5.72450817e-01 -7.21013904e-01 -1.35605872e+00 -9.58334729e-02 4.46757078e-01 1.36407447e+00 -4.54643309e-01 1.24069619e+00 2.82952756e-01 -6.95896447e-01 7.31790304e-01 -3.20329845e-01 5.21100342e-01 4.43183094e-01 -1.59396112e+00 7.06039131e-01 4.50301796e-01 3.49012434e-01 5.43059707e-01 6.68916166e-01 -5.62622488e-01 -1.48005915e+00 -6.37349129e-01 9.57657456e-01 -8.36947322e-01 8.61772001e-01 -8.27417493e-01 -9.62673187e-01 9.81404662e-01 7.13815987e-01 -6.16534770e-01 1.21241295e+00 4.80102748e-01 -1.55653179e-01 4.05130595e-01 -1.13230145e+00 5.91641247e-01 9.35768604e-01 -8.53482842e-01 -8.20596218e-01 5.34936428e-01 1.21686625e+00 -8.71496379e-01 -1.26563442e+00 1.56201601e-01 4.23999578e-01 -8.61484647e-01 8.89565110e-01 -9.62779105e-01 8.14308882e-01 4.25541043e-01 2.86300816e-02 -1.32888699e+00 4.72995341e-01 -9.48322058e-01 -7.75796324e-02 2.00055242e+00 3.76232296e-01 -2.43696108e-01 1.11382055e+00 1.44352198e+00 -1.96699098e-01 -7.03441978e-01 -9.78722155e-01 1.39117807e-01 -6.15280401e-03 -8.84005070e-01 7.87675679e-01 1.13863909e+00 9.06590700e-01 7.46348679e-01 9.55390409e-02 2.53883779e-01 -4.48748283e-02 -5.99107325e-01 1.00552583e+00 -7.62942731e-01 -2.50939816e-01 -2.84351647e-01 -3.78753722e-01 -1.13031924e+00 4.56880748e-01 -6.53027952e-01 6.40143678e-02 -1.54550827e+00 -1.61568552e-01 -3.21431816e-01 5.00440598e-01 9.25677240e-01 1.89206526e-02 -4.43767399e-01 9.29227993e-02 -7.04759836e-01 -1.04009426e+00 6.05036080e-01 5.24414301e-01 -3.08661133e-01 -5.76400459e-01 2.98765659e-01 -3.28992158e-01 8.73350739e-01 4.66958195e-01 9.09187645e-02 -7.81823993e-01 -5.61131418e-01 1.51764438e-01 3.80995452e-01 -1.28671378e-01 -1.16424978e+00 6.57186329e-01 1.45038769e-01 -1.12431802e-01 -4.72923338e-01 3.73897910e-01 -5.41455925e-01 -5.18008649e-01 -1.99856862e-01 -8.33953679e-01 -4.02948737e-01 4.06787992e-01 -3.37932676e-01 -4.71947640e-02 -2.66710550e-01 2.88816482e-01 -4.83227670e-02 -9.10541296e-01 -2.62692366e-02 -9.87738788e-01 9.81530845e-02 8.08671474e-01 1.85068160e-01 -6.58182025e-01 -9.14271653e-01 -8.60166669e-01 8.50008667e-01 -1.85532212e-01 8.89629900e-01 7.57250130e-01 -8.75600398e-01 -7.86047280e-01 4.61844325e-01 1.24353141e-01 3.53266567e-01 4.73131448e-01 1.90556630e-01 -7.43161857e-01 4.59592164e-01 1.77300274e-01 -4.38445330e-01 -1.89730382e+00 -1.53680161e-01 3.77265245e-01 -2.52822578e-01 -5.57807624e-01 1.51491833e+00 -2.21408665e-01 -1.34293628e+00 7.08702385e-01 -7.27309048e-01 -6.91551268e-01 2.24982202e-03 1.07761395e+00 6.86426878e-01 1.90644950e-01 -1.59217283e-01 2.31634736e-01 -2.36921787e-01 -4.43866812e-02 -2.75888830e-01 1.67477977e+00 -3.05508506e-02 1.98430106e-01 5.12594044e-01 8.02613795e-01 -1.52001873e-01 -7.46091008e-01 -1.29983425e-01 -7.70570040e-02 1.57886058e-01 -8.00222531e-02 -9.89531040e-01 -5.45642376e-01 1.05996919e+00 6.02079451e-01 3.61978531e-01 6.32889092e-01 1.97191924e-01 9.27082598e-01 1.06550789e+00 3.72551709e-01 -1.08454609e+00 5.39694019e-02 9.75089133e-01 4.97148126e-01 -1.28039205e+00 -4.90761131e-01 -3.43753040e-01 -6.57295167e-01 1.14559877e+00 1.13374281e+00 2.96185404e-01 4.58021969e-01 5.59078753e-01 4.95502740e-01 -2.76556730e-01 -1.22907126e+00 -2.30854169e-01 -3.77132297e-02 6.30984962e-01 1.08908987e+00 2.00810090e-01 3.71134967e-01 1.08316267e+00 -7.75072038e-01 7.97699206e-03 7.85215199e-01 7.71102905e-01 -4.13385540e-01 -8.95077825e-01 -1.85406819e-01 7.38866702e-02 -4.41787601e-01 -3.75383168e-01 -6.25417888e-01 5.59822321e-01 -9.00273994e-02 1.36641955e+00 5.71739003e-02 -4.47839826e-01 5.20969510e-01 5.56848228e-01 2.36105427e-01 -1.39222765e+00 -1.26693451e+00 -6.17036700e-01 6.70703471e-01 -8.10479045e-01 4.08716202e-02 -5.29129446e-01 -1.30041158e+00 -2.60904759e-01 -5.84589541e-02 3.20754707e-01 6.10238969e-01 1.07047236e+00 4.63873073e-02 6.01136148e-01 1.64659753e-01 -4.68874812e-01 -3.16423267e-01 -1.07347250e+00 -1.51673838e-01 -1.39373079e-01 1.91980436e-01 1.43635318e-01 3.44093889e-02 -9.68832821e-02]
[12.623919486999512, 7.918766975402832]
a8ee294f-4e5c-4c92-9724-fdffcf9c53ab
tmu-nmt-system-with-automatic-post-editing-by
null
null
https://aclanthology.org/2022.wat-1.4
https://aclanthology.org/2022.wat-1.4.pdf
TMU NMT System with Automatic Post-Editing by Multi-Source Levenshtein Transformer for the Restricted Translation Task of WAT 2022
In this paper, we describe our TMU English–Japanese systems submitted to the restricted translation task at WAT 2022 (Nakazawa et al., 2022). In this task, we translate an input sentence with the constraint that certain words or phrases (called restricted target vocabularies (RTVs)) should be contained in the output sentence. To satisfy this constraint, we address this task using a combination of two techniques. One is lexical-constraint-aware neural machine translation (LeCA) (Chen et al., 2020), which is a method of adding RTVs at the end of input sentences. The other is multi-source Levenshtein transformer (MSLevT) (Wan et al., 2020), which is a non-autoregressive method for automatic post-editing. Our system generates translations in two steps. First, we generate the translation using LeCA. Subsequently, we filter the sentences that do not satisfy the constraints and post-edit them with MSLevT. Our experimental results reveal that 100% of the RTVs can be included in the generated sentences while maintaining the translation quality of the LeCA model on both English to Japanese (En→Ja) and Japanese to English (Ja→En) tasks. Furthermore, the method used in previous studies requires an increase in the beam size to satisfy the constraints, which is computationally expensive. In contrast, the proposed method does not require a similar increase and can generate translations faster.
['Mamoru Komachi', 'Seiichiro Kondo']
null
null
null
null
wat-2022-10
['automatic-post-editing', 'automatic-post-editing']
['computer-vision', 'natural-language-processing']
[ 5.61282337e-01 4.26702723e-02 9.07851458e-02 -3.71910363e-01 -1.07924771e+00 -6.40794635e-01 5.13596892e-01 -4.34788585e-01 -4.62662905e-01 1.15061569e+00 3.29140455e-01 -6.68833435e-01 3.60151023e-01 -6.91861153e-01 -7.91705251e-01 -4.25244659e-01 6.42012715e-01 4.83752519e-01 1.33032337e-01 -4.67443943e-01 -4.71953116e-02 1.50956422e-01 -9.90072131e-01 4.77557808e-01 1.20926440e+00 4.81765181e-01 6.72532797e-01 5.82011104e-01 -1.30223006e-01 3.01293522e-01 -5.64387441e-01 -5.60744107e-01 3.82830441e-01 -9.57161248e-01 -8.52348983e-01 -2.54490405e-01 1.75674111e-01 -1.07721716e-01 7.75019974e-02 1.17608404e+00 4.82037574e-01 4.97961156e-02 6.62768364e-01 -7.19274163e-01 -9.31207895e-01 1.04220951e+00 -3.42136979e-01 1.97337791e-01 2.93432653e-01 -7.36855045e-02 9.98695850e-01 -1.37986076e+00 7.04677105e-01 1.09691322e+00 4.12689775e-01 7.90095985e-01 -1.25151575e+00 -6.24267578e-01 -4.91927192e-02 -6.49988204e-02 -1.37616515e+00 -5.86714685e-01 4.33819652e-01 -2.80081183e-01 1.38798976e+00 6.36928082e-01 3.78879219e-01 9.90495324e-01 6.00785494e-01 5.86242020e-01 1.03793299e+00 -9.10951555e-01 1.14090266e-02 1.25713482e-01 6.18225411e-02 4.20644671e-01 4.53472137e-04 -1.62809470e-03 -4.70664233e-01 5.08875549e-02 4.40582573e-01 -4.88331228e-01 -2.18570501e-01 4.12154794e-01 -1.46367300e+00 6.96042299e-01 -2.49185544e-02 5.09936333e-01 -5.55790067e-01 -8.41921717e-02 4.31562901e-01 6.47712588e-01 5.53967595e-01 5.23658097e-01 -6.36394739e-01 -8.98620710e-02 -1.00898898e+00 -6.38918504e-02 5.00424385e-01 1.33924437e+00 7.24735796e-01 7.14586079e-02 -4.68266040e-01 8.80801201e-01 2.01362506e-01 9.29392040e-01 8.13380361e-01 -6.10932827e-01 9.44985807e-01 2.69153386e-01 5.98231964e-02 -6.14219606e-01 -1.79727167e-01 -3.31149459e-01 -9.11235511e-01 -3.18928331e-01 1.83964834e-01 -4.07215416e-01 -9.82486546e-01 1.82541311e+00 5.20961313e-03 -3.73415500e-01 4.32173878e-01 7.68060327e-01 6.50860488e-01 1.20130908e+00 -2.70384640e-01 -7.43857205e-01 1.33280957e+00 -1.24012256e+00 -1.03206110e+00 -3.07571322e-01 7.61986315e-01 -1.31079042e+00 1.42763591e+00 1.23702064e-01 -1.18616438e+00 -7.01848626e-01 -9.29114044e-01 -2.30429873e-01 -1.99767023e-01 5.45550466e-01 1.57385379e-01 5.39947271e-01 -1.15954483e+00 2.49354541e-01 -7.33173966e-01 -3.51665825e-01 -4.38605011e-01 3.15808743e-01 -2.19006300e-01 8.15486386e-02 -1.68504930e+00 1.05131519e+00 2.66382724e-01 3.29738379e-01 -3.35729212e-01 -3.12333465e-01 -8.78149509e-01 1.55420434e-02 3.06222290e-01 -6.71657205e-01 1.38067710e+00 -1.16331565e+00 -1.97335446e+00 6.26796901e-01 -7.41427898e-01 -3.51235688e-01 4.33783442e-01 -2.87377685e-01 -4.00668144e-01 -2.82082230e-01 1.55145511e-01 5.14023662e-01 6.49746537e-01 -8.03827822e-01 -5.99656701e-01 -1.76465176e-02 -3.17105860e-01 2.37506598e-01 -1.41651109e-01 4.62302923e-01 -6.20697856e-01 -8.65569949e-01 1.08390450e-01 -1.27700603e+00 -8.27663094e-02 -7.58280635e-01 -5.26595950e-01 -3.20990086e-01 4.18279618e-01 -1.07750082e+00 1.51958382e+00 -1.93250525e+00 6.01252675e-01 1.80306017e-01 -2.54484445e-01 3.39151770e-01 -4.90852028e-01 7.25484848e-01 1.45345509e-01 2.29839891e-01 -5.16675353e-01 -5.05489767e-01 -7.02882335e-02 1.43919170e-01 -6.34028018e-01 -3.74796838e-02 2.55895108e-01 1.18124223e+00 -8.31478894e-01 -3.77022117e-01 -9.94242281e-02 1.50220215e-01 -4.04644817e-01 8.43833685e-02 -2.91483104e-01 4.00970429e-01 -1.35161743e-01 3.02850217e-01 4.77664769e-01 2.65866548e-01 2.39181012e-01 -6.83900341e-02 -3.88400435e-01 7.49700308e-01 -8.07606339e-01 1.32031143e+00 -6.94201887e-01 6.18172109e-01 -2.02739775e-01 -5.15758514e-01 9.98999536e-01 4.62937921e-01 1.77665427e-02 -5.90310037e-01 7.28665590e-02 7.21377552e-01 1.77298099e-01 -2.89486110e-01 8.85109305e-01 -1.64009809e-01 -2.57665694e-01 5.33953071e-01 -1.53737977e-01 -3.99435729e-01 6.52267456e-01 2.62709465e-02 7.42382228e-01 3.40751976e-01 3.72220814e-01 -2.63257533e-01 7.12825835e-01 1.40541360e-01 8.20403337e-01 6.43289566e-01 1.64796874e-01 7.08914220e-01 2.24942252e-01 -1.27092242e-01 -1.33461022e+00 -8.03195059e-01 2.59168416e-01 9.17033434e-01 -3.56837571e-01 -5.30792594e-01 -9.23307180e-01 -5.08843005e-01 -5.40982902e-01 1.09373546e+00 -2.61330068e-01 -9.93064195e-02 -1.03057826e+00 -5.60964704e-01 6.62444115e-01 4.29456890e-01 2.84193397e-01 -1.29922605e+00 -3.35706294e-01 3.22293699e-01 -7.79980242e-01 -1.04291499e+00 -1.22456217e+00 8.49019289e-02 -7.02642322e-01 -3.70698214e-01 -8.33121002e-01 -9.99790132e-01 7.48614311e-01 1.40233740e-01 8.72758150e-01 -1.76132888e-01 3.32729250e-01 -3.48695427e-01 -5.16928017e-01 -3.02339703e-01 -9.37777996e-01 2.73561299e-01 3.74566823e-01 -1.43177480e-01 5.09596944e-01 -1.37763798e-01 1.21560179e-01 3.02948803e-01 -7.41499782e-01 4.42899883e-01 8.26827049e-01 8.94281924e-01 8.25809836e-01 -1.72994733e-01 6.20535731e-01 -8.72823119e-01 9.53989208e-01 4.10614908e-02 -6.99361205e-01 4.21227783e-01 -5.36923707e-01 7.54637718e-02 1.05909944e+00 -6.22764528e-01 -8.46945643e-01 1.31451026e-01 -2.83484519e-01 -1.06292762e-01 2.78030604e-01 9.04030859e-01 -2.58172512e-01 3.52821440e-01 4.54037994e-01 7.26525366e-01 -1.70557797e-01 -3.91180396e-01 2.73460656e-01 9.51433837e-01 2.61574179e-01 -3.32777619e-01 9.40820038e-01 -3.29889029e-01 -2.66787767e-01 -5.28522074e-01 -6.28607571e-01 -9.75565016e-02 -7.54433692e-01 1.66507885e-02 7.82091081e-01 -6.59443319e-01 -9.86960381e-02 3.55068058e-01 -1.44074500e+00 -3.17420870e-01 -2.30467469e-02 9.01272893e-01 -5.27411580e-01 4.13061112e-01 -8.01366031e-01 -6.36574864e-01 -7.85138309e-01 -1.37643600e+00 7.99186885e-01 -4.57547344e-02 -5.87553501e-01 -5.21890342e-01 -5.15673049e-02 2.68174738e-01 5.87984860e-01 -3.21315885e-01 1.16235268e+00 -5.21918893e-01 -3.21978331e-01 1.00457482e-02 -5.95938042e-02 5.69196105e-01 3.93779814e-01 -1.31433615e-02 -4.60281432e-01 -3.64711404e-01 9.38666835e-02 -8.87732431e-02 5.73083520e-01 2.71932214e-01 5.16770422e-01 -5.22852719e-01 8.04555863e-02 2.62424201e-01 9.51778591e-01 5.31267524e-01 5.60032308e-01 1.37456983e-01 4.78288263e-01 5.08366406e-01 7.97359705e-01 -2.30005890e-01 3.54869694e-01 9.43898439e-01 -2.74169564e-01 -1.74736246e-01 -2.31337234e-01 -3.01080495e-01 9.92750585e-01 1.74646235e+00 -7.97768682e-02 -4.11015689e-01 -7.81082392e-01 3.96056175e-01 -1.87007666e+00 -7.01008916e-01 -4.53092664e-01 2.25438118e+00 1.36638463e+00 1.16527542e-01 -2.70659894e-01 -2.09551841e-01 7.35862136e-01 -1.23175941e-01 -2.16176867e-01 -9.57511008e-01 -1.86382100e-01 1.44916922e-01 3.63695174e-01 9.60453391e-01 -5.88344634e-01 1.46462440e+00 5.77162695e+00 7.27838695e-01 -1.12290454e+00 1.56887397e-01 2.64925629e-01 8.58825371e-02 -4.95849907e-01 -8.77857860e-03 -9.02814984e-01 5.16094506e-01 1.18241513e+00 -3.76403630e-01 8.15156579e-01 3.85409325e-01 5.40992677e-01 1.89032238e-02 -1.10784709e+00 5.89977801e-01 2.82360047e-01 -1.06547332e+00 2.15784416e-01 -7.09960014e-02 8.01109314e-01 -3.52439806e-02 -1.73626259e-01 6.18241549e-01 -1.14554167e-02 -6.99296832e-01 8.61622036e-01 5.35473228e-01 9.58090723e-01 -6.69837952e-01 1.08688104e+00 4.71046627e-01 -9.32258666e-01 3.60414565e-01 -6.27853453e-01 -3.04749645e-02 5.07441819e-01 6.74988806e-01 -7.13060617e-01 7.42095888e-01 3.13897520e-01 2.98019052e-01 -2.23804355e-01 4.37549412e-01 -5.34824550e-01 8.95136356e-01 -2.93886572e-01 -3.14972639e-01 2.44636327e-01 -5.24837971e-01 6.66767120e-01 1.24055302e+00 6.78530514e-01 8.40202160e-03 -7.42204627e-03 8.16331804e-01 -2.20040858e-01 5.17553627e-01 -4.53663558e-01 -1.52826354e-01 4.67443109e-01 8.88172865e-01 -4.19684559e-01 -5.35210907e-01 -3.05763215e-01 1.36598146e+00 2.17693135e-01 4.06532884e-01 -7.21515596e-01 -6.08639181e-01 2.22106293e-01 -2.14460179e-01 1.82050183e-01 -3.65508944e-01 -3.07206422e-01 -1.27814972e+00 3.28019202e-01 -1.19362080e+00 -1.23452805e-01 -9.02473330e-01 -9.80818510e-01 1.19241571e+00 -1.14485048e-01 -1.35579646e+00 -4.87441659e-01 -3.77300501e-01 -2.31656402e-01 1.40655816e+00 -1.32618868e+00 -1.20706451e+00 3.09050590e-01 2.82152206e-01 9.17083025e-01 -9.70187038e-02 9.35928881e-01 3.76426250e-01 -4.85311210e-01 6.45656347e-01 1.07003666e-01 4.20566276e-02 8.44565332e-01 -8.61611724e-01 8.02469313e-01 1.24677527e+00 2.19531998e-01 1.00519574e+00 7.22257733e-01 -1.04308844e+00 -1.33388507e+00 -1.24379408e+00 1.95608246e+00 -4.17397201e-01 5.69389522e-01 -5.80908179e-01 -8.29818904e-01 8.34778666e-01 4.73813087e-01 -5.63154042e-01 5.19651651e-01 -1.27922073e-01 2.90728658e-02 -3.40229739e-03 -7.15180039e-01 8.26625168e-01 7.36199677e-01 -5.27827621e-01 -8.49236667e-01 3.86616439e-01 1.20556569e+00 -5.56781471e-01 -5.37417591e-01 4.22949255e-01 3.80791098e-01 -2.99322784e-01 2.49901295e-01 -6.11552894e-01 5.34711897e-01 -5.87157369e-01 -3.73246461e-01 -1.70412838e+00 -4.67663139e-01 -7.98119843e-01 2.24630669e-01 1.26388514e+00 1.08638132e+00 -6.02001905e-01 7.43498430e-02 4.24157441e-01 -4.47053879e-01 -5.97536325e-01 -1.03034890e+00 -9.58069861e-01 2.50810593e-01 -3.38543981e-01 6.27555549e-01 7.53150702e-01 2.01380206e-03 6.66953444e-01 -7.04535723e-01 -2.26935968e-02 6.73237815e-02 1.27976499e-02 6.64490819e-01 -5.87555647e-01 -1.88544035e-01 -3.03430080e-01 2.42621794e-01 -1.07932520e+00 1.25073671e-01 -1.06777382e+00 3.42246324e-01 -1.61661565e+00 1.91632822e-01 -1.48708329e-01 -1.01909950e-01 5.83655596e-01 -3.50861877e-01 3.43686551e-01 4.10616457e-01 4.00755823e-01 7.28903636e-02 6.78320706e-01 1.14579439e+00 -4.91029993e-02 -5.38138866e-01 2.52131581e-01 -3.98111135e-01 3.21904838e-01 8.63013804e-01 -6.14078045e-01 -2.06994981e-01 -7.15377629e-01 3.69143516e-01 -3.64288539e-02 -3.14472586e-01 -4.34602052e-01 4.01055403e-02 -2.75268048e-01 -4.85500284e-02 -8.01019013e-01 1.09260991e-01 -7.11644709e-01 2.37104923e-01 5.80198467e-01 -3.64965498e-01 7.59082854e-01 2.51371443e-01 3.14257927e-02 -2.21347183e-01 -3.18171054e-01 5.65059900e-01 5.67652546e-02 -1.75306946e-01 -2.96429127e-01 -8.36842895e-01 -2.46569008e-01 7.25445747e-01 5.22205643e-02 -2.26851702e-01 -3.75172794e-01 -3.81855518e-01 1.79879144e-01 2.37430736e-01 6.50758803e-01 5.26032507e-01 -1.37278509e+00 -1.07315195e+00 3.49956751e-01 3.27781658e-03 -2.64621407e-01 -2.46681720e-01 1.04202378e+00 -4.68939036e-01 7.26308227e-01 6.84679896e-02 -2.56073773e-01 -1.20932949e+00 5.03061771e-01 4.38718610e-02 -4.13969666e-01 -4.98607546e-01 6.82279468e-01 -8.32897946e-02 -8.26064408e-01 -1.77481532e-01 -5.30325055e-01 4.36831303e-02 -1.72548383e-01 5.10933578e-01 3.41926664e-02 2.21998051e-01 -8.01387846e-01 -3.04677486e-01 5.00949025e-01 -2.83128351e-01 -4.64819014e-01 1.04314268e+00 -3.22797447e-01 -4.66404498e-01 5.83082676e-01 9.00599599e-01 5.87063968e-01 -4.17946279e-01 -4.12274033e-01 3.92867364e-02 -4.47086878e-02 -9.26415473e-02 -1.05967903e+00 -7.12985396e-01 7.26737499e-01 1.41195133e-01 -3.60315442e-02 1.10372365e+00 -2.88896948e-01 1.14361584e+00 5.41453719e-01 4.43078071e-01 -1.30477071e+00 -5.24132133e-01 1.10421860e+00 1.11631417e+00 -9.65207517e-01 -3.10046673e-01 -5.21402061e-01 -8.73558164e-01 1.10750556e+00 5.01528203e-01 3.37926358e-01 1.29589036e-01 1.74128041e-01 4.47426736e-01 4.60858017e-01 -8.89769971e-01 2.03367956e-02 5.16569138e-01 2.07197815e-01 5.65889478e-01 3.06676835e-01 -1.08456326e+00 6.44992173e-01 -5.36610961e-01 -1.73599049e-02 6.34527862e-01 7.07297862e-01 -2.74800390e-01 -1.31459630e+00 -3.66577655e-01 3.17787647e-01 -2.49282613e-01 -7.19330430e-01 -6.52128994e-01 4.05047745e-01 -1.14862345e-01 1.16735327e+00 -1.49118260e-01 -5.32223940e-01 4.62955624e-01 2.89249390e-01 2.19684139e-01 -8.56896460e-01 -7.26635993e-01 3.59762162e-01 2.43581578e-01 -2.29442328e-01 -1.42663032e-01 -5.98300755e-01 -1.27069557e+00 -4.02926691e-02 -5.99702775e-01 3.85012835e-01 6.86375916e-01 1.07009554e+00 3.78147304e-01 4.88505304e-01 6.32272422e-01 -4.07883048e-01 -7.14491189e-01 -1.41974390e+00 -1.02398664e-01 -2.29488946e-02 1.06775180e-01 -1.10300541e-01 -2.91575164e-01 2.96122491e-01]
[11.645397186279297, 10.24954891204834]
38e427b4-5145-43a0-890e-9023c304f2eb
rigidity-strengthening-is-a-vital-mechanism
1704.05883
null
http://arxiv.org/abs/1704.05883v1
http://arxiv.org/pdf/1704.05883v1.pdf
Rigidity strengthening is a vital mechanism for protein-ligand binding
Protein-ligand binding is essential to almost all life processes. The understanding of protein-ligand interactions is fundamentally important to rational drug design and protein design. Based on large scale data sets, we show that protein rigidity strengthening or flexibility reduction is a pivoting mechanism in protein-ligand binding. Our approach based solely on rigidity is able to unveil a surprisingly long range contribution of four residue layers to protein-ligand binding, which has a ramification for drug and protein design. Additionally, the present work reveals that among various pairwise interactions, the short range ones within the distance of the van der Waals diameter are most important. It is found that the present approach outperforms all the other state-of-the-art scoring functions for protein-ligand binding affinity predictions of two benchmark data sets
[]
2017-03-31
null
null
null
null
['protein-design']
['medical']
[ 1.98635742e-01 -1.12959124e-01 -5.46175420e-01 -3.68692040e-01 -2.94956356e-01 -5.47546148e-01 2.73823947e-01 2.53144175e-01 -4.06955481e-01 1.34033239e+00 2.47396931e-01 -5.66155374e-01 -3.66529971e-01 -5.32307804e-01 -8.29543889e-01 -1.34190965e+00 -8.68668482e-02 6.85826302e-01 5.04996419e-01 -6.48911119e-01 4.47231293e-01 9.72546160e-01 -9.66437638e-01 2.53672421e-01 7.34904706e-01 3.29901069e-01 2.58359551e-01 2.46926650e-01 4.29405496e-02 2.17007160e-01 1.60412505e-01 -2.63543755e-01 -1.43735096e-01 -2.70154208e-01 -8.82209480e-01 -5.77384233e-01 -1.52702674e-01 3.66154790e-01 4.96949069e-02 4.44302619e-01 8.27557445e-01 1.00393839e-01 9.88376677e-01 -6.01640530e-02 -5.99090755e-01 2.28509873e-01 -4.69815314e-01 2.33932018e-01 5.85215807e-01 2.03417554e-01 1.12404037e+00 -1.31860876e+00 7.71440268e-01 1.11739171e+00 3.86765391e-01 5.98479629e-01 -1.49720430e+00 -4.18583483e-01 -1.49040565e-01 2.81238854e-01 -1.29797840e+00 -2.20226198e-01 6.40958905e-01 -6.97477102e-01 1.52392709e+00 3.10744703e-01 4.70025122e-01 7.72659719e-01 6.34807646e-01 -1.10361338e-01 8.82455051e-01 -3.18538040e-01 1.45169869e-01 -4.15685624e-01 2.15746075e-01 5.42391002e-01 3.31640661e-01 1.87717572e-01 -5.89744866e-01 -8.42790246e-01 3.22709650e-01 6.82940334e-02 -4.24867451e-01 -6.58020973e-01 -8.73727620e-01 8.17804098e-01 4.57976043e-01 2.56992966e-01 -3.50233465e-01 3.87955979e-02 1.25859082e-01 -1.42042711e-01 -1.41089112e-02 5.21263123e-01 -8.89124215e-01 -1.73721269e-01 -3.66457403e-01 4.77974236e-01 7.46061742e-01 1.00635715e-01 7.75894821e-01 -6.77963436e-01 1.22284301e-01 7.11259007e-01 5.64056873e-01 3.36983055e-01 9.46067721e-02 -8.16681702e-03 1.15187250e-01 7.14235842e-01 3.15196157e-01 -7.01436996e-01 -7.30583429e-01 -1.25781540e-02 -5.03745735e-01 2.04005037e-02 5.97833872e-01 2.25307375e-01 -5.14634848e-01 1.59757471e+00 4.70429420e-01 -3.41863692e-01 -3.32988538e-02 7.94307172e-01 9.76446629e-01 4.40125048e-01 4.41286147e-01 -7.46378720e-01 1.37024117e+00 -5.46661437e-01 -4.04458940e-01 1.90214574e-01 4.15104747e-01 -8.97457182e-01 7.48203814e-01 2.23835230e-01 -9.23289239e-01 -2.49430820e-01 -8.02651763e-01 9.70128551e-02 -3.16545606e-01 -2.96665370e-01 9.10808802e-01 5.08622766e-01 -3.53677571e-01 8.85583043e-01 -7.16298759e-01 -4.61975276e-01 -2.28704186e-03 8.86297703e-01 -7.52191365e-01 7.34249130e-02 -1.26205194e+00 1.35749912e+00 4.25875306e-01 4.72153425e-02 -4.86446142e-01 -6.67205811e-01 -1.94574147e-01 -2.02145293e-01 2.78887928e-01 -4.60086823e-01 7.31044114e-01 -1.94323376e-01 -1.68441463e+00 8.57371151e-01 -3.60977203e-01 -2.52585918e-01 1.44828200e-01 -8.00711811e-02 -2.28718311e-01 -8.22562426e-02 -5.45501888e-01 3.57935935e-01 1.99651659e-01 -9.34327066e-01 2.15665072e-01 -4.91356909e-01 -2.72390187e-01 2.95716733e-01 1.89615488e-01 1.70237675e-01 1.01089999e-01 -3.30843329e-01 2.41075352e-01 -1.07277107e+00 -5.28147876e-01 -5.55902958e-01 -2.01845273e-01 -3.31641257e-01 4.51718539e-01 -1.28559425e-01 1.31376898e+00 -1.64668989e+00 6.66879654e-01 7.27887690e-01 2.65469283e-01 4.15768385e-01 1.68384388e-01 1.16512096e+00 -5.97666800e-01 -4.44301292e-02 7.45795071e-02 8.25140774e-01 -3.52701694e-01 8.78070593e-02 -2.59533405e-01 6.94015861e-01 -1.79613218e-01 9.64051068e-01 -5.98106563e-01 -6.90656826e-02 1.75174430e-01 6.78007543e-01 -5.78697026e-01 -4.44338098e-02 -4.01980758e-01 7.33546674e-01 -8.32093835e-01 5.47143877e-01 7.89428174e-01 -3.54761839e-01 6.38398111e-01 -3.96398515e-01 -8.68126079e-02 2.58228481e-01 -5.20991087e-01 1.22884643e+00 3.96652997e-01 -1.57200679e-01 -2.71008611e-01 -6.54049873e-01 1.03381360e+00 4.30011690e-01 7.67525971e-01 -3.89188498e-01 4.79182508e-03 4.31140453e-01 7.42126584e-01 -3.76225799e-01 -1.39708623e-01 -4.14470196e-01 3.11827511e-01 4.67123426e-02 -3.17819566e-01 2.20876038e-01 1.14340801e-02 1.79353673e-02 9.02282536e-01 1.55813754e-01 7.57680774e-01 -6.80043101e-01 9.37483788e-01 -1.20020129e-01 4.04742211e-01 1.39695123e-01 -2.58688089e-02 2.81576723e-01 6.26867652e-01 -8.94516110e-01 -1.06922102e+00 -7.72857547e-01 -6.27052188e-01 1.41433740e+00 1.33096902e-02 -5.00046909e-01 -7.24753559e-01 -2.09103495e-01 2.86140144e-01 -1.34199411e-01 -5.98006308e-01 -3.41026247e-01 -5.99780858e-01 -1.39417160e+00 2.70143539e-01 8.49176794e-02 -2.64172763e-01 -1.27368319e+00 -2.27054477e-01 5.75455546e-01 1.74767196e-01 -5.28864443e-01 -2.75216401e-01 7.10851014e-01 -7.70100772e-01 -1.40221786e+00 -6.12350285e-01 -4.76895690e-01 4.83404487e-01 6.86732009e-02 8.87593806e-01 2.04829052e-01 -3.56276244e-01 -3.66312593e-01 -1.14797510e-01 -1.02034315e-01 -3.49091649e-01 4.95782010e-02 4.54614758e-01 -3.87660772e-01 8.95444453e-01 -8.60970438e-01 -9.22229826e-01 7.15752304e-01 -4.12671834e-01 -1.68810174e-01 6.05849147e-01 7.62872458e-01 8.44805002e-01 -5.41819811e-01 5.76062262e-01 -1.00741243e+00 6.28399849e-01 -3.19137186e-01 -4.20216918e-01 1.75672531e-01 -6.44462526e-01 4.92188752e-01 5.82778156e-01 -3.47705930e-01 -7.91658819e-01 4.20338988e-01 -4.12368208e-01 4.31777894e-01 -7.44597390e-02 4.89021599e-01 -4.21167791e-01 -6.03902578e-01 7.42586672e-01 2.23942146e-01 2.54815016e-02 -7.13710368e-01 1.61680728e-01 2.67448992e-01 7.38089578e-03 -9.50571775e-01 4.11805153e-01 3.24694812e-01 5.83081663e-01 -8.95854592e-01 -3.40111464e-01 -5.29102862e-01 -7.95221865e-01 2.89787918e-01 1.00962687e+00 -4.93740082e-01 -1.46664214e+00 -8.91818181e-02 -1.13831568e+00 8.50708783e-02 3.63624722e-01 4.67522830e-01 -5.53423107e-01 5.81275761e-01 -5.61272204e-01 -4.79593486e-01 -3.01979154e-01 -1.60161543e+00 7.84664094e-01 1.80711690e-02 -3.65287393e-01 -7.34780908e-01 6.61664903e-01 4.72653687e-01 1.54086977e-01 3.97206396e-01 1.38710761e+00 -8.51137578e-01 -4.00639415e-01 2.23314073e-02 1.55651793e-02 -3.51433188e-01 1.42860651e-01 1.44854188e-01 -5.67084372e-01 -2.93731779e-01 -4.53773886e-01 -2.03085139e-01 9.49702322e-01 6.03256047e-01 7.51070917e-01 -1.69957832e-01 -5.66186965e-01 4.48517680e-01 1.34151971e+00 5.94408453e-01 1.06326687e+00 2.30988994e-01 7.78626740e-01 5.81561685e-01 6.53532624e-01 3.44877690e-01 -1.77978545e-01 1.05455887e+00 4.93051797e-01 -1.07983746e-01 5.33105612e-01 4.10323329e-02 3.00387263e-01 2.11203977e-01 -8.49847317e-01 -5.90081811e-02 -1.02646458e+00 -1.40523180e-01 -2.00302863e+00 -1.08170736e+00 -5.89918613e-01 2.30656219e+00 1.29944694e+00 9.55597609e-02 2.61914074e-01 -2.20478669e-01 3.34339172e-01 -5.63941076e-02 -7.12547123e-01 -5.16026139e-01 -1.78068191e-01 4.91145164e-01 4.98064011e-01 8.57655466e-01 -8.06330502e-01 9.59286153e-01 7.61457396e+00 8.53517771e-01 -9.53119695e-01 -2.71905154e-01 2.73091882e-01 6.88586896e-03 -1.93972245e-01 1.89909264e-01 -1.01608491e+00 3.60099852e-01 9.05188262e-01 -1.29379202e-02 1.93851382e-01 7.13228643e-01 6.09725237e-01 -1.72326908e-01 -1.14429903e+00 5.21635294e-01 -5.54095328e-01 -1.49161208e+00 1.81901723e-01 4.64496851e-01 4.70069021e-01 -3.16457497e-03 -1.77092925e-01 -4.95022744e-01 -7.90433884e-02 -1.59120989e+00 -8.84858519e-02 5.41106761e-01 6.78123653e-01 -9.00791049e-01 7.03257263e-01 1.65919036e-01 -1.18636501e+00 2.74207234e-01 -6.32553697e-01 -1.04165018e-01 8.58439282e-02 6.40328884e-01 -7.28240609e-01 2.35118985e-01 1.85468972e-01 4.41991061e-01 -2.97701389e-01 6.09150290e-01 4.26033251e-02 2.51508981e-01 -4.04602177e-02 -1.12004235e-01 4.03091349e-02 -5.24519026e-01 3.95034850e-01 1.13354778e+00 -4.08062905e-01 6.54181302e-01 1.52083680e-01 5.35684705e-01 1.14659198e-01 4.85258996e-01 -5.71635425e-01 2.48817466e-02 2.33142227e-01 1.01543903e+00 -5.65746009e-01 1.84965208e-01 -3.68441314e-01 6.80026233e-01 2.67831475e-01 2.70252526e-01 -8.21779191e-01 -2.09484890e-01 1.03437102e+00 3.24820995e-01 2.04740599e-01 -1.37309179e-01 3.63917559e-01 -7.32923865e-01 -1.85670480e-01 -7.81724632e-01 9.73378494e-02 -2.71344185e-01 -9.75595772e-01 1.18132554e-01 -2.31401786e-01 -5.90203464e-01 1.06070630e-01 -9.72123921e-01 -4.06816036e-01 1.07968581e+00 -1.33152032e+00 -8.96231055e-01 2.60093242e-01 4.39348787e-01 1.27802966e-02 -1.69116810e-01 1.06637931e+00 2.34757647e-01 -3.91012341e-01 2.51717478e-01 5.25398970e-01 -5.51972806e-01 8.72331262e-01 -1.00411487e+00 1.25899300e-01 1.31176174e-01 -3.19633842e-01 1.19027674e+00 1.11416483e+00 -8.25489700e-01 -1.71265674e+00 -4.31544691e-01 8.11672270e-01 -7.17402875e-01 5.93332589e-01 -3.23532850e-01 -1.15787494e+00 1.44802257e-01 -9.53024477e-02 -6.78867176e-02 1.29100180e+00 2.10271165e-01 -3.40126604e-01 2.82593787e-01 -9.63522792e-01 3.05456102e-01 9.20752108e-01 -4.55290735e-01 -5.76007903e-01 5.91744125e-01 5.76723218e-01 -2.95167029e-01 -1.12313116e+00 5.21744311e-01 8.82641494e-01 -1.03787303e+00 1.41332471e+00 -1.15161252e+00 1.08244158e-01 -3.96599025e-01 -2.68777430e-01 -6.96407378e-01 -8.04300725e-01 -7.79849529e-01 -1.79363593e-01 5.10451436e-01 8.20061207e-01 -4.26378667e-01 9.59858179e-01 5.86887062e-01 2.03318566e-01 -1.21328914e+00 -1.00490141e+00 -4.48808312e-01 3.42033565e-01 2.40070820e-01 3.11651677e-01 7.01162279e-01 4.68930990e-01 6.47593796e-01 -4.75875288e-01 -2.90913731e-01 2.49471828e-01 5.79718277e-02 6.60843790e-01 -1.46890759e+00 -4.30885792e-01 -3.20907742e-01 -2.24732742e-01 -7.76803732e-01 -7.09964288e-03 -7.48626590e-01 -3.83603781e-01 -1.25851345e+00 7.49607146e-01 -1.25758260e-01 -4.62507367e-01 3.55562270e-01 1.43264355e-02 3.94952223e-02 -2.65957683e-01 2.88189620e-01 -3.54534388e-01 4.83066380e-01 1.20658505e+00 2.18184561e-01 -3.87626320e-01 1.92254558e-02 -6.70793712e-01 7.05093503e-01 5.89064598e-01 -5.55775642e-01 -8.81419405e-02 4.42774028e-01 6.29350424e-01 -1.12006873e-01 -1.79159716e-01 -2.97227293e-01 -2.31357932e-01 -8.63583326e-01 3.07697535e-01 -4.53919142e-01 3.61610740e-01 -8.90085816e-01 5.59063137e-01 8.84660542e-01 -2.45415829e-02 -4.43718880e-02 1.86628774e-02 6.02475703e-01 1.55641481e-01 -2.63333134e-02 1.09229827e+00 -7.35831559e-02 -2.12674588e-01 2.84748971e-01 -3.77948672e-01 -3.00056487e-01 1.05584216e+00 -3.63809824e-01 -1.69800818e-01 2.81168759e-01 -9.40031409e-01 -3.62012625e-01 5.87030947e-01 5.67964464e-02 4.69559968e-01 -1.10223269e+00 -4.32413787e-01 -1.59514323e-02 2.87449658e-01 -5.04611194e-01 1.24960370e-01 8.58855724e-01 -8.10922623e-01 1.04847193e+00 1.34604611e-02 -3.98127943e-01 -1.69851685e+00 5.12370825e-01 5.50562441e-01 -4.42720234e-01 -3.23837012e-01 7.67947137e-01 4.46928382e-01 -1.90161288e-01 -7.55681843e-02 -1.49918929e-01 -2.74506837e-01 -1.80550724e-01 5.47851861e-01 2.46829331e-01 1.13911711e-01 -9.32657003e-01 -9.11259353e-01 1.13378477e+00 -3.62054139e-01 6.41959012e-01 1.51968646e+00 2.98389494e-01 -4.81263518e-01 -1.14837684e-01 9.69811738e-01 2.64165759e-01 -9.59765494e-01 -8.92516300e-02 1.46614656e-01 -2.94997483e-01 -4.83187169e-01 -8.53744686e-01 -3.54786307e-01 6.06587768e-01 5.64289153e-01 -3.81296873e-01 5.46324611e-01 2.84115523e-01 4.20667440e-01 9.00642574e-01 5.21572530e-01 -7.89703667e-01 5.26165776e-02 4.89978254e-01 8.17124665e-01 -1.32448447e+00 5.07750928e-01 -4.67004418e-01 -5.29625475e-01 1.08320343e+00 4.66253638e-01 1.74557343e-01 5.92465580e-01 1.43708456e-02 -4.82215106e-01 -3.96266878e-01 -8.66184473e-01 1.07454676e-02 5.21836698e-01 3.97536814e-01 1.14872563e+00 1.79442629e-01 -1.10040212e+00 6.10616207e-01 2.76727229e-01 -1.55527338e-01 1.86053410e-01 6.36555493e-01 -9.74854946e-01 -1.94811702e+00 -4.34031516e-01 2.30192821e-02 -7.16675937e-01 -2.87485808e-01 -1.03986847e+00 6.50600076e-01 -2.75162570e-02 6.34670913e-01 -6.35893643e-01 -2.12952364e-02 2.84010381e-01 1.00030795e-01 7.24642992e-01 -3.90882313e-01 -5.87017477e-01 2.82578856e-01 -8.95285159e-02 -4.63375151e-01 -4.64782804e-01 -2.06028610e-01 -1.78889179e+00 -6.24107361e-01 -6.95818663e-01 7.18630850e-01 4.21000749e-01 8.92621577e-01 5.64497173e-01 9.32591930e-02 3.52855384e-01 -7.16200948e-01 -2.85654128e-01 -6.69771969e-01 -6.99741244e-01 3.65183085e-01 2.55643010e-01 -9.01180387e-01 -6.21088967e-02 -1.64727509e-01]
[4.8114190101623535, 5.416579246520996]
6f1620a6-9ed8-4b36-abf7-abaa4562b1d4
factify3m-a-benchmark-for-multimodal-fact
2306.05523
null
https://arxiv.org/abs/2306.05523v1
https://arxiv.org/pdf/2306.05523v1.pdf
FACTIFY3M: A Benchmark for Multimodal Fact Verification with Explainability through 5W Question-Answering
Combating disinformation is one of the burning societal crises -- about 67% of the American population believes that disinformation produces a lot of uncertainty, and 10% of them knowingly propagate disinformation. Evidence shows that disinformation can manipulate democratic processes and public opinion, causing disruption in the share market, panic and anxiety in society, and even death during crises. Therefore, disinformation should be identified promptly and, if possible, mitigated. With approximately 3.2 billion images and 720,000 hours of video shared online daily on social media platforms, scalable detection of multimodal disinformation requires efficient fact verification. Despite progress in automatic text-based fact verification (e.g., FEVER, LIAR), the research community lacks substantial effort in multimodal fact verification. To address this gap, we introduce FACTIFY 3M, a dataset of 3 million samples that pushes the boundaries of the domain of fact verification via a multimodal fake news dataset, in addition to offering explainability through the concept of 5W question-answering. Salient features of the dataset include: (i) textual claims, (ii) ChatGPT-generated paraphrased claims, (iii) associated images, (iv) stable diffusion-generated additional images (i.e., visual paraphrases), (v) pixel-level image heatmap to foster image-text explainability of the claim, (vi) 5W QA pairs, and (vii) adversarial fake news stories.
