Datasets:

J0nasW

/

paperswithcode

like 1

58027378-45e6-49cf-81ef-0356437cbb56

impact-of-position-bias-on-language-models-in

2304.13567

https://arxiv.org/abs/2304.13567v2

https://arxiv.org/pdf/2304.13567v2.pdf

Technical Report on Token Position Bias in Transformers

Language Models (LMs) have shown state-of-the-art performance in Natural Language Processing (NLP) tasks. Downstream tasks such as Named Entity Recognition (NER) or Part-of-Speech (POS) tagging are known to suffer from data imbalance issues, specifically in terms of the ratio of positive to negative examples, and class imbalance. In this paper, we investigate an additional specific issue for language models, namely the position bias of positive examples in token classification tasks. Therefore, we conduct an in-depth evaluation of the impact of position bias on the performance of LMs when fine-tuned on Token Classification benchmarks. Our study includes CoNLL03 and OntoNote5.0 for NER, English Tree Bank UD_en and TweeBank for POS tagging. We propose an evaluation approach to investigate position bias in Transformer models. We show that encoders like BERT, ERNIE, ELECTRA, and decoders such as GPT2 and BLOOM can suffer from this bias with an average drop of 3\% and 9\% in their performance. To mitigate this effect, we propose two methods: Random Position Shifting and Context Perturbation, that we apply on batches during the training process. The results show an improvement of $\approx$ 2\% in the performance of the model on CoNLL03, UD_en, and TweeBank.

['Jelena Mitrović', 'Michael Granitzer', 'Mehdi Ben Amor']

2023-04-26

['part-of-speech-tagging', 'named-entity-recognition-ner']

['natural-language-processing', 'natural-language-processing']

[-2.03369051e-01 1.22743413e-01 -2.92889833e-01 -4.34292257e-01 -1.11136639e+00 -5.99596858e-01 3.85150224e-01 5.38776457e-01 -8.65365028e-01 9.40683961e-01 1.62021801e-01 -6.21592224e-01 1.87857509e-01 -7.11099744e-01 -7.34099567e-01 -4.04411942e-01 -9.19130668e-02 4.69447106e-01 1.96800843e-01 -3.87065224e-02 1.36922762e-01 3.56044114e-01 -1.04586339e+00 5.87388456e-01 7.34706879e-01 7.43930578e-01 -5.75419553e-02 5.09041011e-01 -6.07260287e-01 9.88542080e-01 -8.15871716e-01 -7.86100090e-01 3.58948968e-02 4.15123776e-02 -7.90179968e-01 -5.20919442e-01 2.60067910e-01 1.92594156e-01 -4.16414551e-02 9.47284877e-01 1.01851499e+00 -1.65778831e-01 6.76197290e-01 -1.22703850e+00 -2.15298012e-01 1.26452422e+00 -5.27790308e-01 4.55507308e-01 2.37087701e-02 1.25740217e-02 1.15697539e+00 -1.08618283e+00 6.10602856e-01 1.23579431e+00 1.01398528e+00 4.03703243e-01 -1.17760468e+00 -8.94775450e-01 7.00400397e-02 1.17136352e-01 -1.50602996e+00 -6.93252146e-01 2.60955453e-01 -3.06248158e-01 1.40876663e+00 8.68151663e-04 -4.64037843e-02 9.05734777e-01 2.34501615e-01 1.01451349e+00 9.60009813e-01 -7.11714268e-01 1.89797863e-01 3.87477368e-01 3.89357120e-01 3.59739870e-01 1.64537698e-01 -1.24640733e-01 -7.68871903e-01 -2.08053797e-01 1.52256101e-01 -5.96759796e-01 1.05444744e-01 2.83430040e-01 -9.47973669e-01 6.93245113e-01 7.25799426e-02 4.90449607e-01 -3.38529587e-01 7.28101507e-02 7.58263052e-01 2.13604197e-01 7.33109415e-01 5.26329517e-01 -1.16216791e+00 -5.76279044e-01 -1.01452506e+00 2.11105142e-02 1.08282495e+00 9.01745617e-01 5.43863237e-01 -7.27151781e-02 -4.88619834e-01 1.07871425e+00 2.05261797e-01 3.86483938e-01 7.53827393e-01 -3.31362098e-01 1.01415431e+00 4.26436335e-01 -1.78742230e-01 -5.08336425e-01 -4.62015182e-01 -4.69018698e-01 -6.90439105e-01 -3.94065738e-01 3.55337381e-01 -5.79696894e-01 -9.94610310e-01 1.82791901e+00 7.55644590e-02 -8.00566673e-02 2.54793525e-01 2.57018805e-01 7.95962870e-01 6.51813745e-01 6.61056757e-01 -1.18621603e-01 1.56851995e+00 -7.18485355e-01 -6.91693425e-01 -5.09901047e-01 1.16050756e+00 -1.10507250e+00 8.99243236e-01 1.02448843e-01 -1.19251442e+00 -3.59696656e-01 -5.65581918e-01 1.14160944e-02 -4.05808657e-01 3.24952334e-01 3.71716797e-01 8.26566994e-01 -7.97352850e-01 7.67921984e-01 -8.42220008e-01 -3.13628078e-01 3.09414387e-01 3.39445561e-01 -2.60556132e-01 1.47641510e-01 -1.46386039e+00 1.06308651e+00 4.22330707e-01 9.66543274e-04 -2.50384510e-01 -1.12036729e+00 -8.59414637e-01 2.97074139e-01 -7.96917155e-02 -3.99858430e-02 1.38643730e+00 -5.76814532e-01 -1.14153790e+00 9.58047032e-01 -4.56687450e-01 -7.13824630e-01 4.51522201e-01 -2.57067829e-01 -5.73934078e-01 -4.48457003e-01 7.19463080e-02 6.50043488e-01 8.72184038e-02 -7.88858652e-01 -8.23209345e-01 -1.89726442e-01 -3.62739623e-01 1.34728983e-01 -3.54369164e-01 4.42563176e-01 -1.57640532e-01 -5.80659449e-01 -1.37957573e-01 -6.68450236e-01 -2.54208356e-01 -8.41307819e-01 -6.43331766e-01 -4.98340577e-01 2.65823662e-01 -7.39822626e-01 1.56225753e+00 -2.28485370e+00 -6.81745291e-01 1.09058961e-01 -2.10466683e-01 4.79177266e-01 -2.09512025e-01 6.17762327e-01 -2.70240337e-01 5.69276392e-01 1.31866142e-01 -5.35813689e-01 2.00652421e-01 2.54866064e-01 -2.18449950e-01 9.72544700e-02 3.24431062e-01 6.40372634e-01 -8.34995687e-01 -4.91123796e-01 -1.86689168e-01 2.85720378e-01 -4.42197651e-01 9.92545187e-02 1.50101902e-02 -8.18283185e-02 -7.23792836e-02 5.69267452e-01 6.85981452e-01 8.00098777e-02 1.87343553e-01 -5.16660362e-02 -3.76123846e-01 1.25847876e+00 -1.17038810e+00 1.01497078e+00 -7.20101297e-01 6.29128873e-01 -1.65034086e-01 -7.03029335e-01 8.79589081e-01 6.84291899e-01 1.93944275e-01 -7.33135402e-01 -4.42899875e-02 4.24808592e-01 3.01370800e-01 -2.21263379e-01 7.81261861e-01 -3.35030258e-01 -1.99265867e-01 2.12425128e-01 1.68565080e-01 3.16411138e-01 4.69753236e-01 2.74603572e-02 1.33642232e+00 -4.62179720e-01 2.69709408e-01 -2.41850019e-01 2.52372503e-01 -4.21785861e-02 8.71336818e-01 7.19167650e-01 -1.49977192e-01 3.93488407e-01 7.56643116e-01 1.17757758e-02 -1.02472234e+00 -8.79370272e-01 -4.09829766e-01 1.18150365e+00 -6.50349200e-01 -6.09414518e-01 -6.67782426e-01 -8.10891271e-01 7.97891896e-03 1.02624846e+00 -1.67641073e-01 -9.00415108e-02 -6.98797286e-01 -1.30251646e+00 1.17540526e+00 7.01583505e-01 4.87716347e-01 -1.26585519e+00 -1.40962809e-01 4.34692144e-01 -2.72894740e-01 -1.34294236e+00 -2.23706320e-01 8.07557344e-01 -7.82520175e-01 -7.18101919e-01 -5.22630036e-01 -7.26547837e-01 4.59706187e-01 -3.98408532e-01 1.46892786e+00 -3.26474279e-01 1.46100789e-01 -7.94533938e-02 -4.88619894e-01 -7.45437503e-01 -6.66286528e-01 6.57971084e-01 9.83681381e-02 -3.90921354e-01 8.33479345e-01 -3.41105789e-01 -2.49544352e-01 3.08208078e-01 -6.72891617e-01 -3.26321274e-01 8.31783593e-01 7.35422492e-01 4.72859502e-01 1.56481594e-01 7.11113930e-01 -1.19502199e+00 5.74892342e-01 -5.29270172e-01 -3.03725004e-01 1.68885633e-01 -6.97790861e-01 2.00318113e-01 6.69992387e-01 -2.90682226e-01 -9.83530223e-01 -1.05092734e-01 -6.26163363e-01 6.89927191e-02 -2.41354644e-01 6.63009882e-01 -4.14654344e-01 2.35510811e-01 5.90604126e-01 -4.30069230e-02 -7.31860876e-01 -7.86623657e-01 4.63806912e-02 1.02279794e+00 1.69397458e-01 -4.23510641e-01 4.96217430e-01 -7.13104308e-02 -4.50520605e-01 -7.88116515e-01 -8.72774720e-01 -6.91609263e-01 -4.40467656e-01 2.73920596e-01 4.59322810e-01 -1.11220646e+00 -3.60922575e-01 6.07752800e-01 -1.27848864e+00 -4.73973185e-01 -3.78781289e-01 5.53508043e-01 -3.50340232e-02 1.96199909e-01 -9.40495610e-01 -7.78050542e-01 -6.35889709e-01 -8.46213043e-01 8.39802861e-01 2.77587492e-02 -4.53303993e-01 -9.60928917e-01 4.34444994e-02 2.47920349e-01 3.22274536e-01 -3.19164813e-01 1.04907823e+00 -1.33861375e+00 2.20434815e-02 -1.59311846e-01 -1.78568900e-01 6.25260830e-01 -1.63554922e-01 -1.52993456e-01 -1.09920669e+00 -2.47318327e-01 -3.83360356e-01 -9.23190787e-02 7.46618330e-01 2.19683260e-01 8.99178386e-01 -3.49402308e-01 -3.87327433e-01 1.90073684e-01 1.34538352e+00 1.43502355e-01 7.89589524e-01 3.76206905e-01 4.70803022e-01 4.31677490e-01 6.46707356e-01 4.90108460e-01 4.36611861e-01 4.42785174e-01 -1.47729442e-02 -2.56770998e-02 -5.62889203e-02 -3.65108877e-01 6.87231481e-01 9.71450686e-01 2.81771719e-01 -5.00451148e-01 -1.18615794e+00 8.79602671e-01 -1.33076882e+00 -7.53818989e-01 -2.19403133e-01 2.30051446e+00 1.30917621e+00 5.15741885e-01 -1.48837730e-01 3.97501260e-01 7.89115489e-01 -2.17164420e-02 -9.26823691e-02 -7.81346560e-01 -9.52717960e-02 5.36689579e-01 1.02065253e+00 3.00824523e-01 -8.98436248e-01 1.16267633e+00 5.84098434e+00 1.15664113e+00 -1.18897402e+00 1.51589140e-01 8.44816804e-01 1.10796705e-01 -1.83773506e-02 -3.26937921e-02 -1.60618496e+00 6.93286180e-01 1.51870298e+00 5.31782024e-03 -1.44774526e-01 6.99667633e-01 3.13439041e-01 -1.23610437e-01 -1.16254640e+00 6.60288513e-01 -2.77953237e-01 -9.54845786e-01 -2.04142064e-01 5.80528006e-02 4.84450340e-01 5.78488767e-01 -2.06635863e-01 7.84224093e-01 4.39260840e-01 -8.91790986e-01 7.38630712e-01 -5.22901770e-03 6.20582521e-01 -8.65038037e-01 1.13845551e+00 4.26411599e-01 -1.04045081e+00 1.81578442e-01 -4.80254620e-01 5.31581677e-02 3.20959657e-01 1.23441482e+00 -1.30733216e+00 3.91934037e-01 6.37509346e-01 2.32981518e-01 -3.89392644e-01 1.11347830e+00 -3.50567788e-01 1.25005627e+00 -4.12560344e-01 -1.05810389e-01 1.57275438e-01 5.48453093e-01 3.40780258e-01 1.73192227e+00 2.84099042e-01 -1.07617147e-01 -1.91486582e-01 4.17081028e-01 -4.27174479e-01 3.58828366e-01 -1.39937967e-01 1.94976181e-02 8.84057164e-01 1.17335415e+00 -4.58633631e-01 -3.80235374e-01 -3.58210385e-01 5.24023354e-01 3.92934322e-01 7.27842450e-02 -7.59959340e-01 -5.16241252e-01 6.00608647e-01 2.70790547e-01 5.01695931e-01 5.06111681e-02 -2.83038646e-01 -9.23728108e-01 8.77525136e-02 -8.77585590e-01 4.60679591e-01 -3.45005453e-01 -1.39562309e+00 5.32087684e-01 -1.86592013e-01 -8.91638458e-01 -1.84156030e-01 -6.98423266e-01 -5.88576972e-01 9.57380891e-01 -1.84586346e+00 -6.74850464e-01 3.77242863e-01 1.36674196e-01 3.26566935e-01 1.21244108e-02 6.85486794e-01 8.49555790e-01 -6.65887833e-01 1.07663393e+00 2.43117228e-01 6.97284937e-01 9.55109298e-01 -1.42462480e+00 7.60128319e-01 6.64790571e-01 1.92597181e-01 4.53330159e-01 6.06706679e-01 -5.07654369e-01 -7.76648700e-01 -1.34072161e+00 2.03182292e+00 -3.56804669e-01 6.12459898e-01 -3.89443070e-01 -7.84307659e-01 7.58141398e-01 8.63794312e-02 -1.21710829e-01 8.86126995e-01 4.86405641e-01 -2.00625181e-01 -3.12016197e-02 -1.24387777e+00 4.02944237e-01 8.04357052e-01 -3.94175738e-01 -5.48137605e-01 1.87508225e-01 5.52983761e-01 -4.79888916e-01 -1.06255054e+00 3.85932118e-01 3.38228285e-01 -8.17002952e-01 7.68460512e-01 -6.05555177e-01 3.86125594e-01 1.77934721e-01 -5.76556921e-02 -1.53663397e+00 -1.84506521e-01 -4.18799609e-01 2.91414648e-01 1.98629963e+00 9.30524528e-01 -7.14022458e-01 7.58510411e-01 4.31309819e-01 -1.50267214e-01 -6.83865607e-01 -9.96633947e-01 -8.39241087e-01 4.83297586e-01 -7.29940116e-01 4.28091168e-01 8.74920011e-01 2.00807989e-01 4.64016318e-01 6.07810766e-02 1.03845209e-01 1.49035335e-01 -5.01048148e-01 4.02965546e-01 -9.81036246e-01 -7.09123015e-02 -2.47976825e-01 -3.83689702e-01 -9.21836317e-01 2.20538482e-01 -9.34157014e-01 1.93103939e-01 -1.24299252e+00 8.90744030e-02 -8.20308805e-01 -4.35415179e-01 8.31920683e-01 -1.89567730e-01 -9.27067176e-03 2.57394433e-01 -3.79413702e-02 -4.32066917e-01 1.76984489e-01 5.85305095e-01 5.37206903e-02 -6.57365322e-02 3.65717933e-02 -4.93737549e-01 6.82992876e-01 9.38702524e-01 -9.48612988e-01 2.59542286e-01 -6.07692659e-01 5.44061601e-01 -2.56636083e-01 -2.02309415e-02 -9.34904337e-01 9.45468917e-02 3.48843098e-01 2.77909666e-01 -6.84896111e-01 -6.37687556e-03 -5.41265666e-01 -3.78150254e-01 4.30645943e-01 -4.32042152e-01 4.05041933e-01 4.43390876e-01 -3.67754772e-02 -3.99662226e-01 -5.22629976e-01 6.35757029e-01 -1.95285976e-01 -4.78761435e-01 2.63923071e-02 -6.48371816e-01 5.53568065e-01 5.22129536e-01 2.12972015e-01 -2.78803974e-01 1.74356508e-03 -4.48166609e-01 3.66101623e-01 -9.04591754e-03 2.26783723e-01 5.05823232e-02 -1.04836905e+00 -8.38161170e-01 9.31337029e-02 -8.77021812e-03 -1.08597443e-01 1.52841076e-01 9.39643919e-01 -2.51672179e-01 6.61295772e-01 2.45325193e-01 -2.24284619e-01 -1.31208766e+00 7.51341134e-02 3.76250595e-01 -1.05716646e+00 5.68321757e-02 1.10260439e+00 -1.95679173e-01 -7.91191518e-01 2.19958022e-01 -6.53551996e-01 -1.87535912e-01 4.28951234e-01 2.55442381e-01 4.56904143e-01 6.28345251e-01 -4.44141597e-01 -6.28697991e-01 7.54418373e-02 -2.04384878e-01 -9.53363031e-02 1.07117724e+00 1.03192173e-01 -1.48604736e-01 4.27196831e-01 1.16901731e+00 3.40387106e-01 -5.33211827e-01 -2.60280758e-01 5.08591712e-01 -7.98647408e-04 -3.30543816e-02 -1.02424741e+00 -8.92494977e-01 9.65713143e-01 4.00261134e-01 1.73016146e-01 7.80794322e-01 -1.42484143e-01 1.06022882e+00 2.67184258e-01 1.61984041e-01 -1.30802381e+00 -5.36565840e-01 1.00660467e+00 2.42836893e-01 -1.07237220e+00 -3.69206309e-01 -3.97304356e-01 -4.89744991e-01 7.52664268e-01 4.62330043e-01 2.19868094e-01 6.51104152e-01 7.08293498e-01 2.91857839e-01 7.10135028e-02 -1.06906378e+00 -6.84177652e-02 -7.55191296e-02 2.08765298e-01 1.13893998e+00 1.35400593e-01 -6.25446260e-01 8.55856299e-01 -4.46233183e-01 -7.57011995e-02 3.66063416e-01 9.55305755e-01 -1.97820768e-01 -1.38424122e+00 -3.76593560e-01 5.25600195e-01 -1.23569965e+00 -8.21157634e-01 -1.46352366e-01 7.02327549e-01 2.83578068e-01 9.14393067e-01 2.28835836e-01 -1.20936014e-01 5.68885386e-01 5.63105464e-01 1.32153869e-01 -8.08646917e-01 -1.04990244e+00 -2.27794155e-01 5.98823726e-01 -1.50525287e-01 -1.75605968e-01 -8.03314924e-01 -1.38911080e+00 -3.42661649e-01 -5.16922295e-01 5.39596140e-01 7.65137613e-01 9.34025347e-01 4.01236564e-01 5.10359824e-01 4.04014468e-01 -3.16966474e-01 -8.45416665e-01 -1.28855991e+00 -6.46763802e-01 1.23323761e-01 -1.12030312e-01 -2.61514664e-01 -6.20876610e-01 -1.02534622e-01]

[9.867215156555176, 9.532880783081055]

0be85a35-8851-456f-b55c-82f989e4e915

toward-more-accurate-and-generalizable-brain

2306.05255

https://arxiv.org/abs/2306.05255v1

https://arxiv.org/pdf/2306.05255v1.pdf

Toward more accurate and generalizable brain deformation estimators for traumatic brain injury detection with unsupervised domain adaptation

Machine learning head models (MLHMs) are developed to estimate brain deformation for early detection of traumatic brain injury (TBI). However, the overfitting to simulated impacts and the lack of generalizability caused by distributional shift of different head impact datasets hinders the broad clinical applications of current MLHMs. We propose brain deformation estimators that integrates unsupervised domain adaptation with a deep neural network to predict whole-brain maximum principal strain (MPS) and MPS rate (MPSR). With 12,780 simulated head impacts, we performed unsupervised domain adaptation on on-field head impacts from 302 college football (CF) impacts and 457 mixed martial arts (MMA) impacts using domain regularized component analysis (DRCA) and cycle-GAN-based methods. The new model improved the MPS/MPSR estimation accuracy, with the DRCA method significantly outperforming other domain adaptation methods in prediction accuracy (p<0.001): MPS RMSE: 0.027 (CF) and 0.037 (MMA); MPSR RMSE: 7.159 (CF) and 13.022 (MMA). On another two hold-out test sets with 195 college football impacts and 260 boxing impacts, the DRCA model significantly outperformed the baseline model without domain adaptation in MPS and MPSR estimation accuracy (p<0.001). The DRCA domain adaptation reduces the MPS/MPSR estimation error to be well below TBI thresholds, enabling accurate brain deformation estimation to detect TBI in future clinical applications.

['David B. Camarillo', 'Michael M. Zeineh', 'Olivier Gevaert', 'Enora Le Flao', 'Nicholas J. Cecchi', 'Yuzhe Liu', 'Jiawei Sun', 'Xianghao Zhan']

2023-06-08

['unsupervised-domain-adaptation']

['methodology']

[-1.51358217e-01 4.27107327e-02 -1.01776801e-01 -1.22855254e-01 -1.15203154e+00 -1.38404146e-01 2.15156674e-01 -8.57021064e-02 -5.78637242e-01 7.64378130e-01 7.26410687e-01 7.67566590e-03 -2.42471486e-01 -6.31608665e-01 -7.80997634e-01 -8.00160170e-01 -3.34384650e-01 8.33424985e-01 3.50813895e-01 -1.89302951e-01 2.45500758e-01 6.56277120e-01 -1.11134601e+00 2.48222664e-01 7.74007916e-01 7.15255976e-01 6.54367357e-02 4.09655869e-01 6.17676735e-01 6.23080611e-01 -3.78323317e-01 -1.70864150e-01 1.63719952e-01 -2.12628350e-01 -7.75479198e-01 -5.09617031e-01 2.76910186e-01 -6.38756514e-01 -6.60438776e-01 4.18326914e-01 1.07171631e+00 5.20518363e-01 1.22483015e+00 -1.09083903e+00 -4.32933509e-01 3.57076585e-01 -8.05797637e-01 4.54674810e-01 3.01196635e-01 4.72125828e-01 2.48895317e-01 -8.43404114e-01 6.54340565e-01 9.25508797e-01 1.05189717e+00 8.16329122e-01 -9.46278214e-01 -1.03202713e+00 -3.15356046e-01 5.48519671e-01 -1.13440549e+00 -2.14822918e-01 5.45883060e-01 -9.83296514e-01 1.16151655e+00 -2.01153886e-02 8.16268682e-01 1.19355571e+00 8.49389434e-01 3.74755800e-01 1.21585524e+00 1.14936881e-01 4.58959192e-01 -8.12725723e-01 1.77362218e-01 1.74799383e-01 4.40183997e-01 3.07668567e-01 -6.43496454e-01 -9.78806391e-02 6.59422398e-01 -1.63759395e-01 -2.18327135e-01 4.07038569e-01 -1.00144315e+00 7.73888469e-01 4.28095251e-01 -2.17033662e-02 -6.90783203e-01 1.31082341e-01 5.54093122e-01 -1.75921962e-01 5.39635956e-01 1.07751466e-01 -2.08538920e-01 -2.38249213e-01 -1.20922267e+00 6.13563478e-01 1.93609685e-01 5.50171316e-01 9.74956453e-02 3.57239902e-01 -3.77290487e-01 1.04416859e+00 -1.33811370e-01 9.09957528e-01 6.49579525e-01 -9.42330599e-01 4.68408257e-01 2.49085739e-01 -4.31276947e-01 -7.04835951e-01 -1.01284087e+00 -3.81837428e-01 -8.21801484e-01 2.49363735e-01 5.27096868e-01 -4.08861101e-01 -1.06682456e+00 1.67684209e+00 2.06457540e-01 6.36894405e-02 -4.93291080e-01 1.08409572e+00 8.04255784e-01 1.46664977e-01 4.25874442e-01 -8.91151428e-02 1.23402834e+00 -3.66256028e-01 -2.66210854e-01 -4.37119454e-01 7.36947119e-01 -4.13165271e-01 9.80452836e-01 3.10624331e-01 -1.13195026e+00 -8.73914659e-02 -5.37178576e-01 1.40787572e-01 2.61280328e-01 -1.40168473e-01 4.79199439e-02 4.51517344e-01 -7.77304471e-01 4.35334951e-01 -1.08745301e+00 -1.81694061e-01 7.81272590e-01 5.32468319e-01 -5.94464421e-01 -2.93020815e-01 -1.08967018e+00 1.20357120e+00 1.45785630e-01 -1.23866417e-01 -1.02330959e+00 -1.47735941e+00 -5.84981263e-01 -2.67484337e-01 -3.54996443e-01 -8.11970711e-01 9.10577297e-01 -1.76761895e-01 -1.34441793e+00 9.50899661e-01 -2.64676958e-02 -6.34153664e-01 6.98998034e-01 -5.08007944e-01 -2.19722778e-01 4.80542183e-01 2.90008456e-01 4.37715769e-01 4.81973112e-01 -9.34214354e-01 5.00646234e-02 -7.99747288e-01 -7.20965683e-01 1.01882592e-01 6.42909184e-02 5.07683516e-01 3.53573233e-01 -7.73241520e-01 1.20158069e-01 -1.12177813e+00 6.38506934e-02 -3.37600470e-01 -2.35774741e-01 2.16753006e-01 4.93151277e-01 -1.49242091e+00 8.84798586e-01 -1.64487922e+00 -4.59301732e-02 1.29406691e-01 3.36305708e-01 -7.56389722e-02 2.36160859e-01 9.14885551e-02 -3.54291171e-01 -4.42372352e-01 -7.93065965e-01 -3.32953185e-02 -2.60698646e-01 -8.15726370e-02 -4.44095358e-02 9.94039774e-01 -1.12619333e-01 1.06716645e+00 -6.09476030e-01 -4.36900973e-01 -5.70485368e-02 4.27336961e-01 -9.08512175e-01 1.62286505e-01 4.95600492e-01 9.47290242e-01 -1.03869058e-01 6.20341420e-01 8.73350084e-01 6.04186833e-01 -4.64791566e-01 -9.68022197e-02 9.55307037e-02 -1.27666835e-02 -7.01188922e-01 1.36649537e+00 -1.44601077e-01 5.74295461e-01 3.72739211e-02 -9.77720618e-01 9.34601307e-01 3.17675561e-01 1.12957108e+00 -6.93304837e-01 2.18403772e-01 2.77754158e-01 1.73422948e-01 -6.20293617e-01 -6.59887446e-03 -1.13283753e+00 -8.86496007e-02 4.19842660e-01 2.60651596e-02 -2.38663986e-01 -2.39840075e-01 -1.18736923e-01 1.31529796e+00 -1.14127092e-01 -1.63572863e-01 -4.76583809e-01 7.60585070e-02 6.62962571e-02 7.39099622e-01 2.61831135e-01 -4.42517370e-01 1.14178562e+00 3.09933752e-01 -1.96436226e-01 -1.02835500e+00 -1.36188948e+00 -3.64991307e-01 1.16316235e+00 -4.02381599e-01 2.22826913e-01 -1.11977780e+00 -2.28699699e-01 2.85555661e-01 1.03559518e+00 -6.59494400e-01 -7.01389968e-01 -9.76002216e-01 -1.19207060e+00 9.39182460e-01 1.21296418e+00 3.92399818e-01 -1.12775052e+00 -5.41400433e-01 5.99729642e-02 -4.32110101e-01 -1.03834534e+00 -4.33657557e-01 -1.09070137e-01 -1.00430465e+00 -9.06854749e-01 -1.24098194e+00 -4.62041646e-01 4.25735325e-01 -4.31979716e-01 6.70822859e-01 -2.04473123e-01 -1.17034353e-01 5.78226626e-01 -2.04449221e-01 -5.43534040e-01 -3.32541704e-01 -1.25237763e-01 6.26418412e-01 -4.56530482e-01 2.61256188e-01 -7.42722094e-01 -8.66700590e-01 2.32945293e-01 -7.32374966e-01 -2.24615574e-01 5.44241548e-01 7.71287024e-01 4.33043897e-01 -5.49757719e-01 9.87580299e-01 -4.68153983e-01 6.72041833e-01 -7.38203287e-01 1.01530150e-01 -1.75041959e-01 -7.18690336e-01 -2.82795072e-01 2.15163797e-01 -6.22193217e-01 -1.27715886e+00 -4.43821937e-01 -3.80148649e-01 -4.38238055e-01 -4.15731281e-01 4.82852042e-01 8.73297080e-02 8.50191414e-02 1.05323374e+00 6.41531348e-02 1.18525118e-01 -5.23903906e-01 -1.77298293e-01 3.95152032e-01 1.18242753e+00 -9.69942391e-01 4.92794901e-01 2.36167222e-01 2.48547003e-01 -7.55191386e-01 -1.89043835e-01 -3.49195182e-01 -9.40488994e-01 -6.10296249e-01 1.09914160e+00 -7.99644589e-01 -5.88018656e-01 8.45878661e-01 -8.43701363e-01 -8.94164383e-01 -1.60573706e-01 1.08180070e+00 -7.62645543e-01 3.21222067e-01 -8.49296033e-01 -4.88571644e-01 -9.02234972e-01 -1.07134414e+00 9.80051458e-01 -2.05147490e-01 -5.16452491e-01 -7.65711248e-01 3.75025809e-01 7.60590196e-01 5.18782914e-01 8.81083548e-01 1.08444285e+00 -6.26220524e-01 4.34679449e-01 -5.35339236e-01 -1.29748937e-02 2.11329222e-01 -3.10962349e-01 -3.61271799e-01 -8.71282399e-01 -4.51039523e-01 1.15305558e-01 -6.33421838e-02 7.04568267e-01 1.00871634e+00 9.25955534e-01 -8.90081972e-02 -1.37217641e-01 5.69223166e-01 1.04951382e+00 3.00080061e-01 8.87117684e-01 5.94725370e-01 6.79810524e-01 4.53727007e-01 8.83671120e-02 4.77587789e-01 3.51396799e-01 5.17674387e-01 1.51962861e-01 1.63360417e-01 -5.51481187e-01 -5.71185686e-02 6.44002616e-01 8.86127651e-01 -7.10022569e-01 3.30295593e-01 -1.62166345e+00 7.78891146e-01 -1.17421782e+00 -9.98672247e-01 -5.11986256e-01 2.32707953e+00 6.18773341e-01 1.10994406e-01 4.44491535e-01 5.85819073e-02 8.92680526e-01 -4.48427230e-01 -7.80087352e-01 -3.14156681e-01 -2.36040745e-02 7.51505077e-01 5.57913959e-01 6.46558627e-02 -5.42898238e-01 5.18508196e-01 6.41821194e+00 5.01745343e-01 -1.07476783e+00 5.70495188e-01 4.90230173e-01 -4.68337685e-01 7.09616989e-02 -2.44948998e-01 -3.79773140e-01 6.13735139e-01 1.29069304e+00 -1.73696175e-01 3.20914835e-01 5.75484931e-01 2.33530298e-01 -3.09213072e-01 -6.77381158e-01 7.65488029e-01 8.97447839e-02 -9.10738707e-01 -4.66056496e-01 7.86989704e-02 6.53976679e-01 5.40942788e-01 -1.52981922e-01 4.85305339e-01 5.88254025e-03 -1.06185913e+00 8.03887844e-01 7.65406311e-01 1.30652499e+00 -8.36398363e-01 8.04425657e-01 3.58649880e-01 -8.81248653e-01 -1.82324961e-01 -2.16626555e-01 -1.52947605e-01 3.82515460e-01 6.12225533e-01 -7.93043613e-01 3.29295993e-01 9.93865490e-01 2.48662755e-01 -2.71100491e-01 9.66806352e-01 -1.19769260e-01 1.08610952e+00 -5.22465348e-01 5.74782670e-01 -2.01204181e-01 -8.50510969e-06 7.87390709e-01 1.32208073e+00 3.02747995e-01 4.72161770e-01 -3.96178335e-01 8.15916598e-01 1.57489166e-01 4.18965034e-02 -2.47618809e-01 6.32298648e-01 2.78645992e-01 8.26365054e-01 -6.10080779e-01 -4.86740731e-02 -1.06696270e-01 3.96463633e-01 1.09465003e-01 2.73027390e-01 -9.73704040e-01 -2.74931431e-01 5.81105053e-01 8.23650420e-01 -4.45787698e-01 -2.27113470e-01 -8.74912441e-01 -9.50131059e-01 -1.10655554e-01 -5.66522896e-01 5.67361116e-01 -8.38295400e-01 -1.38673830e+00 2.88151503e-01 6.46372020e-01 -1.10886061e+00 -2.81632364e-01 -5.10980189e-01 -9.78646994e-01 8.61839652e-01 -8.44613194e-01 -1.04602110e+00 -3.97491843e-01 6.03209853e-01 1.69304445e-01 -1.32131398e-01 3.93196136e-01 3.27920914e-01 -6.88448489e-01 7.60802329e-01 -1.79617442e-02 3.19420964e-01 9.59497094e-01 -7.81813383e-01 2.17481777e-01 7.53346980e-01 -9.24365401e-01 5.08768260e-01 6.67884767e-01 -1.09571898e+00 -1.12986803e+00 -1.08610380e+00 4.52205598e-01 -4.92969036e-01 4.26827371e-01 1.32083789e-01 -1.17402184e+00 7.19407976e-01 -2.26503640e-01 -1.44755512e-01 6.70893192e-01 -3.80605012e-01 -9.73630399e-02 1.64585754e-01 -1.52587378e+00 3.25925738e-01 1.06414974e+00 -4.13836271e-01 -9.41148341e-01 3.62764418e-01 2.12995276e-01 -5.76451659e-01 -1.43394578e+00 8.64611804e-01 8.10000539e-01 -6.87521219e-01 1.10311759e+00 -7.25112200e-01 5.82317829e-01 1.80300668e-01 5.20954542e-02 -1.32658041e+00 -4.35077399e-01 5.50950058e-02 -1.42940745e-01 8.87551904e-01 -1.50499925e-01 -8.47734451e-01 7.15124249e-01 1.25733066e+00 -6.99365795e-01 -7.87765801e-01 -1.45273840e+00 -8.90792608e-01 1.24070036e+00 -9.04732525e-01 4.26797926e-01 7.79373646e-01 1.17323555e-01 -1.84028208e-01 9.33572724e-02 4.20721620e-02 7.64995635e-01 -4.47143465e-01 3.16806316e-01 -1.28220582e+00 3.40121575e-02 -6.05047584e-01 -6.34729326e-01 -1.96432412e-01 2.97495723e-01 -1.14519739e+00 4.89625186e-02 -1.76570058e+00 4.84084636e-01 -1.42659411e-01 -1.75637007e-01 6.33677125e-01 -4.70458604e-02 1.34756535e-01 1.13219127e-01 2.83336431e-01 4.55136240e-01 5.33308923e-01 1.08145368e+00 -4.64053862e-02 -3.61259542e-02 -8.96458849e-02 -3.90362829e-01 6.71748102e-01 9.06102240e-01 -7.87990928e-01 2.73830388e-02 -3.89657378e-01 -3.25280607e-01 2.54523247e-01 5.61488926e-01 -1.47452736e+00 2.00640574e-01 -2.22587705e-01 8.30529571e-01 -5.39563358e-01 7.66375288e-02 -9.98946428e-02 3.68801534e-01 7.38585830e-01 -9.82014388e-02 1.26300067e-01 4.92172092e-01 1.46157384e-01 2.51592219e-01 2.15865616e-02 1.06931615e+00 2.60068327e-01 -1.36185393e-01 4.14647907e-01 -7.79223740e-01 4.73925263e-01 8.07577431e-01 -4.21228766e-01 -2.07815990e-01 -3.46863687e-01 -8.46973538e-01 -1.97250828e-01 3.84321362e-01 1.47085160e-01 7.88054347e-01 -1.34861875e+00 -9.54308212e-01 1.02691501e-01 -7.80357048e-02 1.97700262e-01 7.06940234e-01 1.59349108e+00 -6.56817555e-01 2.77636707e-01 -5.82875252e-01 -3.94877106e-01 -7.44612753e-01 1.41979635e-01 3.09261113e-01 -1.21284984e-01 -9.24916506e-01 8.03937912e-01 2.67365634e-01 -4.25680578e-01 -3.19703430e-01 -3.57244253e-01 9.85656604e-02 -1.84867889e-01 3.48742664e-01 1.21637583e+00 4.48332191e-01 -1.24136794e+00 -5.91750026e-01 6.89971030e-01 3.80457491e-01 -1.59951419e-01 1.69065928e+00 2.94468701e-01 2.26598009e-02 3.27076800e-02 1.17216074e+00 -1.86143681e-01 -1.27886260e+00 2.26406500e-01 -1.42523870e-01 1.11695424e-01 2.47933283e-01 -9.27589834e-01 -1.28838086e+00 8.81700516e-01 8.49384129e-01 -9.12499428e-01 1.21544325e+00 1.21861845e-01 1.30435002e+00 -2.28543412e-02 4.88205671e-01 -1.17103112e+00 -1.43851876e-01 3.98361832e-01 1.20545733e+00 -5.82086325e-01 8.95698220e-02 1.69507340e-01 -1.04218054e+00 1.09525132e+00 6.60957754e-01 -3.86741519e-01 7.69659340e-01 4.80091602e-01 -2.80669451e-01 -4.17756975e-01 -2.29660615e-01 3.31532508e-01 5.47900558e-01 8.31722379e-01 3.21395755e-01 3.85158569e-01 -5.66353440e-01 1.21011627e+00 -6.27736032e-01 8.10823441e-02 3.41744334e-01 8.61740649e-01 -4.25773412e-01 -4.29226160e-01 -8.64133894e-01 9.11083817e-01 -4.32853371e-01 1.31219879e-01 -1.55341446e-01 7.43733048e-01 1.23511493e-01 5.71385443e-01 3.74125838e-02 -4.84031260e-01 5.90631366e-01 4.32321608e-01 5.53370535e-01 -4.20701146e-01 -7.03838170e-01 -1.29247000e-02 -2.38201126e-01 -3.32103789e-01 -2.53898539e-02 -1.06269681e+00 -1.81803501e+00 -4.03486013e-01 -2.81614602e-01 -3.62865180e-01 3.84598643e-01 1.14273643e+00 2.90476084e-01 3.92022043e-01 2.13292822e-01 -1.04345334e+00 -3.44368339e-01 -1.37405670e+00 -7.96154499e-01 6.16621256e-01 -6.55677170e-02 -1.28334463e+00 -2.23684892e-01 1.77064836e-01]

[14.077093124389648, -1.8916213512420654]

def87520-38a0-4dc4-84e5-ff5cd8612ba9

screenplay-summarization-using-latent

2004.12727

https://arxiv.org/abs/2004.12727v1

https://arxiv.org/pdf/2004.12727v1.pdf

Screenplay Summarization Using Latent Narrative Structure

Most general-purpose extractive summarization models are trained on news articles, which are short and present all important information upfront. As a result, such models are biased on position and often perform a smart selection of sentences from the beginning of the document. When summarizing long narratives, which have complex structure and present information piecemeal, simple position heuristics are not sufficient. In this paper, we propose to explicitly incorporate the underlying structure of narratives into general unsupervised and supervised extractive summarization models. We formalize narrative structure in terms of key narrative events (turning points) and treat it as latent in order to summarize screenplays (i.e., extract an optimal sequence of scenes). Experimental results on the CSI corpus of TV screenplays, which we augment with scene-level summarization labels, show that latent turning points correlate with important aspects of a CSI episode and improve summarization performance over general extractive algorithms leading to more complete and diverse summaries.

['Lea Frermann', 'Frank Keller', 'Pinelopi Papalampidi', 'Mirella Lapata']

2020-04-27

screenplay-summarization-using-latent-1

https://aclanthology.org/2020.acl-main.174

https://aclanthology.org/2020.acl-main.174.pdf

acl-2020-6

['turning-point-identification', 'extractive-document-summarization']

['natural-language-processing', 'natural-language-processing']

[ 6.11650586e-01 3.85260105e-01 -5.96235514e-01 -2.24453852e-01 -1.14955819e+00 -1.03995633e+00 9.94502962e-01 7.16065526e-01 -2.39241421e-01 7.68221080e-01 1.48638391e+00 1.06676966e-01 -2.47396082e-02 -5.36543012e-01 -5.59245229e-01 -1.98468611e-01 9.73289087e-02 4.03923035e-01 1.08808791e-02 -1.39956221e-01 7.05191731e-01 1.58515051e-01 -1.21595097e+00 9.72796381e-01 8.34349275e-01 2.42913619e-01 3.10254395e-01 1.14053917e+00 -3.87663156e-01 1.25486040e+00 -1.20512795e+00 -3.11751962e-01 -3.63264441e-01 -1.09191453e+00 -1.01741076e+00 4.15685147e-01 5.30610263e-01 -4.81382519e-01 -3.86047512e-01 7.14877248e-01 3.29604000e-01 2.45762452e-01 9.98593986e-01 -7.04608381e-01 2.98401341e-02 1.29491055e+00 -5.73309779e-01 4.47456151e-01 7.38527417e-01 -2.16686517e-01 1.53789556e+00 -5.79887629e-01 1.25455892e+00 9.82147813e-01 5.50207198e-01 3.22898567e-01 -1.15347922e+00 1.17961215e-02 3.73334229e-01 1.68259248e-01 -8.28646302e-01 -8.65739465e-01 1.11452317e+00 -3.37531656e-01 1.16258323e+00 7.60420024e-01 8.32042694e-01 1.27473044e+00 3.58199537e-01 1.44148409e+00 4.73684102e-01 -4.60784167e-01 3.21131647e-01 -1.66631401e-01 2.83035070e-01 3.35322917e-01 3.00683886e-01 -8.86793792e-01 -9.05799448e-01 1.84167828e-03 2.82624960e-01 -1.47794619e-01 -1.28062204e-01 8.20911154e-02 -1.31795454e+00 7.87777483e-01 -1.12776056e-01 3.04089218e-01 -6.55700266e-01 1.80570390e-02 9.18841839e-01 -3.80804427e-02 6.09622478e-01 9.08552885e-01 4.60493704e-03 -5.68999827e-01 -1.58222926e+00 6.23293996e-01 1.09131050e+00 1.06740618e+00 4.18925226e-01 -1.05268128e-01 -7.18833089e-01 7.41035163e-01 -4.87020761e-01 1.30538270e-01 2.37696603e-01 -1.16586709e+00 8.23264241e-01 5.45324743e-01 -5.89307211e-03 -1.12860203e+00 -3.77085328e-01 -5.63145101e-01 -6.89310551e-01 -4.62683380e-01 -2.81781167e-01 -2.47119322e-01 -6.31083727e-01 1.28936577e+00 -1.75662428e-01 -2.47320190e-01 2.49893054e-01 4.67305660e-01 1.22658169e+00 1.12458694e+00 -1.46662444e-01 -7.26939440e-01 1.41341078e+00 -1.12518346e+00 -1.13236582e+00 -4.74969506e-01 7.22350001e-01 -7.10251033e-01 1.07942438e+00 5.43475986e-01 -1.39651072e+00 -2.55719841e-01 -1.34739316e+00 -3.12265605e-01 1.03241690e-01 2.85475671e-01 3.74418020e-01 4.83356090e-03 -7.28198588e-01 6.90280199e-01 -7.95152068e-01 -4.73141730e-01 4.32988137e-01 -3.20142329e-01 -4.17849720e-02 1.29986763e-01 -6.93050921e-01 5.61196268e-01 6.40247285e-01 -3.21426243e-01 -7.65371799e-01 -6.23072743e-01 -8.04503143e-01 2.09499553e-01 8.92937005e-01 -7.44206309e-01 1.62178099e+00 -6.92322850e-01 -1.47785306e+00 6.68413579e-01 -4.80279505e-01 -4.66587454e-01 3.69655758e-01 -6.74644768e-01 -7.28843436e-02 7.53246248e-01 3.79215121e-01 4.33792263e-01 6.32838011e-01 -1.34458959e+00 -9.16550279e-01 1.63248017e-01 2.49799475e-01 6.10479832e-01 -3.70268613e-01 2.41662830e-01 -4.89281952e-01 -8.05240214e-01 5.02461866e-02 -6.05960131e-01 -2.42717296e-01 -1.01800597e+00 -1.16123104e+00 -3.22258174e-02 6.04315937e-01 -1.02827847e+00 1.89055061e+00 -1.82968795e+00 6.18916512e-01 -1.57758296e-01 4.60229129e-01 -2.90823251e-01 2.24171262e-02 1.08493245e+00 1.72006652e-01 2.25970283e-01 -2.00361624e-01 -5.31230688e-01 1.47386879e-01 1.30100176e-01 -5.61439872e-01 1.30051941e-01 -5.63591532e-02 9.61953282e-01 -1.17385018e+00 -8.99955511e-01 -8.83287489e-02 -2.48048544e-01 -7.19993651e-01 5.22940643e-02 -5.76378047e-01 3.45243722e-01 -7.01327562e-01 2.39441574e-01 1.01277782e-02 -1.48104548e-01 2.42353436e-02 -2.34551832e-01 -4.66680735e-01 8.63435090e-01 -7.50183165e-01 2.06709123e+00 -2.88362354e-01 1.17229879e+00 -3.29235345e-01 -6.95612788e-01 4.43099439e-01 2.62410522e-01 4.76408422e-01 -3.42281580e-01 3.18852812e-02 -2.43232772e-01 -3.40975225e-01 -6.54671192e-01 1.37773323e+00 9.43463594e-02 -8.17257762e-01 7.37383604e-01 -1.49335591e-02 -5.33643126e-01 8.70351136e-01 9.67878520e-01 1.41181910e+00 -6.03989772e-02 7.64491796e-01 9.11982672e-04 9.99288559e-02 5.83254337e-01 4.01523292e-01 1.01279533e+00 3.67060244e-01 9.43129778e-01 1.15535879e+00 1.04041612e-02 -1.27965915e+00 -8.85285437e-01 3.89488935e-01 9.26456213e-01 -1.06666625e-01 -1.52072954e+00 -9.18240130e-01 -6.23127997e-01 -6.81063890e-01 1.26456928e+00 -3.70348841e-01 2.59807091e-02 -9.40120876e-01 -3.60608160e-01 4.73568946e-01 3.84068072e-01 2.68125683e-01 -1.04286313e+00 -8.80530834e-01 3.39063644e-01 -7.92040706e-01 -1.10138607e+00 -4.53336447e-01 1.38301834e-01 -7.62553632e-01 -7.56341100e-01 -5.25562704e-01 -4.80724722e-01 4.98251557e-01 1.31972373e-01 1.29715538e+00 -2.69877464e-01 5.42903543e-02 4.54353541e-01 -7.73322582e-01 -5.48475862e-01 -6.39030874e-01 6.56866789e-01 -3.81169736e-01 -3.81392628e-01 -2.13815331e-01 -5.38926601e-01 -2.64967710e-01 -5.30653358e-01 -9.88683820e-01 7.18800604e-01 6.23398304e-01 4.97973293e-01 4.91555810e-01 2.03015253e-01 4.21203583e-01 -1.30110323e+00 1.02853084e+00 -3.29617947e-01 2.16878965e-01 2.25986525e-01 8.84019583e-02 4.94255684e-02 7.19637036e-01 -1.65060684e-01 -1.21837413e+00 -4.03062224e-01 -2.76231766e-02 1.41242430e-01 -7.12707415e-02 9.41509783e-01 -9.57245305e-02 1.04495084e+00 7.38226354e-01 3.38476539e-01 -6.39550745e-01 -4.42742378e-01 5.27628362e-01 5.51604390e-01 7.17899919e-01 -5.59972584e-01 5.81972182e-01 5.18431306e-01 -2.38353804e-01 -1.34069669e+00 -1.27840817e+00 -5.74637592e-01 -7.60908782e-01 -5.71684778e-01 7.85084009e-01 -7.33937621e-01 -5.47416992e-02 6.13509379e-02 -1.34432006e+00 -2.44225517e-01 -7.32965946e-01 3.26666862e-01 -8.17977428e-01 5.22423983e-01 -7.37042964e-01 -6.08102143e-01 -4.35289532e-01 -5.42430937e-01 1.14600790e+00 3.10951054e-01 -1.10307980e+00 -8.10118377e-01 2.03232512e-01 5.67816906e-02 -2.52765119e-01 7.21093178e-01 9.62504327e-01 -7.37749219e-01 -3.46347451e-01 -2.30920941e-01 1.89871192e-01 -5.13503104e-02 1.81573153e-01 3.81458521e-01 -7.04822302e-01 6.70326874e-02 1.98757704e-02 -1.40475601e-01 1.12216556e+00 6.31108046e-01 7.84849226e-01 -9.02314663e-01 -4.55514789e-01 3.26548368e-01 1.08507347e+00 1.29672848e-02 5.54966986e-01 4.56977457e-01 8.35462272e-01 7.95278728e-01 6.52142286e-01 8.54454875e-01 3.10058892e-01 2.75647730e-01 -3.58805694e-02 1.10802278e-01 -2.67539114e-01 -6.87416673e-01 4.86790568e-01 1.39782751e+00 4.13687825e-02 -6.28647685e-01 -6.87466204e-01 7.24380314e-01 -1.91732848e+00 -1.38227558e+00 -2.78219253e-01 1.52666914e+00 8.96595180e-01 4.67110306e-01 1.91640705e-01 1.88419715e-01 4.23625857e-01 8.47136021e-01 -2.49979571e-01 -6.39461458e-01 -2.88243711e-01 -1.14784107e-01 2.50123471e-01 3.92040491e-01 -9.41862524e-01 1.01006365e+00 6.45490122e+00 1.00085795e+00 -6.78839684e-01 -1.31715655e-01 4.65809643e-01 -6.26928627e-01 -6.71638787e-01 2.36573264e-01 -7.14394212e-01 2.09160492e-01 8.39406550e-01 -4.67115998e-01 4.93530445e-02 5.59825361e-01 5.33839345e-01 -4.73252624e-01 -1.38241124e+00 7.84509957e-01 4.04419303e-01 -1.76369464e+00 3.15498054e-01 -1.41448930e-01 1.06943548e+00 -4.50571805e-01 -2.42258281e-01 2.17756167e-01 2.49303073e-01 -7.24757135e-01 1.18911731e+00 5.51501215e-01 6.27520561e-01 -8.66809130e-01 4.63505775e-01 6.30553365e-01 -8.99404645e-01 1.22505665e-01 -1.41218260e-01 -2.04273850e-01 6.28977597e-01 5.61794460e-01 -8.59677196e-01 7.47997999e-01 2.53138185e-01 1.13608003e+00 -5.98811805e-01 7.99761891e-01 -6.77650750e-01 8.19144487e-01 -9.79416892e-02 -2.90076673e-01 5.17562091e-01 -1.71843350e-01 1.17807627e+00 1.64683938e+00 1.78608000e-01 3.83063972e-01 1.50614813e-01 5.36478043e-01 -9.20515582e-02 1.46948606e-01 -5.33272386e-01 -3.71718466e-01 2.19439521e-01 1.23343444e+00 -1.21780026e+00 -7.16037214e-01 -5.62077165e-02 1.06673372e+00 3.18958722e-02 4.41530883e-01 -6.45598650e-01 -3.41904402e-01 8.42729285e-02 6.35424405e-02 2.66040087e-01 -2.18960896e-01 -5.98063886e-01 -1.22971594e+00 6.90786215e-03 -8.97996426e-01 4.14731383e-01 -8.03615034e-01 -5.91304421e-01 4.97231454e-01 4.07313794e-01 -1.20661557e+00 -5.77636838e-01 1.89741939e-01 -1.01148522e+00 1.28894910e-01 -1.03043807e+00 -9.23421264e-01 2.34503355e-02 -4.34262082e-02 1.33230221e+00 1.24769740e-01 5.40315926e-01 -2.30341598e-01 -5.73662877e-01 9.31857005e-02 2.12607950e-01 5.92579274e-03 5.78202605e-01 -1.31509769e+00 4.84197587e-01 1.17718220e+00 4.43081766e-01 4.80741292e-01 1.37985241e+00 -9.72459316e-01 -1.10070407e+00 -7.77999699e-01 1.06282103e+00 -4.63285357e-01 5.89995265e-01 -4.54000980e-01 -5.11304140e-01 7.17485368e-01 7.43547082e-01 -1.18504059e+00 7.60974824e-01 3.04411054e-01 4.53595594e-02 1.93470046e-01 -4.36543822e-01 1.10430706e+00 1.15253258e+00 -4.85871613e-01 -1.13712168e+00 4.15755481e-01 8.22529554e-01 -4.31282759e-01 -3.78328532e-01 2.13689469e-02 3.66970539e-01 -7.99751341e-01 7.12446570e-01 -5.47538042e-01 1.23321164e+00 7.97547922e-02 1.02529645e-01 -1.43329978e+00 -2.36125588e-01 -1.04176188e+00 -3.33230674e-01 1.48147237e+00 4.18635428e-01 2.18478724e-01 7.55342305e-01 4.74408688e-03 -6.29553258e-01 -4.27334845e-01 -3.93757492e-01 -3.27172071e-01 -3.89863938e-01 -3.90984595e-01 1.83776662e-01 7.16274202e-01 5.57117105e-01 9.71956313e-01 -4.79519814e-01 -3.04726511e-01 5.49808919e-01 3.14939618e-01 1.00200176e+00 -9.35834944e-01 -1.05387568e-01 -6.98148429e-01 1.08204782e-01 -1.31438816e+00 1.15475282e-01 -6.99658751e-01 2.81648993e-01 -2.26975179e+00 6.67894959e-01 2.32662752e-01 2.85400003e-01 9.45946798e-02 -1.59061924e-01 -2.00872019e-01 2.48881444e-01 4.20939535e-01 -1.20686197e+00 4.25109148e-01 1.15105176e+00 -3.74858618e-01 -6.88419819e-01 -9.21038985e-02 -9.58708465e-01 9.80069518e-01 7.35871196e-01 -5.65717101e-01 -5.80129683e-01 -2.54972547e-01 6.28321409e-01 2.99674779e-01 -1.40871853e-01 -9.56103981e-01 4.54429537e-01 -2.85159975e-01 3.06613386e-01 -1.26105869e+00 2.83941358e-01 -1.25957057e-01 8.65120515e-02 1.39313117e-01 -8.86168480e-01 -1.01074770e-01 6.79698959e-02 3.87753397e-01 -3.57420176e-01 -5.07080615e-01 1.51422232e-01 -2.49061778e-01 -5.67356050e-01 -1.07628874e-01 -8.34827483e-01 2.48583362e-01 7.35509455e-01 -4.55020875e-01 -2.38303617e-01 -7.56744981e-01 -5.63660681e-01 6.07358292e-02 4.68688250e-01 2.13837922e-01 5.70576429e-01 -1.07492852e+00 -1.02145326e+00 -4.75744665e-01 6.30872175e-02 2.33185947e-01 3.73970270e-01 6.23701274e-01 -7.60337830e-01 5.11667192e-01 -6.25811815e-02 -3.93854409e-01 -1.32164621e+00 3.15862954e-01 -5.73078156e-01 -6.07461452e-01 -1.11775506e+00 5.74094534e-01 3.25539678e-01 5.00101745e-01 5.24371490e-02 -6.29806578e-01 -6.99430823e-01 7.80289412e-01 7.57377923e-01 3.88789833e-01 -1.89813852e-01 -5.56810677e-01 1.46619305e-02 2.73990929e-01 -3.95920008e-01 -5.40572286e-01 1.66060746e+00 -3.82712603e-01 -1.44864962e-01 8.60932827e-01 1.00198126e+00 6.63983881e-01 -1.24691677e+00 -9.86301526e-02 3.63110989e-01 -1.02718510e-01 -1.31420076e-01 -4.94841605e-01 -2.89920509e-01 4.97760147e-01 -7.74029613e-01 5.39190412e-01 9.80194688e-01 3.94394457e-01 8.37518215e-01 5.19519210e-01 -1.99787647e-01 -1.41892409e+00 3.37085754e-01 5.78132868e-01 1.07407844e+00 -5.65818310e-01 6.75151289e-01 -3.25731367e-01 -9.32060719e-01 1.15071166e+00 1.30390778e-01 -1.96347937e-01 -2.09027886e-01 2.52326787e-01 -4.26597714e-01 -4.43889111e-01 -9.71773803e-01 -1.95137933e-02 2.74764091e-01 1.79764807e-01 2.36908600e-01 6.87254220e-02 -4.63253528e-01 8.42900395e-01 -6.70834005e-01 -5.35719872e-01 1.13992524e+00 1.17885697e+00 -7.96116948e-01 -6.42308593e-01 -4.22233880e-01 6.58410668e-01 -5.80428600e-01 -5.32504693e-02 -8.69966149e-01 6.58629954e-01 -4.35077637e-01 1.00386608e+00 7.30239600e-02 -1.36792973e-01 4.12011057e-01 -1.87298786e-02 5.19899547e-01 -1.01259816e+00 -6.64159834e-01 4.09444720e-01 7.39156604e-01 -3.10143501e-01 -4.34870273e-01 -9.43330705e-01 -1.31045878e+00 -4.12859559e-01 8.34870711e-02 3.04623455e-01 4.18572962e-01 1.12098086e+00 1.44111499e-01 1.15756309e+00 3.79368633e-01 -1.00434124e+00 -1.12453617e-01 -1.01459873e+00 -3.78381073e-01 1.96402624e-01 4.59904343e-01 5.24590760e-02 -1.39327377e-01 5.55303156e-01]

[12.453764915466309, 9.503264427185059]

41b8d437-96da-4b95-ae62-70126d359809

revisiting-neural-retrieval-on-accelerators

2306.04039

https://arxiv.org/abs/2306.04039v1

https://arxiv.org/pdf/2306.04039v1.pdf

Revisiting Neural Retrieval on Accelerators

Retrieval finds a small number of relevant candidates from a large corpus for information retrieval and recommendation applications. A key component of retrieval is to model (user, item) similarity, which is commonly represented as the dot product of two learned embeddings. This formulation permits efficient inference, commonly known as Maximum Inner Product Search (MIPS). Despite its popularity, dot products cannot capture complex user-item interactions, which are multifaceted and likely high rank. We hence examine non-dot-product retrieval settings on accelerators, and propose \textit{mixture of logits} (MoL), which models (user, item) similarity as an adaptive composition of elementary similarity functions. This new formulation is expressive, capable of modeling high rank (user, item) interactions, and further generalizes to the long tail. When combined with a hierarchical retrieval strategy, \textit{h-indexer}, we are able to scale up MoL to 100M corpus on a single GPU with latency comparable to MIPS baselines. On public datasets, our approach leads to uplifts of up to 77.3\% in hit rate (HR). Experiments on a large recommendation surface at Meta showed strong metric gains and reduced popularity bias, validating the proposed approach's performance and improved generalization.

['Xing Liu', 'Fu Li', 'Zheng Yan', 'Xiao Sun', 'Yueming Wang', 'Zhaojie Gong', 'Jiaqi Zhai']

2023-06-06

['information-retrieval']

['natural-language-processing']

[-1.26059160e-01 -6.83357656e-01 -4.25714552e-01 -4.68825921e-02 -1.05945718e+00 -6.50206029e-01 7.89105594e-01 6.79481745e-01 -6.54149592e-01 4.11796123e-01 3.27659994e-01 -4.10900414e-01 -7.13843942e-01 -7.25107968e-01 -7.01976836e-01 -5.59281409e-01 -3.92413855e-01 1.02400136e+00 4.08464760e-01 -3.71657997e-01 5.56327045e-01 3.51139307e-01 -1.89191520e+00 1.97243243e-01 6.27729237e-01 1.19336879e+00 3.83863211e-01 7.11788952e-01 1.93637032e-02 1.36808217e-01 -3.40813220e-01 -3.32111597e-01 3.68226945e-01 5.60529411e-01 -6.27518773e-01 -5.76338887e-01 7.33067453e-01 -5.97799361e-01 -3.08386505e-01 6.87195122e-01 7.15598285e-01 3.75042200e-01 8.93528700e-01 -9.89702404e-01 -7.72494853e-01 3.60741556e-01 -6.41228199e-01 5.28366625e-01 4.98334378e-01 -2.54327863e-01 1.74735999e+00 -1.28399324e+00 4.62579340e-01 1.19896841e+00 5.81479669e-01 -1.12974212e-01 -1.25066149e+00 -5.00122249e-01 -2.28703823e-02 5.37933856e-02 -1.56626308e+00 -7.63024539e-02 -8.76778364e-02 -1.01635106e-01 1.34334815e+00 6.19874954e-01 5.14234185e-01 7.28985965e-01 3.73622358e-01 1.05328012e+00 5.75498402e-01 -2.60582924e-01 1.36430845e-01 2.78892159e-01 7.12491810e-01 4.52393621e-01 6.67440116e-01 -2.38721613e-02 -8.15033674e-01 -9.58983183e-01 4.25995469e-01 4.64529157e-01 -4.33686487e-02 -1.91687211e-01 -1.07662618e+00 9.67157841e-01 2.37546563e-01 9.15671587e-02 -3.80625546e-01 3.56973648e-01 3.25038493e-01 2.62168348e-01 4.65203285e-01 6.29074454e-01 -5.51527858e-01 -2.46489272e-01 -9.60832119e-01 6.90820098e-01 9.98823285e-01 9.83917832e-01 6.19242847e-01 -6.09917402e-01 -5.46634138e-01 1.10890722e+00 4.43562061e-01 8.50921512e-01 5.51856875e-01 -6.55217052e-01 3.62962514e-01 3.36885989e-01 1.30804613e-01 -9.19408619e-01 -4.04706448e-01 -7.43209243e-01 -5.65551400e-01 -3.99737716e-01 1.23300903e-01 1.79861262e-01 -4.71385628e-01 1.34365535e+00 3.46409202e-01 1.33955881e-01 -2.95540810e-01 8.88934076e-01 5.48654318e-01 9.13293123e-01 2.47722920e-02 -5.11023263e-03 1.74371266e+00 -1.12176359e+00 -3.24492067e-01 8.92295688e-02 9.21492875e-01 -1.05378330e+00 1.04347956e+00 5.06720483e-01 -1.29421175e+00 -3.33804488e-01 -9.33628380e-01 -3.78488958e-01 -4.77671891e-01 -2.15761438e-01 1.13758588e+00 5.57866871e-01 -1.28676653e+00 8.09860468e-01 -5.73654294e-01 -1.91898227e-01 1.37997225e-01 7.82675445e-01 -9.28467978e-03 -4.58393991e-01 -1.28198481e+00 6.00642264e-01 9.25924443e-03 -4.08556581e-01 -3.47129762e-01 -1.17131793e+00 -4.67952728e-01 3.35452884e-01 8.91016573e-02 -1.10887647e+00 1.02909327e+00 -1.80893898e-01 -1.10539246e+00 3.78920853e-01 -1.12491727e-01 -4.75329131e-01 -2.41340399e-02 -6.03247583e-01 -3.70654225e-01 1.28496557e-01 -1.45242037e-02 6.00378275e-01 7.42219150e-01 -7.58796513e-01 -7.87258804e-01 -3.06633413e-01 -1.19825453e-02 5.13032794e-01 -8.36221874e-01 7.14620724e-02 -9.88766134e-01 -5.38117409e-01 1.33595504e-02 -1.17749870e+00 -2.25046664e-01 -1.32210016e-01 4.76131402e-02 -5.55467844e-01 1.95560813e-01 -1.47398263e-01 1.66271830e+00 -2.10141945e+00 -6.38110116e-02 6.17340505e-01 2.33998269e-01 2.71180928e-01 -3.23446065e-01 9.00513947e-01 2.63958812e-01 -2.23748870e-02 3.19336474e-01 -6.44822717e-01 4.67389852e-01 1.15308352e-01 -4.77083087e-01 4.78751987e-01 -1.43222570e-01 9.36113358e-01 -9.59963679e-01 -2.76142925e-01 -1.05235009e-02 6.83639884e-01 -1.03609967e+00 3.22343521e-02 -1.46741956e-01 -4.09428835e-01 -6.17280245e-01 6.14554465e-01 6.36472046e-01 -7.18849361e-01 -5.77475429e-02 -2.07650751e-01 1.38092160e-01 5.47827363e-01 -1.31443036e+00 1.65038562e+00 -6.69114947e-01 1.98236957e-01 -2.75517225e-01 -4.36993241e-01 6.39215469e-01 -1.55013070e-01 5.94396949e-01 -8.62708330e-01 -1.01707064e-01 5.23738444e-01 -2.08354026e-01 -2.19354033e-02 1.18014622e+00 3.70826602e-01 -7.04082921e-02 8.00894022e-01 -4.62209322e-02 1.06481746e-01 3.61031234e-01 6.15585506e-01 1.38706803e+00 -4.30342853e-01 -1.91442236e-01 -6.44256055e-01 2.54234225e-01 -1.65680721e-01 -2.84888804e-01 1.23085797e+00 4.24695969e-01 5.15413940e-01 1.68674171e-01 -4.30567175e-01 -9.37334478e-01 -1.14344764e+00 -6.22426987e-01 1.76014626e+00 4.73914802e-01 -9.33165789e-01 -4.40811217e-02 -3.02601695e-01 4.83754784e-01 3.59547555e-01 -3.82922322e-01 -1.52947202e-01 -3.60472620e-01 -1.09435356e+00 1.76269203e-01 2.02358603e-01 -2.17141375e-01 -5.95467925e-01 -2.53561795e-01 3.49442810e-01 3.17083776e-01 -6.84322774e-01 -9.40372765e-01 2.03556478e-01 -9.37888622e-01 -6.98517084e-01 -9.29448128e-01 -5.55857480e-01 3.72309864e-01 5.87838769e-01 1.46803808e+00 2.76049048e-01 -4.92621660e-01 7.45566010e-01 -4.38638121e-01 7.02662840e-02 2.45699778e-01 2.94910640e-01 4.09155309e-01 -2.59141773e-01 3.64072233e-01 -5.86318851e-01 -1.29784167e+00 3.75499070e-01 -1.14704990e+00 -4.35342431e-01 7.57853210e-01 1.08081090e+00 6.43112183e-01 -1.32163510e-01 2.42976055e-01 -8.40744495e-01 9.57335234e-01 -9.36281741e-01 -4.16008115e-01 1.54290423e-01 -1.20764959e+00 3.39592844e-01 4.34009194e-01 -6.69583023e-01 -5.51374674e-01 -4.15483922e-01 -1.01127833e-01 -3.05273354e-01 4.40658361e-01 4.93686080e-01 6.56676054e-01 2.64732353e-02 5.93743563e-01 1.78244546e-01 -1.90173686e-01 -6.41430974e-01 5.56753099e-01 7.64201462e-01 1.25566050e-01 -1.00244701e+00 4.04332191e-01 3.08160126e-01 2.78776102e-02 -7.17522144e-01 -5.61850190e-01 -1.19226491e+00 -2.07408494e-03 4.71931934e-01 1.14506006e-01 -1.12335956e+00 -9.61970866e-01 -1.52330905e-01 -8.57541382e-01 -1.02165096e-01 -1.19809166e-01 5.76453328e-01 -9.87119004e-02 3.88120741e-01 -9.29036558e-01 -7.58061707e-01 -9.16385949e-01 -1.00351930e+00 1.72478580e+00 1.84493512e-02 -2.52184242e-01 -1.01595128e+00 2.39150062e-01 2.48362079e-01 6.16217077e-01 -6.64327085e-01 8.80483747e-01 -9.57623959e-01 -7.14415789e-01 -5.26966989e-01 -4.58234400e-01 -6.68132082e-02 -3.34552318e-01 -2.60810822e-01 -4.40317988e-01 -6.91377401e-01 -5.90103447e-01 -2.21404538e-01 8.91476512e-01 1.30399123e-01 9.00751829e-01 -1.85856611e-01 -5.79205990e-01 4.54523563e-01 1.37346959e+00 -2.32194155e-01 4.77548510e-01 2.57447571e-01 2.68184990e-01 1.82631612e-01 8.17136467e-01 8.76865327e-01 3.35915983e-01 9.89407599e-01 2.30853662e-01 7.29648247e-02 5.25596477e-02 -1.64474607e-01 4.22310680e-01 1.05190730e+00 2.83059418e-01 -4.37837422e-01 -6.99579895e-01 5.22740662e-01 -1.73816347e+00 -5.91753066e-01 -3.44864249e-01 2.44660544e+00 8.11930239e-01 2.65013933e-01 -7.99586438e-03 -1.78700730e-01 2.38719150e-01 -1.49865463e-01 -4.25667048e-01 -4.78004754e-01 4.24031354e-02 7.23278880e-01 8.37081552e-01 4.70559001e-01 -8.38336349e-01 6.02484822e-01 6.45969152e+00 1.41878891e+00 -7.72885323e-01 1.32352918e-01 4.18733686e-01 -5.90907514e-01 -4.86802042e-01 -2.15681523e-01 -1.52324843e+00 5.47928393e-01 1.25593591e+00 -2.41822496e-01 5.29940784e-01 7.90664196e-01 -2.46166468e-01 -2.68087894e-01 -1.22487867e+00 1.11860824e+00 5.67685552e-02 -1.38485134e+00 2.63156712e-01 3.73983890e-01 8.25266361e-01 4.34637815e-01 5.63997328e-01 6.69229269e-01 3.96120310e-01 -8.01802635e-01 3.36723089e-01 4.11344320e-01 6.28304005e-01 -6.75413489e-01 7.02823818e-01 3.26110780e-01 -1.13556516e+00 -9.08380821e-02 -5.57187319e-01 7.05423206e-02 1.07615501e-01 6.79708600e-01 -7.66204238e-01 4.05153781e-01 8.03668678e-01 5.64226925e-01 -4.83537674e-01 1.12778986e+00 5.79158485e-01 2.91145027e-01 -1.02116847e+00 -5.11298418e-01 3.21544647e-01 -2.27156967e-01 3.02105844e-01 1.40204275e+00 5.83926916e-01 8.50527287e-02 1.51768759e-01 1.73955306e-01 -8.73435363e-02 6.23903811e-01 -2.94623047e-01 3.46546508e-02 5.65716207e-01 1.26618791e+00 -4.73893672e-01 -5.09941339e-01 -3.03016365e-01 9.35159445e-01 4.06253517e-01 1.06240086e-01 -7.05075920e-01 -1.89021572e-01 8.97292137e-01 3.53324145e-01 5.09461224e-01 -8.84850547e-02 2.08094306e-02 -9.76601481e-01 -5.89733059e-03 -8.41250002e-01 6.00456595e-01 -4.25297081e-01 -1.63208973e+00 3.84956867e-01 1.97612703e-01 -1.19805813e+00 -9.32487845e-02 -7.52832413e-01 -1.53822973e-01 9.65285778e-01 -1.52220130e+00 -8.03166330e-01 2.53730953e-01 4.24911082e-01 4.27744716e-01 1.86042096e-02 8.86702299e-01 8.96277189e-01 -2.41399750e-01 9.53780055e-01 8.37437093e-01 -8.17712069e-01 7.91578591e-01 -1.21329772e+00 2.77240574e-01 -1.89176146e-02 4.34505999e-01 1.31764400e+00 6.22263908e-01 -3.74627322e-01 -2.11339498e+00 -7.34948754e-01 8.81897211e-01 -6.00071490e-01 9.46586430e-01 -3.08809489e-01 -7.84809411e-01 9.63037908e-02 -6.13089046e-03 1.35838762e-01 9.32898641e-01 5.47013104e-01 -5.77625930e-01 -1.65537462e-01 -8.05197418e-01 7.65407264e-01 1.01565289e+00 -4.54223752e-01 -3.22345138e-01 7.52219677e-01 7.48817921e-01 -2.89600730e-01 -1.23265374e+00 3.35850179e-01 7.48524904e-01 -6.87237918e-01 1.52674842e+00 -5.00167668e-01 2.32437491e-01 4.33100872e-02 -4.04659092e-01 -8.82876933e-01 -4.54155207e-01 -7.12838113e-01 -6.91895425e-01 6.63349569e-01 4.77259189e-01 -4.65580940e-01 6.51384652e-01 7.10306346e-01 3.83017063e-02 -1.06557214e+00 -6.35287941e-01 -7.95919418e-01 1.76209155e-02 -4.12363619e-01 5.42799950e-01 4.90956604e-01 1.05528221e-01 4.89662647e-01 -3.53017360e-01 1.76814795e-01 5.40036261e-01 2.03455731e-01 6.79484427e-01 -1.24826896e+00 -8.59047115e-01 -6.86122954e-01 -1.13584012e-01 -1.81772482e+00 -1.91884592e-01 -1.17703104e+00 -3.33091438e-01 -1.31183970e+00 3.29361916e-01 -9.38849330e-01 -7.59964526e-01 8.66678879e-02 -1.57009035e-01 5.64164162e-01 -1.66456163e-01 4.68734592e-01 -1.12148511e+00 5.62025726e-01 9.26434755e-01 -7.62441307e-02 -2.05354676e-01 -5.76104224e-02 -6.54427528e-01 2.23372713e-01 3.23227048e-01 -5.24734080e-01 -2.81160057e-01 -4.50469345e-01 7.14730382e-01 -1.72036756e-02 -2.02899545e-01 -6.68076694e-01 6.48928285e-01 2.99982160e-01 8.80342349e-02 -7.99720824e-01 7.09850192e-01 -6.71635747e-01 1.45348966e-01 2.74159193e-01 -3.57661247e-01 2.63300002e-01 2.64523894e-01 7.03150451e-01 2.39062030e-02 -3.16946745e-01 7.48240054e-02 3.26467782e-01 -3.17456156e-01 5.82260609e-01 1.03709428e-03 -7.68813565e-02 5.36891639e-01 2.23600835e-01 -4.39408988e-01 -1.41532451e-01 -3.80424798e-01 2.87039369e-01 1.46610588e-01 4.22220975e-01 4.78178203e-01 -1.41943800e+00 -5.43635428e-01 1.43220380e-01 3.46684784e-01 -2.61000097e-01 2.24213511e-01 8.35345328e-01 -5.07693231e-01 6.62922442e-01 4.86513823e-01 -5.82561672e-01 -1.25004029e+00 4.98360395e-01 -3.44931334e-01 -6.70516253e-01 -3.69866312e-01 9.56046700e-01 1.30072251e-01 -8.34966078e-02 2.70401388e-01 -3.28173079e-02 -4.10599262e-02 1.82243645e-01 8.74993920e-01 4.37580526e-01 2.79002964e-01 -3.14437807e-01 -2.73211330e-01 5.10936439e-01 -5.65584600e-01 -1.30568864e-02 1.31234181e+00 -6.92933425e-02 -2.75106221e-01 1.81965724e-01 1.77530074e+00 1.89651862e-01 -7.58194149e-01 -5.69969893e-01 8.53043720e-02 -6.19191527e-01 3.22087914e-01 -4.61488813e-01 -6.43205762e-01 5.30820847e-01 8.01634490e-01 2.95649678e-01 6.69540167e-01 7.40593299e-02 1.11237276e+00 8.10417533e-01 6.05717123e-01 -8.47941101e-01 1.38492256e-01 5.51188529e-01 5.39304614e-01 -1.18291521e+00 3.88577402e-01 -9.59704742e-02 -2.39644483e-01 8.98029447e-01 1.46379232e-01 -3.05132985e-01 1.02958035e+00 2.14581564e-01 -4.92133230e-01 -3.27870578e-01 -1.22902405e+00 -1.73687071e-01 8.02821815e-01 1.65892101e-03 5.03204226e-01 -2.07684580e-02 -5.97842753e-01 4.65594113e-01 -4.30375487e-02 -2.84513801e-01 -1.20794699e-01 8.01045954e-01 -5.06418943e-01 -1.40980434e+00 -2.29887500e-01 1.03582358e+00 -6.81162834e-01 -7.20938444e-01 4.63044047e-01 4.48591441e-01 -1.91893160e-01 8.25509548e-01 3.51447284e-01 -2.50103176e-01 3.21183294e-01 -2.53875375e-01 3.24808329e-01 -6.81291521e-01 -8.79146397e-01 1.14597514e-01 -5.04229702e-02 -6.10938072e-01 3.09858471e-01 -5.81700265e-01 -9.29367542e-01 -5.69939375e-01 -5.59889972e-01 3.80191386e-01 7.67969429e-01 5.48666835e-01 7.39469886e-01 1.19878165e-01 5.68972170e-01 -9.39086497e-01 -9.40851390e-01 -7.26743460e-01 -7.63121128e-01 6.44553125e-01 6.38695061e-02 -6.66724741e-01 -3.76151055e-01 -6.33916438e-01]

[11.397966384887695, 7.553837299346924]

66da2ffe-d93a-47c1-8484-ee78d6c6ee30

highrisk-prediction-from-electronic-medical

1712.00010

http://arxiv.org/abs/1712.00010v1

http://arxiv.org/pdf/1712.00010v1.pdf

Highrisk Prediction from Electronic Medical Records via Deep Attention Networks

Predicting highrisk vascular diseases is a significant issue in the medical domain. Most predicting methods predict the prognosis of patients from pathological and radiological measurements, which are expensive and require much time to be analyzed. Here we propose deep attention models that predict the onset of the high risky vascular disease from symbolic medical histories sequence of hypertension patients such as ICD-10 and pharmacy codes only, Medical History-based Prediction using Attention Network (MeHPAN). We demonstrate two types of attention models based on 1) bidirectional gated recurrent unit (R-MeHPAN) and 2) 1D convolutional multilayer model (C-MeHPAN). Two MeHPAN models are evaluated on approximately 50,000 hypertension patients with respect to precision, recall, f1-measure and area under the curve (AUC). Experimental results show that our MeHPAN methods outperform standard classification models. Comparing two MeHPANs, R-MeHPAN provides more better discriminative capability with respect to all metrics while C-MeHPAN presents much shorter training time with competitive accuracy.

['Jung-Woo Ha', 'Yun-Geun Lee', 'Jin Joo Park', 'You Jin Kim', 'Jeong Whun Kim', 'Borim Ryu']

2017-11-30

['deep-attention', 'deep-attention']

['computer-vision', 'natural-language-processing']

[ 3.76275294e-02 1.73406303e-01 -2.41052091e-01 -4.40090865e-01 -8.41620445e-01 2.27022439e-01 2.81874478e-01 6.11458361e-01 -4.62413639e-01 8.11519325e-01 6.34489000e-01 -5.96018374e-01 -3.84924710e-01 -1.01486313e+00 -3.54293078e-01 -3.47623855e-01 -6.91235244e-01 6.51641011e-01 7.33092874e-02 -3.52835655e-02 1.18886404e-01 3.21168602e-01 -8.65595818e-01 1.71824589e-01 9.07536685e-01 1.12291849e+00 -6.84695616e-02 8.27916026e-01 1.62359446e-01 1.06981432e+00 -2.34858051e-01 -4.82948542e-01 7.17843547e-02 -5.86262703e-01 -8.09299827e-01 -5.79721689e-01 -4.57155742e-02 -3.39046597e-01 -5.17390072e-01 8.14983368e-01 8.54697108e-01 -1.72743294e-02 8.59452665e-01 -6.41770065e-01 -1.03740156e+00 7.17962682e-01 -4.36512828e-01 6.10638142e-01 -1.48316575e-02 -3.62202600e-02 9.49871659e-01 -5.86062908e-01 4.15578693e-01 1.07960463e+00 8.07419300e-01 4.45987970e-01 -9.91300344e-01 -2.77345330e-01 9.15810838e-02 4.89462167e-01 -1.21029890e+00 2.02758744e-01 3.41635823e-01 -6.03037655e-01 7.38635182e-01 4.37839270e-01 6.00484788e-01 1.22236323e+00 6.30747318e-01 4.70090300e-01 8.88767898e-01 -9.45898294e-02 1.29935801e-01 1.56153766e-02 6.63372636e-01 6.07870936e-01 9.52661559e-02 2.43780151e-01 2.81524330e-01 -2.93754101e-01 1.17575836e+00 4.56671387e-01 -2.60340780e-01 1.78379014e-01 -1.07241571e+00 9.03665602e-01 8.11048448e-01 3.64235491e-01 -9.11275983e-01 -2.63766311e-02 4.92062360e-01 3.39052081e-01 2.74897903e-01 2.70618051e-01 -4.68815386e-01 1.43838212e-01 -6.27287626e-01 5.51180495e-03 5.77919543e-01 5.42330921e-01 -7.04493746e-02 -3.10760434e-03 -6.56691253e-01 7.77838528e-01 7.67433420e-02 2.44887799e-01 9.52516735e-01 -1.85193256e-01 2.21979663e-01 6.70021057e-01 -1.50083408e-01 -8.49620581e-01 -8.39664698e-01 -9.85061944e-01 -1.35529935e+00 -1.43746749e-01 -1.35476869e-02 -3.42135340e-01 -1.08054984e+00 1.44109499e+00 -1.36396065e-01 2.06862390e-01 1.11950390e-01 1.00935781e+00 1.08413732e+00 6.09409273e-01 6.96838439e-01 -9.15339440e-02 1.41828883e+00 -9.15666759e-01 -3.23509961e-01 2.21137658e-01 5.98202050e-01 -5.64280488e-02 7.04440713e-01 1.26833111e-01 -9.71407294e-01 -5.29870629e-01 -7.00702429e-01 -1.92581415e-01 -4.44797099e-01 2.02170894e-01 4.67823714e-01 4.19183135e-01 -7.70654798e-01 8.06356072e-01 -6.95900142e-01 -4.14489448e-01 6.50329590e-01 4.07880157e-01 -9.65715870e-02 1.64015189e-01 -1.44169223e+00 9.19371784e-01 8.03949013e-02 7.02549070e-02 -1.08015227e+00 -9.05942738e-01 -6.45247936e-01 4.53492552e-01 -1.37318566e-01 -9.02515769e-01 9.57744360e-01 -5.87560415e-01 -1.18601918e+00 6.88548505e-01 2.74612010e-01 -9.46608722e-01 6.06778562e-01 -4.76310194e-01 -5.28156519e-01 2.89248321e-02 2.40125898e-02 3.92256796e-01 2.47202292e-01 -4.53778625e-01 -4.30274129e-01 -4.72109318e-01 -3.60226154e-01 -1.97712388e-02 -1.54280752e-01 9.67171602e-03 9.62753743e-02 -8.34519327e-01 -2.55373716e-01 -6.21095598e-01 -1.09135604e+00 -1.11236840e-01 -5.30948043e-01 -2.00372919e-01 4.65820163e-01 -1.13544643e+00 1.33412766e+00 -1.83833551e+00 3.14939916e-01 1.94354042e-01 5.43856740e-01 5.62729418e-01 1.10608608e-01 1.02045730e-01 -3.96762818e-01 3.61813128e-01 -1.72872424e-01 2.24029407e-01 -3.17885011e-01 3.81254442e-02 -1.67060807e-01 2.38444701e-01 3.49806398e-01 1.25917470e+00 -7.13597298e-01 -5.04706621e-01 2.88813323e-01 5.92189193e-01 -6.06712401e-01 1.40190586e-01 -5.26618846e-02 5.68142414e-01 -8.47528338e-01 7.00465322e-01 1.67928576e-01 -8.75123322e-01 2.90661789e-02 6.36815652e-02 1.07558772e-01 4.70478386e-02 -3.79154742e-01 1.18591189e+00 -3.20396423e-01 2.84805506e-01 -6.65401697e-01 -1.24154663e+00 9.58478272e-01 6.17548764e-01 5.29703617e-01 -8.01500857e-01 4.68149632e-01 1.22052670e-01 3.47846568e-01 -7.96959281e-01 -2.44653642e-01 -1.51061222e-01 7.59468749e-02 9.43905786e-02 -1.13257296e-01 9.39012289e-01 -1.05176456e-01 -4.42809388e-02 1.12441730e+00 -3.50179046e-01 7.61483908e-01 -5.71682416e-02 7.65399098e-01 -4.06148344e-01 6.66149497e-01 1.00279820e+00 -3.79649788e-01 5.80708742e-01 7.36530602e-01 -8.76124442e-01 -1.03272176e+00 -7.80559838e-01 -5.05721390e-01 6.80413365e-01 -2.69361198e-01 -5.71901444e-03 -2.00265691e-01 -7.00665474e-01 6.32883757e-02 3.66228819e-01 -9.75355983e-01 -1.04310378e-01 -4.90101755e-01 -1.09102917e+00 4.52607572e-01 1.00836062e+00 4.40542787e-01 -1.19839823e+00 -8.35652530e-01 4.38420206e-01 2.26998121e-01 -5.75560749e-01 6.43475950e-02 3.81849967e-02 -9.67152417e-01 -1.01597643e+00 -1.51624608e+00 -8.92357469e-01 3.25700641e-01 -6.31401718e-01 1.06787395e+00 3.14774550e-02 -6.14565313e-01 -1.18449509e-01 -4.75792348e-01 -5.28055489e-01 -5.53849451e-02 1.36249542e-01 -3.30039084e-01 7.77182952e-02 4.68819857e-01 -7.69497156e-01 -9.63520288e-01 -5.90438433e-02 -1.54461280e-01 -3.90740000e-02 1.23011303e+00 8.33538115e-01 7.06788003e-01 -4.94123340e-01 1.04942334e+00 -1.07148421e+00 5.40981531e-01 -9.82783139e-01 -2.66096562e-01 2.45864645e-01 -8.16384733e-01 -1.47569194e-01 7.99390018e-01 -2.90612012e-01 -5.30308664e-01 -1.69480979e-01 -3.03539664e-01 -4.79286104e-01 -1.77975118e-01 8.33394825e-01 3.35462004e-01 2.79536366e-01 6.08174205e-01 3.08816135e-01 -1.85513481e-01 -8.99837732e-01 -1.14622250e-01 6.87903166e-01 5.17308056e-01 -1.48475796e-01 -4.10618298e-02 3.65242665e-03 1.71155259e-01 -6.77614093e-01 -7.80027270e-01 -2.05474615e-01 -3.09051186e-01 3.53677236e-02 1.12341821e+00 -7.28736401e-01 -6.50770009e-01 4.19882014e-02 -7.84740150e-01 -9.15434211e-02 -1.09825566e-01 9.26636219e-01 -3.03503066e-01 1.23591863e-01 -9.24083948e-01 -7.18033075e-01 -1.00602651e+00 -1.05390799e+00 7.06793666e-01 1.01846583e-01 -1.72504112e-01 -1.07392585e+00 4.77276519e-02 2.52972096e-02 6.42340124e-01 8.38125288e-01 1.65619588e+00 -1.00233650e+00 -2.79467672e-01 -5.06893992e-01 -5.96914351e-01 4.14879508e-02 -1.67695135e-01 -3.37189019e-01 -5.04353821e-01 8.93750191e-02 -3.76373380e-01 5.25378585e-02 9.70200658e-01 1.00974989e+00 1.61626256e+00 -2.70313531e-01 -3.99742007e-01 5.54638743e-01 1.38761187e+00 5.60165882e-01 8.71858776e-01 1.97830155e-01 5.99293888e-01 1.12762332e-01 4.15427871e-02 4.85306978e-01 2.97084570e-01 4.00967360e-01 4.61751044e-01 -3.78978342e-01 -8.51723701e-02 1.21931210e-01 1.38309319e-02 6.91333473e-01 -4.69811171e-01 -8.15508887e-02 -1.07914162e+00 5.84717691e-01 -1.85369694e+00 -7.92410314e-01 -4.21356887e-01 2.13391781e+00 6.61875427e-01 2.22078919e-01 3.06273848e-01 8.19464326e-02 8.93473327e-01 -2.17592016e-01 -7.03115940e-01 -4.85042125e-01 1.14948556e-01 5.55196941e-01 3.78429562e-01 2.85259932e-01 -1.22797549e+00 3.89801174e-01 6.24227476e+00 2.64575601e-01 -1.10511363e+00 1.44652650e-01 1.18644893e+00 6.97299791e-03 1.88945290e-02 -3.09262067e-01 -5.95300078e-01 8.00732315e-01 1.28456235e+00 -1.77838013e-01 -2.73554921e-01 8.33752871e-01 8.09009373e-02 4.76637155e-01 -1.11264598e+00 7.59184837e-01 -2.55779028e-01 -1.27843845e+00 3.37389857e-01 1.83119118e-01 4.12414253e-01 2.65046448e-01 1.13061123e-01 7.00484753e-01 3.76331300e-01 -1.32155669e+00 -2.27625370e-02 9.96857464e-01 9.57050562e-01 -5.52878737e-01 1.22584748e+00 -6.44545481e-02 -8.38791430e-01 -3.02404284e-01 -2.75067329e-01 5.82870133e-02 2.96286821e-01 6.09649658e-01 -5.40287018e-01 7.19757676e-01 6.97857440e-01 1.01961994e+00 -5.60105026e-01 1.33241367e+00 3.87864262e-02 7.98336685e-01 1.25177622e-01 -1.14247136e-01 5.71503639e-01 -1.05584204e-01 4.94159251e-01 1.14529908e+00 3.35481197e-01 3.05318862e-01 1.73449755e-01 7.11939037e-01 5.61284684e-02 4.33664352e-01 -2.85112262e-01 -2.10763067e-01 -3.81315337e-03 9.50441241e-01 -4.66678530e-01 -5.57505429e-01 -4.07400221e-01 5.90462387e-01 9.38242301e-02 1.11153603e-01 -9.98987377e-01 -5.55843651e-01 2.33920753e-01 2.45105445e-01 1.66641802e-01 4.91310000e-01 -4.53924954e-01 -1.04261887e+00 -3.54108930e-01 -5.79560101e-01 8.77070129e-01 -5.63259184e-01 -1.33341980e+00 8.50939631e-01 -6.47514939e-01 -1.16955078e+00 -1.46474227e-01 -5.35869658e-01 -5.57195723e-01 1.01670468e+00 -1.67747855e+00 -1.04351556e+00 -4.19821471e-01 4.40349281e-01 4.64137644e-01 -4.26040351e-01 1.13095641e+00 7.34445214e-01 -9.31448996e-01 6.09290004e-01 -7.63183385e-02 6.44360960e-01 4.40824986e-01 -1.25224400e+00 1.89628750e-01 2.26953015e-01 -4.40267295e-01 5.52309811e-01 2.25295857e-01 -6.97537243e-01 -5.68186939e-01 -1.46772170e+00 1.07463622e+00 -1.76205829e-01 3.25997591e-01 3.83736193e-01 -9.27274585e-01 9.18955624e-01 -2.84828156e-01 -4.26787622e-02 1.07368648e+00 2.67025530e-01 -1.76392943e-01 7.37847313e-02 -9.61500704e-01 4.82955098e-01 7.09023714e-01 -8.90417024e-02 -4.69521284e-01 5.29806614e-01 7.15673804e-01 -1.63613036e-01 -1.46759212e+00 7.10612118e-01 6.91902161e-01 -6.82680786e-01 9.99975026e-01 -1.47558439e+00 1.11530650e+00 1.92233518e-01 -5.76744750e-02 -8.58333647e-01 -9.37484145e-01 -1.13820225e-01 -2.45107129e-01 4.94356275e-01 6.12336755e-01 -5.73358059e-01 6.57098055e-01 4.19080466e-01 -1.50426477e-01 -1.51092625e+00 -6.29560709e-01 -5.33626080e-01 3.89473706e-01 1.48283616e-01 5.48758984e-01 1.09967732e+00 -1.53075382e-01 3.67370009e-01 -6.44669831e-01 1.32845789e-01 3.61488670e-01 1.42930493e-01 9.58696529e-02 -1.57727742e+00 -3.91245514e-01 -5.97799182e-01 -1.02598798e+00 -4.18272704e-01 -3.66593271e-01 -1.10141611e+00 -8.57218862e-01 -1.78922439e+00 4.64281231e-01 -3.96636188e-01 -1.03858757e+00 6.78092957e-01 -2.09339365e-01 -1.27253812e-02 -2.63112485e-01 2.66792327e-01 -3.30908716e-01 4.99424994e-01 9.94640648e-01 -7.09064752e-02 -3.33502352e-01 1.48744896e-01 -8.35739911e-01 5.57285845e-01 7.96062112e-01 -3.17348570e-01 -1.55984312e-01 -3.28739643e-01 1.66739803e-02 5.67769110e-01 6.20441616e-01 -1.17400777e+00 -4.09990847e-02 1.16774350e-01 8.08278084e-01 -4.61708456e-01 -1.03193531e-02 -4.39005673e-01 -1.02801248e-01 1.03572309e+00 -8.54328156e-01 2.77832542e-02 -2.33503491e-01 7.48695850e-01 -8.59962255e-02 7.51056001e-02 7.46296763e-01 -3.56732696e-01 -5.99938810e-01 5.92008829e-01 -4.31769580e-01 -4.24749330e-02 1.04891622e+00 1.80391073e-01 -2.15780064e-01 -3.46332818e-01 -1.36501145e+00 2.37890169e-01 -4.61346716e-01 3.22886437e-01 5.43938339e-01 -1.25233412e+00 -1.08071709e+00 3.55406851e-02 5.40250242e-02 -2.38595143e-01 4.55554724e-01 1.11931777e+00 -6.90300465e-01 7.72098541e-01 -2.93577611e-01 -3.85430276e-01 -9.85972166e-01 8.08395982e-01 5.59580028e-01 -6.39470696e-01 -1.10006928e+00 8.11910689e-01 1.46640375e-01 -3.85355651e-02 3.24387938e-01 -4.06974196e-01 -6.68305933e-01 -9.58676040e-02 6.58573210e-01 4.57221836e-01 1.42623007e-01 -3.91232997e-01 -1.89031184e-01 4.96253848e-01 -6.46680713e-01 2.91855097e-01 1.62701023e+00 4.75310981e-01 9.56028923e-02 1.91526711e-01 1.16422439e+00 -6.53020203e-01 -6.67060673e-01 -3.29308569e-01 -6.03276193e-02 -6.04618490e-02 2.12947324e-01 -1.11888921e+00 -1.22458005e+00 9.63547826e-01 1.04526889e+00 4.69988734e-02 9.39585090e-01 -1.45034909e-01 1.09987044e+00 8.49628374e-02 -1.52736932e-01 -5.93649864e-01 -2.83910722e-01 2.33670309e-01 8.68391573e-01 -1.07042980e+00 -1.92162246e-02 -6.64405078e-02 -6.27805650e-01 1.01909161e+00 4.77191031e-01 -3.93301725e-01 1.12242937e+00 -2.94734359e-01 6.68888912e-02 -3.70324194e-01 -7.38340318e-01 -2.02985674e-01 3.32308888e-01 2.89513320e-01 5.60268342e-01 2.27625981e-01 -6.39060497e-01 1.20219421e+00 -4.69623469e-02 2.78465897e-01 3.31188768e-01 7.68404186e-01 -3.58923703e-01 -7.06881821e-01 1.56904861e-01 1.11358666e+00 -8.10339987e-01 -3.45344156e-01 -3.07652146e-01 4.62299109e-01 -9.64069143e-02 3.01561534e-01 -2.14964002e-02 -3.67497921e-01 2.43785292e-01 2.56306559e-01 4.09223735e-02 -4.85618830e-01 -7.85339653e-01 -6.11791499e-02 -1.27634287e-01 -4.04922068e-01 4.96482477e-02 -3.71464968e-01 -1.02445662e+00 -1.17057911e-03 3.71185429e-02 4.79804203e-02 2.81652123e-01 7.08657801e-01 8.34235370e-01 1.09617662e+00 4.86317515e-01 -1.92537293e-01 -6.59985423e-01 -1.08914959e+00 -5.22524357e-01 3.15761715e-01 4.71800029e-01 -6.24146700e-01 6.96423352e-02 -1.11492656e-01]

[7.949476718902588, 6.413298606872559]

f79a2421-3a6e-4fea-89cd-2e8612dd3e78

class-adaptive-threshold-and-negative-class

2305.01884

https://arxiv.org/abs/2305.01884v1

https://arxiv.org/pdf/2305.01884v1.pdf

Class adaptive threshold and negative class guided noisy annotation robust Facial Expression Recognition

The hindering problem in facial expression recognition (FER) is the presence of inaccurate annotations referred to as noisy annotations in the datasets. These noisy annotations are present in the datasets inherently because the labeling is subjective to the annotator, clarity of the image, etc. Recent works use sample selection methods to solve this noisy annotation problem in FER. In our work, we use a dynamic adaptive threshold to separate confident samples from non-confident ones so that our learning won't be hampered due to non-confident samples. Instead of discarding the non-confident samples, we impose consistency in the negative classes of those non-confident samples to guide the model to learn better in the positive class. Since FER datasets usually come with 7 or 8 classes, we can correctly guess a negative class by 85% probability even by choosing randomly. By learning "which class a sample doesn't belong to", the model can learn "which class it belongs to" in a better manner. We demonstrate proposed framework's effectiveness using quantitative as well as qualitative results. Our method performs better than the baseline by a margin of 4% to 28% on RAFDB and 3.3% to 31.4% on FERPlus for various levels of synthetic noisy labels in the aforementioned datasets.

['S Balasubramanian', 'Bobbili Veerendra Raj Kumar', 'Badveeti Naveen Siva Kumar', 'Darshan Gera']

2023-05-03

['facial-expression-recognition']

['computer-vision']

[ 3.01746190e-01 2.01108456e-01 -6.65913671e-02 -7.21012473e-01 -1.07764196e+00 -4.67217624e-01 -2.08071470e-02 -1.12101465e-01 -5.48075676e-01 1.10369754e+00 -2.16267779e-02 2.46562168e-01 3.69740844e-01 -4.69816387e-01 -5.83198071e-01 -9.40625548e-01 2.08506584e-01 4.08004344e-01 1.48469776e-01 -6.47311211e-02 -8.80749226e-02 2.53337502e-01 -1.64060354e+00 7.48838544e-01 6.26906872e-01 1.30456603e+00 -1.19628787e-01 3.56121778e-01 9.23061743e-03 8.26377809e-01 -1.11205912e+00 -5.04771948e-01 2.45090663e-01 -3.54126394e-01 -6.48081720e-01 3.49597424e-01 4.71866190e-01 -4.20830905e-01 2.52687335e-01 1.42336297e+00 4.52279359e-01 -1.47462428e-01 5.54686964e-01 -1.31409645e+00 -1.86630204e-01 3.64082873e-01 -8.95674706e-01 -2.32304603e-01 2.96318591e-01 -7.91568235e-02 7.54793584e-01 -1.08502448e+00 5.52435219e-01 1.35865831e+00 4.74299550e-01 7.86247015e-01 -1.11358595e+00 -1.16502380e+00 2.05589324e-01 7.84598142e-02 -1.74171424e+00 -8.81031275e-01 7.15995789e-01 -4.22160894e-01 -4.52513322e-02 4.86397505e-01 3.32983702e-01 1.14636612e+00 -2.34738827e-01 6.94184780e-01 1.38604045e+00 -4.19015646e-01 3.73456180e-01 5.04809618e-01 1.19974494e-01 4.97931898e-01 1.09311864e-02 -1.21655449e-01 -5.98153830e-01 -3.05849671e-01 2.75922447e-01 -3.62707049e-01 -3.76090765e-01 1.67751640e-01 -4.89722669e-01 6.65982664e-01 1.73992813e-01 3.84051912e-02 -2.06307322e-01 -3.54920048e-03 4.19246376e-01 3.53681110e-02 5.46757936e-01 -3.80763076e-02 -3.96222115e-01 -7.87513889e-03 -1.03675175e+00 1.35929823e-01 5.54712713e-01 9.98947561e-01 7.74383426e-01 7.43667111e-02 -4.67515409e-01 1.17946613e+00 4.25286025e-01 4.46958780e-01 3.22755754e-01 -1.06943488e+00 2.32274100e-01 2.90544510e-01 5.38931131e-01 -1.00251484e+00 1.27451003e-01 -3.76282066e-01 -8.36368382e-01 4.53653663e-01 6.55551076e-01 -2.82259345e-01 -1.19245934e+00 1.89149261e+00 2.90798217e-01 -9.97487921e-04 -8.50304514e-02 1.23128223e+00 7.30557084e-01 5.10217965e-01 3.37763429e-01 -4.99136209e-01 1.39334702e+00 -6.58837795e-01 -9.78565872e-01 -1.45141363e-01 4.45618808e-01 -9.26245570e-01 1.02068532e+00 7.87748337e-01 -6.93010569e-01 -4.33766335e-01 -9.74974155e-01 3.88419330e-01 -4.11583632e-02 6.67292058e-01 4.13835645e-01 8.21688890e-01 -7.44814634e-01 2.40169510e-01 -4.67089474e-01 1.35230422e-01 8.46110225e-01 3.43662679e-01 -5.05850613e-01 -1.95087790e-01 -1.04586959e+00 4.34103131e-01 2.32304409e-01 2.70584345e-01 -8.03997040e-01 -3.15355808e-01 -5.25697887e-01 -9.18300599e-02 5.17157733e-01 1.03331171e-04 1.32353675e+00 -1.81681216e+00 -1.19122434e+00 9.38438416e-01 -5.11421323e-01 -2.24303871e-01 8.53224277e-01 -1.35609448e-01 -4.99052316e-01 9.62571502e-02 1.11922380e-02 7.57068634e-01 8.30469131e-01 -1.71995366e+00 -7.02177465e-01 -3.01862955e-01 -4.63769361e-02 5.09976372e-02 -8.30458105e-02 1.52392417e-01 -5.68054438e-01 -7.14948773e-01 2.15892106e-01 -9.81566310e-01 7.18077794e-02 4.12741601e-01 -4.52802002e-01 -7.67939538e-02 1.22022474e+00 -5.90773702e-01 1.02899885e+00 -2.31802535e+00 -6.85588956e-01 1.59853786e-01 2.89220214e-01 3.41734856e-01 1.51575923e-01 -2.31438190e-01 -8.07383358e-02 3.58426422e-01 -7.46991634e-02 -4.14752305e-01 -1.85038477e-01 4.00903523e-01 -1.43783122e-01 3.83043826e-01 2.75444061e-01 1.48505084e-02 -8.37279975e-01 -6.16325319e-01 -1.47743493e-01 5.84622562e-01 -4.11596239e-01 2.30864048e-01 -6.97965026e-02 4.93926913e-01 -3.77003610e-01 1.00498593e+00 1.07791424e+00 -5.13438247e-02 4.41874862e-02 -5.22573948e-01 3.94011855e-01 -2.24425733e-01 -1.58950627e+00 9.90002990e-01 -1.66196212e-01 5.69075346e-01 2.55504668e-01 -8.65026832e-01 9.64836061e-01 5.32585144e-01 2.34432876e-01 -3.05236936e-01 2.42862001e-01 1.82215214e-01 3.47050503e-02 -5.07265210e-01 5.32477975e-01 -2.90198088e-01 1.05555236e-01 3.71852890e-04 3.56427506e-02 3.93886149e-01 1.60289556e-01 -1.10949650e-01 7.28172779e-01 -7.94269666e-02 2.69524800e-03 -1.09224990e-02 4.47019279e-01 -3.73344392e-01 1.12254119e+00 6.85012341e-01 -6.32110953e-01 6.85808420e-01 6.73255503e-01 -2.18916729e-01 -8.23674083e-01 -7.70620644e-01 -2.88416773e-01 1.00365901e+00 9.49026085e-03 -2.56476015e-01 -8.11495125e-01 -9.31219935e-01 -4.05264735e-01 3.87357175e-01 -6.75313532e-01 3.75729240e-02 -5.58265373e-02 -7.48532236e-01 5.36329865e-01 1.97739795e-01 7.39307165e-01 -8.65714133e-01 -2.44714811e-01 -9.12549198e-02 -3.72838616e-01 -1.15214133e+00 -3.57165575e-01 2.21877739e-01 -2.93782115e-01 -1.07729888e+00 -6.94508135e-01 -5.83706021e-01 1.14196348e+00 -1.08348265e-01 8.28174174e-01 1.80210337e-01 4.86392230e-02 -2.78289318e-01 -5.79455137e-01 -5.27861416e-01 -4.11523879e-01 -3.59463274e-01 -1.09833360e-01 4.83779281e-01 3.66710812e-01 -3.08050588e-02 -6.30675316e-01 7.20579922e-01 -7.78097510e-01 -1.16691038e-01 2.87846893e-01 1.11115479e+00 8.26405466e-01 1.89435378e-01 6.99577630e-01 -1.23259413e+00 3.66813242e-01 -4.12068874e-01 -4.37673151e-01 1.88430458e-01 -4.10508454e-01 -1.56183153e-01 6.30019903e-01 -7.36696243e-01 -1.00883925e+00 2.69454360e-01 -1.53573960e-01 -6.17500544e-01 -3.48997772e-01 1.14060864e-01 -3.96732390e-01 -1.47786409e-01 6.31443381e-01 -1.60978243e-01 -2.45134041e-01 -2.68577069e-01 -1.59124702e-01 1.00469530e+00 3.87970924e-01 -6.75901711e-01 4.04515058e-01 3.06634605e-01 -1.80702835e-01 -5.32677650e-01 -9.79915798e-01 -2.16377020e-01 -1.68766677e-01 -4.46121156e-01 4.65538859e-01 -1.13742232e+00 -5.29543400e-01 3.72924924e-01 -1.05122900e+00 9.03654844e-02 -1.00904956e-01 3.13880742e-01 -8.34399164e-02 2.50088215e-01 -4.83010054e-01 -1.31427026e+00 -1.93990439e-01 -1.44764066e+00 1.34551585e+00 3.53944212e-01 -3.58231544e-01 -3.78428757e-01 -6.85600162e-01 4.63889062e-01 1.65814072e-01 3.93383831e-01 4.73332554e-01 -6.46855354e-01 -1.59463644e-01 -5.46412587e-01 -1.78378746e-01 7.73550808e-01 1.11835860e-01 2.35149145e-01 -1.41647172e+00 -1.11599259e-01 -1.85601920e-01 -5.99454045e-01 6.67501092e-01 5.94475605e-02 1.43732524e+00 -4.38703567e-01 -2.06319392e-01 2.94724464e-01 1.34701538e+00 3.65460485e-01 8.25591087e-01 -1.67688966e-01 3.35769743e-01 6.36258841e-01 1.10044420e+00 5.68192363e-01 -6.37982637e-02 6.96923375e-01 5.01950145e-01 -8.05112869e-02 2.92374492e-02 -6.81399554e-02 2.01240674e-01 1.69694439e-01 3.73938754e-02 -4.78538096e-01 -5.62105536e-01 3.68887842e-01 -1.58762670e+00 -8.48391294e-01 -8.97663683e-02 2.30057240e+00 1.25979960e+00 1.40181690e-01 -1.65908914e-02 4.87015367e-01 9.63577509e-01 -1.12656474e-01 -2.13586390e-01 -2.22918153e-01 -2.80569136e-01 1.73372239e-01 4.16472435e-01 3.70920569e-01 -1.20521295e+00 9.74953890e-01 5.48509693e+00 1.29073668e+00 -1.31592727e+00 -2.15484463e-02 1.39287424e+00 -1.89942904e-02 -2.22484879e-02 -1.62963197e-02 -9.73217905e-01 8.57181013e-01 6.03832603e-01 3.49085450e-01 -4.90816217e-03 9.86219287e-01 4.90460247e-01 -5.95088005e-01 -8.35759401e-01 1.19523835e+00 -1.60676017e-02 -8.15899789e-01 -6.60058856e-02 -1.56652495e-01 6.82422757e-01 -5.47563434e-01 -3.17518078e-02 2.03076571e-01 1.34488940e-01 -1.20992434e+00 9.38686252e-01 6.12613201e-01 9.92418706e-01 -8.57267857e-01 1.13754821e+00 3.62958640e-01 -7.70422399e-01 1.62321478e-01 -4.79834020e-01 2.95862257e-01 -8.63614529e-02 8.62229347e-01 -8.75289083e-01 1.43167466e-01 8.80730689e-01 2.73249328e-01 -3.59344155e-01 9.97282565e-01 -3.18403423e-01 8.06345522e-01 -3.67315769e-01 1.71275705e-01 -1.29740611e-01 -4.93868068e-03 2.26967961e-01 1.11297750e+00 3.36887002e-01 2.28715822e-01 2.53161550e-01 5.24196029e-01 -3.54379833e-01 1.03248738e-01 -1.84786841e-01 1.21101528e-01 6.75931871e-01 1.28384423e+00 -5.25903404e-01 -4.25055176e-01 -8.79200101e-02 9.27082419e-01 3.18835787e-02 4.87574577e-01 -8.38480651e-01 -3.48820120e-01 3.72318923e-01 1.84687465e-01 3.07728294e-02 2.88852870e-01 -1.74512058e-01 -8.11924815e-01 1.69615284e-01 -9.95783150e-01 3.48778218e-01 -1.02972984e+00 -1.25494432e+00 8.94607306e-01 -2.66565710e-01 -1.35348320e+00 -3.46507728e-02 -3.31356019e-01 -1.16089039e-01 9.68246043e-01 -1.38571215e+00 -8.97845626e-01 -5.11866748e-01 5.03313482e-01 3.21994424e-01 -8.35294425e-02 9.79086459e-01 5.95259547e-01 -4.06850517e-01 1.07792044e+00 -1.74706906e-01 4.69213665e-01 1.10315275e+00 -9.44294810e-01 -7.01187670e-01 5.72600722e-01 -1.08750634e-01 2.00964093e-01 9.42822337e-01 -4.55921710e-01 -6.23400092e-01 -1.08369899e+00 8.56586635e-01 3.30276713e-02 1.72892556e-01 -3.39669198e-01 -9.45921898e-01 3.67232263e-01 -1.93147838e-01 6.71342969e-01 7.79458463e-01 -1.54617026e-01 -3.68693203e-01 -3.82655323e-01 -1.66806817e+00 4.75268662e-01 6.34385526e-01 -2.61846244e-01 1.03827372e-01 3.06057274e-01 2.00342789e-01 -5.61336815e-01 -7.11365104e-01 6.03662848e-01 5.47625542e-01 -9.27845359e-01 4.22638565e-01 -3.88229936e-01 2.92155743e-01 -6.08570337e-01 -4.17395264e-01 -1.07686794e+00 1.94328964e-01 -3.94262612e-01 1.41223282e-01 1.47416270e+00 4.31911558e-01 -2.37479299e-01 1.02437818e+00 5.98595023e-01 1.55634135e-01 -8.62520814e-01 -1.10223961e+00 -3.94378841e-01 -3.77898812e-01 -4.98499364e-01 4.39974368e-01 1.03966427e+00 -4.49692935e-01 3.03590856e-02 -7.81092525e-01 2.39541844e-01 3.84518057e-01 -3.46033812e-01 6.83330536e-01 -9.82037306e-01 -1.20593563e-01 3.07987817e-02 -4.47379708e-01 -7.78459668e-01 3.18936035e-02 -3.21439236e-01 3.55578303e-01 -7.68108845e-01 2.62384027e-01 -7.08571970e-01 -1.98543072e-01 8.66252542e-01 -1.42963290e-01 6.43091679e-01 -1.65964775e-02 1.47870928e-01 -7.12384701e-01 3.71943146e-01 1.21746612e+00 -1.29941106e-01 1.61115527e-01 9.93593186e-02 -7.12028503e-01 8.87410104e-01 6.49826765e-01 -6.52408123e-01 -2.83408701e-01 -1.60969108e-01 -7.61121660e-02 2.02183023e-01 3.25568438e-01 -9.01664257e-01 -1.10936716e-01 -1.72133207e-01 7.09934235e-01 -3.72264028e-01 5.26749372e-01 -1.04927862e+00 1.81377172e-01 -7.03702122e-02 -3.74841273e-01 -4.42854404e-01 6.16242290e-02 4.40859586e-01 -3.79577219e-01 -3.68759304e-01 1.15939152e+00 -5.63467331e-02 -4.08337116e-01 1.68567330e-01 -2.67020851e-01 -1.02439541e-02 9.06470418e-01 -3.46687317e-01 -2.94458747e-01 -6.81492686e-01 -9.34315622e-01 1.49498418e-01 2.71700352e-01 1.29134342e-01 4.73029375e-01 -1.31533599e+00 -5.67948878e-01 1.25872523e-01 3.05907428e-01 1.85469631e-03 1.30742252e-01 6.70529902e-01 -3.09910476e-01 -2.45200962e-01 9.26359147e-02 -7.42236376e-01 -1.70311379e+00 -7.05087036e-02 5.15779495e-01 -6.62431717e-02 1.43850520e-01 9.52929735e-01 1.15980636e-02 1.03394188e-01 5.31140029e-01 -4.48892266e-02 -3.25702518e-01 1.51182741e-01 6.20508611e-01 4.75761145e-02 9.39021260e-02 -9.39443827e-01 -3.24072599e-01 3.18182349e-01 -2.34598354e-01 -2.47826725e-02 8.94534349e-01 -2.27658413e-02 1.29641905e-01 4.70386326e-01 1.15856862e+00 2.58690268e-01 -1.29103160e+00 -1.97827280e-01 -1.26312166e-01 -7.94725060e-01 -8.11818987e-02 -1.14015579e+00 -1.32970417e+00 5.75425684e-01 1.04105616e+00 -1.57080829e-01 1.41747415e+00 -2.45413616e-01 3.92309129e-01 1.37498051e-01 3.63301277e-01 -1.33333814e+00 -6.15157895e-02 2.71342844e-01 8.12408686e-01 -1.53887773e+00 -1.13232033e-02 -5.74895680e-01 -8.12887013e-01 9.74548578e-01 7.59511590e-01 -7.86653347e-03 5.86755991e-01 4.62802559e-01 3.28759253e-01 1.06937125e-01 -6.74120307e-01 1.05595738e-01 2.00414509e-01 5.47588408e-01 5.19600570e-01 5.52546345e-02 -4.08320487e-01 9.51391160e-01 -1.32575095e-01 1.47847220e-01 4.80294794e-01 8.54127526e-01 -4.73936915e-01 -1.07339358e+00 -6.51008964e-01 4.91770476e-01 -9.22247171e-01 1.56012595e-01 -3.22468460e-01 7.45796025e-01 5.70561647e-01 1.18083882e+00 -1.03940576e-01 -4.99924362e-01 1.33187652e-01 4.34870496e-02 2.22284108e-01 -4.96181101e-01 -3.21395159e-01 5.49305797e-01 4.18671072e-01 -2.84886539e-01 -5.86580992e-01 -3.84413600e-01 -1.29613554e+00 -1.69376880e-01 -5.31633377e-01 2.97620058e-01 5.12801766e-01 9.38342571e-01 3.62175032e-02 3.39915812e-01 6.93712473e-01 -3.80211651e-01 -3.83330643e-01 -1.05937111e+00 -6.28414333e-01 5.14609694e-01 2.43652016e-01 -7.19743371e-01 -5.68366885e-01 2.89426476e-01]

[13.596339225769043, 1.6502774953842163]

1718b2b7-4e9a-4626-9574-7484c95fb8d3

gans-and-alternative-methods-of-synthetic

2306.01469

https://arxiv.org/abs/2306.01469v1

https://arxiv.org/pdf/2306.01469v1.pdf

GANs and alternative methods of synthetic noise generation for domain adaption of defect classification of Non-destructive ultrasonic testing

This work provides a solution to the challenge of small amounts of training data in Non-Destructive Ultrasonic Testing for composite components. It was demonstrated that direct simulation alone is ineffective at producing training data that was representative of the experimental domain due to poor noise reconstruction. Therefore, four unique synthetic data generation methods were proposed which use semi-analytical simulated data as a foundation. Each method was evaluated on its classification performance of real experimental images when trained on a Convolutional Neural Network which underwent hyperparameter optimization using a genetic algorithm. The first method introduced task specific modifications to CycleGAN, to learn the mapping from physics-based simulations of defect indications to experimental indications in resulting ultrasound images. The second method was based on combining real experimental defect free images with simulated defect responses. The final two methods fully simulated the noise responses at an image and signal level respectively. The purely simulated data produced a mean classification F1 score of 0.394. However, when trained on the new synthetic datasets, a significant improvement in classification performance on experimental data was realized, with mean classification F1 scores of 0.843, 0.688, 0.629, and 0.738 for the respective approaches.

['Charalampos Loukas', 'Tom OHare', 'Charles MacLeod', 'Christopher MacKinnon', 'Ehsan Mohseni', 'S. Gareth Pierce', 'Shaun McKnight']

2023-06-02

['synthetic-data-generation', 'hyperparameter-optimization', 'synthetic-data-generation']

['medical', 'methodology', 'miscellaneous']

[ 5.94596922e-01 3.42777222e-01 7.89259017e-01 -2.16091439e-01 -1.26109827e+00 -2.22206056e-01 4.06847656e-01 -8.71464089e-02 -2.46792257e-01 5.91589868e-01 -3.31940234e-01 -1.97252288e-01 -4.52824920e-01 -9.28986251e-01 -8.70030999e-01 -1.09357488e+00 -2.29584709e-01 5.56047082e-01 2.66986668e-01 -2.51162767e-01 2.61154085e-01 4.91670042e-01 -1.82466471e+00 4.51721311e-01 6.49488151e-01 1.20854604e+00 2.34393865e-01 8.79663765e-01 3.28037739e-01 5.02236247e-01 -9.19206321e-01 -2.83834115e-02 1.82979614e-01 -5.48859179e-01 -3.15541744e-01 1.17592998e-01 2.21474692e-01 -1.75966740e-01 -1.25198692e-01 8.24395299e-01 1.04505169e+00 -2.07093116e-02 6.64858460e-01 -1.02745891e+00 -3.77674401e-01 6.51301563e-01 -1.63744628e-01 -2.03044891e-01 4.41324741e-01 4.03352290e-01 3.32801372e-01 -7.30207801e-01 6.80281699e-01 8.15296829e-01 9.08378422e-01 6.77993000e-01 -1.33111668e+00 -5.16053140e-01 -8.95860076e-01 -9.81066003e-03 -1.00465107e+00 -2.64595836e-01 9.33175266e-01 -4.59092498e-01 8.02630961e-01 3.34211320e-01 5.96342862e-01 1.18064916e+00 2.46513754e-01 2.29919046e-01 1.46508527e+00 -7.77783334e-01 4.64614153e-01 -4.80957376e-03 -2.19145805e-01 5.38035572e-01 2.71010250e-01 6.66489661e-01 -1.48127675e-01 -1.33128567e-02 7.03478634e-01 -9.14482415e-01 -4.54579532e-01 -1.44804299e-01 -9.93335009e-01 6.57463133e-01 1.48704171e-01 4.52628613e-01 -4.79355752e-01 8.58336780e-03 5.03878117e-01 5.35319567e-01 2.47857258e-01 7.44683802e-01 -4.42707896e-01 3.05918735e-02 -7.38394558e-01 2.32214272e-01 9.30660725e-01 5.37280381e-01 6.43540084e-01 7.42309690e-01 -5.45422062e-02 9.35123861e-01 1.62686259e-01 6.10636234e-01 5.33816516e-01 -9.50080633e-01 -3.73105519e-02 2.04412162e-01 8.07836931e-03 -9.46295321e-01 -6.40650988e-01 -7.21377969e-01 -7.02022493e-01 4.95845020e-01 4.55529064e-01 -3.33294660e-01 -1.37407398e+00 1.56273592e+00 1.10980414e-01 1.08452894e-01 4.07894939e-01 8.17375422e-01 1.06462216e+00 4.96377587e-01 -1.37913957e-01 -7.07124025e-02 8.24278772e-01 -5.58225870e-01 -5.90657532e-01 -3.99352703e-03 6.47398293e-01 -9.68102455e-01 8.74702513e-01 7.20044434e-01 -1.13060284e+00 -7.55748212e-01 -1.51832354e+00 8.64392817e-01 -2.58397907e-01 3.85340631e-01 1.99564770e-01 9.95296061e-01 -1.23323357e+00 7.87287891e-01 -6.62374794e-01 1.14461295e-02 2.43711248e-01 4.61577028e-01 -2.98877895e-01 -1.76740587e-01 -1.20743740e+00 7.18524396e-01 3.49620283e-01 2.61461496e-01 -9.48850274e-01 -8.49957645e-01 -9.29837286e-01 -5.81540838e-02 1.68053955e-01 -1.59776568e-01 1.27747834e+00 -7.81456590e-01 -1.84201789e+00 7.72619486e-01 6.88052535e-01 -2.83912271e-01 5.96896410e-01 2.69936353e-01 -6.55886233e-01 3.45628500e-01 2.61783954e-02 3.81403118e-01 6.57250166e-01 -1.80963540e+00 -6.79892376e-02 2.09418997e-01 -4.09302026e-01 -4.92081493e-01 2.89769560e-01 -2.50650436e-01 -6.09796271e-02 -7.21570551e-01 5.24336100e-01 -8.48600209e-01 -2.03944221e-01 -3.95869464e-01 -3.90025795e-01 3.22815806e-01 5.35050631e-01 -7.04859138e-01 5.43046236e-01 -1.96244431e+00 -3.19407821e-01 7.05776155e-01 -4.28932644e-02 5.05633593e-01 -3.35606039e-01 5.83360851e-01 -5.77462971e-01 -6.12478070e-02 -4.84663069e-01 1.81495324e-01 -1.53878197e-01 -4.70285490e-02 3.31611037e-01 3.50111395e-01 4.56758857e-01 7.91170001e-01 -8.31785679e-01 -1.17063276e-01 3.00648481e-01 4.58031595e-01 -5.14741600e-01 4.30832356e-01 -9.69756320e-02 5.45289516e-01 -1.95324644e-01 5.93817055e-01 1.01202929e+00 2.46707112e-01 1.12806946e-01 -5.95980346e-01 6.14410043e-02 -2.73991346e-01 -1.11740220e+00 1.53166664e+00 -6.47938788e-01 5.99952161e-01 2.71985918e-01 -1.45738411e+00 1.02356517e+00 7.34009683e-01 7.38602340e-01 -1.02641368e+00 3.36749882e-01 6.24791145e-01 5.03121614e-01 -1.02382123e+00 -3.36132534e-02 -4.30978686e-01 2.36811172e-02 4.96757448e-01 4.56453919e-01 -7.17730939e-01 2.54803598e-01 -2.77222574e-01 1.21312773e+00 2.66131908e-01 -4.92643744e-01 -3.33508641e-01 4.97266054e-01 7.63066486e-02 1.10597283e-01 9.64294434e-01 2.03870207e-01 1.19242895e+00 4.60935444e-01 -2.57155329e-01 -1.24646556e+00 -9.81327116e-01 -1.37847066e-01 2.31845886e-01 -2.04613328e-01 8.41479897e-02 -1.07304418e+00 -4.11096722e-01 -3.63813043e-01 7.82626927e-01 -6.82572186e-01 -5.27362287e-01 -6.58824503e-01 -9.97019589e-01 6.90967917e-01 1.24427244e-01 5.38984418e-01 -1.27722728e+00 -5.72422504e-01 4.26174074e-01 9.29622352e-03 -1.12709463e+00 4.21033621e-01 4.28272724e-01 -6.20244384e-01 -1.21157503e+00 -7.30144560e-01 -8.19574833e-01 7.31759548e-01 -2.02011138e-01 1.10857534e+00 3.70643258e-01 -7.14536905e-01 5.19475400e-01 -5.67891359e-01 -4.04128373e-01 -1.32487524e+00 -3.50364506e-01 -1.31385073e-01 -1.57380939e-01 -1.78218886e-01 -5.85383713e-01 -6.29819810e-01 4.04770732e-01 -1.30303907e+00 -2.56059617e-01 8.30598831e-01 1.28707445e+00 4.18772191e-01 2.42253974e-01 7.54814625e-01 -8.29067767e-01 5.69972575e-01 -2.60565102e-01 -2.89606184e-01 3.96024175e-02 -5.86676359e-01 5.53384013e-02 5.75980067e-01 -4.81765479e-01 -1.27398944e+00 -1.04946069e-01 -6.57947421e-01 -1.03782371e-01 -5.11652589e-01 7.17779338e-01 3.09314169e-02 -3.65318954e-01 1.04073191e+00 1.87067032e-01 4.04651165e-01 -1.93961114e-01 -3.15446883e-01 5.86748898e-01 6.11562729e-01 -7.60475338e-01 6.10590160e-01 5.39163798e-02 1.03404693e-01 -9.71082628e-01 -3.68796080e-01 9.37824883e-03 -2.96778768e-01 -6.09331250e-01 4.61657166e-01 -4.11515474e-01 -3.88084352e-01 1.23009574e+00 -8.22426319e-01 -4.96960431e-01 -4.84078348e-01 7.28958905e-01 -6.85445249e-01 4.50807899e-01 -5.32900095e-01 -6.03153348e-01 -3.64151269e-01 -1.30849195e+00 9.08128321e-01 -6.83667511e-02 1.05064161e-01 -7.23286986e-01 -1.33348629e-01 1.91232845e-01 7.05733597e-01 7.11801946e-01 1.11413264e+00 -6.52119219e-01 -1.41030237e-01 -7.59821653e-01 -4.78378981e-02 8.70648324e-01 1.78007975e-01 7.14842379e-02 -1.12262976e+00 -2.96290249e-01 4.24408019e-01 -5.64569771e-01 5.24071515e-01 5.54115832e-01 9.97096360e-01 1.94948569e-01 2.38596741e-02 2.11203218e-01 1.67620850e+00 5.08426726e-01 1.02491736e+00 2.23717183e-01 2.09578872e-01 6.00065291e-01 3.79074126e-01 7.21419528e-02 -5.92710853e-01 5.52017212e-01 4.33721721e-01 -3.45113605e-01 -6.06380284e-01 1.43698931e-01 -5.84795065e-02 6.94204330e-01 -1.10256433e-01 -3.51902127e-01 -1.00237191e+00 3.27151895e-01 -1.11916327e+00 -6.37630224e-01 -3.51086348e-01 1.99386621e+00 6.30052924e-01 4.57035989e-01 -5.11987686e-01 6.43737555e-01 6.60264611e-01 -4.06731248e-01 -2.53498733e-01 -2.69159704e-01 -3.16985637e-01 8.65936816e-01 1.88876003e-01 1.67735800e-01 -7.95952082e-01 3.05891633e-01 6.39400673e+00 8.40340137e-01 -1.26207197e+00 -1.60672054e-01 6.19478405e-01 3.59295756e-01 -1.51729941e-01 -5.05338371e-01 5.08126803e-02 5.04584908e-01 1.27626526e+00 3.60468030e-01 -1.35631217e-02 4.23305750e-01 1.55697644e-01 -4.16315615e-01 -8.11035097e-01 5.36958158e-01 2.37529259e-02 -1.30313694e+00 -5.74136414e-02 -1.95636317e-01 9.93767262e-01 -2.30555475e-01 3.05084568e-02 1.63366377e-01 5.80939278e-02 -1.06507921e+00 6.16159499e-01 5.76443732e-01 9.29894805e-01 -5.00118792e-01 1.24322200e+00 2.08202168e-01 -4.05345112e-01 5.67286275e-02 -1.26859888e-01 1.96610272e-01 8.47593918e-02 7.70383060e-01 -9.60949957e-01 8.61644387e-01 6.11199915e-01 -6.20248131e-02 -4.16923702e-01 1.07323503e+00 2.65091863e-02 1.07334745e+00 -3.80610377e-01 9.78334174e-02 1.30485892e-01 6.12650327e-02 3.71099204e-01 1.10760736e+00 7.80813634e-01 -1.11494340e-01 -2.71933615e-01 6.28153801e-01 3.18776101e-01 -1.34853363e-01 -4.48873013e-01 1.73657879e-01 5.53021133e-02 1.16078174e+00 -1.00354218e+00 -2.96126567e-02 -1.39744729e-01 4.21335697e-01 -4.27690774e-01 3.69900525e-01 -8.24330688e-01 -5.14020562e-01 -1.08602799e-01 2.84456342e-01 3.05186868e-01 -3.39275673e-02 -2.45153189e-01 -4.92938757e-01 -1.41096920e-01 -1.00814295e+00 -1.33226132e-02 -1.14049625e+00 -1.20009744e+00 7.89384067e-01 1.19725838e-01 -1.34660637e+00 -4.91244584e-01 -9.76009250e-01 -6.25868380e-01 9.39016640e-01 -9.80458260e-01 -9.50796425e-01 -5.44839382e-01 -1.01761267e-01 2.97173887e-01 -4.45656806e-01 9.84761000e-01 4.01696026e-01 -1.21210776e-01 5.36254585e-01 1.69327393e-01 -3.14315632e-02 3.91813457e-01 -1.02862191e+00 2.16690525e-01 6.44813538e-01 -3.15840453e-01 7.61388335e-03 9.48549807e-01 -5.22174060e-01 -1.22022665e+00 -6.77746892e-01 1.33423895e-01 1.95394024e-01 4.60866153e-01 -1.92083586e-02 -1.06197035e+00 8.16092454e-03 1.82649374e-01 -4.79775183e-02 5.60820401e-01 -7.28866816e-01 1.81278691e-01 3.08024585e-02 -1.52610004e+00 2.05093935e-01 5.79758525e-01 -1.31426215e-01 -8.39005634e-02 3.75219911e-01 3.98564905e-01 -7.05775917e-01 -1.08538485e+00 8.04578125e-01 4.63015974e-01 -1.05463827e+00 9.85448122e-01 -2.46822998e-01 7.72971153e-01 -1.35043800e-01 -1.56296387e-01 -1.55588269e+00 -3.80825391e-03 -2.86192209e-01 2.62485147e-01 1.00507283e+00 7.12787926e-01 -6.19724095e-01 9.20560122e-01 2.71411806e-01 -5.66783130e-01 -8.44711304e-01 -1.01591825e+00 -5.81970453e-01 8.05592090e-02 -7.69484043e-01 5.03180683e-01 5.27902484e-01 -6.65560484e-01 -1.95689380e-01 -2.04035221e-03 1.80855870e-01 5.97407758e-01 -1.76887989e-01 4.27915543e-01 -1.04330897e+00 -2.81507641e-01 -2.55767733e-01 -7.45129287e-01 -2.07490712e-01 -1.00144841e-01 -5.98420620e-01 3.84074181e-01 -1.30253232e+00 -3.14359963e-01 -6.78740859e-01 -2.83085823e-01 6.35775402e-02 1.55131921e-01 7.29951262e-01 -3.86089087e-01 -1.81420580e-01 4.55517948e-01 9.95133445e-02 1.49602962e+00 -2.71006942e-01 1.44189537e-01 3.05973291e-01 -3.41692895e-01 3.68363857e-01 1.09850252e+00 -3.82499427e-01 -4.26331729e-01 -2.67222881e-01 1.89478487e-01 3.67470741e-01 5.68071246e-01 -1.60890210e+00 -1.67543553e-02 4.06675249e-01 4.94687557e-01 -4.19327170e-01 1.95078358e-01 -7.45789886e-01 6.23239458e-01 7.52358675e-01 -2.41189539e-01 -4.47480530e-01 4.28697348e-01 3.21784467e-01 -3.32148850e-01 -7.79475868e-01 8.25015604e-01 -2.49706626e-01 -6.02191567e-01 -2.86535949e-01 -5.39009273e-01 -2.65386909e-01 8.34861219e-01 -5.37061214e-01 3.61882560e-02 -3.26199889e-01 -1.07087815e+00 -3.59746844e-01 1.16087452e-01 3.07590738e-02 7.99519777e-01 -1.17371750e+00 -9.44913864e-01 7.59422600e-01 -1.75283447e-01 -4.43154760e-02 6.80535316e-01 6.99342847e-01 -1.00773442e+00 8.70517939e-02 -3.74311328e-01 -8.21232378e-01 -8.94719899e-01 1.50170788e-01 7.31958747e-01 -1.48660645e-01 -2.76615620e-01 9.22669530e-01 -2.78016835e-01 -8.00329924e-01 -6.23369850e-02 -3.28983426e-01 -1.24902293e-01 -1.32559985e-01 -8.26668590e-02 4.38383281e-01 9.44424808e-01 -4.56239462e-01 1.59285203e-01 6.58056796e-01 2.91136414e-01 -2.61734873e-01 1.63326824e+00 5.32355249e-01 1.32201137e-02 7.45532587e-02 1.33081365e+00 -2.60574818e-01 -1.02073026e+00 2.06831902e-01 -1.60476774e-01 -1.82625636e-01 1.44832477e-01 -1.19694376e+00 -1.49566436e+00 5.65448046e-01 9.82650042e-01 5.33005953e-01 1.40993047e+00 -2.68859655e-01 5.02088189e-01 2.92666346e-01 3.79300743e-01 -1.02472293e+00 2.00868219e-01 1.18337058e-01 1.02806294e+00 -1.14130461e+00 -2.24965159e-02 -7.06078768e-01 -2.72747725e-01 1.40604818e+00 6.09958708e-01 -1.59823105e-01 8.52871835e-01 5.94589591e-01 3.37412268e-01 -5.89952767e-01 -3.02077740e-01 2.63181657e-01 1.43750325e-01 1.03939033e+00 3.39812964e-01 -1.99485183e-01 -3.63178939e-01 5.33694029e-01 -2.44604111e-01 1.66847229e-01 6.97763085e-01 1.16502857e+00 -1.49719000e-01 -1.07607532e+00 -4.74342227e-01 6.64561808e-01 -5.82267880e-01 2.43525788e-01 1.50311634e-01 1.02832437e+00 1.96507007e-01 9.72904384e-01 -3.31775323e-02 -6.05145931e-01 5.97649217e-01 -2.19639614e-02 5.68714440e-01 -4.77796227e-01 -5.90891182e-01 -2.16314048e-02 3.05012882e-01 -2.54057646e-01 -5.40696740e-01 -4.96064931e-01 -1.06924331e+00 1.10559538e-01 -4.70909625e-01 3.49928886e-01 9.58412647e-01 7.01030195e-01 -6.33167997e-02 1.10561621e+00 7.06290424e-01 -1.16123950e+00 -6.30598545e-01 -1.19184995e+00 -7.04909921e-01 4.85074669e-01 7.83298910e-02 -6.63056433e-01 -5.45905411e-01 8.34821314e-02]

[12.019292831420898, -0.6216424107551575]

899e7a2b-7f6e-4a67-ac85-841a0075366f

ltcr-long-text-chinese-rumor-detection

2306.07201

https://arxiv.org/abs/2306.07201v2

https://arxiv.org/pdf/2306.07201v2.pdf

LTCR: Long-Text Chinese Rumor Detection Dataset

False information can spread quickly on social media, negatively influencing the citizens' behaviors and responses to social events. To better detect all of the fake news, especially long texts which are harder to find completely, a Long-Text Chinese Rumor detection dataset named LTCR is proposed. The LTCR dataset provides a valuable resource for accurately detecting misinformation, especially in the context of complex fake news related to COVID-19. The dataset consists of 1,729 and 500 pieces of real and fake news, respectively. The average lengths of real and fake news are approximately 230 and 152 characters. We also propose \method, Salience-aware Fake News Detection Model, which achieves the highest accuracy (95.85%), fake news recall (90.91%) and F-score (90.60%) on the dataset. (https://github.com/Enderfga/DoubleCheck)

['Ying Shen', 'Guian Fang', 'Mengsha Liu', 'Ziyang Ma']

2023-06-12

['misinformation', 'fake-news-detection']

['miscellaneous', 'natural-language-processing']