['Amitava Das', 'Amit Sheth', 'Aman Chadha', 'Parth Patwa', 'Shreyash Mishra', 'Suryavardan S', 'Dwip Dalal', 'Kinjal Sensharma', 'Shreyas Chatterjee', 'Samahriti Mukherjee', 'Ritvik G', 'Preethi Gurumurthy', 'Janvita Reddy', 'Ishan Paul', 'Harshit Dave', 'Arghya Sarkar', 'Aditya Pakala', 'Adarsh Mahor', 'Anku Rani', 'Khusbu Pahwa', 'Megha Chakraborty']
2023-05-22
null
null
null
null
['fact-verification']
['natural-language-processing']
[ 3.44797254e-01 4.77557391e-01 -4.29969698e-01 -7.52420202e-02 -1.09581268e+00 -1.04636300e+00 9.65257585e-01 4.14302826e-01 2.87054088e-02 6.49892747e-01 9.05814111e-01 -6.56451941e-01 3.19214433e-01 -6.63577557e-01 -7.09059596e-01 -1.94654614e-01 3.55926603e-01 1.43402755e-01 -5.88309020e-02 -5.67276180e-01 6.10530138e-01 5.87710813e-02 -7.58387923e-01 8.10113311e-01 9.02032316e-01 9.47021365e-01 -5.91941833e-01 5.33874750e-01 -1.50689006e-01 1.39908218e+00 -8.48878384e-01 -1.32896030e+00 1.16914652e-01 -6.80492520e-01 -7.69690275e-01 3.29428554e-01 8.95530581e-01 -7.40715802e-01 -6.48500502e-01 1.36655283e+00 1.57104790e-01 -5.28689444e-01 5.17882466e-01 -1.43690073e+00 -1.29751337e+00 6.17381454e-01 -6.81462228e-01 4.32499856e-01 7.12988317e-01 5.22586823e-01 8.42847884e-01 -7.43646622e-01 1.36609554e+00 1.50678253e+00 6.63495243e-01 1.73756748e-01 -1.30177271e+00 -6.73273265e-01 -4.86242622e-01 1.55713245e-01 -1.17363501e+00 -4.29831415e-01 7.44037628e-01 -6.99711144e-01 4.83755976e-01 6.15081787e-01 9.52817321e-01 1.52231586e+00 3.63573879e-01 5.40460050e-01 1.46654570e+00 -7.64692994e-03 -1.46758437e-01 3.74559999e-01 -2.38736719e-01 8.91013563e-01 3.99130136e-01 -2.80881226e-01 -8.07598412e-01 -8.01201642e-01 3.67004305e-01 -3.64311971e-02 -4.52658564e-01 7.98516050e-02 -1.40847874e+00 1.20246160e+00 5.58178127e-01 1.05378643e-01 -3.39028001e-01 1.93248719e-01 4.40604419e-01 4.21517074e-01 8.20315003e-01 5.01714289e-01 2.62121260e-01 -2.64612526e-01 -8.06182683e-01 4.04017240e-01 6.24975145e-01 3.21741611e-01 5.67334294e-01 4.79046740e-02 -1.69120468e-02 4.53544199e-01 -1.64860666e-01 9.97925639e-01 1.41094416e-01 -1.04859650e+00 1.01384318e+00 8.10795784e-01 2.75192171e-01 -2.36605787e+00 -2.69757602e-02 -9.95705798e-02 -6.73623443e-01 -8.18278417e-02 3.70717525e-01 -8.14580619e-02 -5.23493111e-01 1.46684802e+00 2.86453605e-01 -2.05673665e-01 -1.33436978e-01 1.17426324e+00 6.46053791e-01 7.37974346e-01 -1.97375461e-01 1.33993300e-02 1.64952397e+00 -3.44489247e-01 -7.93016315e-01 -2.53842711e-01 3.14986348e-01 -1.00994170e+00 9.61543024e-01 1.30061299e-01 -1.21079695e+00 2.02209011e-01 -8.92062426e-01 -2.41258323e-01 -4.06350017e-01 -4.42834079e-01 3.98724347e-01 7.14950323e-01 -6.31264746e-01 7.69276917e-02 -1.29770204e-01 -3.94867361e-01 9.42945600e-01 -4.87110853e-01 -5.74896455e-01 -3.34978074e-01 -1.24375570e+00 8.15730155e-01 -7.14370012e-02 -6.62131011e-02 -8.45731318e-01 -6.23772204e-01 -8.14301372e-01 -3.10443550e-01 5.09180129e-01 -5.22913158e-01 8.59937012e-01 -1.35595655e+00 -7.05784678e-01 1.30837393e+00 1.39301345e-01 -4.63797480e-01 8.73345375e-01 -1.00776605e-01 -6.96725547e-01 9.22693551e-01 4.51596618e-01 6.97957158e-01 1.30314636e+00 -1.43049729e+00 -1.45731062e-01 -3.62046510e-01 1.72140092e-01 4.67007458e-02 -3.38612914e-01 2.55299658e-01 1.90055728e-01 -1.03448641e+00 2.67632693e-01 -7.91311324e-01 2.93825358e-01 1.76397383e-01 -9.71495271e-01 5.61324298e-01 1.12843060e+00 -1.29110336e+00 8.69486868e-01 -2.23397350e+00 -3.02926362e-01 1.22541726e-01 7.94764459e-01 1.68688092e-02 5.01205819e-03 6.78360224e-01 4.46722358e-02 7.24889219e-01 -1.80400312e-01 1.34537995e-01 -4.87656593e-02 -1.89293087e-01 -8.16585541e-01 1.02339339e+00 2.32735574e-02 1.23056686e+00 -1.03066421e+00 -4.27322030e-01 -1.24236256e-01 3.06492835e-01 -4.48349416e-01 -4.31873649e-01 -2.08050296e-01 3.67333919e-01 -3.21980894e-01 1.01330256e+00 7.65361190e-01 -4.20536816e-01 1.12820424e-01 -4.57588524e-01 -8.00713003e-02 1.58111647e-01 -3.72041404e-01 1.05605447e+00 1.28679261e-01 1.29279232e+00 2.48107076e-01 -4.92167413e-01 5.25836289e-01 2.67242163e-01 2.22907171e-01 -6.52360559e-01 2.55953223e-01 3.81233431e-02 -2.13974044e-01 -7.18115151e-01 7.70087600e-01 -1.60419881e-01 -3.81870389e-01 8.77614439e-01 -5.42799115e-01 -3.91645461e-01 -4.87313241e-01 8.10169578e-01 1.07163858e+00 -6.87017381e-01 3.58144054e-04 1.81574419e-01 -1.27111197e-01 7.35760152e-01 2.45245881e-02 8.35843146e-01 -4.53815609e-01 5.49763858e-01 8.16320419e-01 -6.34958625e-01 -1.32324862e+00 -1.05716383e+00 1.84498772e-01 5.43715298e-01 3.86782646e-01 -1.79145023e-01 -7.51378238e-01 -4.84137207e-01 3.28341514e-01 8.08531582e-01 -6.44442916e-01 -1.08901665e-01 -2.44519383e-01 -4.75436419e-01 9.41505373e-01 -3.17302525e-01 1.00509489e+00 -6.34172022e-01 -5.58742404e-01 -1.10493295e-01 -1.02834070e+00 -1.28573215e+00 -6.19597554e-01 -9.19095159e-01 -4.36927617e-01 -1.36585963e+00 -6.48389578e-01 -2.53084153e-01 6.59377635e-01 8.40354800e-01 1.01334345e+00 4.12310392e-01 -4.29012597e-01 6.09061599e-01 -1.82749838e-01 -4.38587405e-02 -7.40265489e-01 -6.22198820e-01 -1.46926627e-01 1.51820496e-01 -1.44693598e-01 -1.35111451e-01 -6.79915786e-01 2.24344745e-01 -1.17259121e+00 1.64282292e-01 1.95029899e-01 6.27164721e-01 -5.14683202e-02 -1.44665077e-01 4.30202276e-01 -8.65494013e-01 9.32223558e-01 -8.35526884e-01 -2.62586892e-01 2.21994027e-01 -3.11132193e-01 -7.89461434e-01 -3.97469802e-03 -3.51759642e-01 -1.01809251e+00 -7.07769096e-01 3.58228981e-01 -1.57627642e-01 2.24706575e-01 5.76141417e-01 4.08845305e-01 -9.86514837e-02 1.05248165e+00 7.90956169e-02 2.76336640e-01 1.29280046e-01 7.23158836e-01 6.96404457e-01 6.69286311e-01 6.44324720e-03 1.05830169e+00 1.06860447e+00 -5.34089506e-01 -7.86692083e-01 -1.08491063e+00 -6.60612062e-02 1.13372251e-01 -7.28948832e-01 9.00398552e-01 -9.71984684e-01 -9.42671895e-01 3.81224841e-01 -1.41030335e+00 1.13479227e-01 7.71389604e-02 7.62911662e-02 -1.63732097e-01 4.66950357e-01 -1.07467365e+00 -6.76985741e-01 -1.04751445e-01 -8.05354953e-01 7.47047603e-01 -2.31855720e-01 -4.91365939e-01 -6.95629835e-01 -2.43885458e-01 1.39696062e+00 4.85480130e-01 8.12544644e-01 9.77252543e-01 -4.55442935e-01 -7.90474892e-01 -4.26528126e-01 -7.75916159e-01 3.21244402e-03 -5.75581798e-04 -1.58357367e-01 -8.21823299e-01 2.93235816e-02 8.95470977e-02 -8.82842600e-01 6.47945762e-01 1.26948699e-01 6.40505135e-01 -1.17590451e+00 -8.20909515e-02 -1.34330317e-01 1.15110159e+00 -1.39053226e-01 7.57537007e-01 3.15586656e-01 5.90539932e-01 6.92618668e-01 3.97599429e-01 4.48539406e-01 6.80340767e-01 3.81238669e-01 7.01139867e-01 -4.64417003e-02 -1.49270445e-01 -7.03446805e-01 4.57496673e-01 4.82732683e-01 1.47431180e-01 -3.11536342e-01 -9.99626637e-01 5.23862064e-01 -1.54056776e+00 -1.68081915e+00 -4.53547776e-01 1.66439509e+00 5.92228115e-01 8.99586082e-02 1.96922749e-01 -1.03816636e-01 8.91588449e-01 5.57887018e-01 -4.49787915e-01 -3.13836038e-01 -6.04439080e-01 -6.48686588e-01 5.20749152e-01 7.30768681e-01 -8.02780926e-01 9.01179671e-01 6.15246964e+00 7.84243345e-01 -9.95833516e-01 5.08527100e-01 1.19744027e+00 4.81495038e-02 -9.83350098e-01 -4.23963815e-02 1.28667802e-01 4.61519957e-01 2.86755264e-01 -2.70085670e-02 4.99835014e-01 4.48019713e-01 3.56678218e-01 -4.58116829e-01 -1.76979393e-01 9.43057954e-01 6.88209951e-01 -2.11947250e+00 1.65629059e-01 1.28983349e-01 1.04619086e+00 -2.09594548e-01 5.12676597e-01 -2.51940310e-01 1.56396478e-01 -1.01347840e+00 1.28867841e+00 3.92331302e-01 9.81431007e-01 -5.91623008e-01 5.52163064e-01 1.45650819e-01 -3.60226214e-01 -9.02884454e-02 -7.14704543e-02 -7.98873678e-02 5.74309707e-01 8.78327727e-01 -7.62690902e-01 1.28622860e-01 4.41344768e-01 5.22731185e-01 -6.22093081e-01 3.96678120e-01 -3.48629147e-01 6.64041758e-01 8.59179720e-02 6.14833273e-02 3.76237661e-01 -7.15670288e-02 9.30759728e-01 9.72501636e-01 2.81912833e-01 3.58529478e-01 -3.19523871e-01 1.01552665e+00 -4.63106424e-01 -2.99746454e-01 -8.85540843e-01 -7.05882549e-01 3.81141722e-01 1.04107225e+00 -8.06695521e-01 -4.00301844e-01 -2.50932604e-01 1.25414622e+00 7.41487294e-02 3.72901559e-01 -1.03044248e+00 8.01271647e-02 4.43370849e-01 2.82657743e-01 -1.58023149e-01 -2.48219788e-01 -3.70530933e-01 -1.30390620e+00 -1.59558579e-01 -1.27379262e+00 2.13533059e-01 -1.29344046e+00 -1.46400940e+00 4.26237792e-01 -3.80251467e-01 -7.99429893e-01 1.64381117e-01 -1.34222806e-01 -3.21991384e-01 4.39454496e-01 -1.12154984e+00 -1.10003221e+00 -3.67662877e-01 5.36661267e-01 2.09220663e-01 5.26632257e-02 4.07084912e-01 1.81584671e-01 -2.06561178e-01 1.93303898e-01 -1.73349336e-01 1.07108191e-01 7.08972692e-01 -5.91888011e-01 3.82811397e-01 4.54038978e-01 6.06927462e-02 3.25013727e-01 9.83607233e-01 -1.20559692e+00 -1.40717852e+00 -8.14585090e-01 9.72220421e-01 -7.89655626e-01 1.28139448e+00 -3.05230498e-01 -6.93891346e-01 6.69155300e-01 4.62268978e-01 -4.08629864e-01 7.01858163e-01 -2.90502608e-01 -7.35052764e-01 4.18693662e-01 -1.65831935e+00 9.22202528e-01 8.57420743e-01 -1.03398192e+00 -5.35289586e-01 9.46153700e-01 7.60638654e-01 -3.15047741e-01 -4.32541758e-01 -4.13170725e-01 7.31977761e-01 -1.22766030e+00 9.57105398e-01 -6.33737028e-01 1.15464532e+00 3.53802405e-02 -1.84587032e-01 -1.06092250e+00 5.25828674e-02 -7.61408627e-01 1.48592949e-01 1.14501452e+00 4.51697767e-01 -6.11636341e-01 6.74067438e-01 8.75836790e-01 1.75184980e-01 -3.34103942e-01 -9.33674037e-01 -3.67539018e-01 -3.40245098e-01 -4.43302780e-01 2.00005129e-01 1.66709960e+00 2.12135434e-01 1.14535466e-01 -7.32727945e-01 9.86942798e-02 5.49005568e-01 1.77308753e-01 7.29993463e-01 -5.36229789e-01 3.35501693e-02 -3.06169808e-01 -3.32982153e-01 -5.81452310e-01 -1.91967741e-01 -6.74660861e-01 -4.80471760e-01 -1.35484111e+00 7.95935154e-01 -2.22186781e-02 7.17016160e-01 4.50021118e-01 1.48586228e-01 7.83943653e-01 3.31948787e-01 6.68429971e-01 -5.06533146e-01 2.38591015e-01 1.46061516e+00 -5.34862936e-01 4.22726721e-01 -6.02428675e-01 -9.02797997e-01 8.06209385e-01 6.54952168e-01 -6.00134850e-01 -1.42151460e-01 -3.04935992e-01 9.61234927e-01 3.77138346e-01 1.15356755e+00 -3.12579364e-01 -7.44435787e-02 -4.11244571e-01 2.85282850e-01 -5.37839293e-01 5.82463801e-01 -6.39045417e-01 3.27960789e-01 8.20327520e-01 -3.36953551e-01 3.59918952e-01 4.16334607e-02 7.69542456e-01 -3.69547725e-01 7.86024034e-02 5.30587018e-01 -3.11037362e-01 -9.24251005e-02 -2.04329267e-01 -7.35685766e-01 3.51031780e-01 8.46027017e-01 -1.84442848e-01 -1.45109618e+00 -1.33273911e+00 -3.28240424e-01 -1.03793226e-01 6.95560575e-01 1.42302856e-01 7.93462157e-01 -1.39932084e+00 -8.89781713e-01 -3.94441396e-01 1.91407874e-01 -8.11381936e-01 6.23490214e-01 1.11585867e+00 -5.77264190e-01 1.47871211e-01 -3.86551842e-02 -1.45952761e-01 -1.06189871e+00 4.68379200e-01 1.66240502e-02 2.34239340e-01 -6.03222609e-01 7.44114161e-01 -6.01800457e-02 -1.30945034e-02 -4.09904242e-01 -1.53079554e-02 3.47194999e-01 2.31474087e-01 6.60458505e-01 4.02764082e-01 -5.19710064e-01 -1.06060624e+00 -3.17355454e-01 1.50784506e-02 1.21097706e-01 -4.35027361e-01 9.88183022e-01 -4.41005856e-01 -2.84626126e-01 1.43200368e-01 1.22325373e+00 5.77075303e-01 -9.41948116e-01 1.19267181e-01 -1.52409866e-01 -1.11057436e+00 -2.76461571e-01 -1.15985167e+00 -8.79638493e-01 5.10893464e-01 1.08853631e-01 8.46409380e-01 5.80096960e-01 2.67490089e-01 1.25087440e+00 -7.12658465e-03 8.34908932e-02 -7.87717104e-01 7.50125766e-01 7.32507184e-02 1.29258597e+00 -1.43399739e+00 1.43678695e-01 -5.68282843e-01 -1.05930579e+00 8.83274972e-01 2.02380139e-02 1.82441652e-01 1.63420796e-01 -2.38072917e-01 1.95702657e-01 -8.78531158e-01 -3.92267764e-01 4.45151240e-01 1.54971212e-01 3.55395168e-01 -6.26782998e-02 1.60170123e-01 -1.84796795e-01 3.46972793e-01 -2.35654220e-01 -4.35986608e-01 8.37799489e-01 6.32984996e-01 -4.09759045e-01 -2.41276905e-01 -8.67159605e-01 3.82075757e-01 -6.06159806e-01 -2.29833677e-01 -9.78467286e-01 7.48862743e-01 3.71127874e-02 1.23806143e+00 -1.00926973e-01 -3.36746365e-01 5.10062790e-03 -3.35037738e-01 1.25290856e-01 -6.91105723e-02 -7.10450113e-01 -3.44704092e-01 7.34707475e-01 -6.57779276e-01 -4.10470635e-01 -5.68240881e-01 -7.78105557e-01 -1.14108801e+00 -1.02039628e-01 -2.20931157e-01 7.64644623e-01 8.91868889e-01 6.20755732e-01 -2.72486359e-01 3.71120304e-01 -3.59920174e-01 -1.92221761e-01 -6.25887275e-01 -3.21716785e-01 7.05757320e-01 5.73255360e-01 -3.79282296e-01 -7.07048297e-01 1.53668061e-01]
[8.142041206359863, 10.252302169799805]
007956d2-8408-414f-bacb-43a1416675f6
probabilistic-time-series-forecasting-with-1
null
null
http://proceedings.neurips.cc/paper/2020/hash/2f2b265625d76a6704b08093c652fd79-Abstract.html
http://proceedings.neurips.cc/paper/2020/file/2f2b265625d76a6704b08093c652fd79-Paper.pdf
Probabilistic Time Series Forecasting with Shape and Temporal Diversity
Probabilistic forecasting consists in predicting a distribution of possible future outcomes. In this paper, we address this problem for non-stationary time series, which is very challenging yet crucially important. We introduce the STRIPE model for representing structured diversity based on shape and time features, ensuring both probable predictions while being sharp and accurate. STRIPE is agnostic to the forecasting model, and we equip it with a diversification mechanism relying on determinantal point processes (DPP). We introduce two DPP kernels for modelling diverse trajectories in terms of shape and time, which are both differentiable and proved to be positive semi-definite. To have an explicit control on the diversity structure, we also design an iterative sampling mechanism to disentangle shape and time representations in the latent space. Experiments carried out on synthetic datasets show that STRIPE significantly outperforms baseline methods for representing diversity, while maintaining accuracy of the forecasting model. We also highlight the relevance of the iterative sampling scheme and the importance to use different criteria for measuring quality and diversity. Finally, experiments on real datasets illustrate that STRIPE is able to outperform state-of-the-art probabilistic forecasting approaches in the best sample prediction.
['Nicolas Thome', 'Vincent Le Guen']
2020-12-01
null
null
null
neurips-2020-12
['probabilistic-time-series-forecasting']
['time-series']
[-1.17402412e-01 -1.99152544e-01 -1.02071129e-01 -2.78408796e-01 -7.21594274e-01 -7.87662923e-01 1.23916423e+00 2.02439595e-02 1.84475631e-01 7.30355740e-01 3.68551582e-01 -2.85354167e-01 -5.88702381e-01 -7.63026595e-01 -5.41507363e-01 -1.00799012e+00 -4.33691829e-01 5.96590400e-01 2.26862967e-01 -1.55787632e-01 5.00244558e-01 6.42623901e-01 -1.79285097e+00 1.35384649e-01 9.86483395e-01 1.11299217e+00 -1.17114551e-01 6.51913226e-01 6.51348159e-02 7.69262612e-01 -3.59153569e-01 -4.95312870e-01 1.40582249e-01 -1.99544966e-01 -3.54790181e-01 4.14487235e-02 4.28251028e-02 2.35440329e-01 1.82336777e-01 5.56587756e-01 2.70031452e-01 -6.22514226e-02 1.38393879e+00 -1.44309556e+00 -4.39606845e-01 3.44354510e-01 -1.94284618e-01 1.78117961e-01 2.11061686e-01 -7.92284682e-03 1.04741108e+00 -8.50094676e-01 4.12114650e-01 1.06716192e+00 1.09638524e+00 1.17651530e-01 -1.56553936e+00 -3.55702639e-01 2.86558926e-01 7.77368695e-02 -1.31882370e+00 -3.50445718e-01 7.85090923e-01 -9.55790937e-01 5.44288039e-01 4.82968509e-01 4.81512606e-01 1.26384938e+00 6.23227417e-01 6.67415380e-01 1.06970751e+00 -2.01608598e-01 6.53508127e-01 1.07522860e-01 1.65384486e-01 1.95727408e-01 8.21573138e-02 3.14619303e-01 -3.84009421e-01 -7.58888960e-01 5.80685914e-01 1.36219412e-01 -3.06519598e-01 -7.06340969e-01 -1.38003612e+00 9.55343246e-01 -2.37120211e-01 3.81838590e-01 -7.66422033e-01 -1.30393043e-01 2.14716464e-01 2.35080570e-01 9.03775871e-01 2.45608091e-01 -5.10969877e-01 -4.11196947e-01 -1.15068829e+00 5.05444050e-01 1.06935787e+00 6.91385686e-01 2.90438890e-01 7.41026774e-02 -5.97176909e-01 5.54545879e-01 1.79553971e-01 5.89046001e-01 2.93017447e-01 -8.54853988e-01 3.09302688e-01 3.41644645e-01 5.73601782e-01 -1.05354822e+00 -4.45156157e-01 -6.37451231e-01 -1.08145499e+00 -6.87151449e-03 3.48604977e-01 4.71554371e-03 -6.87986493e-01 1.71706295e+00 3.21758389e-01 5.03239453e-01 4.32361104e-03 5.27898967e-01 6.99222535e-02 9.33450282e-01 1.21530980e-01 -4.21932608e-01 9.72631633e-01 -5.18219292e-01 -3.33660811e-01 4.78468716e-01 4.90210980e-01 -5.56948662e-01 8.33405137e-01 5.37738204e-01 -1.03547740e+00 -4.24826771e-01 -5.63243330e-01 4.74754155e-01 -9.92281437e-02 2.77317286e-01 4.81157720e-01 7.48286605e-01 -1.15473568e+00 7.78417289e-01 -9.48570728e-01 -3.08821276e-02 3.79373170e-02 8.83773491e-02 1.45919723e-02 3.30089450e-01 -1.22192252e+00 7.46834099e-01 1.59034759e-01 -9.29515511e-02 -5.24472594e-01 -1.04815495e+00 -5.82191110e-01 3.30085754e-01 -1.87236533e-01 -7.76641488e-01 1.00346947e+00 -6.73496485e-01 -1.48674679e+00 2.53168225e-01 -1.85844392e-01 -5.57381868e-01 8.21359515e-01 2.12588400e-01 -6.72435760e-01 9.07673612e-02 -7.64966309e-02 3.26725215e-01 1.15864062e+00 -1.20720410e+00 -6.47009790e-01 -1.36348575e-01 -3.95850688e-01 -2.41046324e-01 -2.12058410e-01 -3.15986991e-01 1.54414251e-01 -8.45179677e-01 -4.85526696e-02 -1.01630974e+00 -3.18122804e-01 -3.48598897e-01 -1.98118493e-01 -3.36970836e-01 5.34944415e-01 -6.29143238e-01 1.52712798e+00 -1.96261764e+00 2.63278782e-01 4.33462650e-01 3.54401879e-02 -6.84363693e-02 1.94950104e-02 9.21381593e-01 1.02251276e-01 1.46086765e-02 -4.50095236e-01 -5.29547036e-01 2.89164662e-01 2.52919316e-01 -1.00805652e+00 5.93980074e-01 4.34616596e-01 7.92552769e-01 -7.37250268e-01 -3.61285619e-02 3.96478213e-02 6.98820829e-01 -3.41834843e-01 -2.99275923e-03 -3.34827930e-01 6.10557437e-01 -4.10555929e-01 5.69640219e-01 6.74379587e-01 -3.32237929e-01 5.90293817e-02 2.93645114e-01 -3.85925502e-01 1.52869701e-01 -1.14630651e+00 8.50166619e-01 -3.69108886e-01 3.06980699e-01 -4.48613971e-01 -9.70677018e-01 1.15941453e+00 4.01692033e-01 4.93712157e-01 -4.13003236e-01 -1.69467747e-01 3.29554528e-01 -3.85154128e-01 4.11877781e-03 6.05263352e-01 -1.40289009e-01 -9.89668444e-02 4.41859990e-01 -2.49520585e-01 -7.29619642e-04 1.36222959e-01 -1.63687453e-01 8.40651989e-01 8.94885287e-02 7.23430365e-02 -6.91625655e-01 5.61072946e-01 -1.49759576e-01 5.25120378e-01 6.17986441e-01 1.34431332e-01 6.15599096e-01 8.74640405e-01 -4.92942512e-01 -1.22054398e+00 -1.07856071e+00 -4.13376182e-01 9.25753713e-01 -1.27586916e-01 -1.91768363e-01 -3.05955648e-01 -4.20906693e-01 2.77713209e-01 1.13666713e+00 -9.78945076e-01 -3.43330577e-02 -2.83011645e-01 -9.42749321e-01 2.35630423e-01 3.94511282e-01 -2.41994575e-01 -7.34409153e-01 -6.13143742e-01 3.78774703e-01 -2.69049462e-02 -6.79003060e-01 -2.78396130e-01 7.42388749e-03 -1.02169502e+00 -7.41089165e-01 -1.03735697e+00 -1.92345902e-01 4.07666773e-01 9.09343138e-02 1.29940236e+00 -5.68202615e-01 2.80704439e-01 7.25680292e-01 -2.89718479e-01 -2.85092264e-01 -5.52953959e-01 1.90890372e-01 -5.90036958e-02 3.66213053e-01 1.45246476e-01 -9.37600017e-01 -6.76282167e-01 5.00955880e-01 -1.06931651e+00 6.81656674e-02 2.89341003e-01 8.45052958e-01 4.37333286e-01 6.76708370e-02 7.22779512e-01 -5.93239248e-01 8.58822703e-01 -9.49464679e-01 -6.27608657e-01 4.50857490e-01 -8.88433039e-01 2.96379566e-01 7.01929450e-01 -7.03245819e-01 -8.41543734e-01 -7.43103549e-02 7.38727227e-02 -4.76111054e-01 2.77979132e-02 5.64009011e-01 3.53001833e-01 2.31666490e-01 4.35660869e-01 5.48143327e-01 -8.19550753e-02 -3.96730900e-01 3.68197650e-01 3.52944463e-01 1.36888355e-01 -6.52246535e-01 3.86202693e-01 6.44501567e-01 3.26058865e-01 -6.33685410e-01 -5.11040986e-01 -4.06273901e-01 -6.32764399e-01 -1.95106998e-01 3.13547105e-01 -7.38151908e-01 -6.63717330e-01 4.35452849e-01 -9.61744428e-01 -1.55436248e-01 -3.30875605e-01 3.86508495e-01 -8.39779735e-01 1.85886741e-01 -4.67136681e-01 -1.23212206e+00 -1.38065785e-01 -9.25800085e-01 1.40477061e+00 -2.98530012e-01 -2.22560480e-01 -1.44349444e+00 5.53436160e-01 -3.99842471e-01 6.71591580e-01 4.81155038e-01 1.02781928e+00 -6.95699692e-01 -3.65760714e-01 -1.89128220e-01 5.96983992e-02 1.16276830e-01 -1.76992312e-01 4.44297820e-01 -7.59971559e-01 -8.24519619e-02 1.11549176e-01 3.03115308e-01 1.12205184e+00 5.51238179e-01 8.69016945e-01 -3.47829282e-01 -3.30328584e-01 3.66406292e-01 1.13699996e+00 2.04764921e-02 5.59754312e-01 1.64608791e-01 4.59449351e-01 8.87915432e-01 4.32999492e-01 9.50590491e-01 6.00133419e-01 7.04192519e-01 1.61821172e-01 4.50487107e-01 3.24712455e-01 -2.22428441e-01 4.89436090e-01 9.03524816e-01 -7.28649870e-02 -1.99725389e-01 -1.04841840e+00 5.72365105e-01 -2.08897519e+00 -1.28083253e+00 -3.69641036e-01 2.38282967e+00 6.90707028e-01 1.19125452e-02 4.98839706e-01 2.68433034e-01 5.42051494e-01 1.50275499e-01 -3.70315582e-01 -4.32598591e-01 -1.79205060e-01 -1.77825630e-01 3.85091960e-01 5.28238118e-01 -1.14051640e+00 3.93707484e-01 6.59308910e+00 9.60007191e-01 -1.31772864e+00 -1.17362507e-01 8.84307206e-01 1.99042395e-01 -8.51035774e-01 -1.11082584e-01 -9.98024285e-01 9.22652185e-01 1.25124419e+00 -2.44300485e-01 5.42307422e-02 7.53356636e-01 1.79394424e-01 1.44364282e-01 -1.02231610e+00 6.64156616e-01 -2.15568751e-01 -1.42199063e+00 9.17316079e-02 2.21033003e-02 9.47523594e-01 -1.89881742e-01 3.96604270e-01 3.73286456e-01 9.05990750e-02 -9.67887163e-01 9.86405015e-01 1.27988172e+00 2.90888757e-01 -9.50791895e-01 4.72057998e-01 6.86766922e-01 -1.14917052e+00 -1.29837051e-01 -2.69564271e-01 -1.28528640e-01 3.08675408e-01 1.07394016e+00 -6.14278376e-01 6.13809884e-01 3.00404221e-01 7.88385332e-01 -3.42230588e-01 1.04996014e+00 9.79175493e-02 8.99865508e-01 -3.58576566e-01 -9.19722989e-02 2.53054351e-01 -3.66303623e-01 6.57192707e-01 1.36544704e+00 9.30794656e-01 -2.68309563e-01 3.90133634e-02 8.37150455e-01 6.27885699e-01 -1.07616082e-01 -4.76263970e-01 -3.13178822e-02 4.04029876e-01 8.31257224e-01 -5.93208611e-01 -2.28357658e-01 -2.34682765e-02 7.33298957e-01 4.81583849e-02 3.77314359e-01 -7.61705816e-01 1.28291845e-01 5.58554351e-01 1.16824262e-01 7.63295889e-01 -4.95244980e-01 -4.97857034e-01 -1.21203780e+00 6.95042536e-02 -5.30433536e-01 1.70619443e-01 -3.83382112e-01 -1.72475481e+00 7.65247583e-01 1.40057087e-01 -1.52212989e+00 -8.04526329e-01 -3.95317405e-01 -6.53159499e-01 1.04452717e+00 -1.50949168e+00 -1.21319199e+00 2.08501175e-01 4.44671720e-01 2.28057086e-01 -1.02535617e-02 8.21780026e-01 -3.80208455e-02 -3.23964447e-01 2.47202799e-01 5.91576636e-01 -8.33469868e-01 2.64482111e-01 -1.28757203e+00 7.02118874e-01 5.17360508e-01 4.63020131e-02 5.28093696e-01 9.20426130e-01 -6.43019378e-01 -1.26991117e+00 -1.12976265e+00 1.20511103e+00 -7.59036839e-01 9.55728471e-01 -3.43897402e-01 -1.08771372e+00 3.12727064e-01 -3.06771696e-01 -2.01755747e-01 6.91613197e-01 1.57775953e-01 -3.77293795e-01 9.16277617e-02 -1.02306604e+00 5.24250329e-01 6.61540210e-01 -2.91907758e-01 -4.76108223e-01 2.48026505e-01 4.26556766e-01 1.16331112e-02 -9.57651198e-01 6.11758351e-01 7.18942523e-01 -1.21351504e+00 1.02926075e+00 -3.86463642e-01 4.43857610e-01 -1.92243293e-01 -1.56040832e-01 -1.28851926e+00 -6.74095929e-01 -6.67647839e-01 -5.41022122e-01 1.29857004e+00 3.49822849e-01 -9.47009265e-01 4.93626386e-01 6.71247959e-01 1.58828527e-01 -1.17463779e+00 -9.55465317e-01 -1.01667511e+00 2.75474340e-01 -6.29155993e-01 8.21940124e-01 8.80257607e-01 -2.07496077e-01 -3.04384436e-02 -6.82011127e-01 3.34127754e-01 4.77329195e-01 4.99476194e-01 5.63094914e-01 -1.65790462e+00 -2.67588645e-01 -7.55990028e-01 -3.84978563e-01 -8.78044963e-01 1.30961567e-01 -5.06322265e-01 -2.99183637e-01 -1.14266944e+00 -7.86965042e-02 -6.48625016e-01 -3.74997854e-01 -1.91650260e-02 -9.67280492e-02 3.59685533e-02 2.69129104e-03 6.82190776e-01 -4.09269869e-01 9.71403956e-01 7.97623575e-01 2.67111540e-01 -3.30141038e-01 5.60055733e-01 -5.31800568e-01 4.58552718e-01 8.46355796e-01 -3.55846912e-01 -5.56424439e-01 6.02785870e-02 3.63985360e-01 3.47656310e-01 5.73261976e-01 -8.82092178e-01 1.72872737e-01 -2.64305413e-01 2.04073265e-01 -8.23367238e-01 4.24085915e-01 -6.92212403e-01 5.62498868e-01 3.49273682e-01 -4.17477041e-01 1.65918127e-01 8.63813385e-02 7.87315786e-01 -2.44878158e-01 2.50634179e-02 4.52935904e-01 4.15178746e-01 -3.43133152e-01 3.74041170e-01 -3.77913266e-01 -2.11479157e-01 1.09712291e+00 -1.26123771e-01 -3.81463170e-02 -6.14116490e-01 -6.19947910e-01 1.93757042e-01 5.08401096e-01 2.34505326e-01 2.06908777e-01 -1.37201869e+00 -8.57378960e-01 2.35657349e-01 1.04960285e-01 -7.37220824e-01 3.11899602e-01 1.01781797e+00 -2.20121816e-01 6.72310829e-01 1.01252757e-01 -7.75363326e-01 -7.34719038e-01 6.42678678e-01 2.02193949e-02 -5.04894078e-01 -4.77180570e-01 3.48885894e-01 1.45640880e-01 -3.45917344e-01 2.72824436e-01 -6.78242624e-01 -2.89626926e-01 3.81512046e-01 4.93860930e-01 6.31599486e-01 -4.34613004e-02 -6.36163771e-01 -2.60597646e-01 5.81083179e-01 2.02934608e-01 -1.96558028e-01 1.36913681e+00 -2.15063453e-01 4.80069965e-03 8.31665337e-01 8.89349699e-01 6.86334521e-02 -1.76487541e+00 -5.15713450e-03 3.30412000e-01 -3.26638639e-01 -2.20978156e-01 -7.10552394e-01 -4.91371453e-01 8.48409235e-01 3.26856971e-01 1.03938043e+00 9.65923131e-01 -3.00747991e-01 6.31609321e-01 5.89998486e-03 4.79948074e-01 -5.68276167e-01 -5.04125357e-01 7.19177365e-01 1.06540310e+00 -8.86585176e-01 -2.68285394e-01 -2.96348214e-01 -8.59934151e-01 1.24058270e+00 -3.83435309e-01 -2.77225643e-01 9.59240198e-01 2.43707206e-02 -1.99203134e-01 6.26760945e-02 -1.10374069e+00 8.96634981e-02 8.01401436e-01 3.90563995e-01 3.35207909e-01 3.29258800e-01 -4.72093284e-01 7.11140752e-01 -2.82691121e-01 -4.20822650e-02 1.12493470e-01 5.77295005e-01 -3.41186792e-01 -1.17456448e+00 -3.62814814e-01 3.66623491e-01 -3.44818473e-01 5.79003580e-02 -1.52009325e-02 5.71841240e-01 -9.76353288e-02 7.11466610e-01 7.69506693e-02 -2.77579427e-01 1.02594443e-01 1.97818577e-01 1.58125833e-01 -5.59933409e-02 -4.28753495e-01 2.43668575e-02 -3.76757570e-02 -2.45452777e-01 -2.88942367e-01 -1.03801572e+00 -4.83507603e-01 -3.72911155e-01 3.01638376e-02 3.39440554e-01 6.50729954e-01 8.58218551e-01 8.21821332e-01 2.12703690e-01 9.18611526e-01 -1.06243610e+00 -1.03307307e+00 -7.26142764e-01 -7.70450890e-01 2.34391332e-01 3.79708409e-01 -8.57783675e-01 -5.11226237e-01 -1.27486454e-03]
[7.020274639129639, 3.4163975715637207]
a9c1d0f0-4595-4848-a602-f1a6b4375cd4
a-database-for-face-presentation-attack-using
1906.11900
null
https://arxiv.org/abs/1906.11900v1
https://arxiv.org/pdf/1906.11900v1.pdf
A database for face presentation attack using wax figure faces
Compared to 2D face presentation attacks (e.g. printed photos and video replays), 3D type attacks are more challenging to face recognition systems (FRS) by presenting 3D characteristics or materials similar to real faces. Existing 3D face spoofing databases, however, mostly based on 3D masks, are restricted to small data size or poor authenticity due to the production difficulty and high cost. In this work, we introduce the first wax figure face database, WFFD, as one type of super-realistic 3D presentation attacks to spoof the FRS. This database consists of 2200 images with both real and wax figure faces (totally 4400 faces) with a high diversity from online collections. Experiments on this database first investigate the vulnerability of three popular FRS to this kind of new attack. Further, we evaluate the performance of several face presentation attack detection methods to show the attack abilities of this super-realistic face spoofing database.
['Zhengquan Xu', 'Guodong Guo', 'Shan Jia', 'Chuanbo Hu']
2019-06-06
null
null
null
null
['face-presentation-attack-detection', 'small-data']
['computer-vision', 'computer-vision']
[ 2.50297546e-01 -2.26316735e-01 1.31449163e-01 -1.17463671e-01 -1.64686099e-01 -8.27202141e-01 7.99711108e-01 -4.91312742e-01 1.31475747e-01 3.74953568e-01 -9.53007787e-02 -5.08086562e-01 -3.86569910e-02 -6.59333050e-01 -5.62898755e-01 -6.75498784e-01 -6.46034300e-01 -6.27875626e-02 2.35008821e-01 -3.46965313e-01 2.62968034e-01 1.23733139e+00 -1.74148154e+00 2.65912116e-01 1.99543595e-01 8.92597497e-01 -2.83629179e-01 5.33699453e-01 4.87108119e-02 -1.75293982e-01 -1.34156120e+00 -9.67698514e-01 6.03367150e-01 -1.58188134e-01 -2.84863532e-01 1.22205056e-01 7.16682315e-01 -6.76441789e-01 -6.05550945e-01 1.17276955e+00 7.26526558e-01 -4.87495273e-01 5.39077044e-01 -1.71045959e+00 -8.37498844e-01 1.40649125e-01 -7.50196993e-01 3.62741679e-01 1.09332013e+00 1.33573025e-01 -1.67652339e-01 -8.45106423e-01 6.37232721e-01 2.06095409e+00 4.67393368e-01 1.07810235e+00 -8.63777339e-01 -1.42803681e+00 -2.82644600e-01 1.30968109e-01 -1.66149533e+00 -9.27375972e-01 1.01277328e+00 -2.50906825e-01 3.54415208e-01 3.36262107e-01 3.10597718e-01 1.70265687e+00 4.54430468e-02 1.57662198e-01 1.45615900e+00 -3.18522006e-01 -1.23847611e-01 4.48031455e-01 -3.51999402e-02 6.40939057e-01 6.32383227e-01 3.16620618e-01 -5.63995302e-01 -5.75745344e-01 6.31808579e-01 -1.92371026e-01 -5.83497822e-01 6.20059296e-02 -5.80310225e-01 6.61821544e-01 3.03743221e-02 2.92760283e-01 1.15817837e-01 -3.90295744e-01 2.44547069e-01 5.40324032e-01 2.22337708e-01 -3.04038990e-02 -1.95104640e-03 3.00699890e-01 -5.38547099e-01 2.37192258e-01 8.73330832e-01 8.34640980e-01 4.23145503e-01 2.24100515e-01 1.21581376e-01 7.25501835e-01 6.44261539e-01 1.16173065e+00 3.28572452e-01 -1.97913930e-01 4.71795171e-01 6.92259893e-02 -5.60154505e-02 -1.70985627e+00 -2.14517792e-03 8.09631646e-02 -6.50855958e-01 2.69083321e-01 2.24900708e-01 1.00498356e-01 -8.65544379e-01 1.58911717e+00 4.20094788e-01 2.27473423e-01 1.98731124e-01 4.95000154e-01 1.06760657e+00 4.98482555e-01 -2.49363735e-01 -3.93072248e-01 1.52365994e+00 -1.32537141e-01 -8.97036493e-01 4.37667459e-01 2.65129268e-01 -1.15404105e+00 8.60082507e-01 3.98943663e-01 -6.83648825e-01 -4.94352549e-01 -1.16930163e+00 5.30194998e-01 -6.76949978e-01 -1.55942217e-01 2.68268943e-01 2.15963125e+00 -1.12047148e+00 2.85140485e-01 -6.92126751e-02 -2.79629946e-01 7.46316493e-01 5.43317795e-01 -9.45691466e-01 -2.57763177e-01 -1.39283276e+00 6.55371487e-01 5.18240370e-02 1.42176514e-02 -1.17070687e+00 -5.22620261e-01 -7.03235924e-01 -2.01975822e-01 1.11032203e-01 1.98340446e-01 6.31476223e-01 -5.00179350e-01 -1.49231780e+00 1.27926719e+00 7.70555669e-03 -2.09994286e-01 4.59314018e-01 5.72876036e-02 -1.27241969e+00 4.68189061e-01 -2.79805332e-01 3.27337921e-01 1.57294941e+00 -1.35024357e+00 2.43266881e-01 -7.32795775e-01 1.09648012e-01 -6.19240403e-01 -6.73226774e-01 6.71139777e-01 5.52878305e-02 -6.99739397e-01 -8.61702636e-02 -7.13713050e-01 6.37254417e-01 1.79204017e-01 -5.17467618e-01 -1.22509710e-01 1.74979925e+00 -6.69991136e-01 1.00001800e+00 -2.46473336e+00 -5.02528608e-01 3.91195297e-01 -8.68081972e-02 1.02953959e+00 -3.10015768e-01 5.76009512e-01 -3.80444437e-01 7.68544018e-01 1.06578127e-01 -2.85318732e-01 9.70563572e-03 3.66255082e-02 -4.79001731e-01 9.42302823e-01 3.64868253e-01 2.53713191e-01 -5.73968947e-01 -4.33971584e-01 -9.52448975e-03 8.05865884e-01 -3.73315156e-01 1.30479351e-01 2.96416163e-01 2.22255453e-01 -3.76030326e-01 9.79367614e-01 1.65179598e+00 5.42554140e-01 -4.99080261e-03 -7.73031190e-02 3.49864990e-01 -8.87050703e-02 -1.18023944e+00 9.04206395e-01 2.60810554e-03 6.27658129e-01 2.74490893e-01 -4.99180526e-01 1.25451934e+00 4.97463971e-01 9.21464935e-02 -2.73914188e-01 2.51559407e-01 2.53617138e-01 -6.30302429e-02 -7.71250129e-01 1.71716101e-02 2.18483046e-01 2.86727756e-01 5.04261434e-01 8.33421722e-02 -5.42043559e-02 9.94833931e-03 5.68440370e-02 9.34962988e-01 -3.42451751e-01 -1.70297995e-01 -3.97001863e-01 9.75849986e-01 -9.08390403e-01 2.92690814e-01 4.75671440e-01 -5.07633150e-01 3.35145742e-01 5.74229538e-01 -3.23461592e-01 -6.39260054e-01 -9.98574972e-01 -4.69300121e-01 1.61919624e-01 2.76340395e-01 -5.98565757e-01 -7.95183182e-01 -1.07925427e+00 3.01838934e-01 7.47607276e-02 -5.26303530e-01 -2.31613562e-01 -6.47246361e-01 -5.72333813e-01 1.35735059e+00 -2.78215915e-01 8.49010170e-01 -7.45700121e-01 -2.41089180e-01 -3.94027710e-01 2.32988343e-01 -1.33626091e+00 -3.28370869e-01 -8.86657894e-01 -3.77034545e-01 -1.46137524e+00 -8.45307171e-01 -7.45689571e-01 7.73344994e-01 6.80528879e-01 5.91676116e-01 5.27600110e-01 -4.40580606e-01 2.76605159e-01 -4.24922526e-01 -5.64399123e-01 -9.24798012e-01 -6.45250559e-01 7.82206893e-01 4.72961277e-01 3.08947653e-01 -3.94927204e-01 -2.69601554e-01 8.21642637e-01 -1.13053870e+00 -6.91069186e-01 1.86395094e-01 3.87196511e-01 -4.42932874e-01 4.10467088e-01 3.63512039e-01 -6.76619172e-01 6.60527050e-01 -3.36110383e-01 -4.58279908e-01 3.46242458e-01 2.61791684e-02 -3.77468735e-01 3.08454514e-01 -7.95100629e-01 -9.18996274e-01 -2.93865114e-01 -3.01139206e-01 -7.30365753e-01 -5.01753509e-01 -3.59351426e-01 -6.45469010e-01 -8.50413501e-01 7.08176494e-01 2.08258301e-01 4.46258456e-01 -6.61056817e-01 -1.68015644e-01 1.00252461e+00 4.22711253e-01 -3.11732829e-01 1.64125121e+00 4.53383714e-01 2.28820890e-01 -1.41641593e+00 2.22701207e-01 5.22928871e-02 -2.70280153e-01 -5.26687860e-01 3.72279584e-01 -6.16766691e-01 -1.19539094e+00 1.23386765e+00 -1.33318591e+00 5.03228366e-01 5.60795844e-01 3.23100924e-01 -7.46473623e-03 9.66360748e-01 -7.18196630e-01 -1.19136727e+00 1.41080627e-02 -1.26396239e+00 1.22086930e+00 9.86385643e-02 3.20626736e-01 -4.97217983e-01 -3.79812121e-01 2.76632577e-01 5.69333136e-01 6.70913756e-01 7.92429030e-01 -5.77475786e-01 -1.60935298e-01 -4.58150208e-01 -1.54332563e-01 4.14933026e-01 5.55874109e-01 4.06740904e-01 -1.26090109e+00 -5.85292578e-01 3.86839122e-01 -2.38909498e-01 4.70331460e-01 -2.29945883e-01 1.01740813e+00 -4.25450474e-01 -4.12299842e-01 4.90981191e-01 1.13397360e+00 3.96833688e-01 8.22805345e-01 -2.21965611e-01 2.16546416e-01 1.04724371e+00 2.76222348e-01 4.72995609e-01 -4.04507875e-01 8.08100045e-01 5.33842027e-01 2.21315011e-01 -1.37404099e-01 -2.89622575e-01 7.38451481e-01 2.54620552e-01 6.36388734e-02 -5.25236905e-01 -7.24081516e-01 -2.93243587e-01 -6.47527218e-01 -9.86251295e-01 -6.50792802e-03 2.22093987e+00 3.74690801e-01 6.57118782e-02 1.59978911e-01 6.75854981e-01 1.23975074e+00 3.65501702e-01 -2.41390597e-02 -3.51987690e-01 -3.86410773e-01 1.99527293e-01 2.89342046e-01 2.61552513e-01 -1.02038038e+00 7.80421317e-01 6.33592129e+00 9.77551639e-01 -1.20237458e+00 2.57688370e-02 5.27925313e-01 1.14136659e-01 -1.41105726e-01 -3.79898369e-01 -1.34517658e+00 8.45998824e-01 8.29858005e-01 2.20240101e-01 3.10151279e-01 4.03065652e-01 -1.04300924e-01 2.91984141e-01 -6.81853652e-01 1.47047281e+00 6.68554962e-01 -1.03837061e+00 3.94150406e-01 6.14378989e-01 2.83773422e-01 -9.14857328e-01 3.73693585e-01 -1.79021671e-01 -1.31295741e-01 -1.09065628e+00 4.31447983e-01 -1.67178050e-01 1.03167176e+00 -7.32343197e-01 5.16335666e-01 2.41615027e-02 -9.69790876e-01 -2.64592826e-01 -5.88523567e-01 3.73270422e-01 1.44438326e-01 3.66479635e-01 -7.08862603e-01 5.71718216e-01 7.60404944e-01 2.79393226e-01 -7.20701873e-01 7.43764818e-01 3.63063253e-02 3.75473708e-01 -6.61135495e-01 1.55878309e-02 -6.03933781e-02 3.08255106e-01 7.97339261e-01 9.50242817e-01 5.45015335e-01 7.69650638e-02 -4.04151380e-01 5.29448390e-01 -2.59101301e-01 -9.84795466e-02 -1.35836279e+00 -1.80289686e-01 7.14231849e-01 9.85979080e-01 -5.62166452e-01 9.74394903e-02 -2.08028764e-01 7.55357742e-01 -3.39773118e-01 2.51426429e-01 -6.98514700e-01 -3.05705160e-01 9.60845232e-01 1.49946362e-01 2.22820535e-01 -3.78919393e-01 6.67710662e-01 -9.89215374e-01 9.60648656e-02 -1.32728815e+00 2.15647221e-01 -4.31091666e-01 -1.21443594e+00 8.78649652e-01 5.79973832e-02 -1.09697831e+00 7.53347799e-02 -1.01402760e+00 -3.46926302e-01 7.53299057e-01 -1.47751892e+00 -1.02796507e+00 -2.52855837e-01 1.02403915e+00 1.58089504e-01 -7.47197986e-01 8.59452844e-01 5.16923487e-01 -6.51178896e-01 1.02737856e+00 -3.39508027e-01 2.75640041e-01 7.22407103e-01 -3.32711071e-01 6.01168036e-01 5.79064846e-01 5.36706559e-02 7.30638504e-01 4.55102921e-01 -7.26558864e-01 -1.91756535e+00 -4.33122545e-01 7.16090322e-01 -4.07732725e-01 2.01666027e-01 -8.23819995e-01 -7.92213619e-01 7.84331113e-02 7.91258141e-02 9.90677848e-02 7.43470311e-01 -6.10268533e-01 -9.07178998e-01 -1.26815617e-01 -1.95054376e+00 4.27090794e-01 1.35248685e+00 -8.07000637e-01 -1.88414127e-01 4.60956812e-01 4.60988700e-01 5.83448894e-02 -7.56768405e-01 6.03454828e-01 7.95522153e-01 -1.11071658e+00 1.28445566e+00 -5.96772313e-01 -4.47537228e-02 -2.37478852e-01 -3.48884761e-01 -6.97231710e-01 1.80074155e-01 -1.16011298e+00 -1.37643844e-01 1.61630571e+00 -1.61667958e-01 -8.99481475e-01 7.32101619e-01 1.30840586e-02 4.43079382e-01 -1.92349777e-01 -1.07727647e+00 -1.26717484e+00 -3.42279583e-01 -7.06041902e-02 1.19615138e+00 1.01764584e+00 -3.54843289e-01 -4.23981905e-01 -5.63570559e-01 6.29604101e-01 8.73433828e-01 -4.32475984e-01 9.19196665e-01 -1.26244295e+00 -5.63501865e-02 -3.46116811e-01 -9.73582864e-01 -6.63390636e-01 3.40838641e-01 -4.41438198e-01 -6.33495867e-01 -6.58080876e-02 -2.78122604e-01 -2.84128755e-01 1.89034164e-01 1.35133937e-01 2.23449230e-01 7.90058255e-01 2.46835157e-01 1.89424932e-01 5.02905548e-01 3.20029587e-01 1.15507591e+00 -2.02999637e-01 1.99532092e-01 8.61514062e-02 -2.59031594e-01 4.81801778e-01 6.15491629e-01 -4.09478366e-01 -4.82866317e-01 -1.27307698e-01 -4.18245345e-01 1.20030552e-01 4.64905202e-01 -1.06294703e+00 -2.21454829e-01 1.31561875e-01 4.63258088e-01 -4.48286563e-01 5.00224769e-01 -9.43277061e-01 1.50113106e-01 7.71632552e-01 6.47070259e-02 1.93271235e-01 3.09182078e-01 4.28727120e-01 -6.01401553e-02 -2.95940816e-01 8.44162524e-01 -8.97024013e-03 -2.56326765e-01 5.48777640e-01 -2.93935329e-01 -3.85208011e-01 1.24522364e+00 -6.59004271e-01 -7.39030182e-01 -2.77934939e-01 -3.93763691e-01 -5.87604821e-01 4.97608542e-01 6.49906397e-01 9.49070811e-01 -1.29496324e+00 -7.93013215e-01 1.10919046e+00 -8.93861204e-02 -7.99816012e-01 1.99718803e-01 1.46802142e-01 -7.25232601e-01 3.76971394e-01 -7.20626771e-01 -4.34146941e-01 -2.00154901e+00 9.48078752e-01 1.70202941e-01 3.67213398e-01 -2.98001796e-01 8.48169684e-01 6.86992183e-02 -1.44727871e-01 3.04819286e-01 4.12342787e-01 -3.60763937e-01 -6.30756542e-02 1.02574193e+00 2.87359685e-01 8.04860592e-02 -1.09247506e+00 -6.34997249e-01 6.92638159e-01 1.46825388e-02 2.74844710e-02 7.06925213e-01 7.28970543e-02 -1.51725560e-01 -5.10448337e-01 1.63064563e+00 3.05522382e-01 -7.72856295e-01 2.59190667e-02 -2.22664312e-01 -1.42989945e+00 -4.91393179e-01 -1.65182471e-01 -1.15011227e+00 8.79440546e-01 8.70215237e-01 6.27312422e-01 9.41090405e-01 -1.06272124e-01 6.65321410e-01 2.07989216e-01 5.77923238e-01 -2.58960247e-01 3.06996226e-01 7.48838708e-02 1.07167149e+00 -8.96503210e-01 -1.66511387e-01 -1.06821060e+00 3.12875360e-02 1.15517783e+00 4.95403498e-01 7.54856542e-02 1.14810073e+00 3.50721449e-01 4.27337289e-02 -1.81024507e-01 -3.44234824e-01 4.40017611e-01 -5.76199479e-02 1.18230593e+00 -8.89691263e-02 -3.04033071e-01 -2.72098891e-02 1.11012142e-02 -4.20810223e-01 -3.70730728e-01 5.02986312e-01 9.47243392e-01 -4.73385789e-02 -1.34682333e+00 -9.08461511e-01 4.59564514e-02 -7.55705237e-01 4.05167788e-01 -6.87302709e-01 7.99395263e-01 6.78970590e-02 1.29225039e+00 -2.13233367e-01 -7.42731571e-01 1.93461463e-01 3.67032215e-02 6.20554507e-01 -1.88309863e-01 -4.61976171e-01 -3.05377424e-01 -6.33095354e-02 -4.28414017e-01 -5.28821588e-01 -4.47913080e-01 -2.56953508e-01 -1.04900372e+00 -4.02576983e-01 4.99212323e-03 9.34037745e-01 6.03532195e-01 2.86813796e-01 -2.80383229e-01 1.35474217e+00 -9.06726837e-01 -7.29920983e-01 -7.48058975e-01 -8.92027855e-01 7.70433784e-01 5.12801588e-01 -1.05089772e+00 -5.47958195e-01 -9.28242877e-02]
[13.016554832458496, 1.0865051746368408]
4950f89e-6ae9-4f36-8d43-fa562df09307
explanation-based-handwriting-verification
1909.02548
null
https://arxiv.org/abs/1909.02548v1
https://arxiv.org/pdf/1909.02548v1.pdf
Explanation based Handwriting Verification
Deep learning system have drawback that their output is not accompanied with ex-planation. In a domain such as forensic handwriting verification it is essential to provideexplanation to jurors. The goal of handwriting verification is to find a measure of confi-dence whether the given handwritten samples are written by the same or different writer.We propose a method to generate explanations for the confidence provided by convolu-tional neural network (CNN) which maps the input image to 15 annotations (features)provided by experts. Our system comprises of: (1) Feature learning network (FLN),a differentiable system, (2) Inference module for providing explanations. Furthermore,inference module provides two types of explanations: (a) Based on cosine similaritybetween categorical probabilities of each feature, (b) Based on Log-Likelihood Ratio(LLR) using directed probabilistic graphical model. We perform experiments using acombination of feature learning network (FLN) and each inference module. We evaluateour system using XAI-AND dataset, containing 13700 handwritten samples and 15 cor-responding expert examined features for each sample. The dataset is released for publicuse and the methods can be extended to provide explanations on other verification taskslike face verification and bio-medical comparison. This dataset can serve as the basis and benchmark for future research in explanation based handwriting verification. The code is available on github.