[-4.22549039e-01 4.27373089e-02 -6.70589626e-01 6.84401467e-02 -6.57874048e-01 -4.77983296e-01 8.99216890e-01 3.86271268e-01 -2.13397294e-01 9.00706053e-01 6.00061357e-01 -3.20301414e-01 6.11032665e-01 -9.05117452e-01 -4.84015107e-01 -3.07771444e-01 3.57109517e-01 1.88685358e-01 4.90972728e-01 -6.91035748e-01 6.47970378e-01 -1.11599654e-01 -1.02265561e+00 5.32808602e-01 1.29794800e+00 3.84303778e-01 -1.56556010e-01 3.82763714e-01 -1.90464817e-02 1.09265876e+00 -9.89024103e-01 -8.46802711e-01 -3.41845274e-01 -4.97432768e-01 -6.51738286e-01 -9.46074054e-02 1.45440862e-01 -5.13302863e-01 -5.33040285e-01 1.61536872e+00 2.36884087e-01 -4.07114178e-01 5.26718974e-01 -1.07092810e+00 -8.07339668e-01 9.16006148e-01 -8.23819518e-01 5.49731255e-01 4.17024046e-01 -1.23359993e-01 5.99714398e-01 -8.79843354e-01 7.41227388e-01 1.20788300e+00 6.80239797e-01 3.83236468e-01 -3.00790936e-01 -9.10306811e-01 -2.06053227e-01 6.09674118e-02 -1.43499887e+00 -3.22096586e-01 4.03402895e-01 -6.28764987e-01 2.65727609e-01 4.53662127e-01 6.16192400e-01 1.48893046e+00 3.83737952e-01 9.21395719e-01 1.36029434e+00 -6.78729936e-02 -2.58391619e-01 4.56103235e-01 2.67589778e-01 6.80795312e-01 9.33750749e-01 -3.15553732e-02 -3.57896477e-01 -6.25255227e-01 1.60273820e-01 2.09530115e-01 -3.44046831e-01 8.41265440e-01 -1.43099213e+00 1.20704186e+00 4.75208730e-01 4.36214775e-01 -2.68711299e-01 -1.84007987e-01 6.50098205e-01 4.01490748e-01 1.11279821e+00 2.54952729e-01 -1.19331479e-01 -2.31499985e-01 -5.28988540e-01 4.31821674e-01 6.27009153e-01 9.56190825e-01 2.60150284e-01 -1.27695784e-01 -8.26319084e-02 7.60380030e-01 1.53955832e-01 1.18956506e+00 6.51649475e-01 -1.86652631e-01 8.46859455e-01 4.67318684e-01 7.92253554e-01 -1.96133792e+00 -3.77346486e-01 -5.17770708e-01 -8.70774031e-01 -8.66086483e-01 2.88794279e-01 -2.38987625e-01 -3.82171303e-01 1.06248939e+00 3.68868738e-01 2.09575728e-01 -1.17733933e-01 1.10761189e+00 1.11532652e+00 8.73067141e-01 -3.39399993e-01 -3.23455006e-01 1.51644659e+00 -9.52708483e-01 -1.05837786e+00 -4.27059501e-01 9.39826131e-01 -1.14408672e+00 8.04886043e-01 1.41160607e-01 -4.56576139e-01 2.04018503e-01 -9.92623270e-01 1.01131693e-01 -4.92244571e-01 2.01318637e-01 3.28973621e-01 6.93436384e-01 -1.79041177e-01 1.63128734e-01 -3.12842160e-01 -7.79191479e-02 5.45894802e-01 -6.84784710e-01 5.22836447e-02 -2.85332531e-01 -1.70678341e+00 1.01220202e+00 3.41397256e-01 -8.53374153e-02 -5.87678254e-01 -1.04304090e-01 -3.65320951e-01 -4.09736961e-01 3.86639297e-01 -1.88739404e-01 1.18275654e+00 -8.27581286e-01 -1.06763148e+00 9.07703459e-01 2.67687272e-02 -5.05002618e-01 1.10732901e+00 -4.13817763e-01 -9.88850832e-01 1.71021655e-01 5.59818268e-01 -1.74955621e-01 1.00666118e+00 -9.11584258e-01 -4.45775568e-01 -2.29877934e-01 -4.61832166e-01 -1.40196383e-01 -1.34791732e-01 3.74312013e-01 8.11614916e-02 -8.92283678e-01 2.00273499e-01 -9.34869647e-01 2.30766654e-01 -6.42468095e-01 -8.58158410e-01 1.20309800e-01 9.45889950e-01 -1.08191705e+00 1.43932128e+00 -1.81517100e+00 -7.98043668e-01 -7.93947577e-02 5.92539430e-01 6.46353841e-01 2.02724710e-01 7.07943141e-01 4.06481445e-01 3.34679157e-01 1.27886385e-01 2.40710989e-01 -3.35737228e-01 -2.87070423e-01 -7.54718125e-01 1.08373618e+00 -2.62135565e-01 1.02492523e+00 -1.43795276e+00 -1.79866418e-01 -3.03803116e-01 8.48116279e-02 -1.24362223e-02 -4.01164055e-01 -1.03288136e-01 2.70163685e-01 -8.71173859e-01 6.07187271e-01 1.02164960e+00 -4.56656039e-01 2.94393711e-02 2.18951374e-01 -2.92572707e-01 7.87092149e-01 -4.84737664e-01 5.97899199e-01 7.12853596e-02 1.04478192e+00 -3.37155282e-01 -4.12457407e-01 9.64191556e-01 1.74248427e-01 -1.13084644e-01 -7.64098704e-01 5.57153583e-01 3.75839621e-01 -4.31225568e-01 -7.48930395e-01 9.21192408e-01 -4.32456434e-02 -4.73346651e-01 7.32802689e-01 -7.72712529e-01 1.29751161e-01 -2.11135507e-01 6.00039780e-01 8.44570816e-01 -8.46139550e-01 5.40923059e-01 -1.99344292e-01 1.85989022e-01 4.48638052e-01 2.82231867e-01 9.02678788e-01 -3.84560466e-01 2.26401299e-01 6.04641438e-01 -4.79890019e-01 -1.04876482e+00 -3.65161628e-01 -5.86991534e-02 5.86298108e-01 6.65833056e-01 -3.96626770e-01 -5.63458562e-01 -7.81856060e-01 1.24459080e-01 8.39180589e-01 -5.25956750e-01 -1.69680655e-01 -3.48708630e-01 -1.09913576e+00 9.34639394e-01 -3.63722295e-01 9.40140188e-01 -6.38417363e-01 -5.77553138e-02 1.75318867e-01 -1.07760751e+00 -1.28293574e+00 -5.82843304e-01 -9.96831954e-01 -5.38063705e-01 -1.11163974e+00 -4.77052450e-01 -4.03965205e-01 5.63724995e-01 1.10745311e+00 7.69372344e-01 5.60361803e-01 1.73237935e-01 -5.01112401e-01 -7.30019152e-01 -4.39199686e-01 -8.50780666e-01 -1.07641391e-01 1.18726641e-01 -7.00394735e-02 3.98842096e-01 1.57069385e-01 -3.61368328e-01 6.52012110e-01 -7.56265998e-01 3.47987115e-01 2.91654855e-01 9.66642380e-01 -9.81064364e-02 9.33239460e-02 8.65460038e-01 -1.08666718e+00 8.51030588e-01 -1.03281283e+00 -6.30068004e-01 -1.87207863e-01 -4.06217039e-01 -5.94493568e-01 4.18464541e-01 -6.02738380e-01 -8.32656324e-01 -8.41176391e-01 -9.39616875e-04 3.47614080e-01 1.27021357e-01 8.14913511e-01 6.45702541e-01 1.24076605e-01 1.00770247e+00 3.27558577e-01 -2.62638181e-02 -3.24852705e-01 4.21625748e-02 1.32099450e+00 1.18821926e-01 2.22227573e-01 6.92221224e-01 6.52712703e-01 -8.75496745e-01 -1.04462147e+00 -1.38984036e+00 -6.65462971e-01 6.39607087e-02 -2.21041471e-01 2.73184806e-01 -1.13136160e+00 -5.12188435e-01 9.88458335e-01 -1.50012934e+00 5.05472533e-02 6.60165429e-01 5.27683020e-01 3.09717119e-01 7.05861628e-01 -1.03165555e+00 -8.12357903e-01 -3.67133409e-01 -6.33967638e-01 5.74562013e-01 -5.86316995e-02 7.75587857e-02 -6.86158478e-01 -5.65145835e-02 7.49753952e-01 4.74543899e-01 4.46959525e-01 5.47592677e-02 -7.82560050e-01 -3.38220388e-01 -7.03208089e-01 -7.57334828e-01 2.07095761e-02 4.91714589e-02 -8.67441595e-02 -7.00963557e-01 -2.33420894e-01 -3.33687328e-02 -2.92414874e-01 8.99537504e-01 -1.02688357e-01 5.33850551e-01 -1.27657461e+00 -5.28469026e-01 -2.44985953e-01 1.02856266e+00 -3.50002080e-01 7.94169605e-01 4.53093410e-01 5.27995825e-01 3.35571527e-01 9.63698804e-01 7.18418181e-01 5.91752589e-01 5.89980543e-01 4.44086343e-01 3.40204120e-01 1.01081990e-01 -6.15333617e-01 6.62205458e-01 1.17976904e+00 1.69178575e-01 -5.87211728e-01 -9.92123544e-01 5.81823468e-01 -1.74274993e+00 -1.47714877e+00 -9.41757381e-01 1.96106207e+00 7.13002563e-01 2.38873333e-01 2.86001623e-01 -7.72428289e-02 1.25226843e+00 5.34160674e-01 -3.61984745e-02 9.55429394e-03 -2.98545212e-01 -8.78821552e-01 9.31044161e-01 6.79402113e-01 -1.18699038e+00 1.21885037e+00 5.54956245e+00 1.01364088e+00 -1.25474215e+00 5.25002480e-01 5.35607815e-01 3.40122074e-01 -2.97184378e-01 -2.19366536e-01 -9.42568421e-01 1.10520732e+00 8.46570134e-01 -2.06614375e-01 1.92429990e-01 8.07349324e-01 6.24171615e-01 -1.85323492e-01 3.75105828e-01 7.85330355e-01 2.72014350e-01 -1.67999744e+00 -1.26950234e-01 6.53707683e-02 1.09984863e+00 6.39100432e-01 -7.55898356e-02 1.18644208e-01 4.25200582e-01 -7.39176929e-01 9.29867923e-01 2.06085637e-01 5.86714864e-01 -6.37757778e-01 1.24688435e+00 1.04919279e+00 -4.00265694e-01 1.01235062e-01 -6.27022803e-01 -2.88973719e-01 2.40979806e-01 1.23709190e+00 -1.01293635e+00 2.89334834e-01 4.42471355e-01 1.15946209e+00 -4.24409866e-01 8.41870725e-01 -4.82490569e-01 8.72765005e-01 -1.48392320e-01 -7.55395114e-01 3.91398311e-01 -5.40946126e-02 8.99622560e-01 1.44219494e+00 2.72530079e-01 1.64179683e-01 -4.69501615e-02 5.20360768e-01 -3.20233405e-01 1.99846283e-01 -7.20160127e-01 -2.35988110e-01 8.29840124e-01 7.98697412e-01 -5.58896542e-01 -6.50967360e-01 -1.55670807e-01 8.63398731e-01 1.14400692e-01 4.88948524e-02 -1.20787752e+00 -2.88029522e-01 2.97784805e-01 2.79578477e-01 8.36874396e-02 -8.79816338e-02 -1.63758755e-01 -1.72623837e+00 -7.51290396e-02 -1.11650848e+00 2.54736036e-01 -7.06903458e-01 -1.49168217e+00 4.75444347e-01 -3.14591646e-01 -1.47982609e+00 2.34416708e-01 -3.99014354e-02 -4.20699626e-01 3.86250854e-01 -1.48206294e+00 -8.79137337e-01 -5.16189396e-01 2.08247796e-01 2.18620986e-01 5.15807560e-03 3.76068264e-01 3.16851914e-01 -6.04134262e-01 4.54752862e-01 4.70318407e-01 3.98834199e-01 6.87006593e-01 -3.95948529e-01 6.00053906e-01 7.27182329e-01 -3.66653353e-01 5.13616920e-01 9.18467760e-01 -1.10140705e+00 -9.57394481e-01 -1.32560909e+00 1.41073310e+00 -5.64107060e-01 1.08488429e+00 -1.70907438e-01 -7.88960278e-01 5.38175404e-01 -1.09000094e-01 -1.92213044e-01 2.77245879e-01 -3.37744445e-01 -6.24882519e-01 4.74581599e-01 -1.34917593e+00 6.29513979e-01 6.93295121e-01 -5.67856252e-01 -4.34996843e-01 8.70469332e-01 7.90582657e-01 -6.86632931e-01 -3.67558688e-01 -1.78300112e-01 6.17632210e-01 -1.01994729e+00 6.23625040e-01 -5.50713181e-01 6.05350733e-01 -2.68696994e-01 2.59037882e-01 -1.19801009e+00 -2.67441392e-01 -5.70067525e-01 -7.09659830e-02 7.37088680e-01 5.14181793e-01 -1.29664540e+00 3.82986516e-01 -3.11008781e-01 6.09249957e-02 -3.07073951e-01 -6.99808538e-01 -7.89976299e-01 -2.27012515e-01 -1.56422988e-01 5.07501304e-01 1.76035190e+00 2.22642645e-01 2.12226391e-01 -1.00992012e+00 3.79005790e-01 4.74619746e-01 1.72037661e-01 8.23444128e-01 -8.75009537e-01 1.44115463e-01 -2.47684777e-01 -2.24539995e-01 -1.24636567e+00 -2.84715772e-01 -5.02097428e-01 -1.39686763e-01 -1.05776918e+00 3.50076556e-01 -3.36806774e-01 4.88222897e-01 2.01255843e-01 -1.59490228e-01 5.02461791e-01 -3.22948575e-01 6.83133006e-01 -5.67355752e-01 5.77786863e-01 1.60494375e+00 -2.08480760e-01 2.07777862e-02 1.95483744e-01 -7.32938290e-01 8.12220037e-01 1.17125988e+00 -1.03403342e+00 2.16962665e-01 -1.89589426e-01 6.89732194e-01 2.16919601e-01 5.31113088e-01 -2.83850253e-01 -5.45918755e-02 -4.58396882e-01 -1.19249858e-01 -7.73743451e-01 5.73211408e-04 3.01569290e-02 -7.93398730e-03 9.84703898e-01 -2.40076214e-01 1.31889954e-01 -1.52214780e-01 7.82223105e-01 -1.97210148e-01 -1.47567704e-01 8.86534989e-01 -1.96043491e-01 -1.17699027e-01 9.90893412e-03 -7.15372801e-01 3.11888069e-01 7.25348532e-01 3.55413049e-01 -1.51770425e+00 -8.75572026e-01 -2.00286523e-01 -1.71977341e-01 4.53830928e-01 5.40619552e-01 6.24246061e-01 -1.21085525e+00 -1.25046265e+00 -2.22157925e-01 4.32244956e-01 -5.85178614e-01 2.27356434e-01 1.24158144e+00 -8.66797507e-01 4.43280160e-01 1.26734719e-01 3.22949816e-03 -1.10484827e+00 4.74821985e-01 -6.97997678e-03 -8.34490135e-02 -7.25536764e-01 4.52403396e-01 -5.57953179e-01 -2.62406826e-01 -4.44567263e-01 -1.81425363e-02 -1.88430533e-01 8.51432830e-02 1.05750382e+00 7.12585926e-01 -8.19851235e-02 -1.20673573e+00 -1.97858587e-01 -1.43306866e-01 -3.52378160e-01 1.87364310e-01 1.05520034e+00 -4.79109854e-01 -4.63677049e-01 9.04602483e-02 1.11722147e+00 7.36388147e-01 -3.64085585e-01 -3.91229749e-01 3.64406183e-02 -1.00177276e+00 1.79461643e-01 -7.84419060e-01 -6.30494356e-01 2.44547904e-01 -1.21694691e-01 6.76789284e-01 1.24194704e-01 6.38382435e-02 1.25252557e+00 1.67419270e-01 4.05168355e-01 -9.36830223e-01 3.33218992e-01 7.32071102e-01 1.00759828e+00 -1.42357051e+00 4.20053899e-01 -7.72789538e-01 -8.23562145e-01 7.96112299e-01 7.96648264e-02 -2.41038859e-01 4.15829509e-01 -3.19710165e-01 2.36037105e-01 -5.12501359e-01 -4.76511955e-01 2.65606016e-01 2.02720597e-01 1.54965401e-01 1.65529534e-01 4.97236580e-01 -8.44315290e-01 4.20655429e-01 -3.77389431e-01 -3.15089047e-01 1.16659975e+00 6.49576128e-01 -9.38476324e-01 -4.30966675e-01 -4.58444357e-01 5.63349605e-01 -8.17981720e-01 -2.39279509e-01 -5.30964255e-01 6.35758162e-01 -3.36320311e-01 1.20016754e+00 -4.79157418e-01 -5.86943805e-01 -3.19721013e-01 -5.61791062e-01 -2.08309546e-01 -2.26623058e-01 -3.22409213e-01 -2.05584198e-01 7.96054900e-01 -3.13719690e-01 -2.44291812e-01 -4.92434174e-01 -7.93530405e-01 -1.08562148e+00 -9.07690942e-01 2.69067168e-01 5.62282979e-01 1.01631570e+00 4.06544656e-01 -2.30805308e-01 9.18394566e-01 -1.76654875e-01 -6.64918125e-01 -1.16481829e+00 -4.75718170e-01 3.28064322e-01 6.25509799e-01 -6.21241748e-01 -6.65976226e-01 -4.20523673e-01]

[8.140060424804688, 10.26241683959961]

56db144d-56d6-4cea-a414-719f323a2b80

dense-reconstruction-of-transparent-objects

2105.09993

https://arxiv.org/abs/2105.09993v1

https://arxiv.org/pdf/2105.09993v1.pdf

Dense Reconstruction of Transparent Objects by Altering Incident Light Paths Through Refraction

This paper addresses the problem of reconstructing the surface shape of transparent objects. The difficulty of this problem originates from the viewpoint dependent appearance of a transparent object, which quickly makes reconstruction methods tailored for diffuse surfaces fail disgracefully. In this paper, we introduce a fixed viewpoint approach to dense surface reconstruction of transparent objects based on refraction of light. We present a simple setup that allows us to alter the incident light paths before light rays enter the object by immersing the object partially in a liquid, and develop a method for recovering the object surface through reconstructing and triangulating such incident light paths. Our proposed approach does not need to model the complex interactions of light as it travels through the object, neither does it assume any parametric form for the object shape nor the exact number of refractions and reflections taken place along the light paths. It can therefore handle transparent objects with a relatively complex shape and structure, with unknown and inhomogeneous refractive index. We also show that for thin transparent objects, our proposed acquisition setup can be further simplified by adopting a single refraction approximation. Experimental results on both synthetic and real data demonstrate the feasibility and accuracy of our proposed approach.

['Miaomiao Liu', 'Kwan-Yee K. Wong', 'Kai Han']

2021-05-20

['transparent-objects']

['computer-vision']

[ 6.82020605e-01 1.06978863e-01 7.37542510e-01 -2.93675184e-01 3.92700061e-02 -3.73401850e-01 4.46666777e-01 -3.33448529e-01 -2.71314651e-01 5.08458316e-01 -4.85987842e-01 -9.91327465e-02 -3.57111916e-02 -9.87679899e-01 -5.70827365e-01 -9.55610812e-01 1.94732994e-01 9.31067526e-01 6.13359928e-01 -3.49003598e-02 1.96962491e-01 7.67600179e-01 -1.63866162e+00 1.42363489e-01 7.68922746e-01 7.17589200e-01 3.14246714e-01 7.10774243e-01 -4.37426478e-01 3.74862075e-01 -3.66454393e-01 -3.95270437e-01 4.41556215e-01 -3.15323263e-01 -5.86766183e-01 4.99991000e-01 3.72809649e-01 -6.74600780e-01 4.80676442e-02 9.67319608e-01 -3.87634672e-02 -1.12709686e-01 9.46670473e-01 -6.23594880e-01 -3.14978510e-01 7.58467067e-04 -6.37183785e-01 -2.62822926e-01 5.38561165e-01 -2.21962929e-01 3.70680064e-01 -1.08677816e+00 6.26682162e-01 1.03755474e+00 5.40547013e-01 6.57640874e-01 -8.60473871e-01 -2.08892003e-01 -4.52027842e-03 5.51308226e-03 -1.29577684e+00 -4.71894205e-01 9.84472334e-01 -4.89271164e-01 3.06343198e-01 6.01806641e-01 8.25463533e-01 3.09249967e-01 9.70869511e-02 2.63828605e-01 1.25074744e+00 -7.13166118e-01 2.44727716e-01 5.72290838e-01 3.99041146e-01 6.28590286e-01 6.21603191e-01 -8.00605938e-02 5.71221439e-03 -2.93435961e-01 9.11588132e-01 1.86881632e-01 -6.31436586e-01 -6.12606943e-01 -1.01766789e+00 2.64584094e-01 1.55372813e-01 3.03687394e-01 -2.50413001e-01 -2.63318598e-01 -3.75242144e-01 2.83878326e-01 5.61849535e-01 7.38078877e-02 -1.27252147e-01 2.26722911e-01 -6.80737317e-01 4.83820448e-03 1.15478027e+00 8.87407124e-01 8.92192185e-01 -1.06210925e-01 3.30385119e-01 4.99540329e-01 5.56706131e-01 8.56876016e-01 -3.23543549e-01 -8.29126835e-01 2.89486259e-01 4.35242534e-01 6.98119462e-01 -7.22257674e-01 -1.93755776e-01 -2.60370560e-02 -4.58563894e-01 8.07128787e-01 7.57277906e-01 2.14644205e-02 -8.32530558e-01 1.01980686e+00 9.05097425e-01 9.61367115e-02 1.01012014e-01 9.67092454e-01 8.16819429e-01 6.75179422e-01 -8.83214951e-01 -7.98733830e-01 1.14882350e+00 -6.72906756e-01 -7.67069280e-01 4.60463166e-01 1.55478595e-02 -1.01495206e+00 8.18517089e-01 7.86480725e-01 -1.58098733e+00 -1.31650463e-01 -8.11337292e-01 4.38571796e-02 2.17134282e-01 -7.76001662e-02 3.68244082e-01 6.80236280e-01 -8.35626662e-01 4.93935347e-01 -9.63743985e-01 -2.50590056e-01 -4.69710603e-02 5.21609068e-01 -1.51027188e-01 -2.39797816e-01 -3.43889773e-01 7.01885760e-01 -1.00411810e-01 4.82970148e-01 -3.83604914e-01 -5.95627725e-01 -1.72885299e-01 -1.93817109e-01 3.13510120e-01 -7.87290394e-01 9.86745358e-01 -1.10459554e+00 -2.09733915e+00 6.94414675e-01 -3.65700632e-01 3.15480232e-01 7.67110765e-01 -1.76725656e-01 -8.25545192e-02 4.80333507e-01 -5.96361220e-01 -5.40469646e-01 8.75453413e-01 -1.96426475e+00 7.46462941e-02 -5.45860350e-01 1.43197417e-01 2.56558746e-01 -2.78124735e-02 -3.53843533e-02 -3.72690231e-01 1.88762192e-02 6.26245439e-01 -9.64639843e-01 -7.42469132e-02 6.04536593e-01 -4.48454767e-01 3.68922979e-01 1.10620928e+00 -3.25232148e-01 5.01583159e-01 -1.89622343e+00 1.67778999e-01 4.43249792e-01 2.83434629e-01 3.61859530e-01 4.04977538e-02 7.77208686e-01 1.97806358e-01 -3.20473105e-01 -5.09371102e-01 -5.20992041e-01 -5.34064353e-01 2.54321605e-01 -1.72138259e-01 7.15405583e-01 -4.16552752e-01 2.76225269e-01 -6.80545390e-01 -2.47950420e-01 1.17693968e-01 9.51829493e-01 -6.83855772e-01 3.18188578e-01 -1.94985718e-01 8.66455019e-01 -7.66857147e-01 3.43038172e-01 1.23843992e+00 -4.15187776e-02 1.45789579e-01 -2.78709292e-01 -7.03163385e-01 1.50861278e-01 -1.56772828e+00 9.97696459e-01 -6.12922013e-01 1.66028813e-01 7.48992980e-01 -6.34805918e-01 9.40853000e-01 4.87282187e-01 4.85016078e-01 -3.79024416e-01 -1.83870010e-02 4.88974392e-01 -7.28872269e-02 -6.08880937e-01 2.10225835e-01 -6.82440042e-01 8.54070723e-01 5.72042286e-01 -6.78025603e-01 -4.67897922e-01 -2.21131459e-01 -1.59696057e-01 8.11977148e-01 1.70862496e-01 8.69811997e-02 -2.26232857e-01 5.73094487e-01 -3.04072052e-01 2.41070583e-01 2.72569001e-01 6.87985361e-01 7.27439046e-01 -2.29753241e-01 -5.91273487e-01 -8.52050245e-01 -1.29005039e+00 -5.26267529e-01 1.02738328e-01 6.26236439e-01 1.79998934e-01 -6.69878006e-01 4.67209145e-02 -1.67052045e-01 4.27913845e-01 -3.49621207e-01 2.86262184e-01 -7.46941328e-01 -7.84810126e-01 -4.02053833e-01 -3.00105482e-01 5.19601345e-01 -9.20641780e-01 -7.61104047e-01 1.66011453e-01 -2.37577245e-01 -1.07291043e+00 1.11214980e-01 -4.22751993e-01 -1.23988342e+00 -1.23505914e+00 -7.56201982e-01 -4.72150743e-01 1.13788426e+00 6.63177907e-01 1.08768237e+00 3.88207734e-01 -3.06809336e-01 8.63260508e-01 -2.15128899e-01 -3.59813571e-01 -4.77114767e-01 -6.15305841e-01 -2.26016957e-02 7.02846527e-01 -2.34678715e-01 -9.32667553e-01 -7.37117350e-01 5.12103379e-01 -9.58281696e-01 2.42353544e-01 1.23547800e-01 1.00512192e-01 5.10357380e-01 8.74426067e-02 -1.37705564e-01 -1.12055612e+00 -1.63330749e-01 -1.94791973e-01 -8.98559391e-01 1.98756516e-01 -2.08864257e-01 -2.62847811e-01 4.55678731e-01 -2.74985611e-01 -1.44648826e+00 -9.71352831e-02 -2.42404826e-02 -2.34846234e-01 -1.34181589e-01 4.53959331e-02 -1.98470607e-01 -4.95246291e-01 2.21616417e-01 2.02482820e-01 -1.44666865e-01 -9.33928430e-01 -1.58599064e-01 3.91852289e-01 6.67635351e-02 -4.55553740e-01 1.27445769e+00 1.31771219e+00 4.20169890e-01 -1.48903692e+00 -3.71915370e-01 -2.91776747e-01 -7.85982966e-01 -3.58902931e-01 6.44075632e-01 -3.36480916e-01 -1.08816123e+00 5.72683394e-01 -1.41498017e+00 -2.31505707e-01 -3.88161033e-01 7.11057603e-01 -3.64582270e-01 6.34106994e-01 -4.74840939e-01 -1.20424736e+00 -9.57891047e-02 -7.69016981e-01 9.52307582e-01 -6.25582710e-02 2.76095420e-01 -1.15931308e+00 2.97214776e-01 3.29342157e-01 4.75833356e-01 2.57362247e-01 9.10422385e-01 4.03355628e-01 -1.40292358e+00 -3.18502709e-02 -6.00817241e-02 3.07640042e-02 5.77008307e-01 4.30165976e-01 -1.05166423e+00 -2.81844229e-01 6.68736160e-01 1.05060428e-01 5.89794278e-01 3.32332581e-01 6.84297919e-01 -4.34101194e-01 -3.95732343e-01 9.08493280e-01 1.81607366e+00 1.28055230e-01 7.52730727e-01 -7.92612359e-02 9.32008624e-01 9.15045261e-01 5.00897527e-01 4.97938544e-01 2.35981554e-01 9.61865604e-01 9.47595656e-01 -2.30788931e-01 -1.69048831e-01 4.64977145e-01 2.16157705e-01 1.02775145e+00 -9.34890747e-01 -5.06512821e-01 -6.34843171e-01 1.25932992e-02 -1.10618770e+00 -9.04669881e-01 -1.06846225e+00 2.72398019e+00 6.31601810e-01 -2.44747162e-01 -2.15291709e-01 2.99937636e-01 2.26203695e-01 -4.19482142e-01 -2.04887271e-01 -3.08538973e-01 2.22310200e-01 1.47527128e-01 9.81111526e-02 1.31215644e+00 -3.07317317e-01 4.07695115e-01 5.84748983e+00 -2.81745405e-03 -1.28147125e+00 9.76735651e-02 -2.72986352e-01 -9.36203524e-02 -1.09694827e+00 1.17343478e-01 -7.66253889e-01 2.07390264e-01 3.25345367e-01 2.17349365e-01 3.90733778e-01 -7.20825791e-02 3.89079094e-01 -3.47306520e-01 -1.01289833e+00 5.37479579e-01 -1.83007523e-01 -7.44581819e-01 3.21225613e-01 4.99393046e-02 5.54818571e-01 -1.91763520e-01 2.87273787e-02 -6.59436822e-01 -2.12218434e-01 -4.53217685e-01 5.33590078e-01 8.54191065e-01 6.18084371e-01 -1.09838545e-01 2.53928304e-01 6.88195825e-01 -9.52794313e-01 5.05301952e-01 -3.56855482e-01 -9.96246189e-02 2.40421921e-01 1.00486159e+00 -7.95745432e-01 6.11253262e-01 3.80267143e-01 2.84067065e-01 1.03767030e-02 1.27652907e+00 1.47821814e-01 5.55593312e-01 -6.43695951e-01 1.34993568e-01 -2.81922847e-01 -9.73137975e-01 9.64433849e-01 8.02921951e-01 5.07689953e-01 6.67222500e-01 9.40401014e-03 9.93492901e-01 4.20385182e-01 1.76895797e-01 -5.93076766e-01 4.60598260e-01 -1.63435087e-01 1.17041624e+00 -9.21321332e-01 -2.28765123e-02 -5.64572871e-01 9.26607728e-01 -7.82027692e-02 8.33725631e-01 -5.48401117e-01 5.63906729e-02 2.51461655e-01 8.11125338e-01 2.14127243e-01 -3.99649709e-01 -1.00920849e-01 -1.39526320e+00 5.20594656e-01 -2.63753712e-01 -2.55837768e-01 -8.06239963e-01 -8.96666110e-01 6.95583105e-01 9.23872292e-02 -1.44229746e+00 3.02387357e-01 -6.20838106e-01 -6.43605053e-01 9.59020019e-01 -1.95042634e+00 -1.07786858e+00 -4.17087108e-01 9.54934597e-01 2.70890892e-01 6.44067824e-01 8.30487430e-01 7.05015436e-02 1.71147529e-02 -2.14496642e-01 2.38210410e-01 -5.64845383e-01 2.96873808e-01 -9.47310507e-01 -2.96221763e-01 5.83151579e-01 -2.13619798e-01 6.83466971e-01 9.90753114e-01 -5.10724545e-01 -1.69434226e+00 -5.16129375e-01 6.05236828e-01 -3.80533874e-01 1.73801675e-01 -5.72395086e-01 -1.20076120e+00 6.31275892e-01 1.19431473e-01 6.26785159e-02 6.12973988e-01 -2.69564062e-01 -1.22861318e-01 -1.68549404e-01 -1.30512965e+00 2.65100837e-01 9.85224366e-01 -9.94009003e-02 -4.49000150e-01 5.97733974e-01 2.13616177e-01 -4.42193151e-01 -4.77137506e-01 3.89201671e-01 6.42582595e-01 -1.43432379e+00 9.29412246e-01 3.97237763e-02 4.58267927e-02 -3.06172967e-01 6.32800013e-02 -9.25564051e-01 -1.22055374e-01 -6.57277405e-01 1.68897986e-01 9.78044271e-01 2.32781053e-01 -1.03556347e+00 8.23403835e-01 7.20568717e-01 -2.79151827e-01 -4.70256358e-01 -8.28070283e-01 -4.25899297e-01 -1.91519737e-01 -3.99479792e-02 6.33091480e-02 7.66348720e-01 -4.93670195e-01 -1.11468300e-01 -7.22345263e-02 7.96108782e-01 1.16757965e+00 6.82374358e-01 7.60355234e-01 -1.76138258e+00 -4.99521762e-01 2.28218883e-01 3.66616249e-03 -1.08191574e+00 -2.22451210e-01 -5.08767307e-01 -1.60661992e-02 -1.56713498e+00 2.83787757e-01 -9.64562058e-01 4.22176719e-01 -7.97388852e-02 7.66568854e-02 2.82355189e-01 -8.84347707e-02 5.48024654e-01 -8.86477344e-03 5.59095979e-01 1.86191738e+00 2.86135256e-01 -4.23946351e-01 6.77635908e-01 -3.12459678e-03 1.18430555e+00 3.64347130e-01 -4.25094157e-01 -6.27554774e-01 -7.81328261e-01 4.07510519e-01 2.28341267e-01 4.77278799e-01 -7.88151443e-01 -8.98438990e-02 -2.29637057e-01 -6.25836030e-02 -4.54603165e-01 8.67604673e-01 -1.61086380e+00 6.04811430e-01 3.26659828e-01 3.20332706e-01 -3.61366957e-01 -1.27658080e-02 5.24951398e-01 1.60402149e-01 -6.43804312e-01 1.07165933e+00 -2.44006678e-01 1.24354541e-01 2.38838717e-01 -3.64053071e-01 -3.35903287e-01 9.82093036e-01 -5.83774686e-01 -2.14767084e-02 -2.33178005e-01 -8.45390439e-01 -3.52258563e-01 1.03804755e+00 -2.31056243e-01 8.61007035e-01 -9.09856856e-01 -5.48507273e-01 4.62741584e-01 -2.25097179e-01 1.43605590e-01 1.08186193e-01 8.74310613e-01 -1.03336918e+00 -5.09167686e-02 7.76422769e-02 -6.49018526e-01 -1.60263431e+00 2.95227796e-01 6.08432651e-01 1.94116011e-01 -1.00213742e+00 7.03893244e-01 6.84423327e-01 -3.18744332e-01 -2.38568306e-01 -4.89336848e-01 -1.43329456e-01 -3.64120096e-01 5.38078666e-01 4.22959715e-01 1.70423046e-01 -5.31564295e-01 -8.54734629e-02 1.37421310e+00 4.28422689e-02 -1.18300140e-01 1.47651744e+00 -2.85341918e-01 -4.56973135e-01 7.70419180e-01 8.84935737e-01 7.11752295e-01 -1.08782470e+00 -2.55794436e-01 -7.04270840e-01 -6.68103993e-01 -9.28277299e-02 -3.79913002e-01 -1.15183175e+00 1.07873821e+00 6.63865507e-02 3.48200589e-01 9.77786839e-01 -8.15862603e-03 6.32618845e-01 3.89795333e-01 8.39611173e-01 -4.18895692e-01 -6.27249405e-02 1.66349202e-01 8.68247509e-01 -7.85309553e-01 8.42887387e-02 -1.34716201e+00 2.37867549e-01 1.30767739e+00 3.84987652e-01 -4.10507657e-02 9.94267583e-01 4.17611867e-01 2.04058051e-01 -4.15019721e-01 -4.06952500e-01 1.92291364e-01 1.26822606e-01 4.45436090e-01 4.22530413e-01 -2.26692587e-01 -2.89693385e-01 -3.76545817e-01 2.07531020e-01 7.73608834e-02 1.01340771e+00 9.53604877e-01 -5.89042902e-01 -1.16198993e+00 -7.56850123e-01 3.89815457e-02 -3.35419089e-01 2.29344913e-03 -2.41994515e-01 7.84166157e-01 1.66725572e-02 6.44047439e-01 6.39784485e-02 4.14520890e-01 4.76262510e-01 -2.91867644e-01 1.00021899e+00 -9.77254212e-01 -5.59351593e-02 1.81069553e-01 -8.02261084e-02 -2.63695210e-01 -9.98810470e-01 -8.15495074e-01 -1.43271816e+00 3.33629102e-02 -4.42193985e-01 3.35423142e-01 7.58372486e-01 7.73455679e-01 -1.32990852e-01 5.89733683e-02 7.95618236e-01 -1.12815225e+00 -2.92205244e-01 -5.62914789e-01 -9.73385513e-01 2.14591131e-01 7.07682669e-01 -7.40865648e-01 -7.16096401e-01 1.22377530e-01]

[9.674100875854492, -2.996441602706909]

8937aeca-29f4-453b-bc80-bd66797d985a

private-eye-on-the-limits-of-textual-screen

2205.03971

https://arxiv.org/abs/2205.03971v3

https://arxiv.org/pdf/2205.03971v3.pdf

Private Eye: On the Limits of Textual Screen Peeking via Eyeglass Reflections in Video Conferencing

Using mathematical modeling and human subjects experiments, this research explores the extent to which emerging webcams might leak recognizable textual and graphical information gleaming from eyeglass reflections captured by webcams. The primary goal of our work is to measure, compute, and predict the factors, limits, and thresholds of recognizability as webcam technology evolves in the future. Our work explores and characterizes the viable threat models based on optical attacks using multi-frame super resolution techniques on sequences of video frames. Our models and experimental results in a controlled lab setting show it is possible to reconstruct and recognize with over 75% accuracy on-screen texts that have heights as small as 10 mm with a 720p webcam. We further apply this threat model to web textual contents with varying attacker capabilities to find thresholds at which text becomes recognizable. Our user study with 20 participants suggests present-day 720p webcams are sufficient for adversaries to reconstruct textual content on big-font websites. Our models further show that the evolution towards 4K cameras will tip the threshold of text leakage to reconstruction of most header texts on popular websites. Besides textual targets, a case study on recognizing a closed-world dataset of Alexa top 100 websites with 720p webcams shows a maximum recognition accuracy of 94% with 10 participants even without using machine-learning models. Our research proposes near-term mitigations including a software prototype that users can use to blur the eyeglass areas of their video streams. For possible long-term defenses, we advocate an individual reflection testing procedure to assess threats under various settings, and justify the importance of following the principle of least privilege for privacy-sensitive scenarios.

['Kevin Fu', 'Shilin Xiao', 'Wenyuan Xu', 'Shivan Prasad', 'Chen Yan', 'Yan Long']

2022-05-08

['multi-frame-super-resolution']

['computer-vision']

[ 3.39568704e-01 2.84942426e-02 1.02154307e-01 1.39417455e-01 -9.64737058e-01 -1.34288597e+00 5.85917294e-01 -3.89567524e-01 -2.60601521e-01 8.21167529e-02 6.57890365e-02 -9.49887276e-01 8.57994184e-02 -4.04548705e-01 -8.16135466e-01 -1.93432853e-01 1.77311450e-01 -6.26024663e-01 4.24058735e-01 8.60787705e-02 6.30542397e-01 5.86724758e-01 -1.36009347e+00 7.05772579e-01 1.37065351e-01 9.76707578e-01 -3.98188531e-01 1.16353822e+00 3.93735528e-01 8.14177573e-01 -9.77482378e-01 -8.91164541e-01 8.50732744e-01 1.47060558e-01 -3.77150059e-01 8.09291974e-02 1.10794449e+00 -1.15198457e+00 -7.96841502e-01 9.42465365e-01 4.60848212e-01 -2.63257056e-01 -6.37945011e-02 -1.20083058e+00 -7.00008750e-01 -1.38177918e-02 -5.33592284e-01 4.04677749e-01 1.05849969e+00 5.32462239e-01 4.93504137e-01 -2.71589756e-01 2.46123552e-01 9.19676840e-01 7.85110950e-01 4.76963848e-01 -1.20231974e+00 -1.10655642e+00 -3.03727359e-01 -1.55387759e-01 -1.45265794e+00 -6.51036322e-01 4.71783608e-01 -2.08729267e-01 9.12566662e-01 7.95693815e-01 3.02419186e-01 1.80423629e+00 2.91814774e-01 8.73787403e-02 1.33274543e+00 -4.71974939e-01 -1.17581189e-02 8.61178696e-01 3.40947360e-01 7.75269806e-01 6.17786229e-01 7.32595250e-02 -8.29685986e-01 -7.47696042e-01 8.30796480e-01 -3.51282090e-01 -5.67931771e-01 1.92874983e-01 -5.34693122e-01 5.55238008e-01 -6.09107137e-01 -7.10922247e-03 5.98620251e-02 -3.08866054e-01 3.46139148e-02 3.37107986e-01 7.00942799e-02 6.58518136e-01 -2.68475085e-01 -3.89688700e-01 -7.42773116e-01 -1.99319333e-01 1.01704133e+00 1.09118378e+00 1.69084281e-01 -2.02219069e-01 3.14907014e-01 3.15888941e-01 2.53301084e-01 5.61075449e-01 2.73163080e-01 -1.08694971e+00 7.82630324e-01 1.59372076e-01 5.84278524e-01 -1.34048116e+00 2.21201152e-01 2.89019555e-01 4.81224328e-01 4.40750360e-01 8.71096253e-01 -3.56904477e-01 -2.32630804e-01 8.91204357e-01 -1.20991319e-01 3.61631602e-01 -2.75542319e-01 6.52753949e-01 3.71060483e-02 5.40640593e-01 7.87688792e-02 -2.02134460e-01 1.60942280e+00 -5.04889116e-02 -4.23796713e-01 4.53004278e-02 4.03347224e-01 -1.06018782e+00 1.48454082e+00 9.80600297e-01 -8.22757721e-01 -3.28907907e-01 -1.29313421e+00 2.76314497e-01 -4.22730088e-01 -2.34886050e-01 2.41542771e-01 2.02767038e+00 -8.99424613e-01 1.36886597e-01 -7.13773251e-01 -7.45822430e-01 2.18017265e-01 4.19640541e-01 -2.89054722e-01 1.96894199e-01 -9.74678874e-01 8.37847173e-01 -3.13576490e-01 -3.35594803e-01 -6.71529830e-01 -6.61531508e-01 -2.54053026e-01 1.09081380e-02 2.85371274e-01 -2.44167484e-02 9.01831448e-01 -7.32384145e-01 -1.48010015e+00 9.13860261e-01 3.07552040e-01 -5.05399644e-01 5.59125125e-01 -3.97948861e-01 -9.59186792e-01 5.75064123e-01 -4.97178555e-01 -1.14018060e-01 1.15049684e+00 -1.19638789e+00 -6.43239379e-01 -4.26790684e-01 3.99217725e-01 -2.69002635e-02 -1.34202504e+00 5.39208472e-01 -1.33844584e-01 -1.71506673e-01 -6.13394022e-01 -9.77030814e-01 4.53276634e-01 -3.96391526e-02 -3.27946037e-01 5.82122922e-01 1.46618485e+00 -1.18866658e+00 1.18007278e+00 -2.00594926e+00 -9.22779322e-01 3.48195285e-01 2.06562743e-01 6.01676464e-01 1.44141972e-01 2.37709135e-01 9.07436460e-02 7.10737169e-01 7.52199233e-01 2.86352426e-01 -7.22951815e-02 -5.69352210e-01 -7.91198730e-01 5.78082025e-01 -3.64906371e-01 2.90781885e-01 -1.98088929e-01 -1.56681284e-01 4.00892824e-01 6.33012116e-01 -4.02813971e-01 1.09356128e-01 2.30675176e-01 -1.68553546e-01 -3.72936994e-01 7.01056123e-01 7.35535264e-01 -2.39552627e-03 3.14474195e-01 -1.73645511e-01 -2.35544011e-01 6.20554872e-02 -8.90220165e-01 9.14413929e-01 -3.56478930e-01 1.05554831e+00 2.64080286e-01 -3.34370062e-02 6.93280280e-01 1.44536972e-01 1.85734794e-01 -5.65133095e-01 1.61726013e-01 -2.44159579e-01 -5.65747917e-01 -8.06073904e-01 4.79747266e-01 4.96682882e-01 1.13780841e-01 3.98391366e-01 -4.16568190e-01 2.44454309e-01 -4.68592703e-01 2.02115566e-01 1.61417329e+00 -4.52075005e-02 -2.03601450e-01 -1.16778076e-01 3.62977892e-01 -1.83038995e-01 -4.47376817e-01 1.05082381e+00 -5.50658107e-01 5.69708049e-01 5.47256768e-01 -4.43337351e-01 -9.42867577e-01 -9.32218671e-01 -1.06952623e-01 9.61097062e-01 3.00639242e-01 -5.89655340e-01 -1.17697310e+00 -7.05986321e-01 -2.21447334e-01 1.12379360e+00 -1.60456106e-01 -1.13016345e-01 -4.08543527e-01 -7.40187228e-01 1.22184634e+00 -6.88591450e-02 5.13893366e-01 -2.97056079e-01 -8.82400393e-01 -1.60469025e-01 -1.48709670e-01 -1.70865822e+00 -3.98370564e-01 -4.61132795e-01 -5.05858004e-01 -1.50056219e+00 -2.26825371e-01 -1.67900115e-01 4.18113440e-01 7.54015028e-01 2.56279975e-01 -7.27211460e-02 -4.02305096e-01 1.40748119e+00 -3.79377633e-01 -2.92090833e-01 -4.17821407e-01 -4.73380536e-01 6.41261339e-02 1.62689179e-01 8.76953006e-01 -3.38708967e-01 -4.99791533e-01 6.58058584e-01 -8.15775931e-01 -5.79011083e-01 1.30728021e-01 -1.03086255e-01 -1.66926250e-01 5.16676545e-01 -3.60182732e-01 -5.39874554e-01 8.70173872e-01 -4.56836194e-01 -8.42051685e-01 4.30564940e-01 -5.60944915e-01 -5.76807618e-01 6.68123484e-01 -9.42002773e-01 -1.16270149e+00 -1.56838253e-01 4.29567635e-01 -2.81686395e-01 -2.99117863e-01 -4.38929141e-01 -2.37647802e-01 -7.94208050e-01 1.08946562e+00 1.82629451e-01 -1.48816705e-02 -1.27233803e-01 -8.33258871e-03 1.02769494e+00 2.74918199e-01 -5.83816350e-01 1.07868540e+00 7.42364943e-01 -4.92875785e-01 -1.47723150e+00 -2.85695374e-01 -2.69461721e-01 -4.84010130e-02 -4.49742258e-01 6.51984155e-01 -8.35230708e-01 -1.33339787e+00 6.41130328e-01 -9.12280738e-01 -1.70911491e-01 6.00373209e-01 2.17722535e-01 -1.50199071e-01 9.90191758e-01 -7.27755785e-01 -1.13896966e+00 -7.83243030e-02 -8.57549429e-01 9.16473031e-01 2.02953771e-01 -2.63934612e-01 -8.35771918e-01 -3.69006693e-01 1.21501768e+00 3.91604096e-01 1.59638733e-01 4.40933824e-01 -8.38061333e-01 -7.77544975e-01 -8.17784607e-01 -1.57091945e-01 2.44217783e-01 -3.76215763e-02 2.31829315e-01 -1.15730441e+00 -4.02200013e-01 5.57685614e-01 -1.22067831e-01 -1.98784545e-02 -1.03638880e-01 1.19198632e+00 -7.55034208e-01 -1.13203853e-01 5.35099030e-01 1.48168623e+00 3.08953017e-01 1.12368274e+00 8.04389834e-01 6.11021578e-01 5.31080008e-01 2.11800620e-01 5.53407848e-01 -1.34724155e-01 6.26244426e-01 4.02732700e-01 5.69516957e-01 2.56943524e-01 -3.13781410e-01 9.78163064e-01 -5.40958457e-02 1.43883705e-01 -5.28298080e-01 -8.00215960e-01 -3.34695905e-01 -8.58421624e-01 -1.17808771e+00 -2.89180428e-01 2.72311091e+00 2.44171187e-01 6.66509926e-01 2.56418407e-01 -1.55855834e-01 1.20462060e+00 1.09655708e-01 -1.66514561e-01 -4.43885952e-01 9.15605947e-02 1.17930442e-01 1.23365319e+00 5.64762831e-01 -1.05553663e+00 6.31447613e-01 6.62151575e+00 4.96646315e-01 -9.94916916e-01 -3.08435485e-02 7.73666859e-01 -4.71272945e-01 7.07784146e-02 4.10920046e-02 -1.10304058e+00 6.86440289e-01 1.64040542e+00 1.25415206e-01 7.28420436e-01 6.45749629e-01 3.54969174e-01 -1.26273194e-02 -8.45967948e-01 9.03668582e-01 6.03376508e-01 -1.07836211e+00 7.76281506e-02 6.06892705e-01 5.76324128e-02 -5.41580677e-01 5.67613363e-01 -1.86591577e-02 2.11328432e-01 -8.01732063e-01 4.51956123e-01 1.77197024e-01 9.21923041e-01 -5.55980742e-01 1.49961397e-01 1.33154184e-01 -6.61583066e-01 -4.03911978e-01 -2.91863382e-01 3.94502804e-02 -2.99810022e-01 -1.63274825e-01 -8.77100468e-01 -1.62364125e-01 7.81254768e-01 -1.49948031e-01 -9.48012650e-01 5.87577581e-01 1.18620442e-02 1.04649055e+00 -5.44354439e-01 -3.82998362e-02 -2.57751960e-02 7.36956596e-02 6.07861936e-01 1.09159195e+00 4.42009926e-01 3.76684099e-01 -4.15445954e-01 5.97828925e-01 2.96260327e-01 -1.44111872e-01 -7.34732747e-01 -6.09892681e-02 5.77929854e-01 1.17002916e+00 -6.62768126e-01 2.50876069e-01 -7.86168754e-01 1.20181811e+00 -3.38836312e-01 4.93791312e-01 -1.08179343e+00 -3.90934318e-01 5.57601273e-01 7.71375775e-01 -1.69097949e-02 -3.18885863e-01 -4.15899962e-01 -1.19126201e+00 2.01399252e-01 -1.37410712e+00 3.67500275e-01 -9.67873216e-01 -8.86229813e-01 1.76029697e-01 -3.11939195e-02 -1.11685419e+00 1.63811922e-01 -1.06159401e+00 -3.99976969e-01 6.15370989e-01 -7.81385899e-01 -1.16502464e+00 -4.39183176e-01 7.75046647e-01 4.33411807e-01 -3.68615746e-01 7.23302484e-01 3.42327319e-02 -4.43123549e-01 9.20561850e-01 5.19818719e-03 3.84669065e-01 7.99566805e-01 -7.45566368e-01 4.85669017e-01 1.19374359e+00 -1.49196177e-03 1.21262145e+00 7.53852010e-01 -7.83892930e-01 -1.87400496e+00 -4.66609061e-01 1.55391827e-01 -1.36946750e+00 9.24808741e-01 -5.88188767e-01 -7.05385566e-01 8.49333644e-01 2.28445739e-01 -3.80456567e-01 9.64022279e-01 -1.89488724e-01 -8.98169339e-01 -7.73485750e-02 -1.36571074e+00 9.58327770e-01 5.99553108e-01 -9.92121577e-01 -2.49904469e-01 3.39195013e-01 5.43410242e-01 -9.69027206e-02 -7.04353392e-01 -3.44749391e-01 1.14732587e+00 -1.14873576e+00 9.93042469e-01 -3.54517460e-01 -5.25272079e-02 1.24403082e-01 -2.66889870e-01 -3.07049811e-01 1.35641485e-01 -1.42464006e+00 -2.05766648e-01 1.39544141e+00 1.68575034e-01 -7.15757549e-01 1.13647306e+00 1.65413964e+00 5.74949205e-01 1.61779061e-01 -6.85386956e-01 -6.59291446e-01 -2.38791883e-01 -6.16500556e-01 1.20828360e-01 9.38027084e-01 2.61321396e-01 -2.55616963e-01 -6.38041914e-01 9.72385585e-01 9.95722473e-01 -9.38730478e-01 9.08331156e-01 -7.23964691e-01 -3.63969892e-01 -1.54025927e-01 -5.64984560e-01 -7.47107625e-01 -6.34007379e-02 -2.48331100e-01 -8.56104016e-01 -3.67927909e-01 2.64982790e-01 3.48923862e-01 -2.74973884e-02 2.08091602e-01 2.29084447e-01 1.45987675e-01 4.17196155e-01 2.58482069e-01 -4.44228679e-01 -5.51775575e-01 2.18383402e-01 -1.33467233e-03 -1.31885409e-01 -2.33045481e-02 -9.45027173e-01 6.47346616e-01 7.15697944e-01 -2.22401902e-01 -5.61303139e-01 -8.01145583e-02 1.51920661e-01 1.04290493e-01 7.59057403e-01 -1.20226848e+00 2.26396188e-01 -1.16028473e-01 6.69767141e-01 8.90133604e-02 3.52008253e-01 -1.38331878e+00 8.88292193e-02 3.31458300e-01 -4.35052365e-01 -2.99431849e-02 4.79910403e-01 6.35784984e-01 6.36083543e-01 -2.50849873e-01 5.92922509e-01 -1.76710904e-01 -5.01185417e-01 -2.85450190e-01 -9.27507758e-01 -2.43811652e-01 1.17995679e+00 -7.77636111e-01 -9.32461262e-01 -6.39219224e-01 -3.00332487e-01 -3.43938351e-01 8.18557918e-01 5.89357913e-01 7.21635640e-01 -5.83663881e-01 -1.58983573e-01 2.93323398e-01 -2.01349229e-01 -1.16320002e+00 7.56308734e-02 8.25217292e-02 -9.38068449e-01 4.37871546e-01 -2.22631171e-01 -1.30630091e-01 -1.77235925e+00 6.78888083e-01 3.69333029e-01 1.98459655e-01 -5.07707119e-01 4.90127474e-01 -8.07558745e-02 3.14027280e-01 3.56542587e-01 4.03848737e-01 1.41267672e-01 -3.74488920e-01 1.11052108e+00 7.05351412e-01 8.47783089e-02 -3.47115159e-01 -3.12893271e-01 2.67043442e-01 -2.62566864e-01 -3.38708758e-01 7.75427461e-01 -4.24014837e-01 3.35363388e-01 -2.22204849e-01 1.28071713e+00 6.54680490e-01 -1.39393258e+00 3.56481254e-01 -5.33074327e-02 -1.04716814e+00 -6.18626699e-02 -9.12572980e-01 -5.80107749e-01 4.51100290e-01 8.64121139e-01 4.76958543e-01 9.03629482e-01 -4.75856900e-01 9.36772883e-01 4.06266063e-01 6.09063745e-01 -1.03825152e+00 4.44871902e-01 -2.56268829e-01 3.50621611e-01 -8.79136503e-01 2.28271216e-01 -4.91063774e-01 -4.33694422e-01 1.41965806e+00 7.88034141e-01 -4.00126651e-02 4.50656146e-01 6.53609395e-01 -4.71619517e-02 -5.95027320e-02 -4.61781353e-01 7.14419127e-01 -2.30712622e-01 8.03912103e-01 -1.66297499e-02 -1.64293349e-01 2.77098473e-02 5.47478020e-01 -2.18263507e-01 -2.86969602e-01 1.19726610e+00 9.02255714e-01 -3.75249952e-01 -7.67476439e-01 -1.11323988e+00 4.12961245e-01 -1.08144116e+00 -1.16906047e-01 -7.64584720e-01 9.14911866e-01 -2.98834503e-01 1.65780747e+00 -2.63356686e-01 -8.09113085e-01 3.03279459e-01 2.33079359e-01 3.20453644e-01 -1.69869468e-01 -6.08368337e-01 -7.70161897e-02 2.80349880e-01 -7.26267457e-01 1.55486152e-01 -9.91449237e-01 -4.35599297e-01 -7.31745243e-01 -3.33194703e-01 -2.61960238e-01 7.04433024e-01 3.86234254e-01 3.57003301e-01 -2.22836822e-01 6.00142419e-01 -3.36865753e-01 -8.52584481e-01 -4.21060592e-01 -4.14920747e-01 4.46502149e-01 2.07849607e-01 -9.87626687e-02 -8.84461522e-01 2.79107392e-01]

[12.574363708496094, 1.1008154153823853]

d2e5f83e-7822-434d-b20b-efd639566504

efficient-image-retargeting-for-high-dynamic

1305.4544

http://arxiv.org/abs/1305.4544v1

http://arxiv.org/pdf/1305.4544v1.pdf

Efficient Image Retargeting for High Dynamic Range Scenes

Most of the real world scenes have a very high dynamic range (HDR). The mobile phone cameras and the digital cameras available in markets are limited in their capability in both the range and spatial resolution. Same argument can be posed about the limited dynamic range display devices which also differ in the spatial resolution and aspect ratios. In this paper, we address the problem of displaying the high contrast low dynamic range (LDR) image of a HDR scene in a display device which has different spatial resolution compared to that of the capturing digital camera. The optimal solution proposed in this work can be employed with any camera which has the ability to shoot multiple differently exposed images of a scene. Further, the proposed solutions provide the flexibility in the depiction of entire contrast of the HDR scene as a LDR image with an user specified spatial resolution. This task is achieved through an optimized content aware retargeting framework which preserves salient features along with the algorithm to combine multi-exposure images. We show the proposed approach performs exceedingly well in the generation of high contrast LDR image of varying spatial resolution compared to an alternate approach.

['Shanmuganathan Raman', 'Puneet Sharma', 'Govind Salvi']

2013-05-20

['image-retargeting']

['computer-vision']

[ 7.92787194e-01 -2.47649074e-01 3.18174362e-01 2.47028545e-02 -4.26982254e-01 -7.22969115e-01 5.23194015e-01 -3.48294199e-01 -3.35590541e-01 6.90930367e-01 5.27936742e-02 -2.46266991e-01 -2.34972328e-01 -8.11097205e-01 -4.21858519e-01 -8.10502291e-01 3.31106395e-01 -5.29596396e-02 7.00336576e-01 -4.06283379e-01 3.68988097e-01 7.37361312e-01 -2.05767608e+00 1.11224994e-01 7.83021212e-01 6.57987237e-01 1.07128048e+00 9.97359455e-01 -6.83329580e-03 6.75178468e-01 -5.30458808e-01 6.29080608e-02 4.39897180e-01 -5.67361295e-01 -3.88404936e-01 4.64446276e-01 5.50361872e-01 -4.88871396e-01 -3.43436971e-02 1.08512223e+00 5.37003160e-01 1.97070956e-01 4.29918408e-01 -7.91067123e-01 -5.18183589e-01 -4.44024289e-03 -1.07168555e+00 6.29399955e-01 8.27817380e-01 -3.82107273e-02 2.97286719e-01 -6.26846254e-01 7.44513273e-01 1.02069008e+00 -7.41895009e-03 3.18831801e-01 -1.24718666e+00 -3.69114310e-01 -1.27127036e-01 9.79002845e-03 -1.35068381e+00 -3.85435760e-01 9.51374173e-01 -2.52600640e-01 6.88115120e-01 5.84108710e-01 5.04076183e-01 4.57659513e-01 3.25620979e-01 -2.04169542e-01 1.60955966e+00 -5.47806144e-01 6.34140521e-02 4.89585608e-01 -2.02415019e-01 2.97027737e-01 2.43278712e-01 2.10860670e-01 -3.60000819e-01 2.57638395e-01 1.01868522e+00 1.00959986e-01 -6.85475826e-01 -3.09163600e-01 -9.05898690e-01 3.18882346e-01 7.68062174e-02 6.51543081e-01 -1.76345125e-01 -3.83662403e-01 -8.33359640e-03 4.86417234e-01 2.43495941e-01 5.61616421e-01 -1.40252486e-01 1.13863185e-01 -7.87487805e-01 2.59788223e-02 4.40046072e-01 7.91198611e-01 5.58771908e-01 4.04448099e-02 3.71384509e-02 6.64743781e-01 -3.02289929e-02 5.58758020e-01 3.14447731e-01 -9.47680771e-01 4.33567852e-01 3.61181229e-01 2.53282666e-01 -9.26180065e-01 -1.80105567e-01 -2.18516842e-01 -8.35936129e-01 9.88840461e-01 2.71569133e-01 -6.95246458e-02 -6.07750535e-01 1.36764634e+00 4.71065879e-01 -1.05371311e-01 2.61519343e-01 1.06019425e+00 5.09754479e-01 1.01142740e+00 -2.04188406e-01 -5.89738905e-01 1.32020295e+00 -4.65134680e-01 -9.62441683e-01 -1.12765022e-01 -9.09471661e-02 -1.16512990e+00 1.21405351e+00 4.60475534e-01 -1.38639057e+00 -7.28060186e-01 -1.35205078e+00 -9.07909498e-02 -2.47354671e-01 -2.37747170e-02 -5.90202920e-02 8.93671751e-01 -1.17250562e+00 1.99184299e-01 -4.01660614e-03 -4.38420117e-01 -2.53486931e-01 2.98252225e-01 -3.24168056e-01 -3.01957548e-01 -8.07834685e-01 1.09935510e+00 5.84422767e-01 -2.02459782e-01 -2.89906472e-01 -7.19829500e-01 -5.09840786e-01 2.00597957e-01 3.87335420e-01 -3.58438849e-01 5.70974350e-01 -1.30309296e+00 -1.68433523e+00 1.04000199e+00 1.34122878e-01 -5.33762090e-02 7.04443574e-01 9.34680775e-02 -6.25934660e-01 3.63903135e-01 -1.77355170e-01 1.00617848e-01 9.46977079e-01 -1.44847977e+00 -8.97748590e-01 -2.90152937e-01 2.07151651e-01 7.54476666e-01 7.16545135e-02 2.48693004e-01 -2.96807468e-01 -4.01094496e-01 -7.07297847e-02 -6.06116951e-01 1.31893650e-01 -9.48402882e-02 -1.00512989e-01 6.34340227e-01 1.47655869e+00 -5.74168980e-01 1.21180499e+00 -2.32216859e+00 3.01993228e-02 1.20521091e-01 9.68812257e-02 3.78833115e-01 2.02131838e-01 3.85600507e-01 -1.46005452e-01 -1.79036021e-01 -1.52182311e-01 2.17255279e-01 -5.30655265e-01 -2.58530498e-01 -2.51747340e-01 5.57150424e-01 -2.27564827e-01 2.72558838e-01 -5.94206691e-01 -6.47387683e-01 7.10012913e-01 6.20118558e-01 -9.99506488e-02 4.83528107e-01 1.18681088e-01 7.97038376e-01 -1.74038321e-01 2.44447947e-01 1.21018744e+00 1.37031907e-02 3.86705011e-01 -3.55579227e-01 -6.38269603e-01 -5.90031624e-01 -1.48743272e+00 1.18413937e+00 -5.09823084e-01 8.91485214e-01 1.76224485e-01 -3.13423336e-01 1.09206653e+00 1.36481747e-01 3.38113666e-01 -1.17214179e+00 -1.36640742e-01 1.46340162e-01 -2.25958720e-01 -5.04266143e-01 9.97243404e-01 -3.28557760e-01 1.65835500e-01 3.66546422e-01 -5.59282482e-01 -2.50614822e-01 1.18762523e-01 -8.56801122e-02 6.32837713e-01 -1.40768662e-02 5.56148529e-01 -2.98480898e-01 7.49937356e-01 -3.95370543e-01 -4.07938473e-02 5.36430418e-01 2.10099176e-01 8.44254315e-01 2.53977954e-01 1.16818305e-02 -1.59610140e+00 -1.19263697e+00 -2.91014135e-01 8.49013925e-01 6.92130387e-01 2.25097030e-01 -4.58930135e-01 1.42445028e-01 -5.29458165e-01 5.73158622e-01 -3.53004456e-01 2.23585919e-01 -7.25483596e-01 -4.46523041e-01 -1.30621985e-01 -1.36449367e-01 9.48897898e-01 -7.74158776e-01 -1.37644839e+00 -3.49988081e-02 6.58202544e-02 -1.09146535e+00 -4.56947833e-01 1.87205132e-02 -6.97043061e-01 -1.05813718e+00 -8.25498879e-01 -8.52943838e-01 4.89753544e-01 7.13483632e-01 9.74764526e-01 -1.76546261e-01 -3.55754942e-01 3.98048371e-01 -5.33687472e-01 3.15370075e-02 -5.13949394e-01 -2.65155613e-01 -3.11782360e-01 8.47830027e-02 -1.58022925e-01 -5.09018362e-01 -7.96547592e-01 4.20846224e-01 -1.49366522e+00 5.39003909e-01 3.90838027e-01 3.63448024e-01 5.50244391e-01 6.12887561e-01 2.90735871e-01 -8.22435141e-01 7.38202572e-01 -2.98182428e-01 -7.96495616e-01 3.45595509e-01 -4.93229628e-01 -1.10883266e-01 7.82340646e-01 -4.74239707e-01 -1.66085553e+00 -2.21709833e-02 1.19386822e-01 1.80567466e-02 -1.67240843e-01 -2.70008028e-01 -4.02940720e-01 -1.92585379e-01 4.75867152e-01 3.80071342e-01 -8.15506503e-02 -4.60353523e-01 3.28067482e-01 7.64847457e-01 5.45285285e-01 -1.23028412e-01 7.95194268e-01 5.79084992e-01 2.81430751e-01 -9.91583109e-01 -2.52927840e-01 -2.36634016e-01 -4.94415373e-01 -4.51044947e-01 1.01872206e+00 -6.10588372e-01 -5.04339397e-01 6.00340724e-01 -8.56348634e-01 -1.22318864e-01 -2.20363468e-01 2.28623390e-01 -6.35740161e-01 4.01551008e-01 -2.94851154e-01 -7.79830039e-01 -1.07539810e-01 -1.00118721e+00 9.29524899e-01 6.24936938e-01 2.31910259e-01 -8.48136842e-01 1.97579879e-02 4.56142761e-02 6.49941802e-01 4.77332056e-01 9.78717744e-01 1.95345208e-01 -8.79793584e-01 3.58260155e-01 -5.10867894e-01 1.42736569e-01 3.80989254e-01 -1.08905114e-01 -9.68123317e-01 -3.78915936e-01 2.93135792e-01 1.63319424e-01 3.82507384e-01 6.49211347e-01 7.95217633e-01 -1.01018220e-01 -3.76981637e-03 5.58660030e-01 2.30966258e+00 6.29489243e-01 9.83492434e-01 5.59719503e-01 4.53819990e-01 6.66624427e-01 9.34755862e-01 3.79848391e-01 -3.37963045e-01 1.22391641e+00 3.23606044e-01 -6.40118659e-01 -2.99676180e-01 5.37476130e-03 1.09133251e-01 1.20858543e-01 -2.81977467e-02 -4.64505374e-01 -4.30186957e-01 2.88869113e-01 -1.43813312e+00 -1.30105829e+00 -2.06921011e-01 2.41040993e+00 4.50397789e-01 -2.26975054e-01 2.07594857e-02 2.01020136e-01 1.16193223e+00 3.32851976e-01 -3.92035127e-01 -6.66032255e-01 -5.38895130e-01 -1.31033733e-01 6.50406301e-01 6.82855248e-01 -6.86954200e-01 4.05099750e-01 6.43827105e+00 8.42707098e-01 -1.16707647e+00 3.41848917e-02 5.64958572e-01 -1.93593517e-01 -3.83846849e-01 -2.34037846e-01 -4.66178209e-01 7.92112648e-01 6.99214697e-01 -3.90443921e-01 4.98276591e-01 3.69368553e-01 5.73583603e-01 -7.58760512e-01 -6.60065532e-01 1.36368060e+00 2.40822658e-01 -9.40897346e-01 -6.25015274e-02 2.03544080e-01 8.00696969e-01 -7.25624859e-01 6.67481065e-01 -4.55460787e-01 -2.03575581e-01 -8.11009645e-01 4.95527953e-01 6.19854748e-01 1.35018587e+00 -1.03380692e+00 3.26880842e-01 1.92625284e-01 -1.19102836e+00 -2.03672335e-01 -4.36131090e-01 6.62218332e-02 4.38073605e-01 2.88446456e-01 -4.33467478e-01 3.39111149e-01 8.01249027e-01 2.08968610e-01 -6.30692005e-01 8.70858014e-01 3.51770461e-01 -2.43465886e-01 -7.15074129e-03 3.33916843e-01 -1.06880471e-01 -6.12851143e-01 5.97489774e-01 1.03410327e+00 6.61915421e-01 4.04663026e-01 -2.73409039e-01 6.95224345e-01 3.30341756e-01 1.37275204e-01 -1.10544813e+00 4.42065984e-01 3.63326818e-01 9.91468906e-01 -7.22133934e-01 -1.79248676e-01 -5.53549528e-01 1.32437193e+00 -1.87688321e-01 4.90662634e-01 -7.68161416e-01 -4.27377462e-01 2.53061026e-01 6.83054328e-01 3.46479177e-01 -5.25132939e-02 -3.27503271e-02 -8.75079572e-01 2.51351837e-02 -8.72451246e-01 1.25331521e-01 -1.36261559e+00 -8.60870361e-01 8.90639961e-01 3.10691625e-01 -1.48906052e+00 -1.07209861e-01 -4.29689050e-01 -2.73933977e-01 1.16079724e+00 -1.52906716e+00 -9.73151624e-01 -4.74172622e-01 7.74887681e-01 7.39039183e-01 -1.17870502e-01 3.61955494e-01 3.45845222e-01 -4.83734198e-02 -4.55557853e-02 4.58034545e-01 -7.20552683e-01 6.99733317e-01 -1.12326169e+00 -2.57068068e-01 1.06879973e+00 -5.62682927e-01 2.76394427e-01 1.12446117e+00 -5.80538273e-01 -1.17625248e+00 -7.67055273e-01 4.91521209e-01 -2.18626812e-01 5.43689616e-02 -1.36579171e-01 -7.92473078e-01 2.62645721e-01 5.11304557e-01 -4.65224206e-01 4.47612673e-01 -5.13241291e-01 1.04589149e-01 -2.68639266e-01 -1.57912445e+00 6.41582847e-01 6.70570493e-01 -4.51134175e-01 -3.29299718e-01 -1.25555605e-01 6.18676603e-01 -4.22779351e-01 -7.04216838e-01 7.39401430e-02 6.56973600e-01 -1.62656629e+00 9.60703135e-01 2.78578281e-01 2.96734303e-01 -7.01842487e-01 -2.94764370e-01 -9.57054496e-01 -2.48113319e-01 -4.45431381e-01 3.73600453e-01 1.36653876e+00 3.32103819e-02 -5.68721950e-01 3.18089187e-01 6.83425426e-01 3.36135924e-01 -2.42282793e-01 -6.41234696e-01 -4.64791000e-01 -5.24936259e-01 1.30429670e-01 5.10398567e-01 5.90542674e-01 -4.28936630e-01 2.19644040e-01 -7.96573341e-01 2.13797256e-01 6.91084325e-01 2.89576143e-01 5.43278277e-01 -1.10390770e+00 -4.56062526e-01 -8.12649447e-03 -3.65633875e-01 -6.98714375e-01 -3.06522161e-01 -2.02229545e-01 -4.02991593e-01 -1.31523478e+00 6.24772251e-01 -3.44626218e-01 1.41170695e-01 -5.77483058e-01 3.08652800e-02 3.09458673e-01 5.50894797e-01 2.79831648e-01 -3.72504950e-01 2.26278193e-02 1.27827179e+00 2.50941604e-01 -5.30575752e-01 -1.54678702e-01 -4.70440835e-01 3.87194902e-01 7.13326037e-01 -7.68370554e-02 -8.53210449e-01 -3.10732514e-01 2.85358042e-01 3.81337613e-01 2.65310764e-01 -1.03967237e+00 5.97609133e-02 -9.64997336e-02 6.14433587e-01 -5.04874527e-01 3.48589271e-01 -1.24731529e+00 8.61399055e-01 2.01625898e-01 -2.39329323e-01 4.03802544e-01 2.73992009e-02 5.46274364e-01 -8.49384591e-02 -1.82784960e-01 1.39793432e+00 -3.63758206e-01 -9.64785695e-01 -1.32281914e-01 -5.55601478e-01 -3.99932921e-01 1.35075748e+00 -9.14244175e-01 -6.43248141e-01 -4.65160996e-01 -2.94934720e-01 -4.69449192e-01 1.14913356e+00 2.34549567e-01 7.61836290e-01 -1.06089151e+00 -3.59352291e-01 1.78267792e-01 -1.21068284e-01 -4.53451186e-01 8.74335170e-01 4.90299195e-01 -9.25657749e-01 2.64861375e-01 -9.20233846e-01 -2.27651909e-01 -1.67115724e+00 9.61537302e-01 4.51869160e-01 -2.77116686e-01 -8.89033794e-01 -7.22421780e-02 5.63885689e-01 3.85383785e-01 -2.00730488e-01 1.68743894e-01 -5.26744723e-01 -2.12692335e-01 8.41754854e-01 5.65186322e-01 -1.36236608e-01 -8.93058538e-01 -3.98975722e-02 1.10474229e+00 1.84213340e-01 -3.94273579e-01 1.02074647e+00 -1.05700779e+00 1.12777404e-01 4.30262148e-01 1.16463161e+00 3.20185810e-01 -1.29636562e+00 7.35328496e-02 -5.62383652e-01 -1.12749696e+00 1.40148237e-01 -7.79157996e-01 -8.92336071e-01 5.71623266e-01 1.22332978e+00 3.78293872e-01 1.86650932e+00 -1.09812722e-01 2.85201371e-01 -2.73409158e-01 4.17675674e-01 -1.17719162e+00 2.81915534e-02 -1.58610791e-01 7.18378842e-01 -9.93249178e-01 1.11711890e-01 -5.11194229e-01 -9.02494013e-01 1.07914293e+00 4.62861180e-01 -7.03672618e-02 2.05733344e-01 5.49848378e-01 -4.41451445e-02 -1.78335637e-01 -5.52275836e-01 -4.12410885e-01 1.06181003e-01 1.01086521e+00 4.38668549e-01 -2.97639847e-01 -5.92780888e-01 -3.34961087e-01 1.53548226e-01 -2.33765364e-01 9.67293918e-01 7.25188673e-01 -7.42867708e-01 -8.62852931e-01 -6.85451269e-01 1.95134521e-01 -5.85270882e-01 9.02260989e-02 -1.07311621e-01 7.42344022e-01 1.73098445e-01 1.14959681e+00 1.11229010e-01 -9.63981301e-02 3.84129584e-01 -4.51884151e-01 6.23247623e-01 -3.03614885e-01 -3.02221894e-01 1.74446061e-01 -8.30055922e-02 -3.97178203e-01 -5.39484501e-01 -3.69598180e-01 -7.66072392e-01 -3.40673000e-01 7.29374066e-02 -2.07132116e-01 7.26335049e-01 4.12172407e-01 9.65482220e-02 4.54633385e-01 7.53549218e-01 -7.67929256e-01 1.58906549e-01 -4.34717566e-01 -1.38315022e+00 4.59683359e-01 6.10203445e-01 -3.43082488e-01 -3.20575267e-01 2.07842514e-01]

[10.845681190490723, -2.4518094062805176]

199278d0-7c9c-4334-a15a-a59f6252814b

single-stage-heavy-tailed-food-classification

2307.00182

https://arxiv.org/abs/2307.00182v1

https://arxiv.org/pdf/2307.00182v1.pdf

Single-Stage Heavy-Tailed Food Classification

Deep learning based food image classification has enabled more accurate nutrition content analysis for image-based dietary assessment by predicting the types of food in eating occasion images. However, there are two major obstacles to apply food classification in real life applications. First, real life food images are usually heavy-tailed distributed, resulting in severe class-imbalance issue. Second, it is challenging to train a single-stage (i.e. end-to-end) framework under heavy-tailed data distribution, which cause the over-predictions towards head classes with rich instances and under-predictions towards tail classes with rare instance. In this work, we address both issues by introducing a novel single-stage heavy-tailed food classification framework. Our method is evaluated on two heavy-tailed food benchmark datasets, Food101-LT and VFN-LT, and achieves the best performance compared to existing work with over 5% improvements for top-1 accuracy.

['Fengqing Zhu', 'Jiangpeng He']

2023-07-01

['classification-1', 'nutrition']

['methodology', 'miscellaneous']

[ 1.75285414e-01 -3.67138773e-01 -7.39909410e-01 -6.43519521e-01 -5.42278886e-01 -3.22321296e-01 -1.03745156e-03 8.67499530e-01 -2.15278849e-01 3.23637933e-01 3.83300811e-01 -9.14402530e-02 2.44242370e-01 -1.02275860e+00 -8.22962880e-01 -6.95877314e-01 -1.55273959e-01 2.49933168e-01 1.11889407e-01 6.47923127e-02 -2.69908756e-01 -4.75712091e-01 -1.69816947e+00 8.08252633e-01 7.21584618e-01 1.10639477e+00 2.30040938e-01 7.18844354e-01 -2.67262459e-01 8.14745069e-01 -9.13439766e-02 -3.77088308e-01 2.48941526e-01 -5.95717669e-01 -1.92366078e-01 6.63219392e-02 5.76827586e-01 -5.26502490e-01 3.92982103e-02 1.47900259e+00 7.17972815e-01 -1.82468653e-01 8.93909335e-01 -1.21335781e+00 -1.05199683e+00 8.91477108e-01 -1.14741313e+00 3.94361496e-01 3.08206618e-01 1.63917288e-01 7.15532422e-01 -5.43999255e-01 1.53345644e-01 1.28168392e+00 9.09467459e-01 3.24422747e-01 -1.03944516e+00 -9.09142077e-01 1.19674414e-01 4.74911600e-01 -1.05520010e+00 -1.26020119e-01 5.80902219e-01 -6.68031454e-01 7.32207716e-01 1.40612364e-01 8.11177254e-01 8.92683566e-01 4.54987139e-01 1.20403802e+00 1.16920245e+00 -2.31686100e-01 2.40339383e-01 -1.63442820e-01 3.07795614e-01 5.47864079e-01 4.79108572e-01 2.23809943e-01 -2.54775733e-01 1.08127914e-01 2.59603322e-01 2.29163289e-01 2.23132834e-01 -4.81979340e-01 -1.16162813e+00 1.25266433e+00 8.51114213e-01 -1.05222389e-01 -6.28120363e-01 -2.07717776e-01 8.26296806e-01 2.76792288e-01 8.70960176e-01 -3.18562746e-01 -7.26679325e-01 4.58248764e-01 -1.02009821e+00 5.43748200e-01 5.55235565e-01 6.52796924e-01 2.52058208e-01 -2.43247554e-01 -3.43379766e-01 9.02482748e-01 4.96564567e-01 8.35497797e-01 6.16528332e-01 -2.25627393e-01 3.30391675e-01 3.63416582e-01 6.15508109e-02 -1.13041604e+00 -8.37792099e-01 -7.28923798e-01 -1.16566789e+00 1.22857668e-01 5.96076787e-01 5.55252768e-02 -1.12852371e+00 1.65863287e+00 7.30395973e-01 -2.28482515e-01 -8.72533396e-02 1.12631381e+00 1.42413223e+00 6.95202112e-01 1.03918266e+00 -1.91932991e-01 1.71049976e+00 -1.26493311e+00 -7.37901807e-01 -1.12760551e-01 7.13968337e-01 -8.41997087e-01 1.20113397e+00 3.86377573e-01 -8.96551907e-01 -6.97649598e-01 -9.16784883e-01 6.97347820e-02 -4.49348122e-01 2.22787205e-02 6.78983033e-01 5.58946431e-01 -4.93300587e-01 2.93408692e-01 -5.73297441e-01 -3.37572068e-01 8.00167561e-01 4.04109582e-02 -1.57092378e-01 -5.03125429e-01 -1.07858396e+00 3.54371548e-01 6.71210349e-01 -2.65245736e-01 -7.09287286e-01 -1.29543304e+00 -1.01011157e+00 2.24817712e-02 2.01509610e-01 -8.72359812e-01 1.41382861e+00 -1.26056504e+00 -1.03920341e+00 1.07588995e+00 4.15263265e-01 -4.64835376e-01 5.16231120e-01 -1.33822531e-01 -5.68745375e-01 -1.54832527e-01 4.43571925e-01 9.09017682e-01 4.75851238e-01 -8.79699051e-01 -8.00363898e-01 -6.47209406e-01 -2.37570137e-01 1.51576266e-01 -2.83996820e-01 9.12593156e-02 2.64451444e-01 -8.37435842e-01 -2.16192916e-01 -6.13160908e-01 -1.16882794e-01 3.28464389e-01 -4.55426604e-01 -3.99230003e-01 4.55687046e-01 -6.90914989e-01 1.05830276e+00 -2.18246603e+00 -4.69972730e-01 -4.54807222e-01 1.42939389e-01 3.02948594e-01 -2.30153754e-01 -4.10240665e-02 -3.36209774e-01 -2.62069374e-01 2.17308670e-01 2.05805555e-01 2.72061974e-01 1.49438158e-02 2.33870208e-01 5.55517435e-01 -7.32904440e-03 9.41199005e-01 -1.25817931e+00 -6.89264417e-01 4.62887436e-01 2.60280818e-01 -7.15272129e-01 9.13772881e-02 -6.51921034e-01 1.78976208e-01 -4.96123314e-01 8.32518399e-01 1.04927850e+00 -3.79081458e-01 -3.47517692e-02 -7.74773061e-01 -7.62469992e-02 -1.79201409e-01 -9.24795568e-01 1.56360888e+00 6.24522055e-03 -2.62515634e-01 -1.71798781e-01 -1.14262068e+00 4.26944196e-01 4.18913215e-02 6.28555298e-01 -1.20167470e+00 3.39763790e-01 5.59988469e-02 1.49229512e-01 -7.43870139e-01 9.19438675e-02 -3.22859734e-01 -8.78661200e-02 2.77566612e-01 -4.96532489e-03 5.82390606e-01 4.89705801e-01 -2.11342603e-01 5.70825160e-01 2.69055486e-01 6.30941927e-01 -6.22298062e-01 4.97940518e-02 3.74388218e-01 5.21654546e-01 4.86836582e-01 -7.39535332e-01 4.33768749e-01 1.14389069e-01 -8.72282028e-01 -1.11568582e+00 -1.00698793e+00 -1.86724499e-01 1.65230751e+00 -1.00321129e-01 -6.00377843e-02 -6.80302799e-01 -9.54579592e-01 2.25351587e-01 5.95552862e-01 -8.23701680e-01 -2.55536735e-01 -2.13188529e-01 -1.44939435e+00 6.63733855e-02 7.81020224e-01 6.60304844e-01 -9.89117503e-01 -7.24876821e-01 4.53677863e-01 -3.63930881e-01 -8.00555468e-01 -7.78742433e-01 3.90840143e-01 -6.11082554e-01 -1.24861407e+00 -1.16960859e+00 -1.01944256e+00 2.95993179e-01 5.28781593e-01 1.82903755e+00 -1.60907790e-01 -5.04956782e-01 -1.84888825e-01 -6.63876414e-01 -8.79397750e-01 -4.44613636e-01 -5.04362695e-02 -3.47072095e-01 -3.53129089e-01 9.99099970e-01 9.78562087e-02 -1.44996679e+00 1.21058337e-01 -8.64858925e-01 6.71624690e-02 3.74969482e-01 8.86507452e-01 1.04844224e+00 4.82628137e-01 8.30199063e-01 -7.32031167e-01 2.35358886e-02 -8.91236484e-01 -2.29409367e-01 1.74777806e-01 -5.63709617e-01 -3.30013037e-01 5.58228016e-01 -7.32503295e-01 -6.86639428e-01 4.41005051e-01 -2.73725688e-01 2.86496952e-02 -2.73044616e-01 5.56211174e-01 7.27204885e-03 4.32295710e-01 7.64747500e-01 7.56773129e-02 -7.39248544e-02 -5.39066315e-01 2.67687649e-01 5.95777512e-01 3.98893625e-01 -1.80748180e-01 1.78045541e-01 6.91307290e-03 -1.73224181e-01 -4.49630886e-01 -1.39072824e+00 -5.40419519e-01 -2.41508588e-01 -1.08382590e-01 1.18605304e+00 -1.45447075e+00 -5.07541418e-01 7.28662789e-01 -2.24736035e-01 -4.66059983e-01 -2.31870517e-01 6.15054965e-01 -3.57261688e-01 2.31197491e-01 -6.53355896e-01 -3.52763742e-01 -9.84888911e-01 -1.01452649e+00 9.75434542e-01 2.94739127e-01 -1.55091435e-01 -7.15989292e-01 3.88908237e-02 4.41875160e-01 3.15029621e-01 4.63613510e-01 1.29179370e+00 -5.23648739e-01 3.89528364e-01 -4.95693497e-02 -5.37462711e-01 1.17481969e-01 -6.25076098e-03 -2.12847248e-01 -6.97135866e-01 -3.99543315e-01 -2.75606036e-01 -6.76742077e-01 1.02971578e+00 9.98281360e-01 1.18290973e+00 -1.36100933e-01 -2.15469241e-01 4.19588178e-01 1.52289903e+00 -1.06449187e-01 1.98135391e-01 2.36699522e-01 4.93453175e-01 3.87723833e-01 8.05303931e-01 4.96982664e-01 9.34858084e-01 5.20955443e-01 7.00271070e-01 -5.84236622e-01 -7.25064337e-01 -3.81060362e-01 2.07100987e-01 5.51523983e-01 6.43815935e-01 -3.52853298e-01 -4.23450083e-01 6.71183705e-01 -1.62179053e+00 -7.83330381e-01 -5.91921091e-01 2.01259565e+00 9.98133004e-01 -2.01341763e-01 8.48124981e-01 2.18900591e-01 7.16724396e-01 -1.35354400e-01 -1.04164302e+00 -1.07643485e-01 -3.28514650e-02 1.60018541e-02 5.16143262e-01 -3.18519652e-01 -1.79774535e+00 3.86287421e-01 6.23357487e+00 7.32489645e-01 -1.01555848e+00 3.38945955e-01 1.04505324e+00 -1.49034038e-01 1.88879564e-01 -1.11282444e+00 -6.76187158e-01 8.54034722e-01 8.69366765e-01 2.77119756e-01 -2.78050490e-02 9.01081979e-01 7.57037476e-02 -7.08500445e-02 -8.88435543e-01 9.10474062e-01 2.30976800e-03 -7.35702515e-01 -6.80628791e-02 -2.42333800e-01 7.33402967e-01 2.96701014e-01 2.36559451e-01 5.80280244e-01 5.87501705e-01 -7.07191527e-01 1.01898158e+00 -9.70538259e-02 8.05854201e-01 -8.85319233e-01 7.25040615e-01 3.23936343e-01 -1.31859040e+00 -1.90560907e-01 -7.43124545e-01 -1.92860495e-02 -1.87737107e-01 1.19784117e+00 -2.13443533e-01 3.35032254e-01 1.14077401e+00 7.42552936e-01 -3.88614684e-01 1.22968423e+00 3.12751561e-01 6.66012585e-01 -4.06276613e-01 -6.98022842e-02 2.39786431e-01 3.08560342e-01 -2.65692979e-01 1.33564341e+00 3.77886653e-01 8.76032263e-02 9.30684984e-01 4.36468184e-01 8.16794783e-02 6.42709136e-01 -1.50827184e-01 2.67760847e-02 -3.31260055e-01 1.34499025e+00 -1.00781417e+00 -4.50107992e-01 -7.17831194e-01 7.07996488e-01 -2.48728264e-02 -2.12210208e-01 -8.70079339e-01 1.77475810e-01 4.44036484e-01 1.77794054e-01 4.78438139e-01 4.43679929e-01 3.58178169e-02 -1.14813030e+00 -4.84172165e-01 -1.01603627e+00 1.16201580e+00 -6.18234992e-01 -1.95685065e+00 1.15094885e-01 -1.09248698e-01 -9.63021338e-01 1.08626463e-01 -6.06133044e-01 -1.25500292e-01 4.55881417e-01 -1.91915894e+00 -1.42037714e+00 -4.42617655e-01 6.52239740e-01 7.96047747e-01 1.84100240e-01 9.97641683e-01 9.64668632e-01 -5.16652405e-01 9.07021463e-01 1.90192699e-01 6.86924681e-02 7.34880924e-01 -1.45831752e+00 6.77512512e-02 2.71389127e-01 -2.02017576e-01 -2.56441087e-01 8.91989410e-01 -7.89859772e-01 -1.37801218e+00 -1.68104529e+00 5.98931611e-01 8.90158191e-02 3.78416538e-01 -1.03473306e-01 -7.05219686e-01 2.33356535e-01 1.22731037e-01 3.00246686e-01 1.20385075e+00 -3.13146934e-02 -5.68894923e-01 -1.17097557e-01 -1.65281284e+00 1.23839691e-01 7.91990936e-01 2.40491301e-01 -1.66910708e-01 8.74718010e-01 5.24414003e-01 -3.26844752e-01 -1.12396145e+00 5.37662268e-01 7.99644589e-01 -7.82807767e-01 1.22007358e+00 -6.24960124e-01 8.17783058e-01 -1.75436854e-01 -4.73131716e-01 -1.55279744e+00 -8.70005846e-01 4.22002554e-01 -2.90533364e-01 8.71773362e-01 1.53409585e-01 -1.61231950e-01 9.00552332e-01 -3.77741382e-02 3.33475438e-03 -6.65044904e-01 -2.85318494e-01 -4.91625756e-01 4.97073740e-01 1.21136539e-01 8.70673954e-01 1.04387558e+00 -2.08227828e-01 2.23890737e-01 -3.71265113e-01 -1.07031241e-02 1.00053084e+00 5.39292216e-01 4.30827111e-01 -1.07761669e+00 -1.60872757e-01 -2.53624767e-01 -3.35458696e-01 -7.36499071e-01 -2.46004596e-01 -1.11232352e+00 3.40562910e-01 -1.50480676e+00 1.02224457e+00 -2.68964469e-01 -7.05460846e-01 6.09769344e-01 -5.67459524e-01 7.62860119e-01 9.54792928e-03 -5.88330179e-02 -7.15418875e-01 2.26750180e-01 1.45972526e+00 -4.88882452e-01 1.26993164e-01 -9.13135856e-02 -8.31192315e-01 7.47056186e-01 9.65458632e-01 -5.47391951e-01 -5.76356947e-01 -4.83482748e-01 3.32430005e-01 -1.88227862e-01 3.92667204e-01 -9.81276572e-01 -3.62383902e-01 -1.95860699e-01 1.10400474e+00 -1.14078426e+00 -4.23052013e-01 -7.39439189e-01 3.07187915e-01 1.03842223e+00 -4.76585150e-01 -6.63468838e-02 1.51411155e-02 5.92910945e-01 2.05417067e-01 4.38631549e-02 9.63919699e-01 -4.65573221e-01 -6.93799376e-01 5.08086443e-01 -1.23706415e-01 3.23917232e-02 1.12507892e+00 2.16770813e-01 -3.39884251e-01 1.71663519e-02 -7.95272589e-01 3.01685870e-01 1.36825368e-01 4.26851749e-01 1.21678285e-01 -1.56476498e+00 -1.06011248e+00 3.34755927e-01 5.27501583e-01 -3.04933250e-01 7.73960829e-01 7.98684776e-01 -4.32708263e-01 1.32109806e-01 -3.20771575e-01 -7.57192314e-01 -1.28328729e+00 1.15791845e+00 2.09357336e-01 -4.05610412e-01 -9.86129105e-01 6.13473594e-01 7.32321382e-01 -5.22330821e-01 1.63438290e-01 -6.80601120e-01 -4.34753895e-01 8.71206075e-02 1.10082030e+00 2.67772317e-01 1.10770456e-01 -6.03666186e-01 -1.10059217e-01 5.66963315e-01 -2.28712112e-01 1.05983853e+00 1.36910582e+00 -2.43420571e-01 3.42913806e-01 2.57760763e-01 1.10976136e+00 -7.84835100e-01 -1.22649550e+00 -3.35447729e-01 -4.55360621e-01 -2.51126975e-01 4.90055144e-01 -1.50649512e+00 -1.55827296e+00 8.93347681e-01 1.47161245e+00 2.74023890e-01 1.33808553e+00 -1.06415823e-01 1.41838396e+00 -2.61575818e-01 2.42502704e-01 -8.29148233e-01 -8.55494589e-02 -7.64385462e-02 5.61420321e-01 -1.66487336e+00 1.49375319e-01 -5.31274796e-01 -4.77376968e-01 8.42701674e-01 3.92828017e-01 -9.68297496e-02 1.01259708e+00 2.52177984e-01 2.33421028e-01 -2.69540668e-01 -4.30013418e-01 -1.40369594e-01 3.08341831e-01 8.76447082e-01 5.85372627e-01 7.47937083e-01 -3.99124116e-01 7.06802428e-01 -3.03293020e-02 5.68332911e-01 -9.80875269e-02 6.71527028e-01 -6.90808892e-01 -7.96888888e-01 -2.61250049e-01 9.07671094e-01 -1.00416255e+00 -1.28151834e-01 2.99405456e-01 4.36361372e-01 5.25049984e-01 1.05999303e+00 8.17376077e-02 -1.59197226e-01 3.70262951e-01 -3.51969413e-02 6.08132362e-01 -3.58292103e-01 -7.91659534e-01 3.62620652e-01 -5.45349531e-02 -4.74721938e-01 -5.38633347e-01 -7.17786074e-01 -1.02458954e+00 -5.19616604e-01 -1.96565196e-01 -2.85718352e-01 6.14193261e-01 5.63719153e-01 2.02398092e-01 6.00107074e-01 4.26448971e-01 -5.97593606e-01 -1.02338600e+00 -1.02666390e+00 -7.70599782e-01 8.36179912e-01 1.77316472e-01 -7.17472732e-01 5.08842617e-02 2.69020826e-01]

[11.555815696716309, 4.383572578430176]

0fc9c99f-a99c-4ba3-98e8-0928f9858c8b

tsgm-a-flexible-framework-for-generative

2305.11567

https://arxiv.org/abs/2305.11567v1

https://arxiv.org/pdf/2305.11567v1.pdf

TSGM: A Flexible Framework for Generative Modeling of Synthetic Time Series

Temporally indexed data are essential in a wide range of fields and of interest to machine learning researchers. Time series data, however, are often scarce or highly sensitive, which precludes the sharing of data between researchers and industrial organizations and the application of existing and new data-intensive ML methods. A possible solution to this bottleneck is to generate synthetic data. In this work, we introduce Time Series Generative Modeling (TSGM), an open-source framework for the generative modeling of synthetic time series. TSGM includes a broad repertoire of machine learning methods: generative models, probabilistic, and simulator-based approaches. The framework enables users to evaluate the quality of the produced data from different angles: similarity, downstream effectiveness, predictive consistency, diversity, and privacy. The framework is extensible, which allows researchers to rapidly implement their own methods and compare them in a shareable environment. TSGM was tested on open datasets and in production and proved to be beneficial in both cases. Additionally to the library, the project allows users to employ command line interfaces for synthetic data generation which lowers the entry threshold for those without a programming background.

['Samuel Kaski', 'Letizia Iannucci', 'Alexander Nikitin']

2023-05-19

['synthetic-data-generation', 'synthetic-data-generation']

['medical', 'miscellaneous']

[-2.01605096e-01 -3.27521652e-01 6.49806950e-03 -2.20157847e-01 -7.53591120e-01 -7.66351700e-01 7.24260867e-01 2.69031823e-01 -1.48578286e-01 8.59219432e-01 -1.37161627e-01 -4.35648829e-01 -2.53374726e-01 -9.30216968e-01 -3.08114290e-01 -6.79616868e-01 -2.82710075e-01 6.99911416e-01 2.73040116e-01 -8.69345814e-02 2.43398637e-01 7.01849103e-01 -1.87302577e+00 8.35642666e-02 7.41101682e-01 7.67323375e-01 1.65163830e-01 4.91540283e-01 -1.11857772e-01 3.29672486e-01 -7.12757468e-01 -1.15079299e-01 3.62791806e-01 -3.60092819e-01 -1.79248303e-01 -2.19650239e-01 -4.21768188e-01 5.82720898e-02 1.38138846e-01 5.92231452e-01 7.12032020e-01 -2.96611637e-02 4.04147208e-01 -1.86161983e+00 -1.82820454e-01 4.39081937e-01 -8.23612809e-02 6.43177703e-02 3.82611811e-01 2.33996466e-01 3.81450832e-01 -4.62105483e-01 7.01804578e-01 1.03281415e+00 7.82515347e-01 1.12948895e-01 -1.48042643e+00 -6.90121651e-01 -4.20376331e-01 3.30574922e-02 -1.32390010e+00 -3.19231659e-01 7.51512706e-01 -5.92778325e-01 8.85098279e-01 5.50031602e-01 6.29498899e-01 1.29500973e+00 2.36241326e-01 3.80971849e-01 1.31498551e+00 -3.56573045e-01 7.79902160e-01 4.29009914e-01 -1.91243842e-01 1.28928814e-02 1.73611477e-01 2.50454456e-01 -3.72280091e-01 -5.99532306e-01 6.71913087e-01 2.57537067e-01 5.11370860e-02 -4.23497766e-01 -1.31093657e+00 8.21650922e-01 -2.48281121e-01 4.71045971e-01 -5.04031777e-01 -5.30019365e-02 4.28607225e-01 5.46195209e-01 6.32289112e-01 3.77429456e-01 -4.85755980e-01 -5.39223373e-01 -1.17292655e+00 6.57143891e-01 1.08648753e+00 9.81071949e-01 4.06736284e-01 6.07035160e-02 -2.33054757e-02 4.42762047e-01 2.84847438e-01 4.35965866e-01 5.78051627e-01 -8.56978893e-01 1.03669606e-01 4.87704098e-01 1.73831567e-01 -8.95647168e-01 -2.19543457e-01 -3.28730971e-01 -7.00758755e-01 7.27040842e-02 3.64136428e-01 -1.25637740e-01 -6.80342734e-01 1.56713033e+00 4.10836220e-01 5.19248191e-03 -3.82090993e-02 5.33317626e-01 5.19272745e-01 7.84572959e-01 -4.06961180e-02 -6.29852414e-01 7.33527243e-01 -2.74924308e-01 -6.02269828e-01 2.83694953e-01 4.23842639e-01 -9.98148263e-01 9.05898929e-01 4.39017504e-01 -9.89962459e-01 -4.22244132e-01 -7.03289628e-01 5.22801220e-01 -6.74061775e-01 -3.44122887e-01 5.86617529e-01 7.85104394e-01 -1.05556321e+00 7.65730679e-01 -1.19431138e+00 -5.71930528e-01 1.88507512e-01 2.64378458e-01 3.85532854e-03 1.06541336e-01 -1.10514176e+00 6.84370935e-01 3.63917351e-01 -3.60710442e-01 -6.03286445e-01 -8.92203212e-01 -6.58401191e-01 -7.29218572e-02 4.95825410e-02 -6.35959327e-01 1.06925488e+00 -5.27999222e-01 -1.25888026e+00 4.20368671e-01 8.69957730e-02 -4.38710481e-01 9.52569366e-01 1.09790787e-01 -5.85774541e-01 -3.92897278e-01 2.08504364e-01 1.95502490e-01 6.76341474e-01 -9.46496665e-01 -2.36803636e-01 -3.13660949e-01 -5.60061991e-01 -3.21710199e-01 1.46637065e-02 9.08132568e-02 -1.52974367e-01 -8.54732335e-01 -8.28500986e-02 -1.01723659e+00 -4.73825932e-01 -3.17359388e-01 -2.35242918e-01 1.78561658e-02 1.15080559e+00 -4.75155562e-01 1.13223755e+00 -2.08741522e+00 -2.10210428e-01 4.28438395e-01 -2.73972601e-01 7.30536953e-02 2.08301395e-01 1.25764108e+00 -1.20788217e-01 1.74707860e-01 -2.71410853e-01 -2.36643642e-01 1.83741584e-01 1.62251875e-01 -5.24474442e-01 3.45783651e-01 -2.17893850e-02 4.89811212e-01 -7.51509130e-01 -3.54349792e-01 5.39871156e-01 4.45842117e-01 -3.22292179e-01 3.05606902e-01 -4.53411728e-01 7.00348496e-01 -5.10978580e-01 5.28336704e-01 5.10403454e-01 -1.63590327e-01 2.07795590e-01 2.33119562e-01 -4.59348053e-01 6.59993738e-02 -1.39259696e+00 1.43328571e+00 -3.87605578e-01 4.46062773e-01 -4.53285187e-01 -7.19181597e-01 1.16526163e+00 4.17395502e-01 7.27749407e-01 -3.75456303e-01 5.47537878e-02 2.93052465e-01 -2.18038931e-01 -3.27256978e-01 3.77839148e-01 1.39244452e-01 -1.29294917e-01 9.36841965e-01 -1.16817817e-01 -4.90581244e-01 4.58787620e-01 -1.19014725e-01 9.01139557e-01 4.85054255e-01 3.35789919e-01 -1.29342720e-01 -5.87277934e-02 1.02386974e-01 2.60889798e-01 5.29235840e-01 3.05315703e-01 4.59874868e-01 4.55187321e-01 -5.26067257e-01 -1.36069572e+00 -1.13579631e+00 -1.30352452e-01 9.44334805e-01 -3.66195947e-01 -5.19948423e-01 -4.92286623e-01 -4.50284295e-02 2.51404382e-02 1.08045161e+00 -3.86797965e-01 -1.41954338e-02 -2.10378721e-01 -8.87494206e-01 2.20305309e-01 2.29342327e-01 1.07877500e-01 -1.11403716e+00 -9.22792971e-01 4.20113951e-01 -1.75482541e-01 -7.25015044e-01 -8.94459337e-02 8.20601135e-02 -1.04717422e+00 -7.30444908e-01 -4.77279127e-01 -9.41994265e-02 3.22858125e-01 -8.15589949e-02 1.11598170e+00 -3.42929512e-01 -5.68839550e-01 3.30959022e-01 -4.09238547e-01 -8.62177014e-01 -8.17953944e-01 9.69949737e-02 -5.58895580e-02 -2.02244356e-01 3.11835587e-01 -9.53934610e-01 -3.33450049e-01 5.07873237e-01 -1.20838761e+00 -1.11344524e-01 9.11536068e-02 5.13064921e-01 6.66804433e-01 3.86678912e-02 7.41061330e-01 -5.58322966e-01 1.03261948e+00 -7.61249423e-01 -9.79975283e-01 1.97870895e-01 -1.00692546e+00 2.95649134e-02 5.22188067e-01 -4.61580217e-01 -5.99846840e-01 4.28114198e-02 -9.16683525e-02 -5.28033257e-01 -2.43293896e-01 6.76488400e-01 1.57143716e-02 2.64450252e-01 5.75677454e-01 4.33626354e-01 2.14424610e-01 -4.15147096e-01 2.28656545e-01 5.52623332e-01 1.76411971e-01 -4.65823382e-01 5.57243824e-01 2.63855964e-01 -3.61579061e-02 -9.32258427e-01 -3.99929285e-02 -2.09324583e-01 -3.63478988e-01 -2.34276935e-01 2.85790920e-01 -6.82811260e-01 -4.18557465e-01 4.85078514e-01 -8.29186440e-01 -3.41561168e-01 -4.75463808e-01 5.25865018e-01 -7.56423831e-01 2.84919553e-02 -8.80053565e-02 -9.08755422e-01 -2.41235867e-01 -1.15133584e+00 9.43008304e-01 1.41792491e-01 -6.20310366e-01 -1.02799606e+00 4.09608573e-01 -1.64972574e-01 7.97556996e-01 7.92256176e-01 8.06663096e-01 -9.76208806e-01 -6.47975087e-01 -5.71790636e-01 2.35047117e-01 -4.17934544e-02 5.06296679e-02 6.23912513e-01 -9.08530354e-01 -3.96615207e-01 -1.95746613e-03 -7.08108470e-02 1.13991275e-01 5.27885497e-01 1.16839027e+00 -1.39132097e-01 -4.88056749e-01 3.75326574e-01 1.24619889e+00 5.19573987e-01 7.56423175e-01 2.46813864e-01 1.20118663e-01 6.36575997e-01 6.56515956e-01 1.08358765e+00 3.04691344e-01 7.83888280e-01 2.79540896e-01 2.19800882e-02 6.72348440e-01 -1.52869835e-01 1.92571566e-01 7.24152267e-01 -1.72561616e-01 -3.05759996e-01 -9.85004246e-01 4.65316564e-01 -1.93121171e+00 -1.32275009e+00 -3.68073374e-01 2.58181930e+00 7.51473963e-01 9.46472399e-03 3.85342181e-01 3.14282924e-01 5.41423738e-01 -7.53712654e-02 -4.85172808e-01 -3.97958010e-01 1.75429955e-01 2.53281027e-01 3.35040808e-01 -1.03680752e-01 -8.42965126e-01 4.30815637e-01 6.71314573e+00 7.13112056e-01 -1.32876778e+00 6.25583977e-02 5.81132472e-01 1.11458346e-03 -4.46574956e-01 2.88596060e-02 -5.57304084e-01 7.80157089e-01 1.58652794e+00 -9.84739184e-01 4.62070316e-01 9.76919293e-01 5.95959544e-01 -1.36705982e-02 -1.04283261e+00 8.69149625e-01 -3.55632573e-01 -1.46780097e+00 -3.64119321e-01 1.82344049e-01 5.30527115e-01 2.14158222e-01 -7.46877259e-03 2.22680196e-01 3.12368482e-01 -1.03948271e+00 6.36887968e-01 7.42366135e-01 5.91846883e-01 -8.65893066e-01 7.39179671e-01 5.64485669e-01 -9.52556968e-01 2.31499419e-01 -1.45122304e-03 -1.28771574e-03 3.07986259e-01 7.07451165e-01 -9.90540922e-01 7.25727558e-01 7.66666532e-01 3.80979806e-01 -4.90629733e-01 1.17106736e+00 4.50468451e-01 6.04074597e-01 -7.00442135e-01 -1.28344893e-01 -2.06326887e-01 -2.98070371e-01 4.56132174e-01 1.13668287e+00 7.94335485e-01 -3.72249186e-01 2.86054850e-01 1.03642702e+00 6.07078671e-01 9.64508429e-02 -7.86748946e-01 -4.25047070e-01 6.85519338e-01 9.91097152e-01 -7.80456662e-01 8.66016671e-02 -9.45014954e-02 5.27983367e-01 -3.88246566e-01 1.94218740e-01 -8.56360435e-01 -4.29661304e-01 3.74030322e-01 3.65828574e-01 1.04605570e-01 -3.82506162e-01 -1.86651230e-01 -8.66245270e-01 -9.24885422e-02 -1.17157340e+00 4.39094931e-01 -7.92199433e-01 -1.17201149e+00 7.79336333e-01 4.98442203e-01 -1.48979378e+00 -1.03773618e+00 -1.81776449e-01 -5.00236273e-01 1.12268519e+00 -7.15105355e-01 -9.30211604e-01 -3.59359592e-01 6.06133997e-01 2.86247104e-01 -2.74596900e-01 1.00854504e+00 3.74978513e-01 -1.77427605e-01 1.55263811e-01 3.17505360e-01 -4.84077513e-01 5.32453418e-01 -1.05849683e+00 6.90153182e-01 6.15318596e-01 1.45112826e-02 6.85777545e-01 1.08109844e+00 -6.69186890e-01 -1.34340215e+00 -1.26110733e+00 8.38605762e-01 -3.85016173e-01 7.14021146e-01 -4.22270298e-01 -8.79318774e-01 4.38152462e-01 9.85470638e-02 -2.93471277e-01 8.89511704e-01 -1.98905095e-01 3.82537022e-02 -1.33185893e-01 -1.24401128e+00 5.45212626e-01 4.77606565e-01 -2.50224590e-01 -2.50502557e-01 4.90067214e-01 4.60389078e-01 -2.29567200e-01 -1.13394940e+00 2.50149518e-01 4.53979224e-01 -1.18822086e+00 7.32842267e-01 -2.57200480e-01 1.86570898e-01 -4.15579945e-01 -2.25001916e-01 -1.29002607e+00 1.81915328e-01 -1.02273369e+00 2.43760794e-02 1.40435398e+00 4.74108666e-01 -8.95381272e-01 3.73303145e-01 6.38317645e-01 1.33631542e-01 -4.32202786e-01 -7.88567007e-01 -9.35885310e-01 -2.23958388e-01 -8.00491154e-01 1.08544028e+00 1.01490259e+00 -1.28475949e-01 -1.53889820e-01 -2.72959977e-01 -1.91490818e-02 5.58494031e-01 4.57903385e-01 1.11307883e+00 -1.41141105e+00 -3.47306997e-01 -3.92457992e-01 -4.58995402e-01 -1.77379876e-01 -4.02127296e-01 -5.83933711e-01 -2.64983803e-01 -1.34379447e+00 -2.02061102e-01 -6.81531072e-01 4.97790985e-02 2.66484797e-01 3.00572813e-01 -7.44657367e-02 -3.44172940e-02 3.15548360e-01 -5.87806627e-02 4.03170377e-01 6.96640015e-01 4.07662064e-01 -4.40280378e-01 5.15423775e-01 -1.48169592e-01 2.99264044e-01 9.22815442e-01 -6.76678061e-01 -6.82049274e-01 2.74389654e-01 1.06210545e-01 2.73906320e-01 5.10810554e-01 -1.06338561e+00 1.23991799e-02 -3.52285713e-01 2.50035584e-01 -9.31954086e-01 1.04541883e-01 -8.67172182e-01 1.34977281e+00 4.51734990e-01 -1.12602100e-01 4.93013322e-01 3.16823959e-01 3.37566048e-01 -1.26206830e-01 -9.43729188e-03 5.65623403e-01 -1.72050446e-01 -3.56241226e-01 3.70884985e-01 -2.60580093e-01 -2.70848181e-02 1.42306709e+00 -1.70490578e-01 -1.14192612e-01 -6.79277539e-01 -6.01451457e-01 2.69551365e-05 7.56540596e-01 5.68935215e-01 3.39761466e-01 -1.13708854e+00 -6.33500397e-01 5.00449836e-01 2.17339411e-01 -3.58166307e-01 1.22596309e-01 7.02872813e-01 -5.70993960e-01 2.93765515e-01 -2.88359404e-01 -6.90070748e-01 -9.95308042e-01 7.87127972e-01 1.32481307e-01 -1.39359340e-01 -4.76708949e-01 1.76401109e-01 -3.13859046e-01 -5.92849731e-01 2.25613844e-02 -3.04168373e-01 2.28191659e-01 1.49783611e-01 4.87508625e-01 4.36480314e-01 2.32166320e-01 -2.19823867e-01 -2.50812352e-01 8.73571411e-02 3.12564403e-01 -4.09545928e-01 1.64258671e+00 7.97898769e-02 -1.01324797e-01 1.01072824e+00 9.22552228e-01 -1.19539976e-01 -9.94979024e-01 1.60681322e-01 2.69143376e-02 -4.84645218e-01 -2.77141213e-01 -6.72295272e-01 -6.86081469e-01 5.71504653e-01 6.69131935e-01 7.91436493e-01 1.16272795e+00 -1.73154756e-01 2.73423672e-01 -4.15238142e-02 7.67176330e-01 -8.62140119e-01 -4.42877769e-01 7.98779912e-03 9.26786244e-01 -8.94620776e-01 7.27864578e-02 -1.30000561e-01 -5.97685814e-01 9.93963540e-01 -3.75602068e-03 1.69691667e-01 6.90398514e-01 5.07768333e-01 5.99285997e-02 1.73793845e-02 -1.14004993e+00 9.46556851e-02 8.64419267e-02 7.26163208e-01 5.70095181e-01 7.40210861e-02 -4.93234903e-01 2.77697802e-01 -3.23620081e-01 4.67078209e-01 5.15249193e-01 1.07631111e+00 6.28017038e-02 -1.44730651e+00 -4.66240197e-01 4.72202688e-01 -4.62099612e-01 7.97160119e-02 -9.60492417e-02 9.54682648e-01 -5.05250432e-02 9.66688037e-01 1.23789430e-01 -1.55020133e-01 2.14774370e-01 9.84644070e-02 1.09442770e-01 -2.75442928e-01 -6.29284978e-01 2.02782288e-01 3.27147543e-02 -4.30716485e-01 -3.00803602e-01 -1.04709494e+00 -7.58572280e-01 -5.40579379e-01 -1.25179201e-01 3.57731670e-01 1.09685576e+00 6.34092271e-01 6.34437442e-01 2.38362074e-01 7.34896183e-01 -6.59944415e-01 -4.62537795e-01 -8.51461828e-01 -5.15595555e-01 2.17170507e-01 -1.83471039e-01 -4.96939451e-01 -2.34756127e-01 2.62927920e-01]

[7.2125630378723145, 3.6036620140075684]

7cf827bc-f7b5-4975-9b0f-ff032e4ec1a4

ultrahyperbolic-knowledge-graph-embeddings

2206.00449

https://arxiv.org/abs/2206.00449v1

https://arxiv.org/pdf/2206.00449v1.pdf

Ultrahyperbolic Knowledge Graph Embeddings

Recent knowledge graph (KG) embeddings have been advanced by hyperbolic geometry due to its superior capability for representing hierarchies. The topological structures of real-world KGs, however, are rather heterogeneous, i.e., a KG is composed of multiple distinct hierarchies and non-hierarchical graph structures. Therefore, a homogeneous (either Euclidean or hyperbolic) geometry is not sufficient for fairly representing such heterogeneous structures. To capture the topological heterogeneity of KGs, we present an ultrahyperbolic KG embedding (UltraE) in an ultrahyperbolic (or pseudo-Riemannian) manifold that seamlessly interleaves hyperbolic and spherical manifolds. In particular, we model each relation as a pseudo-orthogonal transformation that preserves the pseudo-Riemannian bilinear form. The pseudo-orthogonal transformation is decomposed into various operators (i.e., circular rotations, reflections and hyperbolic rotations), allowing for simultaneously modeling heterogeneous structures as well as complex relational patterns. Experimental results on three standard KGs show that UltraE outperforms previous Euclidean- and hyperbolic-based approaches.

['Steffen Staab', 'Chuan Zhou', 'Shirui Pan', 'Chengjin Xu', 'Mojtaba Nayyeri', 'Shichao Zhu', 'Bo Xiong']

2022-06-01

['knowledge-graph-embeddings', 'knowledge-graph-embeddings']

['graphs', 'methodology']

[-4.48901057e-01 5.06151795e-01 1.57855481e-01 -1.25847146e-01 -5.22304438e-02 -6.25483155e-01 6.16873324e-01 4.26478721e-02 2.27390721e-01 1.34535894e-01 4.34653521e-01 -9.06787962e-02 -5.21852612e-01 -1.07508409e+00 -3.44838649e-01 -7.44810104e-01 -4.44967330e-01 4.47371930e-01 2.56636649e-01 -4.08211023e-01 2.17428301e-02 4.74903971e-01 -1.33612418e+00 -5.11175916e-02 9.35407221e-01 5.76914549e-01 -1.56066790e-02 5.94697297e-01 2.35699087e-01 8.50910246e-01 2.64060451e-03 -6.76606774e-01 1.01911053e-01 -3.43133986e-01 -9.21186507e-01 2.19037458e-01 4.29256618e-01 1.37754232e-01 -1.17693126e+00 1.03290296e+00 1.05005011e-01 9.41313431e-02 7.15957105e-01 -1.53689051e+00 -1.16969335e+00 7.41289079e-01 -2.53819525e-01 -2.34536648e-01 2.93409079e-01 -4.80697423e-01 1.42149127e+00 -1.13667035e+00 6.95155382e-01 1.24546790e+00 4.82084483e-01 -9.54296291e-02 -1.38187468e+00 -1.51480615e-01 -1.41884908e-01 5.35821915e-01 -2.03118157e+00 2.62824036e-02 9.27167952e-01 -5.53158402e-01 3.14158410e-01 4.98900115e-01 1.04237771e+00 7.23865032e-01 2.68780172e-01 3.73537391e-01 1.01007366e+00 -1.03089973e-01 1.54927358e-01 5.30161262e-02 2.02910900e-01 1.06769681e+00 6.67236626e-01 -3.48419815e-01 -5.75509787e-01 -1.31174818e-01 8.49917233e-01 -2.67695785e-02 -5.16766846e-01 -1.23352182e+00 -1.55123770e+00 7.11043358e-01 6.89645708e-01 3.53153646e-01 -2.07217336e-01 -8.74652788e-02 -3.66713181e-02 1.78415269e-01 3.20248567e-02 6.05385482e-01 7.12169334e-02 9.14981663e-02 -3.89028937e-01 -5.09323727e-04 8.32760453e-01 1.39687109e+00 9.51792419e-01 -2.50206850e-02 2.35916227e-01 5.55288970e-01 3.74725044e-01 2.76365548e-01 1.17255293e-01 -8.05343986e-01 3.10821295e-01 1.14246941e+00 -1.69909745e-01 -1.83766448e+00 -6.61881804e-01 -4.81146008e-01 -1.24647510e+00 -5.75778842e-01 1.36116207e-01 5.48129797e-01 -3.05966914e-01 1.59273982e+00 4.20058846e-01 1.91023886e-01 1.06981426e-01 7.64481425e-01 8.50754142e-01 3.49550128e-01 -5.00722647e-01 7.93372393e-02 1.33449769e+00 -5.80020130e-01 -7.57435560e-01 5.99057674e-01 7.89922774e-01 -3.43866199e-01 1.02537549e+00 3.28148343e-02 -1.15130746e+00 -1.87086195e-01 -1.31750691e+00 -1.88734546e-01 -7.68957496e-01 -9.12889093e-02 3.69461119e-01 5.76258063e-01 -1.23055863e+00 4.26943272e-01 -7.56730556e-01 -2.86353022e-01 -1.44140556e-01 1.27129123e-01 -7.13310599e-01 -2.27463514e-01 -1.36186266e+00 6.08347058e-01 4.40843612e-01 2.72215188e-01 -3.54195416e-01 -7.56092250e-01 -1.21628273e+00 -4.68220487e-02 4.16940123e-01 -7.24683225e-01 3.73136342e-01 2.90265590e-01 -1.33014596e+00 7.12351799e-01 6.26589417e-01 9.76620540e-02 3.12536359e-01 2.32939079e-01 -6.55003667e-01 3.54673892e-01 -2.12225720e-01 2.87970826e-02 6.88339770e-01 -1.35704911e+00 -9.44825858e-02 -6.95842266e-01 6.71649814e-01 5.05412579e-01 -7.26680577e-01 -4.34516490e-01 -4.47464168e-01 -6.56836987e-01 8.22015226e-01 -1.38841975e+00 1.18877411e-01 -4.36642319e-01 -6.88648820e-01 -5.95503114e-02 8.44102681e-01 -5.75029492e-01 1.59127831e+00 -2.18570352e+00 9.45879698e-01 4.72168088e-01 8.25673044e-01 -3.05589557e-01 1.01151906e-01 7.97767103e-01 6.64137378e-02 2.09838271e-01 -2.76266932e-01 -5.19472398e-02 3.73272717e-01 3.09362710e-01 5.23488596e-02 7.94328332e-01 -1.06412128e-01 9.15882409e-01 -1.13897920e+00 -6.69657052e-01 2.52702851e-02 6.94089592e-01 -6.03808761e-01 6.37397096e-02 3.12393188e-01 1.92429289e-01 -3.58611703e-01 5.45926630e-01 7.57596374e-01 -3.68805736e-01 6.06392443e-01 -6.78559244e-01 -6.84024841e-02 1.03630042e-02 -1.52137232e+00 1.59067082e+00 -2.62006611e-01 2.07397148e-01 2.39558723e-02 -8.36045802e-01 7.82935977e-01 6.45216703e-02 5.90447307e-01 -2.05917060e-01 1.11575969e-01 1.09785199e-01 4.51746222e-04 -1.51693657e-01 9.86864030e-01 3.15813065e-01 -1.47034287e-01 2.49730587e-01 -7.98715279e-02 -1.60047993e-01 6.08696714e-02 6.58190787e-01 1.22938883e+00 -1.93883553e-01 -6.48732856e-02 -7.74417043e-01 6.12120867e-01 -4.17310983e-01 5.88368535e-01 2.05728963e-01 1.62417069e-01 8.08076560e-01 7.24474669e-01 -8.69984850e-02 -8.93368006e-01 -1.55263555e+00 -3.42599154e-01 6.08517230e-01 7.32431173e-01 -1.04590940e+00 -7.48287439e-01 -4.53954160e-01 1.12991482e-01 2.70456105e-01 -6.75791085e-01 -7.30637729e-01 -4.39834774e-01 -8.36659133e-01 5.23385406e-01 3.40920120e-01 6.30891681e-01 -3.93267125e-01 -2.11306199e-01 2.11005788e-02 -1.65128723e-01 -1.25041091e+00 -7.66183674e-01 -3.77227336e-01 -6.97613537e-01 -1.34202790e+00 -3.26536417e-01 -7.64475465e-01 7.26786196e-01 6.42113209e-01 8.87955427e-01 -8.12328607e-02 -3.83050531e-01 8.72454762e-01 -5.48445880e-01 5.95921457e-01 -2.09675193e-01 4.87021059e-02 4.73408848e-01 6.62231743e-01 -2.52991527e-01 -8.39638174e-01 -7.12890327e-01 5.38662672e-01 -1.15787244e+00 2.14221120e-01 5.21749079e-01 5.79993546e-01 4.78354961e-01 5.34824789e-01 1.40360713e-01 -7.42271662e-01 5.56275070e-01 -5.22833765e-01 -2.33838335e-01 4.55505580e-01 -4.56989408e-01 1.82744890e-01 4.22609121e-01 -1.93910360e-01 -7.28319883e-01 -5.00901580e-01 7.08418190e-01 -4.91402537e-01 6.00530565e-01 9.36119914e-01 -4.61355925e-01 -4.75588441e-01 3.41930598e-01 2.77126729e-01 -8.36017877e-02 -4.56836075e-01 7.82068253e-01 4.78742719e-01 3.71094346e-01 -6.87031746e-01 1.15290976e+00 4.59779203e-01 4.52163041e-01 -1.27150714e+00 -2.88012385e-01 -2.34537914e-01 -1.09441912e+00 -1.67941287e-01 8.99717629e-01 -5.73896706e-01 -6.67582870e-01 4.72050726e-01 -7.50386298e-01 1.94624458e-02 -8.00145790e-02 5.60338438e-01 -5.68023086e-01 6.62155807e-01 -8.44819665e-01 -3.81665707e-01 2.53097564e-01 -8.00543070e-01 8.45005155e-01 -2.29813322e-01 5.25852665e-02 -1.11390460e+00 8.44714716e-02 3.91643256e-01 1.72015905e-01 4.94442463e-01 1.32666159e+00 -1.51174575e-01 -9.42248702e-01 -1.41234651e-01 -2.85143644e-01 1.29246905e-01 1.85901806e-01 1.79862052e-01 -1.77480161e-01 -3.33497316e-01 -1.49928823e-01 -2.16648169e-02 3.57830524e-01 -4.49791133e-01 8.15132260e-01 -4.70709085e-01 -1.12284303e-01 5.28866291e-01 1.41776454e+00 -1.52059078e-01 6.74466193e-01 1.76165432e-01 1.41446483e+00 7.13027596e-01 2.81923354e-01 3.18065852e-01 1.06743968e+00 8.77524376e-01 4.39044088e-01 2.29603097e-01 2.35057380e-02 -3.24293196e-01 2.93893129e-01 1.75382137e+00 -3.80725533e-01 3.63591075e-01 -1.10328770e+00 3.24755490e-01 -1.96053827e+00 -7.02579081e-01 -5.26145279e-01 2.44934225e+00 6.87862635e-01 -2.49196440e-01 2.46686768e-02 1.68203861e-01 8.62524807e-01 2.64669985e-01 -3.53519171e-01 2.65599880e-02 -4.72658992e-01 -2.32542425e-01 3.63102168e-01 5.38078487e-01 -8.35842967e-01 7.02138543e-01 5.46011925e+00 3.28668952e-01 -5.84287643e-01 -3.04244906e-02 -2.21774071e-01 3.22007775e-01 -4.32810456e-01 -2.74036266e-02 -1.85180143e-01 2.55147129e-01 6.11277699e-01 -4.49739099e-01 5.98681867e-01 5.99590063e-01 -4.60231870e-01 4.40298647e-01 -9.77430820e-01 1.11864591e+00 2.12642953e-01 -1.25166583e+00 3.97876263e-01 4.84878987e-01 9.06218767e-01 -3.58577907e-01 2.97575086e-01 2.43104771e-01 3.30043316e-01 -7.16106594e-01 7.70385742e-01 6.06690109e-01 6.05465472e-01 -6.36328995e-01 4.31682527e-01 8.70426930e-03 -1.65386581e+00 1.54168323e-01 -3.80006880e-01 4.08093035e-01 -2.54145771e-01 2.39911258e-01 -2.19423786e-01 1.25361681e+00 6.99082971e-01 9.04135406e-01 -9.05825138e-01 6.58164442e-01 1.52624145e-01 -2.70850640e-02 -4.34590839e-02 2.95012981e-01 1.94306538e-01 -1.04942775e+00 5.54098308e-01 9.21268642e-01 4.14498150e-01 3.57662886e-01 1.17078029e-01 7.66748369e-01 -9.11728367e-02 3.43567967e-01 -8.50534558e-01 -2.32994482e-01 5.29786885e-01 1.50114286e+00 -6.28108799e-01 -1.32193863e-01 -5.65707862e-01 8.27873409e-01 4.16577816e-01 4.97372657e-01 -8.29562962e-01 -5.91607153e-01 8.46619666e-01 3.58670443e-01 1.91967532e-01 -8.12932312e-01 1.72646046e-01 -1.63856947e+00 1.36951625e-01 -6.60528839e-01 4.54974294e-01 -7.82727361e-01 -8.95911217e-01 6.89756870e-01 8.48759264e-02 -1.11085248e+00 3.45566958e-01 -5.69752514e-01 -1.16684221e-01 3.76001507e-01 -8.92547548e-01 -1.45030236e+00 -6.10722780e-01 7.65835524e-01 -5.54583371e-01 1.32996798e-01 7.31947422e-01 3.19267005e-01 -6.70920610e-01 5.02621770e-01 2.68343359e-01 6.31358624e-02 2.18828186e-01 -1.83407831e+00 -2.00238496e-01 4.45051104e-01 3.07623655e-01 9.09935951e-01 3.98389757e-01 -3.30207348e-01 -2.29826021e+00 -1.34000909e+00 5.76773345e-01 -6.07882559e-01 1.24029231e+00 -6.89303041e-01 -1.14278138e+00 8.88452470e-01 -1.81370780e-01 6.55331314e-02 7.56144106e-01 1.83619693e-01 -8.39854121e-01 -2.38931254e-01 -9.21556711e-01 1.05917370e+00 1.64282143e+00 -9.29776847e-01 -3.79240096e-01 3.36370707e-01 8.11332583e-01 -1.81061640e-01 -1.80907547e+00 6.26667917e-01 4.09724653e-01 -1.05979943e+00 9.40137327e-01 -5.10639787e-01 -4.02986482e-02 -5.55327833e-01 -6.06239378e-01 -1.21485388e+00 -6.54682159e-01 -8.40456307e-01 -3.93014848e-01 1.05592358e+00 -3.92530113e-02 -8.70145679e-01 7.62974143e-01 3.90802890e-01 -2.94850200e-01 -8.78648341e-01 -8.55507612e-01 -1.10709500e+00 -7.81327188e-02 3.15582231e-02 7.67069161e-01 1.37604666e+00 5.40113866e-01 3.30842853e-01 -1.86313316e-01 5.26510060e-01 8.94652307e-01 3.62888455e-01 6.43796504e-01 -1.32578993e+00 3.54156941e-02 -4.87065226e-01 -1.19466150e+00 -7.02928066e-01 2.86217593e-02 -1.34981048e+00 -4.33731645e-01 -1.31619370e+00 1.76886156e-01 -5.32289743e-01 -3.30631196e-01 -3.77624333e-02 1.38664424e-01 2.31204435e-01 -2.02421118e-02 3.70903343e-01 -7.28927970e-01 1.07410455e+00 1.28463662e+00 -7.22198635e-02 -2.33308539e-01 -5.95466256e-01 -5.81530452e-01 5.49173594e-01 5.86905122e-01 1.87660813e-01 -6.76978946e-01 -1.03047587e-01 5.33022523e-01 1.66594293e-02 2.56450027e-01 -9.90731001e-01 5.13807356e-01 5.68724945e-02 -4.32101697e-01 -2.57602483e-01 4.61359262e-01 -6.93654418e-01 6.59275711e-01 1.47363901e-01 1.11706406e-01 1.15636945e-01 -3.67963731e-01 8.72769535e-01 -4.10018504e-01 3.56967986e-01 6.36384547e-01 1.89604685e-01 -3.35777909e-01 7.20345378e-01 3.99458595e-03 1.27122030e-01 1.08025146e+00 -3.64499718e-01 -4.72496003e-01 -3.66288036e-01 -1.03225183e+00 1.45425379e-01 7.98393428e-01 5.56034386e-01 7.37810254e-01 -1.96062732e+00 -4.55096215e-01 8.87143463e-02 6.23487711e-01 -4.29533049e-02 3.18767488e-01 1.16892481e+00 -6.04536116e-01 4.29957330e-01 -8.50208923e-02 -4.53692019e-01 -1.12961757e+00 6.85716808e-01 2.71066606e-01 -1.32137358e-01 -8.71666133e-01 2.72567034e-01 6.12371862e-01 -6.89723909e-01 1.20129950e-01 -3.12602818e-01 -2.49686211e-01 1.18879482e-01 -1.58809510e-03 8.54492426e-01 -6.17293306e-02 -1.18257558e+00 -2.91729063e-01 8.24192107e-01 1.13777474e-01 -3.82672064e-02 1.02009714e+00 -1.91235691e-01 -7.14952350e-01 7.67411888e-01 1.43891418e+00 9.70874503e-02 -5.99636495e-01 -4.18986410e-01 1.97101831e-01 -3.82942080e-01 -1.43355101e-01 9.64942351e-02 -9.71693695e-01 5.70389748e-01 -1.53625160e-01 9.55976367e-01 7.02619076e-01 2.02681646e-01 3.92892838e-01 4.38431144e-01 7.60820627e-01 -9.85799074e-01 4.59585696e-01 4.03784782e-01 1.40107644e+00 -4.64378923e-01 -4.46593389e-02 -9.72473025e-01 -5.09168088e-01 1.01110649e+00 5.82207739e-01 -5.49540929e-02 1.20735347e+00 -4.75276262e-01 -4.39501733e-01 -6.62742376e-01 -2.82337129e-01 1.80765986e-02 5.49585164e-01 3.90455306e-01 8.53807777e-02 4.99220163e-01 -1.03123046e-01 4.26532120e-01 -6.52853966e-01 -7.50926912e-01 8.06400657e-01 8.02454650e-01 -1.53336152e-01 -8.33927453e-01 -3.05205792e-01 2.35156432e-01 8.42981860e-02 1.32985637e-01 -6.05978668e-01 9.52063084e-01 -9.32319835e-02 8.93551350e-01 -4.03728299e-02 -1.10227501e+00 5.48162997e-01 -2.89516121e-01 6.69901192e-01 -7.03842700e-01 1.18586101e-01 -3.57665211e-01 -7.13992193e-02 -7.36237228e-01 -1.00013427e-01 -6.02156520e-01 -1.19383228e+00 -6.70005560e-01 -8.69942755e-02 3.94651979e-01 4.61195529e-01 2.82967687e-01 5.56260526e-01 3.60839128e-01 8.69163692e-01 -3.97183746e-01 -4.95605618e-01 -6.97322190e-01 -1.46838212e+00 5.70815623e-01 -8.12981278e-03 -1.16654992e+00 -5.39443076e-01 -2.40283191e-01]

[8.606475830078125, 7.729087829589844]

89800d38-e8ab-472f-86db-f403b1f1ae24

extreme-multi-label-learning-for-semantic

2106.12657

https://arxiv.org/abs/2106.12657v1

https://arxiv.org/pdf/2106.12657v1.pdf

Extreme Multi-label Learning for Semantic Matching in Product Search

We consider the problem of semantic matching in product search: given a customer query, retrieve all semantically related products from a huge catalog of size 100 million, or more. Because of large catalog spaces and real-time latency constraints, semantic matching algorithms not only desire high recall but also need to have low latency. Conventional lexical matching approaches (e.g., Okapi-BM25) exploit inverted indices to achieve fast inference time, but fail to capture behavioral signals between queries and products. In contrast, embedding-based models learn semantic representations from customer behavior data, but the performance is often limited by shallow neural encoders due to latency constraints. Semantic product search can be viewed as an eXtreme Multi-label Classification (XMC) problem, where customer queries are input instances and products are output labels. In this paper, we aim to improve semantic product search by using tree-based XMC models where inference time complexity is logarithmic in the number of products. We consider hierarchical linear models with n-gram features for fast real-time inference. Quantitatively, our method maintains a low latency of 1.25 milliseconds per query and achieves a 65% improvement of Recall@100 (60.9% v.s. 36.8%) over a competing embedding-based DSSM model. Our model is robust to weight pruning with varying thresholds, which can flexibly meet different system requirements for online deployments. Qualitatively, our method can retrieve products that are complementary to existing product search system and add diversity to the match set.

['Inderjit S. Dhillon', 'Japinder Singh', 'Vyacheslav Ievgrafov', 'Nikhil Shandilya', 'Qie Hu', 'Kedarnath Kolluri', 'Kai Zhong', 'Jiong Zhang', 'Choon-Hui Teo', 'Hsiang-Fu Yu', 'Daniel Jiang', 'Wei-Cheng Chang']

2021-06-23

['extreme-multi-label-classification']

['methodology']

[ 1.21487178e-01 -2.60181576e-01 -8.65813494e-01 -5.48204243e-01 -1.00552833e+00 -7.79365599e-01 2.38636628e-01 4.31241989e-01 -3.88406754e-01 -1.30594522e-01 -1.42990127e-02 -4.27958310e-01 -4.70660090e-01 -1.15368390e+00 -5.27026892e-01 -6.12096721e-03 1.48469746e-01 1.12673223e+00 2.49563605e-01 -2.15527222e-01 1.73069835e-01 2.21748158e-01 -1.77861845e+00 4.94670630e-01 6.31770492e-01 1.50249779e+00 2.99800217e-01 5.67690194e-01 -3.92763734e-01 4.63775992e-01 -3.55375141e-01 -7.24606454e-01 4.10749763e-01 1.67600557e-01 -9.50522721e-01 -3.08019608e-01 5.69118381e-01 -3.89184654e-01 -4.97521967e-01 1.01654851e+00 4.25893456e-01 1.18579930e-02 5.13565719e-01 -1.43931425e+00 -1.11962044e+00 6.27831817e-01 -4.78753150e-01 -1.42328134e-02 4.73698288e-01 -2.24998549e-01 1.69483984e+00 -7.52161562e-01 3.87435913e-01 1.37977040e+00 4.47359771e-01 3.21417600e-01 -1.51533735e+00 -7.75133193e-01 1.53080881e-01 3.67986649e-01 -1.50947130e+00 -2.57462054e-01 2.71127343e-01 -1.78759322e-01 1.56571400e+00 4.79559422e-01 2.05426335e-01 8.53258550e-01 -2.07834840e-02 9.91017222e-01 4.95229751e-01 -3.60943973e-01 2.29703486e-01 3.43716949e-01 5.79443991e-01 4.93746489e-01 4.15043741e-01 1.80410638e-01 -6.93380475e-01 -5.62257171e-01 3.10494304e-01 4.67734903e-01 3.19012463e-01 -2.17988659e-02 -6.24905169e-01 1.30502915e+00 4.37523723e-01 -1.01933097e-02 -3.93362522e-01 5.12315273e-01 3.65958780e-01 5.17260015e-01 1.90671504e-01 5.28420746e-01 -7.03065753e-01 -3.43150720e-02 -8.39538693e-01 4.68856454e-01 1.09989977e+00 1.49530804e+00 9.09889877e-01 -3.40596139e-01 -1.75596848e-01 1.06637239e+00 4.97657090e-01 8.40565085e-01 6.24723017e-01 -1.15955472e+00 5.53174853e-01 6.43348753e-01 1.68233942e-02 -1.00648808e+00 -2.36780807e-01 -3.64606082e-01 -2.89208174e-01 -4.25343096e-01 -2.55180206e-02 4.74040002e-01 -6.96226478e-01 1.64405489e+00 1.21605694e-01 -5.25793098e-02 -4.94931191e-02 7.12708890e-01 4.70416874e-01 4.77508932e-01 2.63982266e-01 2.11318716e-01 1.87296724e+00 -1.03732455e+00 -4.55745161e-01 -4.85591650e-01 1.02071118e+00 -9.27885354e-01 9.99708652e-01 1.13739192e-01 -1.07784963e+00 -4.43727881e-01 -1.15208805e+00 -2.57236719e-01 -7.00406253e-01 -3.04761231e-01 1.10038602e+00 6.71896636e-01 -8.40112925e-01 4.23004866e-01 -5.97056329e-01 -2.92248011e-01 1.00554392e-01 6.18818223e-01 1.58089191e-01 -4.32285786e-01 -1.35075247e+00 7.73200393e-01 3.06298167e-01 -1.68573648e-01 -3.56133223e-01 -7.87720919e-01 -8.48392785e-01 5.24256110e-01 4.24706221e-01 -8.04862916e-01 1.62709856e+00 -5.54771721e-01 -1.01396823e+00 6.36517107e-01 -3.86756361e-01 -4.99525934e-01 -1.30131379e-01 -2.61272341e-01 -8.67699385e-01 6.44185543e-02 3.68627042e-01 7.70692945e-01 4.18525219e-01 -7.49906361e-01 -1.03899765e+00 -5.11089206e-01 1.22963853e-01 -5.23827858e-02 -5.12533367e-01 1.51641577e-01 -7.83143818e-01 -3.33283097e-01 1.87975332e-01 -1.06505227e+00 -1.92892283e-01 -2.53859669e-01 1.20165022e-02 -4.26582307e-01 5.16674697e-01 -2.85803914e-01 1.51696289e+00 -2.15801978e+00 -5.04665315e-01 3.73179913e-01 1.86221436e-01 -2.79935300e-02 -5.37769437e-01 5.42258978e-01 3.28043818e-01 1.54471830e-01 4.25614655e-01 -2.17038989e-01 6.31730080e-01 2.97642440e-01 -4.11101937e-01 3.68768685e-02 -5.48018962e-02 1.53346086e+00 -7.20010340e-01 -5.55127025e-01 -1.42466510e-02 1.23347320e-01 -6.76710904e-01 -6.14372864e-02 -4.37780857e-01 -7.87747860e-01 -6.60137415e-01 1.13857901e+00 6.06815755e-01 -8.17224145e-01 1.76572353e-01 -2.57135570e-01 4.73175555e-01 3.46173644e-01 -1.11540008e+00 1.64846456e+00 -1.05477440e+00 2.37808123e-01 -1.15404658e-01 -7.87561238e-01 8.10698986e-01 9.87215713e-02 5.50759375e-01 -1.45305932e+00 -1.96849436e-01 5.46195805e-01 -3.53677809e-01 -2.22810820e-01 7.75697708e-01 3.00708085e-01 -4.95341152e-01 5.93158126e-01 -3.75921018e-02 3.38113517e-01 1.11935228e-01 1.14632025e-01 1.23397672e+00 -4.78081077e-01 -8.56526121e-02 -8.18890855e-02 5.74531183e-02 2.14185819e-01 4.55647290e-01 1.08396518e+00 1.16199918e-01 6.92399368e-02 5.72239608e-02 -5.48846126e-01 -8.97980809e-01 -1.08058298e+00 -2.18932450e-01 1.63679600e+00 5.56288123e-01 -5.63744009e-01 -2.65087038e-01 -4.18049097e-01 7.83604026e-01 7.26194084e-01 -1.12319797e-01 -4.48424995e-01 -4.75793660e-01 -5.45379996e-01 4.05596465e-01 6.50704682e-01 2.25322366e-01 -8.82337809e-01 -5.68643332e-01 5.59987783e-01 -5.30047044e-02 -1.32401729e+00 -7.90223181e-01 2.82934427e-01 -7.91662097e-01 -8.75104070e-01 -2.56430089e-01 -1.03010786e+00 7.22106695e-02 5.28444588e-01 1.50831556e+00 -5.01903594e-02 -5.15289903e-01 3.74710530e-01 -2.32384130e-01 -1.70647919e-01 -1.16948687e-01 4.48067993e-01 3.58284749e-02 -1.87636659e-01 1.33293736e+00 -3.78030717e-01 -9.15423512e-01 6.57004058e-01 -9.37054515e-01 -4.23162252e-01 7.81202197e-01 8.64386499e-01 6.27802968e-01 1.99442193e-01 6.44593775e-01 -8.85307908e-01 7.17642426e-01 -6.71805561e-01 -7.65220881e-01 6.03209436e-01 -1.55573630e+00 3.39191884e-01 3.34788650e-01 -7.08991170e-01 -6.87348843e-01 4.20040116e-02 -3.35212424e-03 -3.95086706e-01 1.45562574e-01 5.21229208e-01 1.65789753e-01 1.30045503e-01 5.54405987e-01 1.20446876e-01 -1.08966462e-01 -6.14258587e-01 5.95167696e-01 1.14048111e+00 3.34705561e-01 -5.82190752e-01 2.51293868e-01 1.18101716e-01 -2.07202017e-01 -2.27335140e-01 -8.04101825e-01 -1.06136131e+00 -6.65189177e-02 4.31954443e-01 4.35889930e-01 -1.10096335e+00 -1.42819619e+00 -2.34957740e-01 -9.80559826e-01 -4.25615981e-02 -3.98627780e-02 4.34490204e-01 -5.20857036e-01 8.01759288e-02 -1.10478699e+00 -6.80830181e-01 -7.28928208e-01 -1.01282108e+00 1.64795494e+00 -2.67116986e-02 -5.28239965e-01 -7.18182743e-01 -2.75168657e-01 5.43843031e-01 6.14358485e-01 -5.51361859e-01 1.28483570e+00 -8.34287286e-01 -9.79093730e-01 -5.45502484e-01 -4.28477645e-01 -1.55161262e-01 -7.45450035e-02 -7.43254602e-01 -8.24038565e-01 -3.65985245e-01 -3.80024195e-01 -3.90457958e-01 8.89537752e-01 1.79409221e-01 1.08196282e+00 -4.31335896e-01 -6.05102956e-01 4.59549397e-01 1.71214259e+00 4.17324871e-01 1.89440459e-01 2.70623893e-01 2.95713872e-01 5.51388323e-01 9.12187636e-01 3.80812496e-01 4.22792137e-01 1.14921761e+00 2.10869759e-01 2.56257445e-01 3.13811898e-02 -7.56054401e-01 1.33377194e-01 6.93955541e-01 7.69587994e-01 -3.76324922e-01 -6.99135482e-01 5.27639806e-01 -2.11186171e+00 -6.40916586e-01 1.21129945e-01 2.29281950e+00 6.09127760e-01 2.80477017e-01 -6.10437691e-02 -2.27310285e-01 3.60776991e-01 -1.02391630e-01 -9.34907019e-01 -5.67614198e-01 9.03148428e-02 3.66106093e-01 1.10978746e+00 5.66711247e-01 -8.64703059e-01 9.82703030e-01 6.60588503e+00 1.19752347e+00 -6.85488462e-01 3.49075526e-01 3.12773675e-01 -4.44405466e-01 -6.67069554e-01 -4.74600308e-02 -1.38934743e+00 7.16563404e-01 1.44314325e+00 -2.55918801e-01 5.55219531e-01 1.13166189e+00 -3.94942373e-01 9.91270468e-02 -1.45886469e+00 1.40880609e+00 -1.34653866e-01 -1.36913979e+00 7.60469437e-02 3.53272974e-01 4.21895951e-01 9.31736976e-02 3.00652713e-01 6.96843863e-01 6.99978471e-01 -9.69445884e-01 4.97591317e-01 1.88549384e-01 7.98557281e-01 -6.44683123e-01 6.95959389e-01 2.36799568e-01 -1.55582380e+00 -4.27079856e-01 -5.34176350e-01 2.85480380e-01 3.59999031e-01 5.50062358e-01 -4.77326244e-01 2.42796198e-01 7.58327007e-01 4.18303162e-01 -4.86172944e-01 6.74526513e-01 4.72590268e-01 2.53644377e-01 -6.39096439e-01 -3.31323653e-01 3.20993602e-01 -1.89990569e-02 -2.75822908e-01 1.14890981e+00 4.27288324e-01 -1.53776929e-01 3.62159640e-01 8.80959630e-01 -1.11832418e-01 1.44461334e-01 -6.41164005e-01 -2.47518793e-01 9.90778804e-01 9.19228673e-01 -3.21987212e-01 -4.19547379e-01 -8.16153228e-01 1.15371764e+00 2.15110824e-01 4.17005897e-01 -8.08845878e-01 -3.68270129e-01 9.58796144e-01 3.52644697e-02 3.88266444e-01 1.93164110e-01 -1.98720962e-01 -7.78314352e-01 2.95810699e-01 -7.58004487e-01 6.55675888e-01 -3.46608698e-01 -1.42083013e+00 2.84073114e-01 -1.73503309e-01 -9.69596863e-01 -5.92020035e-01 -6.21895134e-01 1.20251067e-01 9.01847839e-01 -1.46803546e+00 -1.12751722e+00 8.94823819e-02 2.69653320e-01 6.17046535e-01 -1.95035785e-01 1.10325313e+00 9.30903912e-01 -1.15378417e-01 1.13334084e+00 1.99082300e-01 -2.72998959e-01 4.54122484e-01 -1.09567249e+00 6.99620605e-01 -3.17203663e-02 5.49009502e-01 7.51109362e-01 2.47490808e-01 -4.97461826e-01 -1.76457942e+00 -1.01554978e+00 1.49970925e+00 -5.35761476e-01 7.66603231e-01 -5.14571071e-01 -7.49660075e-01 6.15971327e-01 -2.37316296e-01 -1.93185862e-02 9.16024506e-01 6.71448827e-01 -9.68529463e-01 -4.74880278e-01 -8.56905460e-01 5.49777508e-01 1.22395611e+00 -9.89987910e-01 -1.96938068e-01 5.24027526e-01 1.34385312e+00 3.33274975e-02 -1.23920715e+00 3.55987817e-01 8.93005192e-01 -4.18380708e-01 1.36252427e+00 -7.65287876e-01 -4.10271958e-02 3.81798744e-02 -6.12107098e-01 -6.81245327e-01 -5.97596049e-01 -3.46239775e-01 -4.81555730e-01 8.83302093e-01 8.00105155e-01 -7.79640794e-01 9.92245138e-01 9.80984449e-01 3.42887849e-01 -8.16462815e-01 -5.29788375e-01 -1.10395110e+00 -1.84977978e-01 -5.10290802e-01 1.08274615e+00 5.13110757e-01 5.13755940e-02 5.23996592e-01 -9.51277912e-02 2.16262817e-01 5.93033135e-01 9.03628469e-01 2.29858100e-01 -1.31893098e+00 -7.21546113e-01 -6.65263474e-01 -3.67813766e-01 -1.60354018e+00 2.26900339e-01 -1.12156308e+00 -3.63198638e-01 -1.15410125e+00 3.23507249e-01 -8.08659196e-01 -6.66505933e-01 5.27203500e-01 2.00469494e-01 9.48348641e-02 6.52971789e-02 3.21481764e-01 -8.30128789e-01 2.03647092e-01 4.98191446e-01 -5.82265675e-01 -1.09036379e-01 1.61633864e-01 -8.22069108e-01 1.65525153e-02 5.95680416e-01 -5.74674726e-01 -5.54583728e-01 -5.99932551e-01 6.85264111e-01 2.86224425e-01 5.23297936e-02 -2.33571038e-01 6.27226591e-01 8.55308697e-02 1.14548728e-02 -6.00540817e-01 5.39992154e-01 -9.76469636e-01 2.70933658e-01 4.69805598e-01 -6.69677258e-01 3.48458499e-01 -3.88275944e-02 7.40172088e-01 -4.19023901e-01 -4.54834878e-01 1.28685534e-01 -4.06883359e-02 -8.32181275e-01 4.37487930e-01 -5.55753373e-02 -7.61505216e-02 6.88522279e-01 -9.95213613e-02 -3.50874543e-01 -2.31492296e-01 -5.29579222e-01 6.32288456e-01 1.20874345e-01 8.52329135e-01 4.14639622e-01 -1.69284272e+00 -2.53355145e-01 2.51435816e-01 6.52507961e-01 -4.27115232e-01 7.37042129e-02 3.02925467e-01 -1.94825694e-01 1.10360193e+00 4.00436550e-01 -4.98819917e-01 -1.03467870e+00 9.74345505e-01 -2.83594001e-02 -2.39720911e-01 -2.62037396e-01 6.91253543e-01 2.69009583e-02 -3.32650542e-01 4.52633381e-01 -1.18806720e-01 2.86059409e-01 7.05802301e-03 4.89527553e-01 3.79388571e-01 3.15583140e-01 -2.03595966e-01 -2.74547309e-01 5.82274139e-01 -3.62300098e-01 -3.69934924e-02 8.58772635e-01 -1.82047740e-01 1.25659123e-01 1.16874196e-01 1.81626439e+00 -4.22930717e-01 -5.35905004e-01 -7.94535756e-01 6.24694347e-01 -6.34953260e-01 1.26491219e-01 -7.46913493e-01 -1.06631899e+00 4.88236517e-01 7.87023723e-01 2.57371038e-01 1.02582943e+00 4.26865578e-01 1.40351868e+00 6.52767897e-01 7.91934431e-01 -1.33638191e+00 -1.02665424e-01 6.07916415e-02 2.14582950e-01 -1.35238242e+00 -3.65085930e-01 -4.42490071e-01 -3.77925694e-01 7.37329781e-01 4.30352569e-01 2.35890254e-01 7.84981132e-01 2.93724686e-01 -1.03064485e-01 -4.54891711e-01 -1.25598979e+00 -4.61092025e-01 1.97160989e-01 1.47007495e-01 1.59257010e-01 4.16732728e-01 -9.21742991e-02 5.78552425e-01 -1.78642884e-01 -8.78502056e-02 -3.92558426e-01 7.33609676e-01 -5.42135179e-01 -1.35481644e+00 1.27604544e-01 8.07504416e-01 -3.20733130e-01 -2.71731406e-01 9.74840224e-02 3.92585963e-01 -1.53390303e-01 1.24326706e+00 3.98773730e-01 -5.17711759e-01 4.92316216e-01 2.23481357e-01 2.21170589e-01 -5.24606943e-01 -4.18007016e-01 -1.44789778e-02 1.34125695e-01 -1.08242917e+00 1.78901151e-01 -4.52991009e-01 -1.04040325e+00 -5.67425251e-01 -5.58284819e-01 1.94256574e-01 6.50999606e-01 3.94089937e-01 9.29764330e-01 -5.86759411e-02 6.35105371e-01 9.41684917e-02 -1.13117993e+00 -7.46740997e-01 -7.84525037e-01 5.32332838e-01 1.37333600e-02 -6.93176091e-01 -1.47673637e-01 -3.36067587e-01]

[11.234721183776855, 7.330539703369141]

c171a573-7a85-4b48-9542-7a7f1ff7aa24

first-session-adaptation-a-strong-replay-free

2303.13199

https://arxiv.org/abs/2303.13199v1

https://arxiv.org/pdf/2303.13199v1.pdf

First Session Adaptation: A Strong Replay-Free Baseline for Class-Incremental Learning

In Class-Incremental Learning (CIL) an image classification system is exposed to new classes in each learning session and must be updated incrementally. Methods approaching this problem have updated both the classification head and the feature extractor body at each session of CIL. In this work, we develop a baseline method, First Session Adaptation (FSA), that sheds light on the efficacy of existing CIL approaches and allows us to assess the relative performance contributions from head and body adaption. FSA adapts a pre-trained neural network body only on the first learning session and fixes it thereafter; a head based on linear discriminant analysis (LDA), is then placed on top of the adapted body, allowing exact updates through CIL. FSA is replay-free i.e.~it does not memorize examples from previous sessions of continual learning. To empirically motivate FSA, we first consider a diverse selection of 22 image-classification datasets, evaluating different heads and body adaptation techniques in high/low-shot offline settings. We find that the LDA head performs well and supports CIL out-of-the-box. We also find that Featurewise Layer Modulation (FiLM) adapters are highly effective in the few-shot setting, and full-body adaption in the high-shot setting. Second, we empirically investigate various CIL settings including high-shot CIL and few-shot CIL, including settings that have previously been used in the literature. We show that FSA significantly improves over the state-of-the-art in 15 of the 16 settings considered. FSA with FiLM adapters is especially performant in the few-shot setting. These results indicate that current approaches to continuous body adaptation are not working as expected. Finally, we propose a measure that can be applied to a set of unlabelled inputs which is predictive of the benefits of body adaptation.