['Mohammad Abuzar Shaikh', 'Mihir Chauhan', 'Sargur N. Srihari']
2019-08-14
null
null
null
null
['handwriting-verification']
['computer-vision']
[ 3.80313396e-02 3.42688352e-01 -5.45810342e-01 -9.40360367e-01 -6.88700259e-01 -6.09144807e-01 7.30326355e-01 5.11853881e-02 -1.02141194e-01 9.35576320e-01 2.03296885e-01 -7.17629194e-01 -2.66981423e-01 -6.60022080e-01 -6.58798754e-01 -6.83472216e-01 2.46368408e-01 5.69670677e-01 -2.47365966e-01 2.91278630e-01 6.41738534e-01 1.04855692e+00 -1.19402671e+00 6.56694353e-01 6.96268737e-01 9.59375978e-01 -4.10106778e-01 6.83589399e-01 -2.02773362e-02 7.61501849e-01 -7.57443845e-01 -8.49803865e-01 -5.81052303e-02 -5.51833391e-01 -7.82810569e-01 2.82152444e-01 4.94267851e-01 -3.58410001e-01 -3.49437445e-01 1.09682894e+00 3.36204767e-01 -2.98174210e-02 1.04470563e+00 -1.53207946e+00 -1.15863431e+00 7.03126490e-01 -2.57379144e-01 5.60349375e-02 3.44305277e-01 2.06305146e-01 7.19336987e-01 -1.04491949e+00 5.45963526e-01 1.27556777e+00 4.21668500e-01 7.66163349e-01 -8.13112915e-01 -8.68368983e-01 -1.17628254e-01 7.34375119e-01 -1.04582953e+00 -3.09069633e-01 6.17235482e-01 -3.58978570e-01 5.69676280e-01 2.67846733e-01 1.74072057e-01 1.14509141e+00 5.55574715e-01 6.97855175e-01 1.11242175e+00 -3.29759777e-01 1.39523894e-01 3.62014502e-01 5.61844945e-01 9.54352498e-01 1.90885425e-01 1.74620718e-01 -7.00664163e-01 -1.34826899e-01 6.60926759e-01 -1.29696373e-02 -2.17314750e-01 3.82028997e-01 -6.59798145e-01 9.90516603e-01 3.97173792e-01 2.51125157e-01 -2.75971681e-01 1.96493372e-01 1.86385542e-01 4.02736694e-01 -2.34765381e-01 3.50397378e-01 -2.15985805e-01 1.84303761e-01 -9.39776778e-01 2.43459195e-01 8.82387221e-01 6.14758551e-01 5.70178866e-01 1.16772667e-01 -4.41309035e-01 4.22735512e-01 6.93179667e-01 4.78481323e-01 6.98705912e-01 -8.20518732e-01 2.27863595e-01 4.43175614e-01 -1.43184513e-01 -1.12864733e+00 -7.56953806e-02 -8.26686919e-02 -7.64406741e-01 4.28919703e-01 4.87148792e-01 -1.56558901e-01 -1.07006228e+00 1.22276604e+00 4.23673689e-02 1.13553822e-01 -9.45121944e-02 1.11961591e+00 1.33058727e+00 3.01385522e-01 -1.67281285e-01 1.31247923e-01 1.62281442e+00 -6.30507171e-01 -9.06035006e-01 3.09573323e-01 8.01317394e-02 -7.52275825e-01 7.68900037e-01 6.54092610e-01 -7.38554776e-01 -6.03346527e-01 -1.06022537e+00 -1.38803318e-01 -4.20824707e-01 4.32041615e-01 6.07214510e-01 7.09679365e-01 -8.22455883e-01 7.56764174e-01 -5.66875756e-01 -1.44837946e-01 7.29750276e-01 4.45757329e-01 -7.64534473e-01 8.82534087e-02 -1.01081979e+00 8.82755101e-01 1.43985838e-01 3.70029688e-01 -9.76120591e-01 -2.21193597e-01 -9.79944110e-01 1.01339087e-01 3.65587212e-02 -2.38488257e-01 9.56543505e-01 -7.33570695e-01 -1.20581102e+00 7.73792863e-01 -2.64691353e-01 -1.79309651e-01 6.54087245e-01 8.75529796e-02 -4.50350076e-01 1.26835525e-01 9.37461480e-02 5.24789870e-01 7.45055735e-01 -1.10434031e+00 -1.74197838e-01 -5.36679566e-01 -1.40788600e-01 -3.83022815e-01 -1.09585915e-02 1.22227073e-01 -2.54017085e-01 -6.19560063e-01 1.63909137e-01 -6.43254757e-01 1.62640825e-01 3.44027758e-01 -1.14238179e+00 -5.30836761e-01 1.07211614e+00 -1.01089430e+00 7.10519075e-01 -1.81985784e+00 -3.59883577e-01 4.74208921e-01 3.19656096e-02 3.26332539e-01 -7.21534044e-02 2.78479278e-01 -2.80715942e-01 1.99592322e-01 -2.82447636e-01 -2.94014156e-01 1.17996432e-01 1.70684040e-01 -5.20540535e-01 5.68430662e-01 5.87711394e-01 7.15230167e-01 -7.51976848e-01 -6.30539775e-01 8.86939093e-02 5.22161126e-01 -1.78207844e-01 2.83472151e-01 -5.62722087e-02 3.02724063e-01 -3.15493494e-01 9.87912059e-01 6.30377233e-01 -1.15987808e-01 3.92467491e-02 -1.04497395e-01 2.58263081e-01 1.11388892e-01 -1.21060860e+00 1.12441540e+00 -5.30535653e-02 9.72991645e-01 -4.36858922e-01 -6.15840316e-01 1.27761281e+00 3.96220237e-01 -3.11867625e-01 -6.20602742e-02 4.60752010e-01 5.28985672e-02 3.74101847e-02 -9.12596166e-01 2.81927288e-01 -3.22411418e-01 1.66183785e-01 8.74653637e-01 1.58856556e-01 1.15308382e-01 4.12668027e-02 9.11888778e-02 8.71167958e-01 -1.41482025e-01 6.90262690e-02 -1.04042711e-02 7.08113611e-01 -2.11640492e-01 4.60139841e-01 9.01635468e-01 -1.32983327e-01 6.18062973e-01 8.66305172e-01 -5.72481692e-01 -8.53429079e-01 -1.02748322e+00 -4.48532313e-01 3.32484186e-01 -1.66438639e-01 -2.35798210e-02 -4.74593103e-01 -1.06269312e+00 3.65249187e-01 1.00673497e+00 -9.83197987e-01 -5.18173948e-02 -2.90407389e-02 -2.78961509e-01 8.44029427e-01 9.73799765e-01 6.66597188e-01 -1.27900279e+00 -3.45479965e-01 -3.75278860e-01 5.96453696e-02 -6.74522221e-01 -5.94978988e-01 1.77929446e-01 -6.33791387e-01 -1.39281917e+00 -2.25736633e-01 -4.76928979e-01 9.57316637e-01 -3.79501015e-01 4.24462408e-01 5.13946593e-01 -3.65711719e-01 1.65731430e-01 -9.08718035e-02 -4.56695765e-01 -5.65643311e-01 -3.70766193e-01 1.28614262e-01 1.49632797e-01 6.44405544e-01 -2.34517455e-01 -3.05667520e-01 3.65009069e-01 -9.06407177e-01 -3.72238308e-01 6.28623962e-01 1.07260489e+00 3.44597042e-01 -1.97725236e-01 5.24165332e-01 -9.12185729e-01 9.29802954e-01 -3.34981859e-01 -4.94238406e-01 5.06341517e-01 -6.30270839e-01 3.16037089e-01 5.58412850e-01 -2.97213852e-01 -1.01531482e+00 2.83118691e-02 -2.29850829e-01 -4.66747105e-01 -4.38607246e-01 4.73262817e-01 -2.47482613e-01 9.92128700e-02 6.66310370e-01 2.51370370e-01 1.06242284e-01 -2.77378947e-01 3.11877429e-01 1.04528224e+00 9.66947019e-01 -4.54195887e-01 5.47836483e-01 2.24692106e-01 1.27424419e-01 -5.92898190e-01 -5.25893390e-01 8.69628936e-02 -6.71384037e-01 -2.94837236e-01 8.70395064e-01 -8.57889578e-02 -1.35552895e+00 2.58227080e-01 -1.45261943e+00 1.18387684e-01 6.73096254e-02 6.12850308e-01 -3.16937864e-01 5.01786411e-01 -7.50695109e-01 -1.03307378e+00 -2.74192184e-01 -1.13791323e+00 8.80744994e-01 4.89029408e-01 -3.55975062e-01 -9.77439344e-01 -1.69967577e-01 4.56085563e-01 4.08293307e-02 3.17497194e-01 1.10657179e+00 -1.12552762e+00 -3.38017970e-01 -6.81545436e-01 -3.71000081e-01 3.76264215e-01 1.33949667e-01 3.17632198e-01 -1.39276481e+00 1.68835044e-01 -1.59846470e-01 -5.41918993e-01 9.76408839e-01 2.91343629e-01 1.74417448e+00 -4.78245527e-01 -9.55145657e-02 4.55823213e-01 1.00167680e+00 1.44995242e-01 8.31550181e-01 -5.60831465e-02 3.80580932e-01 5.91150463e-01 3.87643039e-01 4.36767250e-01 1.06430173e-01 3.80576640e-01 3.46774042e-01 7.50620291e-02 -1.89212769e-01 -2.16151670e-01 2.53597945e-01 1.40427146e-02 -2.78410017e-01 -2.70394027e-01 -8.40645790e-01 1.93338767e-01 -1.85455382e+00 -1.13538849e+00 -2.81066716e-01 2.07534623e+00 7.24432409e-01 -1.24268219e-01 -7.39141274e-03 3.21839213e-01 8.64837885e-01 -3.53488505e-01 -8.20514441e-01 -8.14747870e-01 -4.78035398e-02 9.97311547e-02 6.39012828e-02 6.76952183e-01 -8.04376662e-01 6.30498886e-01 6.26120234e+00 4.64780271e-01 -1.28970790e+00 -2.43725017e-01 9.36261833e-01 3.99731934e-01 -3.64503592e-01 -2.68393904e-01 -8.45106602e-01 4.55102861e-01 7.33554125e-01 -2.41794884e-02 1.07321985e-01 7.85313010e-01 1.16728902e-01 3.18736434e-02 -1.32545578e+00 8.70185971e-01 4.52819079e-01 -1.42780304e+00 2.09339187e-01 2.24062093e-02 1.91411853e-01 -5.07582963e-01 1.92354605e-01 -2.48599960e-03 3.57345849e-01 -1.49670112e+00 5.36025345e-01 8.60263646e-01 9.19621170e-01 -7.14275420e-01 1.15313590e+00 9.32605788e-02 -4.89653379e-01 -3.85751203e-02 -3.84449095e-01 2.26886123e-01 -1.58351585e-01 4.67355877e-01 -1.39133084e+00 4.62824702e-01 3.05231243e-01 4.31090266e-01 -4.39703554e-01 9.61605787e-01 -1.08578837e+00 5.62042475e-01 1.73864603e-01 -2.81922847e-01 1.21305726e-01 1.78790078e-01 2.14317158e-01 1.28950036e+00 2.33174667e-01 1.84972540e-01 -3.15541118e-01 1.35756850e+00 -1.73127592e-01 -2.22887725e-01 -4.22347426e-01 -2.06810594e-01 5.50537407e-01 1.35803604e+00 -9.22499478e-01 -4.10395831e-01 3.80701609e-02 9.64738905e-01 1.64910927e-01 1.21441297e-01 -7.74791360e-01 -8.00204217e-01 2.70617485e-01 -3.69611681e-02 1.25016406e-01 1.64029732e-01 -5.90627849e-01 -1.01818550e+00 -1.15278520e-01 -8.50531578e-01 6.61040008e-01 -9.80460227e-01 -1.49612010e+00 7.69604564e-01 -1.15668371e-01 -7.50191927e-01 -4.23441499e-01 -1.08748674e+00 -8.48572373e-01 1.12567985e+00 -1.30448353e+00 -9.85570848e-01 -3.86288315e-01 5.25474608e-01 4.73596066e-01 -5.15941322e-01 8.60997438e-01 -1.75767466e-02 -4.65214789e-01 9.69947517e-01 -3.96327525e-01 6.43621445e-01 7.72194445e-01 -1.22857010e+00 1.30911870e-02 5.66155851e-01 1.36169508e-01 8.68218303e-01 6.83824778e-01 -8.98440480e-01 -1.22586000e+00 -9.55247819e-01 1.32603061e+00 -6.75198257e-01 3.85134429e-01 -6.88218400e-02 -9.81708825e-01 6.64354265e-01 1.72696903e-01 2.22961195e-02 9.61564660e-01 1.68339148e-01 -4.05556887e-01 5.04502095e-02 -1.57446527e+00 2.10170940e-01 3.72997582e-01 -6.30019963e-01 -7.17759550e-01 4.49341834e-01 1.45471916e-02 -3.43464106e-01 -7.06413984e-01 -4.62374557e-03 8.89099360e-01 -8.73746336e-01 3.89001876e-01 -1.10696054e+00 9.96864676e-01 -3.53496850e-01 1.46478729e-03 -9.98425424e-01 -2.00495169e-01 -1.60431951e-01 1.95138846e-02 1.10682380e+00 6.24401391e-01 -5.98983586e-01 8.35151434e-01 1.07961023e+00 -1.02388352e-01 -9.46275353e-01 -8.23258996e-01 -7.80221224e-01 -1.58683434e-02 -4.57888782e-01 7.67726719e-01 1.08775401e+00 2.09775627e-01 1.51746631e-01 -3.09339136e-01 3.90852988e-01 5.58724940e-01 1.20376498e-01 5.92854440e-01 -9.75573063e-01 -4.52746719e-01 -3.54036152e-01 -8.29237461e-01 -4.95304555e-01 3.42522234e-01 -1.12686634e+00 6.74471483e-02 -1.36909473e+00 4.73910451e-01 -8.59694853e-02 -1.30088568e-01 1.09121394e+00 -1.08505771e-01 2.53412187e-01 1.42771333e-01 -1.03266314e-01 -1.75913721e-01 3.89868885e-01 1.02853489e+00 -2.27427036e-01 2.56276101e-01 4.31101732e-02 -7.75096297e-01 5.31637549e-01 6.64843202e-01 -5.46417475e-01 -6.45955279e-02 -2.65116245e-01 -1.83968887e-01 2.93882668e-01 6.38386428e-01 -5.32001197e-01 4.20632631e-01 -1.22352324e-01 1.07470131e+00 -7.25565910e-01 2.02739418e-01 -5.02825797e-01 -4.75052409e-02 6.52294397e-01 -5.64559281e-01 7.89918080e-02 9.26486701e-02 3.39077950e-01 -2.06843555e-01 -7.33079016e-01 4.79427218e-01 5.84687814e-02 -2.77411550e-01 2.41189003e-01 -3.55484009e-01 -6.20946348e-01 7.47849345e-01 -2.97640562e-01 -5.99442780e-01 -6.56835556e-01 -8.89632642e-01 2.21049190e-01 -1.42090227e-02 2.29196906e-01 1.13772583e+00 -1.44442630e+00 -9.06517029e-01 3.22737545e-01 -3.74506786e-02 -6.04454041e-01 5.24042174e-02 5.58615625e-01 -3.83503586e-01 5.22526741e-01 -3.40966344e-01 -3.92278731e-01 -1.53434551e+00 1.92349926e-01 3.66954744e-01 1.58261478e-01 -2.36541167e-01 9.42131877e-01 -2.16949895e-01 -6.12248838e-01 3.22602123e-01 -2.12766722e-01 -3.40523690e-01 -2.02401027e-01 9.18609321e-01 3.62912595e-01 2.17386801e-02 -5.48313975e-01 -4.77844983e-01 2.27207303e-01 -1.69517055e-01 -1.03636280e-01 1.43418443e+00 7.04281807e-01 -1.86681628e-01 2.99728423e-01 8.52093220e-01 -2.09102958e-01 -1.04317439e+00 1.63693875e-01 -1.12645589e-01 -7.35433221e-01 -2.05551550e-01 -1.11518180e+00 -1.16512311e+00 1.05987990e+00 5.42126298e-01 -1.85036734e-01 7.23265886e-01 1.50704756e-01 1.63946524e-01 7.04736769e-01 -7.27175921e-02 -8.64927590e-01 1.67251229e-01 2.26766020e-01 1.34362257e+00 -1.44920838e+00 1.23319432e-01 -6.07241467e-02 -7.42674172e-01 1.79304171e+00 5.43735623e-01 -1.53625250e-01 7.10412025e-01 1.88920796e-01 1.55919939e-01 -5.44160604e-01 -6.34565413e-01 3.48083884e-01 6.48377299e-01 5.50297797e-01 7.45477796e-01 7.46648163e-02 -4.57471579e-01 1.06200993e+00 -3.86862725e-01 -5.12308069e-02 5.64184546e-01 5.48214078e-01 -1.66521430e-01 -1.13579595e+00 -5.52720368e-01 6.18342757e-01 -3.65886092e-01 1.40472308e-01 -8.25941443e-01 7.20402718e-01 2.24331528e-01 1.09018576e+00 5.64872578e-04 -3.58911544e-01 5.03804833e-02 7.26921409e-02 4.18137550e-01 -4.82079893e-01 -4.84704942e-01 -4.97491181e-01 9.41707566e-03 -2.80844420e-01 -1.11719919e-02 -7.80184805e-01 -1.43830347e+00 -3.73323739e-01 -5.69871843e-01 1.98517516e-01 1.03305304e+00 1.21889400e+00 5.91933876e-02 2.68727392e-01 3.92582536e-01 -3.90805066e-01 -5.87105215e-01 -1.05932510e+00 -6.68610334e-01 3.75233591e-01 1.83750212e-01 -5.15118778e-01 -2.04669327e-01 1.40840068e-01]
[11.630926132202148, 2.2039272785186768]
1207a8af-23e6-4f8e-903a-0713b027c318
endowing-robots-with-longer-term-autonomy-by
1809.03979
null
http://arxiv.org/abs/1809.03979v1
http://arxiv.org/pdf/1809.03979v1.pdf
Endowing Robots with Longer-term Autonomy by Recovering from External Disturbances in Manipulation through Grounded Anomaly Classification and Recovery Policies
Robot manipulation is increasingly poised to interact with humans in co-shared workspaces. Despite increasingly robust manipulation and control algorithms, failure modes continue to exist whenever models do not capture the dynamics of the unstructured environment. To obtain longer-term horizons in robot automation, robots must develop introspection and recovery abilities. We contribute a set of recovery policies to deal with anomalies produced by external disturbances as well as anomaly classification through the use of non-parametric statistics with memoized variational inference with scalable adaptation. A recovery critic stands atop of a tightly-integrated, graph-based online motion-generation and introspection system that resolves a wide range of anomalous situations. Policies, skills, and introspection models are learned incrementally and contextually in a task. Two task-level recovery policies: re-enactment and adaptation resolve accidental and persistent anomalies respectively. The introspection system uses non-parametric priors along with Markov jump linear systems and memoized variational inference with scalable adaptation to learn a model from the data. Extensive real-robot experimentation with various strenuous anomalous conditions is induced and resolved at different phases of a task and in different combinations. The system executes around-the-clock introspection and recovery and even elicited self-recovery when misclassifications occurred.
['Sakmongkon Chumkamon', 'Longxin Chen', 'Shuangqi Luo', 'Shuangda Duan', 'Yisheng Guan', 'Hongmin Wu', 'Juan Rojas', 'Dong Liu']
2018-09-11
null
null
null
null
['anomaly-classification']
['computer-vision']
[ 1.92409307e-01 2.80481994e-01 1.13254912e-01 -1.62062924e-02 -2.74299055e-01 -2.31883958e-01 4.38990533e-01 1.04002528e-01 -3.27801228e-01 8.27280939e-01 -1.36077687e-01 -1.06227115e-01 -6.08174801e-01 -1.71571955e-01 -7.08875835e-01 -4.76360828e-01 -5.59794366e-01 1.02146244e+00 4.53695416e-01 -3.50980133e-01 4.96978283e-01 6.95947051e-01 -1.62547827e+00 -1.45722792e-01 9.26151574e-01 4.67712700e-01 4.81285006e-01 1.23806834e+00 4.03343737e-01 9.59385872e-01 -5.43519318e-01 5.66407919e-01 5.75313628e-01 -1.37706861e-01 -5.94148695e-01 4.24968779e-01 -8.33652392e-02 -1.85700968e-01 -4.03216630e-01 9.65918660e-01 2.53155202e-01 7.29870796e-01 6.47781491e-01 -1.69723558e+00 -3.46470743e-01 2.53959835e-01 -4.08591390e-01 3.05348873e-01 4.21380997e-01 1.01382947e+00 1.24250650e-01 -4.68892753e-01 4.88221318e-01 1.46365643e+00 6.21969759e-01 5.46065807e-01 -1.22207892e+00 -1.26759902e-01 4.27324772e-01 3.28739136e-01 -9.08035576e-01 -3.84672254e-01 4.76222873e-01 -6.36486769e-01 1.34097528e+00 -3.67564857e-02 4.64440435e-01 1.58222163e+00 1.04300618e+00 4.03797716e-01 6.46572292e-01 -2.97588413e-04 5.94140828e-01 -5.25091998e-02 -2.27381974e-01 7.57740021e-01 4.03340995e-01 2.02408582e-01 -5.82448006e-01 -3.67804468e-01 8.66480172e-01 2.97552943e-01 2.20204820e-03 -6.39307737e-01 -1.48752844e+00 2.62991667e-01 1.32977301e-02 -1.36083379e-01 -1.02327001e+00 3.12779695e-01 6.45572841e-01 6.72778487e-01 -8.06124657e-02 8.45454574e-01 -5.89482427e-01 -4.18312490e-01 -2.95651138e-01 2.39375189e-01 9.37649548e-01 1.21915317e+00 6.49153411e-01 4.83562022e-01 -9.17973593e-02 5.23097992e-01 -1.37415007e-02 2.78301120e-01 7.01545238e-01 -1.50425017e+00 2.82626539e-01 6.05513155e-01 5.25600553e-01 -9.00244236e-01 -6.40675008e-01 -2.99671441e-01 -5.56817412e-01 8.26190591e-01 1.54919818e-01 -2.73609161e-01 -1.10915172e+00 1.48361135e+00 6.38663888e-01 3.15558136e-01 1.42127320e-01 8.34329545e-01 -6.36444449e-01 3.09765339e-01 -1.46137103e-01 -1.98199555e-01 8.15815747e-01 -8.02681684e-01 -9.29377317e-01 -5.66853702e-01 4.12230343e-01 -3.27585697e-01 9.92650509e-01 9.18735921e-01 -9.57420945e-01 -6.44289374e-01 -1.15403843e+00 3.37314606e-01 -2.14871019e-01 -3.85382950e-01 1.67941257e-01 -2.50997305e-01 -8.13253105e-01 9.85652983e-01 -1.49325991e+00 -6.03005528e-01 -1.32179737e-01 3.75663728e-01 -5.19858181e-01 6.15663938e-02 -6.79407477e-01 1.44664979e+00 5.33856332e-01 4.37128186e-01 -1.67880547e+00 -3.35666925e-01 -7.45918095e-01 -4.90085393e-01 1.02905393e+00 -7.60259092e-01 1.32382894e+00 -7.86515832e-01 -1.92551720e+00 1.71596035e-01 3.30302715e-01 -6.28099084e-01 8.06632102e-01 -5.98536193e-01 -2.81603456e-01 -3.44491415e-02 2.68580824e-01 1.14325441e-01 1.33653128e+00 -1.28039372e+00 -6.37738407e-01 -3.93903285e-01 -3.99449974e-01 5.40501416e-01 3.06181580e-01 -4.76022005e-01 1.72908664e-01 -9.15656537e-02 3.67551476e-01 -1.35977793e+00 -5.05305529e-01 -3.81738544e-01 -5.09437680e-01 1.15918353e-01 9.98602629e-01 -6.33359253e-01 6.10788226e-01 -2.02830958e+00 6.51655078e-01 1.09068580e-01 -8.50304514e-02 -1.32884130e-01 -8.75493214e-02 7.45485485e-01 -1.30071100e-02 -4.62757915e-01 -3.32937330e-01 -2.92722106e-01 2.05870450e-01 6.06793404e-01 -4.46201265e-01 6.25340700e-01 2.59998888e-01 3.42379868e-01 -1.14649463e+00 -5.00786789e-02 2.86266387e-01 -1.50562003e-01 -5.44800341e-01 4.05577809e-01 -5.75351119e-01 9.08113778e-01 -5.89721084e-01 6.33168697e-01 3.57743795e-03 1.18641593e-01 1.89756647e-01 6.36861384e-01 -1.37095049e-01 -1.75618574e-01 -1.29971111e+00 1.76997578e+00 -3.91921788e-01 2.86498189e-01 4.56844389e-01 -9.53917861e-01 8.51062536e-01 2.55707949e-01 4.76694763e-01 -1.54840305e-01 1.67415217e-01 3.19812894e-01 5.21080978e-02 -1.01754057e+00 8.16297293e-01 3.68593067e-01 -2.21099705e-01 2.74257183e-01 7.53717497e-02 -4.36311692e-01 3.14740092e-02 2.19839260e-01 1.89192617e+00 6.50827944e-01 3.02847654e-01 -7.63144419e-02 3.07362348e-01 6.05793238e-01 8.05587888e-01 9.67825711e-01 -6.66794002e-01 2.59584308e-01 4.50893819e-01 -5.05509317e-01 -9.83525872e-01 -1.07439768e+00 4.87212926e-01 1.26860988e+00 2.77747095e-01 1.18542068e-01 -3.45389903e-01 -4.24693912e-01 2.66596586e-01 1.08552003e+00 -6.36103392e-01 -8.26288462e-01 -5.86277306e-01 -3.75511587e-01 -2.63882596e-02 4.68882233e-01 1.85954273e-01 -1.56648076e+00 -1.29396701e+00 5.27786791e-01 1.42444745e-01 -7.57459044e-01 -1.48511499e-01 7.56865740e-01 -8.81400228e-01 -1.29697800e+00 2.40274090e-02 -3.37733328e-01 7.81022489e-01 2.38711247e-03 5.96343935e-01 -1.65330291e-01 -6.82067037e-01 9.91496682e-01 -2.04622611e-01 -3.69934857e-01 -8.15435827e-01 -1.86057657e-01 7.10496724e-01 -1.45753905e-01 -1.52805954e-01 -7.10203111e-01 -2.03184977e-01 3.36820930e-01 -5.92887461e-01 -4.35922682e-01 7.21041858e-01 1.06157398e+00 5.76987326e-01 1.85204163e-01 5.04354835e-01 -7.42452323e-01 9.66048837e-01 -8.02572191e-01 -5.95937550e-01 1.67938188e-01 -7.98946559e-01 2.59241283e-01 4.49670434e-01 -9.28804815e-01 -1.34473407e+00 1.48016408e-01 7.75621712e-01 -8.83986115e-01 -4.79222000e-01 4.21474636e-01 1.30570486e-01 4.41549182e-01 9.79250669e-01 9.78402942e-02 3.04843187e-01 -1.46622494e-01 4.21434730e-01 4.41362768e-01 1.01966631e+00 -7.57644832e-01 7.68529892e-01 2.81852573e-01 -5.70908003e-02 -6.12694144e-01 -3.46383095e-01 -3.61254036e-01 -8.79018486e-01 -4.81963933e-01 6.57604814e-01 -7.12167978e-01 -6.61212206e-01 5.09967208e-01 -8.55669856e-01 -9.79963303e-01 -6.38215661e-01 3.74495685e-01 -1.06232977e+00 1.15410358e-01 -5.13763428e-01 -1.23426008e+00 1.05929986e-01 -9.37263489e-01 7.24741042e-01 1.57405078e-01 -4.08897221e-01 -5.95004499e-01 3.78223121e-01 6.26732111e-02 4.18303043e-01 5.84599733e-01 4.95587587e-01 -8.89364779e-01 -6.07447922e-01 -3.68696958e-01 5.40657282e-01 1.18486993e-01 7.57917389e-02 -9.39574689e-02 -5.89452922e-01 -5.22553086e-01 1.89213574e-01 -4.17064935e-01 3.84458840e-01 9.93466303e-02 6.90172732e-01 -3.94195616e-01 -3.78661633e-01 5.08453697e-02 8.19534957e-01 2.71944135e-01 2.64764369e-01 6.86560750e-01 4.71528977e-01 6.40151024e-01 1.03874624e+00 7.37266600e-01 1.42151877e-01 1.32621706e-01 1.08374488e+00 7.61222303e-01 3.17390501e-01 -2.52421111e-01 7.24978268e-01 3.61888498e-01 2.72525474e-02 1.67272741e-03 -1.01396668e+00 6.52327716e-01 -2.42482495e+00 -9.48413253e-01 4.74504195e-02 2.32094336e+00 4.70080107e-01 5.86645722e-01 -2.17952579e-02 -1.48335278e-01 6.51417732e-01 -2.62053609e-01 -1.34692240e+00 -3.65727454e-01 3.94432485e-01 -5.95432639e-01 6.43538296e-01 5.47673047e-01 -7.61297703e-01 8.75653923e-01 5.70020819e+00 1.53059429e-02 -7.80153930e-01 7.71787912e-02 -1.77600421e-02 -1.90133780e-01 3.75164866e-01 4.54855487e-02 -2.65110731e-01 2.12950528e-01 1.08028460e+00 2.42645154e-03 7.95866489e-01 1.37376881e+00 1.12911485e-01 -5.63386321e-01 -1.24397254e+00 5.36317289e-01 9.97175202e-02 -6.02528155e-01 -7.31091440e-01 -5.58035681e-03 6.47221267e-01 4.48555678e-01 -8.86026174e-02 5.87841094e-01 8.50069582e-01 -5.94302654e-01 9.82327402e-01 9.66122210e-01 2.30521575e-01 -5.76123059e-01 5.70061982e-01 9.07795727e-01 -7.02240705e-01 -7.91196108e-01 -3.64503831e-01 -3.81616831e-01 2.14853659e-01 7.10488558e-02 -1.20189929e+00 4.49682027e-01 6.03676140e-01 4.64870930e-01 -2.70996153e-01 8.03988338e-01 -3.14087331e-01 9.69414506e-03 -3.05428594e-01 1.56253979e-01 -5.48573621e-02 -9.56038833e-02 1.22873986e+00 4.74886388e-01 1.98580489e-01 -1.22521296e-01 6.95218623e-01 8.05583835e-01 7.03475296e-01 -8.09736013e-01 -8.75118196e-01 9.58412439e-02 6.77289128e-01 9.27688897e-01 -6.53543413e-01 -3.01913798e-01 3.46665889e-01 1.23645926e+00 4.45401460e-01 5.08337438e-01 -5.43851495e-01 -2.98623115e-01 6.97488725e-01 2.88894884e-02 -2.10779961e-02 -8.80502582e-01 -9.22795534e-02 -8.54512691e-01 -1.99941292e-01 -8.63942087e-01 2.45102033e-01 -9.36939955e-01 -1.40653217e+00 4.75229770e-01 2.74659246e-01 -1.08678150e+00 -8.17945838e-01 -4.49205667e-01 -6.00928485e-01 5.25466383e-01 -1.07232356e+00 -3.12146395e-01 -4.70881999e-01 7.16967702e-01 8.18815529e-01 -2.93379396e-01 8.25146556e-01 -5.27641535e-01 -8.23392332e-01 -2.95271844e-01 -4.64148186e-02 -5.43160975e-01 9.16729689e-01 -1.30439425e+00 9.58897099e-02 9.70969021e-01 -5.10670066e-01 6.40979886e-01 1.29454207e+00 -1.34050322e+00 -1.84821284e+00 -1.15981269e+00 -7.90817589e-02 -5.79731107e-01 9.54009533e-01 -2.80005068e-01 -1.27771926e+00 1.18128538e+00 9.68262330e-02 -7.48692779e-03 -1.71056435e-01 -9.29681808e-02 9.71457511e-02 6.52204528e-02 -1.06379855e+00 6.14327371e-01 1.05974126e+00 -3.09967399e-01 -8.97586167e-01 6.18378103e-01 7.24392235e-01 -8.12511504e-01 -4.61939633e-01 3.73427540e-01 1.22399129e-01 -7.55754828e-01 5.54899454e-01 -6.42858803e-01 -1.85315356e-01 -3.29348236e-01 1.21904396e-01 -1.50016952e+00 -3.50488603e-01 -1.25003397e+00 -3.16995025e-01 5.37258923e-01 2.03510910e-01 -8.87544453e-01 4.89475787e-01 9.06346917e-01 -5.62138677e-01 -1.96223408e-01 -8.54766607e-01 -1.11108530e+00 -7.08034933e-01 -4.39745784e-01 1.11892700e-01 7.85260081e-01 3.59544665e-01 2.59413093e-01 -5.76669097e-01 5.85790396e-01 7.27773309e-01 -3.19938779e-01 1.18700349e+00 -1.18655086e+00 -2.61817366e-01 -2.27934457e-02 -2.17078224e-01 -3.89480412e-01 3.24658692e-01 -3.93127024e-01 8.75989497e-01 -1.27661371e+00 -3.13260496e-01 -1.94372669e-01 -2.64213532e-01 5.67207694e-01 4.71727997e-02 -7.06688583e-01 -2.44893059e-02 4.83486205e-01 -8.12715054e-01 8.16795170e-01 9.10558224e-01 1.70737803e-01 -7.71201432e-01 2.73405351e-02 5.03703356e-02 8.54077399e-01 9.82747257e-01 -5.74904442e-01 -6.66439593e-01 -2.64206588e-01 2.31065184e-01 5.89371622e-01 4.10650074e-01 -1.38176572e+00 6.71203256e-01 -5.01085758e-01 8.52659345e-02 -4.09117788e-01 3.49418223e-01 -8.87668312e-01 4.51281726e-01 7.80354440e-01 -3.57571095e-01 4.77752030e-01 1.27077371e-01 1.37036264e+00 2.78401285e-01 -5.50264772e-03 4.95020270e-01 -3.85233641e-01 -9.04189348e-01 1.60161212e-01 -7.79538810e-01 -1.79053769e-01 1.40053415e+00 -3.89365077e-01 -1.96480677e-01 -2.93868035e-01 -1.17579484e+00 6.32558227e-01 5.61806619e-01 7.11460412e-01 6.55304790e-01 -7.50760376e-01 -4.71402526e-01 3.96010578e-01 1.02417991e-02 3.12729180e-01 3.06832522e-01 1.01734805e+00 -3.00063103e-01 -1.40121400e-01 -6.66098535e-01 -6.67023540e-01 -5.59136033e-01 6.62119269e-01 3.31302434e-01 -1.09881654e-01 -9.16578829e-01 5.74779630e-01 -3.08712661e-01 -4.55536664e-01 4.24170226e-01 -3.55144292e-01 1.52808756e-01 -4.08726484e-01 2.53906429e-01 7.41440892e-01 -1.54074714e-01 -7.08884671e-02 -3.30714345e-01 -1.46789044e-01 1.18338436e-01 -2.86221385e-01 1.23969364e+00 -3.23363513e-01 -4.80931699e-02 1.12974834e+00 7.35346749e-02 -5.85826159e-01 -2.11287856e+00 -5.17711937e-02 4.38348114e-01 -3.04221451e-01 -2.76367009e-01 -9.56422329e-01 -3.31319422e-01 5.26232600e-01 4.29058671e-01 2.22322926e-01 6.60438418e-01 -4.26095873e-01 3.68689567e-01 8.70576262e-01 7.83393979e-01 -1.79557729e+00 6.46704376e-01 9.39228594e-01 1.37633312e+00 -1.15630150e+00 7.21354038e-02 1.85983971e-01 -1.04726899e+00 8.44804704e-01 1.13041484e+00 -5.50743341e-01 3.35357547e-01 2.54121095e-01 -1.48493037e-01 -1.40874401e-01 -1.24124849e+00 1.61176890e-01 -2.88132131e-01 8.65612984e-01 -6.18257463e-01 5.41523620e-02 3.87286901e-01 3.30035090e-01 1.27673849e-01 -1.43110812e-01 8.39888632e-01 1.47275126e+00 -8.13302815e-01 -3.99624050e-01 -6.38198316e-01 2.32315734e-01 2.18153730e-01 6.23631716e-01 -1.10109553e-01 7.91494787e-01 -1.65296540e-01 9.04511511e-01 -1.70217715e-02 -4.04489756e-01 4.94285256e-01 1.94629282e-01 1.67632326e-01 -9.72660244e-01 -3.70136201e-01 1.98518466e-02 7.24393949e-02 -1.20079279e+00 1.92145035e-01 -9.56020653e-01 -1.61967397e+00 -3.94669734e-03 -1.62878707e-01 -1.35617077e-01 8.10675442e-01 8.81873369e-01 7.25409865e-01 1.04924834e+00 3.19349468e-01 -1.31904936e+00 -1.29575408e+00 -1.23505497e+00 -5.71833670e-01 1.93973288e-01 6.32229686e-01 -9.73419428e-01 -4.54954177e-01 -3.86200473e-02]
[4.472651958465576, 1.412313461303711]
c468d417-0b74-4d29-bd1f-f46af94b85f9
sts-uhh-at-semeval-2017-task-1-scoring
null
null
https://aclanthology.org/S17-2025
https://aclanthology.org/S17-2025.pdf
STS-UHH at SemEval-2017 Task 1: Scoring Semantic Textual Similarity Using Supervised and Unsupervised Ensemble
This paper reports the STS-UHH participation in the SemEval 2017 shared Task 1 of Semantic Textual Similarity (STS). Overall, we submitted 3 runs covering monolingual and cross-lingual STS tracks. Our participation involves two approaches: unsupervised approach, which estimates a word alignment-based similarity score, and supervised approach, which combines dependency graph similarity and coverage features with lexical similarity measures using regression methods. We also present a way on ensembling both models. Out of 84 submitted runs, our team best multi-lingual run has been ranked 12th in overall performance with correlation of 0.61, 7th among 31 participating teams.
['Chris Biemann', 'Sarah Kohail', 'Amr Rekaby Salama']
2017-08-01
null
null
null
semeval-2017-8
['graph-similarity']
['graphs']
[-1.66281238e-01 2.51619041e-01 -3.46661806e-01 -3.43704224e-01 -1.35384297e+00 -6.38465881e-01 7.28186250e-01 7.48846889e-01 -6.81626916e-01 5.37030041e-01 9.04479980e-01 -4.09613885e-02 -2.01354362e-02 -1.55269206e-01 -4.19004023e-01 -1.52459340e-02 9.75685269e-02 7.66607404e-01 2.85683751e-01 -5.64630210e-01 6.06993854e-01 -2.81294018e-01 -1.16105545e+00 5.75049579e-01 9.70661342e-01 7.32058287e-01 1.48493379e-01 4.20895994e-01 -3.28759640e-01 7.92732358e-01 -2.05749750e-01 -8.69169295e-01 -1.28598526e-01 -5.50316572e-01 -1.42726970e+00 -7.14618385e-01 6.28752947e-01 7.82224178e-01 -1.60104588e-01 7.83446550e-01 5.76132059e-01 9.45059806e-02 7.31196046e-01 -1.01750124e+00 -5.88505864e-01 1.39292073e+00 -2.78758734e-01 5.60988665e-01 9.28861320e-01 -3.43588978e-01 1.73495436e+00 -1.08188283e+00 1.13832128e+00 1.16926622e+00 1.10961723e+00 4.57830727e-02 -1.06481314e+00 -4.98952150e-01 -3.11285228e-01 5.47227323e-01 -1.26121771e+00 -4.79559153e-01 2.37764582e-01 -4.81983572e-01 1.77376342e+00 -7.13393614e-02 3.77415240e-01 1.22045410e+00 -1.53529374e-02 6.73417449e-01 1.46672273e+00 -6.71645105e-01 -1.57621186e-02 6.01262338e-02 2.68810302e-01 3.48433524e-01 1.06637768e-01 -3.38303298e-01 -1.07008469e+00 -2.17695460e-01 -7.87524134e-02 -9.22357380e-01 8.10562745e-02 3.08869705e-02 -1.56970632e+00 8.71561766e-01 1.72678947e-01 8.62072825e-01 9.59206596e-02 -2.38407910e-01 8.81654203e-01 7.03298509e-01 8.31671715e-01 7.12139428e-01 -6.12308860e-01 -2.28487208e-01 -1.00932336e+00 2.28524148e-01 9.18932617e-01 1.02911305e+00 4.36209619e-01 -4.03636187e-01 -3.20594609e-01 1.33544874e+00 4.48101796e-02 4.68372285e-01 9.74243283e-01 -5.45223117e-01 7.79009402e-01 4.14865345e-01 -4.63290542e-01 -6.72772467e-01 -3.23073775e-01 -2.22012237e-01 -1.96826413e-01 -5.59376299e-01 1.18045412e-01 1.29991531e-01 -4.44869637e-01 1.62753391e+00 1.52315035e-01 -1.12672351e-01 1.82888284e-01 5.56699872e-01 1.33435011e+00 3.42415720e-01 2.48898923e-01 -1.51023045e-01 1.38616526e+00 -1.00873077e+00 -7.43755341e-01 -1.11895300e-01 1.40928233e+00 -1.54095662e+00 1.03866887e+00 1.58745572e-01 -1.19698894e+00 -4.49388385e-01 -1.16087234e+00 -1.90349475e-01 -6.01687849e-01 -4.09599580e-02 1.32218882e-01 1.74153313e-01 -1.30493140e+00 8.60200286e-01 -3.88550520e-01 -1.23230696e+00 -4.42185625e-02 -6.23684861e-02 -5.92576623e-01 2.86958739e-02 -1.45968366e+00 1.77040982e+00 5.85006237e-01 -6.90316916e-01 -8.53225589e-02 -8.89028072e-01 -8.07982087e-01 -3.88771206e-01 1.53297365e-01 -4.58160758e-01 1.23885870e+00 -6.43551767e-01 -1.13767636e+00 1.54599845e+00 -2.54701346e-01 -5.78575909e-01 3.63701761e-01 -2.34974444e-01 -7.51469851e-01 -1.99956924e-01 6.12114489e-01 2.79213578e-01 1.38123855e-01 -4.99401301e-01 -4.31992769e-01 -3.75752836e-01 -5.24167478e-01 5.66989124e-01 -2.83477187e-01 7.03567863e-01 -1.50497526e-01 -8.50282907e-01 2.60131955e-01 -9.21906471e-01 8.93246830e-02 -6.65241003e-01 -2.63190389e-01 -7.75522828e-01 4.31781828e-01 -1.14342821e+00 1.36649656e+00 -1.66370487e+00 3.11783701e-01 8.38653147e-02 -1.44227920e-02 -6.35951981e-02 -2.27921516e-01 1.15119779e+00 -3.26027364e-01 1.46988735e-01 -1.24263883e-01 -5.80823898e-01 1.93367600e-01 -8.40723738e-02 -1.72636017e-01 2.86387265e-01 -1.21486254e-01 1.06570947e+00 -9.15391028e-01 -7.50829101e-01 -1.57712281e-01 -5.66954054e-02 -8.80675986e-02 1.68916240e-01 8.93921927e-02 3.48116457e-01 -1.97865233e-01 5.15983939e-01 2.93318719e-01 -3.01649962e-02 5.87968528e-01 -4.22601819e-01 -3.48018616e-01 1.13621998e+00 -6.62853241e-01 2.40833449e+00 -6.62120461e-01 6.89306140e-01 -4.69189346e-01 -9.09462273e-01 1.00956964e+00 4.21260267e-01 2.90531605e-01 -1.00498533e+00 6.07658103e-02 7.43620694e-01 -6.77848831e-02 -4.46432501e-01 5.84411979e-01 2.23470684e-02 -5.24569929e-01 7.17030346e-01 3.56491894e-01 -4.55279231e-01 3.92042100e-01 5.45907140e-01 1.22222650e+00 2.49718174e-01 7.14742899e-01 -8.16516876e-01 5.46942174e-01 1.67370483e-01 1.60769656e-01 3.19149792e-01 -1.27563447e-01 6.06216371e-01 4.60283458e-01 -2.43095770e-01 -1.45236349e+00 -1.01153374e+00 -1.75232977e-01 1.24190605e+00 -9.02175754e-02 -8.92300546e-01 -7.63562918e-01 -7.82300353e-01 -7.43710473e-02 7.82379985e-01 -5.26482880e-01 2.11542640e-02 -6.25906646e-01 -3.14229757e-01 9.96688843e-01 4.27025646e-01 2.19316036e-01 -1.02155495e+00 1.80149227e-01 1.51397213e-01 -7.42728889e-01 -1.44592524e+00 -7.43114710e-01 8.98343325e-02 -6.12338543e-01 -8.60941410e-01 -3.91193897e-01 -9.67525482e-01 -1.13650933e-01 1.45978190e-03 1.42689157e+00 -1.50034145e-01 -3.44122410e-01 1.53162196e-01 -8.09361994e-01 -7.98542798e-03 -4.93660152e-01 4.88795847e-01 3.07871461e-01 -5.98233998e-01 5.91521025e-01 -4.53599900e-01 -1.27865657e-01 8.98767263e-02 -2.13986084e-01 -1.58363506e-01 3.01533282e-01 5.65862358e-01 4.68865365e-01 -9.47495461e-01 5.75316727e-01 -6.83762968e-01 6.82670236e-01 -7.12415814e-01 -1.68837812e-02 5.91227531e-01 -7.64788091e-01 1.31197929e-01 3.95867676e-01 7.64031857e-02 -7.71874368e-01 -1.78885505e-01 -1.83148801e-01 1.18806913e-01 2.06769556e-01 6.79953277e-01 2.42966428e-01 2.43797954e-02 7.75334060e-01 -1.32359313e-02 -2.94471920e-01 -6.32232249e-01 4.25454885e-01 7.49729097e-01 5.71081042e-01 -7.98828542e-01 4.99373794e-01 -1.92943662e-01 -3.91520530e-01 -6.44343495e-01 -1.21072829e+00 -7.81824708e-01 -9.58818913e-01 -8.48433673e-02 1.10786450e+00 -1.03515518e+00 -2.08249226e-01 3.22886914e-01 -1.25825858e+00 -1.36410639e-01 -9.77047756e-02 6.96560323e-01 -3.46193880e-01 5.04503310e-01 -6.14266336e-01 -1.68116540e-01 -5.05810976e-01 -6.18985772e-01 9.44672585e-01 -3.22658986e-01 -1.09643829e+00 -1.31767535e+00 8.25577736e-01 7.73566186e-01 4.37078804e-01 3.42624277e-01 7.29306400e-01 -1.31865966e+00 3.29271883e-01 -2.74903998e-02 -1.02741152e-01 1.09505691e-01 -6.17979579e-02 -7.19615966e-02 -8.63486946e-01 -3.82597446e-01 -3.82087648e-01 -8.60642552e-01 7.25895762e-01 1.55056566e-01 6.36979222e-01 -3.09549794e-02 -3.13561469e-01 2.07752913e-01 1.30901587e+00 -3.72454464e-01 2.60071039e-01 5.46642303e-01 7.56667912e-01 7.34307528e-01 7.91709363e-01 2.37837061e-01 8.25612903e-01 1.13463867e+00 -2.16928959e-01 2.54506618e-01 -4.95178908e-01 -4.81670320e-01 6.00566506e-01 1.91337538e+00 -1.14598274e-01 -2.70490646e-01 -1.24996090e+00 6.67664766e-01 -2.03262877e+00 -7.69736886e-01 -7.97188103e-01 2.23846793e+00 8.60505939e-01 8.50223973e-02 2.47643277e-01 -7.45219141e-02 9.70181644e-01 1.21960416e-01 -4.10537906e-02 -8.34724188e-01 -6.95045233e-01 7.57025421e-01 4.63913500e-01 4.94144231e-01 -8.54369760e-01 1.40857959e+00 7.02102661e+00 1.19078529e+00 -6.01364136e-01 5.34214854e-01 5.11749871e-02 -1.07023949e-02 -4.63633239e-01 3.76843810e-01 -7.09194183e-01 4.50992495e-01 1.37095940e+00 -5.90736508e-01 1.67910963e-01 4.66641694e-01 -1.52119562e-01 -4.94794063e-02 -1.14387345e+00 6.55803978e-01 3.98115128e-01 -1.15555739e+00 -7.16357306e-02 -2.08487585e-01 9.30686831e-01 7.56534576e-01 -2.43566096e-01 2.68620133e-01 7.24194348e-01 -1.04894710e+00 8.41810167e-01 3.15006196e-01 8.08458090e-01 -6.63582385e-01 8.38769257e-01 1.32507728e-02 -1.58763778e+00 5.03741205e-01 -3.85798924e-02 -5.00513539e-02 2.13000029e-01 4.48236555e-01 -6.01659358e-01 8.76654446e-01 8.92756701e-01 1.31805813e+00 -7.26686180e-01 6.84108615e-01 -4.22438800e-01 5.91689825e-01 -1.15041420e-01 -6.34932593e-02 2.91897476e-01 -3.53998020e-02 6.85494125e-01 1.69083357e+00 3.62742484e-01 -3.51681948e-01 1.93427905e-01 4.23260093e-01 -1.11826785e-01 8.02518427e-01 -6.45068347e-01 -2.85673719e-02 9.34938967e-01 1.23611689e+00 -6.64380729e-01 -3.73442203e-01 -2.99826860e-01 1.11451387e+00 8.42619300e-01 -1.93084687e-01 -6.69252992e-01 -4.72104877e-01 2.95302540e-01 -1.06787637e-01 1.30280063e-01 -2.19294876e-01 -5.82322180e-01 -1.06851029e+00 -8.47013108e-03 -7.37306535e-01 9.44816232e-01 -8.49816144e-01 -1.74713862e+00 9.61567879e-01 -3.95603664e-02 -1.04966307e+00 -1.43213525e-01 -3.40530306e-01 -6.18315756e-01 9.21259999e-01 -1.16756451e+00 -1.33735216e+00 -6.67831972e-02 4.52188551e-01 6.41360402e-01 -4.48664397e-01 9.00027096e-01 3.78207833e-01 -4.08453375e-01 9.52013671e-01 1.23959333e-01 -2.32919589e-01 1.41630030e+00 -1.24633157e+00 8.94382179e-01 5.43767810e-01 3.65810931e-01 4.29513544e-01 6.57617211e-01 -7.49309897e-01 -8.18859458e-01 -7.54678190e-01 2.11553407e+00 -7.61209965e-01 1.45510614e+00 -1.72567397e-01 -9.70949352e-01 5.21734476e-01 7.97813058e-01 -4.13685799e-01 6.73453152e-01 2.83979177e-01 -8.48351240e-01 2.99711764e-01 -8.63654017e-01 3.41060907e-01 1.38458169e+00 -9.81500149e-01 -1.21259522e+00 6.31940901e-01 7.63105392e-01 -4.11116838e-01 -1.41839743e+00 3.53751779e-01 3.40263069e-01 -7.80129015e-01 5.96257031e-01 -6.67852283e-01 7.30999231e-01 3.70036438e-02 -2.56549001e-01 -1.57535076e+00 -2.21655950e-01 -5.30435562e-01 4.66865420e-01 1.49787700e+00 7.05782950e-01 -6.58907712e-01 1.90931916e-01 -1.13182068e-01 -5.75176716e-01 -5.26578307e-01 -1.01244068e+00 -1.12668312e+00 4.35910106e-01 -4.65256393e-01 2.59275824e-01 1.59689963e+00 6.70708716e-01 6.20907068e-01 1.63000189e-02 -4.86357272e-01 3.92725468e-01 -2.45987236e-01 3.61302346e-01 -1.18290889e+00 -1.11542329e-01 -7.33680010e-01 -4.36276674e-01 -2.46910617e-01 5.53374052e-01 -1.78012574e+00 -6.06886670e-02 -1.53729749e+00 4.79019046e-01 -2.45027184e-01 -2.67820865e-01 5.00310123e-01 -5.18925786e-02 3.13595414e-01 -1.90376133e-01 4.95808303e-01 -8.27016175e-01 3.42372268e-01 6.42835319e-01 1.18260711e-01 2.82232493e-01 -4.75562721e-01 -3.65457654e-01 3.76583755e-01 8.46791089e-01 -6.80267334e-01 -5.63082844e-02 -4.20238823e-01 3.83546799e-01 -3.45295668e-01 -6.31646663e-02 -7.76601315e-01 2.12527439e-01 7.58546367e-02 -2.55404174e-01 -6.24305665e-01 -1.18128344e-01 4.67582518e-04 1.76772073e-01 4.46129978e-01 -6.75749004e-01 4.99692857e-01 -2.32879873e-02 -5.17766960e-02 -4.36477542e-01 -2.70952284e-01 5.63774049e-01 -1.01159878e-01 -6.15635991e-01 8.19194913e-02 -1.98975042e-01 8.55122924e-01 6.66060925e-01 -1.21635132e-01 -4.07985568e-01 2.99968719e-02 -6.74629211e-01 1.54357761e-01 3.23815256e-01 8.54711235e-01 5.21380194e-02 -1.65173554e+00 -1.28753388e+00 -5.21675885e-01 6.96238518e-01 -8.43264818e-01 2.04165876e-02 1.15955198e+00 -2.80976295e-01 4.39874291e-01 -1.01892121e-01 -3.28210801e-01 -1.47758746e+00 7.20733777e-02 -1.66856706e-01 -4.89453614e-01 -4.65287805e-01 8.86511683e-01 -4.63839322e-01 -7.44981587e-01 -1.52839974e-01 1.81735680e-01 -3.21599275e-01 3.15116346e-01 2.18607083e-01 6.81538224e-01 5.98572373e-01 -1.05717885e+00 -5.67511618e-01 9.78928089e-01 -8.68288577e-02 -4.31418031e-01 1.17304587e+00 -1.74409389e-01 -5.22753119e-01 7.91948617e-01 1.49714875e+00 -8.77262279e-02 -1.24646381e-01 -7.13205993e-01 8.32141995e-01 -2.49787755e-02 2.59497412e-03 -9.01639163e-01 -5.18719733e-01 5.36207139e-01 1.85955420e-01 1.03146017e-01 3.91884804e-01 3.69935781e-01 1.02716184e+00 4.16136324e-01 2.28658110e-01 -1.66849482e+00 1.04725875e-01 1.07705963e+00 1.09927857e+00 -1.29439211e+00 4.19134498e-02 -5.04804254e-01 -9.86016214e-01 1.00050259e+00 4.38828737e-01 -5.38319675e-03 4.16927457e-01 1.17547043e-01 -1.08748809e-01 -4.50223386e-01 -8.81056368e-01 -3.55910152e-01 8.26070130e-01 4.59556699e-01 1.17133176e+00 1.06747009e-01 -1.12040508e+00 4.32383925e-01 -8.64086330e-01 -4.89098817e-01 1.42495975e-01 6.71588063e-01 -3.18202347e-01 -1.30535352e+00 1.19800419e-01 1.53711691e-01 -4.45417911e-01 -7.52086937e-01 -8.96793365e-01 8.35917056e-01 -8.72670412e-02 8.59147787e-01 3.56117934e-02 -5.90517461e-01 3.78396124e-01 2.54849583e-01 6.45428360e-01 -6.78434789e-01 -1.01088500e+00 -3.14081609e-01 8.83961320e-01 -7.36502528e-01 -4.61518645e-01 -1.07268059e+00 -1.18562961e+00 -4.92066056e-01 -1.13735914e-01 3.42087120e-01 8.21805418e-01 1.23986554e+00 7.73926266e-03 6.14828244e-02 6.12365425e-01 -2.73036867e-01 -2.85808176e-01 -1.39304399e+00 -3.80728662e-01 4.58935559e-01 -5.73626399e-01 -4.29835588e-01 -4.51578796e-01 -3.23693380e-02]
[10.903227806091309, 9.618900299072266]
579f4c9d-e896-4169-b8f9-fc237b0dc1fc
compound-multi-branch-feature-fusion-for-real-1
2206.02748
null
https://arxiv.org/abs/2206.02748v1
https://arxiv.org/pdf/2206.02748v1.pdf
Compound Multi-branch Feature Fusion for Real Image Restoration
Image restoration is a challenging and ill-posed problem which also has been a long-standing issue. However, most of learning based restoration methods are proposed to target one degradation type which means they are lack of generalization. In this paper, we proposed a multi-branch restoration model inspired from the Human Visual System (i.e., Retinal Ganglion Cells) which can achieve multiple restoration tasks in a general framework. The experiments show that the proposed multi-branch architecture, called CMFNet, has competitive performance results on four datasets, including image dehazing, deraindrop, and deblurring, which are very common applications for autonomous cars. The source code and pretrained models of three restoration tasks are available at https://github.com/FanChiMao/CMFNet.
['Kuan-Hsien Liu', 'Tsung-Jung Liu', 'Chi-Mao Fan']
2022-06-02
compound-multi-branch-feature-fusion-for-real
https://openreview.net/forum?id=WQIdU90Gsu
https://openreview.net/pdf?id=WQIdU90Gsu
null
['image-dehazing']
['computer-vision']
[ 9.55607221e-02 -3.62559944e-01 4.78391722e-02 -1.16267866e-02 -4.61395085e-01 9.63849872e-02 5.63189089e-01 -4.92665589e-01 -1.55700982e-01 7.02725291e-01 3.34073365e-01 -1.47677287e-01 1.51811376e-01 -6.94823742e-01 -7.48199582e-01 -1.05467236e+00 5.67005038e-01 -2.77257949e-01 3.94054770e-01 -4.16548312e-01 4.13975000e-01 2.13371575e-01 -1.65618598e+00 3.17478031e-01 1.12695217e+00 8.99127662e-01 2.85689145e-01 4.50906992e-01 1.92619860e-01 8.03984463e-01 -5.25985420e-01 -2.38663271e-01 8.19396377e-02 -4.68958288e-01 -4.21151966e-01 2.23951727e-01 5.01405895e-01 -5.20683408e-01 -8.03772688e-01 1.41589737e+00 4.41232800e-01 9.94769856e-02 7.03505456e-01 -1.28724122e+00 -1.26947188e+00 -5.36151901e-02 -5.67234516e-01 5.12139380e-01 -5.91840222e-02 3.34473521e-01 4.55553710e-01 -9.66890752e-01 1.28330261e-01 1.47503436e+00 4.85881478e-01 7.59327710e-01 -8.50548625e-01 -8.85973394e-01 9.46089700e-02 8.27903867e-01 -1.22005844e+00 -5.87979674e-01 5.27249515e-01 -3.82561684e-01 5.39407492e-01 6.71264380e-02 2.93747306e-01 8.77487659e-01 5.47203183e-01 8.74012113e-01 1.47154701e+00 -4.86289449e-02 -7.51238465e-02 -1.47678629e-01 1.06670909e-01 3.66379708e-01 2.40175679e-01 4.01190847e-01 -3.66423815e-01 3.10166240e-01 7.69808769e-01 2.98745155e-01 -7.15224504e-01 1.15315311e-01 -1.01893711e+00 6.69379830e-01 8.21576059e-01 5.72442673e-02 -3.06220978e-01 2.91649085e-02 3.01163733e-01 2.76563168e-01 5.78818738e-01 -3.37281883e-01 -6.94086552e-02 4.09665674e-01 -4.17236835e-01 1.81792378e-01 2.82510519e-01 7.14893460e-01 8.25672209e-01 4.17488515e-01 -4.23802882e-02 9.55722690e-01 5.74516773e-01 5.96581817e-01 8.14543366e-01 -8.99049759e-01 9.24957767e-02 2.74260610e-01 1.02412097e-01 -1.14069712e+00 -1.98581055e-01 -5.11526108e-01 -1.29743934e+00 6.78329408e-01 1.74908899e-03 5.73937073e-02 -1.13567448e+00 1.51137054e+00 2.96812594e-01 6.29967034e-01 4.71016854e-01 1.19645786e+00 1.25016320e+00 1.00523484e+00 -3.99957113e-02 -1.97995752e-01 1.32078576e+00 -1.36507761e+00 -8.96429598e-01 -4.07307446e-01 -4.83310036e-02 -1.03729784e+00 9.90686238e-01 7.46725321e-01 -7.73184180e-01 -8.77927840e-01 -1.32194340e+00 -3.02164018e-01 -8.87288153e-02 1.86468169e-01 3.33319455e-01 3.54483902e-01 -9.72328663e-01 4.90925789e-01 -6.01190984e-01 -4.58169729e-01 6.08318806e-01 -1.05592132e-01 -3.74710321e-01 -6.29322767e-01 -1.09163165e+00 1.01516891e+00 2.51693457e-01 4.51902270e-01 -1.33185875e+00 -1.52848780e-01 -5.06536007e-01 -1.15802325e-01 1.54095784e-01 -5.39952338e-01 1.21158135e+00 -9.73470271e-01 -1.36649239e+00 6.35803461e-01 -6.90364614e-02 -3.91694725e-01 3.36072087e-01 -3.81442428e-01 -7.60328293e-01 1.71901695e-02 -1.01870790e-01 5.29837906e-01 1.03854787e+00 -1.43318367e+00 -5.63990951e-01 -3.70810956e-01 -1.17676355e-01 3.56168747e-01 -3.04925770e-01 -7.60984421e-02 -1.36112645e-01 -7.12807178e-01 1.36079744e-01 -6.38270855e-01 -4.54786755e-02 4.71129566e-02 -2.78570980e-01 -1.19875483e-01 1.05822086e+00 -1.11990762e+00 9.82440472e-01 -2.36415744e+00 1.85713731e-02 -4.49576497e-01 7.26561248e-02 6.72772169e-01 -2.75841743e-01 3.25919062e-01 -9.53412578e-02 9.75243896e-02 -4.22332704e-01 -1.14212796e-01 -3.02831799e-01 2.29385123e-01 -4.44722652e-01 6.08576417e-01 -1.10498026e-01 6.41281247e-01 -6.08170509e-01 -1.84263363e-01 3.80686343e-01 7.37951279e-01 -1.11484274e-01 3.26070309e-01 2.01492105e-02 6.38292313e-01 -2.16425925e-01 6.38948619e-01 1.12710607e+00 2.93165427e-02 -4.38071102e-01 -3.49758625e-01 -9.78488252e-02 -1.89786151e-01 -9.55082178e-01 1.44673181e+00 -1.59118876e-01 6.09474301e-01 3.54368910e-02 -1.02487767e+00 1.04412019e+00 2.89257437e-01 6.38738275e-02 -8.37649882e-01 1.77165106e-01 2.84651786e-01 -5.09789027e-02 -6.46050334e-01 1.91591695e-01 -2.88485974e-01 4.88600314e-01 1.10922344e-01 -7.53968656e-02 6.77862987e-02 8.79496336e-02 -1.95869058e-02 9.44873095e-01 1.38885051e-01 2.06006244e-01 -2.11975977e-01 7.83022523e-01 -2.86631808e-02 8.45505297e-01 2.39774317e-01 -3.38110924e-01 9.59469736e-01 7.30334520e-02 -3.08401257e-01 -8.76656950e-01 -9.94391024e-01 -1.79068848e-01 6.16330922e-01 6.19741261e-01 2.06041560e-02 -8.55686784e-01 -2.54239172e-01 -2.79960871e-01 6.02796555e-01 -2.28829071e-01 -5.75578690e-01 -3.58305126e-01 -1.03064549e+00 2.64621139e-01 1.25148803e-01 1.16033196e+00 -1.27818203e+00 -2.01971143e-01 2.41950676e-02 -5.39064407e-01 -9.83112752e-01 -6.37822330e-01 -3.66051078e-01 -8.78386259e-01 -1.19023812e+00 -8.94069433e-01 -1.08317792e+00 5.88545322e-01 1.02918077e+00 8.46731722e-01 4.19583797e-01 -1.14560574e-01 4.98098042e-03 -4.44624990e-01 -4.05866832e-01 -3.67468506e-01 -5.23141980e-01 1.28074810e-02 3.19608659e-01 2.98293710e-01 -8.37428272e-01 -8.70397151e-01 5.86911321e-01 -1.34238780e+00 1.10973440e-01 8.52992415e-01 9.73101139e-01 6.58680499e-01 3.84237617e-01 7.87651539e-01 -4.89558399e-01 7.17876315e-01 -4.80022520e-01 -5.32567143e-01 6.41293824e-02 -7.13990331e-01 -3.04154128e-01 5.33254802e-01 -3.92298371e-01 -1.27079976e+00 -2.93623328e-01 -3.00364375e-01 -5.09206772e-01 -4.59656954e-01 4.71747994e-01 -4.54213530e-01 -1.90360650e-01 7.33405173e-01 5.77389598e-01 3.20271403e-01 -7.91873515e-01 1.16549022e-01 9.23999071e-01 8.26775074e-01 -1.53043345e-01 8.84026051e-01 3.79389852e-01 -2.68327862e-01 -7.45519578e-01 -7.19644666e-01 -2.78122038e-01 -1.96519181e-01 -3.52932394e-01 8.43410611e-01 -1.24830663e+00 -4.96740878e-01 1.20649409e+00 -1.23888457e+00 -3.48896623e-01 1.53415963e-01 4.28722620e-01 -3.93571317e-01 5.54298103e-01 -6.40725076e-01 -3.97448152e-01 -5.05089819e-01 -1.05710161e+00 6.09072149e-01 7.96441138e-01 7.60563433e-01 -6.11882985e-01 -8.81165043e-02 5.95384836e-01 5.60010135e-01 1.64480768e-02 7.74817765e-01 -2.04450488e-01 -8.96843076e-01 3.72943543e-02 -4.71664101e-01 9.55535769e-01 1.79910704e-01 -2.34165192e-01 -9.40897644e-01 -5.01918316e-01 3.10772985e-01 -2.71271169e-01 1.26716185e+00 4.04662907e-01 1.11537790e+00 -3.62403810e-01 -8.28195512e-02 6.24671936e-01 1.44453502e+00 2.28016198e-01 1.25437474e+00 4.24240202e-01 6.02388918e-01 3.92989010e-01 6.67527556e-01 8.08217302e-02 4.15484220e-01 4.01159823e-01 8.26203525e-01 -2.54563630e-01 -5.71597576e-01 -9.47780684e-02 5.89713931e-01 9.85646546e-01 -1.22093178e-01 -3.82570773e-01 -6.62003696e-01 4.18114990e-01 -2.05004001e+00 -9.35305059e-01 -4.04408842e-01 1.95638335e+00 5.69371223e-01 -7.75018800e-03 -1.78090289e-01 1.12426378e-01 9.29551005e-01 1.89109892e-01 -7.96081007e-01 -7.86871761e-02 -3.02586347e-01 -1.36085853e-01 4.02347803e-01 4.39339519e-01 -1.03906763e+00 8.62016618e-01 5.58246374e+00 9.48256135e-01 -1.02754581e+00 4.45891291e-01 7.80123413e-01 3.82195979e-01 -4.74391840e-02 4.15437249e-03 -6.42746389e-01 5.89521170e-01 5.50418556e-01 -2.24924207e-01 6.00685835e-01 5.24035573e-01 2.66988546e-01 -5.66341691e-02 -4.84001994e-01 1.12578630e+00 2.40728587e-01 -9.68357682e-01 1.50609583e-01 -5.02919853e-02 7.04235256e-01 1.61665455e-01 2.52027959e-01 2.54393578e-01 2.47648403e-01 -1.12601459e+00 3.73319238e-01 6.97870374e-01 4.88275141e-01 -6.68256223e-01 7.68243432e-01 5.10930479e-01 -8.64044309e-01 -2.24273652e-01 -8.29583228e-01 -3.02749630e-02 1.84681788e-01 7.52035141e-01 -8.29125494e-02 6.28854752e-01 9.87814307e-01 1.18351328e+00 -5.96501350e-01 1.56921351e+00 -4.88305420e-01 6.52916670e-01 1.99392706e-01 4.77621287e-01 -9.76900756e-02 -3.98688406e-01 5.73455334e-01 8.94575477e-01 5.33374190e-01 1.66285574e-01 1.43472001e-01 6.16501391e-01 -5.93132153e-03 -1.03882477e-01 -3.89956504e-01 3.01770240e-01 2.02509314e-01 1.43015313e+00 -3.84407192e-01 -1.40144810e-01 -5.10151446e-01 8.65710914e-01 -1.20831653e-01 6.29614651e-01 -9.31273580e-01 -3.23490858e-01 7.20645010e-01 -6.02404699e-02 1.59470186e-01 -4.47641835e-02 -5.26181087e-02 -1.25500107e+00 8.74882117e-02 -1.16673052e+00 2.75020242e-01 -1.27379739e+00 -1.42461562e+00 7.65685499e-01 -2.28026927e-01 -1.50980592e+00 4.02106583e-01 -6.61581337e-01 -7.64741361e-01 9.57037210e-01 -1.99943841e+00 -1.24766362e+00 -8.60873282e-01 7.65868247e-01 6.77923918e-01 -3.27620447e-01 4.58258212e-01 5.94158471e-01 -7.29699314e-01 2.39106029e-01 3.75290185e-01 -2.96550486e-02 8.76283467e-01 -7.84907699e-01 2.07521468e-01 1.28229988e+00 -1.88802540e-01 2.71204263e-01 8.03430200e-01 -5.53361535e-01 -1.06031799e+00 -1.27928567e+00 2.87103444e-01 1.69615731e-01 3.45777214e-01 1.66894451e-01 -1.22815847e+00 6.46020710e-01 6.81809843e-01 1.01131879e-01 1.08598970e-01 -4.56160665e-01 -3.35296988e-01 -5.61071336e-01 -1.07789373e+00 4.86190438e-01 8.98595989e-01 -2.30849490e-01 -3.99818569e-01 4.87393469e-01 7.05012381e-01 -3.64604026e-01 -5.18133223e-01 4.62190837e-01 1.62367612e-01 -1.05869877e+00 1.11708510e+00 -1.86880976e-01 5.77206314e-01 -7.10493267e-01 -2.99074322e-01 -1.49023116e+00 -4.20203596e-01 -2.96367198e-01 -1.16498254e-01 1.27547479e+00 -9.75478292e-02 -9.73737061e-01 3.31796288e-01 -4.98764217e-02 -5.43450058e-01 -5.60727477e-01 -8.78393948e-01 -7.73565888e-01 1.37251243e-01 1.93287775e-01 4.56945658e-01 6.36956453e-01 -6.43998086e-01 4.58940387e-01 -7.24323332e-01 3.50447148e-01 7.26095259e-01 -3.22815441e-02 6.47578299e-01 -1.04096878e+00 -1.54508039e-01 -3.70197296e-01 -4.40565497e-01 -1.24447417e+00 -2.40426101e-02 -7.60877788e-01 2.85272211e-01 -1.89538908e+00 2.15977520e-01 -1.63160041e-01 -4.96650845e-01 4.17010009e-01 -2.62246102e-01 4.20635074e-01 6.44858032e-02 5.57007730e-01 -3.30425709e-01 8.73769581e-01 1.55393565e+00 -3.81077915e-01 1.80728585e-01 7.14350343e-02 -9.83574450e-01 7.02784598e-01 1.01812410e+00 -3.16141367e-01 -4.75363016e-01 -7.36753464e-01 -3.30491453e-01 -6.85820952e-02 7.48742640e-01 -1.32228220e+00 3.79165918e-01 -2.66565889e-01 3.17060977e-01 -4.16967511e-01 4.39613700e-01 -6.07197404e-01 2.58466363e-01 5.47357380e-01 -9.54982359e-03 3.00518870e-02 1.83105901e-01 7.69751132e-01 -5.84854841e-01 -2.06119865e-01 1.16239762e+00 -2.27654576e-01 -1.04921484e+00 5.37083149e-01 -3.51793110e-01 -2.61434883e-01 8.62840593e-01 5.65491943e-03 -8.87113810e-01 -4.15907800e-01 -5.06732166e-01 1.50045946e-01 3.32573175e-01 5.08445919e-01 1.09700072e+00 -1.39943469e+00 -1.08153284e+00 -5.85487336e-02 -7.73959458e-02 5.53704053e-02 7.29499340e-01 7.79546559e-01 -5.14780641e-01 -1.15524177e-02 -6.07364774e-01 -4.63025510e-01 -1.37608039e+00 7.80247688e-01 5.59850216e-01 1.41915798e-01 -8.41665328e-01 7.25484312e-01 5.39590538e-01 -3.35631996e-01 4.46434580e-02 5.84209748e-02 -6.02708936e-01 -3.19001198e-01 8.13223779e-01 4.57576334e-01 1.62585810e-01 -8.99507463e-01 -1.28571495e-01 6.13356292e-01 -1.39656112e-01 2.37353593e-01 1.21076441e+00 -3.85305464e-01 -3.65554422e-01 2.88790483e-02 9.93525565e-01 -4.74472404e-01 -1.35505414e+00 -5.07730424e-01 -3.65574449e-01 -6.40752494e-01 2.39525855e-01 -7.85815477e-01 -1.42845464e+00 1.07848585e+00 1.11164212e+00 -7.74425408e-03 1.53091395e+00 -3.52977067e-01 9.61823702e-01 2.22997740e-01 1.48453563e-01 -6.82758212e-01 4.16134655e-01 3.34258884e-01 1.28900123e+00 -1.32436633e+00 9.72912535e-02 -4.63738143e-01 -6.21947527e-01 9.16144133e-01 9.72682357e-01 -3.76188219e-01 6.77983224e-01 -3.01624417e-01 4.17350948e-01 8.92893299e-02 -6.70483708e-01 -1.88782737e-01 -5.60197979e-03 8.18097949e-01 8.83145034e-02 -8.83792788e-02 -4.49264437e-01 4.71616685e-01 6.29132167e-02 9.02660936e-02 8.34516048e-01 6.13236427e-01 -7.60359824e-01 -9.79259431e-01 -6.48041427e-01 4.03871059e-01 -2.27400407e-01 -8.55557919e-02 1.66153535e-01 3.92419577e-01 2.24286392e-01 1.43306994e+00 -2.83049941e-01 -6.58734083e-01 2.67926723e-01 -4.41805750e-01 2.24520251e-01 -3.06248665e-01 -4.50753607e-02 1.00632377e-01 -2.46264607e-01 -3.14704478e-01 -6.31539702e-01 -5.00388205e-01 -1.04359901e+00 -4.49331045e-01 -1.89244419e-01 -3.11258789e-02 3.38189960e-01 8.59706819e-01 3.14718485e-01 7.24664927e-01 6.25117779e-01 -7.57365882e-01 -5.24385750e-01 -1.15200996e+00 -6.57426476e-01 3.33844930e-01 6.06810033e-01 -7.42209375e-01 -3.88648719e-01 2.81756520e-01]
[11.07313060760498, -2.618635892868042]
8f56e30b-87e4-44e2-a812-c6ec8ed70437
vision-through-the-veil-differential-privacy
2306.17794
null
https://arxiv.org/abs/2306.17794v1
https://arxiv.org/pdf/2306.17794v1.pdf
Vision Through the Veil: Differential Privacy in Federated Learning for Medical Image Classification
The proliferation of deep learning applications in healthcare calls for data aggregation across various institutions, a practice often associated with significant privacy concerns. This concern intensifies in medical image analysis, where privacy-preserving mechanisms are paramount due to the data being sensitive in nature. Federated learning, which enables cooperative model training without direct data exchange, presents a promising solution. Nevertheless, the inherent vulnerabilities of federated learning necessitate further privacy safeguards. This study addresses this need by integrating differential privacy, a leading privacy-preserving technique, into a federated learning framework for medical image classification. We introduce a novel differentially private federated learning model and meticulously examine its impacts on privacy preservation and model performance. Our research confirms the existence of a trade-off between model accuracy and privacy settings. However, we demonstrate that strategic calibration of the privacy budget in differential privacy can uphold robust image classification performance while providing substantial privacy protection.
['Balasubramanian Raman', 'Uppala Vivek Narayan', 'Pradeep Singh', 'Kishore Babu Nampalle']
2023-06-30
null
null
null
null
['medical-image-classification']
['medical']
[ 4.28582020e-02 2.54702389e-01 -2.24186808e-01 -5.35204709e-01 -8.92233670e-01 -8.38557422e-01 2.21234202e-01 4.95972425e-01 -6.09512925e-01 6.42293990e-01 1.94773719e-01 -7.72738576e-01 -3.40819627e-01 -7.36366451e-01 -5.10745168e-01 -9.61708069e-01 -2.12983176e-01 -3.08455765e-01 -6.06671453e-01 3.95966738e-01 -1.10280454e-01 7.31989384e-01 -1.09460723e+00 3.94308418e-01 8.65767777e-01 1.04724216e+00 -4.82408524e-01 3.47801000e-01 -6.21071644e-02 8.21635067e-01 -4.29080784e-01 -1.08739316e+00 8.44283581e-01 -1.58145070e-01 -7.82485902e-01 -2.05167398e-01 3.11298132e-01 -6.30393088e-01 -4.61855590e-01 1.13961935e+00 4.60576445e-01 -3.40322942e-01 -1.66456942e-02 -1.55777347e+00 -4.25183684e-01 4.03590202e-01 -2.96222359e-01 3.88587415e-02 -1.87511116e-01 1.42009616e-01 9.61803377e-01 -1.12152971e-01 5.40990353e-01 4.92435127e-01 7.56952882e-01 6.66130126e-01 -1.12968707e+00 -7.95559347e-01 -1.56760335e-01 -1.51892811e-01 -1.17031133e+00 -4.36472714e-01 6.28274024e-01 -2.47438073e-01 4.02924657e-01 8.79842162e-01 6.13240302e-01 9.01359141e-01 5.24712443e-01 6.07022405e-01 1.22109210e+00 -3.14041317e-01 4.08866316e-01 5.72517157e-01 6.19905256e-02 4.83174324e-01 7.95012414e-01 2.58606941e-01 -4.65972900e-01 -9.12443161e-01 5.16314864e-01 4.21138197e-01 -3.99993330e-01 -7.20608771e-01 -7.30239034e-01 6.35407805e-01 2.77721047e-01 1.08377270e-01 -4.17941809e-01 -6.55060634e-02 7.06497788e-01 5.09588420e-01 3.35980177e-01 3.04469436e-01 -6.12389028e-01 1.47965744e-01 -6.12891018e-01 1.59605116e-01 8.58690858e-01 7.88425803e-01 5.77548504e-01 -4.75770861e-01 -1.21355116e-01 2.82938778e-01 1.82185128e-01 -2.26942733e-01 3.87589902e-01 -1.24484503e+00 3.42772871e-01 8.04060757e-01 -1.37208626e-02 -1.06142473e+00 1.12569392e-01 -6.76029682e-01 -8.57547998e-01 3.14642578e-01 3.86070997e-01 -4.24880862e-01 -3.55508029e-02 1.83598161e+00 6.33427680e-01 -2.23158777e-01 4.63141888e-01 6.32282972e-01 1.55724781e-02 -8.54536593e-02 4.72481191e-01 -2.50648677e-01 1.42595208e+00 -5.69327950e-01 -7.91468441e-01 1.65111408e-01 8.62330914e-01 -2.43349195e-01 6.60362661e-01 2.34216735e-01 -8.73111308e-01 3.91452283e-01 -8.83139968e-01 -1.26320515e-02 -2.89925247e-01 -4.26504523e-01 1.13622653e+00 1.33152354e+00 -9.91073132e-01 4.55574006e-01 -1.00797009e+00 -3.16000909e-01 1.03463972e+00 5.99602759e-01 -8.00113022e-01 5.60991094e-03 -1.01708150e+00 4.51489151e-01 2.16515772e-02 -1.12715475e-01 -3.46882701e-01 -1.03671563e+00 -6.44815624e-01 2.61734784e-01 -7.51328189e-04 -9.41945791e-01 1.07559037e+00 -9.96024728e-01 -7.09858716e-01 1.16571271e+00 4.40894455e-01 -9.50387180e-01 1.08238339e+00 3.35997939e-01 -4.81720716e-01 4.44070138e-02 -1.56593904e-01 1.90265447e-01 4.08292443e-01 -1.27037311e+00 -1.00113845e+00 -9.03156459e-01 1.23680033e-01 9.16089714e-02 -8.72076809e-01 1.93501811e-03 1.57616839e-01 -4.88786131e-01 -2.12989315e-01 -6.31642282e-01 -5.44680655e-01 4.18459356e-01 -1.14515685e-01 3.35944712e-01 1.04842389e+00 -6.15445316e-01 1.24941957e+00 -2.48410439e+00 -7.83396900e-01 2.73300350e-01 5.78380227e-01 2.41149776e-03 2.35596567e-01 2.40849748e-01 2.70762116e-01 3.96803945e-01 -2.97887623e-01 -4.29145515e-01 -3.45147178e-02 1.81629598e-01 -5.40876500e-02 7.37471223e-01 -2.82887161e-01 9.01496291e-01 -4.85749543e-01 -6.88022494e-01 -9.54723060e-02 5.62226355e-01 -6.40795350e-01 1.40152842e-01 1.70767292e-01 5.17040610e-01 -7.75597036e-01 7.72273540e-01 9.17925477e-01 -4.31697279e-01 6.88594460e-01 -7.79138952e-02 -1.15777493e-01 -1.63523838e-01 -7.08223343e-01 1.46440089e+00 -1.89010605e-01 1.45537123e-01 8.02531183e-01 -6.45040393e-01 6.37788177e-01 5.94560564e-01 9.21011329e-01 -4.56782967e-01 3.22962373e-01 1.03054687e-01 -2.05351263e-01 -5.37710190e-01 2.96898454e-01 -2.07735270e-01 -4.11485918e-02 5.19001484e-01 -3.44204336e-01 7.48287976e-01 -7.91269720e-01 1.57588601e-01 1.29926503e+00 -4.02894139e-01 3.90713900e-01 -5.39889097e-01 3.30916911e-01 -6.16797293e-03 9.07941580e-01 4.94044483e-01 -8.91187012e-01 3.05250764e-01 4.21413124e-01 -6.46737456e-01 -6.14452660e-01 -6.58559322e-01 -2.76860803e-01 6.51942313e-01 -3.24147902e-02 -3.00773531e-01 -7.94699311e-01 -1.12593532e+00 6.03392363e-01 3.57808948e-01 -6.81996703e-01 -3.83431405e-01 -2.54746228e-01 -6.78330243e-01 8.52361202e-01 1.08468451e-01 5.83630681e-01 -4.92921263e-01 -1.20717990e+00 -5.84366545e-02 1.79773457e-02 -8.71833324e-01 -5.12735546e-01 2.65898228e-01 -1.06308782e+00 -1.24065530e+00 -3.60007256e-01 -3.94133449e-01 7.98405051e-01 2.32101068e-01 7.26238072e-01 1.42525733e-01 -4.36385125e-01 6.72771394e-01 8.77184700e-03 -5.76864421e-01 -5.55717051e-01 -6.38292655e-02 -1.76177353e-01 3.84370685e-01 5.40917516e-01 -6.01353824e-01 -9.79845643e-01 -1.99354207e-03 -1.03341007e+00 -2.90457428e-01 2.88062751e-01 7.41351724e-01 4.07569975e-01 1.14611275e-01 5.93780100e-01 -1.29957318e+00 7.10503697e-01 -6.21837974e-01 -5.43638170e-01 6.36031210e-01 -1.17210448e+00 -2.35280499e-01 6.33501530e-01 2.65965592e-02 -1.21979117e+00 3.28457654e-01 1.33701101e-01 -4.51602429e-01 -3.08594733e-01 3.84682953e-01 -4.44882423e-01 -6.14729226e-01 5.60163498e-01 -3.45754102e-02 6.68337107e-01 -4.47119325e-01 5.88897951e-02 9.44830358e-01 4.28878129e-01 -4.08641607e-01 3.09861213e-01 8.29710901e-01 -5.40405232e-03 -3.02869499e-01 -3.45635623e-01 -3.17360520e-01 -1.98223323e-01 5.05160391e-02 6.02241695e-01 -9.70531821e-01 -1.03404951e+00 2.74617702e-01 -7.94183314e-01 2.25570723e-01 -4.31529552e-01 4.67856586e-01 -4.36908633e-01 5.50148785e-01 -6.16278052e-01 -8.15385401e-01 -8.01876664e-01 -8.85514438e-01 2.90014684e-01 8.89191702e-02 -2.71112978e-01 -9.45805907e-01 -9.15442500e-03 8.24334323e-01 6.08880520e-01 7.18023539e-01 9.40276623e-01 -8.68936121e-01 -7.22537994e-01 -5.03683865e-01 1.08082481e-01 1.80013701e-01 5.38121998e-01 -3.88877869e-01 -1.13193643e+00 -6.84906363e-01 4.50843722e-01 -2.01730892e-01 1.17776625e-01 1.38185918e-01 1.35299480e+00 -9.85808313e-01 -3.53053957e-01 9.43742990e-01 1.60251880e+00 1.13779336e-01 3.96463811e-01 5.88971674e-01 3.37407768e-01 9.09538448e-01 3.58944446e-01 9.70695257e-01 4.20791715e-01 4.47276346e-02 6.40274167e-01 -2.20578074e-01 6.32094920e-01 -2.16340631e-01 -2.39955857e-01 1.92848042e-01 4.40961272e-01 1.18683673e-01 -8.82329345e-01 4.56111133e-01 -1.77235508e+00 -6.71093464e-01 1.33554503e-01 2.42123938e+00 9.30959463e-01 -6.46284282e-01 8.61926451e-02 6.35454198e-04 4.51370806e-01 -8.34666640e-02 -7.50786841e-01 -4.27603215e-01 -2.15943053e-01 -2.20570073e-01 1.09142315e+00 -8.88009276e-03 -8.65283608e-01 2.78639108e-01 6.03859663e+00 1.55170619e-01 -1.17527854e+00 3.49247724e-01 1.16477799e+00 -1.58886671e-01 -5.60007453e-01 -9.20778885e-02 -7.16805160e-02 3.95117253e-01 8.64494681e-01 -8.52571487e-01 4.30007353e-02 9.18320715e-01 1.21582843e-01 3.43114197e-01 -1.10135639e+00 1.01776147e+00 -3.73429388e-01 -1.57748699e+00 -6.43288866e-02 7.99308419e-01 4.51103628e-01 -2.56884813e-01 2.86469460e-01 -2.22206071e-01 3.90383005e-01 -8.41112316e-01 3.30575436e-01 2.22612843e-01 6.31914735e-01 -1.12375224e+00 5.03997982e-01 4.03474838e-01 -6.05957389e-01 -4.23806518e-01 -1.01216234e-01 1.91570073e-01 -1.79903150e-01 2.58313537e-01 -5.56352794e-01 6.50070667e-01 8.77170920e-01 1.61639959e-01 -4.29272741e-01 9.30677474e-01 5.27757585e-01 4.25937831e-01 2.99806315e-02 5.13824642e-01 4.60587032e-02 -2.21156269e-01 2.22967744e-01 9.37870860e-01 1.92198008e-01 2.50921220e-01 -3.04436117e-01 6.53406799e-01 -3.30272794e-01 2.88352519e-01 -9.37468648e-01 -9.66692343e-03 7.13954568e-01 1.29932320e+00 -2.16663837e-01 2.07884908e-01 -6.11755133e-01 1.04025292e+00 1.80297971e-01 4.77005094e-02 -4.49119151e-01 7.37507711e-04 1.31690979e+00 3.99696603e-02 -2.12266430e-01 1.73593894e-01 -8.49599600e-01 -1.04021037e+00 1.75283805e-01 -1.21969533e+00 1.03537333e+00 3.66414696e-01 -1.39664090e+00 3.08463216e-01 -4.30241287e-01 -1.24673450e+00 3.95239294e-02 -3.81671973e-02 -2.78766960e-01 8.24405074e-01 -1.52941513e+00 -1.38404870e+00 -8.08725059e-02 9.50681448e-01 -4.40965414e-01 -2.01689437e-01 1.09900129e+00 3.99038404e-01 -5.24605989e-01 1.28611636e+00 2.65687853e-01 9.56608802e-02 5.07080972e-01 -7.86771536e-01 1.22448914e-01 7.25056469e-01 -2.32570976e-01 1.06943333e+00 3.96239698e-01 -4.84413683e-01 -1.85163379e+00 -1.30154526e+00 9.18311179e-01 -3.84983093e-01 2.50709057e-01 -3.54309261e-01 -1.07924473e+00 8.33117783e-01 1.47749841e-01 4.63790148e-01 1.38270235e+00 -9.69818085e-02 -4.36147034e-01 -3.51473987e-01 -2.06827688e+00 4.85236794e-01 7.13816106e-01 -7.46465266e-01 1.76525965e-01 5.55278324e-02 8.92526567e-01 -9.24110413e-02 -1.13119221e+00 2.91374624e-01 7.20339715e-01 -1.13533533e+00 7.11924016e-01 -8.11796129e-01 -1.00238517e-01 8.03201422e-02 -2.89118350e-01 -5.64146042e-01 -1.74272075e-01 -9.36858475e-01 -2.35316396e-01 1.39277780e+00 3.01974088e-01 -1.17675686e+00 1.50031781e+00 1.81721628e+00 4.82676893e-01 -7.47715175e-01 -1.27855921e+00 -7.70602167e-01 2.87205726e-01 -1.77922711e-01 1.09096360e+00 1.61453533e+00 2.32889459e-01 -7.08729446e-01 -3.60534281e-01 3.53313267e-01 1.12291420e+00 -1.96592864e-02 6.31201148e-01 -1.03059542e+00 -1.75154120e-01 -5.70497625e-02 -5.30325174e-01 -8.41665044e-02 1.79247260e-01 -8.48632455e-01 -6.19982660e-01 -9.75041449e-01 1.75130382e-01 -7.72282660e-01 -7.21314132e-01 7.78562307e-01 9.51384678e-02 -1.68231487e-01 2.23298073e-01 3.30624342e-01 -1.65548742e-01 2.29159236e-01 7.82094181e-01 -4.01943699e-02 -6.13688007e-02 2.44002923e-01 -1.35964417e+00 1.49786040e-01 1.03679943e+00 -4.31631595e-01 -5.61740637e-01 -4.86480087e-01 -7.14945719e-02 2.67433465e-01 4.00065303e-01 -8.13056111e-01 6.08572483e-01 -2.59103149e-01 3.28092463e-02 -6.14612140e-02 -1.65827736e-01 -1.66892838e+00 7.03561842e-01 7.30081737e-01 -4.19348180e-01 8.34518597e-02 1.06335431e-01 7.69037843e-01 -2.54714519e-01 3.09204936e-01 7.42780030e-01 -1.36696219e-01 -3.24962586e-01 6.63263917e-01 -1.24101080e-01 -3.58528674e-01 1.48814929e+00 -4.91505295e-01 -2.65116692e-01 -1.06442034e-01 -4.12098438e-01 4.23046768e-01 9.12624776e-01 1.97266296e-01 4.28476095e-01 -1.02329159e+00 -4.04361904e-01 3.79637957e-01 2.22442031e-01 -1.47579238e-01 6.15752161e-01 7.10778594e-01 -2.86874384e-01 2.86499262e-01 -2.16778874e-01 -1.60447747e-01 -1.63553619e+00 7.41056919e-01 5.80628514e-01 -1.14305250e-01 -8.62850010e-01 5.41056812e-01 7.88154006e-02 -4.15370315e-01 5.90298355e-01 1.91539228e-01 4.27244753e-01 -1.57417640e-01 4.59707081e-01 4.44549382e-01 3.38462442e-01 -2.61077076e-01 -4.93298143e-01 -3.45875889e-01 -2.40070239e-01 1.10346839e-01 1.19090164e+00 -3.59695196e-01 -3.46096903e-01 -2.14406386e-01 1.50861728e+00 2.64046453e-02 -1.39125431e+00 -8.21458548e-02 1.55067191e-01 -9.12741065e-01 1.14009030e-01 -8.54278564e-01 -1.33991957e+00 2.35174671e-01 8.30739737e-01 -6.38765376e-03 1.29498041e+00 -2.96058565e-01 7.42799163e-01 8.88826773e-02 7.99210250e-01 -7.29934692e-01 -8.53513241e-01 -3.33074033e-01 3.58230263e-01 -1.28222501e+00 -6.97117075e-02 -2.76444554e-01 -5.68320215e-01 6.08495653e-01 5.63663065e-01 4.42425191e-01 7.15260148e-01 5.96423090e-01 5.34111202e-01 -1.59836173e-01 -8.22874963e-01 7.53489792e-01 -4.83071208e-01 5.57774663e-01 3.31137955e-01 2.06729248e-01 -5.19105256e-01 9.60015357e-01 -5.19131459e-02 1.80638149e-01 2.51713097e-01 1.58576536e+00 2.98847020e-01 -1.41435158e+00 -3.39763075e-01 5.17992079e-01 -1.05944157e+00 2.06931487e-01 -4.97937530e-01 5.47887862e-01 2.88360361e-02 8.25025499e-01 7.00940983e-03 -2.26087525e-01 1.49298206e-01 1.20739326e-01 4.74320948e-02 -4.47618961e-02 -1.29714680e+00 -3.83059382e-01 -2.21373543e-01 -5.83503246e-01 -1.29581466e-01 -6.38022244e-01 -1.11925077e+00 -4.85204905e-01 -1.35637611e-01 4.48376268e-01 7.77605712e-01 4.19933617e-01 1.04280627e+00 1.24181034e-02 9.93773937e-01 2.79136479e-01 -1.12959540e+00 1.09188855e-01 -9.35427308e-01 4.13146198e-01 8.01536620e-01 2.98883021e-01 -3.72829676e-01 -1.09275982e-01]
[5.98087739944458, 6.607274532318115]
75ab4984-3701-49f3-899a-de0b9feb25c9
skill-critic-refining-learned-skills-for
2306.08388
null
https://arxiv.org/abs/2306.08388v2
https://arxiv.org/pdf/2306.08388v2.pdf
Skill-Critic: Refining Learned Skills for Reinforcement Learning
Hierarchical reinforcement learning (RL) can accelerate long-horizon decision-making by temporally abstracting a policy into multiple levels. Promising results in sparse reward environments have been seen with skills, i.e. sequences of primitive actions. Typically, a skill latent space and policy are discovered from offline data, but the resulting low-level policy can be unreliable due to low-coverage demonstrations or distribution shifts. As a solution, we propose fine-tuning the low-level policy in conjunction with high-level skill selection. Our Skill-Critic algorithm optimizes both the low and high-level policies; these policies are also initialized and regularized by the latent space learned from offline demonstrations to guide the joint policy optimization. We validate our approach in multiple sparse RL environments, including a new sparse reward autonomous racing task in Gran Turismo Sport. The experiments show that Skill-Critic's low-level policy fine-tuning and demonstration-guided regularization are essential for optimal performance. Images and videos are available at https://sites.google.com/view/skill-critic. We plan to open source the code with the final version.