['Richard E. Turner', 'Rahaf Aljundi', 'Daniel Olmeda Reino', 'Yuriko Kobe', 'Aristeidis Panos']

2023-03-23

['class-incremental-learning']

['computer-vision']

[ 4.52588201e-01 -5.38734868e-02 -4.66053635e-01 -4.01972592e-01 -7.39719570e-01 -2.88923591e-01 7.47300386e-01 2.45060965e-01 -7.14853942e-01 5.93999088e-01 -3.20216082e-02 -1.02109507e-01 -2.61244833e-01 -5.22429943e-01 -7.42994726e-01 -7.05032647e-01 -1.38462827e-01 4.77909148e-01 5.65375626e-01 -1.58962727e-01 6.34716451e-02 4.36470985e-01 -1.99063814e+00 3.49939644e-01 4.82644886e-01 9.74145532e-01 -1.12402342e-01 9.19993937e-01 -4.18771058e-02 7.52021313e-01 -6.54725015e-01 -4.25197542e-01 4.18714136e-01 -7.05635190e-01 -9.18060243e-01 2.25158080e-01 8.83316159e-01 -2.42610827e-01 -4.90696393e-02 5.36004245e-01 8.11741829e-01 3.40403616e-01 5.50945818e-01 -1.16800714e+00 -3.31192195e-01 3.95012438e-01 -4.64860499e-01 5.64379156e-01 3.55483711e-01 3.04148644e-01 5.96723080e-01 -9.65075672e-01 7.05761015e-01 1.00855565e+00 9.68366802e-01 8.01368713e-01 -1.44924688e+00 -6.36933804e-01 4.32932079e-01 2.23516807e-01 -1.09180057e+00 -7.67902970e-01 4.78787929e-01 -4.93762195e-01 1.14068818e+00 2.99124241e-01 7.13495731e-01 9.81742561e-01 2.85441935e-01 8.07683945e-01 1.08754158e+00 -8.91130269e-01 3.38244051e-01 2.36785993e-01 4.33589369e-01 6.67972505e-01 1.26517788e-01 1.43486522e-02 -8.38814259e-01 -1.62928421e-02 2.06752792e-01 -2.30143458e-01 -1.60869777e-01 -4.84107107e-01 -8.70006442e-01 7.29580879e-01 2.17305481e-01 2.43745014e-01 -3.31520230e-01 1.76447537e-02 6.69031382e-01 5.84698915e-01 5.85946918e-01 3.40365827e-01 -4.19850856e-01 -1.26600802e-01 -1.14850950e+00 2.61252970e-01 9.17658091e-01 6.52451754e-01 7.51280129e-01 -6.22859672e-02 -5.15227735e-01 1.07913983e+00 -1.52747825e-01 1.65916562e-01 9.17019129e-01 -8.73292863e-01 7.62091279e-02 3.49995971e-01 -3.70589674e-01 -2.74937451e-01 -4.99214381e-01 -5.54894149e-01 -3.72458935e-01 4.54807639e-01 5.17407477e-01 -5.57223661e-03 -1.27198732e+00 1.85480320e+00 6.21003985e-01 1.38857812e-01 1.77058633e-02 4.01715547e-01 7.11717904e-01 2.26149037e-01 2.89917350e-01 -4.99687463e-01 1.15145159e+00 -1.00787938e+00 -5.01002550e-01 -3.21335942e-01 5.67308247e-01 -7.08858609e-01 1.37201238e+00 4.63060021e-01 -9.74185348e-01 -8.49986434e-01 -1.22814143e+00 1.77388519e-01 -4.44374710e-01 2.25785207e-02 5.18468142e-01 8.13951135e-01 -1.08354568e+00 8.10791731e-01 -9.29939926e-01 -7.93144166e-01 3.82796794e-01 6.00353956e-01 -4.09042716e-01 -1.71069771e-01 -8.67252588e-01 1.03062308e+00 3.62762868e-01 -2.99984068e-01 -7.03500926e-01 -6.88284874e-01 -6.47567034e-01 -1.70694530e-01 4.11095768e-01 -7.88648963e-01 1.62745857e+00 -1.32084906e+00 -1.67758739e+00 8.66788208e-01 -1.66395307e-01 -5.08841455e-01 4.85714376e-01 -1.89755261e-01 -3.52928817e-01 1.33157313e-01 -5.37740858e-03 7.77692497e-01 1.05812752e+00 -1.10769212e+00 -6.08664274e-01 -3.93247575e-01 1.03866257e-01 2.80776769e-01 -4.76365298e-01 -1.72483981e-01 -6.13083601e-01 -4.44048017e-01 -8.99818987e-02 -1.09017122e+00 1.19281739e-01 -9.91194472e-02 -2.03246716e-03 -9.04066488e-02 7.30790496e-01 -3.78922403e-01 1.41857553e+00 -2.19804668e+00 8.79850090e-02 7.17086196e-02 -2.04044715e-01 4.55316693e-01 -1.15497194e-01 3.69535089e-01 -3.22064400e-01 -1.50818661e-01 -2.51113027e-01 -6.59719884e-01 -1.34313613e-01 4.65900302e-01 1.71777420e-02 4.29383487e-01 -5.80843724e-02 5.73713243e-01 -6.21640503e-01 -5.47882259e-01 4.41376343e-02 2.44667679e-01 -7.15091586e-01 1.01440534e-01 1.11337483e-01 6.80088475e-02 2.08746672e-01 5.72110295e-01 3.11891258e-01 -1.51919320e-01 -1.49594352e-01 -2.60304242e-01 -4.89573032e-02 -3.67785469e-02 -1.08135116e+00 1.57153058e+00 -4.50383604e-01 6.11168265e-01 -3.13366562e-01 -8.85313630e-01 5.20190835e-01 2.20215619e-01 3.29019278e-01 -6.67672098e-01 -5.73561601e-02 1.14592746e-01 1.63813114e-01 -5.47958910e-01 2.86993533e-01 -2.79856771e-01 1.67205527e-01 3.82758737e-01 5.82354367e-01 2.64874566e-02 4.50183243e-01 2.35103980e-01 1.18784225e+00 2.62664467e-01 6.94392264e-01 -1.28537312e-01 5.18084228e-01 -2.07912534e-01 3.36277425e-01 1.02579296e+00 -3.44559640e-01 6.29831076e-01 1.26600996e-01 -3.24370414e-01 -7.58835673e-01 -9.76984739e-01 -2.40399584e-01 1.69412315e+00 -1.40269369e-01 -4.08942819e-01 -9.41291451e-01 -9.14615929e-01 7.49535784e-02 7.74134874e-01 -8.89847636e-01 -5.45975864e-01 -3.65822464e-01 -9.11956549e-01 4.05993074e-01 5.69524348e-01 5.43061256e-01 -1.12197351e+00 -8.01898479e-01 2.34315589e-01 2.39388868e-01 -6.80444479e-01 -3.66778940e-01 6.21922791e-01 -7.98871696e-01 -1.08862710e+00 -6.35392189e-01 -3.53531003e-01 5.41666448e-01 -5.57399914e-02 9.25470948e-01 -1.87105164e-02 -2.94591010e-01 9.11873639e-01 -4.45680737e-01 -5.56225955e-01 -4.83710825e-01 3.56310248e-01 1.74769521e-01 9.75131765e-02 3.30858320e-01 -3.67150068e-01 -4.90537643e-01 2.75894701e-01 -8.92349184e-01 -1.70472622e-01 7.68551290e-01 9.42304552e-01 5.29519796e-01 -3.50461185e-01 6.64541483e-01 -1.30477273e+00 6.01427436e-01 -4.27003562e-01 -1.01246424e-01 2.65156806e-01 -9.99694407e-01 -1.01518570e-04 5.07173836e-01 -7.58435309e-01 -1.07840049e+00 1.90928981e-01 -3.24290484e-01 -4.21522111e-01 -1.94097742e-01 5.16086757e-01 7.47821182e-02 -3.47981453e-01 1.16322279e+00 1.99499968e-02 1.46683201e-01 -4.67975497e-01 2.28005528e-01 6.14956498e-01 6.65666342e-01 -5.39459169e-01 6.22291505e-01 2.36624077e-01 -1.87023431e-01 -7.93336749e-01 -9.21947777e-01 -5.74351966e-01 -8.79726410e-01 -4.46222752e-01 5.80390155e-01 -6.77666783e-01 -2.70182937e-01 6.46582067e-01 -4.97618824e-01 -8.63862455e-01 -8.70174527e-01 4.04268056e-01 -5.87688684e-01 1.44441426e-01 -5.38814902e-01 -5.94794989e-01 -2.62118459e-01 -9.51450408e-01 8.67425740e-01 2.95323789e-01 -4.52897400e-01 -8.89677763e-01 2.65740484e-01 9.04095024e-02 4.20435160e-01 6.82772845e-02 7.05927372e-01 -8.00798059e-01 -3.21942358e-03 -3.00569713e-01 2.95486957e-01 4.85189229e-01 1.60783499e-01 -1.56106248e-01 -1.38828576e+00 -6.05212569e-01 -1.80557102e-01 -4.74720746e-01 1.12034130e+00 3.49665999e-01 9.60420251e-01 -1.54626578e-01 -2.06249610e-01 5.88335574e-01 1.15633762e+00 2.45473400e-01 5.45595169e-01 5.78195453e-01 3.00417215e-01 3.81481171e-01 5.14541686e-01 2.21801236e-01 1.99392349e-01 8.43897641e-01 3.98466811e-02 -8.05199519e-02 -6.18458688e-01 -1.18209654e-02 6.03714705e-01 6.20264769e-01 -1.16921216e-01 -1.46744400e-01 -7.43465543e-01 4.96047318e-01 -1.63013685e+00 -1.02974081e+00 3.97849381e-01 2.48356915e+00 1.05769122e+00 4.58120733e-01 4.59686756e-01 4.28253859e-01 3.08122009e-01 -6.94150999e-02 -6.59347236e-01 -5.40107548e-01 1.89293906e-01 5.50071955e-01 4.59420115e-01 5.41270614e-01 -1.26739728e+00 7.43180156e-01 6.90740681e+00 7.44782567e-01 -1.23564947e+00 3.77173632e-01 4.87698495e-01 -3.51770222e-01 2.24388346e-01 -1.14795469e-01 -1.04668391e+00 4.32199001e-01 1.27875471e+00 -6.71370104e-02 3.35953802e-01 9.69519794e-01 -3.28322262e-01 -5.61625540e-01 -1.39859879e+00 6.67983294e-01 5.47214925e-01 -9.85840023e-01 -6.54500797e-02 -1.73260108e-01 5.99399567e-01 8.87207240e-02 1.37043372e-01 7.27872670e-01 -7.75757525e-03 -5.54476321e-01 7.50211298e-01 5.08278668e-01 9.36500609e-01 -5.86408615e-01 6.95722401e-01 2.89063513e-01 -1.04372942e+00 -2.81370074e-01 -5.79610094e-02 9.35964473e-03 -1.54944196e-01 2.88266063e-01 -9.33326721e-01 3.87340367e-01 7.16840327e-01 4.47114289e-01 -1.11002147e+00 1.21498358e+00 -7.68214539e-02 9.23946500e-01 -4.03731823e-01 2.87714303e-01 -1.45635149e-03 4.11326110e-01 4.43444461e-01 1.53059721e+00 7.57979527e-02 3.29768099e-02 2.53265142e-01 1.28093824e-01 2.03372210e-01 2.36874908e-01 -5.02413869e-01 3.84477437e-01 2.72968620e-01 1.03630817e+00 -8.70124221e-01 -5.67380071e-01 -3.43857259e-01 1.08245707e+00 2.22863197e-01 2.84810007e-01 -4.95756716e-01 -3.02138329e-01 2.95442224e-01 1.83063492e-01 3.75241429e-01 1.70137435e-01 1.37485430e-01 -9.89008188e-01 -2.20900282e-01 -1.03084767e+00 8.37104440e-01 -5.83914220e-01 -1.18029189e+00 5.15107393e-01 4.57803935e-01 -1.07354498e+00 -5.47272563e-01 -4.74356890e-01 -5.38566947e-01 5.11851847e-01 -1.23545349e+00 -9.93870556e-01 -4.24356163e-01 6.49292946e-01 8.75586092e-01 -1.67118758e-01 9.30988908e-01 2.65631586e-01 -7.21730053e-01 1.01020300e+00 -6.01972193e-02 -3.23132336e-01 9.51602519e-01 -1.18470132e+00 1.69054389e-01 8.80006492e-01 2.92295933e-01 6.09557152e-01 7.34383106e-01 -5.15636146e-01 -9.50971603e-01 -9.77339506e-01 6.05554104e-01 -3.32648516e-01 1.53938681e-01 -2.81933904e-01 -1.16702867e+00 6.56002998e-01 1.41950548e-02 1.82648346e-01 7.82954931e-01 2.95025110e-01 -1.67880401e-01 -3.92709106e-01 -1.10613036e+00 3.62499446e-01 9.12428081e-01 -5.57976782e-01 -6.01089299e-01 2.03845203e-01 3.69494438e-01 -5.50447881e-01 -7.22482681e-01 3.89148504e-01 8.25910628e-01 -1.05598962e+00 9.03195918e-01 -4.51168835e-01 6.93173930e-02 -7.81194866e-02 -1.33775696e-01 -1.37947237e+00 -3.10683668e-01 -3.47365886e-01 -2.63673723e-01 1.15868831e+00 3.17313075e-01 -7.45403767e-01 6.07840478e-01 4.75448638e-01 -2.28929266e-01 -8.89099956e-01 -9.16937411e-01 -9.73689318e-01 -8.05154890e-02 -4.68079329e-01 2.79334962e-01 7.33742297e-01 -4.71646748e-02 4.55961853e-01 -1.81551337e-01 -2.28500769e-01 4.31329697e-01 -2.49322131e-01 1.02617049e+00 -1.09845698e+00 -6.83289886e-01 -3.50883663e-01 -4.23584431e-01 -5.30905187e-01 4.12833095e-02 -9.45803225e-01 3.79781686e-02 -1.09558785e+00 2.53643095e-01 -2.28463814e-01 -5.19674957e-01 9.11051631e-01 -3.93353373e-01 5.32464445e-01 3.74527007e-01 3.30652356e-01 -5.59857428e-01 1.84973389e-01 6.81489825e-01 -9.00995359e-03 -4.70571637e-01 1.32529303e-01 -4.30824250e-01 7.81288862e-01 5.92156947e-01 -3.34076375e-01 -5.70018828e-01 1.19362950e-01 -1.09763727e-01 -4.20113802e-01 2.90962249e-01 -1.40379047e+00 3.42351854e-01 1.51592955e-01 5.99949539e-01 -2.09035844e-01 3.08828384e-01 -5.66777468e-01 1.59991398e-01 5.13779104e-01 -4.38395888e-01 -1.46102846e-01 5.72898507e-01 4.92550671e-01 5.07497899e-02 -5.20512998e-01 1.08876789e+00 -6.67640716e-02 -9.23745096e-01 8.46689939e-02 -3.38669509e-01 9.21062082e-02 9.98029232e-01 -4.89504248e-01 -1.05828322e-01 -2.82316983e-01 -1.00733137e+00 -1.21747531e-01 4.15989876e-01 1.69580013e-01 2.74871588e-01 -1.08987415e+00 -2.92227536e-01 5.15069962e-01 3.25192153e-01 -4.37561691e-01 1.76458016e-01 1.08618212e+00 -1.70895964e-01 -3.23045924e-02 -1.89811647e-01 -6.97061956e-01 -1.56518269e+00 5.02006650e-01 3.79777163e-01 -1.89735904e-01 -5.76972306e-01 1.01102507e+00 -1.32242171e-02 -2.60100603e-01 4.27307457e-01 -1.92780271e-01 -2.05255404e-01 4.05799270e-01 5.54163575e-01 2.80905902e-01 4.20028985e-01 -5.09173214e-01 -2.38432765e-01 5.52457690e-01 -3.53475690e-01 -2.46332988e-01 1.09789681e+00 -9.36252177e-02 3.20706815e-01 9.26467538e-01 1.02926469e+00 -1.67316511e-01 -1.22177589e+00 -2.49529690e-01 -6.48260713e-02 -3.68342638e-01 9.43929330e-02 -9.87879813e-01 -7.43972600e-01 6.82239652e-01 1.07681489e+00 -9.54782069e-02 1.41270697e+00 -6.81760386e-02 3.58341694e-01 4.97219712e-01 1.96808472e-01 -1.19979417e+00 2.30096474e-01 2.49909088e-01 7.48582423e-01 -1.07630873e+00 3.75840396e-01 -2.15233844e-02 -6.16125882e-01 1.01714694e+00 6.58766747e-01 1.61524102e-01 5.53244233e-01 1.91885263e-01 -1.86736267e-02 -1.59060091e-01 -9.04756188e-01 -3.75150234e-01 4.62841809e-01 5.06479800e-01 4.06386167e-01 -2.72315115e-01 -2.76876718e-01 3.52994442e-01 -1.13165170e-01 1.84172168e-01 3.24153066e-01 1.12380874e+00 -4.28863972e-01 -1.10975969e+00 -3.30896556e-01 6.86328113e-01 -3.65194350e-01 7.51853883e-02 -3.68277073e-01 1.16935658e+00 4.81155425e-01 5.56163609e-01 -9.23731737e-03 -3.84001791e-01 5.58680236e-01 7.14483678e-01 7.52704561e-01 -8.38388026e-01 -7.98103631e-01 1.13170601e-01 9.00182724e-02 -6.08298123e-01 -6.52242064e-01 -9.97553229e-01 -9.16775823e-01 2.86199991e-02 -5.55016220e-01 -6.64738119e-02 5.68234563e-01 1.05314124e+00 1.57790333e-01 5.69131434e-01 2.91594565e-01 -1.13961399e+00 -6.42078757e-01 -1.06672776e+00 -3.53951901e-01 4.17701215e-01 3.49593341e-01 -8.96402776e-01 -5.22029877e-01 3.58329684e-01]

[9.87344741821289, 3.083477735519409]

0111d12d-ced4-4b66-8a3a-536462e9cb7b

unleashing-the-power-of-visual-prompting-at

2212.10556

https://arxiv.org/abs/2212.10556v2

https://arxiv.org/pdf/2212.10556v2.pdf

Unleashing the Power of Visual Prompting At the Pixel Level

This paper presents a simple and effective visual prompting method for adapting pre-trained models to downstream recognition tasks. Our method includes two key designs. First, rather than directly adding together the prompt and the image, we treat the prompt as an extra and independent learnable component. We show that the strategy of reconciling the prompt and the image matters, and find that warping the prompt around a properly shrinked image empirically works the best. Second, we re-introduce two "old tricks" commonly used in building transferable adversarial examples, i.e., input diversity and gradient normalization, into visual prompting. These techniques improve optimization and enable the prompt to generalize better. We provide extensive experimental results to demonstrate the effectiveness of our method. Using a CLIP model, our prompting method sets a new record of 82.8% average accuracy across 12 popular classification datasets, substantially surpassing the prior art by +5.6%. It is worth noting that this prompting performance already outperforms linear probing by +2.1% and can even match fully fine-tuning in certain datasets. In addition, our prompting method shows competitive performance across different data scales and against distribution shifts. The code is publicly available at https://github.com/UCSC-VLAA/EVP.

['Cihang Xie', 'Yuyin Zhou', 'Alan Yuille', 'Huiyu Wang', 'Chen Wei', 'Xianhang Li', 'Junyang Wu']

2022-12-20

['visual-prompting']

['computer-vision']

[ 2.89372444e-01 -3.21776532e-02 -3.48001182e-01 -4.53292668e-01 -1.12351632e+00 -1.04492092e+00 7.32644320e-01 -3.33238184e-01 -4.51158375e-01 6.90521479e-01 3.35332006e-01 -3.65623325e-01 2.62311280e-01 -2.52897114e-01 -1.01913083e+00 -7.12059796e-01 3.45577776e-01 1.01408958e-01 1.27843395e-01 -1.82479963e-01 3.22123468e-01 5.05810857e-01 -1.12238896e+00 3.93244773e-01 7.55888164e-01 8.35209131e-01 1.55330777e-01 8.10170412e-01 6.51335046e-02 7.15613544e-01 -6.79988027e-01 -5.89392126e-01 5.54794312e-01 -3.53153795e-01 -7.39180028e-01 -5.78723326e-02 1.12202418e+00 -6.48809612e-01 -6.07215405e-01 8.15361619e-01 6.24272525e-01 1.94036096e-01 7.02342451e-01 -1.36622441e+00 -1.13631153e+00 4.61486995e-01 -6.48036540e-01 4.92287993e-01 -6.91863475e-03 5.31679809e-01 8.42147946e-01 -9.44991589e-01 5.98815501e-01 1.15061998e+00 7.32038796e-01 9.75663364e-01 -1.36659789e+00 -9.69996691e-01 5.71599126e-01 2.84664452e-01 -1.10691559e+00 -7.50764072e-01 7.09018648e-01 -5.63324213e-01 8.79944384e-01 3.68136913e-01 2.61536151e-01 1.63244557e+00 -6.88745931e-04 7.71520257e-01 1.16286612e+00 -3.00713599e-01 8.13137144e-02 2.65332401e-01 1.37186706e-01 4.34519291e-01 3.44843157e-02 4.30478841e-01 -5.24416268e-01 -1.20431758e-01 7.34950542e-01 -1.09200761e-01 -4.36072111e-01 -2.56352931e-01 -8.56355250e-01 7.56491184e-01 6.28908157e-01 -5.39536886e-02 -7.88082853e-02 3.26975405e-01 2.58001089e-01 2.87840486e-01 5.48997343e-01 5.46091795e-01 -4.44268525e-01 1.03659958e-01 -7.85930395e-01 2.06750721e-01 6.41864359e-01 9.07047629e-01 4.92891103e-01 2.38912210e-01 -3.88192534e-01 7.55778730e-01 5.22652194e-02 4.98796493e-01 3.17142606e-01 -1.07545114e+00 4.75502342e-01 1.02271579e-01 1.00884579e-01 -8.92353594e-01 -1.84707358e-01 -6.24004424e-01 -6.48061454e-01 2.64785439e-01 6.75916731e-01 -2.97930658e-01 -1.26002514e+00 2.07786179e+00 1.33022442e-01 2.89067179e-01 -2.58716404e-01 7.36160398e-01 8.64088058e-01 7.15392590e-01 4.03100550e-01 1.37214765e-01 1.07507145e+00 -1.32252955e+00 -3.59145254e-01 -5.77884853e-01 2.09684715e-01 -7.83744454e-01 1.37444961e+00 2.06448451e-01 -1.01447308e+00 -5.72036862e-01 -9.83802438e-01 -6.55319616e-02 -2.64735103e-01 4.09449674e-02 6.78729296e-01 4.54830557e-01 -1.20425546e+00 4.59539145e-01 -6.63574994e-01 -4.43823338e-01 6.94558740e-01 1.45218089e-01 -4.23081905e-01 -2.40620852e-01 -8.16346467e-01 9.05015945e-01 -2.45878119e-02 -3.35710943e-01 -1.01910937e+00 -1.14552248e+00 -6.04997873e-01 1.22408923e-02 3.01430672e-01 -7.07240403e-01 1.49174201e+00 -1.12556279e+00 -1.40769458e+00 8.62941325e-01 -2.16249093e-01 -4.22732860e-01 6.58933640e-01 -4.29301172e-01 -1.98828027e-01 9.64005515e-02 -8.02946910e-02 1.02451789e+00 1.19067371e+00 -1.47128940e+00 -3.42071533e-01 -7.01644793e-02 2.44916566e-02 2.34173730e-01 -5.18978715e-01 6.98383003e-02 -6.64511919e-01 -9.63765860e-01 -3.35675418e-01 -8.68149459e-01 -2.18336746e-01 2.82310069e-01 -4.16096866e-01 1.78102061e-01 7.13745832e-01 -8.02612185e-01 8.62382889e-01 -2.33742452e+00 8.10080692e-02 -7.24408552e-02 3.73135179e-01 3.85475785e-01 -5.39657235e-01 2.88911492e-01 -3.06182861e-01 2.59915233e-01 -2.56275028e-01 -4.51663285e-01 2.69391537e-02 1.10609442e-01 -7.56180227e-01 4.12563115e-01 2.80206531e-01 1.29537320e+00 -6.26172125e-01 -1.03795767e-01 6.29360601e-02 4.71996009e-01 -7.54082739e-01 2.70853519e-01 -1.20703623e-01 6.85481310e-01 1.51693998e-02 5.67861736e-01 6.73252702e-01 -4.60147172e-01 -5.70987873e-02 -1.79791555e-01 1.97558373e-01 1.80062339e-01 -8.70839477e-01 1.50221729e+00 -1.99354246e-01 8.10380876e-01 9.91590396e-02 -9.39291477e-01 8.23487461e-01 -6.58424273e-02 1.27329648e-01 -7.90994644e-01 -5.00172600e-02 -1.93950653e-01 -1.72727242e-01 -3.50803435e-01 4.23963070e-01 -4.87071611e-02 6.77108765e-02 2.82035500e-01 2.26247326e-01 -1.64294928e-01 -1.34637356e-01 3.42084825e-01 1.19479871e+00 1.22467212e-01 2.20775440e-01 -1.43559128e-01 1.18681625e-01 -7.89067969e-02 5.82745433e-01 9.40166831e-01 -4.43323553e-01 7.85461962e-01 4.02463019e-01 -3.14480782e-01 -1.09832060e+00 -1.27743220e+00 7.17893317e-02 1.42198217e+00 1.18812166e-01 -1.09059498e-01 -6.60658121e-01 -8.25236738e-01 2.15854689e-01 7.14375615e-01 -8.43254328e-01 -3.07278693e-01 -5.82575858e-01 -7.02185929e-01 5.97278535e-01 8.10524642e-01 4.22608405e-01 -7.50912607e-01 -6.37213290e-02 -1.14338771e-01 -7.29972944e-02 -1.04438770e+00 -7.73117661e-01 2.95102507e-01 -6.93470955e-01 -6.35303676e-01 -8.30679238e-01 -7.48724937e-01 7.49690473e-01 3.69804978e-01 1.12713790e+00 1.35733217e-01 -2.54429907e-01 4.88609433e-01 -1.92289814e-01 -2.07685038e-01 -4.36965317e-01 2.26193860e-01 -2.24984437e-02 -1.54590800e-01 -7.94276744e-02 -5.92340469e-01 -6.36542618e-01 3.02275270e-01 -7.27124691e-01 4.59594354e-02 5.93689620e-01 1.01031268e+00 5.76881170e-01 -6.63222015e-01 5.81254303e-01 -8.86908710e-01 5.38534939e-01 -5.23362815e-01 -4.98274744e-01 2.20237076e-01 -5.11250019e-01 4.88264039e-02 7.71072745e-01 -8.55772078e-01 -1.01108301e+00 1.57766432e-01 -1.82630226e-01 -6.27682567e-01 -3.55214059e-01 -8.82849917e-02 -1.20041877e-01 -3.65755409e-01 9.49563503e-01 8.56370330e-02 -1.48546502e-01 -5.71586430e-01 7.16493905e-01 4.08126593e-01 9.72996712e-01 -6.16993010e-01 1.14573538e+00 4.20249403e-01 -3.82655412e-01 -4.06985581e-01 -7.88108408e-01 -2.20057100e-01 -3.66291910e-01 -7.49067217e-02 5.53338587e-01 -9.09820557e-01 -4.22801226e-01 5.12460709e-01 -1.08429897e+00 -8.84151280e-01 -3.24790895e-01 1.98355854e-01 -4.08514559e-01 3.45696300e-01 -6.87346637e-01 -2.90212333e-01 -2.93860137e-01 -1.07124782e+00 7.88208604e-01 3.20279270e-01 -2.72945434e-01 -9.05552745e-01 9.73249674e-02 4.90982175e-01 7.96635091e-01 2.90657878e-02 6.95051730e-01 -8.48146915e-01 -5.57676554e-01 -6.43387586e-02 -4.07048523e-01 4.67413664e-01 8.19784775e-02 9.77908820e-02 -1.25540006e+00 -5.38302541e-01 -1.85369223e-01 -5.50397396e-01 1.00376344e+00 2.22245604e-01 1.29745936e+00 -5.42530537e-01 -3.41536283e-01 1.06647968e+00 1.26399136e+00 1.40855312e-01 7.26014495e-01 3.22161943e-01 7.56243706e-01 2.64996946e-01 2.53351301e-01 2.51623362e-01 2.93599159e-01 6.57988846e-01 4.09252316e-01 -3.18001509e-01 -5.76604843e-01 -4.54899013e-01 3.87765825e-01 5.84380269e-01 1.83266625e-01 -4.26156431e-01 -8.43496501e-01 3.81371617e-01 -1.61185861e+00 -1.11662233e+00 3.19189757e-01 2.14975309e+00 1.02940989e+00 1.24597348e-01 9.84507427e-02 -3.39632988e-01 6.52943611e-01 2.25273356e-01 -9.51276779e-01 -3.94272506e-01 -1.61664084e-01 2.78238446e-01 7.06065416e-01 9.16049480e-01 -1.13860798e+00 1.12530637e+00 6.74489403e+00 8.08067977e-01 -1.37212241e+00 1.37041956e-01 8.51749361e-01 -3.07116985e-01 -4.56015289e-01 -5.45684360e-02 -8.87533844e-01 5.29201925e-01 5.74555337e-01 -2.43755460e-01 7.06761122e-01 8.27444255e-01 -1.50362283e-01 1.64902151e-01 -1.13911426e+00 9.03602183e-01 1.66187301e-01 -1.41850233e+00 7.32435957e-02 -1.18808948e-01 8.55146825e-01 1.71110854e-01 5.84831238e-01 3.63236904e-01 5.77107966e-01 -1.28690827e+00 7.67154992e-01 3.37676346e-01 1.02575755e+00 -4.41612065e-01 1.63892686e-01 1.26719892e-01 -7.99862266e-01 -1.11969367e-01 -1.34155408e-01 -6.11589625e-02 -1.75938103e-02 2.56216764e-01 -7.36747980e-01 1.46032736e-01 8.43820632e-01 4.79414076e-01 -8.29033494e-01 1.06099439e+00 -4.74310964e-01 8.16449463e-01 -2.81279713e-01 2.25855485e-01 6.20460995e-02 3.60262454e-01 4.95573372e-01 1.38171768e+00 1.94255274e-03 7.17945173e-02 1.57585721e-02 6.83101833e-01 -3.41818392e-01 -3.25303487e-02 -5.59299707e-01 6.62299618e-02 7.08157599e-01 1.24539697e+00 -4.24531907e-01 -3.76869529e-01 -3.50189149e-01 1.22974527e+00 6.78614616e-01 7.21619129e-01 -1.19887102e+00 -3.55589122e-01 7.32686162e-01 -8.80899727e-02 4.32433337e-01 -1.89766333e-01 -4.81329620e-01 -1.13580120e+00 1.12636656e-01 -1.07345343e+00 2.41226077e-01 -7.55585849e-01 -1.51120842e+00 6.68457687e-01 -1.71964057e-02 -1.00189745e+00 -1.12215497e-01 -6.41249955e-01 -7.23409176e-01 8.55731785e-01 -1.38431883e+00 -1.08514774e+00 -5.66702068e-01 5.54778695e-01 5.20603836e-01 -4.52466048e-02 6.34186387e-01 3.43869954e-01 -6.71142101e-01 1.26723623e+00 8.40288103e-02 7.67795667e-02 1.23199272e+00 -1.09791100e+00 8.73929143e-01 1.08043003e+00 2.08075449e-01 5.46026647e-01 7.41392672e-01 -4.22430187e-01 -1.23016417e+00 -9.16796088e-01 4.85413522e-01 -7.79068649e-01 6.89566195e-01 -5.82859695e-01 -1.18410909e+00 9.85883772e-01 4.10223454e-01 1.36717245e-01 5.18787920e-01 -1.54971868e-01 -9.42143798e-01 -1.99141562e-01 -1.17249703e+00 9.24400568e-01 1.28242743e+00 -4.27509189e-01 -4.47977185e-01 3.15511018e-01 8.99475396e-01 -7.23172784e-01 -5.37966847e-01 5.02614141e-01 4.44014758e-01 -8.29697013e-01 1.22232544e+00 -9.27402079e-01 4.52828258e-01 1.10975802e-02 -3.26481432e-01 -1.53102577e+00 -6.67155564e-01 -7.82935441e-01 -1.46926016e-01 1.34426606e+00 5.48297763e-01 -8.30521524e-01 8.67293715e-01 8.06815624e-01 -3.41037773e-02 -7.83109426e-01 -7.18412161e-01 -9.00782108e-01 4.57168221e-01 -3.03579003e-01 4.51918423e-01 1.03146362e+00 -1.97653532e-01 3.67368788e-01 -5.50878704e-01 1.22720324e-01 5.51737070e-01 -7.37614557e-02 9.88276005e-01 -6.50083601e-01 -6.86315358e-01 -5.50515234e-01 -5.35771176e-02 -1.41561055e+00 1.26362815e-01 -9.08195674e-01 -4.38495865e-03 -1.25555313e+00 4.25056785e-01 -5.41139841e-01 -3.43135715e-01 8.82664859e-01 -5.17191410e-01 3.85382861e-01 4.25888985e-01 2.01781482e-01 -3.20205331e-01 4.16059226e-01 1.25945771e+00 -2.89060861e-01 -1.30205065e-01 -2.09992513e-01 -1.20078802e+00 5.85038960e-01 1.26288700e+00 -4.40309107e-01 -3.77170682e-01 -7.73848832e-01 -1.98331058e-01 -2.70693094e-01 6.43033028e-01 -7.40712881e-01 2.84482419e-01 -2.83167422e-01 6.01650894e-01 -5.28891496e-02 3.98815721e-01 -4.89467680e-01 -7.84838274e-02 4.12183493e-01 -5.31803787e-01 1.43548146e-01 7.04218268e-01 4.98777121e-01 2.55769879e-01 -3.80084217e-02 9.91387665e-01 1.00977138e-01 -9.71757889e-01 3.23302835e-01 -1.86041474e-01 3.43735069e-01 8.42740119e-01 1.35517120e-01 -9.64454770e-01 -6.19400024e-01 -4.67486948e-01 8.01583827e-02 5.81389427e-01 6.22533023e-01 4.55228806e-01 -1.22463584e+00 -7.34481454e-01 1.06075846e-01 -2.02400219e-02 -3.73136044e-01 2.83725232e-01 5.71591377e-01 -1.93441898e-01 9.57027525e-02 -3.17773551e-01 -5.41664720e-01 -1.19578576e+00 7.09599078e-01 3.35294634e-01 5.38705522e-03 -5.20228148e-01 1.30087149e+00 5.50957024e-01 -3.92819315e-01 5.56273282e-01 -3.28169018e-02 2.22327635e-01 -2.40011722e-01 5.98867714e-01 2.13215798e-01 -1.45688578e-01 -3.29040349e-01 -3.85323495e-01 6.01752639e-01 -4.32664901e-01 -2.63622791e-01 1.11073530e+00 8.83352906e-02 3.24667633e-01 -1.35593424e-02 1.30647933e+00 1.78595379e-01 -1.80152595e+00 -2.68056899e-01 -4.23030466e-01 -5.57597518e-01 -1.96156681e-01 -1.13764739e+00 -1.17277074e+00 8.16037297e-01 5.08465290e-01 -9.07097757e-02 1.20788896e+00 1.67893365e-01 5.37661791e-01 3.52096081e-01 6.40874952e-02 -6.33363485e-01 3.21139485e-01 3.94774020e-01 1.17309654e+00 -1.37254107e+00 -1.57673145e-03 -3.30387861e-01 -7.11059570e-01 6.18096769e-01 1.03142309e+00 -1.79987729e-01 2.33359069e-01 5.93896806e-01 3.51956934e-01 2.53215998e-01 -8.87756348e-01 1.73423678e-01 3.27034682e-01 8.28921437e-01 1.33402959e-01 -8.91664103e-02 1.37507275e-01 5.94409883e-01 -1.98356763e-01 -1.46961719e-01 1.95017293e-01 7.48221159e-01 -4.05979812e-01 -1.04372001e+00 -3.88643563e-01 3.26776475e-01 -3.70147377e-01 -2.64917642e-01 -5.43333888e-01 7.86035180e-01 -1.34018540e-01 7.85539329e-01 -4.56925742e-02 -6.68229282e-01 2.85384119e-01 2.51570251e-02 6.06873214e-01 -5.00388324e-01 -5.69130599e-01 -3.35613012e-01 -1.19055845e-01 -6.91972315e-01 1.62598237e-01 -4.72571343e-01 -8.84521604e-01 -5.07357240e-01 -6.29161671e-02 -2.48796344e-01 4.72294480e-01 6.21228337e-01 6.13167763e-01 3.44611287e-01 6.66190743e-01 -1.13048148e+00 -8.23674560e-01 -7.34973252e-01 8.56369287e-02 4.43478525e-01 4.75998342e-01 -4.89840657e-01 -6.04917526e-01 2.37938359e-01]

[10.221307754516602, 2.1849350929260254]

70de9844-dfad-4f3e-9aa7-f9f80b92a24c

voice2mesh-cross-modal-3d-face-model

2104.10299

https://arxiv.org/abs/2104.10299v1

https://arxiv.org/pdf/2104.10299v1.pdf

Voice2Mesh: Cross-Modal 3D Face Model Generation from Voices

This work focuses on the analysis that whether 3D face models can be learned from only the speech inputs of speakers. Previous works for cross-modal face synthesis study image generation from voices. However, image synthesis includes variations such as hairstyles, backgrounds, and facial textures, that are arguably irrelevant to voice or without direct studies to show correlations. We instead investigate the ability to reconstruct 3D faces to concentrate on only geometry, which is more physiologically grounded. We propose both the supervised learning and unsupervised learning frameworks. Especially we demonstrate how unsupervised learning is possible in the absence of a direct voice-to-3D-face dataset under limited availability of 3D face scans when the model is equipped with knowledge distillation. To evaluate the performance, we also propose several metrics to measure the geometric fitness of two 3D faces based on points, lines, and regions. We find that 3D face shapes can be reconstructed from voices. Experimental results suggest that 3D faces can be reconstructed from voices, and our method can improve the performance over the baseline. The best performance gains (15% - 20%) on ear-to-ear distance ratio metric (ER) coincides with the intuition that one can roughly envision whether a speaker's face is overall wider or thinner only from a person's voice. See our project page for codes and data.

['Ulrich Neumann', 'Chin-Cheng Hsu', 'Ke Xu', 'Cho-Ying Wu']

2021-04-21

['face-model']

['computer-vision']

[ 9.12431106e-02 7.24697769e-01 -2.56771734e-03 -2.76729494e-01 -7.47081220e-01 -4.93474633e-01 5.78697979e-01 -6.98102534e-01 1.34685293e-01 5.01559615e-01 4.52875197e-01 -1.11055195e-01 3.00445203e-02 -6.66191697e-01 -7.96313107e-01 -6.27883255e-01 1.34724617e-01 2.38721967e-01 -2.94898152e-01 -7.60581866e-02 -3.50160301e-02 8.40201437e-01 -1.91689885e+00 1.43572196e-01 3.42887551e-01 9.27426219e-01 -5.62982820e-02 5.47579408e-01 -1.26464382e-01 7.91992992e-02 -4.71304178e-01 -6.08123779e-01 5.45880377e-01 -5.77138960e-01 -4.06970084e-01 5.90440214e-01 8.09758067e-01 -5.40547371e-01 -9.03373212e-02 7.28490889e-01 7.45722473e-01 -2.96510398e-01 9.30177271e-01 -1.21839619e+00 -6.94386005e-01 3.06939989e-01 -4.88661081e-01 -4.59511191e-01 6.35022640e-01 2.38014668e-01 7.81572819e-01 -1.17689872e+00 6.45850599e-01 1.62127829e+00 6.16315305e-01 9.19156611e-01 -1.27068257e+00 -7.03340590e-01 -6.37935475e-02 -4.74888027e-01 -1.39289761e+00 -1.14315128e+00 8.54209661e-01 -4.79712754e-01 5.43680370e-01 1.47838935e-01 5.60663879e-01 1.28856754e+00 -2.63681442e-01 2.96034753e-01 1.23492610e+00 -6.42266512e-01 6.49883226e-02 4.69972372e-01 -6.43354952e-01 9.13462758e-01 1.36181861e-01 4.16319370e-01 -8.05061221e-01 -3.76716591e-02 1.16078746e+00 -5.83873093e-01 -2.51563638e-01 -3.82851571e-01 -9.56879556e-01 6.93461597e-01 6.05073273e-02 1.39912009e-01 -1.76995412e-01 -1.24382779e-01 -1.56503528e-01 2.66213149e-01 3.70540947e-01 4.88832921e-01 -2.79076815e-01 1.59274399e-01 -1.02784693e+00 1.95585892e-01 8.54098380e-01 9.64552402e-01 5.85633337e-01 3.64821464e-01 7.33213946e-02 9.41976190e-01 5.40125847e-01 8.23595345e-01 1.84644818e-01 -1.35855281e+00 -7.09770471e-02 2.66157031e-01 -3.77959237e-02 -9.56974149e-01 -1.04600742e-01 -2.40637168e-01 -4.78529483e-01 2.92366028e-01 5.34156322e-01 -3.15305769e-01 -8.04773867e-01 1.74487650e+00 4.01186079e-01 2.23787814e-01 5.44976071e-02 8.22351336e-01 1.09400809e+00 2.21131682e-01 -3.99544150e-01 -5.26393235e-01 1.15702367e+00 -4.59326863e-01 -5.19482911e-01 -9.48550645e-03 2.65885890e-01 -8.48842859e-01 1.10306907e+00 3.10352057e-01 -1.32186484e+00 -5.21491051e-01 -7.90594339e-01 2.09924504e-01 -5.08071519e-02 1.57940894e-01 5.05255461e-01 1.17639911e+00 -1.28206050e+00 4.34079856e-01 -4.51528281e-01 -5.49178481e-01 4.73466188e-01 2.71367311e-01 -6.96271896e-01 1.96602285e-01 -7.97144651e-01 8.42906892e-01 -4.01751280e-01 -8.20077583e-02 -9.04041946e-01 -7.96087503e-01 -8.01408470e-01 -2.60901719e-01 2.27915227e-01 -7.72736609e-01 1.08941936e+00 -1.05278599e+00 -1.70973694e+00 1.07808232e+00 -3.50538760e-01 2.38792179e-03 4.16544199e-01 1.39392883e-01 -2.24771291e-01 3.48222077e-01 -1.67408705e-01 1.05789530e+00 1.08726525e+00 -1.81095934e+00 1.00572268e-02 -4.58218902e-01 -9.53301787e-02 7.79465809e-02 -2.65209824e-01 -5.95593154e-02 -2.71229595e-01 -4.64851707e-01 8.26134235e-02 -8.97353768e-01 1.17918663e-01 4.79274809e-01 -3.44367981e-01 1.37178138e-01 6.95048749e-01 -5.22182226e-01 5.37239492e-01 -2.08122802e+00 -2.17703462e-01 2.99518257e-01 -5.23443567e-03 1.94680616e-02 -3.34748209e-01 2.66417086e-01 -2.11035922e-01 4.63899016e-01 -1.14507206e-01 -4.22647148e-01 1.68603864e-02 2.68970877e-02 -3.36918056e-01 4.55644935e-01 3.78718615e-01 7.71241128e-01 -4.03957427e-01 -6.00496769e-01 -8.13152269e-02 6.40216529e-01 -7.85284340e-01 7.78523088e-02 6.45415336e-02 6.94910407e-01 -1.67022243e-01 8.66459846e-01 6.96688592e-01 1.16359249e-01 2.35032469e-01 -2.11549580e-01 8.09810981e-02 1.81950256e-02 -1.01545882e+00 1.54445398e+00 -5.43897748e-01 4.95656639e-01 3.17155421e-01 -7.82329440e-01 1.03674829e+00 5.71838915e-01 3.43217403e-01 -4.83637184e-01 1.48756444e-01 2.50938296e-01 1.09446198e-01 -4.51608539e-01 -1.06178828e-01 -5.37351370e-01 2.91897148e-01 4.82817829e-01 2.97600627e-01 -5.19158959e-01 -3.18924665e-01 -1.19367294e-01 6.42136276e-01 9.04929414e-02 -8.66573527e-02 -3.37710828e-01 2.42373690e-01 -7.31039047e-01 3.21024448e-01 3.59827310e-01 4.82252687e-02 7.91151524e-01 7.08131135e-01 1.33808255e-01 -9.93979275e-01 -1.41900539e+00 -3.32051843e-01 8.36908817e-01 -1.20255291e-01 -1.46835119e-01 -1.07185566e+00 -4.72717792e-01 3.79486419e-02 6.38126910e-01 -5.13774216e-01 -4.37954441e-03 -1.81760281e-01 -3.18325728e-01 8.35766912e-01 3.15249264e-01 1.17374644e-01 -8.57317686e-01 -3.16612452e-01 -4.72151369e-01 1.91870444e-02 -1.15555096e+00 -3.64857018e-01 -3.77761155e-01 -8.30594122e-01 -8.26282024e-01 -7.99162924e-01 -7.84615457e-01 8.94423068e-01 -6.32531345e-02 9.35295403e-01 9.44893062e-02 -2.93887854e-01 8.02226365e-01 -1.06534608e-01 -6.06651545e-01 -4.30915952e-01 -3.50901872e-01 5.68535924e-01 2.15860471e-01 -3.23742232e-03 -9.85133588e-01 -5.55327296e-01 6.06396258e-01 -6.01875961e-01 3.18165682e-02 4.79221106e-01 5.10664761e-01 4.19765323e-01 -1.73699036e-01 9.54051197e-01 -5.51450968e-01 4.57562268e-01 -1.56103805e-01 -2.42784917e-01 -2.20042244e-02 -3.48774731e-01 3.49117927e-02 6.30709901e-02 -4.70997095e-01 -1.10646927e+00 1.35127932e-01 -2.27818042e-01 -6.94674313e-01 -5.23210406e-01 -1.17794216e-01 -6.26348197e-01 -1.40806705e-01 7.22327411e-01 6.20543547e-02 4.87099916e-01 -3.35772693e-01 5.37100375e-01 7.18079269e-01 4.09264803e-01 -8.68879139e-01 9.91458535e-01 6.10205173e-01 6.29023835e-02 -1.32583606e+00 -3.76084775e-01 6.73568845e-02 -7.09510446e-01 -5.31982541e-01 6.94153190e-01 -8.10110450e-01 -7.94498324e-01 2.36910731e-01 -1.01251221e+00 -2.68412113e-01 -2.01550931e-01 6.05955303e-01 -7.30772793e-01 1.76999986e-01 -4.13370043e-01 -1.13160217e+00 -4.12811339e-02 -1.04755890e+00 1.30166876e+00 1.04931772e-01 -3.03916752e-01 -8.52955878e-01 -2.24153087e-01 4.91633922e-01 3.06254089e-01 2.91731566e-01 9.61034119e-01 -3.78595620e-01 -4.07144547e-01 1.44119620e-01 7.33099207e-02 3.90691727e-01 4.18691993e-01 2.51575708e-01 -1.48566985e+00 -1.55469954e-01 9.12066177e-03 -4.25684214e-01 4.95085984e-01 4.13861603e-01 9.93185580e-01 -4.18438435e-01 -1.94943547e-01 4.78325576e-01 9.06377137e-01 9.60463099e-03 7.16060162e-01 -5.61094463e-01 4.24859375e-01 1.35632193e+00 1.14289403e-01 3.01789761e-01 9.16393623e-02 3.98322374e-01 3.32603127e-01 -8.07417706e-02 -5.00891149e-01 -5.83764076e-01 4.78091508e-01 7.52425373e-01 -2.84013301e-01 -5.92009798e-02 -5.88313282e-01 4.25022155e-01 -9.05060768e-01 -8.28746855e-01 3.17534685e-01 2.32863283e+00 7.38065779e-01 4.37117741e-02 2.28167757e-01 1.44716844e-01 7.22168207e-01 -1.95656806e-01 -4.50229883e-01 -3.20231199e-01 -7.30800480e-02 6.09199762e-01 1.58937782e-01 5.37311554e-01 -6.05105758e-01 7.89934695e-01 7.26886368e+00 5.34021974e-01 -1.33329797e+00 -2.12860197e-01 8.26581895e-01 -4.14756447e-01 -5.89782536e-01 -2.27979586e-01 -7.46587873e-01 1.28814071e-01 8.48078072e-01 3.69790234e-02 4.31332678e-01 6.06980205e-01 4.02367830e-01 1.80612430e-01 -1.40219927e+00 1.06800938e+00 2.27498770e-01 -1.03492212e+00 2.91162550e-01 3.83835226e-01 5.67864656e-01 -6.81230843e-01 5.46598911e-01 1.49133146e-01 -1.68050393e-01 -1.49290943e+00 7.44416714e-01 7.89807260e-01 1.32094526e+00 -6.69510365e-01 1.85935512e-01 3.44883740e-01 -9.24243569e-01 2.07640111e-01 -1.09048069e-01 2.19597250e-01 7.49520538e-03 4.05723751e-01 -9.94203031e-01 3.24359834e-01 5.35087466e-01 2.17844576e-01 -3.95181894e-01 5.82946301e-01 -3.31018381e-02 5.86712122e-01 -3.13130707e-01 3.15878689e-01 -3.08734655e-01 -2.83450633e-02 6.46010876e-01 8.50288391e-01 7.34867930e-01 1.01705100e-02 -1.86569333e-01 1.17358053e+00 -2.40085989e-01 2.31788039e-01 -1.00415301e+00 -1.69639826e-01 4.72797900e-01 1.22842968e+00 -6.04608655e-01 1.67625412e-01 -4.96039629e-01 6.55272007e-01 -1.18528910e-01 2.95266420e-01 -5.23967087e-01 -3.31848003e-02 6.57664537e-01 6.99292719e-01 2.38121092e-01 -1.13684051e-01 -2.06388786e-01 -7.59717166e-01 -1.48191694e-02 -1.00079858e+00 -3.43608916e-01 -1.04378200e+00 -1.38746226e+00 4.37116623e-01 3.46669629e-02 -8.41548085e-01 -3.12456012e-01 -7.88026392e-01 -6.00630164e-01 7.91303992e-01 -1.14350653e+00 -1.20421040e+00 -9.37354714e-02 5.83754182e-01 3.82945091e-01 -3.17927122e-01 9.33022201e-01 1.76416561e-01 -2.10691050e-01 1.02231812e+00 -5.18990874e-01 2.18509704e-01 8.75114202e-01 -9.48250949e-01 2.05457091e-01 3.17763299e-01 3.26606929e-01 6.99476779e-01 5.03163218e-01 -4.75878090e-01 -1.61952579e+00 -8.21766794e-01 6.33297443e-01 -6.32689357e-01 2.35527456e-01 -4.72302258e-01 -6.78262234e-01 3.70429456e-01 1.74082682e-01 4.85604107e-02 8.05305839e-01 1.01869375e-01 -4.33404326e-01 -1.89048693e-01 -1.31380570e+00 6.27315998e-01 1.39136016e+00 -6.18524849e-01 -4.37542796e-01 3.56634296e-02 4.68407869e-01 -6.77921921e-02 -9.41511035e-01 4.40201998e-01 8.78193915e-01 -1.06985021e+00 9.39484954e-01 -4.85616922e-01 3.90572667e-01 -1.10006846e-01 -3.24489653e-01 -1.16102409e+00 1.75469697e-01 -7.26720691e-01 1.97220713e-01 1.44139588e+00 5.91422558e-01 -5.52608490e-01 6.95986569e-01 4.33741271e-01 8.44842568e-02 -6.83858752e-01 -8.62448275e-01 -6.99107289e-01 3.61314267e-01 -4.72902179e-01 6.28717244e-01 9.15186644e-01 -2.82508194e-01 3.14854980e-01 -1.12011544e-01 1.86099112e-01 5.48226833e-01 -3.81568335e-02 8.90780687e-01 -1.31989384e+00 -1.07880399e-01 -4.64312226e-01 -3.10412616e-01 -8.50562096e-01 4.61191416e-01 -1.04584205e+00 -4.17369194e-02 -1.22460866e+00 -4.52249348e-02 -2.67512321e-01 2.37100035e-01 3.20027143e-01 4.60293144e-01 2.63990909e-01 1.03212893e-01 -1.01977907e-01 1.77495733e-01 4.88912970e-01 1.64692330e+00 1.74435884e-01 -1.59022465e-01 -1.50198743e-01 -9.46872234e-01 1.05171311e+00 6.85307801e-01 -8.92458558e-02 -6.83281183e-01 -1.29169881e-01 -1.24200992e-01 3.21490318e-01 5.39360702e-01 -7.83407092e-01 -8.72233063e-02 -4.85606045e-02 6.13130450e-01 -1.77201144e-02 5.64462185e-01 -6.35963619e-01 2.11916730e-01 -1.41197406e-02 -2.75843680e-01 -3.91774684e-01 2.22523779e-01 3.78082991e-01 1.52189702e-01 -6.07475527e-02 8.56399894e-01 -2.89772004e-01 2.86411624e-02 3.03928733e-01 -2.93774843e-01 -1.63007841e-01 7.31783807e-01 -4.45539415e-01 7.40708318e-03 -7.19154358e-01 -9.06726182e-01 -2.61524767e-01 5.86662054e-01 3.48915815e-01 7.02707767e-01 -1.47561669e+00 -9.20530915e-01 5.76058567e-01 -9.09552127e-02 -2.76690483e-01 -3.49926054e-02 7.57452309e-01 -3.08268726e-01 3.19326133e-01 -1.59181595e-01 -6.21044695e-01 -1.26331913e+00 3.90796781e-01 5.83193243e-01 4.76998836e-01 -2.77534753e-01 8.38113666e-01 5.97736239e-01 -6.00192010e-01 2.63434649e-01 -8.36998448e-02 1.36642214e-02 1.85619697e-01 2.30671272e-01 2.11277574e-01 -3.45860459e-02 -8.76416445e-01 -2.50062495e-01 8.40505481e-01 3.16419214e-01 -3.98901224e-01 9.79406476e-01 1.01134405e-02 1.17679007e-01 4.34323251e-01 1.10990906e+00 6.24959767e-01 -1.16597664e+00 1.24759354e-01 -4.84095395e-01 -4.86760765e-01 -1.78103715e-01 -5.30575931e-01 -1.30885696e+00 9.47559893e-01 5.58138371e-01 -8.37587193e-02 1.03208005e+00 2.70135701e-01 4.23878759e-01 1.51107147e-01 3.38410795e-01 -7.38848746e-01 4.13381577e-01 5.20284474e-02 1.24877095e+00 -9.74024355e-01 -1.97609141e-01 -6.42795801e-01 -5.92733383e-01 9.06398118e-01 6.14499032e-01 1.50542436e-02 8.87073874e-01 3.47789407e-01 5.22887111e-02 -6.95210099e-02 -7.87644207e-01 -3.18460852e-01 4.89853531e-01 8.49176884e-01 5.58545470e-01 -1.05546162e-01 1.36330321e-01 7.80210674e-01 -8.11044455e-01 -3.39477062e-01 2.67467380e-01 4.36295092e-01 -4.04067695e-01 -1.02992880e+00 -4.81476158e-01 4.55369562e-01 -3.92240077e-01 1.17973059e-01 -7.28144467e-01 7.21303463e-01 3.46466690e-01 1.04270756e+00 2.21371531e-01 -3.34472835e-01 3.55276495e-01 4.47871745e-01 9.73224640e-01 -7.23346889e-01 4.69548218e-02 3.27558696e-01 1.11971170e-01 -3.39089394e-01 -5.59214711e-01 -5.53146362e-01 -1.05772603e+00 -2.82356501e-01 -4.22090352e-01 -1.88220263e-01 5.97944796e-01 6.79385483e-01 3.44488114e-01 8.57025161e-02 7.90051281e-01 -8.77701700e-01 -7.08141550e-02 -8.53412271e-01 -7.72971869e-01 1.27150178e-01 2.80045033e-01 -9.08905625e-01 -6.74734533e-01 3.58380914e-01]

[12.92284870147705, -0.313340425491333]

0c464d0c-540a-43e7-a482-063398c8449c

multi-2oie-multilingual-open-information-1

https://aclanthology.org/2020.findings-emnlp.99

https://aclanthology.org/2020.findings-emnlp.99.pdf

Multi\^2OIE: Multilingual Open Information Extraction Based on Multi-Head Attention with BERT

In this paper, we propose Multi$^2$OIE, which performs open information extraction (open IE) by combining BERT with multi-head attention. Our model is a sequence-labeling system with an efficient and effective argument extraction method. We use a query, key, and value setting inspired by the Multimodal Transformer to replace the previously used bidirectional long short-term memory architecture with multi-head attention. Multi$^2$OIE outperforms existing sequence-labeling systems with high computational efficiency on two benchmark evaluation datasets, Re-OIE2016 and CaRB. Additionally, we apply the proposed method to multilingual open IE using multilingual BERT. Experimental results on new benchmark datasets introduced for two languages (Spanish and Portuguese) demonstrate that our model outperforms other multilingual systems without training data for the target languages.