['Wei Zhan', 'Masayoshi Tomizuka', 'Kenta Kawamoto', 'Chen Tang', 'Catherine Weaver', 'Ce Hao']
2023-06-14
null
null
null
null
['hierarchical-reinforcement-learning']
['methodology']
[-3.20611931e-02 -2.48545483e-02 -4.79071647e-01 -2.07968682e-01 -9.02139604e-01 -4.68174517e-01 5.73257029e-01 -1.58495083e-01 -6.73612356e-01 1.07195795e+00 4.51840609e-01 -7.32212067e-02 -6.29694164e-02 -3.07898611e-01 -9.00638163e-01 -8.06569695e-01 -4.29342717e-01 5.22313952e-01 2.84846812e-01 -2.27803007e-01 3.03599596e-01 3.66146535e-01 -1.65449798e+00 1.49760067e-01 1.06020188e+00 5.63033938e-01 5.87967932e-01 8.36888552e-01 2.97921151e-01 1.16444671e+00 -4.95151520e-01 2.97836572e-01 5.86742282e-01 -5.42136550e-01 -5.18708348e-01 2.99045928e-02 2.10210010e-01 -6.33340657e-01 -4.11557972e-01 1.00153112e+00 3.47988516e-01 7.13213563e-01 3.05938423e-01 -1.27229857e+00 -2.66966790e-01 6.12908840e-01 -4.60607260e-01 1.95010707e-01 3.15502703e-01 9.02681828e-01 9.29530859e-01 -4.58255827e-01 4.47518706e-01 1.53954840e+00 2.42606223e-01 6.57446802e-01 -1.34419358e+00 -7.00296998e-01 6.98678613e-01 2.31706038e-01 -8.84912133e-01 -2.72117555e-01 5.92635274e-01 -4.95236039e-01 9.59842205e-01 -3.50776881e-01 1.06231689e+00 1.50041783e+00 2.60960311e-01 1.08667243e+00 1.49462938e+00 -9.73487943e-02 4.76754278e-01 -3.08614135e-01 -3.82128090e-01 9.19645786e-01 -1.03583649e-01 8.11085165e-01 -5.61141193e-01 -1.52199492e-01 1.23657262e+00 8.19609985e-02 1.58334106e-01 -6.69149220e-01 -1.25780559e+00 9.07147706e-01 3.47134292e-01 -6.11569174e-02 -7.88773417e-01 6.29473209e-01 3.17078769e-01 5.81614971e-01 1.10556535e-01 7.81475842e-01 -2.83310264e-01 -6.07298315e-01 -8.96578193e-01 6.98731244e-01 6.33999705e-01 7.89365053e-01 5.95964789e-01 4.52957213e-01 -4.75765437e-01 7.26843059e-01 1.07578374e-01 5.04027009e-01 4.98380005e-01 -1.64366078e+00 3.93093973e-01 1.26816139e-01 5.93720376e-01 -5.20488679e-01 -3.05511087e-01 -3.69020998e-01 -1.50362119e-01 8.89711022e-01 6.10606790e-01 -4.49058652e-01 -1.08514690e+00 1.86736286e+00 3.54489684e-01 3.64948004e-01 9.42749009e-02 1.17592490e+00 1.14323668e-01 5.39939940e-01 3.18800509e-01 -1.31278217e-01 8.25342059e-01 -1.36482596e+00 -6.78833604e-01 -4.61513400e-01 4.20474738e-01 -3.82863015e-01 1.43895483e+00 5.16343951e-01 -1.29997432e+00 -7.22914457e-01 -8.56641591e-01 3.61738980e-01 1.01899609e-01 2.09695593e-01 6.50414705e-01 -1.56604439e-01 -8.50009799e-01 9.60455656e-01 -1.14893651e+00 -1.18875183e-01 2.93516099e-01 3.49719465e-01 2.73466073e-02 1.37123048e-01 -1.18906820e+00 1.06304133e+00 3.10211599e-01 4.36338484e-02 -2.01308084e+00 -1.85623139e-01 -7.22178996e-01 -2.17631087e-02 9.95188236e-01 -5.35023510e-01 1.72432983e+00 -1.09207034e+00 -2.06574202e+00 4.32555765e-01 2.88829237e-01 -6.23887956e-01 6.12316787e-01 -5.59060574e-01 2.21962798e-02 8.20279941e-02 3.08090717e-01 7.78871477e-01 1.23439813e+00 -1.17857778e+00 -7.77554810e-01 -7.57618248e-02 4.84691173e-01 7.65865088e-01 8.34834501e-02 -2.09867954e-01 -1.80014879e-01 -5.48916340e-01 -3.31897408e-01 -1.18480885e+00 -6.90974116e-01 -2.90122420e-01 1.49966896e-01 -1.34110078e-01 3.31052840e-01 -4.82607424e-01 8.67528975e-01 -2.02530575e+00 7.04319060e-01 -9.43190232e-02 5.41245677e-02 -2.80114096e-02 -3.71548414e-01 3.40355575e-01 2.43581280e-01 -4.70754236e-01 4.45518419e-02 -3.49896729e-01 1.19427450e-01 5.03320694e-01 -3.10227990e-01 5.83065748e-01 -9.22653377e-02 9.13973987e-01 -1.36083329e+00 -3.11447144e-01 2.63161331e-01 4.43745963e-02 -7.60803223e-01 4.58399206e-01 -6.64823651e-01 1.05955994e+00 -7.14302838e-01 5.15895486e-01 -5.06687202e-02 -1.26617089e-01 3.15963328e-01 3.43477041e-01 -1.71244949e-01 3.13627541e-01 -1.13942039e+00 2.06785703e+00 -4.57938939e-01 1.93408623e-01 4.16345865e-01 -8.52500916e-01 6.98432922e-01 1.02978937e-01 6.89501405e-01 -6.96849942e-01 4.25609015e-03 1.31249055e-01 1.17650941e-01 -5.55934548e-01 6.75102651e-01 -9.95584801e-02 -3.87895197e-01 3.26758504e-01 7.75800571e-02 -4.13586855e-01 2.48764336e-01 2.98562109e-01 1.00476944e+00 9.76824820e-01 1.76919833e-01 -1.50237784e-01 1.30359128e-01 3.63488019e-01 7.54170954e-01 1.13620031e+00 -4.33871299e-01 1.11564524e-01 6.87426388e-01 -3.19736153e-01 -1.02306342e+00 -9.70520198e-01 3.44879508e-01 1.46093702e+00 1.18887179e-01 -2.61235833e-01 -4.85606551e-01 -8.09359372e-01 2.60872543e-01 6.66224182e-01 -7.22504735e-01 -2.13765338e-01 -8.27727079e-01 -1.39386311e-01 2.77288169e-01 5.02786577e-01 2.49840200e-01 -1.45335150e+00 -1.07538426e+00 4.12054926e-01 6.98801130e-02 -8.97472978e-01 -4.75822419e-01 4.81125742e-01 -9.26261902e-01 -9.64229822e-01 -7.63635218e-01 -3.77099395e-01 5.62148452e-01 -8.39710515e-03 1.02201498e+00 -3.74108069e-02 -5.32468036e-02 6.83065295e-01 -4.24697220e-01 -1.93031535e-01 -3.56668204e-01 -2.38318771e-01 5.55486321e-01 -3.64153922e-01 -1.82721272e-01 -6.95517123e-01 -5.67068040e-01 2.50980884e-01 -4.69606459e-01 1.23461500e-01 7.29371428e-01 1.17447841e+00 5.87409496e-01 -2.10883453e-01 3.49120975e-01 -5.60122550e-01 8.23293865e-01 -4.19053316e-01 -1.14403486e+00 -5.82158566e-02 -5.75168788e-01 3.19700897e-01 6.11084104e-01 -8.29990149e-01 -9.97556686e-01 1.58502698e-01 9.67780575e-02 -8.68186295e-01 -1.69025198e-01 2.71547228e-01 3.80518228e-01 2.25521475e-01 7.86544740e-01 2.26821080e-02 9.30611193e-02 -3.27674329e-01 5.16152382e-01 4.01672833e-02 3.80970210e-01 -1.10437131e+00 6.60165071e-01 2.29732126e-01 -1.90888375e-01 -3.52999866e-01 -1.01175249e+00 -2.86524594e-01 -4.75577652e-01 -5.52817881e-01 8.27932835e-01 -1.18936336e+00 -8.95917654e-01 2.24763080e-01 -5.78093588e-01 -1.34048283e+00 -8.60040486e-01 8.30307782e-01 -1.19848764e+00 2.88346671e-02 -6.92164898e-01 -8.68369699e-01 2.54878730e-01 -1.48118377e+00 9.93036628e-01 2.65078843e-01 -9.45512652e-02 -7.94329286e-01 2.89599389e-01 1.16384342e-01 2.53441930e-01 2.08210021e-01 2.88564563e-01 -1.18473619e-01 -6.04164898e-01 3.44097346e-01 2.94570327e-01 2.46174231e-01 -1.33328572e-01 -4.19781446e-01 -5.60691714e-01 -7.43606389e-01 -4.42426242e-02 -1.04279065e+00 8.14544141e-01 7.10803628e-01 1.09898221e+00 -2.70258605e-01 -8.71014781e-03 6.11122787e-01 1.11547315e+00 1.67810291e-01 1.44196615e-01 5.39854407e-01 6.02348149e-01 4.81671453e-01 1.33919477e+00 7.50972331e-01 7.62965754e-02 7.47665823e-01 6.15274668e-01 7.27169737e-02 -5.96017763e-02 -7.25282252e-01 7.72492170e-01 4.80730951e-01 -2.98195541e-01 3.47767740e-01 -6.62266791e-01 3.36723596e-01 -2.33060598e+00 -1.16587365e+00 4.79670346e-01 1.90972662e+00 8.78835618e-01 3.25948805e-01 5.88143051e-01 -6.10026538e-01 2.23582134e-01 3.33139420e-01 -1.23629999e+00 -3.40928257e-01 2.54948080e-01 -9.78606418e-02 6.25378430e-01 8.04461002e-01 -8.82611096e-01 1.42605007e+00 6.46550560e+00 7.89005041e-01 -9.10352290e-01 1.51361853e-01 1.38261884e-01 -7.99960196e-01 -2.00686842e-01 2.41483837e-01 -8.39405894e-01 4.00400877e-01 8.02283406e-01 -1.10253073e-01 9.91411805e-01 1.06340373e+00 6.03145480e-01 -4.24747735e-01 -1.01661766e+00 8.20704639e-01 -3.57380450e-01 -1.10591769e+00 -5.54632962e-01 2.16730908e-01 8.52978230e-01 2.31389433e-01 1.18492752e-01 1.07793057e+00 1.31099367e+00 -9.92897630e-01 9.52497065e-01 4.95326132e-01 8.41518819e-01 -6.87012374e-01 1.83110267e-01 6.52981699e-01 -9.88696635e-01 -6.44944072e-01 -3.26215297e-01 -3.66622001e-01 1.94377482e-01 -5.57436533e-02 -5.58048844e-01 8.66650939e-02 6.59773409e-01 9.63923037e-01 -1.20541476e-01 6.14498258e-01 -6.72128856e-01 5.56138396e-01 -1.80513129e-01 -2.43676323e-02 6.65305197e-01 -4.26821470e-01 7.10539043e-01 8.03904295e-01 5.07794283e-02 2.18245298e-01 9.85741854e-01 8.34161818e-01 3.54372352e-01 -2.44708195e-01 -5.60710549e-01 -1.54022485e-01 2.69457579e-01 1.02237856e+00 -4.56865698e-01 -4.18420345e-01 -6.57554492e-02 9.08945680e-01 7.88418710e-01 6.20371103e-01 -1.04290652e+00 2.28402764e-01 9.37040925e-01 -1.96862876e-01 2.12426782e-01 -5.95051050e-01 1.42244667e-01 -1.33871734e+00 -4.07926500e-01 -1.24058938e+00 3.20059717e-01 -6.00706220e-01 -9.81054723e-01 3.29129040e-01 2.52744764e-01 -1.34844160e+00 -6.18328094e-01 -2.80249238e-01 -2.24031806e-01 8.11608672e-01 -1.51620901e+00 -6.59186661e-01 -7.55719189e-03 7.22076178e-01 9.99585927e-01 -2.87102014e-01 6.57948196e-01 -1.36571720e-01 -4.70984489e-01 1.62426814e-01 1.89690217e-01 -2.77056783e-01 7.29621410e-01 -1.44057858e+00 6.96341172e-02 6.76818311e-01 -4.44218097e-03 2.91273504e-01 1.00721431e+00 -8.65269363e-01 -1.27838898e+00 -5.30780792e-01 -1.49060503e-01 -2.28102386e-01 8.98919463e-01 -2.50279397e-01 -6.20208323e-01 8.41161609e-01 1.26453355e-01 -1.97718889e-01 1.47813708e-01 2.65342265e-01 1.05265833e-01 1.56112149e-01 -8.43885481e-01 8.45288873e-01 1.01882732e+00 -3.70099068e-01 -6.35848224e-01 4.29690927e-01 6.52029097e-01 -7.87210107e-01 -7.33187139e-01 2.56451428e-01 5.39598227e-01 -8.43580842e-01 8.40533137e-01 -8.24548781e-01 3.13727826e-01 -2.20394775e-01 1.03850894e-01 -1.68274426e+00 -5.53457201e-01 -7.01458514e-01 -3.85495871e-01 3.71324837e-01 6.56633228e-02 -3.52135092e-01 9.45943356e-01 1.96477905e-01 -2.17394456e-01 -8.33668172e-01 -6.60004556e-01 -1.03230762e+00 -6.28975704e-02 -2.17944801e-01 2.58083731e-01 6.28324449e-01 4.42468608e-03 2.85403222e-01 -9.54191446e-01 8.72891396e-02 7.63687849e-01 3.90093297e-01 1.03198338e+00 -8.37389886e-01 -8.29992354e-01 -3.84822965e-01 2.65044272e-01 -1.58840990e+00 4.67216104e-01 -5.56680441e-01 4.77318406e-01 -1.41722715e+00 -7.02743679e-02 -5.79191029e-01 -3.20939928e-01 6.52029037e-01 -1.25670448e-01 -4.47021246e-01 4.20449018e-01 4.61548299e-01 -9.20342863e-01 1.02672303e+00 1.85400057e+00 8.35120603e-02 -5.77260196e-01 2.28027953e-03 -2.88952857e-01 7.84442365e-01 1.14828146e+00 -6.84551775e-01 -5.69723845e-01 -3.24669749e-01 -6.55888692e-02 5.57274997e-01 3.08039397e-01 -1.07544088e+00 2.48502735e-02 -8.25581789e-01 2.12005973e-01 -2.65815824e-01 7.24028170e-01 -6.69108033e-01 -2.78858840e-01 7.46403515e-01 -7.94943869e-01 1.85647160e-01 8.03668424e-02 8.56038272e-01 -7.31501654e-02 -1.75679103e-02 8.18463266e-01 -6.04737520e-01 -9.56009984e-01 3.76487792e-01 -6.14895105e-01 1.09645225e-01 1.15283382e+00 -1.22730039e-01 -2.05858368e-02 -5.68901837e-01 -1.11596942e+00 8.47336471e-01 4.98237222e-01 4.48619515e-01 6.11361027e-01 -1.17336583e+00 -6.50765240e-01 7.88279846e-02 -1.96360618e-01 -1.02493301e-01 1.89216509e-01 8.91363740e-01 -7.80508369e-02 1.61343753e-01 -7.76295841e-01 -5.59592962e-01 -1.00520265e+00 5.76462686e-01 4.35660154e-01 -6.26770496e-01 -9.03770924e-01 7.86796987e-01 7.77657554e-02 -3.89488339e-01 5.47883749e-01 -3.58040065e-01 -2.52655864e-01 -2.34828591e-01 2.42026359e-01 3.45772505e-01 -7.37091959e-01 -2.83533216e-01 1.37826167e-02 2.55257994e-01 2.94917971e-02 -6.08518958e-01 1.34081030e+00 -8.58806223e-02 3.65943044e-01 6.55725300e-01 5.36150515e-01 -1.83201566e-01 -2.36312699e+00 -1.45992979e-01 -1.13091581e-01 -6.67396724e-01 -3.89449820e-02 -7.65975595e-01 -6.36684835e-01 5.78271747e-01 4.56099212e-01 -1.62579358e-01 8.31476986e-01 -1.72896877e-01 4.77184355e-01 5.21016300e-01 7.85738170e-01 -1.55982578e+00 6.68064356e-01 8.56621861e-01 9.21549857e-01 -1.27198029e+00 4.85630184e-02 3.74356657e-01 -1.17639494e+00 7.08850801e-01 9.94913042e-01 -5.33422351e-01 2.19458014e-01 3.53395075e-01 3.39818597e-02 -2.51095146e-01 -1.15756571e+00 -5.55183947e-01 -5.87289184e-02 4.47351635e-01 -2.11041458e-02 2.95291573e-01 -1.26080766e-01 8.75896066e-02 -8.62130299e-02 6.41024262e-02 4.15167898e-01 1.15544415e+00 -6.77260220e-01 -1.19864559e+00 -3.90581310e-01 4.39215153e-01 -2.44504556e-01 2.07509413e-01 1.07560962e-01 5.11526883e-01 -1.29929528e-01 7.30834544e-01 -1.84778780e-01 -2.39819512e-01 3.30019861e-01 -1.33647218e-01 8.37271094e-01 -8.72196794e-01 -6.85492814e-01 2.88742661e-01 1.70510918e-01 -1.23679698e+00 -2.47366488e-01 -8.83893132e-01 -1.32883835e+00 -8.01730007e-02 1.67654365e-01 4.33380812e-01 3.48881066e-01 6.92554891e-01 1.14488825e-01 7.69677758e-01 4.87740636e-01 -1.24663138e+00 -1.22032917e+00 -8.90169322e-01 -6.67802989e-01 4.17408198e-01 5.63254893e-01 -1.09410846e+00 -2.88609117e-01 -2.03096107e-01]
[4.168341159820557, 1.5664150714874268]
77de38be-5ba9-4e0e-99df-f592419ba16c
tyger-task-type-generic-active-learning-for
2205.11279
null
https://arxiv.org/abs/2205.11279v1
https://arxiv.org/pdf/2205.11279v1.pdf
Tyger: Task-Type-Generic Active Learning for Molecular Property Prediction
How to accurately predict the properties of molecules is an essential problem in AI-driven drug discovery, which generally requires a large amount of annotation for training deep learning models. Annotating molecules, however, is quite costly because it requires lab experiments conducted by experts. To reduce annotation cost, deep Active Learning (AL) methods are developed to select only the most representative and informative data for annotating. However, existing best deep AL methods are mostly developed for a single type of learning task (e.g., single-label classification), and hence may not perform well in molecular property prediction that involves various task types. In this paper, we propose a Task-type-generic active learning framework (termed Tyger) that is able to handle different types of learning tasks in a unified manner. The key is to learn a chemically-meaningful embedding space and perform active selection fully based on the embeddings, instead of relying on task-type-specific heuristics (e.g., class-wise prediction probability) as done in existing works. Specifically, for learning the embedding space, we instantiate a querying module that learns to translate molecule graphs into corresponding SMILES strings. Furthermore, to ensure that samples selected from the space are both representative and informative, we propose to shape the embedding space by two learning objectives, one based on domain knowledge and the other leveraging feedback from the task learner (i.e., model that performs the learning task at hand). We conduct extensive experiments on benchmark datasets of different task types. Experimental results show that Tyger consistently achieves high AL performance on molecular property prediction, outperforming baselines by a large margin. We also perform ablative experiments to verify the effectiveness of each component in Tyger.
['Xinchao Wang', 'Tingyang Xu', 'Jian Tang', 'Jiashi Feng', 'Kaixin Wang', 'Kuangqi Zhou']
2022-05-23
null
null
null
null
['molecular-property-prediction']
['miscellaneous']
[ 4.35953587e-01 -5.14848046e-02 -6.03118360e-01 -3.38656276e-01 -1.00831890e+00 -6.84688032e-01 2.98061192e-01 5.86078346e-01 -3.91027749e-01 9.69397545e-01 -1.39963076e-01 -3.05594891e-01 -1.09337382e-01 -8.67433548e-01 -8.99703443e-01 -1.02301729e+00 1.01446941e-01 6.49442196e-01 6.61040097e-02 1.75156236e-01 1.61810130e-01 4.60714012e-01 -1.04050565e+00 1.20667383e-01 1.41016006e+00 1.00303864e+00 2.88410723e-01 1.07535638e-01 -9.06942561e-02 6.23325109e-01 -4.33866799e-01 -4.06348258e-01 2.36342400e-02 -4.37896162e-01 -8.35612178e-01 -7.22462460e-02 9.57406387e-02 6.07490204e-02 1.50499985e-01 9.26943362e-01 6.31415963e-01 1.86423659e-01 8.26571763e-01 -1.00810409e+00 -4.52251405e-01 4.72524524e-01 -1.97597921e-01 -1.96118608e-01 2.43205830e-01 3.56806755e-01 1.25661826e+00 -9.86971259e-01 5.04290521e-01 9.32621777e-01 3.82785678e-01 5.74971139e-01 -1.62212658e+00 -8.58491302e-01 3.53528947e-01 9.12274495e-02 -1.30887651e+00 -5.58952212e-01 9.17677104e-01 -5.17017365e-01 7.01539636e-01 2.08785325e-01 4.15205419e-01 1.25542009e+00 -2.98981019e-03 9.07552600e-01 8.10893416e-01 -9.71582904e-02 7.82891095e-01 2.93382227e-01 2.85019595e-02 5.37577450e-01 3.46886367e-01 -3.01636606e-01 -6.52939916e-01 -6.69612765e-01 2.50403345e-01 8.84641632e-02 -4.91207004e-01 -8.19145799e-01 -1.11974859e+00 1.04511118e+00 4.13057864e-01 7.30983913e-02 -5.72195232e-01 -6.38344660e-02 3.10033500e-01 6.78721294e-02 5.64998686e-01 1.12471020e+00 -8.12590480e-01 1.62524164e-01 -5.35822809e-01 5.83031140e-02 7.55281806e-01 6.23514831e-01 9.02916014e-01 -3.00537825e-01 -3.54521632e-01 8.39449763e-01 4.55723584e-01 3.58489491e-02 2.42640182e-01 -2.87003011e-01 3.38965535e-01 9.72301245e-01 9.75222215e-02 -6.94957793e-01 -2.96587318e-01 -5.41349471e-01 -5.94467402e-01 8.98536891e-02 3.02247375e-01 -1.60432518e-01 -8.99236917e-01 1.79925346e+00 5.08071482e-01 1.45153478e-01 9.62321088e-02 8.15579295e-01 6.71011627e-01 7.15485871e-01 6.13080800e-01 -4.11691219e-01 1.04359186e+00 -9.25848603e-01 -5.13590336e-01 -2.04348162e-01 9.16306138e-01 -4.76180851e-01 1.12721896e+00 4.78947699e-01 -6.04157746e-01 -3.28307748e-01 -1.16429806e+00 1.38304502e-01 -4.98028725e-01 2.22125828e-01 8.03456962e-01 4.39928114e-01 -3.80918205e-01 6.17646992e-01 -7.12264121e-01 7.59710930e-03 8.47362638e-01 5.50328553e-01 -4.43975061e-01 -2.98961420e-02 -1.28068614e+00 5.86565852e-01 8.44149530e-01 -9.25674886e-02 -1.16768026e+00 -7.90437698e-01 -7.97684610e-01 1.38258502e-01 6.71174526e-01 -5.96329868e-01 8.21002543e-01 -1.09990048e+00 -1.66990852e+00 4.86568660e-01 9.24881697e-02 -3.20318341e-01 3.64954948e-01 -1.02170318e-01 -1.13989823e-01 6.96809292e-02 -8.53587016e-02 5.50531685e-01 6.93314493e-01 -1.22383738e+00 -3.89133483e-01 -4.14134443e-01 2.15726107e-01 2.08797306e-01 -8.63164604e-01 -3.09960127e-01 -4.65908438e-01 -4.73027110e-01 -1.01148494e-01 -9.57709312e-01 -4.07148153e-01 8.31371844e-02 -5.42352796e-01 -5.40111899e-01 5.87831914e-01 -3.35271478e-01 1.33026540e+00 -1.95746136e+00 3.89808029e-01 3.87205988e-01 4.93410915e-01 3.74543846e-01 -1.63045079e-01 4.25190806e-01 -1.57106906e-01 2.08287865e-01 -4.02946830e-01 -1.82575032e-01 -1.22215413e-01 -8.39510933e-02 -2.33868703e-01 3.47762525e-01 5.15055418e-01 9.14104939e-01 -1.20410073e+00 -4.20488566e-01 -7.85961822e-02 3.42382222e-01 -5.40731609e-01 5.07573724e-01 -7.89031446e-01 6.27349854e-01 -8.11403096e-01 8.09169292e-01 3.83364916e-01 -6.47757411e-01 4.02296662e-01 -2.79777110e-01 2.09526774e-02 3.86770457e-01 -8.62134218e-01 1.79254532e+00 -5.61307251e-01 -2.06160937e-02 -4.46238846e-01 -1.22065377e+00 9.91198480e-01 4.32327837e-01 5.37714660e-01 -3.46561641e-01 -1.16510525e-01 4.71814722e-01 1.79762200e-01 -4.04496104e-01 -1.23877361e-01 1.05326682e-01 4.23049927e-02 3.51859629e-01 1.65517852e-01 2.71496385e-01 1.94951922e-01 3.72556085e-03 1.13699222e+00 2.55263567e-01 5.09472072e-01 -1.40535608e-01 6.59471452e-01 8.98124576e-02 9.23448741e-01 5.00215232e-01 -4.76186946e-02 1.17058724e-01 5.88842571e-01 -5.07120132e-01 -7.52554357e-01 -6.23123765e-01 -2.10994631e-01 1.18515325e+00 2.33199894e-01 -5.36686540e-01 -4.96571064e-01 -1.18632853e+00 -1.17841393e-01 5.99028707e-01 -6.39261723e-01 -4.70092416e-01 -3.14010173e-01 -1.11903799e+00 1.75320059e-01 2.84792244e-01 1.64889723e-01 -1.05263674e+00 -2.05527186e-01 4.14199650e-01 2.70148128e-01 -7.67842770e-01 -1.96016118e-01 6.21590137e-01 -6.65727437e-01 -1.15551770e+00 -6.60024285e-01 -6.86693668e-01 7.36103952e-01 -1.35711402e-01 9.55056667e-01 -1.19731627e-01 -6.18771724e-02 -2.01754496e-01 -4.32498902e-01 -4.57733452e-01 -1.72758594e-01 4.59989309e-01 -1.84719399e-01 2.12077588e-01 3.43439698e-01 -4.00155425e-01 -7.96653688e-01 2.81220615e-01 -8.29149604e-01 6.56018853e-02 1.00622475e+00 1.05947423e+00 8.64852369e-01 4.18044180e-02 8.84093106e-01 -1.25272381e+00 6.31581903e-01 -5.77382684e-01 -5.70381880e-01 5.08982956e-01 -6.97114348e-01 2.84910202e-01 1.06310523e+00 -6.85377300e-01 -6.54382706e-01 5.27249873e-01 -1.08189031e-01 -2.92851985e-01 -7.26616159e-02 7.71481395e-01 -8.23705852e-01 -1.51270241e-01 7.39604712e-01 2.73908824e-01 -1.93981960e-01 -4.48006541e-01 2.86512543e-02 5.86487532e-01 -1.91991240e-01 -6.56230032e-01 7.86713898e-01 1.03588216e-01 -1.24787673e-01 -4.53315556e-01 -8.85431230e-01 -3.75842869e-01 -4.62111115e-01 1.53382376e-01 7.12963045e-01 -8.18979919e-01 -7.58687854e-01 1.38535395e-01 -9.88464892e-01 -4.18256730e-01 -7.18917772e-02 6.21292353e-01 -4.76815879e-01 1.45727560e-01 -1.41123936e-01 -6.62329376e-01 -5.05102217e-01 -1.53307474e+00 1.03352439e+00 1.04056455e-01 -2.39092231e-01 -9.76039469e-01 2.79414386e-01 2.74066865e-01 1.47786871e-01 3.09442818e-01 1.35460436e+00 -1.16864204e+00 -7.45086730e-01 -2.20524952e-01 7.19500929e-02 1.57213956e-01 2.99601465e-01 -7.85476714e-02 -1.14335680e+00 -2.98435211e-01 -4.26226646e-01 -6.70288503e-01 1.00407732e+00 1.03007041e-01 1.65575612e+00 -4.06873941e-01 -5.86236954e-01 5.91427088e-01 1.25729275e+00 4.98214990e-01 5.01354814e-01 2.35089287e-01 7.66267419e-01 4.93174553e-01 7.70574093e-01 2.47817978e-01 4.32078801e-02 8.19337845e-01 4.64001536e-01 -2.89669693e-01 3.18252295e-01 -3.09255630e-01 2.80050039e-01 4.11843449e-01 8.24978724e-02 -4.75092828e-01 -8.46216500e-01 1.70744315e-01 -1.94203770e+00 -6.45105004e-01 3.63276482e-01 2.46360350e+00 1.48940301e+00 2.62531251e-01 1.71664089e-01 1.70592427e-01 4.24923509e-01 -1.32785719e-02 -1.06920183e+00 -3.19528617e-02 9.85739678e-02 2.87458181e-01 2.92175263e-01 2.48357862e-01 -1.34099209e+00 1.03885651e+00 4.45416737e+00 1.23855340e+00 -1.36704040e+00 -1.85557172e-01 8.40115964e-01 3.00696880e-01 -3.99354786e-01 1.63367391e-01 -7.14569509e-01 5.86947739e-01 6.37491405e-01 -4.44005392e-02 1.27720982e-01 9.39800918e-01 1.16845809e-01 2.71778077e-01 -1.41627586e+00 8.87840867e-01 -2.52747685e-01 -1.39278436e+00 2.54470468e-01 1.16911836e-01 5.71079969e-01 -4.50135589e-01 2.15777252e-02 3.59685749e-01 1.16145432e-01 -1.15692437e+00 3.59301895e-01 4.02820796e-01 6.83019519e-01 -6.82545483e-01 5.52019179e-01 4.44323182e-01 -1.02659917e+00 -3.89779583e-02 -4.31085616e-01 4.45371777e-01 -2.74867624e-01 6.56342983e-01 -1.02498412e+00 5.58381379e-01 1.84249312e-01 8.76078188e-01 -4.67501104e-01 1.20737255e+00 -3.05366069e-01 7.14742303e-01 -1.77013371e-02 -2.38814250e-01 8.01837891e-02 -2.53882647e-01 2.99196929e-01 8.89262080e-01 6.17582500e-02 -1.38625205e-01 6.02900207e-01 9.17250037e-01 -4.72768337e-01 4.76363182e-01 -4.76341903e-01 -5.14982045e-01 6.90788209e-01 1.34718883e+00 -5.59286952e-01 -1.87570989e-01 -2.23874062e-01 9.36913311e-01 5.36161244e-01 3.38214368e-01 -7.60346711e-01 -5.15936732e-01 6.02133930e-01 4.44516912e-02 5.58280610e-02 7.04051927e-02 -1.73149481e-01 -9.73338187e-01 -9.45974737e-02 -8.91635597e-01 3.94471437e-01 -1.53174892e-01 -1.40931857e+00 4.73636597e-01 -3.65952253e-01 -1.33780110e+00 5.48641607e-02 -6.15411520e-01 -5.60059249e-01 9.02982175e-01 -1.56341338e+00 -9.27456558e-01 -1.98772848e-01 2.71349907e-01 4.11583155e-01 -1.62031814e-01 1.12296128e+00 3.43584746e-01 -1.06271446e+00 7.99154699e-01 9.21247825e-02 3.17173311e-03 8.54032874e-01 -1.27687490e+00 6.88425750e-02 3.49687725e-01 2.38753576e-02 8.84683907e-01 4.49043751e-01 -6.20703578e-01 -1.66954827e+00 -1.29271257e+00 6.64110482e-01 -2.95971543e-01 6.48181736e-01 -7.08755314e-01 -1.01533651e+00 2.76997566e-01 -4.37048763e-01 1.19226821e-01 1.18425620e+00 2.20585749e-01 -3.67312908e-01 -1.28534377e-01 -9.30725813e-01 3.86551946e-01 9.23073590e-01 -4.47458655e-01 -2.94220578e-02 7.04863906e-01 7.80143797e-01 -2.54573077e-01 -9.36101973e-01 6.60317183e-01 3.79939288e-01 -3.55300128e-01 1.02902186e+00 -9.83312845e-01 3.15255880e-01 -4.37093943e-01 1.40089348e-01 -1.28363383e+00 -3.45124394e-01 -4.56901103e-01 -2.94735849e-01 1.03794658e+00 8.83750379e-01 -6.73587680e-01 7.56576717e-01 4.47930604e-01 -4.09469493e-02 -1.32726073e+00 -5.23435652e-01 -6.95141733e-01 -1.18223123e-01 7.18645304e-02 6.82293713e-01 1.17345417e+00 -1.66196913e-01 7.75997579e-01 -2.94846982e-01 1.36927813e-01 3.59327376e-01 3.60134900e-01 7.29354620e-01 -1.52446032e+00 -4.22243327e-01 -2.88730711e-01 -2.66360343e-01 -9.20589685e-01 4.05701995e-01 -1.04701221e+00 3.55524500e-03 -1.49189210e+00 3.31145853e-01 -7.87724972e-01 -5.46618998e-01 7.14561522e-01 -6.31886601e-01 -7.99370781e-02 -2.74247319e-01 4.18168604e-01 -8.25019658e-01 7.72599161e-01 1.15305758e+00 -4.17770177e-01 -4.37078327e-01 1.15158763e-02 -8.52291524e-01 4.93374228e-01 7.53164887e-01 -4.93603319e-01 -6.10558450e-01 -7.28712007e-02 4.26571280e-01 -2.10246474e-01 1.73493057e-01 -7.26529837e-01 1.06774606e-01 -5.07428825e-01 4.92665857e-01 -1.61542058e-01 3.23533952e-01 -8.04327786e-01 1.20492265e-01 4.17948723e-01 -6.43099964e-01 -6.46743000e-01 -1.60272166e-01 7.71081090e-01 -1.10359743e-01 -1.64561346e-01 8.34994674e-01 6.87941015e-02 -6.00205719e-01 9.50827956e-01 1.63781732e-01 -6.17518201e-02 1.07635653e+00 -2.76237186e-02 -3.47743064e-01 1.41421974e-01 -5.35803199e-01 2.36229628e-01 5.56929946e-01 9.73426104e-02 4.34965730e-01 -1.17329240e+00 -5.41413128e-01 -7.90360868e-02 7.63368249e-01 2.29557559e-01 -1.36827499e-01 7.45857656e-01 -2.43197694e-01 3.71271521e-01 1.49749026e-01 -3.66812050e-01 -9.32552755e-01 6.10864520e-01 3.49406213e-01 -3.41519535e-01 -1.98279768e-01 9.17623043e-01 4.33857709e-01 -4.18152511e-01 3.33244294e-01 6.45399615e-02 -3.91452163e-01 2.12475792e-01 4.40806717e-01 -1.01583265e-01 3.67923856e-01 -1.72189653e-01 -5.05897760e-01 2.89865881e-01 -3.29766810e-01 5.60018241e-01 1.48932028e+00 6.13061249e-01 8.07346702e-02 3.02471519e-01 1.27048337e+00 -3.76160704e-02 -1.24938297e+00 -3.37558538e-01 3.29641372e-01 -2.94552594e-01 1.13734409e-01 -8.73811364e-01 -8.74011338e-01 8.23954940e-01 4.89895076e-01 -7.69427866e-02 9.64354694e-01 -7.11396635e-02 6.72926605e-01 6.64603472e-01 2.32014269e-01 -1.04053402e+00 3.12890619e-01 1.67672455e-01 6.73337579e-01 -1.37399817e+00 7.32591748e-02 -4.50154722e-01 -5.40706098e-01 1.10933769e+00 6.98693216e-01 2.96461403e-01 3.72144639e-01 -1.34285495e-01 -1.87793300e-01 -3.82950366e-01 -8.79906833e-01 -2.12994099e-01 5.26062787e-01 3.60008538e-01 7.80830622e-01 1.68491364e-01 -3.41649413e-01 5.55730224e-01 3.58365715e-01 -1.17675133e-01 -1.56022325e-01 9.56383348e-01 -5.22491693e-01 -1.48279321e+00 3.55376713e-02 5.74442327e-01 -2.43242592e-01 -1.12762585e-01 -8.16492558e-01 4.08631027e-01 2.23268658e-01 6.85001612e-01 -4.00597155e-01 -3.95941466e-01 1.80318519e-01 6.08329736e-02 2.55932331e-01 -9.35732424e-01 -3.68824750e-01 5.14000189e-03 -4.82635945e-03 -3.23105633e-01 -2.21430436e-01 -2.35980347e-01 -1.19509447e+00 1.05007678e-01 -5.40439308e-01 3.91281992e-01 4.59609449e-01 7.17350066e-01 3.93153578e-01 3.54078144e-01 9.26012039e-01 -4.38054144e-01 -6.41385138e-01 -7.32567012e-01 -3.35280091e-01 4.56125587e-01 1.90196037e-02 -7.99754322e-01 -2.27008551e-01 -1.51753768e-01]
[5.202031135559082, 5.933951377868652]
1d5475db-3242-40ba-a7c8-abb8ba30bf33
contrastive-code-representation-learning
2007.04973
null
https://arxiv.org/abs/2007.04973v4
https://arxiv.org/pdf/2007.04973v4.pdf
Contrastive Code Representation Learning
Recent work learns contextual representations of source code by reconstructing tokens from their context. For downstream semantic understanding tasks like summarizing code in English, these representations should ideally capture program functionality. However, we show that the popular reconstruction-based BERT model is sensitive to source code edits, even when the edits preserve semantics. We propose ContraCode: a contrastive pre-training task that learns code functionality, not form. ContraCode pre-trains a neural network to identify functionally similar variants of a program among many non-equivalent distractors. We scalably generate these variants using an automated source-to-source compiler as a form of data augmentation. Contrastive pre-training improves JavaScript summarization and TypeScript type inference accuracy by 2% to 13%. We also propose a new zero-shot JavaScript code clone detection dataset, showing that ContraCode is both more robust and semantically meaningful. On it, we outperform RoBERTa by 39% AUROC in an adversarial setting and up to 5% on natural code.
['Joseph E. Gonzalez', 'Ion Stoica', 'Ajay Jain', 'Pieter Abbeel', 'Paras Jain', 'Tianjun Zhang']
2020-07-09
null
https://aclanthology.org/2021.emnlp-main.482
https://aclanthology.org/2021.emnlp-main.482.pdf
emnlp-2021-11
['code-summarization', 'type-prediction', 'method-name-prediction']
['computer-code', 'computer-code', 'natural-language-processing']
[ 4.87905979e-01 3.68487924e-01 -4.75797802e-01 -4.26974714e-01 -1.21298444e+00 -1.00384653e+00 4.99404311e-01 5.35067856e-01 -8.51429403e-02 3.11230600e-01 5.08241773e-01 -6.53562784e-01 5.47930777e-01 -9.26949799e-01 -1.42624640e+00 -5.06736189e-02 -3.64696421e-02 -3.62186581e-02 1.70458958e-01 -1.59455627e-01 4.78072613e-01 -1.04143523e-01 -1.65722454e+00 8.41099858e-01 9.40725863e-01 3.13483506e-01 2.69451022e-01 1.11006832e+00 -4.18744683e-01 1.16593587e+00 -8.74980211e-01 -7.02511072e-01 6.75288737e-02 -4.76211160e-01 -1.16405857e+00 -5.81534445e-01 7.51103878e-01 -2.61199474e-01 -1.38264582e-01 1.16305304e+00 2.13811040e-01 -1.82372957e-01 2.78376043e-01 -1.05371547e+00 -1.02840841e+00 1.30123210e+00 -2.42546350e-01 9.40001979e-02 5.35306454e-01 2.26109847e-01 1.38404667e+00 -6.56352937e-01 9.98705685e-01 1.10776269e+00 1.18306267e+00 8.49302471e-01 -1.59206676e+00 -2.90629864e-01 -1.18548766e-01 -8.02894831e-02 -7.72819281e-01 -3.45378935e-01 6.03493750e-01 -5.25190949e-01 1.46398318e+00 4.62425768e-01 8.61074850e-02 1.46221817e+00 1.92976117e-01 6.86214149e-01 7.58813322e-01 -3.91312301e-01 2.16794103e-01 -5.38610034e-02 4.17599976e-01 1.02176607e+00 2.42136225e-01 1.21258069e-02 -5.49067482e-02 -7.35110879e-01 5.52349947e-02 -1.64651629e-02 -1.59554690e-01 -1.91750199e-01 -1.10843360e+00 8.73682261e-01 5.74375153e-01 1.60291135e-01 2.66776770e-01 7.87378967e-01 8.69436741e-01 5.30235052e-01 3.19511592e-02 1.06324780e+00 -6.42325819e-01 -3.77474785e-01 -6.57557905e-01 3.84815991e-01 1.00349605e+00 1.14309406e+00 9.44272518e-01 2.29078740e-01 -5.89904338e-02 6.96025729e-01 -5.12035303e-02 4.23491001e-01 7.83482432e-01 -9.91606474e-01 5.35547376e-01 7.57597446e-01 -2.51090676e-01 -7.21134484e-01 6.56208172e-02 -1.43375307e-01 -2.63821065e-01 3.49759579e-01 3.03259701e-01 8.32337886e-02 -6.09258354e-01 1.88474309e+00 -5.23886690e-03 -6.98448643e-02 1.54062301e-01 4.03964281e-01 8.12036455e-01 5.13879895e-01 2.43947372e-01 4.44119513e-01 1.14846623e+00 -7.92839885e-01 -1.35457233e-01 -4.26482290e-01 1.13588738e+00 -6.82709455e-01 1.46076596e+00 5.35703264e-02 -8.53401542e-01 -4.36352223e-01 -1.09523630e+00 -4.67690349e-01 -2.57960200e-01 -3.98353487e-03 7.93832302e-01 6.52342021e-01 -9.67655182e-01 8.16606879e-01 -8.26086104e-01 -2.01691478e-01 4.33682919e-01 -1.39362440e-01 -3.98877740e-01 7.90325701e-02 -6.81057155e-01 6.74203217e-01 2.64622569e-01 -6.47497833e-01 -1.42357671e+00 -1.22817969e+00 -1.20478415e+00 2.57565111e-01 2.98884928e-01 -6.68832958e-01 1.64638889e+00 -1.55783916e+00 -1.13853383e+00 1.08866894e+00 -3.28619391e-01 -6.25231862e-01 2.43106857e-01 -2.33191937e-01 -3.33267033e-01 -9.35117379e-02 3.27612489e-01 2.55346894e-01 6.62089944e-01 -1.36337018e+00 -2.95293808e-01 -1.50507882e-01 4.89984542e-01 -4.23014224e-01 1.02991968e-01 2.03933746e-01 -2.52001211e-02 -8.98916900e-01 -4.55236077e-01 -7.52996802e-01 -1.44278035e-01 3.64634804e-02 -5.42214453e-01 -1.18641503e-01 5.87621868e-01 -1.02180409e+00 9.24236238e-01 -2.11913705e+00 3.64200361e-02 -1.39513716e-01 4.03105885e-01 -3.65087166e-02 -4.83973294e-01 4.08103555e-01 -3.00742984e-01 5.65991938e-01 -8.12667906e-01 4.62310910e-02 2.53248870e-01 4.38181125e-02 -7.13210702e-01 3.57511938e-01 4.33599949e-01 1.10639787e+00 -1.05766177e+00 -5.55015542e-02 -3.64108920e-01 2.05433786e-01 -1.18878782e+00 3.86819690e-01 -9.10349727e-01 -1.20953716e-01 -2.31904432e-01 6.78642154e-01 4.40926731e-01 -1.89949542e-01 3.02574545e-01 2.66736299e-01 1.43914029e-01 7.63979733e-01 -4.18251216e-01 2.26139069e+00 -1.09073663e+00 7.88796484e-01 -1.80286676e-01 -1.05726075e+00 8.35477412e-01 2.19432991e-02 5.31694200e-03 -5.46884239e-01 -2.40192801e-01 2.50973046e-01 -1.60469890e-01 -7.44748294e-01 7.62974858e-01 5.41485958e-02 -6.98111773e-01 6.68258905e-01 1.21701948e-01 -1.61456138e-01 -5.71359843e-02 4.63428110e-01 1.65634000e+00 6.73130572e-01 4.15656745e-01 -4.37812835e-01 2.15112641e-01 2.37841189e-01 6.74533844e-01 1.04782569e+00 1.22526884e-01 5.74878812e-01 1.09540057e+00 -4.93500620e-01 -1.28847277e+00 -1.15399241e+00 1.51580453e-01 1.44622326e+00 -3.18994783e-02 -8.47949386e-01 -1.04905188e+00 -1.26714182e+00 1.60763651e-01 1.10119867e+00 -8.29518080e-01 -4.71161246e-01 -9.75667715e-01 -4.69826072e-01 1.29099429e+00 7.58965313e-01 2.83460170e-02 -9.68531907e-01 -6.44509554e-01 1.45421475e-01 -2.08404794e-01 -6.00440621e-01 -5.32298863e-01 3.72038007e-01 -7.31403530e-01 -1.48998499e+00 -2.92773694e-01 -1.00494587e+00 7.78314352e-01 -1.88974395e-01 1.70936346e+00 4.93496835e-01 -4.46504593e-01 1.92684487e-01 -2.68625289e-01 9.56985801e-02 -1.47769678e+00 1.38279751e-01 -6.81791246e-01 -7.37217963e-01 2.98026890e-01 -7.25724339e-01 -2.69073755e-01 -1.58892617e-01 -8.98638427e-01 -1.24757193e-01 4.20520335e-01 8.71322274e-01 2.68593520e-01 -5.77334166e-01 4.21076298e-01 -1.55064774e+00 3.66705507e-01 -8.31236601e-01 -7.51929820e-01 2.47075438e-01 -4.09034342e-01 5.63473701e-01 1.24919343e+00 -2.27968827e-01 -1.33548713e+00 1.41016832e-02 -3.18559408e-01 -9.51375738e-02 -1.28093079e-01 3.96041930e-01 -1.73706084e-01 1.31373927e-01 1.48221004e+00 2.84261525e-01 -7.14437887e-02 -6.11159980e-01 7.56622970e-01 5.22173584e-01 9.71696913e-01 -1.16804278e+00 8.55493009e-01 2.81907052e-01 -4.39512491e-01 -3.82479906e-01 -3.96132797e-01 5.35846874e-02 -3.36797297e-01 5.88918686e-01 5.85516870e-01 -8.26524138e-01 -4.60190177e-01 2.65876442e-01 -1.40680540e+00 -5.39141119e-01 -4.62315381e-01 -2.90836573e-01 -6.21620774e-01 5.86966872e-01 -7.61756957e-01 -1.37485579e-01 -3.01816672e-01 -1.35060918e+00 1.16151965e+00 -3.09226930e-01 -3.91154289e-01 -7.40211427e-01 2.09677011e-01 5.84759042e-02 4.52599347e-01 5.03314614e-01 1.55086768e+00 -7.77073681e-01 -7.19461918e-01 8.83255154e-02 -1.32167965e-01 2.58326322e-01 1.45990485e-02 1.42676279e-01 -1.07364428e+00 -7.72487521e-02 -1.20312132e-01 -4.76614892e-01 9.69892502e-01 -2.53175050e-01 1.43496633e+00 -8.74743581e-01 -2.61325747e-01 8.19241822e-01 1.72247171e+00 -7.85223618e-02 6.49125516e-01 2.86103129e-01 6.98919475e-01 4.32508767e-01 -1.20820850e-01 2.27859244e-01 2.95808673e-01 2.31095657e-01 7.46415257e-01 3.84975225e-01 -3.99430782e-01 -5.70452154e-01 6.02497578e-01 6.44268930e-01 5.09346128e-01 1.99968219e-02 -1.00933588e+00 8.33849192e-01 -1.33005583e+00 -1.17396462e+00 -1.18106909e-01 2.09234452e+00 1.36520016e+00 -1.21065103e-01 -4.27766070e-02 -2.10637167e-01 6.15513384e-01 1.02546178e-01 -4.68307137e-01 -6.96254969e-01 8.17944482e-02 5.45848548e-01 4.71190304e-01 5.29123068e-01 -9.44482327e-01 9.28168654e-01 6.05846739e+00 4.84014183e-01 -9.69342947e-01 3.74953777e-01 4.77584779e-01 2.58041650e-01 -1.10873723e+00 4.82110053e-01 -4.14168954e-01 5.09739876e-01 1.18049467e+00 -3.57994705e-01 7.09065557e-01 1.59617364e+00 -4.47782636e-01 3.66359726e-02 -1.49682593e+00 2.97586352e-01 1.83798835e-01 -1.70872855e+00 -4.99483608e-02 -4.43968952e-01 8.19837630e-01 2.94297606e-01 -2.55862474e-01 7.47045279e-01 7.85869241e-01 -1.03546786e+00 8.68926346e-01 2.27894485e-01 9.42850411e-01 -6.13743901e-01 3.26194078e-01 6.77314177e-02 -9.49773490e-01 -8.65141153e-02 -4.43398446e-01 2.11169973e-01 -5.22941351e-01 3.32379133e-01 -8.01826239e-01 3.35337162e-01 4.90779966e-01 7.75258482e-01 -1.05533946e+00 7.08141327e-01 -3.69752735e-01 6.50211036e-01 1.93734452e-01 -1.62300125e-01 -5.10904565e-03 5.18264234e-01 4.66603369e-01 1.72833836e+00 2.28496134e-01 -5.00290215e-01 1.93248000e-02 1.59160900e+00 -6.37457132e-01 -2.61757731e-01 -9.04694021e-01 -1.98926225e-01 5.38486958e-01 9.64978695e-01 -2.86211103e-01 -6.58356249e-01 -4.63464916e-01 1.14080107e+00 4.54413891e-01 2.46775687e-01 -9.71042931e-01 -1.12040174e+00 1.00357962e+00 -1.67603701e-01 3.76098990e-01 1.90804422e-01 -3.24393421e-01 -1.35430634e+00 2.29252353e-01 -1.42413461e+00 2.73707569e-01 -6.24288142e-01 -9.53000426e-01 5.92099428e-01 -2.11323529e-01 -8.16397727e-01 -4.38237786e-01 -5.12376130e-01 -8.83350134e-01 7.47709870e-01 -1.27588212e+00 -9.94357049e-01 -1.97873428e-01 1.05919903e-02 6.48480654e-01 -2.04381838e-01 1.12320542e+00 -2.19729226e-02 -1.55044168e-01 9.08261538e-01 2.68558383e-01 5.03192782e-01 4.38481450e-01 -1.69772291e+00 1.22450709e+00 1.27042317e+00 9.03950110e-02 1.15154243e+00 7.02287316e-01 -6.70235217e-01 -1.73697960e+00 -1.58500373e+00 7.43398547e-01 -8.26040328e-01 7.23732233e-01 -5.49712479e-01 -1.21246731e+00 1.04494154e+00 2.86053538e-01 2.42445365e-01 4.83142674e-01 4.64153755e-03 -1.50681591e+00 1.95880920e-01 -1.17528439e+00 6.19739175e-01 1.19644630e+00 -1.19180274e+00 -1.07864523e+00 2.15036273e-01 1.13557661e+00 -3.69135946e-01 -8.53933811e-01 9.10722166e-02 3.59793484e-01 -1.03189862e+00 8.68703723e-01 -9.01709437e-01 1.19416606e+00 -1.62568823e-01 -3.55762690e-01 -1.22644651e+00 2.56438442e-02 -7.93516695e-01 -1.26746714e-01 1.31815588e+00 3.55083108e-01 -3.08456123e-01 6.47177637e-01 1.51546299e-01 -4.48787063e-01 -2.14132458e-01 -5.80665946e-01 -7.97561586e-01 4.67440307e-01 -5.96148729e-01 6.80318534e-01 1.12306833e+00 2.74950445e-01 1.69306755e-01 1.45054623e-01 5.56719787e-02 4.49500501e-01 5.08705378e-01 8.73415709e-01 -8.58758390e-01 -8.06913972e-01 -6.43328488e-01 -3.06199431e-01 -9.02647674e-01 7.64369428e-01 -1.65157127e+00 3.03531498e-01 -1.07675314e+00 4.17344093e-01 -2.93204665e-01 9.52289775e-02 6.77425683e-01 -3.17948103e-01 -8.24436098e-02 -2.19844595e-01 6.53927028e-02 -2.94274807e-01 1.99225739e-01 4.31698024e-01 -5.27873814e-01 2.68354386e-01 -2.49957487e-01 -1.07862616e+00 6.18319571e-01 7.12919593e-01 -8.61072898e-01 -2.59505153e-01 -7.27171838e-01 4.02362436e-01 1.02215245e-01 6.33965731e-01 -7.47234881e-01 -3.06426048e-01 5.14210714e-03 -2.99025606e-02 6.66595474e-02 -2.35012650e-01 -3.22555661e-01 7.54118115e-02 7.89429665e-01 -9.48186398e-01 2.21362963e-01 3.15648943e-01 6.99200034e-01 1.02206618e-01 -6.80383027e-01 7.94689834e-01 -6.00324869e-01 -9.84526694e-01 -2.23001942e-01 -5.07436275e-01 7.12119937e-01 6.47137523e-01 9.77611616e-02 -8.94754350e-01 8.15057680e-02 -1.80827752e-01 -3.95939767e-01 8.88517320e-01 7.40646303e-01 3.14582705e-01 -9.39897895e-01 -6.14681840e-01 2.48935714e-01 4.81444329e-01 -3.08635205e-01 -4.92309667e-02 2.84447879e-01 -9.33132350e-01 5.59352525e-02 -1.73816755e-01 -2.92614430e-01 -1.28169632e+00 7.22890615e-01 3.26312244e-01 3.39141674e-02 -7.22609818e-01 8.51926684e-01 2.20274672e-01 -9.62685883e-01 1.36707664e-01 -8.60167265e-01 4.22233760e-01 -5.41875184e-01 5.71036279e-01 2.36496478e-01 1.07681878e-01 -2.51794606e-01 -4.48699445e-01 2.64653891e-01 -7.13463053e-02 3.29323322e-01 1.34414589e+00 5.79966664e-01 -4.96644884e-01 2.38061205e-01 1.68037772e+00 5.37056804e-01 -1.20189440e+00 -1.78504035e-01 6.05911791e-01 -4.73157942e-01 -4.09003586e-01 -1.08227253e+00 -8.38230789e-01 7.41590202e-01 6.78585172e-02 3.64976265e-02 7.15963423e-01 1.19694039e-01 6.67921662e-01 8.40894461e-01 3.16450179e-01 -5.42253315e-01 1.53927907e-01 5.96670687e-01 8.20344329e-01 -1.02205765e+00 -3.66210997e-01 -5.21011911e-02 -2.90076941e-01 1.22466326e+00 6.81340575e-01 -3.54597837e-01 1.03395693e-01 6.95606589e-01 -1.23650290e-01 6.48070350e-02 -8.29960048e-01 2.70280063e-01 -3.11509054e-02 7.82258451e-01 7.32662559e-01 -7.45039657e-02 2.45831702e-02 7.34263241e-01 -2.95123130e-01 -3.33056360e-01 9.38460410e-01 1.07791865e+00 -2.75949955e-01 -1.13474357e+00 -1.36147767e-01 4.44297314e-01 -8.23093295e-01 -6.18704855e-01 -4.40047354e-01 7.86248922e-01 -8.84639397e-02 5.67968905e-01 -1.03409644e-02 -3.30924153e-01 -1.71051715e-02 2.21850872e-01 4.15562212e-01 -9.77295220e-01 -1.18357682e+00 -6.27498627e-01 1.97043613e-01 -8.20480049e-01 1.16370663e-01 -5.16853929e-01 -1.40227020e+00 -3.07166517e-01 -4.17279787e-02 1.35863096e-01 6.51626289e-01 4.55576122e-01 7.01698244e-01 7.36629188e-01 4.61415410e-01 -3.14939231e-01 -8.94422054e-01 -6.47233427e-01 2.08748817e-01 8.01522434e-01 5.88911653e-01 -2.61315778e-02 -4.47756678e-01 4.69995171e-01]
[7.534585952758789, 7.873610973358154]
c4b2d7cc-b67c-44e1-bcd4-8944044a8166
open-vocabulary-panoptic-segmentation-with-2
2303.11324
null
https://arxiv.org/abs/2303.11324v1
https://arxiv.org/pdf/2303.11324v1.pdf
Open-vocabulary Panoptic Segmentation with Embedding Modulation
Open-vocabulary image segmentation is attracting increasing attention due to its critical applications in the real world. Traditional closed-vocabulary segmentation methods are not able to characterize novel objects, whereas several recent open-vocabulary attempts obtain unsatisfactory results, i.e., notable performance reduction on the closed vocabulary and massive demand for extra data. To this end, we propose OPSNet, an omnipotent and data-efficient framework for Open-vocabulary Panoptic Segmentation. Specifically, the exquisitely designed Embedding Modulation module, together with several meticulous components, enables adequate embedding enhancement and information exchange between the segmentation model and the visual-linguistic well-aligned CLIP encoder, resulting in superior segmentation performance under both open- and closed-vocabulary settings with much fewer need of additional data. Extensive experimental evaluations are conducted across multiple datasets (e.g., COCO, ADE20K, Cityscapes, and PascalContext) under various circumstances, where the proposed OPSNet achieves state-of-the-art results, which demonstrates the effectiveness and generality of the proposed approach. The code and trained models will be made publicly available.
['Hengshuang Zhao', 'Antonio Torralba', 'Ser-Nam Lim', 'Shuang Li', 'Xi Chen']
2023-03-20
null
null
null
null
['panoptic-segmentation', 'open-vocabulary-panoptic-segmentation']
['computer-vision', 'computer-vision']
[ 3.32136184e-01 7.67564168e-03 -2.18736783e-01 -1.76439837e-01 -8.83649290e-01 -5.37769258e-01 4.14847702e-01 -1.11725509e-01 -5.54125190e-01 5.46152294e-01 -5.39903715e-02 -6.22325838e-02 -3.21664242e-03 -5.05955637e-01 -5.29642582e-01 -7.00734079e-01 1.46283999e-01 3.19893897e-01 3.62082154e-01 -2.70079434e-01 8.09810460e-02 -1.45616978e-02 -1.60028005e+00 -3.86490434e-01 1.13622797e+00 1.17542624e+00 4.50380266e-01 4.36568081e-01 -1.39946729e-01 1.31737098e-01 -2.31688321e-01 -7.04572260e-01 3.27336699e-01 -7.43816942e-02 -7.19337881e-01 3.78635556e-01 5.33583641e-01 -8.96494836e-02 -3.65587175e-01 1.15686142e+00 5.16329169e-01 8.79037157e-02 5.09819806e-01 -8.33695889e-01 -7.48344064e-01 5.00431180e-01 -7.20788181e-01 2.67357498e-01 4.81941737e-03 2.25571200e-01 1.36318326e+00 -6.85932755e-01 7.13947356e-01 8.22243273e-01 4.31500971e-01 4.88206416e-01 -1.15278316e+00 -7.90601075e-01 2.21678525e-01 1.70871705e-01 -1.76015973e+00 -3.98409992e-01 6.33500516e-01 -3.34923089e-01 6.79074585e-01 4.18635130e-01 8.20766985e-01 9.65954959e-01 -3.82862310e-03 8.82701278e-01 8.24046373e-01 -2.97363997e-01 -4.85150740e-02 3.47710222e-01 1.54142216e-01 6.67387307e-01 2.93225884e-01 -8.87355804e-02 -2.58175552e-01 3.57634842e-01 7.28103280e-01 -1.15984596e-01 -4.04234201e-01 -3.60826433e-01 -1.17945993e+00 8.35206747e-01 3.87397796e-01 3.07645351e-01 -1.84543446e-01 -7.53592923e-02 6.60044372e-01 4.89283353e-02 6.23887360e-01 3.85640591e-01 -2.84937829e-01 -1.25920251e-01 -1.00899673e+00 1.02086455e-01 3.92635584e-01 1.39965725e+00 6.10197127e-01 -2.78388429e-02 3.11875138e-02 1.01728201e+00 1.74781859e-01 5.31222045e-01 5.08565247e-01 -6.98787272e-01 4.62325096e-01 4.47916001e-01 -1.78589240e-01 -1.11828744e+00 -1.98014736e-01 -4.45530802e-01 -7.74981678e-01 -4.32620555e-01 7.99369588e-02 6.73829839e-02 -9.63686585e-01 1.65454733e+00 3.58609498e-01 3.18095654e-01 2.45513365e-01 8.80296826e-01 1.01458275e+00 7.62578368e-01 1.03326790e-01 -2.93971390e-01 1.53661871e+00 -1.12717092e+00 -1.01147330e+00 -3.09011132e-01 3.76392901e-01 -8.99312317e-01 1.37393904e+00 3.00989330e-01 -8.90418887e-01 -6.38577938e-01 -1.16623771e+00 -2.95129806e-01 -4.26993757e-01 3.56247872e-02 7.36803830e-01 5.16094685e-01 -7.15532780e-01 1.15319662e-01 -6.05841637e-01 -3.25086981e-01 7.27380812e-01 4.08600718e-01 -2.92438865e-01 -3.38775441e-02 -1.16603565e+00 4.50445473e-01 6.74995840e-01 5.49877957e-02 -5.68512082e-01 -5.60818493e-01 -1.01990032e+00 -1.08783217e-02 8.07150185e-01 -3.26236546e-01 1.04388213e+00 -9.00919437e-01 -1.25379992e+00 1.04688978e+00 1.34971619e-01 -5.17425656e-01 3.97451818e-01 -4.80368882e-02 -5.01447618e-01 4.15033221e-01 2.12972626e-01 1.16302228e+00 9.77644742e-01 -9.46874321e-01 -7.95490503e-01 -3.74660999e-01 1.00918114e-01 3.58351797e-01 -7.26809144e-01 -5.23148254e-02 -9.74528551e-01 -9.33538377e-01 8.53181705e-02 -8.72613549e-01 -2.31932566e-01 1.00367246e-02 -3.90653968e-01 -1.58971146e-01 7.99329579e-01 -5.92335701e-01 1.49112952e+00 -2.54733682e+00 9.78459939e-02 -1.82877958e-01 3.58140171e-01 5.49610138e-01 4.51858416e-02 -1.23725021e-02 1.74505889e-01 1.80702910e-01 -3.80687505e-01 -1.34089470e-01 3.30775194e-02 2.74477929e-01 -2.09199131e-01 4.52118099e-01 1.92989677e-01 1.07740831e+00 -6.57030761e-01 -9.59890783e-01 4.50599909e-01 3.38551968e-01 -4.99812990e-01 1.40560120e-02 -4.95457858e-01 3.81360978e-01 -6.08904839e-01 6.18473887e-01 6.79344058e-01 -4.53021765e-01 -1.83450714e-01 -5.57457842e-02 -2.05775768e-01 -2.31638953e-01 -1.14271855e+00 1.92135346e+00 -3.94018620e-01 7.00389802e-01 3.74430418e-02 -8.59296262e-01 8.53339314e-01 3.35289180e-01 3.00101608e-01 -8.97803366e-01 4.78748590e-01 3.05188209e-01 -2.97763318e-01 -5.64538956e-01 8.47808301e-01 -8.20183232e-02 -2.78271824e-01 8.53719488e-02 1.51004270e-01 -2.81159043e-01 5.18205643e-01 1.80047601e-01 5.21493614e-01 -1.22290388e-01 5.53198278e-01 -2.30986387e-01 7.18240261e-01 1.84088144e-02 5.40618181e-01 2.72044390e-01 -5.41973948e-01 5.06905675e-01 1.59401581e-01 -9.54112113e-02 -1.03792930e+00 -9.33538079e-01 -4.60684627e-01 8.69110525e-01 8.15767109e-01 -4.46403921e-01 -9.15656269e-01 -4.70059425e-01 -3.57363224e-01 4.13485497e-01 -4.88576621e-01 8.22707042e-02 -4.02409345e-01 -4.24564898e-01 6.86944664e-01 4.00874138e-01 7.44124532e-01 -9.13625479e-01 -4.28002238e-01 -7.94057734e-03 -4.30595368e-01 -1.49389160e+00 -8.92141998e-01 -1.35320470e-01 -6.73057795e-01 -8.96726251e-01 -8.66071403e-01 -1.33022165e+00 5.13833582e-01 4.58946496e-01 7.67163932e-01 -1.85781643e-01 -3.34422261e-01 6.78506568e-02 -3.89071792e-01 -3.27568293e-01 -1.79994807e-01 2.43601963e-01 -2.33524665e-01 1.14565812e-01 5.46744704e-01 -2.19516918e-01 -5.44921517e-01 4.51279193e-01 -9.24328446e-01 1.96331397e-01 5.35998046e-01 8.75485063e-01 1.08492696e+00 9.36545711e-03 6.29806876e-01 -7.50739038e-01 4.00637805e-01 -2.19915673e-01 -8.75852525e-01 2.20146343e-01 -5.75506032e-01 -2.77248234e-01 6.94241107e-01 -5.36545932e-01 -1.00561130e+00 -7.61270225e-02 -4.24047291e-01 -4.66067642e-01 -1.05598003e-01 2.43696630e-01 -2.56119967e-01 -6.82432353e-02 2.97195822e-01 4.08751637e-01 -1.20981075e-01 -3.50964904e-01 5.66633642e-01 1.13629341e+00 6.22175992e-01 -2.66619802e-01 5.94564319e-01 5.66956103e-01 -5.17633080e-01 -1.12898934e+00 -7.90896535e-01 -7.09296167e-01 -5.04046321e-01 -1.13778211e-01 1.20419347e+00 -1.09214735e+00 -3.81632030e-01 5.17114043e-01 -8.63934100e-01 1.24927595e-01 -4.57049221e-01 3.98646116e-01 -5.97899854e-01 5.02150238e-01 -3.20559144e-01 -4.14374113e-01 -4.90078032e-01 -1.52294934e+00 1.22133470e+00 5.12372136e-01 1.79279577e-02 -9.35055971e-01 -2.64030159e-01 7.82978296e-01 1.47336110e-01 3.78552601e-02 6.72281623e-01 -7.37771809e-01 -8.11522961e-01 -1.63471535e-01 -5.33499658e-01 5.09696245e-01 6.55398816e-02 -1.52822405e-01 -9.90154028e-01 -2.61438966e-01 -1.37436941e-01 -5.62684417e-01 7.02012241e-01 3.49491209e-01 1.19761884e+00 1.69585906e-02 -2.70840853e-01 8.15257370e-01 1.44245815e+00 1.53287500e-01 5.36561847e-01 9.85726044e-02 5.91887295e-01 3.93537581e-01 9.05241191e-01 2.98723817e-01 3.53153378e-01 7.62056947e-01 2.98810244e-01 -2.02319294e-01 -2.10033044e-01 -2.45259240e-01 -6.97019249e-02 1.31477344e+00 3.26575756e-01 -3.89084280e-01 -6.00388706e-01 8.57081175e-01 -1.50447273e+00 -7.30839908e-01 -4.82208766e-02 1.97761834e+00 9.69397426e-01 2.54877388e-01 -2.26469383e-01 5.05389180e-03 7.54293323e-01 5.88578999e-01 -6.40069723e-01 -3.50632668e-01 -2.49115333e-01 1.84656650e-01 4.64753002e-01 2.92284369e-01 -1.33454370e+00 1.31403279e+00 6.03256464e+00 1.58407581e+00 -1.20744288e+00 2.80228049e-01 5.71636140e-01 -7.11003393e-02 -3.02769125e-01 -2.86634177e-01 -7.90830255e-01 5.60822189e-01 6.58450067e-01 -1.30310193e-01 3.30673277e-01 6.64479554e-01 -2.16297377e-02 -1.04479278e-02 -7.52094150e-01 1.20817113e+00 2.52082914e-01 -1.48596728e+00 -6.32513175e-03 5.57114780e-02 6.85747266e-01 7.51467375e-03 2.39342287e-01 3.22470903e-01 -2.66349137e-01 -8.53557825e-01 7.28660345e-01 -2.21959036e-02 1.21350491e+00 -7.36941397e-01 6.02356374e-01 1.62395060e-01 -1.23696518e+00 2.70979740e-02 -2.27088347e-01 2.11422518e-01 3.46831590e-01 2.34259084e-01 -3.91922623e-01 6.17905676e-01 6.68440521e-01 8.67916822e-01 -4.68766958e-01 6.78070128e-01 -1.21533021e-01 6.64756060e-01 -2.87484914e-01 -6.98575750e-02 5.76155603e-01 -4.10433024e-01 6.24767005e-01 1.14011860e+00 -1.57620147e-01 9.25443545e-02 2.00981930e-01 6.68778718e-01 -2.12696359e-01 5.65300465e-01 -5.95387697e-01 -2.66257912e-01 2.39310429e-01 1.10466027e+00 -8.88229966e-01 -2.94597149e-01 -8.38485658e-01 7.85635829e-01 1.81407884e-01 2.18174562e-01 -1.18959272e+00 -5.93837142e-01 7.23115385e-01 -8.79280269e-02 8.46914411e-01 -2.83244834e-03 -2.67274410e-01 -1.23215723e+00 1.45311460e-01 -8.89267802e-01 3.27154428e-01 -4.73656744e-01 -1.04523969e+00 6.46036386e-01 -1.49346918e-01 -1.23810053e+00 2.02123150e-01 -4.07760650e-01 -8.63750279e-03 5.23378551e-01 -1.59840381e+00 -1.24242938e+00 -2.02926472e-01 5.44919550e-01 9.95291948e-01 -1.05944954e-01 6.22864425e-01 7.01200008e-01 -8.61105919e-01 8.07595849e-01 2.78315872e-01 1.12038083e-01 5.30296743e-01 -9.09218669e-01 3.66161466e-01 9.52564776e-01 4.05935168e-01 2.83920497e-01 6.23514235e-01 -4.36313391e-01 -1.19477034e+00 -1.21793687e+00 6.42457366e-01 -5.77245951e-02 7.60831296e-01 -6.65761054e-01 -8.21099877e-01 5.64020455e-01 2.77414769e-01 5.31946123e-02 6.58923686e-01 -2.65572250e-01 -1.68255195e-01 -1.32762343e-01 -9.67278838e-01 9.18225169e-01 1.12573123e+00 -2.85489708e-01 -6.55906618e-01 2.32721135e-01 1.30172837e+00 -5.06041050e-01 -9.48927343e-01 3.46344471e-01 4.59421992e-01 -6.51328683e-01 8.09583902e-01 -2.43158564e-01 3.26966584e-01 -1.24238148e-01 -4.23372328e-01 -8.48157465e-01 -9.38662421e-03 -8.48915458e-01 1.34932190e-01 1.23658049e+00 4.25358951e-01 -6.35621250e-01 5.59753418e-01 1.02602758e-01 -2.29692668e-01 -1.05833459e+00 -1.04145312e+00 -6.58258140e-01 -5.99339837e-03 -6.20131195e-01 5.27013719e-01 7.78218329e-01 -1.03148982e-01 5.63447356e-01 -4.21380848e-01 5.36641628e-02 4.26639795e-01 2.60581344e-01 7.54704058e-01 -9.20807302e-01 -1.55950755e-01 -3.76373500e-01 -7.01573133e-01 -1.63960660e+00 2.02327177e-01 -7.86542416e-01 5.25164790e-02 -1.23760402e+00 2.63193160e-01 -6.19027436e-01 -4.79093865e-02 7.27897808e-02 -2.63874352e-01 6.38096511e-01 1.65613651e-01 8.31778571e-02 -1.02613699e+00 8.51059556e-01 1.71384907e+00 -3.21342498e-01 -5.27816825e-02 -2.25461587e-01 -9.43362057e-01 6.54553175e-01 5.43265402e-01 -2.95902759e-01 -6.50199473e-01 -5.25375605e-01 -7.75948092e-02 -2.35765994e-01 -6.77718723e-04 -7.89401233e-01 3.54107171e-02 1.17085278e-02 -1.38457179e-01 -5.29878616e-01 4.67196763e-01 -8.23981762e-01 -4.70450260e-02 1.49315745e-01 -2.02154607e-01 -3.00995767e-01 3.16408575e-01 8.45800042e-01 -4.89740878e-01 -2.09258407e-01 8.78640592e-01 1.03420883e-01 -1.10239780e+00 5.89443743e-01 -3.11781242e-02 5.62256277e-01 1.39193892e+00 -6.99518025e-01 -4.72133346e-02 -1.12384781e-01 -6.44358456e-01 5.45152009e-01 5.18143952e-01 6.50973737e-01 6.13728464e-01 -1.06230688e+00 -6.04318261e-01 4.73963946e-01 4.69691575e-01 2.17644393e-01 4.35315341e-01 9.32628095e-01 -5.25230169e-01 5.97902358e-01 3.90359536e-02 -9.22410250e-01 -1.27004039e+00 6.50977373e-01 1.97734982e-02 -2.82207996e-01 -9.06516671e-01 8.45722914e-01 5.00033796e-01 -3.35085273e-01 3.88530314e-01 -4.30833369e-01 -2.07239166e-01 2.19459206e-01 4.38530385e-01 1.39089197e-01 -6.35325536e-02 -8.80402982e-01 -9.08120051e-02 7.81591535e-01 -3.96487832e-01 1.45991638e-01 1.01061785e+00 -4.88485903e-01 2.10882142e-01 3.25708568e-01 1.27291083e+00 -1.81810662e-01 -1.25362337e+00 -3.73691142e-01 -4.37854260e-01 -5.37639737e-01 1.59687296e-01 -3.90021652e-01 -1.09416950e+00 8.34517360e-01 6.79935575e-01 1.26563296e-01 1.24998212e+00 1.83989868e-01 1.31666279e+00 1.44021511e-02 4.55260128e-01 -1.30842483e+00 -1.89432316e-02 2.96270669e-01 5.52595437e-01 -1.43220258e+00 -3.89157273e-02 -6.91231310e-01 -7.97267795e-01 6.30697489e-01 4.70257223e-01 1.01262115e-01 5.91779172e-01 -4.87972125e-02 1.58961698e-01 -2.20033586e-01 -5.27372837e-01 -3.13275516e-01 1.52569309e-01 5.32312274e-01 9.23084095e-02 1.85625896e-01 -4.22689587e-01 7.05057383e-01 -2.45892048e-01 -1.94772065e-01 3.45892966e-01 5.75412095e-01 -3.74671310e-01 -7.39921212e-01 -1.54184237e-01 5.26020646e-01 -5.72657645e-01 -7.45210052e-02 -9.18489248e-02 1.09034550e+00 2.56671429e-01 8.59749258e-01 2.34511599e-01 -1.41934574e-01 3.20933133e-01 -3.67364585e-01 3.05848211e-01 -5.61049938e-01 -2.96721488e-01 3.04241419e-01 -5.84897473e-02 -4.62929338e-01 -5.29922962e-01 -6.85952842e-01 -1.28941631e+00 -1.69843331e-01 -8.92153978e-01 6.95658401e-02 5.13510883e-01 8.99890602e-01 3.67933065e-01 5.59595704e-01 3.97345543e-01 -4.24280077e-01 -3.91600966e-01 -8.51319671e-01 -6.30725682e-01 3.91594678e-01 3.12676787e-01 -6.98127508e-01 -8.59073326e-02 1.90750793e-01]
[9.636449813842773, 0.7192903161048889]
1817b2ec-576e-4c29-a638-c0891ccb6552
simpleclick-interactive-image-segmentation
2210.11006
null
https://arxiv.org/abs/2210.11006v3
https://arxiv.org/pdf/2210.11006v3.pdf
SimpleClick: Interactive Image Segmentation with Simple Vision Transformers
Click-based interactive image segmentation aims at extracting objects with a limited user clicking. A hierarchical backbone is the de-facto architecture for current methods. Recently, the plain, non-hierarchical Vision Transformer (ViT) has emerged as a competitive backbone for dense prediction tasks. This design allows the original ViT to be a foundation model that can be finetuned for downstream tasks without redesigning a hierarchical backbone for pretraining. Although this design is simple and has been proven effective, it has not yet been explored for interactive image segmentation. To fill this gap, we propose SimpleClick, the first interactive segmentation method that leverages a plain backbone. Based on the plain backbone, we introduce a symmetric patch embedding layer that encodes clicks into the backbone with minor modifications to the backbone itself. With the plain backbone pretrained as a masked autoencoder (MAE), SimpleClick achieves state-of-the-art performance. Remarkably, our method achieves 4.15 NoC@90 on SBD, improving 21.8% over the previous best result. Extensive evaluation on medical images demonstrates the generalizability of our method. We further develop an extremely tiny ViT backbone for SimpleClick and provide a detailed computational analysis, highlighting its suitability as a practical annotation tool.
['Marc Niethammer', 'Gedas Bertasius', 'Zhenlin Xu', 'Qin Liu']
2022-10-20
null
null
null
null
['interactive-segmentation']
['computer-vision']
[ 3.65457535e-01 5.91842592e-01 -1.92250803e-01 -4.10984129e-01 -7.78306365e-01 -4.21974152e-01 1.64275438e-01 -1.69411376e-01 -4.77479607e-01 1.34219885e-01 1.36091754e-01 -3.98828208e-01 5.13016403e-01 -5.78294635e-01 -1.04486823e+00 -7.71538854e-01 1.67396933e-01 3.46394628e-01 8.06721866e-01 1.72362030e-02 6.51955511e-03 1.72164068e-01 -1.30009782e+00 5.62163353e-01 6.46730423e-01 1.07213485e+00 5.13328552e-01 5.25606036e-01 1.51302576e-01 6.09448195e-01 -4.71697897e-01 -5.40585816e-01 4.11743879e-01 -1.15912370e-01 -9.82462943e-01 -4.14841867e-04 7.03400612e-01 -7.63928473e-01 -3.84903103e-01 5.48481047e-01 5.21239042e-01 -2.45010138e-01 3.31820548e-01 -1.05199277e+00 -6.13640308e-01 9.16692078e-01 -5.80908835e-01 2.40295120e-02 -2.27002171e-03 3.16199154e-01 1.31907451e+00 -8.34519148e-01 6.89464331e-01 6.49820924e-01 1.08439207e+00 5.87638438e-01 -1.39002562e+00 -6.61850452e-01 2.49098554e-01 1.90177947e-01 -1.34925199e+00 -1.35455951e-01 4.54459190e-01 -3.80711555e-01 1.03638756e+00 1.10941634e-01 1.00011218e+00 1.00789988e+00 2.01437876e-01 1.40545106e+00 9.76106763e-01 -9.65302959e-02 1.42881960e-01 -3.24629508e-02 9.43389386e-02 1.12370908e+00 -5.53670414e-02 -1.84712455e-01 -4.04925436e-01 2.42805630e-01 8.58810723e-01 1.85135789e-02 -3.20584387e-01 -4.28732812e-01 -1.22892833e+00 9.34347630e-01 1.00795734e+00 1.18147107e-02 -1.92984730e-01 2.60326028e-01 2.86188990e-01 -1.62457436e-01 1.57318473e-01 5.11420846e-01 -4.07678604e-01 8.09177160e-02 -1.32000673e+00 8.21828321e-02 7.90309608e-01 1.04196119e+00 6.54252589e-01 -1.33056074e-01 -4.26456898e-01 6.75317109e-01 1.11476757e-01 1.78696647e-01 3.22866857e-01 -1.06431651e+00 5.82738109e-02 6.11457646e-01 -3.14296514e-01 -6.36916518e-01 -4.52241838e-01 -8.10787022e-01 -5.99487603e-01 9.81524736e-02 2.09255338e-01 -1.16600923e-01 -1.26545179e+00 1.53521848e+00 4.61242437e-01 2.31355414e-01 -4.66329813e-01 9.59957063e-01 9.53803480e-01 4.64155167e-01 5.92343509e-02 3.77850831e-01 1.32475841e+00 -1.61785305e+00 -2.17313811e-01 -2.17789382e-01 4.80231315e-01 -5.31156838e-01 1.39467883e+00 5.66672921e-01 -1.02396035e+00 -4.33892995e-01 -1.17032802e+00 -5.04538894e-01 -2.14467585e-01 5.57345301e-02 9.10527289e-01 4.27935779e-01 -1.26879859e+00 5.75156629e-01 -1.09110868e+00 -1.83984011e-01 6.99678302e-01 5.87453544e-01 -2.76803315e-01 4.37317975e-02 -8.57493818e-01 4.71229255e-01 3.35775048e-01 -4.80773710e-02 -9.16523337e-01 -1.04978335e+00 -6.12812936e-01 7.92769641e-02 5.17428160e-01 -1.00384045e+00 1.53572524e+00 -7.27034867e-01 -1.62031353e+00 8.16644251e-01 -2.35951021e-02 -8.19396615e-01 4.99815643e-01 -2.02453792e-01 2.42422875e-02 5.22935092e-01 2.67418921e-01 1.44207036e+00 1.07420385e+00 -1.06349075e+00 -7.77108610e-01 -8.27243477e-02 1.92163333e-01 6.44537061e-02 -4.98217076e-01 -3.49164695e-01 -1.20791638e+00 -8.55068624e-01 1.27205014e-01 -1.20854437e+00 -3.57442498e-01 9.82004702e-02 -7.30020523e-01 -2.18685180e-01 6.84031844e-01 -6.71453476e-01 1.31742561e+00 -2.10898256e+00 -5.58475740e-02 1.48289641e-02 6.66015267e-01 9.89743993e-02 3.75547558e-02 2.26410732e-01 1.59814194e-01 5.92011958e-02 -4.22699034e-01 -4.96265978e-01 4.78912629e-02 1.67302385e-01 -1.59395263e-01 2.88019210e-01 1.19966164e-01 1.22989321e+00 -5.27585626e-01 -6.24593318e-01 1.80423528e-01 4.85355675e-01 -1.13467646e+00 2.30479911e-01 -2.61508495e-01 2.00981528e-01 -2.99887657e-01 7.39469171e-01 4.92998213e-01 -7.66973734e-01 -3.24954093e-02 -6.61713719e-01 -1.76993147e-01 2.49420881e-01 -6.96269453e-01 1.97122192e+00 -7.14901788e-03 6.66969478e-01 2.48471964e-02 -8.69012177e-01 4.00298357e-01 2.05831118e-02 5.30112028e-01 -6.51142776e-01 1.93790775e-02 2.31320839e-02 -2.62785614e-01 -3.58359039e-01 4.14021343e-01 1.21089242e-01 -7.31735379e-02 2.29984283e-01 2.21248046e-01 -6.86839921e-03 4.23938185e-02 4.81006116e-01 1.37740099e+00 1.82000324e-01 -6.02984466e-02 -1.45366579e-01 -8.60787481e-02 1.90950140e-01 5.99468112e-01 1.04621983e+00 -2.02588826e-01 1.13132489e+00 2.84886390e-01 -3.38639587e-01 -1.11065638e+00 -1.16088927e+00 -1.56010538e-01 1.21531570e+00 2.30058670e-01 -6.35717213e-01 -1.21575749e+00 -1.04565215e+00 -8.66871774e-02 4.44239229e-01 -7.65179574e-01 1.29355967e-01 -5.60768783e-01 -5.88638544e-01 6.64392531e-01 8.78242970e-01 7.62265563e-01 -8.77224922e-01 -6.79565191e-01 2.84830540e-01 -2.08329916e-01 -1.43549931e+00 -6.79334044e-01 2.28304058e-01 -6.84645772e-01 -9.12368178e-01 -8.12362254e-01 -9.45716083e-01 5.94130039e-01 3.06105226e-01 1.09409726e+00 7.95220435e-02 -4.50931728e-01 4.17385131e-01 -3.00464958e-01 -2.75987267e-01 -1.09392265e-02 5.66872299e-01 -4.23194677e-01 -1.55052781e-01 1.19639739e-01 -5.67570686e-01 -1.37738419e+00 3.46755207e-01 -9.19148386e-01 6.73984945e-01 8.44311595e-01 8.19006681e-01 8.65857303e-01 -3.52261275e-01 1.30691350e-01 -1.14245570e+00 1.34079922e-02 -2.39726052e-01 -3.92195940e-01 -1.47773698e-02 -6.76122248e-01 -3.66377980e-02 4.69889939e-01 -1.30766347e-01 -8.46013069e-01 3.86953712e-01 -6.13021135e-01 -4.36593592e-01 -1.96943760e-01 3.34972858e-01 1.35531157e-01 -1.55551821e-01 3.51987273e-01 1.62155837e-01 -1.01275183e-01 -6.58228815e-01 4.52410072e-01 5.79394460e-01 7.35006154e-01 -2.29600310e-01 5.98205209e-01 6.01549745e-01 -3.42649251e-01 -6.39371753e-01 -1.07346058e+00 -5.01221716e-01 -5.67526877e-01 7.90864006e-02 1.23861432e+00 -1.08241737e+00 -8.70689452e-01 3.45957845e-01 -8.13982666e-01 -8.50858986e-01 -1.63104534e-01 6.04159087e-02 -4.44982290e-01 3.46918941e-01 -9.87475991e-01 4.72150482e-02 -5.73256731e-01 -1.35986042e+00 1.37010992e+00 6.50938675e-02 -2.06376478e-01 -6.65111899e-01 -2.36280277e-01 8.63785207e-01 3.73275876e-01 1.92931592e-01 7.66643167e-01 -4.76137817e-01 -9.55989122e-01 -3.61826457e-02 -3.74542892e-01 3.76737416e-01 -2.06180528e-01 -2.03814879e-01 -1.06232309e+00 -3.83599520e-01 -1.84759602e-01 -3.44025224e-01 1.18289924e+00 4.03307110e-01 1.64859343e+00 -2.20150411e-01 -5.41052163e-01 1.04697633e+00 1.26799047e+00 -1.15761362e-01 6.27996624e-01 3.57518166e-01 9.38749075e-01 4.04732712e-02 2.51112700e-01 2.41113082e-01 7.34350860e-01 6.62758410e-01 4.65726823e-01 -5.70390940e-01 -4.52276826e-01 -3.34359944e-01 2.42020816e-01 8.91463816e-01 1.41988769e-01 3.64439525e-02 -9.08905745e-01 4.15261298e-01 -1.53939390e+00 -7.11513460e-01 1.17945105e-01 1.89531446e+00 1.08429396e+00 4.19991374e-01 9.88315120e-02 -1.48878396e-01 3.22791696e-01 5.30218519e-02 -7.68671453e-01 -4.94114608e-02 1.03844807e-01 5.17691553e-01 6.80846930e-01 3.55930775e-01 -1.39354384e+00 1.22792470e+00 6.49169064e+00 9.67150986e-01 -1.30229950e+00 2.23982930e-01 6.05542421e-01 -5.81133254e-02 -1.25146747e-01 2.32150443e-02 -1.11553705e+00 4.65643317e-01 6.17860079e-01 4.51520056e-01 2.45248750e-01 1.02616084e+00 5.66271842e-02 -5.69072329e-02 -1.11592412e+00 8.63560557e-01 -1.31342225e-02 -1.62862337e+00 1.95509598e-01 -4.81775776e-02 5.49304128e-01 4.48042631e-01 1.76149234e-01 4.34507042e-01 1.63102463e-01 -1.03076792e+00 7.44016290e-01 1.11107789e-01 7.78244615e-01 -3.42247963e-01 4.21134859e-01 2.77778916e-02 -9.98081625e-01 -1.68501418e-02 -6.99961111e-02 3.38543445e-01 9.87064317e-02 4.68725830e-01 -1.11589015e+00 1.87077254e-01 1.06774783e+00 7.73290098e-01 -8.84390414e-01 1.17875612e+00 -1.50769725e-01 1.12327349e+00 -4.84577209e-01 2.79574454e-01 4.44876850e-01 2.73221999e-01 3.50116491e-01 1.60076058e+00 -1.56112403e-01 -2.03826129e-02 3.10931891e-01 7.23577201e-01 -2.01194003e-01 -3.91272083e-02 -1.26928911e-01 5.03631942e-02 2.82084554e-01 1.28335166e+00 -9.88117099e-01 -4.06620473e-01 -5.76142430e-01 1.39795470e+00 3.72261345e-01 3.73803169e-01 -1.11852765e+00 -5.18537700e-01 4.47426587e-01 2.43254021e-01 9.27551568e-01 -1.26070321e-01 -4.07853127e-01 -1.11472392e+00 -1.88529901e-02 -1.00827050e+00 3.04577321e-01 -6.20269954e-01 -1.10690331e+00 5.24775088e-01 -2.83720404e-01 -9.79461193e-01 1.76888466e-01 -6.06256425e-01 -2.75802433e-01 2.37873927e-01 -1.40163922e+00 -1.52685702e+00 -4.63699877e-01 7.11270273e-01 7.72931397e-01 2.57743120e-01 7.62004435e-01 3.24577093e-01 -8.30294967e-01 9.54871297e-01 -1.79782152e-01 3.14769119e-01 5.71129620e-01 -1.56101525e+00 5.16580045e-01 6.39732301e-01 1.69078797e-01 5.79631627e-01 5.39538383e-01 -5.73336482e-01 -1.43877232e+00 -1.17518914e+00 3.08860719e-01 -3.22151691e-01 5.23877382e-01 -6.24978423e-01 -7.36896157e-01 1.03491426e+00 4.31783468e-01 -4.39348444e-02 8.69518638e-01 6.26431182e-02 -4.93528098e-01 -1.77254692e-01 -7.86784530e-01 8.69242311e-01 1.13601780e+00 -3.27693373e-01 -4.56157237e-01 3.48302156e-01 1.15525448e+00 -6.30720258e-01 -1.02287853e+00 4.23637390e-01 6.52828932e-01 -1.17269289e+00 1.05904102e+00 -6.38080388e-02 5.17436981e-01 -2.63760835e-01 5.47897369e-02 -9.15006399e-01 -6.32367373e-01 -6.90146208e-01 -1.53410777e-01 9.30388927e-01 5.37635684e-01 -3.12216401e-01 1.22199202e+00 7.08167076e-01 -5.28049469e-01 -1.21787262e+00 -6.28981590e-01 -3.76054823e-01 7.11255595e-02 -4.59573627e-01 4.19533074e-01 5.85348666e-01 -2.64436603e-01 5.12114763e-01 -1.48503095e-01 1.27830341e-01 6.19406521e-01 2.09988266e-01 8.09755921e-01 -9.07514215e-01 -7.70709515e-01 -4.26336050e-01 -2.32297376e-01 -1.71290076e+00 -1.70373335e-01 -9.66017365e-01 4.88695018e-02 -1.64110303e+00 3.59783441e-01 -5.13230920e-01 -2.74183184e-01 9.04054284e-01 -1.52917907e-01 7.59263813e-01 2.27717459e-01 1.88071728e-01 -8.23091745e-01 4.65971887e-01 1.33367515e+00 -3.01293820e-01 -1.97998434e-01 -4.02958915e-02 -8.79448354e-01 7.33403444e-01 7.60474920e-01 -2.48366311e-01 -4.61981535e-01 -6.74613297e-01 4.39370237e-03 -3.06009918e-01 3.65801960e-01 -1.12204993e+00 4.45165128e-01 2.60266542e-01 4.19362932e-01 -7.39002228e-01 3.85675669e-01 -6.33920848e-01 -1.21146441e-01 2.04190254e-01 -3.60499829e-01 -1.06472373e-01 2.16557696e-01 5.66798627e-01 -4.37654704e-02 4.96778339e-02 7.14990079e-01 -8.81997272e-02 -8.30618620e-01 5.72979927e-01 -2.91683763e-01 1.01422906e-01 9.14648473e-01 -4.63382363e-01 -3.92819680e-02 3.67328175e-03 -8.69689703e-01 2.44216591e-01 5.00667036e-01 2.16745943e-01 5.94317913e-01 -8.02281916e-01 -3.88213038e-01 1.52303487e-01 -4.74104984e-03 2.89060146e-01 2.86261380e-01 1.00606918e+00 -6.86108708e-01 4.33599085e-01 1.15874954e-01 -8.05880547e-01 -1.16118622e+00 5.02248347e-01 2.12073475e-01 -2.17583865e-01 -1.30865788e+00 1.13542259e+00 5.44857442e-01 -1.65925458e-01 4.97802377e-01 -5.06063998e-01 2.21735425e-02 -1.42829001e-01 4.28175598e-01 6.13812497e-03 1.97775304e-01 -1.42104506e-01 -1.74625605e-01 3.61372530e-01 -4.53750014e-01 -2.11067894e-03 1.36222482e+00 -1.74045041e-01 2.33675819e-02 1.92919061e-01 1.25940394e+00 -2.25809395e-01 -1.59752595e+00 -9.21510234e-02 -3.16622108e-01 -8.33014399e-02 2.96744525e-01 -8.89840066e-01 -1.31469023e+00 8.78278852e-01 5.42331696e-01 2.87702214e-02 1.12242115e+00 4.95336577e-02 1.38247514e+00 3.48806739e-01 3.57574522e-01 -1.05037522e+00 1.66672304e-01 2.63563573e-01 7.26296723e-01 -1.17695653e+00 -2.59173065e-01 -5.41484654e-01 -7.22507000e-01 8.05504560e-01 7.18042552e-01 -1.16640076e-01 7.67640471e-01 5.62886178e-01 -4.58508059e-02 -2.65078813e-01 -5.72241008e-01 -1.53558910e-01 1.85894012e-01 3.81578326e-01 3.55300307e-01 1.26266554e-01 4.36224006e-02 6.19096041e-01 -4.47915375e-01 1.19856693e-01 2.23166674e-01 7.75510311e-01 -3.27855617e-01 -8.65389228e-01 8.59945565e-02 6.94268048e-01 -5.80373824e-01 -3.90502840e-01 -5.57340197e-02 7.17143357e-01 2.06966683e-01 6.12090230e-01 4.75078747e-02 -5.60006440e-01 2.58508831e-01 -1.14170842e-01 3.53045672e-01 -6.26097441e-01 -8.70483398e-01 1.76486015e-01 -2.58149475e-01 -1.00007296e+00 -8.68312716e-02 -5.11490822e-01 -1.38282967e+00 -1.61055595e-01 -2.52328992e-01 -6.59113005e-02 4.40030187e-01 7.50318170e-01 6.89076483e-01 6.15372062e-01 2.63877213e-01 -9.98589218e-01 -2.20710948e-01 -4.85611826e-01 -2.73755342e-01 2.14006051e-01 3.43935847e-01 -4.79436874e-01 -1.14367045e-02 2.16579378e-01]
[9.64409065246582, 0.12779301404953003]
d581db44-60d3-491b-80c1-26e5d422dbe1
intervention-generalization-a-view-from
2306.04027
null
https://arxiv.org/abs/2306.04027v1
https://arxiv.org/pdf/2306.04027v1.pdf
Intervention Generalization: A View from Factor Graph Models
One of the goals of causal inference is to generalize from past experiments and observational data to novel conditions. While it is in principle possible to eventually learn a mapping from a novel experimental condition to an outcome of interest, provided a sufficient variety of experiments is available in the training data, coping with a large combinatorial space of possible interventions is hard. Under a typical sparse experimental design, this mapping is ill-posed without relying on heavy regularization or prior distributions. Such assumptions may or may not be reliable, and can be hard to defend or test. In this paper, we take a close look at how to warrant a leap from past experiments to novel conditions based on minimal assumptions about the factorization of the distribution of the manipulated system, communicated in the well-understood language of factor graph models. A postulated $\textit{interventional factor model}$ (IFM) may not always be informative, but it conveniently abstracts away a need for explicit unmeasured confounding and feedback mechanisms, leading to directly testable claims. We derive necessary and sufficient conditions for causal effect identifiability with IFMs using data from a collection of experimental settings, and implement practical algorithms for generalizing expected outcomes to novel conditions never observed in the data.