['Pilsung Kang', 'Yukyung Lee', 'Youngbin Ro']

2020-11-01

findings-of-the-association-for-computational

['open-information-extraction']

['natural-language-processing']

[ 1.35268671e-02 2.12883130e-01 -6.06281698e-01 2.77121142e-02 -1.40955245e+00 -9.91197646e-01 7.07503319e-01 3.41855437e-01 -1.04265320e+00 1.12875974e+00 3.06728274e-01 -8.55263710e-01 6.76371902e-02 -7.75799274e-01 -1.14846003e+00 5.19656278e-02 2.21193373e-01 7.59777546e-01 1.35552004e-01 -5.08128941e-01 8.45051110e-02 -2.35986009e-01 -1.06877363e+00 6.36085391e-01 1.09686053e+00 9.66878533e-01 2.31129646e-01 6.16333425e-01 -5.02013862e-01 9.34386492e-01 -5.00374913e-01 -1.09517336e+00 -5.91537468e-02 7.91703314e-02 -1.50787854e+00 -7.54903436e-01 2.99418420e-01 -1.42284393e-01 1.60686690e-02 9.29617405e-01 6.02191985e-01 -1.71806976e-01 4.36211705e-01 -9.63652849e-01 -7.19835937e-01 1.28206658e+00 -3.81734222e-01 2.86810338e-01 6.86096728e-01 -8.68836865e-02 1.53414190e+00 -9.58332419e-01 9.96386528e-01 1.27976632e+00 6.45026088e-01 1.78331584e-01 -1.10206044e+00 -5.95055759e-01 1.90838903e-01 6.21636748e-01 -1.12468421e+00 -1.72549501e-01 5.40222883e-01 -1.57869279e-01 1.60256195e+00 2.31592998e-01 3.85723025e-01 1.27532303e+00 -1.34321064e-01 1.23625076e+00 9.71482694e-01 -6.52057171e-01 -3.51377636e-01 1.87201935e-04 4.92044061e-01 8.26330364e-01 -1.87604260e-02 -7.49434978e-02 -1.90142781e-01 -2.01945886e-01 1.07590184e-01 -4.23770875e-01 -2.32865721e-01 1.36272665e-02 -1.55391073e+00 9.63644505e-01 2.73029625e-01 2.36408144e-01 -3.59010041e-01 2.97549516e-02 1.08915520e+00 5.23480415e-01 2.28625536e-01 4.18593884e-01 -1.04314959e+00 -2.36885667e-01 -4.57781255e-01 6.80654645e-02 1.11874962e+00 1.14314592e+00 5.63018620e-01 -4.14425671e-01 -5.20669997e-01 1.04762757e+00 2.02494144e-01 6.93311155e-01 5.17594099e-01 -7.99902916e-01 1.18132174e+00 5.69131374e-01 1.04665689e-01 -4.04997587e-01 -2.97726721e-01 -4.90816534e-01 -4.91924018e-01 -6.03966177e-01 2.40491509e-01 -3.69253725e-01 -6.47088468e-01 1.79578197e+00 1.99174732e-01 -3.17664355e-01 6.41864300e-01 4.01769042e-01 1.22349358e+00 7.82363057e-01 2.97794670e-01 -1.22067146e-01 1.85556841e+00 -1.39004552e+00 -1.02148700e+00 -3.05251211e-01 9.98307109e-01 -7.35608518e-01 1.20376718e+00 2.46236980e-01 -1.08557940e+00 -3.10750455e-01 -1.04972351e+00 -5.72134197e-01 -7.61709809e-01 2.56050855e-01 6.06071293e-01 2.93947905e-01 -5.55984378e-01 1.41119525e-01 -3.21295142e-01 -9.71167386e-02 1.72058061e-01 1.35579318e-01 -4.13402528e-01 7.65274689e-02 -1.92699349e+00 8.01156521e-01 1.00128841e+00 5.56325838e-02 -7.25637376e-01 -4.36333269e-01 -1.20931280e+00 1.59085840e-01 8.35784793e-01 -7.01373041e-01 1.41366100e+00 -3.75184327e-01 -1.29112017e+00 8.82139981e-01 -1.48913935e-01 -8.09324801e-01 2.37266585e-01 -6.98029101e-01 -3.48820299e-01 1.61287338e-01 5.15593588e-01 6.73900545e-01 3.42043996e-01 -9.05903757e-01 -5.11172950e-01 -1.68432981e-01 2.71030217e-01 5.76588362e-02 -3.33521336e-01 3.04471195e-01 -6.33186102e-01 -6.83263540e-01 -2.78712273e-01 -7.44904757e-01 1.13066427e-01 -8.05603325e-01 -6.41830146e-01 -8.14267516e-01 5.65996826e-01 -1.12716639e+00 1.69730592e+00 -1.76137829e+00 2.12553769e-01 -5.63121261e-03 -4.90610339e-02 6.31749928e-01 -1.53153211e-01 6.55455589e-01 2.72215735e-02 3.74169677e-01 -4.31420952e-01 -1.80343091e-01 2.82711178e-01 2.37936631e-01 -3.83145005e-01 -1.35653883e-01 1.06050767e-01 1.42813253e+00 -1.05841565e+00 -8.78402054e-01 -8.87236372e-02 3.06510538e-01 -4.85419244e-01 1.35481611e-01 -7.28711128e-01 1.10739090e-01 -3.53452414e-01 6.47053599e-01 3.26780289e-01 -3.75568271e-01 5.45225859e-01 -2.56471276e-01 -1.00952543e-01 8.23112607e-01 -7.03976274e-01 1.99854290e+00 -7.58812904e-01 3.03624153e-01 -6.97599351e-02 -7.54192233e-01 5.82723141e-01 6.88192546e-01 -1.40211396e-02 -1.05164790e+00 2.45321661e-01 7.30468631e-01 -1.91267401e-01 -6.06559157e-01 4.29854751e-01 2.95297027e-01 -5.73267281e-01 3.48370373e-01 2.40970999e-01 8.04295316e-02 8.06446970e-01 5.00546336e-01 8.32115591e-01 4.54970211e-01 4.20634508e-01 -1.69980943e-01 1.00225282e+00 -2.32960284e-01 7.11879611e-01 8.08481872e-01 5.89658804e-02 -3.77572738e-02 6.80023849e-01 -4.43231940e-01 -9.97439861e-01 -6.47703707e-01 -2.74391919e-01 1.23528898e+00 -8.40850100e-02 -7.54693270e-01 -7.68777370e-01 -1.26277137e+00 -1.83857962e-01 6.73004687e-01 -3.73259634e-01 3.11486244e-01 -9.21154797e-01 -6.38790250e-01 1.04710329e+00 6.15462899e-01 6.83147073e-01 -1.45206189e+00 -5.14158010e-01 6.70926392e-01 -1.20996296e+00 -1.44390166e+00 -4.14550871e-01 3.30838740e-01 -2.78683752e-01 -1.04609227e+00 -6.70169055e-01 -1.01855218e+00 -1.01101115e-01 -5.59209764e-01 1.53057694e+00 -1.40373990e-01 7.06018955e-02 8.06184411e-02 -4.54981953e-01 -3.64068300e-01 -1.89391360e-01 7.18982220e-01 -4.70694840e-01 -3.55504483e-01 3.79553437e-01 -1.37191668e-01 -3.02431464e-01 -2.61112005e-02 -7.67298102e-01 4.84777652e-02 5.94109118e-01 1.29750216e+00 5.26405215e-01 -6.49459124e-01 8.49839807e-01 -1.07690132e+00 7.11080492e-01 -6.25749409e-01 -6.45725846e-01 6.70022607e-01 -6.42618835e-01 5.75771153e-01 6.65723860e-01 -1.96799442e-01 -1.21682572e+00 -5.51387429e-01 -6.25029206e-01 7.41035342e-02 2.44277909e-01 8.93390894e-01 -4.07739997e-01 4.14099574e-01 2.58502007e-01 2.12333072e-02 -5.63835144e-01 -8.21117520e-01 6.72679722e-01 8.20502818e-01 6.97230577e-01 -8.66975904e-01 2.13328212e-01 2.07889944e-01 -3.46118867e-01 -3.94549668e-01 -9.56942737e-01 -2.43623033e-01 -4.88059908e-01 4.43593472e-01 8.70997667e-01 -9.43436861e-01 -1.00120950e+00 3.52496713e-01 -1.53519905e+00 -1.53653726e-01 8.59668013e-04 4.35935378e-01 -4.90079761e-01 5.16444147e-01 -1.11185622e+00 -4.88713384e-01 -9.53588784e-01 -1.30874169e+00 1.19413233e+00 -1.97746485e-01 -1.43239841e-01 -1.09498394e+00 1.43107876e-01 6.23541713e-01 5.52005321e-02 -4.51310985e-02 1.28810704e+00 -8.41820061e-01 -3.72303367e-01 1.33431301e-01 -5.11695683e-01 2.07739770e-01 -5.64275026e-01 -3.84101301e-01 -5.84195018e-01 -1.66337639e-01 -5.62088728e-01 -9.56382811e-01 1.13873065e+00 -8.92775692e-03 8.57642055e-01 -6.23044848e-01 -4.12143320e-01 7.77913690e-01 1.43123114e+00 2.24151224e-01 4.56954092e-01 9.19233859e-01 7.59268939e-01 5.18137872e-01 7.12639749e-01 1.19903311e-01 8.90440941e-01 4.92616862e-01 2.88905948e-01 8.03167559e-03 -1.43567041e-01 -6.00674987e-01 3.71762246e-01 1.07905221e+00 2.24832669e-01 -3.34228009e-01 -9.65237498e-01 7.94188321e-01 -1.92216277e+00 -7.33255744e-01 -6.91182166e-03 1.91773307e+00 1.24361002e+00 1.78260058e-01 -7.98763409e-02 1.09662406e-01 4.15384829e-01 1.14576839e-01 -2.65850216e-01 -6.99685454e-01 -5.42801678e-01 3.23883235e-01 5.03913403e-01 5.47239482e-01 -1.23332107e+00 1.19184852e+00 5.44905758e+00 1.13376534e+00 -7.33027160e-01 3.51383239e-01 3.74552906e-01 2.06729129e-01 -6.95012569e-01 2.34414056e-01 -1.27559912e+00 4.50254530e-01 1.05867600e+00 3.42937671e-02 1.75416917e-01 4.51591164e-01 -6.60108447e-01 2.34252051e-01 -9.87902761e-01 7.47384548e-01 4.29908782e-02 -1.28593969e+00 -4.59271409e-02 -1.58993080e-01 4.09434140e-01 3.83772314e-01 -1.79622620e-01 8.05239618e-01 7.06909001e-01 -7.44242191e-01 8.87623072e-01 1.38806522e-01 9.23259199e-01 -6.47864759e-01 8.85850966e-01 4.17607069e-01 -1.36754978e+00 -3.17059368e-01 7.56945163e-02 3.30038488e-01 3.45799893e-01 1.21725895e-01 -6.29908860e-01 9.69909310e-01 7.22712576e-01 5.51174462e-01 -5.73283851e-01 5.50652683e-01 -7.14318931e-01 5.31520426e-01 -3.47409129e-01 -2.10013971e-01 5.49411118e-01 -3.71811092e-02 6.94197655e-01 1.50090408e+00 8.26972164e-03 -3.51234287e-01 3.29040259e-01 6.03927195e-01 -5.55592597e-01 4.60476547e-01 -6.29219592e-01 -2.32691407e-01 4.74408805e-01 1.05705202e+00 -2.04720646e-01 -7.83187568e-01 -7.29580343e-01 8.71728063e-01 8.34238887e-01 3.17229003e-01 -8.75633359e-01 -7.53416479e-01 -1.50312111e-01 -5.06592274e-01 5.66718876e-01 7.17361569e-02 2.33619139e-01 -1.43021500e+00 1.71899259e-01 -1.25659478e+00 1.04638481e+00 -7.17394888e-01 -1.03145719e+00 9.89102066e-01 1.28693044e-01 -6.72515213e-01 -6.28260672e-01 -4.03995454e-01 1.98678166e-01 8.18248034e-01 -2.02468753e+00 -1.57849789e+00 6.15192413e-01 5.59532285e-01 6.01728857e-01 -3.38219292e-03 8.59384239e-01 6.48252249e-01 -7.04033196e-01 7.35933602e-01 1.34913614e-02 7.27870822e-01 5.10894656e-01 -1.29516065e+00 6.24188304e-01 6.72082126e-01 3.19801271e-01 7.77726591e-01 2.21491143e-01 -7.57499456e-01 -1.28027737e+00 -8.28801036e-01 1.47303116e+00 -3.95216405e-01 7.86609888e-01 -4.96301085e-01 -8.57055306e-01 1.06105280e+00 8.30474317e-01 -8.96561444e-02 5.45082033e-01 3.42372417e-01 -6.73702419e-01 1.90706581e-01 -7.84852266e-01 4.16954309e-01 1.05640793e+00 -7.52406597e-01 -1.05118752e+00 2.69867331e-01 1.29401100e+00 -2.44604751e-01 -1.16215312e+00 8.84779692e-01 6.38369262e-01 -3.93892646e-01 1.17653322e+00 -7.91613400e-01 3.25347334e-01 1.04814149e-01 -2.23795384e-01 -1.12869704e+00 -1.61059704e-02 -7.38055229e-01 -3.94060493e-01 1.30380929e+00 9.26430464e-01 -7.70432711e-01 1.18561685e-02 -4.70901653e-02 -4.01049480e-02 -9.50472355e-01 -9.17375684e-01 -7.06489861e-01 2.63575852e-01 -4.84895051e-01 7.38340616e-01 7.19773054e-01 2.48840377e-01 1.07465935e+00 -2.67628342e-01 -1.88516185e-01 6.28836453e-01 4.20905977e-01 3.92912626e-01 -9.96872425e-01 -2.89064020e-01 -2.12485567e-01 3.37119967e-01 -1.10500169e+00 7.20646918e-01 -1.18310893e+00 -4.01036590e-01 -1.52779806e+00 3.26603383e-01 -3.12626719e-01 -2.69739151e-01 6.61455095e-01 -4.08841163e-01 4.74427491e-02 1.81660708e-02 2.42210124e-02 -9.03431952e-01 7.43325531e-01 8.86578560e-01 -3.45863819e-01 1.50774285e-01 -4.44054991e-01 -4.95467991e-01 7.65341401e-01 5.39463460e-01 -3.66943985e-01 -2.76600756e-02 -7.64005423e-01 8.63413393e-01 2.78018296e-01 6.25485629e-02 -4.95588958e-01 4.27329391e-02 3.74272168e-01 -1.09187111e-01 -9.62141752e-01 1.32113993e-01 -5.74399889e-01 -5.15017450e-01 4.72872078e-01 -4.95861679e-01 4.31358427e-01 3.07187319e-01 3.54128242e-01 -3.58060837e-01 -4.29974228e-01 2.76423723e-01 -3.12902749e-01 -8.81806910e-01 2.02969730e-01 -1.30336508e-01 7.86425591e-01 6.67984188e-01 5.35568774e-01 -6.13575220e-01 -8.27420577e-02 -6.28222287e-01 7.46378481e-01 -8.91975164e-02 5.42467415e-01 4.71284807e-01 -1.21408045e+00 -8.42248321e-01 2.71768570e-01 4.26028550e-01 -2.26031914e-01 -1.03897892e-01 7.50405490e-01 -3.91335636e-01 1.14642644e+00 1.48641959e-01 -2.69240260e-01 -1.12741792e+00 8.89233887e-01 1.02810496e-02 -1.14969850e+00 -4.82494563e-01 8.47944617e-01 -1.77001089e-01 -1.11141384e+00 2.50959128e-01 -1.95555836e-01 -5.64470768e-01 1.37439042e-01 2.43061766e-01 2.69408524e-01 1.51040092e-01 -6.65610194e-01 -2.89828658e-01 3.33904654e-01 -3.12284052e-01 -4.37254578e-01 1.00328147e+00 -3.02180380e-01 -3.61969292e-01 3.62541229e-01 1.36960649e+00 6.91216737e-02 -5.12149215e-01 -6.79101765e-01 4.64679688e-01 4.02908847e-02 -7.48726726e-02 -1.06890821e+00 -6.14880800e-01 9.96150434e-01 2.03655913e-01 -9.61358175e-02 7.60547996e-01 1.44792482e-01 1.47264671e+00 6.55037820e-01 3.33262801e-01 -1.14466083e+00 -3.65497947e-01 9.52448964e-01 7.71858454e-01 -1.36290598e+00 -4.48291004e-01 -2.76140660e-01 -5.12589455e-01 6.73363209e-01 5.19336104e-01 4.92601991e-01 4.14741844e-01 3.20496172e-01 -3.60706039e-02 -3.01869839e-01 -1.02924788e+00 -4.21145022e-01 3.08949471e-01 1.11427009e-01 6.52077556e-01 1.79582778e-02 -8.72852147e-01 7.81841874e-01 -1.82994738e-01 2.68086344e-02 -7.52523309e-03 9.94782507e-01 7.22036362e-02 -1.52530503e+00 -6.01484701e-02 2.31610611e-01 -1.08512855e+00 -7.33866632e-01 -1.18512101e-01 8.33424866e-01 7.96561465e-02 7.79624820e-01 -1.63389176e-01 -1.12480102e-02 2.02334777e-01 5.72936714e-01 4.29926932e-01 -4.22683448e-01 -1.02821743e+00 1.41137779e-01 7.16768682e-01 -4.85716909e-01 -2.26343840e-01 -5.01444876e-01 -1.37045896e+00 5.40131666e-02 -3.71394128e-01 5.32016814e-01 6.94809318e-01 8.99687052e-01 4.94640738e-01 5.76806903e-01 2.46355042e-01 -3.79605554e-02 -5.91795981e-01 -1.21870756e+00 8.30257405e-03 5.57150841e-01 3.03105474e-01 -6.15383148e-01 -5.31841209e-03 -2.11015001e-01]

[9.881592750549316, 8.936219215393066]

f0ee7bb7-5162-442c-84b6-d7dd46b7f5d7

towards-flow-graph-prediction-of-open-domain

2305.19497

https://arxiv.org/abs/2305.19497v1

https://arxiv.org/pdf/2305.19497v1.pdf

Towards Flow Graph Prediction of Open-Domain Procedural Texts

Machine comprehension of procedural texts is essential for reasoning about the steps and automating the procedures. However, this requires identifying entities within a text and resolving the relationships between the entities. Previous work focused on the cooking domain and proposed a framework to convert a recipe text into a flow graph (FG) representation. In this work, we propose a framework based on the recipe FG for flow graph prediction of open-domain procedural texts. To investigate flow graph prediction performance in non-cooking domains, we introduce the wikiHow-FG corpus from articles on wikiHow, a website of how-to instruction articles. In experiments, we consider using the existing recipe corpus and performing domain adaptation from the cooking to the target domain. Experimental results show that the domain adaptation models achieve higher performance than those trained only on the cooking or target domain data.

['Shinsuke Mori', 'Hirotaka Kameko', 'Keisuke Shirai']

2023-05-31

['reading-comprehension']

['natural-language-processing']

[ 1.45855457e-01 4.63831484e-01 -1.87451035e-01 -4.13059443e-01 -2.47429401e-01 -7.20092177e-01 3.79167795e-01 5.57088792e-01 6.50460552e-03 4.57738787e-01 6.61544681e-01 -6.33271754e-01 -1.29101619e-01 -1.31760550e+00 -7.83560514e-01 1.26293391e-01 1.25326216e-01 3.69997293e-01 4.33925152e-01 -3.98284018e-01 3.52197379e-01 -3.33974391e-01 -1.25164616e+00 6.97136402e-01 1.44895351e+00 3.73055935e-01 5.20010889e-01 9.13837373e-01 -7.98194110e-01 1.45691729e+00 -4.07832503e-01 -5.10068774e-01 6.95823459e-03 -7.33582139e-01 -1.22905564e+00 -2.75784910e-01 6.32302880e-01 -1.39073551e-01 -2.34197155e-01 9.03093040e-01 -1.78402185e-01 5.26308179e-01 8.76087546e-01 -1.28711438e+00 -7.72106886e-01 1.29357123e+00 9.97877494e-02 2.09804308e-02 9.58288014e-01 -3.58437657e-01 1.13162386e+00 -4.24603432e-01 1.02117312e+00 8.89356136e-01 8.20021093e-01 5.34395576e-01 -8.92031312e-01 -3.99333030e-01 1.59577936e-01 5.48581064e-01 -9.67426121e-01 -1.04986522e-02 8.52395117e-01 -6.69385791e-01 1.12654531e+00 -8.71978551e-02 5.99440396e-01 7.34472036e-01 2.23853081e-01 9.45976496e-01 7.28019118e-01 -7.70539939e-01 -9.54197645e-02 -6.93360250e-03 9.79806721e-01 1.22989357e+00 2.45377973e-01 -3.41639817e-01 -7.48661876e-01 2.14916497e-01 3.63657773e-01 -3.82381558e-01 -4.67384040e-01 -5.89723408e-01 -1.06260359e+00 9.85726535e-01 -1.47679886e-02 3.01020384e-01 -2.67138146e-02 -3.37764055e-01 6.36063099e-01 4.67101425e-01 2.57129043e-01 6.25736594e-01 -6.25479937e-01 -4.10073578e-01 -7.56446719e-01 4.09491867e-01 1.61705399e+00 1.49087238e+00 6.75340712e-01 -4.33747053e-01 -3.19524668e-02 6.89001858e-01 4.75342155e-01 7.55525753e-02 2.91277796e-01 -6.25434279e-01 1.14839864e+00 1.08806431e+00 -2.08838582e-02 -1.15356660e+00 -5.68087280e-01 4.53286946e-01 -2.88645506e-01 -3.62748116e-01 8.34484816e-01 -2.10041180e-01 -6.77369833e-01 1.52411389e+00 4.25557464e-01 -2.56677828e-04 3.67123246e-01 5.23248553e-01 1.65860462e+00 9.43200171e-01 5.83689451e-01 9.43371207e-02 1.35786915e+00 -1.45430696e+00 -1.07921076e+00 -2.93391883e-01 1.13175154e+00 -6.37608647e-01 1.24808967e+00 3.34379673e-01 -8.51848066e-01 -7.34983206e-01 -1.01374054e+00 -6.87026203e-01 -8.05089414e-01 1.70460030e-01 6.26421452e-01 4.26428646e-01 -3.71918172e-01 6.90823674e-01 -5.95658481e-01 -6.79520726e-01 -2.15520542e-02 -1.84573695e-01 -3.85832846e-01 -4.35336649e-01 -1.33543336e+00 9.02863622e-01 9.59033787e-01 -3.42461824e-01 -6.46667898e-01 -1.18818438e+00 -1.40328252e+00 3.62575084e-01 6.44283891e-01 -6.24435782e-01 1.26393783e+00 -5.04956603e-01 -1.53911090e+00 7.67276347e-01 8.34079087e-02 -1.92773834e-01 2.07432121e-01 -5.93346655e-01 -4.38769966e-01 -7.18811676e-02 2.73946136e-01 1.25964269e-01 3.87595922e-01 -8.44567835e-01 -7.48305678e-01 -1.69899359e-01 3.74079555e-01 1.86410859e-01 -2.13354945e-01 -1.07309178e-01 -2.93304056e-01 -4.74738091e-01 -1.30831013e-02 -7.31992602e-01 1.03846207e-01 -6.61889434e-01 -3.85948658e-01 -7.12976933e-01 5.37836969e-01 -1.13301003e+00 1.84257531e+00 -2.06608081e+00 1.84753388e-01 -7.93239567e-03 2.67616391e-01 7.78213069e-02 -1.82157189e-01 7.57504821e-01 -1.04583688e-01 2.19407622e-02 -1.39215142e-01 3.36306214e-01 2.81369388e-01 1.71534330e-01 -1.41483173e-01 -3.33984382e-02 -1.94450319e-01 5.23196936e-01 -1.06039751e+00 -7.38904834e-01 2.27844551e-01 -2.47817010e-01 -8.32433879e-01 7.87483335e-01 -7.53244400e-01 1.60600111e-01 -7.73444533e-01 7.49275759e-02 5.82337499e-01 -3.65464725e-02 6.30747676e-01 -3.75669837e-01 -7.61144981e-02 7.71369815e-01 -1.23083901e+00 2.25232434e+00 -5.85732758e-01 5.15688896e-01 -1.04169235e-01 -1.13204205e+00 1.17700815e+00 2.17790082e-01 3.46898884e-01 -4.07857656e-01 -2.55404077e-02 -1.71050563e-01 -2.38410607e-02 -1.03717661e+00 7.82951653e-01 9.15149897e-02 -5.36332607e-01 6.22386575e-01 6.09482706e-01 -4.44564790e-01 9.81052160e-01 5.03617287e-01 1.09357905e+00 8.98884892e-01 5.92543960e-01 -4.62159365e-01 7.22577214e-01 8.40611875e-01 3.25002223e-01 4.72705066e-01 -6.64017797e-02 6.64888844e-02 6.50247395e-01 -4.08942431e-01 -7.22316742e-01 -9.32282269e-01 2.40047351e-01 1.56256425e+00 8.82804617e-02 -1.32920063e+00 -9.71431792e-01 -1.12600338e+00 -1.42220274e-01 1.32029653e+00 -4.50988233e-01 -1.48139581e-01 -8.35046589e-01 -1.75087065e-01 4.33285236e-01 5.66378415e-01 6.47840023e-01 -9.02717173e-01 -5.54067314e-01 3.31710637e-01 -8.44376743e-01 -1.26326823e+00 -5.58312953e-01 1.45008922e-01 -6.58533335e-01 -1.59335339e+00 1.66837215e-01 -1.24457836e+00 4.96218860e-01 -1.08515238e-02 1.60238278e+00 5.85752241e-02 1.40319662e-02 6.01178169e-01 -7.01359272e-01 -6.56194031e-01 -9.12535489e-01 2.93583602e-01 -6.94464087e-01 -7.64899552e-01 8.39922607e-01 -1.40133267e-02 -1.71127226e-02 5.98840937e-02 -7.12348998e-01 5.96185446e-01 -1.78846642e-01 5.36219835e-01 8.85338262e-02 1.47354633e-01 2.08341569e-01 -1.56880891e+00 8.64391685e-01 -5.65252721e-01 -3.95932794e-01 6.88120723e-01 -6.45598471e-01 3.66423637e-01 9.38744605e-01 -1.70728713e-01 -1.72205031e+00 1.97401252e-02 9.29124206e-02 2.52519846e-01 -3.44834805e-01 8.45341921e-01 -2.73326457e-01 4.72162068e-01 7.71355331e-01 -2.41641894e-01 -4.16481942e-01 -3.89043987e-01 7.95457959e-01 3.28348756e-01 5.15188098e-01 -1.17235279e+00 5.26135385e-01 -4.66302961e-01 -1.04649372e-01 -7.00135887e-01 -9.58898902e-01 -6.66944504e-01 -9.72158313e-01 -2.71376044e-01 1.12435961e+00 -8.24629962e-01 -4.01828796e-01 1.34063631e-01 -1.07344115e+00 -7.39938676e-01 -1.17667168e-01 4.40316260e-01 -6.40266001e-01 4.48608965e-01 -8.04419518e-01 -1.44273922e-01 -1.51229411e-01 -6.98101938e-01 4.76108938e-01 3.93817514e-01 -5.50387621e-01 -1.48218429e+00 4.53100830e-01 6.53334796e-01 4.27519828e-02 2.97669560e-01 1.49973929e+00 -9.51097608e-01 -3.93847853e-01 1.62521183e-01 2.97251139e-02 -6.64406270e-02 1.60651118e-01 2.69990116e-01 -3.55552524e-01 3.68488967e-01 -1.96130499e-01 -3.36155385e-01 4.65965539e-01 1.26680797e-02 8.62090170e-01 -2.06210509e-01 -2.85413414e-01 5.52692413e-01 1.46819997e+00 1.36499524e-01 6.64653420e-01 5.75152040e-01 9.53201056e-01 8.52656126e-01 9.60912585e-01 1.73497036e-01 8.88020992e-01 7.83676505e-02 -9.48073417e-02 3.24902952e-01 -3.22127312e-01 -1.07284105e+00 4.10702258e-01 1.20611537e+00 -9.15740654e-02 -3.77194554e-01 -1.11350870e+00 7.42749691e-01 -1.85604489e+00 -9.15753484e-01 -7.52737999e-01 1.66334188e+00 1.04300106e+00 -1.36133701e-01 -1.73343137e-01 -2.28185236e-01 5.19408286e-01 1.11768857e-01 -6.00754889e-03 -8.16131532e-01 4.93599951e-01 4.33873147e-01 1.86917305e-01 5.53817928e-01 -1.21576226e+00 1.16801167e+00 5.96340370e+00 5.37696779e-01 -3.59134823e-01 -1.81920489e-03 -1.84145808e-01 5.86376786e-01 -2.24681869e-01 1.30395815e-01 -8.64882827e-01 -1.04206912e-01 1.13861883e+00 -3.82626265e-01 3.88250083e-01 9.28713083e-01 7.35573769e-02 -2.39796191e-01 -1.46725690e+00 3.76493722e-01 1.67343915e-01 -1.18177640e+00 -7.11557344e-02 -6.57704651e-01 7.11810470e-01 -4.41757828e-01 -8.75375688e-01 9.74174440e-01 9.21529830e-01 -7.73106992e-01 5.12935460e-01 3.45698416e-01 1.98130310e-01 -4.52286869e-01 4.51020926e-01 6.78863287e-01 -1.38751006e+00 2.00254872e-01 -3.85586798e-01 -6.69099465e-02 -8.84349793e-02 1.12587243e-01 -1.15843511e+00 9.16296542e-01 6.56235754e-01 9.94099557e-01 -6.75827980e-01 5.98020315e-01 -8.27825069e-01 8.52202117e-01 -4.58086133e-02 -3.46161276e-01 7.83802755e-03 -7.42648721e-01 2.24726140e-01 1.50015104e+00 2.18881264e-01 3.06767732e-01 3.14919442e-01 1.01737487e+00 -5.89120388e-03 5.80614507e-01 -8.62818062e-01 -3.04815382e-01 1.43894553e-01 1.27702200e+00 -5.43268740e-01 -3.90671313e-01 -9.34706926e-01 7.14479327e-01 5.49145103e-01 2.07031026e-01 -9.06571567e-01 -6.29963875e-01 2.28050247e-01 1.00725800e-01 2.47975126e-01 -2.68266141e-01 -2.58118600e-01 -1.27173781e+00 -3.09609622e-01 -9.03345227e-01 8.61742675e-01 -8.94423127e-01 -1.19424975e+00 1.45106986e-01 3.58484000e-01 -9.76353526e-01 -1.78410470e-01 -8.22933972e-01 -8.04313064e-01 6.10888183e-01 -1.43192971e+00 -1.15191162e+00 -4.91207600e-01 8.30097735e-01 6.35827601e-01 1.49663106e-01 1.04780197e+00 1.79487094e-01 -4.64145958e-01 1.90893367e-01 -1.48598209e-01 5.50371230e-01 8.41006100e-01 -1.59886730e+00 1.66063651e-01 1.01200056e+00 3.00674796e-01 7.07404852e-01 7.89645612e-01 -9.32456911e-01 -1.43211949e+00 -1.14552295e+00 1.14804626e+00 -5.39684772e-01 8.57166827e-01 -3.82185519e-01 -1.13520396e+00 1.07149720e+00 7.24992394e-01 -4.67823595e-01 1.08484280e+00 4.22365636e-01 -2.46026024e-01 3.18508953e-01 -9.29498076e-01 5.54232299e-01 1.11033392e+00 -4.05044585e-01 -1.46366727e+00 3.32137585e-01 6.33154452e-01 -8.15408528e-01 -1.51628876e+00 1.31183207e-01 1.25813693e-01 -6.75631523e-01 7.27098763e-01 -8.99366796e-01 1.00272787e+00 -1.70370921e-01 1.49704563e-02 -1.65134180e+00 -3.02892178e-01 -3.88722569e-01 -1.37335241e-01 1.43381846e+00 6.54467583e-01 -3.71439062e-04 7.80240417e-01 8.94323409e-01 -4.81029212e-01 8.12103525e-02 -1.85305044e-01 -2.84611374e-01 3.21382612e-01 -3.67658556e-01 4.19118971e-01 1.34381676e+00 8.33953500e-01 8.46518338e-01 7.26733059e-02 7.66176805e-02 2.93761641e-01 4.93978053e-01 1.03638554e+00 -1.30811954e+00 -9.12577212e-02 -3.88907678e-02 3.37219775e-01 -1.07021713e+00 4.66133296e-01 -1.30466056e+00 3.19742024e-01 -2.12708616e+00 7.83536211e-02 -1.01547800e-01 3.41352969e-01 5.14742911e-01 -5.11174142e-01 -7.78146327e-01 3.24387878e-01 -2.57879108e-01 -7.28537560e-01 1.58773914e-01 1.50951529e+00 -2.75818706e-01 -3.28531772e-01 -3.80796969e-01 -4.44725126e-01 8.82086396e-01 8.27424705e-01 -3.40190679e-01 -6.85620070e-01 -4.04767752e-01 3.01582426e-01 1.96464553e-01 -2.74250716e-01 -9.10768270e-01 2.85491288e-01 -5.30172586e-01 -7.19697326e-02 -5.98943710e-01 -3.87253881e-01 -8.17260325e-01 1.01037482e-02 2.28993088e-01 -5.43455243e-01 -7.01841433e-03 3.54729980e-01 3.03702384e-01 -3.53181124e-01 -8.15347850e-01 3.69439960e-01 -3.83799762e-01 -1.36171842e+00 -2.11336821e-01 -5.50313115e-01 5.88990986e-01 1.16422915e+00 -7.52105638e-02 -6.77317739e-01 -9.88248512e-02 -7.76929080e-01 2.98671067e-01 9.66883227e-02 6.25538826e-01 4.20795560e-01 -1.07617974e+00 -5.86884558e-01 1.84298173e-01 3.42360973e-01 1.63342342e-01 1.22014254e-01 4.05053794e-01 -1.00826311e+00 5.53955138e-01 -3.15636754e-01 -2.00267553e-01 -1.33271623e+00 6.49985075e-01 1.32016584e-01 -8.81440818e-01 -6.44156396e-01 4.40522671e-01 1.60135597e-01 -1.08335793e+00 -1.72933079e-02 -8.64233255e-01 -8.87823105e-01 -7.40376413e-02 2.22330526e-01 4.31218028e-01 -2.35644635e-02 -2.15432420e-01 5.84436627e-03 4.89852935e-01 1.50455371e-01 5.35075724e-01 1.36937571e+00 -7.13427812e-02 -1.91182807e-01 2.83874214e-01 8.88158083e-01 1.96458295e-01 -7.05283940e-01 -6.83211014e-02 3.98423672e-01 -1.69506535e-01 -1.50924191e-01 -7.93431342e-01 -6.66262329e-01 7.54297197e-01 -3.71736437e-02 4.22857642e-01 7.97532558e-01 -5.56873120e-02 6.59419358e-01 5.03649950e-01 3.58198106e-01 -1.35225606e+00 -5.44861034e-02 1.01762092e+00 6.32306576e-01 -1.03053224e+00 1.91786848e-02 -1.08083010e+00 -8.02284658e-01 1.68518889e+00 1.14655173e+00 3.64180207e-02 8.12075436e-01 1.35273054e-01 -6.74557686e-02 -4.31014538e-01 -6.14161134e-01 -4.67209555e-02 3.41785312e-01 7.27117598e-01 9.23879683e-01 2.60811090e-01 -6.23473167e-01 1.05019271e+00 -2.86740124e-01 4.39493358e-01 7.88931668e-01 9.59528685e-01 -4.51168299e-01 -1.21373260e+00 -3.62944216e-01 3.85568619e-01 -2.68097937e-01 -2.14887053e-01 -3.45755666e-01 1.13695097e+00 8.56158286e-02 1.24418056e+00 -8.81598517e-03 -1.20155826e-01 7.53376544e-01 4.68574971e-01 7.79577255e-01 -9.78610218e-01 -9.14877892e-01 -4.32382762e-01 8.42179477e-01 -5.29562950e-01 -7.30782866e-01 -4.61084276e-01 -1.71157038e+00 -3.83708298e-01 -2.51865596e-01 5.43785334e-01 3.64425689e-01 8.87372673e-01 -1.51862320e-03 7.82010436e-01 -2.93888408e-03 2.88368464e-02 -2.19364226e-01 -1.10428011e+00 -5.25134623e-01 9.87772346e-01 -3.87715220e-01 -5.23206770e-01 1.21902442e-02 6.64078593e-01]

[9.83618450164795, 8.073934555053711]

a559fd21-2b0e-4bc7-87fd-0ecbbab95985

evaluating-methods-for-extraction-of-aspect

https://aclanthology.org/2022.lrec-1.407

https://aclanthology.org/2022.lrec-1.407.pdf

Evaluating Methods for Extraction of Aspect Terms in Opinion Texts in Portuguese - the Challenges of Implicit Aspects

One of the challenges of aspect-based sentiment analysis is the implicit mention of aspects. These are more difficult to identify and may require world knowledge to do so. In this work, we evaluate frequency-based, hybrid, and machine learning methods, including the use of the pre-trained BERT language model, in the task of extracting aspect terms in opinionated texts in Portuguese, emphasizing the analysis of implicit aspects. Besides the comparative evaluation of methods, the differential of this work lies in the analysis’s novelty using a typology of implicit aspects that shows the knowledge needed to identify each implicit aspect term, thus allowing a mapping of the strengths and weaknesses of each method.

['Thiago Alexandre Salgueiro Pardo', 'Mateus Machado']

lrec-2022-6

['aspect-based-sentiment-analysis']

['natural-language-processing']

[-9.46451095e-04 4.60972399e-01 -4.53090072e-01 -1.96032673e-01 -4.35418576e-01 -8.60824585e-01 9.60162580e-01 8.89893234e-01 -5.33323288e-01 5.78905940e-01 4.83192176e-01 -3.17756921e-01 -4.24850166e-01 -7.48272717e-01 -1.57337606e-01 -4.67829227e-01 2.05718353e-01 5.93248487e-01 2.96072457e-02 -7.16302872e-01 5.59605777e-01 4.00022030e-01 -1.49182749e+00 4.55895364e-01 6.83197677e-01 8.83682787e-01 -1.63613290e-01 1.03119083e-01 -8.76312077e-01 7.46479630e-01 -9.14394259e-01 -7.93679893e-01 -1.55734584e-01 -1.91772789e-01 -8.16926837e-01 -9.29518342e-02 2.20611736e-01 2.55021602e-01 5.92474699e-01 7.59143233e-01 2.49776885e-01 -1.73585057e-01 9.61576104e-01 -8.75219643e-01 -4.06193346e-01 7.54673362e-01 -3.35976332e-01 2.17801437e-01 7.13493109e-01 -2.50376850e-01 1.33547771e+00 -8.21377516e-01 7.00698018e-01 8.26060891e-01 8.09346557e-01 1.81339949e-01 -9.10328448e-01 -1.26524553e-01 4.12428230e-01 1.37092158e-01 -1.01345921e+00 -4.69787866e-02 8.41110051e-01 -6.32602334e-01 1.35230410e+00 1.48778513e-01 1.08039463e+00 7.06455767e-01 2.35620290e-02 6.49599969e-01 1.48614419e+00 -7.96397090e-01 5.26981354e-01 8.98573160e-01 5.23757935e-01 2.05660358e-01 6.31740153e-01 -3.34135771e-01 -6.77826881e-01 -3.64411712e-01 -1.55009786e-02 -2.68114120e-01 -1.08142480e-01 -2.64394015e-01 -7.23873198e-01 9.49869454e-01 -1.59396157e-01 1.02578223e+00 -4.02789921e-01 -5.03015935e-01 5.02723575e-01 2.59551316e-01 8.46319675e-01 6.84022903e-01 -9.90105748e-01 -3.39585364e-01 -9.96339262e-01 1.11750878e-01 1.26078129e+00 7.21251190e-01 1.04700613e+00 -1.51361227e-01 5.43434471e-02 4.90568697e-01 3.76939505e-01 4.38157827e-01 6.08859360e-01 -3.43951255e-01 2.03783050e-01 1.35140204e+00 -1.98951624e-02 -9.47485745e-01 -3.02661091e-01 -5.20241320e-01 -1.80525944e-01 7.91729689e-02 1.25543728e-01 -2.14992404e-01 -6.92747414e-01 1.05962753e+00 4.04429674e-01 -2.80223101e-01 2.65324891e-01 3.44765425e-01 8.39055240e-01 2.27066323e-01 1.82745233e-01 -3.29280198e-01 1.71682465e+00 -5.84762812e-01 -8.01591575e-01 -3.44632983e-01 5.42446673e-01 -9.42545772e-01 9.96725917e-01 3.20810765e-01 -6.53683662e-01 -1.37117207e-01 -9.34267759e-01 2.24739388e-01 -1.29727721e+00 1.51628181e-01 9.89081264e-01 9.12878931e-01 -8.70166719e-01 3.31825048e-01 -3.16232353e-01 -3.75632852e-01 5.47711775e-02 2.63889819e-01 -4.03006554e-01 4.43833590e-01 -1.17044818e+00 1.28339243e+00 4.97522086e-01 -2.06193268e-01 1.43904477e-01 -8.77547503e-01 -1.01520979e+00 4.08460274e-02 5.35570621e-01 -5.08553386e-01 1.01217711e+00 -1.37262321e+00 -1.21843600e+00 1.01942515e+00 -4.38386858e-01 -2.31013343e-01 6.91622123e-02 -3.87383878e-01 -5.66130102e-01 2.99746424e-01 2.93396056e-01 -7.03031197e-02 6.48102582e-01 -1.28715050e+00 -8.13593984e-01 -4.05904233e-01 5.11413097e-01 2.45438050e-02 -4.02716458e-01 4.62188013e-02 -1.09498166e-01 -5.81391811e-01 -3.14137757e-01 -7.79122770e-01 -2.11063042e-01 -5.70534289e-01 5.11809848e-02 -3.66609186e-01 8.43146503e-01 -4.98575896e-01 1.36123371e+00 -2.19998384e+00 -4.33197096e-02 3.42882097e-01 1.32096276e-01 3.15306723e-01 1.25003964e-01 8.58174086e-01 4.98214066e-02 2.69352674e-01 -2.11109474e-01 -1.95278108e-01 5.49718216e-02 1.97202086e-01 -3.62581819e-01 -9.82629601e-03 4.59131926e-01 8.68201494e-01 -9.07770813e-01 -4.00760472e-01 9.09066126e-02 6.09460354e-01 -1.77477360e-01 -1.34215057e-01 -2.50698566e-01 -5.65521717e-02 -5.32224417e-01 6.31061554e-01 4.82481331e-01 5.65389842e-02 4.52937156e-01 -4.21310097e-01 -3.87343675e-01 7.30522096e-01 -1.13100517e+00 1.28566444e+00 -7.23308325e-01 6.15261912e-01 -4.12130177e-01 -8.67412925e-01 7.67700136e-01 4.98608172e-01 4.96346205e-01 -5.50270259e-01 2.08888859e-01 5.67172527e-01 -2.27411330e-01 -4.52658981e-01 7.08732069e-01 -4.41555738e-01 -3.48572321e-02 6.79100633e-01 2.80205220e-01 -5.12571216e-01 6.78722441e-01 1.40517861e-01 5.70363343e-01 4.90645505e-02 1.02872360e+00 -5.42820811e-01 9.01741326e-01 2.42281199e-01 2.17778802e-01 3.08211356e-01 2.84039825e-01 5.21451056e-01 8.33941221e-01 -5.57276070e-01 -7.04925656e-01 -5.98826349e-01 -1.30133480e-01 7.86679268e-01 -2.18837425e-01 -8.19754362e-01 -5.72100163e-01 -1.06253624e+00 -8.89524966e-02 9.04625237e-01 -9.38869894e-01 2.03111649e-01 -4.98939395e-01 -7.54440665e-01 4.68028113e-02 1.83061361e-01 -3.55423838e-02 -9.99274611e-01 -8.62809062e-01 1.36947796e-01 -8.87761414e-02 -1.11651480e+00 5.30588143e-02 3.61397564e-01 -8.45115185e-01 -1.23307586e+00 -2.87099481e-01 -4.05245304e-01 6.39493048e-01 -3.42482030e-02 1.51626003e+00 8.19377601e-02 1.04940757e-01 6.42867386e-01 -8.04245651e-01 -1.10830104e+00 -2.54478246e-01 4.01106089e-01 -3.47530186e-01 -1.32860422e-01 1.06595051e+00 -5.48845828e-01 -3.24796140e-01 -1.14902951e-01 -1.14840376e+00 -5.56987047e-01 8.47876906e-01 4.35485214e-01 5.59027553e-01 -8.62641335e-02 2.17390999e-01 -1.44656801e+00 8.30757976e-01 -5.43727875e-01 -2.81605542e-01 2.67918557e-01 -1.04275870e+00 1.55596301e-01 3.15748245e-01 -3.94804031e-01 -8.84023428e-01 -2.82532930e-01 -2.96718657e-01 4.78289276e-01 -1.93746418e-01 9.40980494e-01 -1.15760989e-01 4.09188233e-02 4.30903733e-01 1.78500026e-01 -1.02049991e-01 -4.82757002e-01 3.39910626e-01 4.60104525e-01 -3.72929692e-01 -3.91648710e-01 6.49192095e-01 6.27537966e-01 -2.12663606e-01 -9.52669144e-01 -1.28687131e+00 -7.57577062e-01 -7.71630168e-01 -1.06344163e-01 5.27425706e-01 -8.32424045e-01 -1.86256215e-01 1.76528960e-01 -1.17451596e+00 2.26530790e-01 -9.87406194e-01 3.48856747e-01 -1.88069716e-01 5.54760396e-01 -1.75678045e-01 -9.09912467e-01 -5.50823092e-01 -8.74750674e-01 1.06357336e+00 2.98649877e-01 -6.91921294e-01 -1.32764733e+00 4.01572198e-01 2.62519330e-01 6.86308026e-01 1.09835245e-01 1.12077451e+00 -1.04562175e+00 -8.89101997e-02 -4.47047353e-01 2.21263349e-01 4.03754652e-01 4.79448110e-01 1.99156366e-02 -1.02841020e+00 9.62758064e-02 4.34289783e-01 6.28727674e-02 5.43918848e-01 5.55078089e-02 -1.39807649e-02 -3.60632390e-01 -5.35338111e-02 1.28488969e-02 1.65777421e+00 3.69134136e-02 5.87077737e-01 9.45970535e-01 2.31114089e-01 1.02284563e+00 7.69199193e-01 2.74841577e-01 4.40583140e-01 6.74864590e-01 2.12336302e-01 3.37512977e-02 -1.47977564e-02 4.93061543e-02 4.73981053e-01 9.45474446e-01 -2.03242555e-01 1.78122148e-01 -5.78834116e-01 7.48895049e-01 -1.48185825e+00 -7.11225927e-01 -2.22974151e-01 1.98732567e+00 6.91622615e-01 3.40225071e-01 2.22997099e-01 4.53429341e-01 1.21469595e-01 3.61771435e-01 2.32416123e-01 -8.60079229e-01 -3.08410138e-01 6.08239651e-01 4.18685041e-02 4.57743108e-01 -1.01283586e+00 7.08233058e-01 6.42456388e+00 7.70414114e-01 -1.05438566e+00 2.22260877e-01 1.93103552e-01 1.85419381e-01 -7.50818372e-01 3.38373810e-01 -8.27582002e-01 2.67835736e-01 8.39153528e-01 -1.67454034e-01 -4.03199382e-02 9.19509470e-01 1.32844299e-01 -3.81782264e-01 -7.12746859e-01 3.29320908e-01 5.94387174e-01 -9.92598474e-01 4.06363398e-01 -1.74215715e-02 5.65258861e-01 -9.66410562e-02 -2.40458429e-01 1.43503726e-01 -1.75003201e-01 -7.06876516e-01 6.80877924e-01 3.42622101e-01 2.18415827e-01 -7.00051010e-01 1.30057144e+00 -3.66698876e-02 -1.07603395e+00 1.48102283e-01 -1.52819082e-01 -3.52719396e-01 1.73705891e-01 1.01192665e+00 -3.65897506e-01 9.22664225e-01 5.42320073e-01 6.39883578e-01 -6.04430974e-01 7.47147620e-01 -7.60274053e-01 5.87627828e-01 -2.58275181e-01 -4.27917451e-01 3.36743206e-01 -4.30547863e-01 6.05801940e-01 1.51988447e+00 2.00005487e-01 -2.88777560e-01 -1.06839105e-01 5.00574946e-01 4.97094333e-01 7.61758327e-01 -6.74785316e-01 -1.99413955e-01 -8.33506659e-02 1.54983616e+00 -9.13655639e-01 -3.08410883e-01 -7.28219211e-01 4.70241815e-01 1.70938611e-01 1.48485020e-01 -1.41405374e-01 -3.67823392e-01 5.35907745e-01 2.42139295e-01 5.75192571e-01 -1.33730620e-01 -5.46805918e-01 -1.08322799e+00 4.14481282e-01 -9.71925616e-01 5.70484400e-01 -3.87157708e-01 -1.19593263e+00 8.79640043e-01 7.37643987e-02 -1.22530341e+00 -4.18809116e-01 -8.47525120e-01 -5.60915589e-01 8.07061195e-01 -1.77328646e+00 -1.56493628e+00 -1.65655874e-02 1.38673097e-01 3.56101930e-01 -1.96342751e-01 8.49456549e-01 2.27317214e-01 1.84551269e-01 2.23461017e-01 -4.51711863e-01 -2.00874463e-01 4.37259823e-01 -1.58944941e+00 2.31178641e-01 5.73472559e-01 3.39639634e-01 7.43857324e-01 9.87132668e-01 -5.77406704e-01 -1.00611508e+00 -5.34750462e-01 1.65770721e+00 -8.17948401e-01 9.98079658e-01 -8.21729302e-02 -7.24681616e-01 4.86921847e-01 5.27626574e-01 -7.00666726e-01 1.22208977e+00 5.82002759e-01 -6.24260008e-01 -5.96427941e-04 -9.44835663e-01 3.77749681e-01 3.76625955e-01 -6.73592508e-01 -1.21387589e+00 -1.75073698e-01 4.99876738e-01 4.19139266e-02 -8.01089346e-01 4.14492935e-01 6.22205853e-01 -1.01566160e+00 7.71488011e-01 -3.83337557e-01 3.85068208e-01 -4.18716282e-01 1.15049705e-01 -1.29933250e+00 1.24614656e-01 -1.31294981e-01 2.54892558e-02 1.61142182e+00 9.86472547e-01 -8.47892940e-01 6.96495473e-01 2.11878002e-01 2.98361659e-01 -8.02920341e-01 -8.36488008e-01 -4.16901469e-01 1.03423983e-01 -6.01423800e-01 5.07008731e-01 8.53237867e-01 4.09272045e-01 6.72542989e-01 -9.99889225e-02 -1.80976570e-01 1.59986522e-02 1.97812885e-01 5.77290595e-01 -1.25416279e+00 -4.17405814e-01 -6.40236676e-01 -5.66705763e-01 -3.19403887e-01 1.01369061e-01 -4.12208617e-01 -4.64696020e-01 -1.60414696e+00 2.76230693e-01 -1.58836931e-01 -2.95286387e-01 3.04903179e-01 -3.86248708e-01 2.58232445e-01 1.85346469e-01 1.31196797e-01 -5.23574829e-01 2.83289552e-01 7.11033702e-01 -2.79225618e-01 -1.93441063e-01 2.34375402e-01 -1.18791449e+00 8.95301521e-01 7.62621403e-01 -7.34763563e-01 -2.53876925e-01 -1.46674231e-01 1.09705997e+00 -8.74305904e-01 -1.73279583e-01 -6.01655543e-01 2.93754674e-02 1.01398379e-02 -1.73720405e-01 -5.54253936e-01 2.14014381e-01 -1.09893429e+00 -6.06485717e-02 5.68853617e-01 1.72782480e-03 3.41823518e-01 3.38705063e-01 1.84587911e-01 -6.23568475e-01 -8.47172737e-01 1.59970745e-01 -3.42367679e-01 -7.79860675e-01 -3.16557765e-01 -5.85003614e-01 2.83197492e-01 8.27475786e-01 -2.15666994e-01 -3.19952331e-02 -3.02231848e-01 -3.72013807e-01 -3.37410241e-01 4.55832332e-01 3.55763853e-01 2.58697569e-01 -7.52073348e-01 -5.24620235e-01 -2.67182201e-01 4.47659224e-01 -4.35817748e-01 1.01790436e-01 9.65837479e-01 -3.11105400e-01 3.97325933e-01 5.36794998e-02 -1.49128705e-01 -1.17641687e+00 6.78153098e-01 1.50857210e-01 -7.73991466e-01 -2.93367654e-01 1.33265451e-01 -1.78586692e-01 -5.36456048e-01 -2.75787354e-01 -1.18917398e-01 -1.12996674e+00 7.97351778e-01 5.73917508e-01 1.12720646e-01 3.28625411e-01 -7.57647574e-01 -4.86210018e-01 9.19858873e-01 -4.18768413e-02 -2.53479838e-01 1.34589231e+00 -1.24492973e-01 -5.93019187e-01 7.85146773e-01 7.21355200e-01 8.82059872e-01 -1.80173501e-01 -1.21093087e-01 5.32360137e-01 -1.11109897e-01 -2.89127916e-01 -1.09779477e+00 -8.59610498e-01 6.11408591e-01 3.20623040e-01 5.97247064e-01 1.02611721e+00 1.07629791e-01 3.53342384e-01 2.69277304e-01 3.53447139e-01 -1.18355930e+00 -2.74511039e-01 6.81217194e-01 5.07177532e-01 -1.17544651e+00 3.51156741e-01 -7.69418657e-01 -5.89503348e-01 1.18535197e+00 1.22103289e-01 2.00506896e-02 8.94203305e-01 5.03704965e-01 5.87002277e-01 -6.55751646e-01 -5.17327547e-01 -6.88835382e-01 5.21290898e-01 7.17668414e-01 8.53676140e-01 -4.89789173e-02 -1.35047936e+00 6.88523948e-01 -3.01709145e-01 -1.59294724e-01 5.52123427e-01 1.19067585e+00 -1.97059661e-01 -1.56089604e+00 -1.42066739e-02 5.01468837e-01 -9.85407174e-01 -4.15721208e-01 -9.68558311e-01 1.00268054e+00 3.15156162e-01 1.01556337e+00 -3.87497514e-01 1.08864643e-02 5.55390596e-01 1.24981828e-01 1.51672572e-01 -8.04112256e-01 -1.30535161e+00 -1.38744861e-01 3.61947447e-01 -2.22539335e-01 -9.10485387e-01 -7.34907925e-01 -5.83953500e-01 2.12452412e-01 -3.64846677e-01 7.96365559e-01 1.18498349e+00 1.48882735e+00 3.22491169e-01 3.13946187e-01 3.07791412e-01 -3.83494616e-01 -1.21803634e-01 -9.12002683e-01 -4.46614832e-01 5.78000963e-01 1.10553555e-01 -5.07533908e-01 -6.95484102e-01 -1.96091339e-01]

[11.264972686767578, 6.8067522048950195]

613e5362-3579-4ccd-9ee6-3236c67c9505

supervised-visual-attention-for-simultaneous

2201.09324

https://arxiv.org/abs/2201.09324v2

https://arxiv.org/pdf/2201.09324v2.pdf

Supervised Visual Attention for Simultaneous Multimodal Machine Translation

Recently, there has been a surge in research in multimodal machine translation (MMT), where additional modalities such as images are used to improve translation quality of textual systems. A particular use for such multimodal systems is the task of simultaneous machine translation, where visual context has been shown to complement the partial information provided by the source sentence, especially in the early phases of translation. In this paper, we propose the first Transformer-based simultaneous MMT architecture, which has not been previously explored in the field. Additionally, we extend this model with an auxiliary supervision signal that guides its visual attention mechanism using labelled phrase-region alignments. We perform comprehensive experiments on three language directions and conduct thorough quantitative and qualitative analyses using both automatic metrics and manual inspection. Our results show that (i) supervised visual attention consistently improves the translation quality of the MMT models, and (ii) fine-tuning the MMT with supervision loss enabled leads to better performance than training the MMT from scratch. Compared to the state-of-the-art, our proposed model achieves improvements of up to 2.3 BLEU and 3.5 METEOR points.

['Lucia Specia', 'Ozan Caglayan', 'Veneta Haralampieva']

2022-01-23

['multimodal-machine-translation']

['natural-language-processing']

[ 4.35202777e-01 -3.76122333e-02 -4.10251737e-01 -1.14975259e-01 -1.10257912e+00 -7.00142205e-01 1.06467187e+00 -9.08689864e-04 -2.59738892e-01 6.39913678e-01 3.01235884e-01 -7.38426805e-01 5.12067199e-01 -1.84149921e-01 -8.01052928e-01 -4.84392285e-01 5.11532009e-01 5.67492127e-01 -5.66663705e-02 -4.03838903e-01 2.80745655e-01 1.44535556e-01 -9.39288914e-01 5.86540461e-01 1.10907257e+00 7.43683815e-01 3.15794855e-01 4.32268918e-01 -3.05261821e-01 4.36129332e-01 -4.72490877e-01 -8.48531306e-01 2.01348558e-01 -7.45688498e-01 -9.20283377e-01 3.15141529e-01 5.30806363e-01 -5.45220412e-02 -2.21403360e-01 8.98315012e-01 5.74078977e-01 -3.23549181e-01 5.52817106e-01 -1.17754352e+00 -1.13846064e+00 4.86078948e-01 -9.42941725e-01 1.68169722e-01 4.44408923e-01 2.98616052e-01 1.06109047e+00 -1.40624702e+00 7.62201905e-01 1.40472722e+00 1.38227507e-01 4.56544638e-01 -1.36774778e+00 -3.95634770e-01 4.24432978e-02 2.14003220e-01 -1.10173583e+00 -5.67184389e-01 6.65005207e-01 -2.98341483e-01 1.05072474e+00 2.20997140e-01 4.70469207e-01 1.26483965e+00 3.04887354e-01 1.02490032e+00 1.47230160e+00 -8.89539480e-01 -1.65992364e-01 4.24854755e-01 -3.61958355e-01 7.66362906e-01 -2.29319736e-01 -8.91453326e-02 -7.09647954e-01 2.84877032e-01 5.98739684e-01 -2.59067118e-01 -3.22541356e-01 -3.24881196e-01 -1.71474421e+00 7.03217089e-01 4.19791281e-01 4.57652032e-01 -1.74346492e-01 8.03513154e-02 2.86892056e-01 4.28408951e-01 7.21502781e-01 3.17231953e-01 -7.39638880e-03 -1.57515705e-01 -1.04960871e+00 -2.11207479e-01 3.79781365e-01 1.03712654e+00 5.42416930e-01 -8.05198550e-02 -5.04383266e-01 8.85045648e-01 3.96670908e-01 8.48397434e-01 2.82683313e-01 -6.30949855e-01 1.18182862e+00 7.53278673e-01 6.86447993e-02 -7.08627701e-01 -3.79907303e-02 -4.60369289e-01 -6.91476643e-01 -3.21405493e-02 3.82340372e-01 1.40400347e-03 -9.53989267e-01 1.57436955e+00 1.55408308e-01 -4.19601858e-01 7.62432162e-03 1.13055694e+00 6.25878811e-01 7.98650265e-01 -1.95453122e-01 -2.50297278e-01 1.29204226e+00 -1.36323667e+00 -8.66963565e-01 -3.59291553e-01 5.34949303e-01 -1.28157508e+00 1.52598071e+00 1.80684905e-02 -1.17774248e+00 -4.10379618e-01 -9.80484366e-01 -1.98754773e-01 -1.46702513e-01 3.12442988e-01 3.81282687e-01 3.81192118e-01 -1.28433514e+00 1.88627928e-01 -8.33925664e-01 -6.95002019e-01 4.70057607e-01 3.10855180e-01 -4.40417826e-01 -2.91418672e-01 -1.11431074e+00 1.27287412e+00 -2.63895184e-01 1.36461705e-01 -7.65280366e-01 -2.50515074e-01 -7.64631927e-01 -9.22887325e-02 2.63996094e-01 -9.39456642e-01 1.25787628e+00 -1.32473350e+00 -1.65277743e+00 1.00943625e+00 -5.70388675e-01 -1.99386120e-01 5.72270572e-01 -2.00801864e-01 -8.34889188e-02 4.21744585e-01 1.08733937e-01 9.94255424e-01 8.11291516e-01 -1.43543398e+00 -3.58649194e-01 -3.45424682e-01 2.04252258e-01 5.76742113e-01 -4.16933715e-01 4.00837898e-01 -8.28312516e-01 -6.92747414e-01 -2.07243636e-01 -1.08978891e+00 -1.42246643e-02 1.49475977e-01 -3.46022606e-01 4.96619269e-02 7.76379466e-01 -8.88432980e-01 1.06068981e+00 -1.84653449e+00 6.69356704e-01 -2.20427915e-01 4.88754213e-02 8.71378183e-02 -2.60704845e-01 7.19889104e-01 2.36784056e-01 1.46016836e-01 -3.53405803e-01 -7.55791724e-01 6.16665520e-02 1.12934634e-01 -3.87592763e-01 3.08160454e-01 4.69509929e-01 1.48140872e+00 -6.87626839e-01 -6.21735275e-01 4.44062464e-02 5.02533495e-01 -5.46292178e-02 1.66662782e-01 -1.93760276e-01 7.25646436e-01 -8.89409259e-02 8.48683536e-01 4.35159773e-01 -5.24873018e-01 2.00346619e-01 -9.64965150e-02 -1.79511115e-01 2.45990902e-01 -2.95923322e-01 1.99589336e+00 -5.45674086e-01 1.01904821e+00 4.35721502e-02 -7.42736936e-01 6.23905718e-01 4.66558188e-01 1.04105018e-01 -1.14326775e+00 1.42395049e-01 2.09467039e-01 1.96796115e-02 -3.84287119e-01 6.42572284e-01 -1.36979148e-01 7.12370276e-02 5.28753042e-01 1.26717180e-01 7.53487572e-02 2.11320192e-01 2.97077119e-01 7.13653982e-01 3.83692265e-01 6.57875016e-02 -4.43220101e-02 4.00141805e-01 3.13713968e-01 1.35150090e-01 3.89432669e-01 -6.54683188e-02 6.85072601e-01 3.79370511e-01 -3.81827913e-02 -1.21351469e+00 -7.64532983e-01 1.34280175e-01 1.02030861e+00 3.01314473e-01 -2.64608800e-01 -8.88879776e-01 -8.37015033e-01 -3.48209262e-01 5.55805683e-01 -5.20339787e-01 8.36573690e-02 -6.17030263e-01 -7.22283602e-01 3.70104402e-01 5.86546779e-01 5.49316585e-01 -8.57480109e-01 -4.14853603e-01 -1.39555052e-01 -6.52278185e-01 -1.35355330e+00 -7.62811184e-01 -1.25218466e-01 -1.11718357e+00 -5.17343819e-01 -1.16251135e+00 -7.79613793e-01 7.39686191e-01 4.61301357e-01 1.17825663e+00 -3.58011946e-02 1.27069443e-01 4.20428723e-01 -3.96737576e-01 -1.26336247e-01 -5.30472517e-01 7.53077790e-02 -1.13637611e-01 1.28627762e-01 7.48697519e-02 -2.81385422e-01 -5.48328519e-01 5.19650578e-01 -7.52214730e-01 6.53216124e-01 9.57145095e-01 9.51871216e-01 4.87428308e-01 -8.51167798e-01 3.03381532e-01 -5.09698391e-01 6.08461618e-01 -1.88085407e-01 -4.74329621e-01 5.07248402e-01 -6.88549697e-01 3.45653743e-02 4.12887096e-01 -4.00076985e-01 -9.37633574e-01 -1.72702029e-01 1.98323965e-01 -4.59728360e-01 5.28851114e-02 6.42913818e-01 -1.08721353e-01 -1.64768040e-01 4.33188885e-01 3.06360453e-01 3.07311583e-02 -3.54458481e-01 5.09290278e-01 8.09478343e-01 3.38234037e-01 -4.76907462e-01 9.31552172e-01 3.77092838e-01 -5.03384843e-02 -4.60487217e-01 -5.20831585e-01 -2.24516690e-01 -7.90738761e-01 -2.99647927e-01 8.63302767e-01 -8.66282880e-01 -2.68846333e-01 1.02088526e-01 -1.29705453e+00 -3.95224959e-01 3.75830561e-01 3.85571390e-01 -5.16182125e-01 4.99621481e-01 -6.79126084e-01 -7.19384968e-01 -4.59461540e-01 -1.48473394e+00 1.40382445e+00 1.92471314e-02 -1.39585376e-01 -9.23197269e-01 1.19842310e-03 7.82634377e-01 5.25377810e-01 -1.42761707e-01 9.54039872e-01 -2.95713335e-01 -8.47818911e-01 1.04547448e-01 -4.49792176e-01 -8.16732869e-02 7.88598955e-02 -1.49483040e-01 -8.30560803e-01 -4.02863950e-01 -2.93663532e-01 -3.15355003e-01 8.40965033e-01 3.20709050e-02 4.05919909e-01 -2.38382190e-01 -3.33348125e-01 3.20932746e-01 1.20111191e+00 7.14379773e-02 5.71081460e-01 3.72548163e-01 7.44489312e-01 6.02877557e-01 7.70926654e-01 -1.52834088e-01 5.11882782e-01 1.17486012e+00 5.18648088e-01 -6.47969782e-01 -4.47174489e-01 -2.65949339e-01 6.47645772e-01 9.37592924e-01 -2.20955670e-01 -4.31108713e-01 -8.89789224e-01 5.71592212e-01 -1.98932183e+00 -7.07870543e-01 9.71547067e-02 2.11474562e+00 8.34292531e-01 -1.67189371e-02 3.74712497e-02 -8.13132450e-02 6.70730233e-01 6.94660246e-02 -1.86752796e-01 -5.48612177e-01 -2.07434162e-01 -2.99157370e-02 1.79213762e-01 5.86570680e-01 -8.52577269e-01 1.11723948e+00 6.14068317e+00 6.58899009e-01 -1.26150024e+00 2.43124902e-01 6.51088893e-01 -1.16275798e-03 -4.67638701e-01 9.48732495e-02 -3.98737967e-01 3.02791148e-01 8.44552815e-01 2.03880012e-01 4.99906898e-01 2.42750242e-01 3.58244747e-01 -1.02513671e-01 -1.22134602e+00 9.73271966e-01 3.43525231e-01 -1.24655497e+00 2.16067314e-01 3.66937369e-01 8.21403444e-01 -1.09756142e-02 4.11010236e-01 1.84311628e-01 -1.21452048e-01 -9.93565440e-01 8.82230759e-01 3.06765765e-01 1.01186073e+00 -5.50493777e-01 7.05764055e-01 2.70707428e-01 -9.65257704e-01 1.62843242e-01 -6.35631830e-02 9.94531810e-02 3.11346173e-01 2.84564376e-01 -9.20583963e-01 8.74604702e-01 4.42573518e-01 7.50133693e-01 -7.07017541e-01 8.06258440e-01 -4.10312414e-01 7.60147691e-01 2.79630795e-02 -1.61166806e-02 4.35140043e-01 -2.64157176e-01 5.95633507e-01 1.38614118e+00 3.45103323e-01 -3.63407761e-01 6.37844577e-02 9.62756872e-01 -3.31985712e-01 3.48914653e-01 -6.49684012e-01 -3.19463104e-01 1.22657523e-01 1.19636810e+00 -6.10636413e-01 -3.39533210e-01 -7.22679913e-01 1.44176269e+00 3.80282074e-01 6.19777977e-01 -8.50358546e-01 1.95562001e-02 3.19183409e-01 -1.38796493e-02 3.37805122e-01 -4.27050471e-01 -5.43255627e-01 -1.32698309e+00 2.58367896e-01 -1.07273507e+00 1.34930713e-02 -9.98028159e-01 -9.73049879e-01 7.59438396e-01 -2.21925586e-01 -1.40368593e+00 -2.24644631e-01 -6.33764863e-01 -3.57924879e-01 1.12721598e+00 -1.59338403e+00 -1.52631032e+00 6.30108640e-02 3.67304534e-01 8.27633619e-01 -1.67875901e-01 8.09036076e-01 2.78369755e-01 -6.13557100e-01 8.51742089e-01 7.59736542e-03 6.24461249e-02 1.00780189e+00 -1.06975400e+00 4.27733302e-01 9.98790383e-01 5.09234369e-01 6.64673567e-01 6.90576375e-01 -5.70859134e-01 -1.83381414e+00 -8.86987686e-01 1.28136015e+00 -6.94840550e-01 6.19894028e-01 -5.38076282e-01 -7.00715303e-01 5.81841826e-01 1.05770576e+00 -3.26276779e-01 4.40798789e-01 -1.19982675e-01 -4.81419116e-01 1.24302683e-02 -8.24930489e-01 8.24989974e-01 8.22488546e-01 -5.72254062e-01 -5.42364836e-01 2.81096637e-01 6.69209123e-01 -5.40046990e-01 -6.49689317e-01 2.85441846e-01 4.29624081e-01 -5.75498164e-01 7.78697014e-01 -4.38182354e-01 9.57199931e-01 -4.51374799e-01 -2.05775425e-01 -1.37218523e+00 -1.39513060e-01 -6.13526940e-01 -7.50557184e-02 1.11655486e+00 9.08131599e-01 -2.26276517e-01 3.39977056e-01 2.29516253e-01 -2.70353049e-01 -9.07201409e-01 -9.52528358e-01 -5.02969742e-01 7.60944188e-02 -1.81226432e-01 1.99836940e-01 9.13894117e-01 2.40869969e-01 8.56707752e-01 -7.01538742e-01 1.18091544e-02 3.47583890e-01 5.02829254e-01 7.37352252e-01 -5.43869138e-01 -3.05232376e-01 -6.03623688e-01 -1.38246030e-01 -1.34638476e+00 -9.38065425e-02 -1.12310779e+00 -5.85862920e-02 -1.97172058e+00 6.97515249e-01 1.28369838e-01 -3.36003751e-02 6.39888763e-01 -3.42749298e-01 6.59755111e-01 3.65861028e-01 4.21892852e-01 -7.00377226e-01 7.89284170e-01 1.57023573e+00 -4.29413974e-01 4.17437963e-02 -3.13805133e-01 -6.98070824e-01 1.71310157e-01 5.87982118e-01 -2.35662282e-01 -2.40124643e-01 -9.27408516e-01 2.29390666e-01 1.82884797e-01 3.45438927e-01 -4.60438341e-01 7.64544532e-02 -1.17401645e-01 4.16876346e-01 -5.42290986e-01 3.87197614e-01 -7.69409537e-01 -1.96241334e-01 1.52749717e-01 -3.70384991e-01 6.21708512e-01 3.03980231e-01 4.71833825e-01 -3.43245953e-01 2.01719522e-01 5.34044385e-01 -1.84139777e-02 -2.39625096e-01 3.77204269e-02 -4.04985785e-01 -1.22446388e-01 6.31729186e-01 -1.19419448e-01 -4.66967195e-01 -4.88083988e-01 -3.89208734e-01 3.76669049e-01 6.43418252e-01 6.41060829e-01 6.05633736e-01 -1.63049304e+00 -8.51617038e-01 -2.94393003e-01 3.50640953e-01 -5.78419685e-01 -2.54606009e-01 1.53401887e+00 -1.15731120e-01 6.75051689e-01 -1.56482980e-01 -9.70604062e-01 -1.46890485e+00 6.30542397e-01 1.53755024e-01 -3.11943620e-01 -5.24101317e-01 4.38735366e-01 2.39935338e-01 -2.96589583e-01 1.30262524e-01 -1.52365893e-01 -1.09592319e-01 -1.35441095e-01 4.52349901e-01 1.04438260e-01 2.13148132e-01 -9.37270105e-01 -4.11489129e-01 7.47952461e-01 -8.44629779e-02 -6.79071128e-01 1.02309191e+00 -5.26644409e-01 -1.52676672e-01 5.32815695e-01 9.08478379e-01 2.88860928e-02 -9.88189638e-01 -4.45170373e-01 -5.71580045e-02 -5.40360153e-01 -1.66592270e-01 -1.29937720e+00 -9.29020941e-01 1.33992517e+00 5.44705749e-01 -2.55429268e-01 1.14687455e+00 1.40801698e-01 7.08908021e-01 1.24130256e-01 3.84993881e-01 -8.25902164e-01 2.30210423e-01 5.03313005e-01 1.11655104e+00 -1.55264509e+00 -2.35014126e-01 -2.15095490e-01 -9.32712913e-01 9.03052330e-01 4.00248200e-01 4.58374292e-01 -2.62833953e-01 -9.05736408e-04 4.67662781e-01 3.98354307e-02 -8.88357162e-01 -1.55217886e-01 8.54696810e-01 2.89773405e-01 6.91318035e-01 -8.31813142e-02 -4.28202331e-01 2.38500550e-01 1.89031273e-01 -2.72471458e-01 1.82837099e-01 8.37993741e-01 -2.68803090e-01 -1.33038139e+00 -5.36395729e-01 -1.37684392e-02 -4.23028886e-01 -4.12292540e-01 -8.79773378e-01 7.25639105e-01 -3.99011582e-01 1.14249933e+00 -2.86819488e-01 -4.46763426e-01 2.49823064e-01 2.56769210e-01 8.40725601e-01 -4.78772879e-01 -6.75489545e-01 4.28760439e-01 1.48220167e-01 -5.82894027e-01 -6.20168328e-01 -5.87458968e-01 -9.90995586e-01 -1.27111256e-01 -2.66082555e-01 -1.18439030e-02 8.77469361e-01 1.05703890e+00 4.39929456e-01 3.62519890e-01 5.46675801e-01 -9.26278949e-01 -2.57816285e-01 -1.16220129e+00 2.36942738e-01 2.61135042e-01 3.60771626e-01 -5.50448179e-01 -1.23948604e-01 1.68574661e-01]

[11.439160346984863, 1.4904249906539917]

442058fe-911d-4d64-9083-adce28209e8b

face-recognition-accuracy-across-demographics

2206.01881

https://arxiv.org/abs/2206.01881v2

https://arxiv.org/pdf/2206.01881v2.pdf

Face Recognition Accuracy Across Demographics: Shining a Light Into the Problem

We explore varying face recognition accuracy across demographic groups as a phenomenon partly caused by differences in face illumination. We observe that for a common operational scenario with controlled image acquisition, there is a large difference in face region brightness between African-American and Caucasian, and also a smaller difference between male and female. We show that impostor image pairs with both faces under-exposed, or both overexposed, have an increased false match rate (FMR). Conversely, image pairs with strongly different face brightness have a decreased similarity measure. We propose a brightness information metric to measure variation in brightness in the face and show that face brightness that is too low or too high has reduced information in the face region, providing a cause for the lower accuracy. Based on this, for operational scenarios with controlled image acquisition, illumination should be adjusted for each individual to obtain appropriate face image brightness. This is the first work that we are aware of to explore how the level of brightness of the skin region in a pair of face images (rather than a single image) impacts face recognition accuracy, and to evaluate this as a systematic factor causing unequal accuracy across demographics. The code is at https://github.com/HaiyuWu/FaceBrightness.