['Ricardo Silva', 'Jakob Zeitler', 'Jialin Yu', 'David S. Watson', 'Gecia Bravo-Hermsdorff']
2023-06-06
null
null
null
null
['causal-inference', 'experimental-design', 'causal-inference']
['knowledge-base', 'methodology', 'miscellaneous']
[ 8.36162150e-01 1.11653440e-01 -4.75253880e-01 -1.76756874e-01 -2.63944864e-01 -6.17029369e-01 4.84966964e-01 3.62887353e-01 -2.44076550e-01 1.12082434e+00 7.71420971e-02 -9.91159797e-01 -9.36836362e-01 -5.23792624e-01 -1.17041218e+00 -6.82350576e-01 -4.79975492e-01 2.63809085e-01 -3.95930558e-01 1.61879152e-01 1.56030566e-01 6.46661997e-01 -1.38047338e+00 -3.39419752e-01 7.15613723e-01 4.53442931e-01 2.08788544e-01 6.89239025e-01 4.65926200e-01 4.99470145e-01 -3.56328517e-01 -7.39232674e-02 1.97379515e-01 -7.46858537e-01 -5.65920413e-01 1.05864041e-01 1.14101157e-01 -3.02298874e-01 -2.14202002e-01 9.58053350e-01 4.69591737e-01 2.00588688e-01 7.97034204e-01 -1.36306882e+00 -4.95466918e-01 6.98760986e-01 -4.56336290e-01 1.86429724e-01 2.61785567e-01 2.37141535e-01 9.74600971e-01 -4.16117072e-01 5.08520484e-01 1.25617635e+00 3.72992486e-01 1.36386275e-01 -1.87245154e+00 -7.81541526e-01 3.37192357e-01 -2.36894175e-01 -1.22309399e+00 -5.17131746e-01 5.72754264e-01 -6.21376812e-01 2.55476654e-01 3.64372164e-01 5.36329448e-01 1.46464598e+00 4.90646929e-01 2.93369591e-01 1.30543864e+00 -3.93735349e-01 5.27273536e-01 -6.64588884e-02 4.65037562e-02 5.54549158e-01 8.53291392e-01 8.12967539e-01 -5.51871121e-01 -4.10177499e-01 9.07888472e-01 -4.27421881e-03 -4.64960098e-01 -5.56221068e-01 -1.15916908e+00 9.06960905e-01 1.40285105e-01 1.35534838e-01 -3.64430457e-01 9.40737426e-02 1.44887865e-01 4.83817399e-01 8.54845718e-02 8.31266701e-01 -5.60158551e-01 2.18025401e-01 -5.58714211e-01 4.09605712e-01 6.51678860e-01 7.65801370e-01 6.27146363e-01 -3.08798961e-02 3.82006355e-02 8.74544531e-02 7.14723766e-02 5.56261182e-01 -6.94126040e-02 -9.92822587e-01 1.68581665e-01 3.05032760e-01 6.30617619e-01 -1.10174513e+00 -6.54325247e-01 -4.32485938e-01 -8.53179634e-01 1.07380018e-01 7.01050282e-01 -5.76253831e-01 -6.45832062e-01 2.21920609e+00 3.55064839e-01 3.33388597e-01 -3.29318553e-01 7.66048014e-01 1.14220642e-02 3.91758531e-01 1.96697980e-01 -9.45882559e-01 8.89549017e-01 9.75074023e-02 -7.05723584e-01 -3.27260584e-01 6.69786990e-01 -4.63900596e-01 1.22172642e+00 2.95600891e-01 -8.97921920e-01 -1.81065604e-01 -1.02113426e+00 3.66439074e-01 1.90848671e-02 -1.45339563e-01 9.90788341e-01 6.18608832e-01 -6.05151057e-01 7.23439395e-01 -8.29708278e-01 -3.68606448e-01 1.91052303e-01 5.54273903e-01 -3.68540019e-01 -1.00953683e-01 -1.16879046e+00 8.56637776e-01 2.81317294e-01 2.81428248e-01 -1.25128543e+00 -9.38908041e-01 -6.64032876e-01 1.79782808e-01 8.93628597e-01 -8.47669244e-01 9.49601352e-01 -5.03277540e-01 -1.08697534e+00 2.29785681e-01 -9.30372179e-02 -2.22736135e-01 2.95696408e-01 4.77660485e-02 -2.65069991e-01 9.02375579e-02 5.11005260e-02 1.97251678e-01 8.01880479e-01 -1.11097836e+00 -9.34495777e-02 -5.13259053e-01 3.02036643e-01 -1.03637449e-01 1.14020348e-01 3.76742184e-02 4.20757234e-01 -5.79328835e-01 5.28305396e-02 -1.10756648e+00 -4.95522976e-01 -3.27749327e-02 -5.25872946e-01 2.54350156e-01 3.50843132e-01 -3.60857844e-01 1.23909271e+00 -2.07063770e+00 1.11886241e-01 3.31576794e-01 1.55447826e-01 -2.09551170e-01 6.04729354e-02 6.47431493e-01 -5.78550816e-01 3.55263770e-01 -3.05857182e-01 1.74180597e-01 1.08747140e-01 1.92992494e-01 -4.15053815e-01 9.35837567e-01 2.27772474e-01 7.23237634e-01 -7.98842609e-01 -2.89406866e-01 1.45144269e-01 1.07133523e-01 -7.09195316e-01 6.86532259e-02 -3.08330595e-01 7.63655245e-01 -5.79834282e-01 1.67664945e-01 3.12821358e-01 -4.41114575e-01 5.04822195e-01 1.94803327e-01 -1.66338474e-01 1.78579390e-01 -1.50863123e+00 1.26047969e+00 -2.52902240e-01 1.45426214e-01 7.42831379e-02 -1.58852243e+00 4.32175249e-01 3.66164029e-01 4.67458189e-01 -2.61510134e-01 3.82739604e-01 -5.61244003e-02 3.99798095e-01 -5.74694753e-01 -7.14911744e-02 -6.92238986e-01 -3.27387661e-01 5.97231627e-01 3.45516279e-02 6.44494668e-02 9.77809355e-02 9.01466981e-02 1.11773288e+00 -1.55301586e-01 4.55151796e-01 -7.52138257e-01 -4.99866903e-02 -7.09532201e-02 8.15574884e-01 1.12557411e+00 1.44730598e-01 2.89877653e-02 9.53870833e-01 1.36177251e-02 -9.76653039e-01 -1.14362788e+00 -4.01709616e-01 6.75576389e-01 -2.10794300e-01 -1.75731376e-01 -3.22672248e-01 -3.14675927e-01 1.14141211e-01 8.04589152e-01 -9.28865314e-01 -4.04784709e-01 -1.45211279e-01 -1.02933037e+00 2.23254159e-01 2.41090789e-01 2.04818435e-02 -4.96695161e-01 -5.28406620e-01 1.47006169e-01 2.31730014e-01 -7.39339769e-01 -2.01274663e-01 5.25565207e-01 -7.46733725e-01 -1.35946429e+00 -1.94244310e-01 -3.41401905e-01 8.59811962e-01 1.25189811e-01 7.96438158e-01 -8.91124755e-02 -3.85718286e-01 2.90082127e-01 6.97606802e-02 -5.43304861e-01 -4.35459375e-01 -6.02199435e-01 4.69866782e-01 -1.14787355e-01 4.61369231e-02 -7.40065157e-01 -5.72378695e-01 2.91377515e-01 -9.16907430e-01 -1.44869387e-01 5.46817958e-01 1.11961246e+00 4.22769934e-01 5.02011478e-01 6.90427184e-01 -8.92216921e-01 4.58031446e-01 -6.44462943e-01 -9.13932800e-01 2.75136530e-01 -6.48609400e-01 3.27579051e-01 6.52666509e-01 -7.39929974e-01 -1.00513482e+00 -5.98297939e-02 3.76605362e-01 -2.26281717e-01 -2.79103965e-01 9.83089030e-01 -5.28293073e-01 2.22625017e-01 7.85730004e-01 -3.33415180e-01 -2.68128254e-02 -3.94042522e-01 3.98081005e-01 7.57729411e-02 1.02503404e-01 -1.09993076e+00 5.71643829e-01 2.99951136e-01 6.35753810e-01 -8.09314013e-01 -8.29792500e-01 1.03678755e-01 -5.04608035e-01 -2.29727793e-02 6.74387932e-01 -6.42313957e-01 -1.09125268e+00 -1.46397144e-01 -4.89697039e-01 -4.81709152e-01 -3.89203668e-01 9.43297088e-01 -7.09692597e-01 7.33390450e-02 -2.57627487e-01 -8.97083819e-01 6.43137813e-01 -1.07023823e+00 5.35140216e-01 -1.28044277e-01 -4.49436873e-01 -1.08912110e+00 6.20983262e-03 -2.39600353e-02 -6.28065690e-03 4.28588033e-01 1.28849900e+00 -4.71650243e-01 -3.62116337e-01 -3.44577551e-01 2.73237377e-02 -1.04488559e-01 3.77158910e-01 3.58943455e-02 -7.01529741e-01 -3.75140548e-01 3.21540326e-01 -2.05827564e-01 5.93671978e-01 9.08783972e-01 1.13177395e+00 -6.35286748e-01 -4.53912914e-01 1.89533740e-01 1.17179430e+00 3.64156604e-01 1.86692730e-01 -3.15065593e-01 4.29577947e-01 7.46568501e-01 3.25496107e-01 4.93822366e-01 4.15160693e-02 4.93040860e-01 1.67073801e-01 -9.07722570e-04 5.81170022e-01 -5.81890821e-01 1.90330848e-01 1.63288370e-01 1.77151307e-01 -3.25563729e-01 -6.35336816e-01 2.82752275e-01 -1.64002335e+00 -1.02101219e+00 -1.33716583e-01 2.69920540e+00 8.67518067e-01 1.36277556e-01 3.00908059e-01 1.28730312e-01 7.92019188e-01 -3.21832180e-01 -7.38705099e-01 -1.64738759e-01 -6.77805543e-02 2.16865346e-01 6.13162577e-01 5.85595727e-01 -6.75691187e-01 1.32537782e-01 7.23218250e+00 3.83758038e-01 -8.44930053e-01 -1.65081024e-01 7.23136425e-01 -1.40445992e-01 -3.85456949e-01 5.92703044e-01 -5.54027021e-01 2.29341850e-01 1.27420342e+00 -6.22957706e-01 4.14247453e-01 1.87733859e-01 7.48756111e-01 -4.23003256e-01 -1.50538063e+00 3.34091574e-01 -3.95348847e-01 -8.26361597e-01 -1.95434064e-01 2.53866136e-01 6.06298089e-01 -5.91548622e-01 1.51488692e-01 -2.04525515e-02 7.83707559e-01 -1.20571911e+00 4.68241334e-01 3.73737037e-01 7.14072049e-01 -5.61941445e-01 1.82239398e-01 6.71308875e-01 -5.98967731e-01 -3.45869094e-01 -3.07025075e-01 -6.75256789e-01 -5.81655018e-02 7.47283161e-01 -7.15031505e-01 4.61283267e-01 2.74294913e-01 3.72295082e-01 -1.39914602e-01 8.07478845e-01 -2.45683849e-01 8.81778896e-01 -5.45281351e-01 7.18152896e-02 -1.72420412e-01 -1.88631445e-01 4.83722985e-01 5.99002302e-01 4.23866570e-01 4.37501669e-01 1.00272566e-01 1.00291395e+00 2.07942426e-01 -2.42313207e-03 -9.81095850e-01 -5.07218540e-01 3.95256519e-01 6.55381083e-01 -7.12624788e-01 -3.17919180e-02 -5.00276208e-01 3.15902889e-01 1.00631500e-02 6.08653128e-01 -6.94149435e-01 1.47977605e-01 4.70176965e-01 2.26246089e-01 5.77228842e-03 -3.37140828e-01 -3.01691711e-01 -1.21797800e+00 -1.36358276e-01 -9.72081065e-01 6.99308872e-01 -5.45746148e-01 -1.24128020e+00 -4.81968939e-01 6.64088190e-01 -8.47489417e-01 -3.34531307e-01 -4.89231765e-01 -3.29127133e-01 9.22898710e-01 -7.13393927e-01 -7.07090795e-01 5.40567875e-01 4.85932320e-01 1.95980612e-02 3.44864905e-01 8.16981792e-01 -4.45676334e-02 -8.54920208e-01 2.82976151e-01 4.98953648e-02 -4.06727761e-01 6.19477808e-01 -1.06960773e+00 -1.11702621e-01 9.48844552e-01 1.97822228e-02 1.11643457e+00 1.40489221e+00 -8.41135740e-01 -1.68749499e+00 -7.88273931e-01 6.05569541e-01 -2.32532695e-01 1.18349612e+00 -5.44039845e-01 -7.63110161e-01 1.00449193e+00 -2.34942302e-01 -2.46543422e-01 6.05726182e-01 5.83833933e-01 -1.78203732e-01 1.43198043e-01 -8.70024562e-01 9.67477262e-01 1.11076427e+00 -3.61096084e-01 -4.00558025e-01 4.22531635e-01 6.12614393e-01 -7.77601600e-02 -8.74292612e-01 4.36212808e-01 3.60025138e-01 -4.75003898e-01 8.96262228e-01 -1.23233056e+00 2.81981677e-01 -3.19566697e-01 -2.06158414e-01 -1.40578628e+00 -3.37628782e-01 -8.38459313e-01 3.05825412e-01 1.03097188e+00 6.26053333e-01 -5.88778377e-01 3.45003158e-01 1.01406133e+00 1.63544521e-01 -3.01253527e-01 -7.74418414e-01 -8.09429049e-01 4.30590451e-01 -5.76553226e-01 3.74090731e-01 1.19354975e+00 3.29616010e-01 4.43187565e-01 -4.85700041e-01 5.72277844e-01 7.78341830e-01 2.29426786e-01 7.24977314e-01 -1.25026691e+00 -7.17200458e-01 -2.46715605e-01 -2.55724758e-01 -7.61049211e-01 1.94261953e-01 -6.90200448e-01 -6.19517118e-02 -1.05927718e+00 3.43493432e-01 -5.50800622e-01 -2.59889960e-01 2.93634206e-01 -2.83780783e-01 -4.25306946e-01 -1.29554391e-01 -3.08711737e-01 -5.43975607e-02 4.79793936e-01 1.21585631e+00 -1.67673901e-02 -2.74016649e-01 2.64943570e-01 -1.06438291e+00 6.28590703e-01 6.72074080e-01 -5.73331237e-01 -9.14375067e-01 1.04598533e-02 5.41471541e-01 8.17852378e-01 8.20617795e-01 -3.27392846e-01 5.66017106e-02 -7.55224645e-01 3.49937856e-01 -9.83473212e-02 -4.34063561e-02 -8.56095254e-01 7.33762324e-01 4.77715254e-01 -6.78584754e-01 -7.31989741e-02 3.81912261e-01 8.78033102e-01 4.40582007e-01 -1.94830701e-01 4.13510442e-01 -4.66131642e-02 -1.09006606e-01 8.54807571e-02 -3.70667756e-01 1.28101036e-01 8.29780936e-01 1.91548422e-01 -4.12337244e-01 -3.79350007e-01 -1.06233382e+00 2.07477868e-01 4.41398203e-01 1.99422270e-01 2.64263809e-01 -1.03176343e+00 -5.24902105e-01 1.54725805e-01 -9.66852084e-02 -4.33289349e-01 5.41197479e-01 9.94416833e-01 3.06475043e-01 4.32943255e-01 8.52013156e-02 -3.26018751e-01 -7.83909559e-01 1.19087970e+00 1.46851510e-01 1.47537723e-01 -5.46103418e-01 4.14755702e-01 7.88709819e-01 5.67378774e-02 -1.30154833e-01 -2.99591064e-01 2.75354564e-01 -2.23335147e-01 4.13280636e-01 2.51243651e-01 -2.32934833e-01 -2.01789394e-01 -1.48303837e-01 -5.77274291e-03 1.01807341e-01 -1.12511054e-01 1.28194690e+00 -3.18702757e-01 -4.37313206e-02 7.92095304e-01 8.53258550e-01 -3.20729055e-02 -1.43839657e+00 -4.76714559e-02 -7.39779100e-02 -3.47534835e-01 2.83847880e-02 -7.69872367e-01 -6.85108125e-01 6.34306848e-01 2.65469968e-01 2.60839760e-01 1.02183414e+00 -2.21173130e-02 -2.44984210e-01 4.30955321e-01 4.18143302e-01 -6.88613296e-01 -3.34625185e-01 -1.64183095e-01 8.84916365e-01 -1.06068695e+00 3.30557853e-01 -3.21995705e-01 -2.13948965e-01 6.44266844e-01 9.66749191e-02 -2.37094983e-02 9.61840868e-01 3.95720214e-01 -5.75983882e-01 -2.49040872e-01 -8.40134799e-01 4.87199286e-03 9.91246775e-02 4.55674410e-01 2.77536035e-01 1.95507988e-01 -4.78644222e-01 6.51719987e-01 -1.90876409e-01 4.92532514e-02 7.39263415e-01 9.44514394e-01 -2.25981444e-01 -9.37597871e-01 -4.92236316e-01 7.49872565e-01 -5.38986266e-01 4.54707518e-02 -1.88300431e-01 1.12616193e+00 -1.41742110e-01 1.26053381e+00 -1.44667923e-01 -3.22479978e-02 3.33382517e-01 4.74992171e-02 5.07263243e-01 -5.19238293e-01 1.45453036e-01 2.29105353e-01 1.36629194e-01 -4.71351177e-01 -4.85607296e-01 -1.12648618e+00 -8.55425179e-01 -5.44324279e-01 -6.25491142e-01 1.89308032e-01 1.95248961e-01 1.20810711e+00 2.20830828e-01 5.42155921e-01 5.98716378e-01 -4.26011145e-01 -6.89782023e-01 -6.29322767e-01 -8.96797121e-01 7.35891312e-02 3.98990124e-01 -1.02024257e+00 -7.48240054e-01 1.91203192e-01]
[7.844381332397461, 5.262531280517578]
a4099606-d1a6-428a-bd2e-ab99183a9440
cross-lingual-citations-in-english-papers-a
2111.05097
null
https://arxiv.org/abs/2111.05097v2
https://arxiv.org/pdf/2111.05097v2.pdf
Cross-Lingual Citations in English Papers: A Large-Scale Analysis of Prevalence, Usage, and Impact
Citation information in scholarly data is an important source of insight into the reception of publications and the scholarly discourse. Outcomes of citation analyses and the applicability of citation based machine learning approaches heavily depend on the completeness of such data. One particular shortcoming of scholarly data nowadays is that non-English publications are often not included in data sets, or that language metadata is not available. Because of this, citations between publications of differing languages (cross-lingual citations) have only been studied to a very limited degree. In this paper, we present an analysis of cross-lingual citations based on over one million English papers, spanning three scientific disciplines and a time span of three decades. Our investigation covers differences between cited languages and disciplines, trends over time, and the usage characteristics as well as impact of cross-lingual citations. Among our findings are an increasing rate of citations to publications written in Chinese, citations being primarily to local non-English languages, and consistency in citation intent between cross- and monolingual citations. To facilitate further research, we make our collected data and source code publicly available.
['Tornike Tsereteli', 'Michael Färber', 'Tarek Saier']
2021-11-07
null
null
null
null
['cross-lingual-entity-linking', 'citation-intent-classification']
['natural-language-processing', 'natural-language-processing']
[-6.86332107e-01 -5.09255111e-01 -1.04474556e+00 1.82961062e-01 -1.13885760e+00 -9.82172132e-01 1.30717862e+00 5.94959557e-01 -5.93685329e-01 9.48804855e-01 6.92065656e-01 -9.40717280e-01 -3.88283819e-01 -4.68631715e-01 -6.31244659e-01 -2.43543684e-01 5.30749202e-01 1.59274936e-01 -2.85192281e-01 2.00484186e-01 1.11361647e+00 4.08013612e-01 -1.22686207e+00 -3.48636717e-01 1.05620384e+00 2.96986818e-01 3.71193051e-01 9.41672921e-02 -9.55034077e-01 2.93171197e-01 -8.89113486e-01 -4.19866204e-01 -2.34729499e-01 -3.48198563e-01 -6.67675853e-01 -4.43127871e-01 5.87206542e-01 4.41206723e-01 -4.17829990e-01 8.70306015e-01 1.82930842e-01 -3.53509068e-01 6.96850300e-01 -8.28292966e-01 -1.08032835e+00 9.24487829e-01 -6.79618955e-01 6.24451816e-01 3.22184026e-01 -1.69091329e-01 1.03552222e+00 -7.23935783e-01 1.23848593e+00 1.13321602e+00 4.02251035e-01 -2.84419339e-02 -9.49091375e-01 -8.01856458e-01 -1.32530257e-01 2.40890533e-01 -1.32304323e+00 -4.91081506e-01 9.19659913e-01 -9.14308190e-01 4.19436693e-01 2.82732002e-03 5.64938128e-01 1.52054763e+00 4.48352903e-01 -7.21572042e-02 1.64094925e+00 -9.10767972e-01 -5.92096767e-04 2.97073841e-01 2.35246211e-01 8.55222065e-03 8.00160289e-01 -2.84063250e-01 -8.07075143e-01 -2.41679877e-01 3.86235625e-01 4.78295796e-02 -2.42956191e-01 5.05130589e-01 -1.49151444e+00 7.02576518e-01 9.25850868e-02 1.10505497e+00 -4.36350077e-01 -2.26931632e-01 4.32433695e-01 3.60259831e-01 5.50748706e-01 4.39572096e-01 -2.54521072e-01 -6.58328950e-01 -1.15327299e+00 7.36431628e-02 1.09290290e+00 9.06341076e-01 5.03356516e-01 -1.40759304e-01 1.32458150e-01 9.19593334e-01 2.44173989e-01 6.22043908e-01 5.23692906e-01 -1.01435208e+00 4.15564060e-01 5.52621126e-01 -2.24430516e-01 -1.21998358e+00 1.02935702e-01 -8.83559525e-01 -5.05320609e-01 -2.88860738e-01 3.47460896e-01 3.38037387e-02 -2.85731941e-01 1.43963015e+00 -5.30015528e-01 -4.83426869e-01 -1.60670519e-01 7.78586328e-01 9.80814457e-01 6.50091887e-01 2.13059306e-01 -5.69986761e-01 1.24978220e+00 -5.33319712e-01 -8.26236367e-01 -9.47436274e-05 6.23423278e-01 -1.16492188e+00 9.01786208e-01 4.12257724e-02 -9.96790409e-01 -2.52360940e-01 -7.75312006e-01 -2.51309097e-01 -7.47083485e-01 -2.04038452e-02 2.84495622e-01 2.54633069e-01 -8.74159992e-01 3.69773835e-01 -3.66470188e-01 -6.68389976e-01 4.76407379e-01 -3.60169441e-01 -1.64623350e-01 -1.43890321e-01 -9.66493666e-01 1.25173438e+00 -4.49267700e-02 -2.93725491e-01 6.67856261e-02 -9.99037802e-01 -2.18957394e-01 -2.19186902e-01 2.39074439e-01 -1.96707278e-01 8.20679486e-01 -7.96057165e-01 -8.11890900e-01 1.08271039e+00 -4.44088787e-01 5.55642135e-02 4.88166690e-01 3.01751606e-02 -7.69003749e-01 -1.56525463e-01 7.98055708e-01 -3.98828425e-02 7.33557716e-02 -1.18208206e+00 -8.55335236e-01 -4.27011311e-01 -3.76000255e-01 -2.44796634e-01 -3.95711601e-01 4.66639072e-01 -7.42915869e-01 -8.55740249e-01 -3.92219052e-03 -6.77594483e-01 3.75436723e-01 -3.54623556e-01 -1.23007484e-01 -7.40487635e-01 7.67620623e-01 -8.50807309e-01 1.46154106e+00 -1.92869639e+00 1.99355289e-01 -2.95520364e-03 2.72395104e-01 -5.04005909e-01 2.10567862e-01 7.67819107e-01 4.11071599e-01 7.51894891e-01 2.36106906e-02 -2.45367792e-02 -2.90326029e-02 1.92642003e-01 -3.24206740e-01 7.61817813e-01 -2.29072899e-01 7.84749985e-01 -9.21317458e-01 -7.85371542e-01 -1.33953795e-01 3.95630032e-01 2.33950913e-01 -2.93672085e-01 1.08421832e-01 4.96354520e-01 -6.47443593e-01 1.15133560e+00 3.47411960e-01 -2.17945397e-01 -8.92349053e-03 4.50385101e-02 -1.19003117e+00 7.86698580e-01 -4.88025099e-01 1.50106823e+00 -5.51156938e-01 1.39386475e+00 -9.24947932e-02 -7.07113624e-01 7.62197137e-01 1.47555724e-01 3.19915086e-01 -1.17681026e+00 1.50722101e-01 9.04406905e-01 1.74021125e-01 -2.05697060e-01 5.58447659e-01 1.10517398e-01 -1.10451147e-01 6.26816034e-01 -2.47640863e-01 7.68419877e-02 7.30943143e-01 1.93035975e-01 5.91636598e-01 4.60783243e-02 1.51446372e-01 -7.28471160e-01 1.89420655e-01 3.63638252e-01 4.62527275e-01 8.40539396e-01 2.82413006e-01 3.44424248e-01 5.60570836e-01 9.07599255e-02 -1.29336035e+00 -5.63090086e-01 -8.91780376e-01 8.36103380e-01 -2.17916429e-01 -3.16011429e-01 -3.18082124e-01 9.71103385e-02 2.87019134e-01 8.78330052e-01 -3.57937753e-01 2.75411427e-01 -5.49111187e-01 -4.32420641e-01 4.57190692e-01 1.38423324e-01 2.04089046e-01 -1.00275052e+00 -4.69266355e-01 1.17229044e-01 -9.57387537e-02 -8.73661518e-01 -1.39430434e-01 -3.04864906e-02 -8.85894358e-01 -9.93415713e-01 -1.15863669e+00 -6.57260060e-01 3.28904033e-01 4.06874306e-02 1.19474554e+00 2.84286082e-01 -7.54662007e-02 2.42777154e-01 -2.43669063e-01 -6.72851562e-01 -4.65777040e-01 7.25539386e-01 -1.27849430e-01 -7.29743600e-01 6.79622591e-01 -1.54831678e-01 -1.28707429e-02 -4.01293576e-01 -4.88550961e-01 -4.56726491e-01 7.42170215e-01 4.79907602e-01 3.33578408e-01 -4.36947316e-01 8.23771715e-01 -7.80458450e-01 7.79431999e-01 -1.10807073e+00 -4.31593448e-01 8.38704929e-02 -1.21166909e+00 -2.53023468e-02 1.62875086e-01 -2.59896815e-01 -9.52706099e-01 -1.28618348e+00 3.48766267e-01 -1.59339726e-01 -1.28396451e-01 1.36449957e+00 5.04372954e-01 4.53997590e-02 2.25477085e-01 -5.38660921e-02 -1.02129718e-02 -8.61170352e-01 -9.26035866e-02 1.12690246e+00 5.09014606e-01 -1.05313122e+00 5.87948024e-01 4.10348224e-03 -9.36096683e-02 -1.14336920e+00 -7.45814919e-01 -4.78149563e-01 -6.56101704e-01 -3.72462571e-01 4.69922215e-01 -8.68979037e-01 -4.90556210e-01 3.80997844e-02 -1.10388088e+00 2.37165332e-01 1.08385272e-01 9.62827802e-01 2.36395359e-01 1.74010053e-01 -5.80265403e-01 -5.74131906e-01 -5.30848317e-02 -1.11313963e+00 5.94765186e-01 3.25226247e-01 -6.81641877e-01 -1.32829821e+00 2.66740561e-01 4.50993180e-01 5.58741271e-01 2.93434471e-01 1.21755075e+00 -3.71367753e-01 -4.18569148e-01 -5.31095117e-02 -3.53635609e-01 -3.40276718e-01 2.89481670e-01 5.93999743e-01 -3.98936689e-01 -3.10109407e-01 -2.16305196e-01 -1.21907614e-01 7.35555828e-01 5.15779257e-01 9.21575963e-01 -2.42645189e-01 -5.87098718e-01 1.67965636e-01 1.62354970e+00 2.15268865e-01 1.57666415e-01 1.31737030e+00 6.84052408e-01 7.40961373e-01 8.58882740e-02 2.86256075e-01 4.60418761e-01 3.91692549e-01 -1.61964074e-01 3.52058858e-01 -4.33880389e-02 1.17476717e-01 -1.56148598e-02 1.50916898e+00 -5.80546319e-01 -2.49338746e-01 -1.36064112e+00 1.07925057e+00 -1.22861350e+00 -9.58135664e-01 -7.63853312e-01 2.10957456e+00 1.02537191e+00 1.52839899e-01 -1.07251987e-01 -2.90917873e-01 5.60982406e-01 1.81419373e-01 -1.61049515e-01 -4.75928158e-01 -6.14624500e-01 1.94387674e-01 7.72895575e-01 3.19695890e-01 -3.96144599e-01 5.40499151e-01 6.55012369e+00 5.88934243e-01 -1.53291547e+00 1.23893335e-01 3.31993759e-01 -1.18096888e-01 -6.65895224e-01 3.63007694e-01 -5.31771779e-01 9.31121111e-01 1.11768341e+00 -6.69587970e-01 -4.24708650e-02 6.05532110e-01 2.65441954e-01 -8.00905302e-02 -6.69402301e-01 8.75896811e-01 1.35598555e-02 -1.67787457e+00 -2.34660488e-02 5.32206237e-01 7.23431706e-01 3.19500357e-01 5.91033548e-02 2.57359654e-01 -1.26491934e-01 -1.01505947e+00 1.17027378e+00 4.60317850e-01 9.05556619e-01 -6.83055818e-01 7.40951300e-01 1.49137452e-01 -4.06209499e-01 2.48269349e-01 -3.47106576e-01 -3.09733599e-01 -2.02359296e-02 7.26677060e-01 1.55874804e-01 6.02668047e-01 9.38364029e-01 1.00957823e+00 -6.03253782e-01 7.98388779e-01 7.67234489e-02 1.09135604e+00 -1.05372695e-02 -1.87566563e-01 4.91615772e-01 -6.38774395e-01 5.89670599e-01 1.53108692e+00 6.10824049e-01 -3.51710618e-02 -2.82383740e-01 1.06485629e+00 -5.97343624e-01 1.67148292e-01 -6.13654435e-01 -7.97558308e-01 1.14883006e+00 1.01667011e+00 -6.17183864e-01 -1.07786894e-01 -1.13740551e+00 4.28563982e-01 4.12694335e-01 4.42640096e-01 -3.33599985e-01 -4.03192669e-01 3.28241467e-01 3.55998486e-01 -1.05472527e-01 -3.78938794e-01 -6.04191422e-01 -9.76926267e-01 1.68780550e-01 -7.09301531e-01 5.90796806e-02 -4.36680019e-01 -1.58187950e+00 2.48730153e-01 4.06259065e-03 -8.14295352e-01 2.57496983e-02 -4.36967045e-01 -4.90127951e-01 1.39117563e+00 -1.47788274e+00 -9.48550701e-01 -4.72361408e-02 -1.41439468e-01 4.22363967e-01 -4.32412773e-01 4.85010028e-01 4.24614102e-01 -4.93754148e-01 2.46843591e-01 8.53556335e-01 7.25761405e-04 1.03222013e+00 -9.06291187e-01 1.78676490e-02 3.22790742e-01 3.68121229e-02 1.14605689e+00 6.45261586e-01 -9.72942472e-01 -1.67678821e+00 -4.69908386e-01 1.68800533e+00 -6.53406560e-01 1.21023655e+00 3.63899887e-01 -1.12502646e+00 7.51208842e-01 8.38784039e-01 -6.33478999e-01 7.39110112e-01 6.72925115e-01 -1.61427587e-01 2.52451390e-01 -5.34977317e-01 7.67916799e-01 8.77514958e-01 -6.90113783e-01 -1.00571597e+00 -4.64209020e-02 2.34117940e-01 -1.01279706e-01 -1.34992492e+00 -9.47577506e-02 7.36012399e-01 -2.90071040e-01 6.77644908e-01 -3.72804374e-01 9.15780425e-01 1.45036027e-01 8.98275971e-02 -1.04326618e+00 -5.94550848e-01 -3.11227471e-01 2.17079327e-01 1.79870963e+00 4.52879608e-01 -5.97978771e-01 2.29487062e-01 3.80954176e-01 -3.37077588e-01 -5.19520104e-01 -1.03471589e+00 -5.86492717e-01 8.68646622e-01 -1.54404134e-01 1.72769710e-01 1.66366851e+00 5.10909297e-02 1.55565485e-01 2.95131862e-01 -5.45646548e-01 7.10571647e-01 5.09412944e-01 3.37489814e-01 -1.78477430e+00 4.55014139e-01 -1.12366116e+00 -7.84713849e-02 -4.57384706e-01 5.60802519e-01 -9.96105194e-01 -5.47965765e-01 -1.72782946e+00 5.74707568e-01 -5.87235570e-01 -1.14470087e-01 1.80893019e-01 -1.73553899e-02 -1.57907512e-02 4.35502641e-02 1.17011845e+00 1.25425737e-02 1.52907092e-02 1.03191960e+00 -6.43541990e-03 -5.59153184e-02 -5.94463766e-01 -9.71565008e-01 6.22509539e-01 5.67023575e-01 -6.73941851e-01 3.26436281e-01 -6.88221693e-01 1.88602149e-01 -9.02848616e-02 2.34574869e-01 -5.34016490e-01 4.49677706e-01 -5.55988848e-01 3.78271103e-01 -7.52992272e-01 -3.66654336e-01 -5.49523294e-01 2.47043714e-01 1.73214078e-01 -4.79912609e-01 5.21639526e-01 2.10317284e-01 2.18476042e-01 -4.65077937e-01 -3.58747780e-01 1.65948808e-01 -2.28199139e-01 -3.65831763e-01 -4.68399003e-02 -5.65402806e-01 4.48693484e-01 6.44616246e-01 -3.29726398e-01 -5.04373610e-01 1.57080684e-02 8.66262689e-02 -4.40893173e-02 9.57605898e-01 7.29635417e-01 -7.94361383e-02 -1.26412702e+00 -1.02021933e+00 -4.72536206e-01 2.19819173e-01 -4.61218983e-01 -4.19922739e-01 9.81494308e-01 -4.36708361e-01 1.02075160e+00 -2.01633170e-01 -1.66659340e-01 -7.65091240e-01 2.85592616e-01 -3.43885064e-01 3.48051459e-01 -6.95471406e-01 1.06355861e-01 -3.65284830e-01 -8.79055187e-02 1.81966648e-01 1.35070994e-01 -3.10882747e-01 4.96775419e-01 -3.41550191e-03 8.55004966e-01 -1.15329392e-01 -1.13696039e+00 -3.73287350e-01 6.80156291e-01 3.14403586e-02 -3.35093528e-01 1.21642005e+00 -3.23964536e-01 -6.20594859e-01 1.41936362e+00 1.39649212e+00 5.37874758e-01 -1.30587175e-01 -1.59132600e-01 4.75322992e-01 -4.93928909e-01 3.98751825e-01 -7.08332241e-01 -8.84447694e-01 5.14547765e-01 6.97601512e-02 1.25686139e-01 3.81302923e-01 4.27864611e-01 3.67460549e-01 1.03292512e-02 1.97920799e-01 -1.30657268e+00 -6.11621439e-01 3.96420896e-01 9.51123536e-01 -1.09252322e+00 2.90097564e-01 7.19039142e-02 -2.01665804e-01 1.36641657e+00 2.40275934e-01 4.16616499e-01 5.94388247e-01 1.31812915e-01 2.31645733e-01 -4.02135432e-01 -4.51221704e-01 2.56053835e-01 3.94886881e-01 3.98022272e-02 1.31434941e+00 -1.46752343e-01 -1.40442860e+00 4.21982020e-01 -3.61972690e-01 -8.57223116e-04 7.66190648e-01 1.08976364e+00 -6.61344603e-02 -1.23992372e+00 -7.03767836e-01 6.91039205e-01 -1.30237949e+00 -1.07040823e-01 -6.53130472e-01 1.13832855e+00 -2.24091977e-01 8.15365613e-01 4.65511918e-01 2.98695326e-01 -9.46282316e-03 3.75506878e-02 2.87348717e-01 -3.09587032e-01 -4.22791511e-01 1.68933555e-01 1.40094489e-01 1.91128805e-01 -7.44999707e-01 -1.14858425e+00 -1.15004945e+00 -8.86281550e-01 -2.65555054e-01 5.92935264e-01 1.16162515e+00 1.17777944e+00 4.39833134e-01 6.58950031e-01 1.65331215e-01 -4.36439723e-01 -2.09665880e-01 -1.09103918e+00 -5.42123675e-01 3.85282598e-02 1.61715016e-01 -6.99489653e-01 -7.30310559e-01 1.90641358e-01]
[9.559171676635742, 8.23935604095459]
83cdfdd7-fd5f-4ec0-aa4c-9e8999d369e1
tfdet-target-aware-fusion-for-rgb-t
2305.16580
null
https://arxiv.org/abs/2305.16580v1
https://arxiv.org/pdf/2305.16580v1.pdf
TFDet: Target-aware Fusion for RGB-T Pedestrian Detection
Pedestrian detection is a critical task in computer vision because of its role in ensuring traffic safety. However, existing methods that rely solely on RGB images suffer from performance degradation under low-light conditions due to the lack of useful information. To address this issue, recent multispectral detection approaches combine thermal images to provide complementary information. Nevertheless, these approaches have limitations such as the noisy fused feature maps and the loss of informative features. In this paper, we propose a novel target-aware fusion strategy for multispectral pedestrian detection, named TFDet. Unlike existing methods, TFDet enhances features by supervising the fusion process with a correlation-maximum loss function. Our fusion strategy highlights the pedestrian-related features while suppressing the unrelated ones. TFDet achieves state-of-the-art performances on both KAIST and LLVIP benchmarks, with a speed comparable to the previous state-of-the-art counterpart. Importantly, TFDet performs remarkably well under low-light conditions, which is a significant advancement for road safety.
['Hui-Liang Shen', 'Zehua Sheng', 'Xiaohan Zhang', 'Xue Zhang']
2023-05-26
null
null
null
null
['pedestrian-detection']
['computer-vision']
[ 2.74875164e-01 -8.69732440e-01 2.61999965e-01 -2.08636492e-01 -8.22993279e-01 -2.64598846e-01 6.21212900e-01 6.27386793e-02 -7.58204281e-01 7.58718848e-01 -2.79170990e-01 -1.42524630e-01 2.39454865e-01 -8.55182827e-01 -4.09711927e-01 -1.26433170e+00 6.57751501e-01 -2.74893492e-01 8.25823247e-01 -3.36008310e-01 -8.32703188e-02 3.16598892e-01 -1.89026570e+00 9.01355445e-02 1.03207910e+00 1.04397309e+00 2.02613235e-01 3.93912643e-01 2.57693738e-01 3.19818467e-01 -3.61435831e-01 -6.62698984e-01 4.73315001e-01 -1.51473999e-01 -1.40242547e-01 1.19850270e-01 6.29816830e-01 -2.32437059e-01 -4.34279680e-01 1.26747894e+00 7.54087865e-01 1.65161327e-01 3.64430755e-01 -1.37406027e+00 -1.71157047e-01 -2.92406499e-01 -1.02521002e+00 3.27013642e-01 1.42808348e-01 5.74506640e-01 8.30489039e-01 -8.12515438e-01 9.29605681e-03 1.10699344e+00 5.54893374e-01 2.29948834e-01 -1.30395842e+00 -6.69906318e-01 6.29766956e-02 5.57957172e-01 -1.41746175e+00 -4.43160087e-01 7.89102376e-01 -3.14947277e-01 6.03996813e-01 4.75802541e-01 7.48328030e-01 8.58330369e-01 6.64727539e-02 8.13233912e-01 1.48637354e+00 -3.68441701e-01 -3.62771936e-02 1.75106049e-01 5.55592105e-02 6.64729476e-01 6.50479436e-01 5.23601532e-01 -4.87595320e-01 7.28622675e-02 4.31225568e-01 8.45840797e-02 -1.79902062e-01 -2.74108976e-01 -1.13326550e+00 6.94290936e-01 6.43988013e-01 3.36582810e-02 -4.00719970e-01 3.52222323e-02 3.14872742e-01 -7.25446716e-02 3.86014611e-01 -1.91972896e-01 -5.28036281e-02 2.79313147e-01 -8.09307575e-01 1.82460263e-01 6.05947077e-02 5.78056514e-01 8.96189094e-01 -4.52289954e-02 -3.33445072e-01 7.53238618e-01 3.84942293e-01 8.09001863e-01 -6.56859502e-02 -6.56042218e-01 4.54062521e-01 5.18539965e-01 1.78756267e-01 -8.28455746e-01 -4.52772409e-01 -6.75176263e-01 -9.38458979e-01 6.89361274e-01 6.13548934e-01 -2.76576858e-02 -7.88558304e-01 1.44136262e+00 5.99333227e-01 -3.66789661e-02 -1.17516982e-04 1.23015559e+00 9.31111336e-01 4.17094797e-01 3.21724832e-01 -1.87300548e-01 1.47268689e+00 -9.55640674e-01 -4.39348251e-01 -4.59004641e-01 9.34569687e-02 -9.90455031e-01 7.71764278e-01 4.17839438e-01 -6.34459853e-01 -7.28682816e-01 -9.40033257e-01 9.85452831e-02 -3.82189214e-01 4.27416116e-01 5.74817181e-01 1.25318837e+00 -8.96798670e-01 1.23908021e-01 -4.68064368e-01 -6.39875829e-01 4.18910861e-01 1.12777613e-01 -3.76423180e-01 -2.16758206e-01 -9.87108052e-01 9.55224633e-01 3.86804998e-01 2.85783142e-01 -5.37897229e-01 -4.97375757e-01 -7.70453930e-01 -3.03034812e-01 5.90346634e-01 -7.91299403e-01 9.19752240e-01 -4.82163072e-01 -1.24377894e+00 7.03750432e-01 -2.52636343e-01 -4.31624621e-01 9.63225722e-01 -3.32343668e-01 -4.67766106e-01 1.10099062e-01 1.84307769e-01 5.02528608e-01 8.38394642e-01 -1.26849008e+00 -1.10642874e+00 -3.11330914e-01 3.77521478e-02 1.67522550e-01 -3.65500659e-01 1.84972569e-01 -6.03047729e-01 -3.86161357e-01 -6.86461702e-02 -7.97318578e-01 -3.96936685e-01 2.89981365e-01 -4.75571215e-01 -3.30004864e-03 1.06102395e+00 -3.50550622e-01 8.98909152e-01 -2.12458849e+00 -3.86420965e-01 -1.50081113e-01 1.94026232e-01 7.59504437e-01 5.91445342e-03 1.16259873e-01 3.25282305e-01 -3.31568211e-01 -2.39100218e-01 -3.17067504e-01 -2.34511018e-01 -1.00270979e-01 1.23227738e-01 7.50764906e-01 2.25128666e-01 7.20683992e-01 -1.05823171e+00 -6.66145563e-01 9.87835884e-01 8.30397367e-01 2.51369365e-02 -1.42184645e-01 2.67497689e-01 4.84652013e-01 -4.39052761e-01 8.35402608e-01 1.11238086e+00 1.30371764e-01 -3.21600229e-01 -5.55492878e-01 -6.58251464e-01 -2.02884018e-01 -1.11209643e+00 1.28737390e+00 -1.89515844e-01 6.69293046e-01 8.81595239e-02 -7.55997181e-01 8.16319466e-01 -3.36645879e-02 5.34561694e-01 -9.33167577e-01 8.42628404e-02 2.60087937e-01 -1.77352414e-01 -3.70620698e-01 6.91386580e-01 -2.28171363e-01 9.91791561e-02 1.00439321e-02 -2.82462299e-01 9.16902348e-02 2.86493719e-01 -5.73935993e-02 8.48309636e-01 1.41285136e-01 2.96789885e-01 -2.29852125e-02 9.00628567e-01 -1.31753206e-01 5.85187018e-01 8.17677081e-01 -7.83796549e-01 5.99799156e-01 -5.86254932e-02 -3.06913942e-01 -7.77783334e-01 -1.25879717e+00 -2.13941917e-01 9.03888106e-01 5.35523415e-01 -2.02649429e-01 -4.99035716e-01 -6.37293518e-01 6.13903254e-02 5.99912882e-01 -4.22295958e-01 -2.65177339e-01 -3.44022661e-01 -1.42173016e+00 4.72220033e-01 5.20645678e-01 1.23593771e+00 -5.12405992e-01 -1.05736673e+00 2.09254175e-01 -5.28542519e-01 -1.37278354e+00 -1.79196954e-01 -7.93202817e-02 -4.03109819e-01 -1.17195177e+00 -8.73497605e-01 -2.44025066e-01 4.02802676e-01 1.14791095e+00 7.33771503e-01 -9.50243324e-02 -6.98959649e-01 2.29359090e-01 -2.53574282e-01 -4.44912642e-01 2.57355887e-02 -2.92311132e-01 -5.22108972e-02 4.52417552e-01 4.41741824e-01 -4.12798971e-02 -1.06459236e+00 5.25517464e-01 -4.43656296e-01 1.24943117e-02 7.69673526e-01 8.69316399e-01 4.43473279e-01 4.42991376e-01 3.27773750e-01 -2.81203002e-01 2.47972347e-02 6.11652583e-02 -7.27376759e-01 1.97078511e-01 -4.39858407e-01 -3.36625636e-01 4.34262514e-01 -2.34021228e-02 -1.51788092e+00 4.30504143e-01 -7.81262070e-02 2.74760816e-02 -4.49286252e-01 -2.28518277e-01 -2.89364547e-01 -5.30798912e-01 6.78916156e-01 3.01723391e-01 -1.82935417e-01 -3.88729095e-01 2.65663624e-01 5.60149789e-01 6.52809501e-01 -3.09307188e-01 9.35609996e-01 8.49549413e-01 4.31066066e-01 -1.13756609e+00 -8.26730013e-01 -1.04831231e+00 -6.77516282e-01 -6.80008233e-01 7.51681209e-01 -9.92183924e-01 -7.15266585e-01 7.64535189e-01 -9.82492328e-01 2.70615220e-01 7.86163658e-02 5.36713600e-01 -2.33810961e-01 7.28307009e-01 -2.57122427e-01 -1.38281941e+00 -2.87068158e-01 -1.21923137e+00 1.11885548e+00 4.95862484e-01 5.61971247e-01 -5.61219633e-01 -3.30639571e-01 6.15020573e-01 4.86988515e-01 3.85738641e-01 3.57182324e-01 1.70286875e-02 -6.51200771e-01 -3.19662273e-01 -7.67669320e-01 3.48317564e-01 1.25308931e-01 3.03319320e-02 -1.33917630e+00 -1.44614577e-01 -3.40548724e-01 -1.78881958e-02 1.56083560e+00 4.97528911e-01 5.91715515e-01 4.07240003e-01 -4.84217703e-01 3.73354584e-01 1.61580348e+00 -2.61691604e-02 5.54165363e-01 5.81727803e-01 6.51575923e-01 7.20494270e-01 8.94930065e-01 3.24520200e-01 3.62799436e-01 8.93015862e-01 6.32486403e-01 -5.17072737e-01 -3.91096592e-01 2.65689731e-01 3.55456561e-01 -6.66152239e-02 -2.74257630e-01 -1.54144198e-01 -7.95295238e-01 4.71832037e-01 -2.19650388e+00 -1.03406036e+00 -8.49642456e-01 2.39752197e+00 3.80827427e-01 2.58783489e-01 6.66518569e-01 4.28619981e-01 1.00410569e+00 1.77824467e-01 -4.48255271e-01 2.84530580e-01 -5.22633314e-01 -1.49855018e-01 7.76686430e-01 1.15401328e-01 -1.65410328e+00 8.37567329e-01 5.98676252e+00 9.00585711e-01 -9.32409406e-01 2.89220423e-01 5.16294003e-01 -2.62220185e-02 3.52231950e-01 -2.81497985e-01 -8.21748614e-01 5.11042953e-01 5.15097201e-01 1.57781839e-01 -4.58807573e-02 5.60644507e-01 4.23119009e-01 -7.40742266e-01 -3.65625203e-01 1.15449095e+00 9.68082249e-02 -7.06601143e-01 -3.29315931e-01 6.64118230e-02 4.69025731e-01 2.21723784e-03 2.09300622e-01 -4.84429188e-02 1.29862964e-01 -6.05161011e-01 7.84130573e-01 4.21164632e-01 5.73480487e-01 -9.29198444e-01 8.04225981e-01 2.07297683e-01 -1.63648951e+00 -1.44052222e-01 -4.92573470e-01 -3.18883993e-02 3.84752870e-01 9.32160020e-01 -3.23431492e-01 9.68175650e-01 8.46228898e-01 6.79072797e-01 -8.85141790e-01 1.64894068e+00 -4.02670979e-01 4.51542139e-01 -5.08862376e-01 -3.84112308e-03 2.54279405e-01 -2.94132501e-01 6.82091355e-01 1.35623538e+00 2.41206825e-01 -2.13900834e-01 4.15156156e-01 6.58571422e-01 5.19836903e-01 -7.63411522e-02 -4.80433673e-01 5.57463765e-01 1.74612641e-01 1.59749758e+00 -8.34001362e-01 -1.96293682e-01 -8.07565153e-01 9.48004305e-01 -1.71842530e-01 3.64103436e-01 -9.13467050e-01 -2.86103457e-01 6.97481513e-01 7.35826567e-02 4.34416085e-01 -2.17555135e-01 -3.74059200e-01 -9.97183383e-01 2.85231590e-01 -2.80367523e-01 3.92315000e-01 -5.20121634e-01 -1.10644805e+00 4.37531561e-01 -5.06932139e-02 -1.48348653e+00 3.65792513e-01 -7.33901680e-01 -5.19473791e-01 7.94311345e-01 -2.03614712e+00 -1.64524233e+00 -6.38838172e-01 6.41345918e-01 3.70242685e-01 7.86409304e-02 2.82479018e-01 5.40234506e-01 -8.58869970e-01 4.31141645e-01 9.00819972e-02 -6.37698993e-02 8.61930251e-01 -1.07084763e+00 2.78803587e-01 1.43690491e+00 -2.62590766e-01 1.31816164e-01 8.03656340e-01 -4.56344604e-01 -1.11575747e+00 -1.24535406e+00 4.92915660e-01 -2.10137486e-01 4.27931905e-01 -1.72882378e-01 -5.98684847e-01 -5.85027635e-02 9.31976140e-02 2.99240500e-01 6.14844382e-01 -1.75937936e-01 -4.39098120e-01 -4.85982567e-01 -1.14324188e+00 4.36665028e-01 9.85942543e-01 -2.77020127e-01 -2.56169915e-01 9.36005041e-02 1.63452074e-01 -8.05348437e-03 -3.15721124e-01 4.47469383e-01 7.28494704e-01 -1.40502572e+00 1.38100028e+00 5.28340749e-02 -1.84260726e-01 -8.62155735e-01 -1.68822452e-01 -1.16713238e+00 -5.10656118e-01 -4.39821810e-01 1.75158486e-01 1.35420227e+00 7.22090453e-02 -6.83908641e-01 5.65686047e-01 3.28651011e-01 4.07254398e-02 -2.30043679e-01 -1.13058031e+00 -1.03852594e+00 -2.67371207e-01 -4.59787071e-01 3.51832271e-01 3.24477911e-01 -5.91295362e-01 1.70818821e-01 -5.71339011e-01 2.50837594e-01 1.47435510e+00 1.52930580e-02 7.64715910e-01 -1.36486828e+00 1.22198500e-02 -5.46013772e-01 -5.93029976e-01 -6.42699599e-01 -1.99198529e-01 -3.03295255e-01 2.43704334e-01 -1.57973409e+00 4.97252375e-01 -2.49098048e-01 -5.56307495e-01 3.00180465e-01 -7.25801229e-01 8.16548049e-01 3.48364651e-01 2.77400352e-02 -8.64195049e-01 5.22686481e-01 1.05257559e+00 -3.15891117e-01 1.07253656e-01 3.46893370e-02 -6.47421122e-01 5.68274558e-01 8.11033428e-01 -1.60055831e-01 7.50845298e-03 -9.20051858e-02 -2.03470066e-01 -5.47136128e-01 1.02825356e+00 -1.46284652e+00 3.55047226e-01 -2.23453254e-01 6.46331608e-01 -9.62183654e-01 6.69524491e-01 -7.24280715e-01 -5.52053824e-02 7.19998956e-01 3.03427994e-01 -3.86485159e-01 2.37978369e-01 7.60495067e-01 -1.21833473e-01 1.26915917e-01 1.11430180e+00 -5.02984673e-02 -1.07127333e+00 1.27175525e-01 -5.24872124e-01 -4.06232148e-01 1.23673427e+00 -3.70850533e-01 -7.39153862e-01 -1.81394853e-02 -4.62340154e-02 1.47519097e-01 4.22937244e-01 2.86164850e-01 5.81039488e-01 -1.29979074e+00 -8.78591776e-01 2.57035434e-01 3.88950378e-01 -2.57874250e-01 5.34618020e-01 1.21830475e+00 -1.28391951e-01 4.76125330e-01 -3.63604665e-01 -8.29119802e-01 -1.61336112e+00 5.45728803e-01 2.94003516e-01 -1.21134883e-02 -6.48278654e-01 6.09343946e-01 3.92581403e-01 2.86786258e-02 -6.81424290e-02 4.72231880e-02 -1.87962189e-01 1.44633308e-01 8.23880970e-01 7.46512890e-01 2.39773542e-01 -1.07741416e+00 -5.91823101e-01 8.24927688e-01 2.86033079e-02 4.05915640e-02 9.84567523e-01 -3.67033869e-01 1.18025534e-01 1.03878208e-01 8.28079700e-01 -1.80384099e-01 -1.46457684e+00 -3.62322330e-01 -3.41331631e-01 -8.64328861e-01 3.64347786e-01 -7.61488438e-01 -1.09702265e+00 1.04211903e+00 9.17888284e-01 3.66070382e-02 1.35081959e+00 -1.92988187e-01 7.71434844e-01 1.36379302e-01 4.00927454e-01 -1.15947616e+00 7.77529925e-02 1.89597651e-01 4.55307901e-01 -1.70042014e+00 2.36377716e-01 -8.26241493e-01 -4.96487826e-01 1.05606878e+00 6.37731373e-01 4.13227558e-01 3.93194377e-01 4.76279743e-02 1.49202004e-01 1.03320934e-01 -1.93812832e-01 -1.18561566e+00 4.16677892e-01 9.10594165e-01 1.63139418e-01 1.08572267e-01 -3.21842492e-01 1.11187749e-01 2.35437915e-01 -2.02099815e-01 7.42731690e-02 8.95894766e-01 -7.24979758e-01 -9.68504429e-01 -8.17392290e-01 3.15745026e-01 -2.80804038e-01 2.04178002e-02 -3.40983599e-01 6.91668212e-01 2.67989546e-01 1.50887358e+00 -3.91762763e-01 -6.03026390e-01 5.17554343e-01 -2.69674271e-01 5.30886471e-01 -1.12675220e-01 -6.05044246e-01 3.42183352e-01 3.78976539e-02 -6.07216477e-01 -7.80102074e-01 -9.70441580e-01 -7.10366786e-01 -2.35250160e-01 -6.78796232e-01 -3.34195316e-01 5.49707949e-01 8.70641947e-01 3.45546678e-02 4.54351276e-01 6.41066134e-01 -9.26634669e-01 -2.45550603e-01 -6.81749463e-01 -5.53299487e-01 4.36332434e-01 4.86447901e-01 -1.04700673e+00 -1.56422436e-01 -2.56635576e-01]
[9.828218460083008, -1.390444040298462]
37b69289-49fb-4ed9-8598-52d4cd85f980
improving-toponym-resolution-with-better
2305.11315
null
https://arxiv.org/abs/2305.11315v1
https://arxiv.org/pdf/2305.11315v1.pdf
Improving Toponym Resolution with Better Candidate Generation, Transformer-based Reranking, and Two-Stage Resolution
Geocoding is the task of converting location mentions in text into structured data that encodes the geospatial semantics. We propose a new architecture for geocoding, GeoNorm. GeoNorm first uses information retrieval techniques to generate a list of candidate entries from the geospatial ontology. Then it reranks the candidate entries using a transformer-based neural network that incorporates information from the ontology such as the entry's population. This generate-and-rerank process is applied twice: first to resolve the less ambiguous countries, states, and counties, and second to resolve the remaining location mentions, using the identified countries, states, and counties as context. Our proposed toponym resolution framework achieves state-of-the-art performance on multiple datasets. Code and models are available at \url{https://github.com/clulab/geonorm}.
['Steven Bethard', 'Zeyu Zhang']
2023-05-18
null
null
null
null
['toponym-resolution']
['natural-language-processing']
[-2.17765898e-01 1.65802911e-01 -3.03191334e-01 -3.99608910e-01 -1.01548696e+00 -5.05436182e-01 8.23486507e-01 8.25916231e-01 -4.50615525e-01 8.16811442e-01 1.04893184e+00 -3.70577067e-01 -1.49914518e-01 -1.25249600e+00 -5.68970203e-01 -1.55356333e-01 9.76954773e-02 5.53839087e-01 1.74872890e-01 -2.42452264e-01 4.61552501e-01 2.89674371e-01 -1.58309495e+00 3.31441909e-01 1.13853979e+00 1.00017822e+00 9.83269364e-02 1.90058321e-01 -6.52797222e-01 9.52168286e-01 -2.30072811e-01 -3.78106594e-01 1.80975229e-01 1.18992776e-01 -8.01277220e-01 -9.67600822e-01 4.66117769e-01 -1.62599176e-01 -3.81241560e-01 1.34658504e+00 2.62051642e-01 4.18830141e-02 6.70035243e-01 -1.06644654e+00 -1.04474211e+00 1.05000794e+00 -5.55961370e-01 2.27348626e-01 7.34170377e-01 -5.42214274e-01 1.11632991e+00 -1.01816893e+00 6.74856246e-01 9.87816513e-01 7.92498350e-01 -9.80322994e-03 -4.38754857e-01 -7.98255503e-01 -6.20831326e-02 -1.21955876e-03 -1.92029119e+00 -4.84516323e-01 2.77294278e-01 -5.96906424e-01 9.78619576e-01 2.12358281e-01 2.72340447e-01 5.28677642e-01 1.11081824e-01 1.39770597e-01 6.85674965e-01 -4.59257096e-01 2.64395297e-01 -3.74716282e-01 8.49073529e-02 6.19079590e-01 6.03256345e-01 -1.65947318e-01 -4.43203986e-01 -4.79189306e-01 4.73435372e-01 2.08145037e-01 6.15429878e-02 -9.73615423e-02 -1.14823699e+00 5.42852044e-01 6.36951208e-01 4.23754424e-01 -6.99930072e-01 2.86272645e-01 1.90518767e-01 -2.42405236e-01 6.30420268e-01 4.04256552e-01 -1.25918359e-01 -2.46047713e-02 -1.20123339e+00 9.46085602e-02 6.65012538e-01 1.04077590e+00 1.03357744e+00 -3.29967827e-01 -1.04730045e-02 8.28187108e-01 3.80322725e-01 3.09266627e-01 4.32206601e-01 -8.54511201e-01 1.10644710e+00 1.00180984e+00 5.56302607e-01 -1.31717002e+00 -3.97992700e-01 -7.02836290e-02 -5.38030744e-01 -3.34154248e-01 1.10128574e-01 -1.19889237e-01 -1.03465760e+00 1.52426827e+00 3.64788264e-01 4.87533569e-01 6.07279874e-02 6.73175991e-01 1.07830262e+00 7.72617102e-01 4.90386933e-01 3.99694949e-01 1.48762822e+00 -4.24465567e-01 -5.49751878e-01 -1.27125010e-01 5.21737695e-01 -4.40227628e-01 6.51810706e-01 -5.08778214e-01 -7.70816386e-01 -1.01320609e-01 -7.93558180e-01 -3.21000010e-01 -1.09120059e+00 2.59849280e-01 5.42347193e-01 3.97185683e-01 -1.19833827e+00 3.76873851e-01 -6.66024387e-01 -6.81338668e-01 1.51182875e-01 4.82636467e-02 -5.41437984e-01 1.26891747e-01 -1.70900083e+00 7.47298539e-01 6.57092035e-01 -9.57486928e-02 -2.97725290e-01 -8.39012802e-01 -1.27034795e+00 3.40499729e-01 -5.90468682e-02 -6.02136552e-01 7.41733313e-01 -7.32168928e-02 -7.16884851e-01 8.08086753e-01 -3.60314041e-01 -4.45236117e-01 1.64596066e-01 -8.59574825e-02 -1.00058019e+00 -7.92495385e-02 1.02323580e+00 3.59477818e-01 4.96017076e-02 -9.27350044e-01 -1.23160911e+00 -4.83639747e-01 -7.02056438e-02 1.97823837e-01 -2.31844082e-01 1.45431414e-01 -7.85553515e-01 -7.62969434e-01 4.73754615e-01 -5.33854187e-01 -1.83559254e-01 -5.03127813e-01 -4.73577082e-01 -1.78740561e-01 1.36844084e-01 -1.17029679e+00 1.94107533e+00 -1.96004057e+00 -4.97991651e-01 6.02701664e-01 1.62014470e-01 -2.38502115e-01 -7.34284446e-02 7.60055602e-01 -1.93615422e-01 3.99896741e-01 -1.06174812e-01 1.96666233e-02 1.03078641e-01 -1.89357907e-01 -4.88590747e-01 2.05481514e-01 -1.27140030e-01 6.36722147e-01 -1.16562605e+00 -5.15077531e-01 1.02539837e-01 3.98582846e-01 -3.82821232e-01 -4.30031002e-01 1.82042032e-01 -4.06419449e-02 -5.60402334e-01 8.16224933e-01 6.54016018e-01 -1.57968834e-01 3.09490830e-01 -1.90220833e-01 -5.18011451e-01 6.87001348e-01 -1.33692706e+00 1.56071711e+00 -4.71823543e-01 3.25564712e-01 -4.15224373e-01 -4.56765145e-01 1.10315919e+00 2.38529921e-01 8.17181587e-01 -7.60441065e-01 -8.24147612e-02 4.66179967e-01 -8.02781343e-01 -4.07084912e-01 1.08303952e+00 3.60077798e-01 -7.20506132e-01 3.51033241e-01 -1.76909938e-01 3.24912637e-01 2.89935827e-01 3.10324043e-01 1.11825192e+00 2.13788092e-01 6.37832284e-01 -3.02963525e-01 4.20105517e-01 3.37603062e-01 5.76830626e-01 6.58811331e-01 7.45607540e-02 3.94550234e-01 2.97920734e-01 -7.56295323e-01 -1.07333994e+00 -1.17384768e+00 -9.95430425e-02 1.08796966e+00 2.00068310e-01 -6.42480731e-01 -6.58146143e-01 -3.29120606e-01 3.55963886e-01 8.54894698e-01 -7.44640768e-01 1.15889646e-01 -3.22220325e-01 -4.92306948e-01 8.20783079e-01 5.01480460e-01 4.52980548e-01 -7.54360616e-01 -3.52798671e-01 3.05841863e-01 -5.39238513e-01 -9.07752931e-01 -3.09824377e-01 -4.16119359e-02 -4.84398216e-01 -9.21974957e-01 -5.14365792e-01 -7.07812369e-01 6.75417066e-01 -3.78272049e-02 9.23541605e-01 -2.93060038e-02 2.49332160e-01 -1.28698647e-01 -3.58150303e-01 4.68783910e-05 -7.74524510e-02 3.11600238e-01 1.95775390e-01 -2.09261194e-01 7.97722220e-01 -5.66400588e-01 -4.40583050e-01 8.57206658e-02 -6.22474611e-01 -9.70046595e-02 1.20345697e-01 3.04740578e-01 6.59624040e-01 7.72761628e-02 4.50070441e-01 -7.78967977e-01 6.53102517e-01 -9.20221448e-01 -8.27042282e-01 5.33605039e-01 -3.61784518e-01 1.90321982e-01 5.23942053e-01 3.45678508e-01 -1.03863859e+00 2.46207476e-01 -7.05286041e-02 9.51225907e-02 -1.41640574e-01 1.17768788e+00 -7.45912790e-02 2.54486263e-01 6.62209094e-01 3.79414596e-02 -7.45975077e-01 -6.07904553e-01 5.13039708e-01 1.01002622e+00 1.02373457e+00 -7.70908177e-01 7.25571036e-01 5.07158399e-01 -4.39639270e-01 -2.95508951e-01 -6.39294446e-01 -8.38063538e-01 -7.88562000e-01 1.02900073e-01 9.01208222e-01 -1.37966871e+00 -1.92951664e-01 -4.72457940e-03 -1.11385560e+00 5.74810468e-02 -4.69863676e-02 4.47427839e-01 -1.23639427e-01 5.93209006e-02 -4.06426102e-01 -5.17682552e-01 -3.87865961e-01 -6.56917751e-01 1.03924084e+00 3.36539537e-01 -3.36454123e-01 -8.62030447e-01 3.05322617e-01 2.48879969e-01 2.85451621e-01 4.23762679e-01 9.30539429e-01 -8.40315580e-01 -3.61562729e-01 -4.37149942e-01 -5.77862203e-01 -6.67534411e-01 5.74180007e-01 -1.59323569e-02 -6.72060490e-01 1.75670058e-01 -8.31665874e-01 5.05624056e-01 7.65222430e-01 2.18879938e-01 8.97701323e-01 -6.24373734e-01 -8.07001352e-01 8.46472859e-01 1.72851396e+00 2.78507560e-01 7.61164963e-01 7.89841533e-01 7.42641509e-01 3.68218660e-01 5.49607337e-01 5.60414314e-01 9.71239328e-01 5.39628565e-01 3.36449355e-01 9.64800194e-02 1.13948338e-01 -7.89874792e-01 -1.70213729e-01 5.65579116e-01 -1.01768747e-01 -2.52121091e-01 -1.66029811e+00 1.10807228e+00 -1.89835715e+00 -1.28765023e+00 -7.86158815e-02 2.09216094e+00 7.60900855e-01 -3.21656883e-01 -1.16648249e-01 -4.25692588e-01 1.14596462e+00 2.36040667e-01 -2.12061524e-01 -1.96923554e-01 -5.24828397e-02 7.33895227e-02 9.91714239e-01 6.80562854e-01 -1.24733865e+00 1.40726388e+00 6.09420156e+00 4.34111953e-01 -1.03896224e+00 4.42190766e-02 3.92944068e-01 5.95071092e-02 -6.28626823e-01 2.04161853e-01 -9.73526359e-01 7.97079980e-01 8.30991089e-01 -5.55067897e-01 4.16500270e-01 8.30963910e-01 3.12128961e-01 -4.97350581e-02 -5.19598961e-01 7.92266250e-01 -1.32191777e-01 -1.72548473e+00 2.40353480e-01 -2.32175346e-02 8.61491323e-01 4.61818278e-01 -1.37642682e-01 9.39338654e-02 1.05021381e+00 -7.87289083e-01 1.00454104e+00 6.24907255e-01 1.19020736e+00 -6.35046899e-01 7.33507693e-01 -1.19332083e-01 -1.80863500e+00 -2.80139804e-01 -4.55384910e-01 1.12720355e-02 2.34431952e-01 6.89914167e-01 -7.73039222e-01 7.69842029e-01 1.09962380e+00 7.62518764e-01 -6.57002449e-01 1.14355075e+00 -3.49953949e-01 2.29160398e-01 -4.51543182e-01 2.71183759e-01 2.78426647e-01 -2.28811771e-01 2.77972788e-01 1.18259978e+00 1.08730125e+00 1.25948101e-01 -1.67276934e-01 5.57383716e-01 -3.26396674e-01 1.99757427e-01 -7.90285230e-01 1.45021483e-01 1.46783280e+00 9.07151461e-01 -6.31092250e-01 -4.92650956e-01 -2.77110845e-01 5.51234007e-01 5.31817913e-01 4.35344875e-01 -7.82327771e-01 -7.33049154e-01 7.36655354e-01 3.07116836e-01 1.14773750e-01 -6.22422732e-02 -4.20325547e-01 -1.11282957e+00 1.78719885e-04 -4.04246062e-01 8.93378198e-01 -1.02054346e+00 -8.63872886e-01 4.42595929e-01 8.46045092e-02 -1.33023083e+00 -4.32624400e-01 -1.81236461e-01 -6.15273058e-01 1.01209366e+00 -1.39961720e+00 -1.20812392e+00 -5.14352143e-01 4.02254999e-01 -5.97208552e-02 -1.38467550e-01 7.08060205e-01 6.89970255e-01 -3.48252743e-01 4.07433391e-01 4.19285953e-01 6.53136849e-01 7.21908331e-01 -1.17721570e+00 9.33627129e-01 1.04281390e+00 -6.48611337e-02 1.05579007e+00 3.70025039e-01 -1.21685135e+00 -7.28584826e-01 -1.44862688e+00 1.65015864e+00 -5.10125279e-01 1.06497467e+00 -9.98641625e-02 -5.95343769e-01 8.31062853e-01 -9.59933177e-02 -4.98524122e-02 6.51184142e-01 1.47527710e-01 -6.41478002e-01 -1.47994608e-01 -1.14497769e+00 6.91621125e-01 1.15357530e+00 -7.57484972e-01 -6.92932785e-01 5.39763384e-02 6.17677629e-01 -3.86225849e-01 -1.12140751e+00 2.29687300e-02 6.74284875e-01 -4.23771977e-01 1.10626602e+00 -3.43837976e-01 4.44740206e-01 -7.50724792e-01 -4.99898195e-01 -1.31537902e+00 -5.82697511e-01 -2.78672963e-01 3.05657119e-01 1.38102579e+00 8.24536383e-01 -8.26521575e-01 6.05086029e-01 6.44229949e-01 -1.32260323e-01 -1.68949634e-01 -1.09564722e+00 -5.87555945e-01 -7.44626066e-03 -2.40364417e-01 1.59079349e+00 1.45173204e+00 2.97841221e-01 -1.65529937e-01 -2.98868686e-01 5.81753790e-01 4.99163955e-01 -4.59664762e-02 2.97799885e-01 -1.52264273e+00 6.92463934e-01 -2.62810469e-01 -3.95117342e-01 -3.22248429e-01 5.29158600e-02 -1.03596604e+00 -5.88652976e-02 -2.20374322e+00 -1.57100931e-01 -8.57912183e-01 -4.80699658e-01 1.01185298e+00 -8.96669775e-02 2.30926514e-01 -7.45997876e-02 5.53764999e-01 -5.95527530e-01 7.56916776e-02 3.03102493e-01 -1.63219854e-01 -2.97687441e-01 -5.19590497e-01 -1.04420102e+00 6.08767688e-01 9.14786220e-01 -7.27160811e-01 3.39635089e-02 -8.50742996e-01 7.81544149e-01 4.83911810e-03 2.44021177e-01 -1.30688536e+00 7.56218314e-01 -3.43452573e-01 3.30887347e-01 -7.75680363e-01 -1.82048120e-02 -8.18081141e-01 6.15955114e-01 2.48810425e-01 -2.41284862e-01 5.35217822e-01 5.21318875e-02 2.99290031e-01 -3.55343491e-01 -1.25307128e-01 1.75759733e-01 -2.01817930e-01 -1.07100105e+00 3.59218776e-01 -2.84959763e-01 8.43269657e-03 6.43956304e-01 -3.47745717e-01 -7.23715484e-01 -1.81810677e-01 -6.21488094e-01 3.06992799e-01 8.57506573e-01 6.26128376e-01 3.08014423e-01 -1.72446871e+00 -5.80602169e-01 -9.81438607e-02 5.14458656e-01 -1.65860891e-01 3.11227478e-02 4.58816350e-01 -9.77017820e-01 5.46411157e-01 -2.51932383e-01 1.84497125e-02 -7.54386544e-01 3.10786366e-01 2.76894122e-01 -1.76279604e-01 -4.61789548e-01 4.32359815e-01 -1.99548185e-01 -7.06425130e-01 -1.00835316e-01 -4.08479482e-01 -5.84325850e-01 4.38032180e-01 6.48117244e-01 3.81570637e-01 1.74324691e-01 -1.00135171e+00 -7.87341714e-01 6.77998364e-01 3.53467017e-01 -1.79019585e-01 1.54191601e+00 -2.69771695e-01 -3.37686986e-01 -4.43400443e-02 9.32681501e-01 3.07646990e-01 -6.66853607e-01 -3.23221117e-01 6.07528150e-01 -4.02186036e-01 -1.20703995e-01 -7.02350855e-01 -8.86882007e-01 2.47272328e-01 3.15041125e-01 -3.25832330e-02 8.60790610e-01 6.96364641e-02 5.40907562e-01 1.80878401e-01 7.13972449e-01 -1.28929603e+00 -7.34202921e-01 8.94021511e-01 6.35743737e-01 -8.18674564e-01 4.94055077e-02 -3.07312191e-01 -3.11599374e-01 7.56791711e-01 2.29580179e-01 1.67459592e-01 7.87515879e-01 2.72085816e-01 1.24476030e-01 -3.23162824e-01 -3.03305984e-01 -3.67292404e-01 1.33911103e-01 7.08761573e-01 4.60207820e-01 2.23300219e-01 -2.47258082e-01 5.76996863e-01 -6.15561604e-01 1.91648021e-01 3.12543720e-01 7.29190171e-01 -5.76644003e-01 -7.44053841e-01 -6.69311762e-01 5.98840058e-01 -3.91347885e-01 -6.13584399e-01 -2.57469147e-01 5.17942548e-01 3.08888614e-01 7.43619382e-01 5.27134001e-01 -3.67091537e-01 1.53733939e-01 5.24807349e-02 -4.50146317e-01 -6.67258799e-01 -4.66412693e-01 -4.10608292e-01 3.21535140e-01 -5.73840261e-01 -9.71549377e-02 -6.13348484e-01 -1.42554557e+00 -4.13809210e-01 1.00011103e-01 5.00714600e-01 6.93253815e-01 6.21563852e-01 7.94836402e-01 2.21660405e-01 2.59982347e-01 -4.88763839e-01 1.34010583e-01 -7.01746345e-01 -3.90388995e-01 3.33663613e-01 3.86614241e-02 -5.72494507e-01 -1.03402644e-01 -7.57643878e-02]
[9.373022079467773, 9.082247734069824]
4ca5751e-806a-4b7a-9096-bfc3b1fcb15c
pose-disentangled-contrastive-learning-for
2211.13490
null
https://arxiv.org/abs/2211.13490v2
https://arxiv.org/pdf/2211.13490v2.pdf
Pose-disentangled Contrastive Learning for Self-supervised Facial Representation
Self-supervised facial representation has recently attracted increasing attention due to its ability to perform face understanding without relying on large-scale annotated datasets heavily. However, analytically, current contrastive-based self-supervised learning (SSL) still performs unsatisfactorily for learning facial representation. More specifically, existing contrastive learning (CL) tends to learn pose-invariant features that cannot depict the pose details of faces, compromising the learning performance. To conquer the above limitation of CL, we propose a novel Pose-disentangled Contrastive Learning (PCL) method for general self-supervised facial representation. Our PCL first devises a pose-disentangled decoder (PDD) with a delicately designed orthogonalizing regulation, which disentangles the pose-related features from the face-aware features; therefore, pose-related and other pose-unrelated facial information could be performed in individual subnetworks and do not affect each other's training. Furthermore, we introduce a pose-related contrastive learning scheme that learns pose-related information based on data augmentation of the same image, which would deliver more effective face-aware representation for various downstream tasks. We conducted linear evaluation on four challenging downstream facial understanding tasks, ie, facial expression recognition, face recognition, AU detection and head pose estimation. Experimental results demonstrate that our method significantly outperforms state-of-the-art SSL methods. Code is available at https://github.com/DreamMr/PCL}{https://github.com/DreamMr/PCL
['Shaoze Feng', 'Kejun Liu', 'Zhe Chen', 'Yibing Zhan', 'Wenbin Wang', 'Yuanyuan Liu']
2022-11-24
null
http://openaccess.thecvf.com//content/CVPR2023/html/Liu_Pose-Disentangled_Contrastive_Learning_for_Self-Supervised_Facial_Representation_CVPR_2023_paper.html
http://openaccess.thecvf.com//content/CVPR2023/papers/Liu_Pose-Disentangled_Contrastive_Learning_for_Self-Supervised_Facial_Representation_CVPR_2023_paper.pdf
cvpr-2023-1
['head-pose-estimation', 'facial-expression-recognition']
['computer-vision', 'computer-vision']
[ 2.42366735e-02 1.28780678e-01 -3.51287305e-01 -8.40252221e-01 -7.81485617e-01 -2.90426642e-01 5.22641599e-01 -4.79141563e-01 -5.35399653e-02 5.84358215e-01 3.82101953e-01 2.61283845e-01 -8.29438865e-02 -3.98224920e-01 -7.71319687e-01 -8.96171451e-01 -6.08716570e-02 2.32522249e-01 -5.66418052e-01 -3.06131274e-01 -2.26655364e-01 5.97959280e-01 -1.74290013e+00 2.64655858e-01 4.84434873e-01 1.15025151e+00 -1.23922899e-01 6.02944009e-02 3.45104724e-01 5.43479919e-01 -1.81378633e-01 -4.49569225e-01 2.18716428e-01 -3.43969882e-01 -4.87821251e-01 1.20230101e-01 7.72203445e-01 -4.85786557e-01 -5.43230116e-01 8.70292425e-01 6.27582014e-01 -2.57120103e-01 6.02311194e-01 -1.58546627e+00 -5.10376513e-01 2.22119629e-01 -1.05103517e+00 7.06136078e-02 3.36409211e-01 1.42799497e-01 9.63588238e-01 -1.34221697e+00 2.96683818e-01 1.50432384e+00 4.57327902e-01 9.45989311e-01 -1.10274136e+00 -1.49186337e+00 3.01720798e-01 8.15133825e-02 -1.65150404e+00 -1.04907072e+00 1.18826807e+00 -2.07936555e-01 3.96358490e-01 2.24972099e-01 5.15424728e-01 1.35711753e+00 -1.38043478e-01 9.99445617e-01 1.17011774e+00 8.89119634e-04 -1.94026977e-01 1.07972315e-02 -1.67511180e-01 1.18318999e+00 1.70785159e-01 1.87858686e-01 -9.77929413e-01 -1.39788702e-01 8.41012359e-01 1.04724253e-02 -4.03471708e-01 -5.30695498e-01 -7.88662791e-01 7.14569390e-01 3.86897802e-01 -7.25833625e-02 -1.01841964e-01 1.99673310e-01 4.48045999e-01 1.11018248e-01 8.47474337e-01 -6.40930980e-02 -6.81757629e-01 9.81611535e-02 -6.03994727e-01 8.18156525e-02 4.30759490e-01 7.68709362e-01 1.02030015e+00 2.12426811e-01 -7.40313828e-02 7.85625637e-01 6.87927663e-01 4.91224587e-01 2.02702045e-01 -8.36043477e-01 3.19393158e-01 6.37257338e-01 -5.33221602e-01 -1.09792757e+00 -4.99739140e-01 -4.66004193e-01 -9.14008439e-01 1.34862944e-01 1.83588892e-01 -3.12492758e-01 -6.30296111e-01 2.47938061e+00 3.52960348e-01 4.34125781e-01 -1.54474601e-01 9.24292743e-01 1.13743365e+00 3.34590733e-01 1.79539025e-01 -3.66882265e-01 1.41339338e+00 -6.62296772e-01 -6.88988090e-01 -2.75580764e-01 5.36446393e-01 -4.40308243e-01 8.42623055e-01 2.56433845e-01 -9.52876508e-01 -4.58215803e-01 -8.66818488e-01 -4.40634936e-02 1.02242390e-02 4.28450286e-01 1.01303589e+00 6.24877810e-01 -9.41245258e-01 2.47183323e-01 -7.17534602e-01 -2.51096915e-02 1.03819633e+00 6.22948050e-01 -9.13211882e-01 -2.23670825e-02 -1.01250553e+00 6.21170700e-01 -6.99259713e-02 3.54360700e-01 -1.15030706e+00 -9.51140881e-01 -1.21658170e+00 1.03703916e-01 5.20628154e-01 -6.45021021e-01 1.02408755e+00 -1.08692861e+00 -1.50017285e+00 1.11766207e+00 -2.75627732e-01 2.83950210e-01 1.58212274e-01 -2.10391968e-01 -2.38742307e-01 2.58476973e-01 8.73516276e-02 9.11053538e-01 1.24000514e+00 -1.38976336e+00 6.60373271e-02 -1.12914860e+00 -1.85622856e-01 3.35996151e-01 -7.02292562e-01 -5.53335100e-02 -4.48205948e-01 -6.20779634e-01 2.00729981e-01 -8.44261527e-01 2.62072206e-01 6.21931374e-01 -2.38727391e-01 -2.71390885e-01 9.10921931e-01 -4.69473064e-01 9.26510453e-01 -2.41843915e+00 3.87523204e-01 -1.45543605e-01 4.65259314e-01 1.72836408e-01 -5.07363319e-01 7.74740353e-02 -6.09032691e-01 -1.43957198e-01 -1.32862508e-01 -8.31792355e-01 -7.61145726e-02 4.96207811e-02 -2.03297600e-01 8.46207082e-01 5.08205473e-01 1.01474798e+00 -7.90131271e-01 -4.42891449e-01 -9.33829974e-03 6.69984579e-01 -8.39816809e-01 2.44562298e-01 -6.95710555e-02 7.45164156e-01 -5.34361899e-01 7.86331534e-01 8.73844028e-01 -8.10867101e-02 1.14305548e-01 -7.08155334e-01 2.34539211e-01 -7.75491372e-02 -8.33363950e-01 1.93497491e+00 -4.39048737e-01 3.34884316e-01 2.73547620e-01 -9.63264883e-01 8.65417480e-01 3.68066400e-01 5.54545105e-01 -6.15853131e-01 2.52399176e-01 -5.11380620e-02 -1.96614698e-01 -4.90523517e-01 -2.97662258e-01 -2.80095071e-01 9.53182653e-02 2.60413706e-01 4.68428642e-01 1.89139783e-01 -4.50211823e-01 7.18925744e-02 7.56111443e-01 4.18998271e-01 3.13826025e-01 -3.01716238e-01 6.55335903e-01 -7.33572185e-01 7.07232833e-01 -9.05864686e-02 -3.35169584e-01 5.83610535e-01 6.13921940e-01 -1.36213928e-01 -3.21474940e-01 -1.08690977e+00 -2.65969813e-01 1.21997261e+00 -4.76658307e-02 -4.65094000e-01 -5.65114856e-01 -7.67928660e-01 1.45547658e-01 2.46039346e-01 -8.29335809e-01 -5.15911281e-01 -3.99594903e-01 -7.82995343e-01 7.62998164e-01 5.39163649e-01 5.53421378e-01 -8.03122699e-01 9.77778509e-02 -4.43110108e-01 -4.78128567e-02 -9.66389179e-01 -6.60986304e-01 -3.29923406e-02 -5.95239699e-01 -1.07602596e+00 -5.90637684e-01 -6.51764333e-01 9.77168024e-01 4.26596075e-01 7.88825035e-01 5.65314591e-02 -3.03787529e-01 5.36036611e-01 -1.16872102e-01 -4.08940881e-01 2.59359539e-01 -2.16696233e-01 2.93780327e-01 5.01560986e-01 4.45722878e-01 -8.29295337e-01 -7.34730661e-01 3.27469647e-01 -6.49576485e-01 -7.67857709e-04 6.68234468e-01 9.09219801e-01 3.27739716e-01 -4.54217225e-01 8.78493130e-01 -7.47138202e-01 3.21982771e-01 -6.10415578e-01 -2.74726450e-01 5.15278615e-03 -3.18075329e-01 -3.37754861e-02 3.35871220e-01 -3.91807824e-01 -1.14231300e+00 2.94982225e-01 -1.53663084e-01 -8.27679753e-01 -2.44246975e-01 2.38624215e-01 -8.57615530e-01 -2.59628862e-01 4.50713903e-01 3.61111313e-01 5.49544632e-01 -3.73005301e-01 3.52321953e-01 5.57226002e-01 2.93247968e-01 -7.35387146e-01 9.12290037e-01 6.11330211e-01 9.98134911e-02 -7.70205617e-01 -1.17229843e+00 -2.78246790e-01 -5.55751979e-01 -3.12831402e-01 5.53751886e-01 -1.39628136e+00 -1.06128502e+00 4.95887190e-01 -1.07576990e+00 1.79777592e-02 2.09301505e-02 3.44498605e-01 -6.47419512e-01 3.13349992e-01 -4.31154877e-01 -7.90864408e-01 -2.81878442e-01 -1.02906132e+00 1.51324618e+00 1.50393113e-01 -2.14285716e-01 -6.91141725e-01 -3.63091491e-02 6.95309401e-01 7.74841523e-03 2.29988798e-01 7.00683653e-01 -4.67928797e-01 -3.92169058e-01 -6.55458942e-02 -4.28443015e-01 3.76002997e-01 2.39956230e-01 -3.60294342e-01 -1.42126024e+00 -4.85301584e-01 -1.17045427e-02 -7.97044516e-01 8.16844761e-01 1.08008541e-01 1.48575866e+00 -4.87179279e-01 -2.63848186e-01 9.71219182e-01 9.62946415e-01 -2.56144047e-01 5.42151153e-01 -2.48027593e-01 9.88865077e-01 8.80960703e-01 3.58874023e-01 5.60835302e-01 4.93958890e-01 7.46283531e-01 6.40797555e-01 -2.07882985e-01 -1.64879307e-01 -5.92357814e-01 4.75001007e-01 4.66423690e-01 -1.19615659e-01 9.88291875e-02 -4.55076486e-01 1.63692310e-01 -1.68272722e+00 -7.45939434e-01 2.35743180e-01 1.97989094e+00 8.33952904e-01 -3.71573299e-01 1.06660485e-01 1.38841374e-02 3.29501092e-01 3.64139855e-01 -8.02609086e-01 6.54270872e-02 -6.29412830e-02 2.91557461e-01 -2.91199161e-04 1.81582406e-01 -1.02587759e+00 1.04971862e+00 4.92262220e+00 8.69727433e-01 -1.15338600e+00 2.59442151e-01 7.68036842e-01 -2.43803427e-01 -3.20253789e-01 -3.34058791e-01 -8.04883957e-01 2.40413144e-01 5.23108661e-01 -4.85504344e-02 2.63173133e-01 8.12306106e-01 2.03295946e-01 2.58007675e-01 -1.31502843e+00 1.46513879e+00 4.69745547e-01 -9.38834608e-01 3.15044284e-01 1.18046328e-01 2.52363890e-01 -3.85970980e-01 5.40965438e-01 4.66304481e-01 -3.64994973e-01 -1.23134279e+00 4.63311970e-01 3.44126195e-01 1.19489384e+00 -7.80294240e-01 3.88277143e-01 2.22243488e-01 -1.05743396e+00 -5.63105615e-03 -9.49962884e-02 -6.74617440e-02 -1.91792801e-01 3.12627256e-01 -4.02240187e-01 4.37151045e-01 5.44321597e-01 1.03757513e+00 -4.32667077e-01 3.38075757e-01 -3.22140634e-01 5.26778936e-01 -2.37142425e-02 3.66494417e-01 -1.32563904e-01 -1.15001775e-01 6.00858748e-01 8.45582306e-01 -1.44190211e-02 4.08109307e-01 -5.09864315e-02 9.23073411e-01 -2.60240525e-01 1.55352831e-01 -6.81407213e-01 6.64704964e-02 2.10429862e-01 1.45561004e+00 -1.56284243e-01 2.38078460e-01 -5.45475364e-01 1.02487516e+00 6.16577327e-01 3.46929580e-01 -7.73224294e-01 2.93731511e-01 1.14785945e+00 2.34238639e-01 2.92989407e-02 -1.08876318e-01 -2.74374224e-02 -1.33870888e+00 2.41240766e-02 -9.56257641e-01 3.12188774e-01 -7.10849941e-01 -1.37778616e+00 6.40167713e-01 1.27959087e-01 -1.15207767e+00 2.04950478e-03 -6.05309963e-01 -4.78074372e-01 6.10982656e-01 -1.56513953e+00 -1.64049900e+00 -4.28458035e-01 8.12705398e-01 4.23177630e-01 -3.59534234e-01 1.04933763e+00 5.03237069e-01 -9.97940123e-01 1.14066434e+00 -5.55923581e-01 2.39675060e-01 8.40840459e-01 -7.07549751e-01 -2.71290749e-01 3.79809529e-01 6.91091940e-02 7.03706503e-01 3.93548906e-01 -3.86286169e-01 -1.69098389e+00 -1.11649561e+00 3.54087949e-01 -4.61945057e-01 4.91889358e-01 -6.92587256e-01 -8.29781175e-01 7.32876658e-01 -2.31629431e-01 4.78808492e-01 1.03510296e+00 2.53377020e-01 -8.65011752e-01 -5.00507534e-01 -1.08722568e+00 5.49709141e-01 1.45806932e+00 -7.32184708e-01 -1.62123889e-01 3.14638466e-01 4.67931896e-01 -2.09143698e-01 -6.29061341e-01 7.55975544e-01 7.08321035e-01 -9.20890510e-01 9.72062171e-01 -7.37640917e-01 5.20444810e-01 -1.64787322e-02 -1.86146140e-01 -1.10282946e+00 -1.90070227e-01 -6.24451578e-01 -3.43833745e-01 1.34903967e+00 1.27985850e-01 -7.34959185e-01 9.50081885e-01 3.77867937e-01 8.35769251e-03 -1.16590559e+00 -1.02975690e+00 -5.99293232e-01 -2.49998197e-02 -1.67717889e-01 4.34387207e-01 1.00096548e+00 4.79425900e-02 6.67087615e-01 -5.36881030e-01 2.78134048e-01 8.11330616e-01 -7.93612301e-02 6.82373345e-01 -1.20118642e+00 -1.14883542e-01 -2.75764912e-01 -5.85164487e-01 -9.91132081e-01 8.66254449e-01 -1.06539309e+00 -2.67064273e-01 -8.08675408e-01 6.43256605e-01 -3.15832913e-01 -1.28873602e-01 8.72129619e-01 -2.09230304e-01 4.85002607e-01 1.08166434e-01 3.01317573e-02 -5.77003360e-01 1.20459950e+00 1.46987963e+00 -5.27951587e-03 1.78905174e-01 -8.67921412e-02 -1.05946052e+00 7.85213709e-01 7.26232469e-01 -3.35018605e-01 -6.35351658e-01 -2.94182688e-01 -4.75284532e-02 7.59263113e-02 5.76802254e-01 -5.67840278e-01 2.97949687e-02 1.06379054e-01 6.35359049e-01 -2.74692684e-01 7.37566233e-01 -7.00713694e-01 -1.18539430e-01 2.24305779e-01 -1.18447162e-01 -3.05161923e-01 4.53103095e-01 5.58802545e-01 -2.33565778e-01 2.05762729e-01 8.45422864e-01 5.66129237e-02 -5.30800104e-01 9.15913224e-01 1.18829034e-01 -2.18969155e-02 1.03505552e+00 -1.46318614e-01 -8.46996903e-02 -5.45574486e-01 -6.96394444e-01 1.54275224e-01 5.87230809e-02 5.65533638e-01 7.94857800e-01 -1.40299213e+00 -8.89856517e-01 5.12368500e-01 3.46145302e-01 -6.79674631e-05 4.74773109e-01 8.23970556e-01 1.43593047e-02 3.09875935e-01 -3.13382447e-01 -5.08997619e-01 -1.46626329e+00 4.22922522e-01 3.67158830e-01 2.06397325e-01 -2.48447776e-01 1.13093317e+00 8.78670037e-01 -5.92528462e-01 1.77515715e-01 4.44501549e-01 -3.91197950e-01 2.12808713e-01 5.93676448e-01 2.07772389e-01 -1.32538959e-01 -1.04846931e+00 -5.80168247e-01 7.04770267e-01 -1.55038208e-01 1.93172827e-01 1.49776745e+00 -3.95818613e-02 -1.92057580e-01 1.44865319e-01 1.70371401e+00 -1.13939464e-01 -1.40914679e+00 -2.80544341e-01 -5.17205715e-01 -4.53539610e-01 2.43277952e-01 -4.76292253e-01 -1.52966988e+00 8.28127325e-01 6.92075789e-01 -7.85742402e-01 1.39577866e+00 1.80182323e-01 4.09835130e-01 8.80852863e-02 4.83372122e-01 -4.88866717e-01 5.56127667e-01 2.13890582e-01 1.32323539e+00 -1.51604831e+00 1.37547448e-01 -9.08732295e-01 -5.47458053e-01 1.02777696e+00 1.12496197e+00 4.45761681e-02 8.90856743e-01 2.65721023e-01 -9.50492397e-02 -5.46667576e-01 -7.39363194e-01 -2.31398106e-01 5.53407073e-01 4.58561391e-01 5.82426667e-01 -3.77484672e-02 1.69554185e-02 9.46801543e-01 -1.23359427e-01 -1.52960777e-01 -1.23878166e-01 5.83247244e-01 2.09469590e-02 -9.25678194e-01 -1.53518334e-01 3.21230292e-01 -1.49891272e-01 -8.84080119e-03 -4.86718178e-01 7.32341051e-01 2.22545654e-01 6.49494588e-01 1.07990518e-01 -3.42269510e-01 1.78086847e-01 -8.68217200e-02 8.57798934e-01 -7.97405541e-01 -6.17662109e-02 4.43993956e-02 -1.13230146e-01 -7.94264734e-01 -3.33721012e-01 -6.96568251e-01 -1.21834254e+00 -1.47001997e-01 -2.31412351e-01 -1.88318595e-01 3.34941119e-01 8.57370555e-01 5.08502066e-01 2.42476910e-01 9.72208440e-01 -1.02576149e+00 -5.76684177e-01 -9.08522427e-01 -5.67656994e-01 3.98608536e-01 4.47779834e-01 -1.07397938e+00 -4.24496114e-01 -2.38876969e-01]
[13.246153831481934, 0.6103165149688721]