['Kevin W. Bowyer', 'Michael C. King', 'K. S. Krishnapriya', 'Vítor Albiero', 'Haiyu Wu']

2022-06-04

['unsupervised-face-recognition']

['computer-vision']

[ 4.04324055e-01 -1.05530895e-01 8.90904814e-02 -7.05559373e-01 -2.49120682e-01 -4.27164614e-01 3.77804786e-01 -4.21275795e-01 -2.34127954e-01 3.97606403e-01 3.94943804e-02 -1.77852511e-01 6.73368797e-02 -7.19937384e-01 -4.94833082e-01 -7.69736648e-01 3.25863987e-01 -1.35355294e-01 -2.72639662e-01 6.50023073e-02 3.13009828e-01 6.38093770e-01 -1.96463966e+00 1.82577178e-01 5.56614935e-01 8.82952154e-01 -2.55991995e-01 5.54892421e-01 -2.98070759e-02 1.56669497e-01 -7.93710768e-01 -5.89954317e-01 6.12197280e-01 -6.62313819e-01 -5.18384874e-01 2.27811664e-01 1.19441581e+00 -6.18435383e-01 -3.34842526e-03 1.11042941e+00 8.60263646e-01 -1.81751758e-01 5.43756664e-01 -1.28074992e+00 -6.71540737e-01 1.82364300e-01 -9.73007143e-01 9.07316208e-02 5.91541111e-01 3.45497370e-01 2.15488985e-01 -7.81517982e-01 3.23878407e-01 1.56876194e+00 5.34732401e-01 9.10833597e-01 -1.36857009e+00 -1.06590199e+00 -1.55057892e-01 4.08364944e-02 -1.78403211e+00 -1.01860261e+00 6.03649914e-01 -1.79646626e-01 4.28486764e-01 4.45565283e-01 6.30855858e-01 9.57891762e-01 2.09514587e-03 -1.43689558e-01 1.55785859e+00 -7.06712067e-01 -6.41641617e-02 4.57807034e-01 -1.09868929e-01 5.88177979e-01 4.26389933e-01 1.34375811e-01 -4.72800642e-01 -1.98212117e-01 8.60923529e-01 -8.05641711e-02 -3.35232973e-01 4.94012646e-02 -5.32112002e-01 4.76898968e-01 7.09638745e-02 3.69639188e-01 -1.33846089e-01 -2.00952157e-01 3.96377742e-02 5.37465155e-01 1.37353465e-01 3.10851067e-01 -1.67507321e-01 -9.92345512e-02 -7.48326123e-01 -1.16807394e-01 9.69146848e-01 4.63621438e-01 8.56051803e-01 -8.30551758e-02 -9.59657505e-02 1.08853614e+00 2.01894343e-01 6.12227201e-01 3.36646199e-01 -9.42396402e-01 6.23305254e-02 4.92977440e-01 1.06835134e-01 -1.15724301e+00 -1.32669643e-01 1.99304625e-01 -4.04018462e-01 6.56156659e-01 9.76169884e-01 -2.27922529e-01 -9.45024371e-01 1.94584167e+00 3.05328816e-01 -2.93582771e-02 -1.40721485e-01 9.29679036e-01 7.04414904e-01 2.11250722e-01 1.10281996e-01 -3.64457220e-01 1.80370724e+00 -2.27670729e-01 -6.42287850e-01 -3.46376538e-01 2.71733612e-01 -1.07612753e+00 1.30479133e+00 1.89262748e-01 -1.01192021e+00 -5.90917289e-01 -1.08352876e+00 3.13713729e-01 -3.72209549e-01 1.32814899e-01 1.58456713e-01 1.57414365e+00 -1.21067393e+00 3.98923904e-01 -3.00035894e-01 -7.98769116e-01 3.75936240e-01 3.58708203e-01 -7.23971188e-01 -1.86768696e-01 -9.86072123e-01 1.03636515e+00 -2.23425686e-01 2.16108322e-01 -5.09197950e-01 -6.83823943e-01 -6.60610497e-01 -1.39148086e-01 -4.84654568e-02 -2.52071649e-01 7.37232327e-01 -1.76529098e+00 -1.40667975e+00 1.21754515e+00 -3.12096864e-01 5.32147773e-02 4.02037531e-01 3.16881053e-02 -7.56314278e-01 9.07919705e-02 -2.08723009e-01 6.94746852e-01 1.11269701e+00 -1.48012066e+00 4.58792113e-02 -9.22451913e-01 -2.81940758e-01 2.24274471e-01 -4.22899693e-01 5.35625041e-01 -3.25674005e-02 -2.19713226e-01 -1.89244583e-01 -7.73100495e-01 3.64419132e-01 1.21639170e-01 5.83207645e-02 -9.34008323e-03 9.67316210e-01 -8.04211080e-01 1.11020708e+00 -2.33255029e+00 -5.81522405e-01 4.99274909e-01 -8.97904113e-02 3.97977829e-01 -2.56944239e-01 2.00302064e-01 -2.81070918e-01 4.64607596e-01 -1.01671871e-02 -8.91300756e-03 -2.51204967e-01 1.16037793e-01 1.22626849e-01 7.88555205e-01 3.39110255e-01 2.66485661e-01 -4.61530060e-01 -5.09869039e-01 7.82726146e-03 7.94815302e-01 -2.54049152e-01 2.01743111e-01 6.06711507e-01 3.14192086e-01 3.23224999e-02 9.29611206e-01 1.43393552e+00 3.79686981e-01 2.35320345e-01 -1.70522973e-01 -1.86152667e-01 -2.84091681e-01 -1.14116216e+00 7.59343803e-01 -3.13082546e-01 5.57991564e-01 2.45153919e-01 -4.98253137e-01 1.18103385e+00 3.38843673e-01 7.01158792e-02 -8.67141247e-01 2.50542521e-01 2.71805108e-01 3.70932370e-01 -3.98320168e-01 2.81086504e-01 -2.82054961e-01 5.76715529e-01 3.37021738e-01 -3.69279355e-01 -1.23929188e-01 -1.04921639e-01 -1.76097199e-01 6.82596505e-01 -3.69755208e-01 9.39223394e-02 -3.99365842e-01 5.66202462e-01 -8.00480306e-01 3.80393863e-01 6.72145724e-01 -7.12884843e-01 7.94125259e-01 5.94973385e-01 -1.86048180e-01 -9.86780584e-01 -1.07499695e+00 -6.10410988e-01 7.84306288e-01 1.60916522e-01 -5.35123609e-02 -1.09272993e+00 -3.76679748e-01 9.13944319e-02 4.00201559e-01 -7.05332696e-01 -2.69092649e-01 -2.98802406e-01 -6.31063402e-01 6.63041532e-01 2.33500391e-01 6.78984702e-01 -9.01968122e-01 -6.01674914e-01 -6.32478595e-01 7.28584006e-02 -7.23356962e-01 -4.65617865e-01 -3.94250304e-01 -6.77358091e-01 -1.10485351e+00 -7.85248399e-01 -6.76061451e-01 1.24542117e+00 1.93644017e-01 1.13076389e+00 6.56900167e-01 -5.69450021e-01 5.82451820e-01 -3.78784180e-01 -3.76742780e-01 -4.34415162e-01 -4.30301696e-01 2.41677254e-01 3.45584869e-01 5.86022854e-01 -1.18863501e-01 -9.06650722e-01 7.91996896e-01 -7.90903747e-01 -3.93413186e-01 3.33226383e-01 5.73102832e-01 9.84219238e-02 5.89999696e-03 4.70819682e-01 -6.10007524e-01 4.69663829e-01 -3.32416832e-01 -3.83594751e-01 3.64537269e-01 -4.80413377e-01 -3.76511246e-01 2.81568319e-01 -4.63169128e-01 -1.34731030e+00 -1.00977477e-02 4.97566089e-02 -1.53484836e-01 -4.95678842e-01 -2.71688312e-01 -2.65869617e-01 -4.77351159e-01 8.03776145e-01 -2.35371873e-01 5.88299572e-01 -2.19507515e-01 -1.46912441e-01 1.00048208e+00 2.50275165e-01 -4.17351007e-01 6.94161177e-01 6.22377455e-01 -1.49408758e-01 -1.09253669e+00 -1.55548975e-01 -1.83741376e-01 -5.50444305e-01 -6.06658697e-01 6.78464711e-01 -6.56626523e-01 -7.82532573e-01 7.19576299e-01 -6.73476100e-01 -3.08459103e-01 2.54877478e-01 4.55019504e-01 -9.20524728e-03 4.08431113e-01 -1.92567900e-01 -1.10262012e+00 -1.74188063e-01 -1.11223507e+00 8.78369212e-01 6.93506658e-01 -3.75255287e-01 -7.21100032e-01 -2.80384153e-01 4.33342516e-01 4.95489448e-01 7.68038854e-02 5.39055586e-01 -2.39967793e-01 1.30795792e-01 -1.40116170e-01 -4.20765996e-01 4.69819158e-01 6.26919091e-01 5.58350146e-01 -1.22859824e+00 -4.85745609e-01 7.27225468e-02 -3.35716248e-01 4.83287990e-01 3.04361790e-01 1.07321286e+00 -2.30054349e-01 -5.48607744e-02 3.60042214e-01 1.35936284e+00 2.48450294e-01 1.18531251e+00 1.88763458e-02 2.64015913e-01 9.74297404e-01 3.90837282e-01 3.27337831e-01 -9.22190472e-02 7.31460989e-01 2.29383111e-01 -3.80206764e-01 -1.57546476e-01 -4.41355631e-02 6.92741394e-01 2.72236275e-03 -1.33927569e-01 2.87025347e-02 -7.56188750e-01 2.81458646e-01 -8.78552854e-01 -1.06198907e+00 -2.13531181e-02 2.73083448e+00 7.65198350e-01 -5.80227137e-01 3.48530799e-01 1.32553698e-02 1.19969106e+00 -2.03334063e-01 -2.35412464e-01 -8.06017458e-01 -6.97676465e-02 1.91627055e-01 4.61968154e-01 5.28660715e-01 -6.60421729e-01 5.33453465e-01 7.14086294e+00 5.79630256e-01 -1.21170461e+00 -2.35048369e-01 1.16600621e+00 -2.22214445e-01 -1.53123438e-01 -1.55559937e-02 -6.60533011e-01 6.81808591e-01 9.97293115e-01 -9.56797972e-03 4.73961025e-01 5.15826881e-01 2.94073731e-01 -4.53792065e-01 -7.60507524e-01 1.05556333e+00 3.82428527e-01 -3.49521220e-01 -2.80098654e-02 2.57976413e-01 4.11314249e-01 -6.38622940e-01 4.61679131e-01 -2.18762904e-01 -2.17915848e-01 -1.25627077e+00 4.58786011e-01 4.23837274e-01 9.61592913e-01 -7.47525334e-01 6.96780086e-01 -3.67836177e-01 -9.99625802e-01 -5.18904775e-02 -3.94050151e-01 -1.80992916e-01 -2.89865553e-01 3.93201083e-01 -6.66732728e-01 -6.82444945e-02 9.02992249e-01 -5.24150208e-02 -8.43391180e-01 6.76981509e-01 6.30812943e-02 4.72726971e-01 -2.80454367e-01 2.82247812e-01 -4.72631007e-01 -2.67574012e-01 2.40998462e-01 9.05291438e-01 4.46881950e-01 1.52109265e-01 -3.75021189e-01 8.17105055e-01 -1.69853214e-02 2.52546370e-01 -6.84707820e-01 1.12658314e-01 7.73304522e-01 1.39076650e+00 -7.42023230e-01 -2.83783853e-01 -6.83384776e-01 9.82365131e-01 -2.20469255e-02 3.71215284e-01 -7.60840476e-01 -1.93435490e-01 8.62010419e-01 2.93543994e-01 -7.46245533e-02 2.62107372e-01 -3.73205274e-01 -5.98678708e-01 1.03292100e-01 -1.04572821e+00 4.24829572e-01 -6.50079608e-01 -1.23189294e+00 3.08561534e-01 4.44913469e-03 -8.81328583e-01 1.73341811e-01 -5.27152777e-01 -7.86283851e-01 1.17097187e+00 -1.02264810e+00 -7.93046355e-01 -6.47194624e-01 6.16096020e-01 9.13945287e-02 4.24000667e-03 6.56445146e-01 3.83014143e-01 -8.42660487e-01 1.13972163e+00 -2.90259510e-01 1.06861152e-01 1.04369402e+00 -8.15224409e-01 1.55397996e-01 6.88984156e-01 -3.50220561e-01 9.65793908e-01 5.53058147e-01 -4.21331793e-01 -1.49664867e+00 -5.09958863e-01 5.47184885e-01 -6.06490374e-01 -4.88154814e-02 -2.78236330e-01 -9.33824956e-01 1.82892710e-01 3.04036826e-01 -1.84337795e-01 9.43573594e-01 7.28455186e-02 -5.03050268e-01 -3.15771967e-01 -1.85927868e+00 7.43343234e-01 8.49746287e-01 -5.55853307e-01 -5.83016910e-02 -8.18779394e-02 -7.93215036e-02 6.72353730e-02 -1.17381799e+00 3.90148491e-01 1.09489834e+00 -1.36227500e+00 9.06433344e-01 2.69254576e-02 1.77707717e-01 -2.82458037e-01 3.91759463e-02 -1.01142192e+00 -1.08783171e-01 -3.99141520e-01 5.76980233e-01 1.59903133e+00 4.03602690e-01 -1.10057557e+00 5.73554218e-01 1.08504570e+00 5.18937647e-01 -4.93908435e-01 -8.62362087e-01 -8.08989167e-01 8.42662305e-02 1.78214908e-01 6.87718809e-01 7.10934103e-01 4.43482324e-02 -3.86371315e-01 -1.86707452e-01 2.21083537e-01 3.64862919e-01 -3.80218178e-01 8.24674129e-01 -8.46414685e-01 1.24184802e-01 -4.49446172e-01 -5.57599843e-01 -1.61891878e-01 1.36784948e-02 -3.58583182e-01 -8.17682520e-02 -8.28907669e-01 4.14917350e-01 -2.73620307e-01 -3.59841101e-02 5.13575375e-01 -2.56196260e-01 8.41550052e-01 1.82663202e-01 3.34297679e-02 2.66114533e-01 -4.28776778e-02 1.05831790e+00 1.85998648e-01 -1.00681663e-01 -3.73701215e-01 -9.72165763e-01 5.28399825e-01 9.86342251e-01 -1.04262687e-01 -1.57021552e-01 -2.71230370e-01 -1.80600092e-01 -3.46282303e-01 4.69484061e-01 -1.08672488e+00 -2.02351332e-01 -1.52144969e-01 9.24170911e-01 1.77448094e-01 2.73015469e-01 -9.05814409e-01 2.55578578e-01 5.60736477e-01 8.27276558e-02 1.27280653e-01 4.18813974e-01 -4.12412658e-02 9.52818915e-02 -2.50639796e-01 1.28724480e+00 -1.35551721e-01 -5.37921965e-01 5.52953966e-02 -4.64514107e-01 -3.12930405e-01 9.50685084e-01 -8.01080763e-01 -5.05845070e-01 -3.38905394e-01 -2.53062308e-01 -2.42407024e-01 1.11448407e+00 4.58951175e-01 5.23195744e-01 -1.26020908e+00 -7.16314256e-01 7.51457453e-01 9.50522497e-02 -9.11725461e-01 4.86258477e-01 9.74775910e-01 -5.16714811e-01 -2.27650836e-01 -6.08138561e-01 -3.78494233e-01 -1.92397070e+00 1.96685374e-01 6.04336858e-01 5.08253455e-01 4.27357368e-02 9.14641201e-01 2.53208727e-01 -1.92643091e-01 -7.22340867e-02 2.06842050e-01 -1.93617530e-02 8.04975927e-02 8.23867381e-01 6.02208078e-01 8.80120173e-02 -9.76879418e-01 -4.99633968e-01 8.78019273e-01 -1.47990361e-01 -8.08072686e-02 4.46908921e-01 -2.33502850e-01 -3.34868014e-01 2.55285446e-02 1.24829006e+00 2.29627445e-01 -1.13080490e+00 4.02007848e-01 -5.93367100e-01 -1.43196249e+00 -1.52649179e-01 -7.34142363e-01 -1.17159402e+00 5.28314829e-01 1.31353211e+00 2.92116553e-01 1.40866911e+00 -8.67763460e-02 2.09596559e-01 -1.20975718e-01 2.29807869e-01 -1.18722904e+00 3.63422767e-03 2.81325635e-02 1.01617301e+00 -1.37288380e+00 1.21884033e-01 -4.99716491e-01 -6.86168790e-01 9.62797105e-01 9.32539463e-01 2.24716261e-01 4.74983871e-01 2.86623925e-01 4.59100276e-01 -1.62012517e-01 -2.59486109e-01 -1.68114394e-01 3.62126566e-02 8.88015032e-01 7.66729712e-01 3.26681808e-02 -4.49283361e-01 3.49331051e-02 -3.82027954e-01 -1.67434275e-01 3.46350759e-01 7.79084325e-01 -2.76543200e-01 -1.17725945e+00 -9.45321858e-01 6.49458110e-01 -5.39746761e-01 2.36650273e-01 -9.29733455e-01 6.71867013e-01 3.23258489e-01 1.22424912e+00 5.67105114e-01 -3.33905607e-01 1.41678557e-01 2.52658546e-01 7.73736119e-01 -3.54295403e-01 -5.25883853e-01 -3.35325807e-01 6.81681260e-02 -4.56003457e-01 -4.51523870e-01 -7.14027882e-01 -8.93533647e-01 -8.53810489e-01 -4.33047116e-01 -8.91243294e-02 6.41438544e-01 4.50329483e-01 3.71781796e-01 -2.07873181e-01 8.13850641e-01 -6.94947481e-01 -7.83473253e-02 -9.83034790e-01 -8.03739011e-01 6.87895894e-01 1.96383938e-01 -7.73390114e-01 -7.56777823e-01 -6.13645511e-03]

[13.035174369812012, 1.1558433771133423]

66f65283-82e5-4966-85d3-8a92770172c7

always-valid-risk-monitoring-for-online

2211.10363

https://arxiv.org/abs/2211.10363v1

https://arxiv.org/pdf/2211.10363v1.pdf

Always Valid Risk Monitoring for Online Matrix Completion

Always-valid concentration inequalities are increasingly used as performance measures for online statistical learning, notably in the learning of generative models and supervised learning. Such inequality advances the online learning algorithms design by allowing random, adaptively chosen sample sizes instead of a fixed pre-specified size in offline statistical learning. However, establishing such an always-valid type result for the task of matrix completion is challenging and far from understood in the literature. Due to the importance of such type of result, this work establishes and devises the always-valid risk bound process for online matrix completion problems. Such theoretical advances are made possible by a novel combination of non-asymptotic martingale concentration and regularized low-rank matrix regression. Our result enables a more sample-efficient online algorithm design and serves as a foundation to evaluate online experiment policies on the task of online matrix completion.

['Wenjie Li', 'Chi-Hua Wang']

2022-11-18

['matrix-completion']

['methodology']

[ 3.23832817e-02 1.17384037e-02 -1.94401935e-01 -1.60957679e-01 -1.02647674e+00 -5.15125096e-01 2.32827470e-01 2.70347953e-01 -4.80815589e-01 7.90132582e-01 -1.49252981e-01 -5.53171754e-01 -5.79841793e-01 -5.19775927e-01 -1.25189805e+00 -9.54969227e-01 -5.26567280e-01 5.31456232e-01 -1.50188535e-01 2.13556319e-01 4.27794047e-02 4.24640894e-01 -1.44775045e+00 -3.31065744e-01 7.75431633e-01 8.71311903e-01 -1.44460285e-02 7.02098668e-01 1.65366307e-01 3.77855301e-01 1.15050912e-01 -5.27010322e-01 6.86321616e-01 -5.70245624e-01 -3.72614861e-01 1.32065117e-01 5.34149468e-01 -1.93683058e-01 -1.44936189e-01 1.28002334e+00 5.53348958e-01 2.38623470e-01 6.67901933e-01 -1.12024963e+00 -3.66325378e-01 9.58598316e-01 -6.47848308e-01 2.55032301e-01 1.07459471e-01 -1.97144479e-01 1.31727862e+00 -8.91816258e-01 3.51966262e-01 8.46108079e-01 4.71653461e-01 2.91043401e-01 -1.28280044e+00 -8.96605611e-01 9.78436600e-03 9.18653756e-02 -1.44894791e+00 -4.25201178e-01 4.90590841e-01 -5.26931584e-01 1.87216501e-03 3.01009208e-01 5.20476460e-01 1.00497746e+00 1.19034126e-01 9.32874680e-01 1.17439222e+00 -6.36813402e-01 5.53489864e-01 2.45900348e-01 1.06279314e-01 7.53441691e-01 6.31260395e-01 1.69255063e-01 -6.17591083e-01 -2.44667932e-01 9.59760666e-01 1.41103705e-02 -1.47264540e-01 -7.08195984e-01 -9.33306277e-01 9.22847867e-01 -7.13066831e-02 3.44040602e-01 -3.60661119e-01 2.14697167e-01 2.83011943e-01 5.52086353e-01 6.43978119e-01 1.70719922e-01 -3.91570270e-01 -3.15505803e-01 -1.00272381e+00 2.11381406e-01 1.03540230e+00 9.87907648e-01 6.35264456e-01 2.33821258e-01 -3.27839643e-01 4.49310303e-01 1.58496410e-01 7.76258409e-01 -1.66616648e-01 -7.78191030e-01 5.79531431e-01 -5.76152988e-02 3.50526333e-01 -8.26497257e-01 -1.62080884e-01 -9.82009709e-01 -8.77888501e-01 -1.22135125e-01 1.05326819e+00 -1.25196576e-01 -4.79658581e-02 1.87669420e+00 5.30024350e-01 4.51332599e-01 -4.19779927e-01 8.17088127e-01 -3.67757708e-01 4.56247509e-01 -1.35560274e-01 -5.52886546e-01 8.56274307e-01 -3.60277027e-01 -5.95060110e-01 2.79474050e-01 8.66979301e-01 -2.78695554e-01 1.20720112e+00 6.50609672e-01 -9.50896680e-01 -7.65099227e-02 -8.60079408e-01 2.33487055e-01 5.81874102e-02 -6.08069301e-02 9.90571320e-01 1.06621611e+00 -8.65754545e-01 8.26505959e-01 -7.76147783e-01 -1.56607971e-01 4.76758778e-01 2.54246414e-01 -2.70826757e-01 -1.77567720e-01 -8.33065033e-01 2.37731978e-01 -1.72195546e-02 2.53442973e-01 -9.39051330e-01 -1.11547923e+00 -5.99188685e-01 1.72339320e-01 6.77243769e-01 -6.97069705e-01 9.86059487e-01 -8.85167241e-01 -1.35865808e+00 7.10367799e-01 2.53264024e-03 -8.65407705e-01 9.12457883e-01 -2.80809164e-01 1.77948624e-01 5.66667551e-03 -1.72247380e-01 -2.62293577e-01 1.20384014e+00 -9.09372926e-01 -4.18269306e-01 -6.34500384e-01 5.50878979e-02 -2.62933765e-02 -4.78734255e-01 -2.51278311e-01 -7.86795840e-02 -3.09575051e-01 -2.33834669e-01 -9.32483375e-01 -5.07780373e-01 -4.90731522e-02 -1.96868420e-01 -1.16205037e-01 1.74902938e-02 -6.18619502e-01 1.16234624e+00 -2.09244514e+00 4.76786718e-02 3.77953470e-01 1.34491444e-01 -3.00701082e-01 -9.03391391e-02 5.50828457e-01 4.40671854e-02 1.20977730e-01 -1.12328231e-01 -4.92441833e-01 3.16949099e-01 -1.93811119e-01 -5.48728108e-01 9.73273993e-01 -3.00472200e-01 5.33576250e-01 -9.43379402e-01 -3.09431732e-01 1.46204576e-01 5.38871922e-02 -8.05999398e-01 -1.80688798e-02 -3.82364169e-02 3.82146716e-01 -3.89527261e-01 1.89010203e-01 5.67447186e-01 -3.08573782e-01 1.55673906e-01 2.54058540e-01 1.29061684e-01 -2.06643119e-01 -1.38490975e+00 1.39856541e+00 -6.85529470e-01 4.70135301e-01 5.04752636e-01 -1.16102660e+00 3.95551831e-01 2.34345794e-01 5.79644859e-01 -7.78183118e-02 3.21774989e-01 1.94601387e-01 -1.16790533e-01 -3.04062843e-01 1.49938658e-01 -4.40465361e-01 -1.62337888e-02 4.54249740e-01 -1.01295382e-01 3.35307240e-01 3.88849109e-01 2.86310852e-01 1.11770046e+00 -1.30790398e-01 7.81944469e-02 -6.17169321e-01 4.19758916e-01 -5.82297683e-01 3.17935258e-01 1.31267321e+00 2.87851244e-02 2.45099783e-01 6.40976131e-01 1.41496256e-01 -1.16595399e+00 -1.16814947e+00 -2.72642225e-01 1.34794033e+00 -1.48532331e-01 -2.33726814e-01 -6.03038907e-01 -6.01480424e-01 3.57514143e-01 4.87915128e-01 -7.76328802e-01 3.13992091e-02 -8.16940367e-02 -6.09250426e-01 1.54221818e-01 2.89069861e-01 -1.51458441e-03 -3.92986178e-01 1.38216838e-01 -1.66259948e-02 2.49475241e-01 -1.09979928e+00 -6.41576290e-01 3.07763182e-02 -1.14185059e+00 -1.10307431e+00 -7.32225835e-01 -2.72975773e-01 7.37851501e-01 3.80003929e-01 8.25700760e-01 -1.02289908e-01 -2.54480124e-01 9.19520080e-01 -1.51277736e-01 -2.71164745e-01 -3.46664190e-01 9.89457741e-02 3.11680168e-01 5.63734353e-01 1.01955436e-01 -7.14933872e-01 -5.59186697e-01 2.05605924e-01 -1.04145730e+00 -1.99856997e-01 6.92684650e-01 7.13650763e-01 4.87281948e-01 6.08782582e-02 7.04968035e-01 -1.03519094e+00 6.36626422e-01 -7.83826411e-01 -1.30063760e+00 1.64014399e-01 -8.83605421e-01 3.93684536e-01 9.64567184e-01 -6.31175101e-01 -6.50443912e-01 -5.87941073e-02 3.37459087e-01 -4.30271536e-01 3.18415761e-01 6.49603248e-01 -1.16538078e-01 -7.85925686e-02 5.28513730e-01 3.61061156e-01 6.84786141e-02 -3.89386117e-01 4.34390545e-01 3.50118160e-01 1.57256037e-01 -8.52810860e-01 1.24949181e+00 5.60929060e-01 5.60615063e-01 -1.01457846e+00 -9.82278228e-01 -6.77966416e-01 -4.40397352e-01 -3.38510245e-01 3.68826747e-01 -8.60663235e-01 -1.03873730e+00 1.45862088e-01 -4.79688555e-01 -5.08432209e-01 -3.69150102e-01 7.36239135e-01 -7.41080701e-01 5.53229511e-01 -5.11819541e-01 -1.62929106e+00 9.98700634e-02 -4.48261201e-01 7.38436341e-01 -6.51719943e-02 2.75899768e-01 -1.24154449e+00 5.06593660e-02 5.02218664e-01 2.55514622e-01 -2.73603145e-02 5.16212940e-01 -5.37826002e-01 -9.30343926e-01 -3.77412796e-01 -1.51229456e-01 3.27960640e-01 -4.31965023e-01 -5.47450148e-02 -5.87194383e-01 -4.50379759e-01 7.89462551e-02 -2.25332156e-01 8.89842927e-01 5.65389335e-01 1.29093361e+00 -4.50445235e-01 1.05648919e-03 4.42962825e-01 1.45562088e+00 -4.15960461e-01 3.32700521e-01 -9.95567515e-02 4.26320761e-01 6.66709661e-01 4.97616708e-01 1.01229095e+00 1.29258156e-01 3.89981955e-01 3.01098935e-02 4.28680420e-01 4.38046396e-01 -5.07131338e-01 6.04353607e-01 8.19717705e-01 1.68526724e-01 1.62019670e-01 -4.61903960e-01 4.51917112e-01 -1.79603815e+00 -1.21429980e+00 -1.64889500e-01 2.93305087e+00 1.10754991e+00 -4.33354266e-02 3.53347361e-01 1.04302853e-01 7.86658108e-01 -2.58538336e-01 -4.95643198e-01 7.49226287e-03 -9.69197452e-02 2.78978437e-01 9.45572376e-01 6.39095962e-01 -9.39672887e-01 5.45779824e-01 5.53118706e+00 1.28584385e+00 -6.88889027e-01 2.89408982e-01 3.37887555e-01 -2.26106226e-01 -3.42564285e-01 2.15542927e-01 -7.99993396e-01 4.33483064e-01 1.06749952e+00 -3.26541871e-01 7.90121138e-01 1.12198603e+00 4.77498889e-01 -2.84369826e-01 -1.41740751e+00 1.28309739e+00 -1.31595403e-01 -1.11358356e+00 -4.50171024e-01 7.15335131e-01 8.31530869e-01 -4.54693675e-01 4.11792725e-01 4.76320356e-01 2.30223864e-01 -7.79344022e-01 5.10304451e-01 5.17886519e-01 5.58811545e-01 -8.81835222e-01 3.57230246e-01 5.05747318e-01 -9.53773379e-01 -2.79057443e-01 -4.28384721e-01 -1.14950135e-01 -1.42728835e-01 1.26264155e+00 -4.96505111e-01 3.97634745e-01 1.20837726e-01 4.87125099e-01 -4.56806242e-01 1.26902199e+00 1.07658720e-02 1.15048707e+00 -5.97309887e-01 -2.38684297e-01 -1.91478342e-01 -7.28613436e-01 5.84249318e-01 9.47505772e-01 1.58953309e-01 -2.74414480e-01 4.08344567e-02 7.22739756e-01 -3.06488395e-01 5.12086749e-01 -5.76079667e-01 -3.02003235e-01 4.74621475e-01 1.29217279e+00 -7.15785682e-01 1.52638406e-02 -2.63886631e-01 8.23048592e-01 3.53375226e-01 2.53532082e-01 -8.12062740e-01 -3.36856022e-02 4.03121322e-01 4.37280238e-01 4.21249032e-01 -6.28453910e-01 -3.72550577e-01 -1.32118428e+00 2.43334487e-01 -6.74552262e-01 4.04614955e-01 1.42636374e-01 -1.59732294e+00 -3.08224559e-01 1.03717610e-01 -9.33004141e-01 -2.38939032e-01 -5.86930513e-01 -4.64072585e-01 4.88907218e-01 -1.05225730e+00 -7.95782149e-01 1.53804630e-01 4.55367416e-01 1.41017660e-01 -3.63007262e-02 4.10606563e-01 3.85761201e-01 -7.36081719e-01 7.99538791e-01 8.07951808e-01 -1.26930445e-01 5.38680494e-01 -1.43058193e+00 -2.61031121e-01 1.02497065e+00 5.23649037e-01 5.77352047e-01 1.01052797e+00 -5.60773075e-01 -1.96841300e+00 -6.84753239e-01 5.13696253e-01 -5.32860398e-01 1.20649910e+00 -7.94420600e-01 -6.38932109e-01 3.17958176e-01 -4.04799074e-01 4.19398397e-02 7.55567014e-01 7.86925912e-01 -2.97663987e-01 -4.19113696e-01 -9.24773633e-01 6.38914049e-01 1.12201393e+00 -5.87027669e-01 2.03880847e-01 6.91182554e-01 4.05492485e-01 -2.81660911e-02 -9.10821319e-01 -4.12237644e-02 5.96538901e-01 -7.49404430e-01 8.09124947e-01 -7.40379035e-01 2.74373263e-01 1.75410092e-01 -1.47695959e-01 -6.90098941e-01 9.01313126e-02 -1.28775227e+00 -3.74966145e-01 1.06191623e+00 4.78863597e-01 -6.75947607e-01 9.85916972e-01 6.67395771e-01 3.30872744e-01 -7.44995952e-01 -1.03580439e+00 -1.28534019e+00 1.90361381e-01 -6.58958852e-01 -1.34679824e-01 7.21008599e-01 -9.51178744e-02 1.81429803e-01 -6.62629902e-01 1.25015885e-01 1.17664599e+00 -1.42107746e-02 1.11243582e+00 -1.19863009e+00 -8.89313102e-01 -4.58880901e-01 -2.31774792e-01 -1.20996904e+00 4.01172578e-01 -1.05125678e+00 -9.15384218e-02 -6.96345985e-01 3.90543908e-01 -6.47219718e-01 -1.76148832e-01 -1.93387568e-01 -1.00396827e-01 -1.49668336e-01 2.11617351e-01 -3.26075032e-02 -8.01642895e-01 7.85914779e-01 1.03093886e+00 3.16940069e-01 -1.97163187e-02 4.98798490e-01 -6.26419544e-01 2.39872694e-01 4.87987101e-01 -5.33116519e-01 -4.64112580e-01 2.72693425e-01 7.62746751e-01 3.16574901e-01 3.44821721e-01 -7.74004877e-01 1.43002257e-01 -1.24404080e-01 -1.60739962e-02 -3.60956818e-01 -7.56363347e-02 -7.82272637e-01 -1.28409103e-01 4.84366864e-01 -6.44866049e-01 -4.14300054e-01 -3.64706278e-01 1.27611828e+00 3.85053724e-01 -4.90591735e-01 6.43195629e-01 2.96055436e-01 1.20081142e-01 6.37242794e-01 -3.10078323e-01 4.79288757e-01 8.28746796e-01 -3.21990787e-03 4.14779663e-01 -7.60455608e-01 -7.61446238e-01 1.79874167e-01 8.76692683e-02 -5.83981387e-02 3.11814904e-01 -1.05016196e+00 -8.07535827e-01 -6.11185888e-03 -2.21757561e-01 -2.71466345e-01 4.79116857e-01 1.23199463e+00 -2.51561459e-02 3.42880338e-01 3.67997825e-01 -4.13021743e-01 -9.04897213e-01 5.84500730e-01 3.59116048e-02 -4.43466127e-01 -3.20346892e-01 7.11574197e-01 1.45078838e-01 -6.43241107e-02 5.13802886e-01 -7.72266388e-02 4.54038799e-01 -1.12034373e-01 4.85915363e-01 5.81665039e-01 -3.99858765e-02 1.40205041e-01 8.76790583e-02 -2.63459757e-02 -4.45234552e-02 -4.72264260e-01 1.13600516e+00 -3.84960592e-01 -1.74885709e-02 9.51222360e-01 1.12863612e+00 3.42293739e-01 -1.49761438e+00 -4.37057763e-01 2.99097598e-01 -5.74492395e-01 -9.68998447e-02 -1.54899061e-01 -7.65463412e-01 7.11487949e-01 4.75675493e-01 2.52414018e-01 7.97359228e-01 -1.60440076e-02 6.39321923e-01 4.97241646e-01 8.04568887e-01 -1.27185559e+00 1.19897425e-01 4.34277058e-01 7.30698526e-01 -1.21465635e+00 -2.78748497e-02 -2.46956185e-01 -1.26818940e-01 9.68913853e-01 1.16457835e-01 -3.73037815e-01 9.46864903e-01 1.52398288e-01 -6.35635078e-01 2.09918767e-01 -6.71455979e-01 -2.36059338e-01 3.87826324e-01 2.78643727e-01 3.78297329e-01 1.58539951e-01 -6.13741994e-01 7.57590413e-01 -2.62089431e-01 -7.46952593e-02 6.07360721e-01 5.64213395e-01 -3.76802802e-01 -1.11380529e+00 -1.83486268e-01 6.06635928e-01 -6.64366782e-01 -1.41653657e-01 -7.78980777e-02 6.38840735e-01 -6.63205862e-01 6.70544624e-01 -2.60564923e-01 3.02486867e-02 -2.66287327e-01 1.51169956e-01 9.83178318e-01 -5.19867301e-01 -2.76421070e-01 -9.41082835e-02 -1.89441651e-01 -3.45506549e-01 2.56603789e-02 -8.71514857e-01 -6.87357426e-01 -5.71363509e-01 -5.96457183e-01 3.36014420e-01 7.43699253e-01 1.10185421e+00 2.30861709e-01 -2.34686196e-01 1.09302402e+00 -5.30432820e-01 -1.48473132e+00 -8.53486359e-01 -1.01095021e+00 3.17562997e-01 -1.26318308e-02 -4.69054490e-01 -8.38613451e-01 -1.72600135e-01]

[6.833423614501953, 4.419346809387207]

d5ee8b6d-2a55-45a0-8589-c63d158186f7

elementwise-language-representation

2302.13475

https://arxiv.org/abs/2302.13475v1

https://arxiv.org/pdf/2302.13475v1.pdf

Elementwise Language Representation

We propose a new technique for computational language representation called elementwise embedding, in which a material (semantic unit) is abstracted into a horizontal concatenation of lower-dimensional element (character) embeddings. While elements are always characters, materials are arbitrary levels of semantic units so it generalizes to any type of tokenization. To focus only on the important letters, the $n^{th}$ spellings of each semantic unit are aligned in $n^{th}$ attention heads, then concatenated back into original forms creating unique embedding representations; they are jointly projected thereby determining own contextual importance. Technically, this framework is achieved by passing a sequence of materials, each consists of $v$ elements, to a transformer having $h=v$ attention heads. As a pure embedding technique, elementwise embedding replaces the $w$-dimensional embedding table of a transformer model with $256$ $c$-dimensional elements (each corresponding to one of UTF-8 bytes) where $c=w/v$. Using this novel approach, we show that the standard transformer architecture can be reused for all levels of language representations and be able to process much longer sequences at the same time-complexity without "any" architectural modification and additional overhead. BERT trained with elementwise embedding outperforms its subword equivalence (original implementation) in multilabel patent document classification exhibiting superior robustness to domain-specificity and data imbalance, despite using $0.005\%$ of embedding parameters. Experiments demonstrate the generalizability of the proposed method by successfully transferring these enhancements to differently architected transformers CANINE and ALBERT.

['Jeeeun Kim', 'Dunam Kim']

2023-02-27

['document-classification']

['natural-language-processing']

[ 5.48935056e-01 1.55322626e-01 -3.22799832e-01 -1.25175431e-01 -7.49271631e-01 -7.10057259e-01 4.68073159e-01 4.91991520e-01 -7.61074662e-01 5.97685635e-01 1.49014935e-01 -8.36515844e-01 -3.82811087e-03 -1.02897537e+00 -6.72934592e-01 -5.30193090e-01 -1.70480758e-01 2.80106634e-01 -2.92188842e-02 -2.98628062e-01 2.99835414e-01 6.11043394e-01 -1.55742097e+00 4.37496483e-01 5.25189102e-01 1.05497718e+00 1.30374759e-01 6.61239326e-01 -7.24195302e-01 5.83911598e-01 -7.34241068e-01 -6.32286668e-01 4.34675664e-01 1.11022003e-01 -9.31118011e-01 -3.06945550e-03 3.73781741e-01 -1.78390503e-01 -3.19402218e-01 8.77824306e-01 3.60624820e-01 7.79613853e-02 8.93209338e-01 -9.03392732e-01 -1.12979770e+00 6.98580623e-01 -5.74832618e-01 5.72455637e-02 3.52383584e-01 -2.96908617e-02 1.68526185e+00 -9.51214552e-01 4.82313126e-01 1.30379057e+00 5.46852827e-01 4.34436828e-01 -1.34136367e+00 -7.45618165e-01 2.28592068e-01 4.48297821e-02 -1.20336711e+00 -5.04743382e-02 5.34601092e-01 -4.75563705e-01 1.52541625e+00 3.23302239e-01 2.82064378e-01 9.04390514e-01 3.52635771e-01 6.38287723e-01 7.76405931e-01 -6.28053010e-01 7.39434883e-02 3.59058559e-01 5.23904145e-01 8.16867590e-01 4.80399460e-01 -3.08285266e-01 -1.19845159e-01 -3.50996464e-01 5.01603484e-01 2.81673670e-01 -5.52340485e-02 -3.61285359e-01 -1.06605637e+00 1.02579796e+00 4.21274513e-01 3.45324963e-01 -1.99335158e-01 4.56738055e-01 6.02853298e-01 3.75055134e-01 1.79335341e-01 3.73288035e-01 -4.96539861e-01 1.22873202e-01 -5.42775333e-01 1.53358057e-01 5.56313396e-01 1.26729262e+00 8.79940033e-01 2.37465858e-01 -2.27520674e-01 7.54441857e-01 6.01136051e-02 4.31826562e-01 7.55394697e-01 -4.98579592e-01 6.21901870e-01 8.78051460e-01 1.25595564e-02 -8.09415936e-01 -1.13060519e-01 -5.12627661e-01 -7.30357170e-01 1.04550514e-02 2.52781752e-02 2.09290430e-01 -8.93666029e-01 1.75256371e+00 -9.62370262e-02 -1.69588596e-01 4.56382781e-02 2.62965769e-01 3.25384378e-01 8.26966286e-01 4.06593263e-01 1.42281398e-01 2.16020894e+00 -7.81616449e-01 -4.48265940e-01 -1.74349844e-01 8.27014029e-01 -7.76184261e-01 1.23615718e+00 2.07848445e-01 -9.41923320e-01 -6.23654842e-01 -1.25883842e+00 -3.46970230e-01 -9.70849752e-01 8.94649774e-02 5.13388813e-01 1.07087231e+00 -9.74630237e-01 4.86209005e-01 -2.45197460e-01 -1.43917084e-01 2.82577336e-01 5.97911537e-01 -5.40249288e-01 1.51313758e-02 -1.44081855e+00 7.60808229e-01 5.21530688e-01 -2.77679622e-01 -5.82376778e-01 -6.63861632e-01 -9.99295831e-01 3.54573697e-01 5.89768663e-02 -5.48387468e-01 8.26838911e-01 -5.40585458e-01 -1.31738031e+00 7.82750845e-01 -3.93452317e-01 -5.58694661e-01 -3.33318487e-02 1.33630028e-02 -4.98109341e-01 8.84001888e-03 1.19608104e-01 5.45104325e-01 1.11658752e+00 -9.55019951e-01 -7.85638750e-01 -3.38588268e-01 2.78328896e-01 -1.23001002e-01 -8.71418953e-01 3.26212272e-02 -1.61561385e-01 -9.46869552e-01 -2.09471248e-02 -8.68346810e-01 -1.36938691e-01 -1.52956620e-01 -2.18058184e-01 -3.32368970e-01 6.01036608e-01 -6.04264438e-01 1.57420206e+00 -2.33032274e+00 1.09664798e-01 2.88605541e-01 3.97333413e-01 3.95752400e-01 -3.50338489e-01 6.70874715e-01 -4.14144397e-01 4.60984141e-01 -6.06486678e-01 -3.21926057e-01 3.52373093e-01 2.63839066e-01 -5.08015633e-01 3.52081031e-01 3.53348494e-01 9.31758881e-01 -7.79518902e-01 -1.94972843e-01 1.67743087e-01 5.92963219e-01 -7.46972084e-01 -8.97532180e-02 -1.06348194e-01 -4.59695011e-01 -4.61617500e-01 5.09760320e-01 6.49728417e-01 -1.80473313e-01 3.43294084e-01 -1.35898218e-01 -2.32545938e-02 4.17467773e-01 -1.20795846e+00 1.60368097e+00 -5.96188843e-01 3.41295987e-01 -1.57779261e-01 -1.25613153e+00 1.09994864e+00 2.88278580e-01 2.50121653e-01 -5.55324376e-01 9.21256617e-02 3.68882835e-01 -1.34500921e-01 -1.05503418e-01 9.86283839e-01 -3.42008799e-01 -6.29413605e-01 7.55034268e-01 1.23479247e-01 8.75958204e-02 1.84559330e-01 1.41327456e-01 1.17193663e+00 -2.29119152e-01 4.40526962e-01 -2.15470091e-01 8.78470540e-01 -3.17308545e-01 4.26751941e-01 7.07296073e-01 -6.42835274e-02 9.42701027e-02 7.12917686e-01 -3.44038367e-01 -1.36257756e+00 -1.15712857e+00 -1.46397352e-01 1.28303158e+00 -2.48571306e-01 -7.05795765e-01 -6.32025123e-01 -6.64807320e-01 2.95574963e-01 8.37452292e-01 -7.50457525e-01 -1.62789881e-01 -7.23645806e-01 -6.52170420e-01 8.64483595e-01 7.65192866e-01 6.36527687e-02 -1.01540816e+00 -6.55013144e-01 4.96139050e-01 2.97094762e-01 -8.04602861e-01 -5.49360394e-01 5.40815771e-01 -6.85041606e-01 -7.77572453e-01 -6.36403501e-01 -9.83322442e-01 6.77891731e-01 1.66848395e-02 8.18946123e-01 -2.37285644e-01 -4.45896566e-01 2.26032853e-01 -4.46679682e-01 -2.84919500e-01 -2.76280761e-01 2.04901516e-01 -1.14032277e-03 6.27962947e-02 6.09946191e-01 -4.59094971e-01 -6.01454794e-01 -7.63299987e-02 -1.25134659e+00 -4.95356262e-01 7.64877200e-01 1.15660882e+00 4.79180574e-01 -3.33964340e-02 3.19172680e-01 -1.04127264e+00 9.36865568e-01 -2.63029814e-01 -3.27240080e-01 1.70372084e-01 -7.83629775e-01 5.14059007e-01 1.10148799e+00 -2.45868534e-01 -4.74740356e-01 -3.40145618e-01 -8.60230252e-02 -3.55775684e-01 7.21232221e-02 3.72580737e-01 -5.33930026e-02 3.39214295e-01 4.89608824e-01 4.09193039e-01 8.95593762e-02 -5.63566208e-01 8.04248333e-01 8.70789886e-01 2.62323260e-01 -8.42621148e-01 7.64840007e-01 3.35211188e-01 -1.80147931e-01 -6.89141691e-01 1.05476258e-02 -4.05501038e-01 -5.01219690e-01 6.24368131e-01 8.07493031e-01 -6.71780825e-01 -7.52223074e-01 -1.97570585e-02 -1.22606587e+00 3.33375365e-01 -8.03876102e-01 2.92027503e-01 -3.69384676e-01 6.81111097e-01 -7.85301983e-01 -6.06493294e-01 -5.53808272e-01 -1.32439637e+00 1.05635631e+00 -3.59322727e-01 -4.31225330e-01 -7.93992758e-01 -1.18948892e-01 1.56902179e-01 3.22085470e-01 -1.72446638e-01 1.78710914e+00 -7.97466397e-01 -2.63339877e-01 -6.29046798e-01 -3.40746850e-01 7.08192170e-01 1.62259430e-01 -3.75677615e-01 -9.52731133e-01 -4.56558079e-01 -1.22205719e-01 -2.19015451e-03 8.67069006e-01 -1.01134449e-01 1.23790443e+00 -4.49308515e-01 -2.76641220e-01 3.35689425e-01 1.49830067e+00 4.47688997e-01 6.79805219e-01 4.77562279e-01 5.25806129e-01 5.11211812e-01 6.84221182e-03 4.85851228e-01 1.12949818e-01 3.95945102e-01 3.12718362e-01 3.02669276e-02 -4.24072566e-03 -1.86737701e-01 5.35504162e-01 8.16220701e-01 1.04277678e-01 -2.82121867e-01 -6.67599797e-01 6.04006350e-01 -1.22674990e+00 -1.05163479e+00 3.33753347e-01 2.33378959e+00 7.44286597e-01 1.99130088e-01 -1.54580384e-01 4.36207920e-01 5.35927653e-01 4.32152510e-01 -3.46200228e-01 -1.14607191e+00 -5.10973558e-02 1.06308174e+00 7.64251053e-01 4.78236616e-01 -7.54117608e-01 9.28816795e-01 5.94776201e+00 1.20540965e+00 -9.45686102e-01 1.56065017e-01 3.27602893e-01 7.87318859e-04 -8.62494588e-01 3.78559120e-02 -8.68394375e-01 4.65049028e-01 1.05648959e+00 -8.78799930e-02 1.78435519e-01 7.59909153e-01 -2.00818628e-01 4.42920715e-01 -1.26109481e+00 8.70498657e-01 6.16581775e-02 -1.31622517e+00 6.63847268e-01 1.95471704e-01 2.05951244e-01 -5.11890471e-01 4.86596942e-01 5.52404821e-01 1.05018452e-01 -1.25711155e+00 6.18883967e-01 -1.70273319e-01 1.15395164e+00 -9.10254836e-01 6.15127504e-01 -3.78815494e-02 -1.65633535e+00 -4.15282339e-01 -5.44062555e-01 1.13044754e-01 -4.48375456e-02 2.47989163e-01 -9.47266936e-01 7.50763774e-01 4.23669189e-01 4.01893556e-01 -1.86443284e-01 7.04777986e-02 8.23631063e-02 1.19206235e-01 -5.99591620e-02 -1.95705310e-01 5.75557470e-01 -1.84422791e-01 3.62655401e-01 1.67573273e+00 4.53644186e-01 -1.87664330e-02 -1.27803653e-01 5.96797168e-01 -3.40311378e-01 3.11619222e-01 -6.71749711e-01 -3.19863081e-01 6.23077452e-01 9.05782580e-01 -4.44820136e-01 -6.43506289e-01 -5.42598784e-01 1.07331753e+00 2.37765551e-01 2.63083786e-01 -7.81368494e-01 -9.65465665e-01 1.14297163e+00 2.59725302e-01 7.31212080e-01 -1.76738113e-01 -1.98933885e-01 -9.55548704e-01 1.30848259e-01 -8.99878919e-01 5.51700592e-01 -3.13551873e-01 -1.24456871e+00 8.59056830e-01 -2.53048897e-01 -1.36604249e+00 -1.50847554e-01 -1.14389586e+00 -1.94612384e-01 1.23808336e+00 -1.48065352e+00 -9.42973077e-01 3.94378185e-01 5.96309006e-01 5.65279007e-01 -3.94848704e-01 1.33827639e+00 2.39260793e-01 -4.86397833e-01 1.11161697e+00 2.73317814e-01 2.61289358e-01 5.05872846e-01 -1.21112132e+00 5.35064340e-01 6.57679379e-01 1.10199071e-01 1.13610387e+00 4.07630920e-01 -3.23476553e-01 -1.47990429e+00 -1.09591424e+00 1.25371385e+00 -3.49992514e-01 9.00189400e-01 -7.20645308e-01 -6.81125939e-01 8.92976463e-01 3.13847423e-01 -1.62110090e-01 1.07968020e+00 5.39913448e-03 -8.70837867e-01 -1.44001693e-01 -1.21166420e+00 6.37613416e-01 9.08960879e-01 -1.00695765e+00 -8.63876700e-01 5.62947728e-02 9.94035065e-01 7.30874762e-02 -1.12855482e+00 2.64280796e-01 5.98568797e-01 -5.29950082e-01 1.23429704e+00 -7.84041345e-01 2.08090842e-01 -2.62670875e-01 -4.96372730e-01 -8.96701992e-01 -5.91343284e-01 -3.48282993e-01 8.96893144e-02 8.57496679e-01 6.81218028e-01 -1.03511298e+00 7.45454788e-01 3.80517036e-01 -1.48997545e-01 -8.59576881e-01 -1.08333194e+00 -8.00765872e-01 3.55007857e-01 -5.79249680e-01 9.29503441e-01 9.62891757e-01 2.51197100e-01 3.54046851e-01 -2.86314398e-01 -9.59433988e-02 4.23442543e-01 -4.20255167e-03 3.84923488e-01 -9.77761209e-01 -4.58514482e-01 -6.17507100e-01 -6.43779874e-01 -1.09863317e+00 2.37758756e-01 -1.34138846e+00 -5.54894090e-01 -1.15608358e+00 -1.05646737e-02 -5.57280183e-01 -8.67031336e-01 6.29482687e-01 4.66826782e-02 1.08579755e-01 2.04467013e-01 4.79705930e-02 1.07878618e-01 3.90211731e-01 7.74052322e-01 -6.18236184e-01 1.04142480e-01 -5.56412578e-01 -9.08945978e-01 3.18493247e-01 6.10878229e-01 -3.78125072e-01 -5.05584359e-01 -5.82846999e-01 5.44895902e-02 -1.30159691e-01 -9.21304710e-03 -6.69279873e-01 -6.37209862e-02 1.96361229e-01 2.24904403e-01 -3.80362242e-01 4.17277187e-01 -9.62558746e-01 7.28094205e-02 6.76414847e-01 -4.40667480e-01 5.12560189e-01 2.98820674e-01 5.92921495e-01 -1.03868984e-01 -3.67199779e-01 4.55553412e-01 -1.83603048e-01 -7.24226296e-01 3.39969456e-01 -4.56247121e-01 -3.45957041e-01 1.02419972e+00 -5.48808515e-01 -1.89430177e-01 2.34855622e-01 -5.95900536e-01 -1.80247858e-01 3.28649372e-01 4.84956443e-01 6.19273961e-01 -1.23822522e+00 -5.62534332e-01 5.77543437e-01 2.25160912e-01 -2.28039071e-01 1.31883770e-01 3.20080310e-01 -5.63891292e-01 8.47362339e-01 -2.13005587e-01 -2.62493432e-01 -8.87767613e-01 8.54304552e-01 -5.63539229e-02 -6.21391416e-01 -5.76075196e-01 9.58454370e-01 3.15703064e-01 -3.65473419e-01 1.07939869e-01 -6.10898018e-01 -2.05723688e-01 3.64533454e-01 5.20426691e-01 1.65537059e-01 2.66931444e-01 -5.33601344e-01 -4.74104792e-01 7.45695293e-01 -4.60728705e-01 -2.13522255e-01 1.13560438e+00 2.13439777e-01 -3.15539777e-01 2.91769475e-01 1.64662993e+00 1.98417827e-01 -4.71199363e-01 -3.05612475e-01 1.51694091e-02 -2.72393376e-01 -4.14988458e-01 -3.03107738e-01 -8.29572499e-01 1.00481915e+00 6.30914629e-01 1.03543013e-01 8.62876952e-01 -1.40228838e-01 1.00395894e+00 3.23507041e-01 4.16397095e-01 -1.05529439e+00 -7.16866404e-02 4.68176574e-01 4.61311758e-01 -5.39120138e-01 -2.01324344e-01 -1.91484764e-01 -3.41762424e-01 1.13726938e+00 6.22831844e-02 -1.49346203e-01 4.40514803e-01 2.73940533e-01 -2.40563393e-01 -6.69921935e-02 -6.17299974e-01 -2.25358028e-02 4.44265194e-02 4.10157859e-01 6.26536906e-01 3.54491204e-01 -6.66757941e-01 4.92262840e-01 -1.80714890e-01 -4.24666017e-01 1.94803029e-01 1.16133988e+00 -5.01401544e-01 -1.71620965e+00 -2.82845885e-01 5.56604445e-01 -5.29494405e-01 -4.15085554e-01 5.99115379e-02 7.58412182e-01 2.93032497e-01 5.69179177e-01 3.21112037e-01 -4.83066052e-01 5.21776259e-01 5.13844252e-01 1.73823178e-01 -7.26159215e-01 -7.88548648e-01 -1.59358665e-01 -1.21707991e-01 -3.32350343e-01 6.70114383e-02 -5.51507235e-01 -1.34682226e+00 -4.96163398e-01 1.22420840e-01 2.97801673e-01 5.81478596e-01 4.35128152e-01 4.63464379e-01 5.58329403e-01 6.07289314e-01 -4.34772462e-01 -8.16533804e-01 -8.42463315e-01 -7.74104536e-01 5.41030824e-01 1.90246895e-01 -5.09920299e-01 -4.44278777e-01 -2.91886013e-02]

[10.594846725463867, 8.63782787322998]

02969134-978f-48a8-80c8-413671c4cfac

towards-protein-protein-docking-with

1605.09266

http://arxiv.org/abs/1605.09266v1

http://arxiv.org/pdf/1605.09266v1.pdf

Towards protein-protein docking with significant structural changes using CABS-dock

The protein-protein interactions (PPIs) are crucial for understanding the majority of cellular processes. PPIs play important role in gene transcription regulation, cellular signaling, molecular basis of immune response and more. Moreover, a disruption of hese mechanisms is frequently postulated as a possible cause of diseases such as Alzheimer's or cancer. For many of biologically relevant cases the structure of protein-protein complexes remain unknown. Therefore computational techniques, including molecular docking, have become a valuable part of drug discovery pipelines. Unfortunately, none of the widely used protein-protein docking tools is free from serious limitations. Typically, in docking simulations the protein flexibility is either completely neglected or very limited. Additionally, some knowledge of the approximate location and/or the shape of the active site is also required. Such limitations arise mostly from the enormous number of degrees of freedom of protein-protein systems. In this paper, an efficient computational method for protein-protein docking is proposed and initially tested on a single docking case. The proposed method is based on a two-step procedure. In the first step, CABS-dock web server for protein-peptide docking is used to dock a peptide, which is the appropriate protein fragment responsible for the protein-protein interaction, to the other protein partner. During peptide docking, no knowledge about the binding site, nor the peptide structure, is used and the peptide is allowed to be fully flexible. In the second step, the docked peptide is used in the structural adjustment of protein complex partners. The proposed method allowed us to obtain a high accuracy model, therefore it provides a promising framework for further advances.

[]

2016-05-30

['molecular-docking']

['medical']

[ 1.48593470e-01 -1.96575925e-01 -2.04052255e-01 -6.51454777e-02 -2.04524621e-01 -6.26548111e-01 1.77399874e-01 5.76221883e-01 -5.12667596e-01 1.32938206e+00 -3.83304775e-01 -3.94839108e-01 8.44963416e-02 -6.57388091e-01 -7.82944322e-01 -1.22516418e+00 1.68062076e-01 6.01540446e-01 4.86002326e-01 -2.21788242e-01 2.37207532e-01 7.79852808e-01 -1.28092360e+00 -1.61056310e-01 9.55630660e-01 3.70793104e-01 5.56733429e-01 3.13568145e-01 -3.60010386e-01 1.13580665e-02 -3.03982675e-01 -2.44402334e-01 -9.69569087e-02 -5.46237230e-01 -4.55015779e-01 -2.82961100e-01 -5.12947738e-01 3.42720717e-01 4.36392784e-01 8.61354113e-01 7.14990795e-01 1.48598969e-01 6.00673556e-01 -5.33087552e-01 2.13566080e-01 -2.68519610e-01 -3.69141430e-01 -1.42920673e-01 4.26878273e-01 -4.24432829e-02 7.70651042e-01 -1.07677627e+00 7.90921032e-01 8.34937751e-01 2.80339837e-01 3.79971355e-01 -1.42907858e+00 -6.29871845e-01 -1.49837688e-01 1.67598754e-01 -1.34143746e+00 -1.10949866e-01 7.20140517e-01 -4.61413056e-01 9.03571904e-01 1.90002769e-01 9.07023847e-01 6.72597885e-01 5.37605703e-01 -1.44957453e-02 9.40576255e-01 -4.07809645e-01 4.32483524e-01 1.72983080e-01 6.19406775e-02 3.37877959e-01 2.48486385e-01 -1.38377547e-01 -4.01412785e-01 -7.07315207e-01 2.26414993e-01 6.39591813e-02 -3.88083607e-01 -7.73563445e-01 -8.74620914e-01 7.64050066e-01 2.72731394e-01 5.48773766e-01 -4.64673549e-01 -4.36491877e-01 2.50327229e-01 -2.15821967e-01 -1.34882659e-01 2.03677312e-01 -5.32054543e-01 -3.47069502e-02 -4.48866844e-01 4.64246124e-01 8.15416098e-01 2.16689613e-02 6.56905651e-01 -4.98943835e-01 4.31945622e-01 5.44549823e-01 3.17733824e-01 1.14415839e-01 3.54609936e-01 -5.12146652e-02 6.08973168e-02 6.78783178e-01 5.34864426e-01 -9.27331984e-01 -5.47554791e-01 -1.02498658e-01 -7.39026487e-01 2.68992484e-01 5.82206190e-01 -5.31882644e-02 -5.04304767e-01 1.74453795e+00 7.65077829e-01 -2.55572554e-02 1.98349610e-01 9.04945672e-01 5.06356001e-01 7.29433060e-01 4.33809310e-01 -7.84205616e-01 1.65797555e+00 -4.26733077e-01 -4.52843457e-01 3.03351909e-01 3.75793189e-01 -1.03747082e+00 6.70835018e-01 3.35839510e-01 -8.14571261e-01 -3.80702823e-01 -1.04177916e+00 2.98627347e-01 -4.64756250e-01 -1.28729769e-03 4.33751374e-01 4.37154412e-01 -4.42618281e-01 6.44876063e-01 -9.74971414e-01 -5.93587637e-01 -1.63506866e-01 8.43663692e-01 -6.61457181e-01 9.03496752e-04 -1.18071342e+00 9.64692652e-01 7.20156312e-01 1.33183494e-01 -2.23653302e-01 -1.65222809e-01 -2.59595931e-01 1.11236975e-01 -7.97249097e-03 -6.01047993e-01 6.74509466e-01 -6.05907798e-01 -1.44867265e+00 5.68661988e-01 -4.11703616e-01 -1.69447154e-01 1.41657606e-01 1.98608920e-01 -1.68368056e-01 -5.75151220e-02 -1.03931077e-01 3.43327403e-01 1.47789598e-01 -8.69421899e-01 -5.20449467e-02 -5.39374352e-01 -3.00513387e-01 5.70491195e-01 4.44156498e-01 1.98123515e-01 -2.81981021e-01 -3.98567766e-01 3.58565152e-01 -9.80584145e-01 -4.69670385e-01 -1.44555509e-01 -2.74488628e-01 -1.95195004e-01 5.04895508e-01 -2.43948832e-01 8.72442365e-01 -2.08594346e+00 4.16240543e-01 5.15913904e-01 4.08591591e-02 5.76928973e-01 2.24239439e-01 1.04419947e+00 -5.62247217e-01 -1.09253109e-01 4.14255221e-04 3.87351900e-01 -5.31099439e-01 -1.18811674e-01 -1.03011923e-02 5.09386361e-01 -1.68323219e-02 4.05301422e-01 -4.70943063e-01 -1.84330866e-01 2.24418312e-01 8.24924350e-01 -2.95927435e-01 2.29168504e-01 -4.19539928e-01 9.71608460e-01 -8.74696970e-01 4.07864898e-01 8.27503145e-01 -6.57459646e-02 7.00985491e-01 -3.09328526e-01 -3.02989066e-01 7.33671486e-02 -1.07255590e+00 1.17577052e+00 2.57270664e-01 -2.40443140e-01 -8.24867785e-02 -8.59552085e-01 1.02454841e+00 7.25028634e-01 6.90833509e-01 -4.26605701e-01 7.25685284e-02 3.66439492e-01 4.18445319e-01 -2.27998286e-01 -2.78083324e-01 -3.55623603e-01 3.81791353e-01 1.18112281e-01 -4.95106548e-01 3.60821486e-01 1.88883364e-01 -7.76538178e-02 6.65447533e-01 2.54067659e-01 8.25564146e-01 -1.12047903e-01 8.95435214e-01 2.41553038e-01 7.44422078e-01 -5.38826995e-02 1.63357574e-02 1.58565879e-01 6.80592835e-01 -5.86828053e-01 -9.98179197e-01 -5.86309373e-01 -3.77448082e-01 5.52760303e-01 1.45835340e-01 -3.57593924e-01 -8.23515236e-01 -1.09414257e-01 -1.99724108e-01 1.93136081e-01 -2.15707242e-01 -7.55638778e-02 -3.70482206e-01 -1.25202417e+00 2.08083495e-01 -3.15215975e-01 1.49295688e-01 -1.13291836e+00 -3.86368066e-01 6.88525438e-01 3.58444080e-03 -5.31928837e-01 -5.76235913e-02 5.30931056e-01 -8.93179834e-01 -1.45937657e+00 -6.12333417e-01 -6.07717752e-01 5.45478702e-01 2.22147435e-01 3.96992534e-01 1.28270194e-01 -2.01705113e-01 -6.38745725e-01 -1.92708343e-01 -3.43532413e-01 -3.43797982e-01 -1.34602502e-01 1.78807318e-01 -2.37589285e-01 5.06696224e-01 -8.09158504e-01 -6.91869199e-01 5.74216545e-01 -6.42781019e-01 -1.80066433e-02 5.90426981e-01 8.67807090e-01 9.42775965e-01 7.77969211e-02 5.61154425e-01 -7.75698960e-01 6.36373043e-01 -3.78711760e-01 -8.14562678e-01 1.13349833e-01 -3.19487780e-01 2.24114805e-01 7.17399776e-01 -3.54179561e-01 -8.65586102e-01 5.88523030e-01 -5.17936349e-01 1.06819399e-01 -3.01548630e-01 7.06054449e-01 -8.84329796e-01 -2.76322067e-01 3.38831246e-01 4.13340300e-01 2.99641430e-01 -6.93411708e-01 -3.93814564e-01 2.85885930e-01 9.25552696e-02 -5.94093978e-01 4.88435715e-01 -3.11722923e-02 3.27982396e-01 -7.73636520e-01 -1.75215498e-01 -5.41330576e-01 -5.67237556e-01 3.01825911e-01 1.02739906e+00 -6.42379582e-01 -1.18396389e+00 2.27435082e-01 -1.20540535e+00 2.06099093e-01 5.34564495e-01 7.94793963e-01 -2.23116517e-01 6.79361224e-01 -3.08387607e-01 -4.96236354e-01 -3.71429116e-01 -1.73267317e+00 4.77764845e-01 2.60619730e-01 -8.79851803e-02 -6.87692046e-01 5.27449012e-01 3.38711053e-01 -4.79549877e-02 3.48191440e-01 1.37174034e+00 -9.87099946e-01 -4.72903967e-01 -3.24792325e-01 8.99284333e-03 -4.64848764e-02 1.90822884e-01 1.01120725e-01 -4.43848491e-01 -1.45056620e-01 -1.20980099e-01 -1.19497888e-01 2.94656873e-01 3.09302062e-01 5.48963726e-01 -2.44044662e-02 -7.12940454e-01 2.43539006e-01 1.35983658e+00 6.95655227e-01 7.44964123e-01 3.99881750e-01 4.51280564e-01 5.28309762e-01 1.08220851e+00 2.96476632e-01 -1.15815245e-01 1.19499230e+00 4.37355518e-01 -1.18595958e-01 5.82484186e-01 -2.11857714e-05 1.86256886e-01 2.27309763e-01 -4.35891241e-01 -2.93207139e-01 -8.72968435e-01 -5.67051210e-02 -1.74786448e+00 -9.11719441e-01 -3.60405028e-01 2.57659149e+00 9.93218720e-01 4.07357849e-02 2.60068238e-01 6.02112003e-02 6.99142396e-01 -4.55690175e-01 -7.27648795e-01 -4.20537263e-01 -1.14186034e-02 2.90418416e-01 2.27340907e-01 5.72956145e-01 -6.94573879e-01 8.24744701e-01 5.72672319e+00 5.70070505e-01 -1.20263064e+00 -2.91436195e-01 9.05218869e-02 3.11336368e-01 1.11113794e-01 4.06042159e-01 -8.80955458e-01 5.28340936e-01 6.97813272e-01 2.15164870e-02 2.00193048e-01 6.16825640e-01 7.34779656e-01 -3.94081742e-01 -6.51497424e-01 7.17136800e-01 -6.57541633e-01 -1.14536440e+00 -2.38379300e-01 5.48937023e-01 1.55307338e-01 -3.11057389e-01 -4.40937132e-01 -2.77811259e-01 -2.66127408e-01 -8.62848461e-01 3.47537063e-02 4.39183652e-01 4.84108627e-01 -9.89519656e-01 1.06581450e+00 5.61181724e-01 -9.47781384e-01 5.47878802e-01 -4.57747132e-01 7.49827400e-02 1.35915518e-01 7.07948267e-01 -1.05332685e+00 3.46287519e-01 2.46859640e-01 7.35232979e-02 -5.10802977e-02 9.93084371e-01 -4.81502526e-03 2.72245437e-01 -3.55186403e-01 -1.66548770e-02 -4.01107594e-03 -6.44725025e-01 4.29684401e-01 6.18517816e-01 5.16080521e-02 5.05300999e-01 3.78266603e-01 5.19331574e-01 1.40343487e-01 7.65215635e-01 -3.77103448e-01 -1.67900443e-01 4.13424641e-01 1.17146456e+00 -8.03098500e-01 7.07747936e-02 -3.79023194e-01 8.53923738e-01 2.47353852e-01 3.30461472e-01 -7.74999559e-01 -2.39820272e-01 9.66640353e-01 2.28579298e-01 1.97128534e-01 -7.39193410e-02 6.13407850e-01 -8.39389205e-01 -1.01933300e-01 -8.84726644e-01 8.79314914e-02 -4.74256724e-01 -6.50617123e-01 2.38185465e-01 -2.01329455e-01 -9.30046380e-01 -8.31617787e-02 -4.08948392e-01 -5.90472341e-01 1.24758637e+00 -1.21588719e+00 -8.68068337e-01 5.44021055e-02 4.35421884e-01 1.46621928e-01 5.09040207e-02 1.25387645e+00 2.14436844e-01 -8.02004576e-01 1.56395704e-01 4.88472551e-01 -4.29687411e-01 9.55828607e-01 -8.66639674e-01 -2.12197974e-01 3.44560802e-01 -4.34460998e-01 1.05077028e+00 1.21948993e+00 -8.69990885e-01 -1.12609577e+00 -5.46181679e-01 9.52182829e-01 3.32682393e-02 4.21051592e-01 -3.60711783e-01 -1.02267158e+00 2.80138820e-01 -1.83786690e-01 -1.18172660e-01 1.14083827e+00 -1.54275253e-01 1.92129433e-01 1.38188273e-01 -1.13688958e+00 4.78634596e-01 3.08723599e-02 -4.68340702e-02 -3.46188784e-01 4.54656541e-01 4.28705752e-01 -4.46657717e-01 -1.02164626e+00 1.57430485e-01 6.53469324e-01 -1.07387793e+00 1.11581874e+00 -5.83467782e-01 -4.26803321e-01 -7.32570648e-01 -2.87924074e-02 -8.05159688e-01 -2.82908827e-01 -2.94463962e-01 2.23414481e-01 8.71722281e-01 2.82572955e-01 -5.56510746e-01 9.28339243e-01 5.48304319e-01 1.96932070e-02 -8.72462153e-01 -9.44985211e-01 -2.40246803e-01 -2.06399471e-01 1.68837041e-01 4.06023562e-01 7.35299587e-01 1.83900177e-01 6.01041317e-01 -2.82672942e-01 3.46904173e-02 3.43773514e-01 9.99627868e-04 7.94195592e-01 -1.53638434e+00 -3.04278910e-01 -3.90415527e-02 -4.52561259e-01 -3.37251157e-01 -1.26277953e-01 -4.43730921e-01 -3.71119261e-01 -1.26407957e+00 3.30448925e-01 -2.60624588e-01 -1.58273965e-01 3.64714891e-01 -7.88985863e-02 -9.21733901e-02 -1.60242289e-01 3.92262965e-01 4.26344946e-03 4.31826264e-01 9.80836451e-01 2.34097451e-01 -5.55782199e-01 4.60068166e-01 -3.86167228e-01 6.13573551e-01 8.89521658e-01 -6.84465587e-01 -3.72584820e-01 5.96779764e-01 2.93825328e-01 1.13482773e-01 -1.68604627e-02 -6.06132388e-01 4.81988601e-02 -4.42393541e-01 3.34474593e-01 -6.49825871e-01 4.33103591e-01 -1.13659501e+00 9.34704840e-01 8.05635989e-01 3.60876203e-01 -2.90616229e-02 1.47222549e-01 5.92977047e-01 -2.28950590e-01 -3.93417597e-01 1.00835812e+00 -1.23083256e-01 -1.72666386e-01 -3.88583802e-02 -5.50850928e-01 -6.76377475e-01 1.23402071e+00 -5.86875498e-01 1.88516393e-01 1.54402122e-01 -1.22451460e+00 -2.16908514e-01 8.30128074e-01 -7.30457380e-02 2.65750438e-01 -8.62035036e-01 -1.12588055e-01 1.53545365e-02 1.86849639e-01 -1.02841802e-01 2.62248874e-01 1.03891361e+00 -8.46451938e-01 9.34962213e-01 -2.08585545e-01 -3.79227489e-01 -1.76113284e+00 7.93355942e-01 4.34885979e-01 -2.40228117e-01 -1.66402042e-01 2.19400659e-01 3.13050359e-01 -1.87464803e-01 -6.24916777e-02 4.34593298e-02 -5.37712753e-01 -1.42470300e-01 5.36206186e-01 7.58602694e-02 1.99714407e-01 -9.59711552e-01 -5.36360860e-01 7.34920144e-01 -3.17776203e-01 4.06787455e-01 1.47834849e+00 2.83488509e-04 -4.52398896e-01 -7.48665556e-02 8.29838157e-01 1.03023976e-01 -8.06821287e-01 9.46176350e-02 -3.07721496e-02 -1.06606863e-01 -2.24917680e-01 -7.35811234e-01 -4.01824564e-01 6.72931612e-01 6.73129320e-01 -4.84990299e-01 9.38750267e-01 -1.11636020e-01 3.69713962e-01 5.32477558e-01 5.10600984e-01 -6.01304710e-01 -3.59560221e-01 1.61334857e-01 4.92816120e-01 -1.03981376e+00 9.05936584e-02 -6.24102354e-01 -5.21305442e-01 1.15262687e+00 4.57951486e-01 3.15429002e-01 4.21307832e-01 -3.18230130e-02 -2.35169351e-01 -2.91769892e-01 -6.78242147e-01 1.03329502e-01 1.14664182e-01 3.93489152e-01 8.91183913e-01 -2.35451117e-01 -1.07669497e+00 6.92197740e-01 2.62723863e-01 4.10962291e-02 1.91315740e-01 8.66883159e-01 -5.83608389e-01 -1.88975847e+00 -7.47782588e-01 -2.44999066e-01 -6.20348394e-01 7.71532208e-02 -6.46771848e-01 9.13175702e-01 2.54433334e-01 6.00719094e-01 -5.34401596e-01 1.41819134e-01 3.15083444e-01 2.95575321e-01 2.91006505e-01 -4.98449147e-01 -4.52292264e-01 5.77064037e-01 -2.32900590e-01 -3.02697748e-01 -4.67764765e-01 -6.69954836e-01 -1.79757881e+00 -2.70685673e-01 -6.65917277e-01 8.95584762e-01 1.03058648e+00 9.64269578e-01 4.61046576e-01 3.68863195e-02 3.52147341e-01 -7.17322767e-01 -5.44450916e-02 -5.59960663e-01 -7.11897552e-01 3.46751288e-02 -8.40628222e-02 -8.35128963e-01 9.93759334e-02 -8.50319937e-02]

[4.780991077423096, 5.348245620727539]

749530ff-d22c-4130-9d40-399bc769f66f

geometric-matrix-completion-via-sylvester

2206.09477

https://arxiv.org/abs/2206.09477v1

https://arxiv.org/pdf/2206.09477v1.pdf

Geometric Matrix Completion via Sylvester Multi-Graph Neural Network

Despite the success of the Sylvester equation empowered methods on various graph mining applications, such as semi-supervised label learning and network alignment, there also exists several limitations. The Sylvester equation's inability of modeling non-linear relations and the inflexibility of tuning towards different tasks restrict its performance. In this paper, we propose an end-to-end neural framework, SYMGNN, which consists of a multi-network neural aggregation module and a prior multi-network association incorporation learning module. The proposed framework inherits the key ideas of the Sylvester equation, and meanwhile generalizes it to overcome aforementioned limitations. Empirical evaluations on real-world datasets show that the instantiations of SYMGNN overall outperform the baselines in geometric matrix completion task, and its low-rank instantiation could further reduce the memory consumption by 16.98\% on average.

['Hanghang Tong', 'Fei Wang', 'Changhe Yuan', 'Boxin Du']

2022-06-19

['matrix-completion']

['methodology']

[ 3.88800979e-01 2.58910805e-01 -4.10860091e-01 -3.26478630e-01 -3.52556735e-01 -1.26139000e-01 4.80823815e-01 -3.69891003e-02 -3.15811366e-01 6.19138062e-01 7.40273148e-02 -5.19667566e-01 -4.98051047e-01 -6.74030364e-01 -5.39039433e-01 -3.96277428e-01 -4.87806588e-01 6.74739003e-01 -2.78254449e-01 -1.55361116e-01 -2.25667849e-01 3.93782854e-01 -8.38151991e-01 1.16621874e-01 9.06922102e-01 1.01189125e+00 -1.02087706e-01 1.45259306e-01 6.16567768e-02 9.55461919e-01 -8.63177404e-02 -6.23751283e-01 4.44706351e-01 1.84035391e-01 -7.61201739e-01 -1.14973970e-01 9.44525719e-01 -2.73824662e-01 -7.20860362e-01 9.26678956e-01 4.12675530e-01 2.39081442e-01 6.43890798e-01 -1.61757803e+00 -7.66497493e-01 8.15229416e-01 -9.92100477e-01 2.54370715e-03 1.35995999e-01 -3.46127123e-01 1.53262329e+00 -1.04020762e+00 4.91652399e-01 1.21452379e+00 1.16357386e+00 1.84086919e-01 -1.35267591e+00 -1.10941291e+00 3.65001380e-01 5.25504127e-02 -1.58306742e+00 -2.80729473e-01 8.48378837e-01 -2.71535158e-01 8.96294415e-01 -7.40609365e-03 2.60917008e-01 1.07128906e+00 1.19587744e-03 8.70943964e-01 1.07422268e+00 -1.54819608e-01 -3.03749651e-01 -3.63057435e-01 2.61002690e-01 1.10518551e+00 5.22547245e-01 2.43759155e-02 -5.99315643e-01 -2.65170217e-01 9.23757374e-01 8.60780105e-03 1.18443444e-01 -5.74735940e-01 -1.20631671e+00 7.11014926e-01 7.80156195e-01 -9.78650600e-02 -2.57857621e-01 3.13282520e-01 4.09299016e-01 2.15512335e-01 6.61379814e-01 5.20391881e-01 -2.97246128e-01 5.05347848e-01 -8.07049513e-01 -2.09702812e-02 8.71544659e-01 1.04374301e+00 7.17736900e-01 2.08376214e-01 -1.80002227e-02 9.14791644e-01 4.23965037e-01 2.60276407e-01 2.53466815e-02 -7.31525481e-01 8.83439064e-01 8.06030691e-01 -6.48715377e-01 -1.53087723e+00 -9.13752377e-01 -9.30378616e-01 -1.58510137e+00 -2.00127244e-01 3.31426889e-01 -2.95702040e-01 -6.55396879e-01 1.99713421e+00 2.73077965e-01 5.79661846e-01 -1.74375236e-01 7.19397902e-01 9.82231140e-01 2.69470364e-01 2.07497075e-01 8.82352069e-02 1.14109576e+00 -1.45285976e+00 -6.23849750e-01 -4.41827923e-01 8.92647564e-01 -6.50831282e-01 8.11976671e-01 3.57084513e-01 -7.61708140e-01 -4.83395904e-01 -1.18125200e+00 -9.69271064e-02 -2.60815620e-01 4.06075388e-01 1.60793555e+00 4.30669934e-01 -1.07841027e+00 6.52959824e-01 -7.20393479e-01 -2.89239436e-01 4.76197153e-01 8.06086838e-01 -6.98330283e-01 -3.56188081e-02 -1.01343596e+00 5.42091310e-01 7.28496611e-01 3.94559354e-01 -5.58186114e-01 -6.84652805e-01 -9.12622869e-01 5.87696657e-02 8.68797302e-01 -1.01216280e+00 7.46688306e-01 -4.44447130e-01 -1.18045366e+00 5.88710845e-01 2.12660506e-01 -4.73157167e-01 2.44222254e-01 -4.41819310e-01 -5.11569738e-01 -1.36257922e-02 -7.52613321e-02 5.41863620e-01 4.96905118e-01 -1.02219355e+00 -3.66599470e-01 -3.79364610e-01 1.04027040e-01 3.05227011e-01 -8.31121922e-01 -2.12031156e-01 -5.74052870e-01 -9.39825416e-01 3.77071351e-01 -1.17048323e+00 -4.68903750e-01 -9.32295918e-02 -7.86021352e-01 -2.85628408e-01 7.88741887e-01 -5.11286557e-01 1.32134449e+00 -1.84197795e+00 9.94482115e-02 5.61022878e-01 9.10548091e-01 2.03197137e-01 -6.23408735e-01 5.55896342e-01 -3.66445303e-01 -1.40473992e-01 -9.46596786e-02 -5.88851988e-01 -1.48294270e-02 2.04963431e-01 -1.33506328e-01 4.08584207e-01 1.30200252e-01 1.05521107e+00 -9.11483586e-01 -5.66892564e-01 -3.80795337e-02 4.78044689e-01 -4.00930941e-01 -8.87322575e-02 -3.82468142e-02 2.23651931e-01 -2.98678309e-01 6.98787630e-01 6.51086628e-01 -8.38502526e-01 5.33411622e-01 -6.17538095e-01 4.97246683e-01 2.53906578e-01 -1.49470913e+00 1.93428147e+00 -3.82614404e-01 3.62835616e-01 1.94334567e-01 -1.01797688e+00 8.53121817e-01 1.22203857e-01 7.50558555e-01 -5.16194701e-01 -4.95240465e-02 4.89921570e-02 1.70699403e-01 -1.46802992e-01 4.52536672e-01 4.09476042e-01 6.22836649e-02 6.88935757e-01 1.41273990e-01 5.03114223e-01 4.08628136e-01 7.43068218e-01 1.24608994e+00 2.07832232e-01 2.16720402e-01 -2.60708004e-01 5.35219789e-01 -3.04878414e-01 5.04042566e-01 6.31266415e-01 2.33237237e-01 2.84567386e-01 4.81865346e-01 -5.32982111e-01 -7.30145991e-01 -1.08700061e+00 1.52686864e-01 1.37038994e+00 -9.42742638e-03 -7.68458843e-01 -2.24656075e-01 -9.23866630e-01 4.78422828e-02 2.82701463e-01 -4.49610591e-01 -6.81783333e-02 -7.12927639e-01 -1.08916140e+00 9.33748066e-01 6.54619694e-01 7.64829457e-01 -5.58120966e-01 4.37704355e-01 1.03705890e-01 -1.98856100e-01 -1.56504560e+00 -5.00760138e-01 -8.30148384e-02 -9.89125907e-01 -1.14879084e+00 -1.41633287e-01 -7.45704472e-01 8.38812411e-01 4.00397688e-01 1.20824337e+00 5.28018661e-02 -1.10164098e-01 1.90291196e-01 -2.73296535e-02 -1.12203397e-01 -1.26924649e-01 7.17106760e-01 1.68135673e-01 8.90024081e-02 6.66346848e-02 -1.07013321e+00 -4.93350863e-01 2.11933613e-01 -6.94671810e-01 3.89769793e-01 1.08513975e+00 8.80895019e-01 4.17988330e-01 7.85661265e-02 7.79513776e-01 -1.44327724e+00 7.54126728e-01 -3.76674533e-01 -4.55525428e-01 3.02295029e-01 -1.28383589e+00 1.00618806e-02 5.15234232e-01 -3.60500604e-01 -8.60753596e-01 2.36017630e-01 1.93221912e-01 -3.02896321e-01 2.68385351e-01 9.13684011e-01 -1.25225663e-01 -3.37135464e-01 4.48314786e-01 -2.61597186e-01 1.89263329e-01 -4.96589392e-01 6.17151380e-01 1.22291118e-01 6.68086112e-01 -7.62216032e-01 1.25038218e+00 3.07276249e-01 6.98093057e-01 -5.35832524e-01 -1.16510332e+00 -4.44647700e-01 -7.12418139e-01 1.19278776e-02 4.90154654e-01 -1.33599353e+00 -1.01902974e+00 2.91664988e-01 -1.04860580e+00 -1.20945275e-01 1.56701609e-01 5.49866080e-01 -1.82928413e-01 6.11923873e-01 -9.38078821e-01 -4.92770970e-01 -6.58712626e-01 -7.70332694e-01 8.32345843e-01 -3.79636399e-02 -2.28742689e-01 -1.36515534e+00 -1.87003016e-01 6.88928664e-01 3.32102925e-01 2.84592122e-01 1.08460009e+00 -8.17131579e-01 -6.78102672e-01 -2.98276454e-01 -7.28789270e-01 2.85879076e-01 -1.54631678e-02 -3.27062339e-01 -7.92083383e-01 -5.30522048e-01 -6.25470698e-01 -5.01955390e-01 1.02495039e+00 -4.88032438e-02 1.13320518e+00 -4.96806413e-01 -4.89893019e-01 9.15622771e-01 1.40520406e+00 -3.95907104e-01 3.54726195e-01 1.40552297e-01 1.45046651e+00 3.61855537e-01 2.09692776e-01 1.20166846e-01 8.28454852e-01 6.11938477e-01 6.80419385e-01 -6.22835338e-01 -2.99222112e-01 -2.60827184e-01 2.15286851e-01 1.31344438e+00 -2.55288839e-01 -3.94544639e-02 -8.81772816e-01 1.97315082e-01 -2.10981393e+00 -6.94297254e-01 -3.00489336e-01 1.79828393e+00 6.86648250e-01 1.40997797e-01 2.11519301e-01 -1.62366852e-01 4.31204855e-01 3.97088975e-01 -3.97363901e-01 -1.58684712e-03 -3.19115788e-01 1.78539157e-01 6.89234555e-01 3.29619259e-01 -1.33933067e+00 1.01668656e+00 6.18262291e+00 9.71615732e-01 -7.66193092e-01 -1.07733086e-01 5.08306026e-01 1.24223813e-01 7.05639347e-02 2.99483258e-02 -7.79206157e-01 -2.05644950e-01 5.92664957e-01 -4.85362625e-03 6.29650831e-01 7.08817661e-01 -2.74583071e-01 2.86263973e-01 -1.27653706e+00 8.83698225e-01 1.88599855e-01 -1.28756785e+00 1.53395876e-01 1.91497102e-01 8.79120588e-01 2.93714881e-01 1.83461264e-01 4.66101557e-01 6.84511840e-01 -1.15747654e+00 4.76106023e-03 4.57091302e-01 7.88195491e-01 -6.28690541e-01 5.97581983e-01 1.07219785e-01 -1.43295383e+00 6.85612857e-02 -1.67410821e-01 -3.28352779e-01 1.61814671e-02 7.70123005e-01 -1.14481616e+00 1.13285112e+00 2.34314516e-01 9.27013695e-01 -9.40705180e-01 7.78778851e-01 -2.15846628e-01 7.08547950e-01 -4.22879666e-01 2.36879736e-01 2.60697275e-01 -5.62772453e-01 2.42577866e-01 1.26565719e+00 -3.91571261e-02 -1.97780117e-01 6.55698180e-01 6.19647920e-01 -4.29548442e-01 2.50535578e-01 -5.93898118e-01 -7.88308606e-02 4.72285509e-01 1.82947755e+00 -5.70620000e-01 -1.12802990e-01 -5.99049449e-01 7.56440222e-01 7.10851073e-01 5.65204382e-01 -8.76520395e-01 -2.85740405e-01 2.86065578e-01 -5.59800304e-02 -8.02595764e-02 -4.12457854e-01 -5.80718994e-01 -1.08820903e+00 1.02357432e-01 -9.32222664e-01 6.30599260e-01 -4.19128507e-01 -1.60573137e+00 7.02019334e-01 -5.41581586e-02 -8.87937725e-01 -1.43591553e-01 -6.75606906e-01 -4.41054434e-01 6.52097285e-01 -1.36487806e+00 -1.82821214e+00 -3.32979858e-01 4.83058214e-01 -9.58598927e-02 -4.01126057e-01 9.30863380e-01 7.46133029e-01 -1.01176167e+00 1.00588000e+00 -2.52133384e-02 5.56222737e-01 7.55697966e-01 -1.28034902e+00 5.22957206e-01 8.00033569e-01 4.68465269e-01 8.36169302e-01 2.78507531e-01 -6.68662071e-01 -1.41914439e+00 -1.44762194e+00 7.38180161e-01 -1.70103148e-01 9.74089742e-01 -4.69508260e-01 -7.75116324e-01 1.04609954e+00 1.35061324e-01 5.85917793e-02 8.13572407e-01 7.36023307e-01 -8.43176842e-01 -4.84834164e-01 -5.68424702e-01 8.34766865e-01 1.49083650e+00 -6.41071439e-01 -3.20601761e-02 7.79742479e-01 5.82677543e-01 -4.09289062e-01 -1.16792846e+00 8.60273838e-01 4.32728499e-01 -5.92575431e-01 1.30073786e+00 -7.58516312e-01 4.95633841e-01 -1.46917105e-01 6.18041260e-04 -9.85213101e-01 -5.94991684e-01 -7.38832653e-01 -4.75105464e-01 1.34483480e+00 5.92103958e-01 -7.09573925e-01 1.11129379e+00 5.27970076e-01 -1.24370262e-01 -1.12859547e+00 -6.81249022e-01 -8.98099899e-01 -2.88100243e-01 -2.59944737e-01 4.24981058e-01 1.43588066e+00 -2.18428925e-01 1.20449054e+00 -7.17986643e-01 2.46557772e-01 9.46364164e-01 -2.32371539e-02 1.06978309e+00 -1.54439890e+00 -5.03622353e-01 -4.08727407e-01 -1.99876338e-01 -1.20981038e+00 5.59999287e-01 -1.48442400e+00 -5.65678716e-01 -1.57030439e+00 2.53408074e-01 -8.18657339e-01 -4.64450032e-01 7.83639789e-01 -2.58692354e-01 3.34129512e-01 1.82060674e-01 2.38067418e-01 -9.35982347e-01 5.31623483e-01 1.17386770e+00 -2.33113885e-01 -1.89168621e-02 -8.76345858e-02 -8.07453990e-01 9.15587008e-01 6.65377498e-01 -3.79012853e-01 -6.94909453e-01 -6.51967824e-01 5.34721434e-01 -1.05048552e-01 3.04548621e-01 -1.11380577e+00 5.58123648e-01 1.82255566e-01 3.08614880e-01 -6.39038324e-01 4.59954858e-01 -8.73305142e-01 1.66121647e-01 2.19272867e-01 -2.75492668e-01 4.64352012e-01 4.85964268e-02 8.30851972e-01 8.91062338e-03 2.17367083e-01 2.58235365e-01 3.03449512e-01 -5.67195475e-01 7.75003493e-01 2.92759240e-01 2.99697593e-02 7.45822966e-01 -1.46374032e-01 -5.01198769e-01 -2.73299754e-01 -5.04241407e-01 4.29639757e-01 -9.50962901e-02 3.67185742e-01 4.13821876e-01 -1.50760853e+00 -7.39303648e-01 5.86701520e-02 -3.07694171e-02 2.15413302e-01 5.43097854e-02 1.15833187e+00 -3.17311555e-01 3.15658987e-01 -5.20198569e-02 -4.35277015e-01 -1.20601678e+00 5.41948140e-01 4.77703065e-02 -1.01530921e+00 -5.86337566e-01 8.20092976e-01 1.86332881e-01 -7.17527568e-01 3.75872672e-01 7.29303807e-02 -1.48472637e-01 -1.79411434e-02 5.96069284e-02 6.25011802e-01 2.79056579e-02 -4.90071207e-01 -2.08539829e-01 4.20098126e-01 -5.01607716e-01 2.66894311e-01 1.35925460e+00 -3.05211302e-02 -4.95886683e-01 1.85215250e-01 1.09395552e+00 -4.17207666e-02 -8.35695922e-01 -9.02184546e-01 2.72701263e-01 -5.84996156e-02 -7.91522488e-02 -8.43730092e-01 -1.33988833e+00 5.84792256e-01 1.13262609e-01 -6.33807927e-02 1.18971705e+00 -2.27267101e-01 8.97696853e-01 8.98049712e-01 2.20495701e-01 -9.06818509e-01 1.38189375e-01 7.12310851e-01 6.79167628e-01 -1.21354830e+00 6.56456411e-01 -1.08010447e+00 -4.10306782e-01 1.00149393e+00 9.10213828e-01 -1.75988570e-01 7.29563355e-01 1.33972168e-01 -1.68326851e-02 -4.57345992e-01 -8.64716589e-01 -5.81229739e-02 6.12425327e-01 5.58465481e-01 5.02920747e-01 -4.89251092e-02 -1.30547941e-01 3.36369723e-01 -2.46398404e-01 -3.12105894e-01 1.82922885e-01 3.05467844e-01 2.02914745e-01 -1.12270617e+00 1.12485485e-02 7.66879499e-01 -3.18689704e-01 -3.58668000e-01 -4.12757099e-01 1.00340927e+00 -3.83766666e-02 8.25334251e-01 -9.48857069e-02 -7.15730488e-01 2.84608215e-01 -1.91793337e-01 3.47313106e-01 -5.75109780e-01 -6.74930692e-01 3.52092162e-02 5.65044522e-01 -5.74659348e-01 -5.98954797e-01 -3.34384412e-01 -1.15275967e+00 -5.13889968e-01 -5.11764348e-01 -1.01114742e-01 4.15566146e-01 9.23103094e-01 5.37246764e-01 7.18980253e-01 3.70984972e-01 -4.36890692e-01 -7.62821496e-01 -1.07354617e+00 -5.50723970e-01 4.71560538e-01 -9.68853831e-02 -6.35100067e-01 -3.85451317e-02 -2.44461998e-01]

[7.2095465660095215, 6.230625629425049]

742b6f2d-b849-4c30-b82a-83651892e57d

dynamic-algorithms-for-matroid-submodular

2306.00959

https://arxiv.org/abs/2306.00959v1

https://arxiv.org/pdf/2306.00959v1.pdf

Dynamic Algorithms for Matroid Submodular Maximization

Submodular maximization under matroid and cardinality constraints are classical problems with a wide range of applications in machine learning, auction theory, and combinatorial optimization. In this paper, we consider these problems in the dynamic setting where (1) we have oracle access to a monotone submodular function $f: 2^{V} \rightarrow \mathbb{R}^+$ and (2) we are given a sequence $\mathcal{S}$ of insertions and deletions of elements of an underlying ground set $V$. We develop the first parameterized (by the rank $k$ of a matroid $\mathcal{M}$) dynamic $(4+\epsilon)$-approximation algorithm for the submodular maximization problem under the matroid constraint using an expected worst-case $O(k\log(k)\log^3{(k/\epsilon)})$ query complexity where $0 < \epsilon \le 1$. Chen and Peng at STOC'22 studied the complexity of this problem in the insertion-only dynamic model (a restricted version of the fully dynamic model where deletion is not allowed), and they raised the following important open question: *"for fully dynamic streams [sequences of insertions and deletions of elements], there is no known constant-factor approximation algorithm with poly(k) amortized queries for matroid constraints."* Our dynamic algorithm answers this question as well as an open problem of Lattanzi et al. (NeurIPS'20) affirmatively. As a byproduct, for the submodular maximization under the cardinality constraint $k$, we propose a parameterized (by the cardinality constraint $k$) dynamic algorithm that maintains a $(2+\epsilon)$-approximate solution of the sequence $\mathcal{S}$ at any time $t$ using the expected amortized worst-case complexity $O(k\epsilon^{-1}\log^2(k))$. This is the first dynamic algorithm for the problem that has a query complexity independent of the size of ground set $V$.

['Morteza Monemizadeh', 'Peyman Jabbarzade', 'Mohammadtaghi Hajiaghayi', 'Samira Goudarzi', 'Leyla Biabani', 'Kiarash Banihashem']

2023-06-01

['combinatorial-optimization', 'open-question']

['methodology', 'natural-language-processing']

[ 2.19718456e-01 2.98613280e-01 -3.32368076e-01 -8.46178606e-02 -8.15472424e-01 -1.06678224e+00 -4.11213845e-01 3.46425146e-01 -9.14381027e-01 7.30262458e-01 -6.10261381e-01 -6.27928436e-01 -6.77389562e-01 -1.31005490e+00 -1.18406951e+00 -9.10132825e-01 -6.10718727e-01 9.71445978e-01 8.89651924e-02 -3.39982092e-01 1.87669292e-01 2.73503631e-01 -1.53920197e+00 2.18536258e-02 6.17932975e-01 1.34667027e+00 9.43706185e-02 7.98676968e-01 -2.80072480e-01 2.88230002e-01 -3.92935693e-01 -7.15478957e-01 9.91528571e-01 -4.45867687e-01 -1.08010173e+00 3.59181195e-01 3.92298162e-01 5.08594839e-03 -6.18228316e-01 1.45524335e+00 1.23077519e-01 -9.69130322e-02 -1.64436065e-02 -1.46458566e+00 -1.72359675e-01 1.22014749e+00 -9.38548505e-01 2.86715448e-01 5.93724363e-02 -3.72077487e-02 1.34623837e+00 -2.37490728e-01 6.35607421e-01 8.06266963e-01 -3.75594497e-02 2.12937757e-01 -1.27183807e+00 -8.63990247e-01 3.39325160e-01 1.98044226e-01 -1.62830293e+00 7.18967542e-02 3.95698875e-01 -8.35830122e-02 8.13685715e-01 9.87201154e-01 5.25950491e-01 -3.19672644e-01 -2.08896860e-01 6.09023035e-01 1.03325868e+00 -5.67122161e-01 3.23668808e-01 2.49956504e-01 4.74749086e-03 8.14438343e-01 6.98013186e-01 -2.49077320e-01 -4.07746106e-01 -2.86694199e-01 3.26647371e-01 -8.64612907e-02 -3.22227269e-01 -3.59114170e-01 -9.29474413e-01 1.07833648e+00 1.82680055e-01 3.54664326e-01 7.37134833e-03 8.07581782e-01 6.08236268e-02 1.01295841e+00 1.30803674e-01 4.91368711e-01 -6.98244393e-01 -1.82843171e-02 -8.67464662e-01 5.40644884e-01 1.12135255e+00 1.32054877e+00 9.19202745e-01 -9.03692096e-02 3.62390727e-01 3.14451486e-01 -1.43651217e-01 9.33977544e-01 -3.30322891e-01 -1.20846069e+00 1.03866136e+00 7.09245682e-01 3.23846638e-01 -7.68839955e-01 -1.87525034e-01 -3.71288985e-01 -5.58589518e-01 1.08602934e-01 6.57167077e-01 -1.56994581e-01 -3.67602587e-01 2.08241296e+00 3.44277292e-01 -5.46366215e-01 -2.69295424e-01 5.04572332e-01 9.73811895e-02 7.19279170e-01 -6.00679576e-01 -1.00370252e+00 1.19057143e+00 -5.07447779e-01 -4.85609144e-01 -8.15952290e-03 9.43541229e-01 -5.63772321e-01 7.91863441e-01 7.07430005e-01 -1.84549248e+00 3.12648833e-01 -8.94346535e-01 3.87987763e-01 -3.01619798e-01 -5.80735326e-01 9.64659989e-01 1.22427344e+00 -1.27321410e+00 7.99056813e-02 -2.84185082e-01 1.16605841e-01 7.50420019e-02 8.88949811e-01 -2.68126547e-01 -4.59443152e-01 -6.06201351e-01 2.78981060e-01 2.50793129e-01 -5.49128577e-02 -6.69136941e-01 -4.97128069e-01 -7.31287062e-01 -3.03649474e-02 1.17743552e+00 -4.13922817e-01 1.02560461e+00 -6.00388229e-01 -6.95141673e-01 9.52278972e-01 -2.06893206e-01 -4.87061888e-01 5.40285230e-01 4.50751632e-01 2.53587395e-01 2.36482531e-01 -1.08084045e-01 1.36016637e-01 2.28069052e-01 -1.02890348e+00 -9.82549489e-01 -9.32404459e-01 9.22127604e-01 2.41732091e-01 -2.91234434e-01 6.08725548e-02 -3.74217510e-01 -2.67389357e-01 6.14408195e-01 -1.09249210e+00 -6.44386113e-01 -1.81216210e-01 -7.00143501e-02 -5.69704780e-03 2.51060456e-01 -1.17512599e-01 1.49362814e+00 -1.94823778e+00 4.53269541e-01 6.95576012e-01 3.24667454e-01 -2.41733417e-01 -3.90949175e-02 5.90307713e-01 2.16598153e-01 3.96568865e-01 -4.09964234e-01 -1.97098181e-01 1.37501851e-01 2.87697881e-01 -7.65179396e-02 7.08158374e-01 -9.70234096e-01 5.69877267e-01 -4.08337742e-01 -1.15129836e-01 -4.27607954e-01 -4.77805048e-01 -7.55715787e-01 -3.02978784e-01 -8.27294528e-01 -4.86685246e-01 -1.22484706e-01 6.89739704e-01 1.17475212e+00 -2.49700814e-01 5.75375557e-01 3.64043772e-01 -4.34129313e-02 -2.98991203e-01 -2.03420615e+00 1.49553216e+00 -2.36413687e-01 9.06838551e-02 8.16740692e-01 -1.12891030e+00 3.45963955e-01 1.11195892e-02 9.33507264e-01 -6.81408525e-01 1.73417479e-01 6.07669413e-01 -2.03755379e-01 -2.57742375e-01 5.31007469e-01 -4.43868369e-01 -6.07957423e-01 8.68095696e-01 -4.44580883e-01 -1.23466775e-01 3.59957278e-01 4.51934427e-01 1.51070797e+00 -8.04256082e-01 -3.47813189e-01 -3.21073294e-01 3.46903235e-01 8.27583447e-02 5.90745807e-01 1.08378923e+00 2.49102905e-01 5.28023057e-02 9.04520869e-01 -3.69653881e-01 -8.28842580e-01 -8.54721129e-01 1.00413971e-01 1.34528220e+00 4.72970039e-01 -4.04927313e-01 -8.01807940e-01 -3.99347574e-01 2.09116921e-01 5.49805522e-01 -7.17661500e-01 3.57505262e-01 -5.89481831e-01 -9.11251485e-01 2.92595893e-01 3.40772599e-01 4.42732394e-01 -6.90404713e-01 -3.39751631e-01 3.00310284e-01 -7.22761601e-02 -1.07311058e+00 -8.30991209e-01 5.30497134e-01 -9.25601304e-01 -1.07840908e+00 -3.61670673e-01 -5.85826218e-01 8.76473963e-01 4.88336176e-01 6.73223078e-01 1.82921067e-01 -3.80541354e-01 6.47870064e-01 -2.11966306e-01 -4.75318402e-01 1.13276884e-01 -6.23564832e-02 -1.24922976e-01 -3.18540752e-01 2.30261773e-01 -4.02581781e-01 -5.87255776e-01 2.55641550e-01 -1.36179793e+00 -4.96214390e-01 2.37284169e-01 4.08346236e-01 1.06970596e+00 5.70444942e-01 4.16350551e-02 -1.03490984e+00 2.41024524e-01 -3.43815893e-01 -1.49167430e+00 2.40072057e-01 -8.83223951e-01 6.73011765e-02 6.68771684e-01 -1.69677645e-01 -3.79213870e-01 -2.63224114e-02 2.65649974e-01 -1.89549968e-01 6.11367762e-01 6.75058186e-01 -3.84541094e-01 -2.69362241e-01 3.86338085e-01 6.24626935e-01 4.95632663e-02 -3.14449161e-01 2.98883229e-01 3.36249828e-01 2.99447119e-01 -7.58454025e-01 8.91954541e-01 7.85219133e-01 6.33053243e-01 -5.70700765e-01 -4.55374032e-01 -4.35230792e-01 5.47205620e-02 2.78880537e-01 2.32387617e-01 -6.66868865e-01 -1.49461484e+00 1.66744336e-01 -6.91004813e-01 -3.73406321e-01 -6.39210761e-01 1.83729142e-01 -8.84615541e-01 5.50953031e-01 -3.31407756e-01 -1.17687392e+00 -2.79684365e-01 -1.05768275e+00 4.01490539e-01 -1.70597181e-01 5.28351843e-01 -4.65888798e-01 -2.22983688e-01 7.89807379e-01 2.76045710e-01 1.28052488e-01 1.11845934e+00 -3.50110263e-01 -1.41258454e+00 -3.93562764e-01 -1.08881310e-01 1.45912573e-01 -5.27115703e-01 -7.64413834e-01 -2.06954375e-01 -7.30531812e-01 3.02626621e-02 -8.75230655e-02 6.70515418e-01 3.00493300e-01 1.50636685e+00 -1.00953400e+00 -1.74868748e-01 7.39163041e-01 1.93820298e+00 2.68238395e-01 6.70732617e-01 2.34532908e-01 1.06716909e-01 2.74956763e-01 6.93224907e-01 8.87709498e-01 4.02385294e-01 4.60626990e-01 8.98828804e-01 3.64579052e-01 7.96450138e-01 2.94730753e-01 1.67391315e-01 3.83745402e-01 -7.01871738e-02 -5.19806981e-01 -3.60122263e-01 8.62733722e-01 -1.63891113e+00 -9.45727468e-01 -1.21318400e-01 2.71575284e+00 9.21245635e-01 2.57992834e-01 3.92125309e-01 2.67808974e-01 4.51858521e-01 -4.05358933e-02 -3.94046634e-01 -9.05409396e-01 -3.65704209e-01 7.45272696e-01 1.33098710e+00 6.33807719e-01 -4.53446060e-01 6.43924654e-01 4.71090031e+00 9.54777181e-01 -7.03319132e-01 1.50083378e-01 6.39201581e-01 -8.04159999e-01 -9.36216176e-01 2.02057332e-01 -8.10068905e-01 5.43753684e-01 8.37037444e-01 -4.70683903e-01 9.97840703e-01 9.92595673e-01 -2.63446540e-01 -6.22819304e-01 -1.08493316e+00 1.07954431e+00 1.48278102e-01 -1.35244429e+00 -3.01613182e-01 8.24182153e-01 8.01901519e-01 -3.86180937e-01 2.61759579e-01 2.39665098e-02 3.37886870e-01 -1.07262993e+00 5.92080057e-01 -2.43051186e-01 1.06991351e+00 -1.14448750e+00 5.70765555e-01 6.90771222e-01 -1.32227921e+00 -5.91950297e-01 -4.80970889e-01 9.98105630e-02 1.18860282e-01 2.95910746e-01 -4.54623997e-01 4.24499810e-01 9.73948896e-01 -7.77618825e-01 1.90186873e-01 9.76399958e-01 4.15455729e-01 1.43108502e-01 -9.96419013e-01 -1.24098174e-01 2.63741583e-01 -3.08175385e-01 4.91831630e-01 9.45398152e-01 3.36423606e-01 8.00706029e-01 2.40623191e-01 3.83140236e-01 -5.11699080e-01 2.99216002e-01 -3.51517349e-01 -1.87641844e-01 5.97790360e-01 8.01962197e-01 -7.41467059e-01 3.09883263e-02 -1.73890993e-01 7.25777626e-01 -8.44733045e-02 -1.43108517e-02 -8.76605690e-01 -5.64805090e-01 6.41568065e-01 6.26804948e-01 7.52356529e-01 -4.43849564e-01 -5.33585727e-01 -8.23389173e-01 4.02727306e-01 -7.56117284e-01 1.05059314e+00 3.66138518e-02 -6.45177841e-01 2.20306799e-01 2.31024280e-01 -4.99085635e-01 2.22726822e-01 -4.85905707e-01 -4.00967486e-02 6.34000242e-01 -1.20640481e+00 -6.69843197e-01 3.61134149e-02 8.74305427e-01 2.80195028e-02 3.12307011e-02 5.76009929e-01 1.87358528e-01 -2.23755627e-03 9.70340073e-01 3.49169105e-01 -4.33905691e-01 -5.10058366e-02 -1.31739843e+00 -4.24038321e-01 7.77778327e-01 -2.22912561e-02 3.89957994e-01 8.00857246e-01 -2.93540537e-01 -2.03891087e+00 -7.25803077e-01 9.79033351e-01 -5.79014346e-02 3.68088573e-01 -3.55268538e-01 -3.66476297e-01 7.79414773e-01 -3.56728099e-02 -1.85263287e-02 8.25173497e-01 -1.68193191e-01 -4.25877035e-01 -7.15892494e-01 -1.57814431e+00 4.22203273e-01 1.33610129e+00 -2.25813478e-01 2.46540189e-01 4.67633098e-01 8.62775147e-01 -4.83025283e-01 -9.05140042e-01 4.38753277e-01 3.11957091e-01 -6.89609349e-01 5.91367424e-01 -4.09999758e-01 -1.97482824e-01 -2.37063542e-01 -7.31215417e-01 -3.84835422e-01 4.52525429e-02 -1.05693841e+00 -2.79368967e-01 3.93828869e-01 7.79552877e-01 -5.65910697e-01 1.22085941e+00 6.58909976e-01 2.16251060e-01 -1.07679260e+00 -1.40503967e+00 -7.12512970e-01 3.65520477e-01 -5.44590652e-01 6.51448667e-01 7.39870071e-01 1.46715492e-01 -3.67662966e-01 -2.72937775e-01 2.22023219e-01 6.13907397e-01 5.21555722e-01 8.70420873e-01 -7.08934486e-01 -1.09366679e+00 -2.56348372e-01 -1.31590068e-01 -1.06269062e+00 -4.16609436e-01 -9.82482910e-01 -2.82429367e-01 -1.13474524e+00 5.85902035e-01 -9.37572479e-01 -1.31505713e-01 4.41615254e-01 4.68526304e-01 -1.34136409e-01 4.88027573e-01 -2.16165856e-01 -9.14149284e-01 -2.31775902e-02 1.04185367e+00 -1.60006151e-01 -1.89172238e-01 3.46069038e-01 -9.56998289e-01 2.78475374e-01 4.39190388e-01 -5.12967229e-01 -4.34826523e-01 -5.92561722e-01 1.00172246e+00 7.92203486e-01 -6.79321960e-02 -3.55051458e-01 3.35469872e-01 -7.10715115e-01 -4.38863367e-01 -6.64695084e-01 4.50746238e-01 -9.15257454e-01 4.47358072e-01 7.99486995e-01 -1.51242524e-01 5.08814156e-01 3.37799564e-02 6.05711281e-01 1.98944122e-01 -5.83513737e-01 8.16923261e-01 -4.79204595e-01 -1.25416711e-01 7.25029111e-01 -1.00861125e-01 6.63287416e-02 1.54647005e+00 -4.96483415e-01 -4.88201469e-01 -6.16741478e-01 -6.70852244e-01 3.87987614e-01 4.87280875e-01 -3.45728368e-01 4.55120504e-01 -6.27728999e-01 -4.93075371e-01 -9.66690555e-02 -2.70031039e-02 4.61982727e-01 4.98082072e-01 9.35911000e-01 -6.60755455e-01 4.61136341e-01 2.84146339e-01 -3.00898343e-01 -1.28860986e+00 8.71252835e-01 2.90339381e-01 -5.67885399e-01 8.61941427e-02 1.33083451e+00 -2.48751730e-01 4.14009718e-03 5.16982377e-01 -2.69195549e-02 8.08251262e-01 -1.36811156e-02 2.85741538e-01 5.94617844e-01 3.58502865e-02 -1.24229416e-01 -1.71500042e-01 1.42270774e-01 -3.48132849e-01 -5.78910947e-01 1.44621122e+00 -1.87330216e-01 -6.68146908e-01 -9.14624482e-02 1.07287180e+00 2.48007730e-01 -5.23433805e-01 -5.01757562e-01 -2.47056469e-01 -7.13891864e-01 -4.70329344e-01 -4.33275998e-01 -1.41551089e+00 3.90525609e-01 4.61251408e-01 3.80097091e-01 1.37816274e+00 3.26894313e-01 8.08822095e-01 5.90049326e-01 1.19292295e+00 -1.37080467e+00 -2.81453896e-02 2.51367033e-01 5.68207860e-01 -6.29707873e-01 1.47875533e-01 -4.82842445e-01 -2.54450411e-01 8.36497962e-01 5.95225632e-01 1.46303013e-01 7.03601837e-01 4.59801614e-01 -5.84086478e-01 -2.12298527e-01 -8.61821592e-01 -1.51951283e-01 -4.72501963e-01 -1.81552649e-01 -1.18953943e-01 2.20668152e-01 -1.00448430e+00 7.55294979e-01 -4.10076171e-01 -1.82765007e-01 7.64633060e-01 1.37228143e+00 -8.80386651e-01 -1.48951888e+00 -4.14463758e-01 6.15197301e-01 -7.99569011e-01 -6.37872443e-02 -1.27161831e-01 5.53708553e-01 4.38370854e-01 9.08495724e-01 -6.42615780e-02 -1.17029347e-01 8.90911072e-02 -1.55183688e-01 8.24404895e-01 -6.25533760e-01 -5.82383931e-01 1.56428561e-01 -1.13732353e-01 -3.16133231e-01 1.34234697e-01 -5.42113006e-01 -1.32415962e+00 -8.33701789e-01 -5.49224138e-01 4.16544020e-01 9.06110764e-01 6.64005280e-01 5.87863624e-02 -2.73863405e-01 9.97890532e-01 2.84962028e-01 -1.05175424e+00 -4.31352317e-01 -1.16426337e+00 5.23791052e-02 -2.38846868e-01 -3.81151438e-02 -5.66919744e-01 -3.07377189e-01]

[6.567547798156738, 4.8372273445129395]

9a5f5a48-f221-49a0-b53c-b908f7d19261

bangla-handwritten-digit-recognition-and

2103.07905

https://arxiv.org/abs/2103.07905v1

https://arxiv.org/pdf/2103.07905v1.pdf

Bangla Handwritten Digit Recognition and Generation

Handwritten digit or numeral recognition is one of the classical issues in the area of pattern recognition and has seen tremendous advancement because of the recent wide availability of computing resources. Plentiful works have already done on English, Arabic, Chinese, Japanese handwritten script. Some work on Bangla also have been done but there is space for development. From that angle, in this paper, an architecture has been implemented which achieved the validation accuracy of 99.44% on BHAND dataset and outperforms Alexnet and Inception V3 architecture. Beside digit recognition, digit generation is another field which has recently caught the attention of the researchers though not many works have been done in this field especially on Bangla. In this paper, a Semi-Supervised Generative Adversarial Network or SGAN has been applied to generate Bangla handwritten numerals and it successfully generated Bangla digits.

['Md Fahim Sikder']

2021-03-14

['handwritten-digit-recognition']

['computer-vision']

[ 8.77235979e-02 -1.58537850e-01 4.42681164e-01 -3.56546760e-01 -1.95169643e-01 -4.54656035e-01 8.46781194e-01 -4.32687253e-01 -3.05218786e-01 1.07483280e+00 9.22082439e-02 -3.93155485e-01 3.84624712e-02 -9.70721960e-01 -2.46869639e-01 -7.01258481e-01 3.19958240e-01 4.15261954e-01 1.12574674e-01 -3.73765320e-01 5.37022352e-01 9.12105083e-01 -9.94650364e-01 2.47856542e-01 4.44585830e-01 4.17556852e-01 -1.27406001e-01 9.67459202e-01 -3.75439562e-02 1.30340111e+00 -9.44204092e-01 -7.31514513e-01 5.62181354e-01 -9.62616324e-01 -6.59154058e-01 2.66965240e-01 1.08265541e-01 -3.88417155e-01 -7.23261058e-01 7.38172531e-01 9.36724007e-01 -7.31545761e-02 8.15727651e-01 -1.05431259e+00 -1.09090745e+00 7.72418320e-01 -7.72412002e-01 1.38178915e-01 -5.05766720e-02 8.37524831e-02 2.30884939e-01 -7.63517559e-01 4.05932128e-01 1.03982043e+00 6.04382873e-01 5.56888223e-01 -6.02918029e-01 -5.80662310e-01 -7.54271746e-01 1.83482334e-01 -1.30775249e+00 1.41589493e-01 9.42614615e-01 -1.98296025e-01 1.06193376e+00 3.40909213e-02 4.76888418e-01 7.22183406e-01 4.65106308e-01 6.72631085e-01 1.56765914e+00 -6.75659239e-01 1.11845106e-01 2.42173851e-01 9.86638665e-02 4.21745569e-01 2.59057581e-01 1.34342298e-01 1.25493724e-02 6.19127154e-01 9.58996892e-01 -2.49616489e-01 2.51586705e-01 3.25659066e-01 -9.69313860e-01 8.64765048e-01 2.73666501e-01 7.01774597e-01 -4.33768034e-01 -9.99329891e-03 5.09949744e-01 5.11613905e-01 -9.67339426e-02 3.26952487e-01 -1.10341787e-01 -4.64526623e-01 -1.04350877e+00 2.77594417e-01 6.48439586e-01 9.23917770e-01 1.68874815e-01 9.31360543e-01 1.57736912e-01 8.63345563e-01 1.72512621e-01 5.23713291e-01 4.72202420e-01 -2.24017993e-01 4.52638805e-01 7.79089928e-01 -6.92368522e-02 -1.14870501e+00 -8.28565434e-02 -5.01505286e-02 -1.13553512e+00 8.25305223e-01 6.55577660e-01 -6.76671326e-01 -1.43676519e+00 8.08409691e-01 -1.30810589e-01 -2.41081938e-01 4.84298289e-01 1.02953756e+00 7.24621475e-01 1.19311833e+00 -5.70967495e-02 4.27527905e-01 1.12501490e+00 -8.47042024e-01 -6.43875062e-01 7.95937777e-02 -2.29684770e-01 -1.46387815e+00 5.66881001e-01 7.58202553e-01 -9.89283204e-01 -8.38231146e-01 -1.29697728e+00 1.19640782e-01 -6.58135831e-01 3.72659326e-01 7.57287204e-01 1.22643709e+00 -7.70201683e-01 3.27369273e-01 -5.95395029e-01 -6.34175360e-01 6.29244030e-01 5.06487072e-01 -2.39544392e-01 -2.57279783e-01 -8.89036953e-01 1.29200780e+00 6.17421865e-01 3.03732485e-01 -8.61538470e-01 1.31774306e-01 -3.08884203e-01 -3.75763893e-01 -2.69605577e-01 -1.28567293e-01 7.54657030e-01 -1.10081923e+00 -1.77196324e+00 9.41452026e-01 5.24910986e-01 -7.18676090e-01 7.80995488e-01 2.00497448e-01 -9.17298377e-01 -1.48058087e-01 -4.26444978e-01 8.32710087e-01 6.98265135e-01 -6.29317045e-01 -6.01276517e-01 -5.30120492e-01 -1.95022672e-01 1.02186203e-01 -1.07197449e-01 3.99340004e-01 1.56230152e-01 -1.10210371e+00 -6.18569292e-02 -7.90527523e-01 1.40950372e-02 -4.74964947e-01 -4.58303720e-01 -1.07852787e-01 1.17492533e+00 -9.89175200e-01 8.36787939e-01 -1.82766104e+00 -1.19133465e-01 2.68935561e-01 -5.39548874e-01 7.56044269e-01 4.42132466e-02 8.64110708e-01 1.56280436e-02 -6.56405911e-02 -3.75172317e-01 3.14831823e-01 -1.65119439e-01 2.68675715e-01 -3.43606293e-01 4.14954424e-01 4.81653869e-01 7.19620764e-01 -3.70428443e-01 -2.23217264e-01 4.51513141e-01 7.20465899e-01 -9.75017995e-02 -2.74426062e-02 6.84988871e-02 2.60704547e-01 -2.45310277e-01 9.81591046e-01 9.66890216e-01 3.51128370e-01 -2.55803764e-01 3.07900854e-03 -2.06208691e-01 -2.62264371e-01 -1.32597303e+00 1.12587500e+00 -3.55321288e-01 9.41028059e-01 -4.92498219e-01 -1.07082307e+00 1.63395035e+00 3.67050946e-01 2.59891525e-02 -4.90434617e-01 4.26832825e-01 6.19698584e-01 5.03067553e-01 -4.34340239e-01 6.14873648e-01 -2.19181612e-01 1.52191073e-01 4.11055565e-01 3.41355205e-02 -3.78318578e-01 2.99045801e-01 -2.37497613e-01 7.81700969e-01 5.15889466e-01 3.28897268e-01 1.12032235e-01 8.66065741e-01 5.68716228e-01 -4.88784462e-02 7.05917835e-01 -7.15047196e-02 7.84171999e-01 6.92629889e-02 -6.36449993e-01 -1.71425462e+00 -6.29834652e-01 -3.13722074e-01 3.04094881e-01 -3.47520411e-01 4.10582215e-01 -6.40300632e-01 -2.94827163e-01 -2.12519079e-01 3.61450881e-01 -3.66527557e-01 2.03577876e-01 -8.98939073e-01 -8.49657059e-01 1.15729463e+00 7.34161735e-01 1.34841681e+00 -1.78473377e+00 -4.57567930e-01 5.41585445e-01 6.37283862e-01 -8.51345599e-01 1.76935986e-01 -6.30726991e-03 -8.45724583e-01 -7.93816447e-01 -1.30986619e+00 -1.24788988e+00 6.72073305e-01 -1.89924479e-01 5.67010880e-01 -4.08537775e-01 -8.17449331e-01 -1.45279139e-01 -6.35221958e-01 -8.91368568e-01 -7.67894983e-01 -9.14721787e-02 -4.16792840e-01 -1.64731041e-01 7.08099723e-01 -2.95610309e-01 -2.84010351e-01 1.80408776e-01 -8.35068464e-01 -1.02727331e-01 9.22033429e-01 7.28399396e-01 1.42981112e-01 3.53398085e-01 9.15459871e-01 -8.98284674e-01 6.59344256e-01 -3.08905184e-01 -7.27044344e-01 -2.95647271e-02 -1.68757677e-01 -1.48486987e-01 9.49706316e-01 -1.97956488e-01 -9.58302200e-01 4.05589640e-02 -4.88350600e-01 2.65934348e-01 -6.03673935e-01 5.02880931e-01 6.78231120e-02 -2.12320685e-01 6.65468454e-01 4.83470827e-01 2.48584338e-02 -3.03424269e-01 -1.05792470e-02 1.27157950e+00 5.91661096e-01 -2.16710716e-01 8.59609902e-01 1.95102841e-01 3.96953337e-02 -8.90100360e-01 -1.59141347e-01 1.79632157e-02 -3.89520526e-01 -1.85313314e-01 9.27529395e-01 -5.52661657e-01 -2.93425560e-01 1.35728872e+00 -1.16820657e+00 1.83978733e-02 4.19631116e-02 5.23551524e-01 -1.51276931e-01 1.53679863e-01 -6.17947221e-01 -9.22552049e-01 -4.97160614e-01 -1.01971936e+00 2.22819597e-01 4.61685956e-01 -3.11493501e-02 -8.64711404e-01 -6.77410066e-02 2.93640256e-01 7.42480695e-01 5.56445003e-01 6.24333203e-01 -5.90978980e-01 -5.45650244e-01 -7.93755352e-01 -1.82942793e-01 7.96321034e-01 4.81281489e-01 2.97741324e-01 -6.67482376e-01 -4.08645049e-02 -1.66997194e-01 -3.78772825e-01 5.45336068e-01 5.40201701e-02 7.11160779e-01 -1.78811178e-01 3.68793964e-01 3.37509125e-01 1.60050774e+00 8.73614550e-01 1.27178335e+00 5.53833067e-01 4.81869340e-01 5.42695373e-02 3.81870568e-01 5.36238611e-01 -1.90111488e-01 3.37812811e-01 1.68030307e-01 8.69072750e-02 -4.83745009e-01 -1.15528822e-01 3.06526572e-01 5.97919703e-01 -5.10380149e-01 -3.89654130e-01 -9.78606641e-01 4.28233027e-01 -1.06019819e+00 -1.12184250e+00 -4.47677732e-01 1.80271161e+00 8.52613807e-01 1.07920185e-01 1.75455809e-01 8.79142582e-01 8.26891601e-01 5.70487510e-03 -2.75023162e-01 -9.62929904e-01 -3.64661515e-01 7.94438064e-01 3.76560301e-01 1.44501075e-01 -1.08214831e+00 9.91501570e-01 5.25404072e+00 4.82409716e-01 -1.60030866e+00 -4.43307340e-01 7.19726562e-01 6.38909578e-01 3.65813136e-01 -1.70274913e-01 -8.28735292e-01 5.75664282e-01 6.77832723e-01 3.75393890e-02 3.74324322e-01 8.96213889e-01 -3.16506401e-02 -8.52686986e-02 -4.83029604e-01 8.14423144e-01 2.57016569e-01 -1.03191698e+00 2.83795595e-01 -8.38718489e-02 1.09821618e+00 -4.72136468e-01 1.52524248e-01 2.17201188e-01 4.33725089e-01 -1.42888296e+00 1.65568963e-01 3.10185641e-01 6.60531521e-01 -1.18431091e+00 1.23945844e+00 2.05566064e-01 -5.65800726e-01 -4.37101200e-02 -7.65193999e-01 -1.73829287e-01 -2.36668047e-02 2.05891013e-01 -1.17764485e+00 3.01817864e-01 4.27294612e-01 5.56302011e-01 -5.89299679e-01 1.26513875e+00 -3.21739644e-01 9.27351296e-01 -1.63176030e-01 -4.68059570e-01 7.67336190e-01 -4.17674035e-01 2.04263493e-01 1.35928702e+00 5.96833885e-01 2.04961196e-01 -4.80158210e-01 5.61027706e-01 2.47023664e-02 3.39923292e-01 -6.57664895e-01 -4.14596051e-01 2.13003263e-01 1.14239740e+00 -8.71841431e-01 -2.99393296e-01 -2.71853685e-01 1.03605294e+00 -4.28323567e-01 1.57453772e-02 -9.27501023e-01 -9.93284225e-01 -1.99199468e-01 -4.60689962e-02 3.81692618e-01 -3.80826920e-01 -5.07803500e-01 -6.55197442e-01 -2.89008439e-01 -1.32562792e+00 2.55905509e-01 -7.43372202e-01 -1.10324359e+00 7.32940137e-01 -4.44358915e-01 -1.17314148e+00 -1.88570768e-01 -1.05350125e+00 -8.32045794e-01 1.11821759e+00 -1.02290964e+00 -1.33752084e+00 -5.32606959e-01 4.82269794e-01 7.56941020e-01 -1.05334532e+00 8.06447685e-01 4.08268392e-01 -5.96663833e-01 6.18557334e-01 2.89296597e-01 7.09711134e-01 5.48496962e-01 -1.10355306e+00 2.96495318e-01 1.14402843e+00 2.32696488e-01 3.91492993e-01 6.50202692e-01 -5.48039258e-01 -1.31275129e+00 -9.92976427e-01 8.53113055e-01 -1.12193346e-01 3.18409741e-01 1.91924006e-01 -6.44766212e-01 5.68407536e-01 8.12565148e-01 -2.90261567e-01 3.61116737e-01 -8.41062367e-01 1.27293065e-01 -1.97328210e-01 -1.49123216e+00 4.99261409e-01 2.31159016e-01 6.26290729e-03 -6.42995954e-01 8.95619094e-02 -3.02418560e-01 -4.46425885e-01 -7.12031424e-01 5.44171706e-02 4.26362127e-01 -9.75397110e-01 6.55430079e-01 -5.54227233e-01 8.79038334e-01 -5.11002719e-01 -2.46490389e-02 -9.28889573e-01 8.47627968e-02 -5.49692929e-01 2.18010917e-01 1.57869077e+00 3.19856197e-01 -5.17010748e-01 9.12459731e-01 3.48017395e-01 -1.72818284e-02 -7.09954143e-01 -4.05023992e-01 -5.86914659e-01 3.56879443e-01 1.12794265e-01 5.07376671e-01 9.13596392e-01 -6.50403500e-01 2.12475941e-01 -7.83673048e-01 -1.47515774e-01 4.29308772e-01 1.38019741e-01 8.23751926e-01 -7.83712745e-01 -2.89226443e-01 -2.04294443e-01 -9.35393751e-01 -5.77851832e-01 -4.62204933e-01 -8.47206056e-01 -4.13537025e-01 -1.67528355e+00 -4.52824943e-02 -2.46839076e-01 -5.20092919e-02 5.78335941e-01 2.38317192e-01 7.59407461e-01 2.63132930e-01 -3.20152342e-02 2.39222959e-01 -1.00361146e-01 1.44339156e+00 -1.91669032e-01 3.66804674e-02 1.20654993e-01 -4.69230950e-01 3.62688422e-01 1.31169450e+00 -2.84386188e-01 -1.58797130e-01 -3.20117563e-01 -3.77941579e-02 -9.27676260e-02 2.16684669e-01 -1.37808621e+00 8.15919861e-02 3.04659270e-02 9.96732712e-01 -7.96827257e-01 8.03298056e-02 -7.78863430e-01 2.40002230e-01 5.97979486e-01 -1.78288817e-01 2.98251837e-01 5.40770181e-02 1.53861416e-03 -5.10606110e-01 -4.49607819e-01 1.20482898e+00 -3.57110202e-01 -1.03219557e+00 6.34371489e-02 -5.71868837e-01 -1.98000893e-01 1.30783784e+00 -7.91564405e-01 -1.66636065e-01 -1.04862534e-01 -5.08377314e-01 -4.03179467e-01 1.32595748e-01 4.30510014e-01 7.71411598e-01 -1.25040483e+00 -1.24449515e+00 1.10353127e-01 -3.47629666e-01 -3.74481499e-01 5.69506809e-02 3.08933109e-01 -1.47906327e+00 6.24439597e-01 -1.16727591e+00 2.32105479e-02 -1.23844910e+00 2.02943459e-01 1.86028451e-01 -2.03723125e-02 -5.06396532e-01 6.52076185e-01 -1.05970037e+00 -2.06930101e-01 1.24289654e-01 1.83590561e-01 -5.28224409e-01 -1.08288966e-01 3.40554655e-01 5.96807182e-01 3.68032493e-02 -6.10834181e-01 -1.17279023e-01 6.33438349e-01 -2.54015684e-01 -4.82197143e-02 1.50596380e+00 5.67113280e-01 -1.54049516e-01 4.43770327e-02 1.00783217e+00 -2.70251967e-02 -9.92046118e-01 3.35248560e-01 -6.93257079e-02 -5.08464694e-01 -2.57367998e-01 -1.35667610e+00 -1.15495074e+00 1.13320971e+00 7.61628568e-01 5.32530732e-02 1.05654669e+00 -6.91840291e-01 7.13948429e-01 5.40501952e-01 4.02035713e-01 -1.21991956e+00 1.80680588e-01 6.72984004e-01 1.02671802e+00 -1.24350631e+00 -5.42173907e-02 1.69878885e-01 -8.59801233e-01 1.59886181e+00 6.16806269e-01 -6.97686374e-01 2.66153753e-01 6.92950189e-01 3.87993217e-01 2.49383226e-01 6.93109930e-02 2.10969150e-01 -1.36884063e-01 7.97410965e-01 9.44405437e-01 -9.82911959e-02 -7.80777514e-01 2.98463672e-01 -3.87720525e-01 4.00315821e-01 9.64951873e-01 1.16443372e+00 -2.96898991e-01 -1.43367982e+00 -5.76883018e-01 4.30857211e-01 -9.50023592e-01 -9.39242095e-02 -5.07076323e-01 1.17247355e+00 5.84579073e-03 7.46604383e-01 -2.08511218e-01 -2.55180150e-01 1.28849909e-01 7.86907896e-02 7.75529683e-01 -3.32594514e-01 -8.10725331e-01 -5.49836218e-01 -3.80608998e-02 3.28184575e-01 -4.16535765e-01 -4.85822618e-01 -1.29183161e+00 -7.07581103e-01 -7.54212290e-02 -5.34073226e-02 1.04841173e+00 4.83597904e-01 -4.29139659e-02 4.07468081e-01 5.81602275e-01 -3.30041856e-01 -7.18440175e-01 -1.16568351e+00 -6.34981573e-01 1.57441765e-01 -2.15398297e-01 -4.23105657e-02 1.44049957e-01 2.00686038e-01]

[11.849250793457031, 2.677915573120117]

2e3bf60e-da47-4002-843c-5b1d40a4bff9

generalized-approach-to-matched-filtering

2104.03961

https://arxiv.org/abs/2104.03961v3

https://arxiv.org/pdf/2104.03961v3.pdf

Generalized Approach to Matched Filtering using Neural Networks

Gravitational wave science is a pioneering field with rapidly evolving data analysis methodology currently assimilating and inventing deep learning techniques. The bulk of the sophisticated flagship searches of the field rely on the time-tested matched filtering principle within their core. In this paper, we make a key observation on the relationship between the emerging deep learning and the traditional techniques: matched filtering is formally equivalent to a particular neural network. This means that a neural network can be constructed analytically to exactly implement matched filtering, and can be further trained on data or boosted with additional complexity for improved performance. Moreover, we show that the proposed neural network architecture can outperform matched filtering, both with or without knowledge of a prior on the parameter distribution. When a prior is given, the proposed neural network can approach the statistically optimal performance. We also propose and investigate two different neural network architectures MNet-Shallow and MNet-Deep, both of which implement matched filtering at initialization and can be trained on data. MNet-Shallow has simpler structure, while MNet-Deep is more flexible and can deal with a wider range of distributions. Our theoretical findings are corroborated by experiments using real LIGO data and synthetic injections, where our proposed methods significantly outperform matched filtering at false positive rates above $5\times 10^{-3}\%$. The fundamental equivalence between matched filtering and neural networks allows us to define a "complexity standard candle" to characterize the relative complexity of the different approaches to gravitational wave signal searches in a common framework. Finally, our results suggest new perspectives on the role of deep learning in gravitational wave detection.

['Szabolcs Márka', 'Zsuzsa Márka', 'John Wright', 'Imre Bartos', 'Doğa Veske', 'Robert E. Colgan', 'Mariam Avagyan', 'Jingkai Yan']

2021-04-08

['gravitational-wave-detection']

['miscellaneous']

[-3.99758518e-02 -1.02884814e-01 1.74895689e-01 -3.64561439e-01 -4.48565423e-01 -6.04058683e-01 8.28747690e-01 -4.16303515e-01 -5.40268362e-01 4.45942283e-01 -1.18682414e-01 -5.02589941e-01 -5.88043928e-01 -1.03280163e+00 -6.53612614e-01 -9.70774233e-01 -2.69884467e-01 6.08933091e-01 3.65446478e-01 -4.75166559e-01 2.84684867e-01 5.41255414e-01 -1.33055878e+00 -2.17839822e-01 6.08849406e-01 1.49156487e+00 8.75382721e-02 6.95104599e-01 4.79349978e-02 2.44091004e-01 -5.33653736e-01 -5.71826935e-01 8.38044822e-01 -4.34237003e-01 -3.67981434e-01 -6.50563240e-01 3.88921380e-01 -7.17692226e-02 -7.48392105e-01 1.29774356e+00 9.76130247e-01 2.20689505e-01 3.81633461e-01 -7.55708933e-01 -4.00238037e-01 6.43216789e-01 -8.38852152e-02 5.76546073e-01 -1.94288418e-01 4.41608489e-01 9.02783155e-01 -7.78290749e-01 2.14238986e-01 7.02137053e-01 8.35974753e-01 3.97818536e-01 -1.04767394e+00 -7.89217234e-01 -2.48959959e-01 1.51762038e-01 -9.92163777e-01 -3.60980421e-01 7.74409115e-01 -5.66279173e-01 7.08040893e-01 1.07873268e-01 6.49186909e-01 8.58914375e-01 1.96608201e-01 3.57364386e-01 1.18093812e+00 -7.01849222e-01 2.78101176e-01 -2.82340825e-01 2.49925986e-01 6.78044975e-01 4.90426213e-01 9.63452935e-01 -6.82747900e-01 -7.23375231e-02 7.49207377e-01 -3.32419634e-01 -4.88744557e-01 -2.15460539e-01 -9.01120007e-01 1.12041116e+00 5.18939376e-01 3.90343070e-01 2.27973815e-02 2.79749393e-01 1.25756949e-01 4.82160956e-01 3.86581689e-01 8.83675635e-01 -4.44903553e-01 2.01481417e-01 -1.12351656e+00 6.45721138e-01 7.54384935e-01 5.42503476e-01 7.78980851e-01 4.17852372e-01 -1.31402567e-01 4.43096042e-01 3.12245727e-01 6.83035970e-01 5.38433313e-01 -5.89675009e-01 1.72281876e-01 2.71449864e-01 9.33339596e-02 -7.18582988e-01 -8.03602993e-01 -1.29559410e+00 -8.32413733e-01 5.84319413e-01 8.16296875e-01 -3.54316711e-01 -9.20091152e-01 1.69182289e+00 1.71516314e-02 3.04175586e-01 1.35982841e-01 9.77897763e-01 7.33660340e-01 3.40021998e-01 -5.93954921e-01 9.62246805e-02 1.34490538e+00 -4.12285089e-01 -3.19182068e-01 -5.26941359e-01 2.75792837e-01 -7.39738822e-01 7.94182062e-01 4.75425541e-01 -1.08773339e+00 -4.62607384e-01 -1.27290237e+00 2.04937950e-01 -3.72606605e-01 -6.72428384e-02 8.23589444e-01 9.23497736e-01 -8.69481504e-01 9.24510121e-01 -6.54066205e-01 7.22178295e-02 1.75151110e-01 5.56665480e-01 2.43352741e-01 5.24432182e-01 -1.32340240e+00 7.00233161e-01 3.68246615e-01 2.10992187e-01 -6.59000874e-01 -7.99144983e-01 -4.11536455e-01 2.05559820e-01 1.35559723e-01 -7.52321303e-01 1.33461249e+00 -4.08672512e-01 -1.55777836e+00 6.75264180e-01 2.05270350e-01 -1.00324380e+00 4.31608319e-01 -2.60223728e-02 -5.70108652e-01 -3.13464589e-02 -3.04981530e-01 -4.44364585e-02 8.31160009e-01 -7.55727470e-01 -7.25679576e-01 -2.75855154e-01 2.14632899e-01 -4.01670128e-01 2.00574566e-02 4.95375060e-02 -1.87001318e-01 -7.76881039e-01 5.32581806e-01 -8.13584626e-01 -1.54331163e-01 -3.39417905e-01 6.87340647e-02 -8.10948610e-02 3.38026702e-01 -3.17113310e-01 8.56779099e-01 -1.95651090e+00 9.26344004e-03 3.31584811e-01 2.78804153e-01 4.17298317e-01 3.91556658e-02 2.43220076e-01 -1.05513722e-01 -7.98477605e-02 -1.53910235e-01 -4.30250093e-02 4.10347700e-01 -5.89221753e-02 -3.86048585e-01 6.71990335e-01 -1.24983221e-01 9.07167017e-01 -6.59552038e-01 3.60353023e-01 1.91452116e-01 2.72355616e-01 -7.67257810e-01 1.60640091e-01 -2.29543343e-01 6.20230138e-01 -1.57855272e-01 3.88798654e-01 8.88143718e-01 -1.28635690e-01 -1.56226188e-01 -3.53895456e-01 -3.81007791e-01 3.49953443e-01 -1.23799872e+00 1.38886082e+00 -3.72415900e-01 7.54168034e-01 3.05734277e-01 -1.47533286e+00 1.19592881e+00 1.05121378e-02 1.23935051e-01 -1.15581369e+00 5.18548250e-01 3.72225076e-01 4.18122470e-01 -4.19332266e-01 3.86132151e-01 -6.19680285e-01 -1.31367743e-01 3.31903934e-01 4.08402354e-01 -4.89768013e-03 1.15969643e-01 -2.29824215e-01 1.11152112e+00 -1.50512815e-01 -1.09467924e-01 -5.51281273e-01 4.14771408e-01 -3.03125739e-01 6.34141743e-01 1.43290544e+00 -1.00785747e-01 3.90080273e-01 2.57245451e-01 -7.91657984e-01 -8.04014981e-01 -1.04597187e+00 -4.22664613e-01 1.13365710e+00 1.20359883e-01 -1.16223469e-01 -3.42148393e-01 -3.32768232e-01 -1.70214772e-02 3.96938562e-01 -5.23683608e-01 -2.59559453e-01 -8.04827571e-01 -1.40032053e+00 5.87469518e-01 4.07263011e-01 6.13936901e-01 -9.32340443e-01 -5.78444362e-01 1.35177106e-01 1.67925864e-01 -8.05323243e-01 1.97131515e-01 5.63162982e-01 -5.82655549e-01 -1.06970704e+00 -4.30451006e-01 -6.78280950e-01 1.53673515e-01 -4.88181785e-02 1.23243928e+00 9.43513364e-02 -3.57671864e-02 -4.31050733e-02 -3.98767084e-01 -5.42935371e-01 -2.93281227e-01 2.96373572e-02 2.71337032e-01 4.48520072e-02 4.63838756e-01 -7.56217003e-01 -7.03736603e-01 2.27880016e-01 -6.28337145e-01 -3.12323391e-01 8.41306210e-01 9.83128905e-01 -2.22705547e-02 1.45935789e-01 5.97117543e-01 -5.04604816e-01 4.71699208e-01 -2.93701053e-01 -1.21299624e+00 -5.66547588e-02 -4.94921654e-01 5.60339570e-01 5.33479631e-01 -1.72826663e-01 -9.62488890e-01 -2.86824673e-01 -6.39544129e-01 -1.98098332e-01 5.66011854e-02 5.83393097e-01 5.82737699e-02 -6.72116101e-01 7.71328866e-01 2.84929067e-01 -1.63296878e-01 -7.29631603e-01 3.26136798e-01 3.39288652e-01 1.04486346e+00 -6.78281069e-01 1.13260841e+00 5.48552513e-01 2.18214214e-01 -5.35664797e-01 -8.32469106e-01 -2.76721567e-01 -4.71779108e-01 -1.30940214e-01 5.57070315e-01 -6.63412988e-01 -1.01878417e+00 4.36531961e-01 -9.24819469e-01 -1.64729267e-01 -2.84270823e-01 9.34337556e-01 -3.68984044e-01 1.70814425e-01 -4.22052234e-01 -8.41301024e-01 -3.65280569e-01 -9.83060658e-01 5.75959086e-01 4.56395417e-01 3.58179450e-01 -1.02910602e+00 1.87463045e-01 -8.52102712e-02 7.59186625e-01 1.27907544e-01 7.73521185e-01 -7.94291615e-01 -5.56831777e-01 -4.30866778e-01 -2.37590626e-01 1.55169636e-01 -2.95313150e-01 -4.91924703e-01 -1.15107751e+00 -5.23898125e-01 5.43825328e-01 -4.73274402e-02 1.28901482e+00 6.24639750e-01 7.75815248e-01 -7.02840313e-02 -6.72419518e-02 1.22540772e+00 1.47878730e+00 3.00056517e-01 3.68809044e-01 5.13781309e-01 7.80733898e-02 3.20082039e-01 2.26543266e-02 4.66746837e-01 -2.09954843e-01 5.40668726e-01 6.84712529e-01 -4.38236818e-02 -1.90067682e-02 1.12963624e-01 2.35823914e-02 6.40217304e-01 -2.40651861e-01 -5.66928573e-02 -7.07386732e-01 2.25926533e-01 -1.59494269e+00 -1.19037032e+00 -1.98745057e-01 2.55450678e+00 5.22091210e-01 4.94676590e-01 5.20880669e-02 2.11290017e-01 5.25021374e-01 8.09482411e-02 -3.70627850e-01 -3.93027849e-02 -3.46702605e-01 9.49617565e-01 6.52659476e-01 6.71451151e-01 -1.13280654e+00 4.86149997e-01 6.72778082e+00 7.09530771e-01 -1.34262061e+00 1.79005384e-01 1.18233494e-01 -1.41885743e-01 -2.06334919e-01 4.59426381e-02 -8.99944723e-01 5.57677567e-01 8.15075874e-01 -3.00233215e-01 4.20664668e-01 6.32401526e-01 3.63757163e-02 1.49655178e-01 -9.40585315e-01 1.03728902e+00 -1.48114413e-01 -1.86434293e+00 -3.71094763e-01 2.42533423e-02 4.61635292e-01 5.50639391e-01 1.08089661e-02 5.45315564e-01 3.77897382e-01 -7.02049971e-01 7.97891736e-01 6.14541590e-01 3.69124621e-01 -6.63489103e-01 9.67413485e-01 4.38542157e-01 -1.03651452e+00 -1.64869517e-01 -6.78796589e-01 -3.94378275e-01 1.15536436e-01 8.04533005e-01 -5.32035470e-01 8.19653332e-01 7.67127872e-01 1.54077888e-01 -5.55388272e-01 1.52035594e+00 -8.22708905e-02 7.92854905e-01 -6.02668643e-01 -1.22966908e-01 3.00495267e-01 -2.30134740e-01 7.31348932e-01 1.19581974e+00 5.60021281e-01 -1.27740443e-01 2.88984515e-02 9.76234496e-01 -3.27115762e-03 -2.41535559e-01 -2.96154082e-01 1.13887683e-01 2.62665123e-01 1.27714789e+00 -7.44595051e-01 -2.65046537e-01 -6.21214926e-01 2.27359891e-01 1.08168595e-01 2.92738885e-01 -7.35364258e-01 -5.20714998e-01 5.56236625e-01 4.20312062e-02 5.51045358e-01 -4.23861623e-01 -3.17368090e-01 -1.20766580e+00 -1.38661608e-01 -5.12182474e-01 4.51723129e-01 -2.71733761e-01 -1.39649200e+00 7.15607524e-01 -1.40421227e-01 -1.20197809e+00 -2.34079972e-01 -1.14737201e+00 -8.66251945e-01 9.82347667e-01 -1.60979366e+00 -6.43148661e-01 -3.61023933e-01 4.04038817e-01 -2.40834635e-02 -5.84660053e-01 4.83276188e-01 5.03784359e-01 -3.92712444e-01 4.75476384e-01 4.18528587e-01 2.74643838e-01 3.88007790e-01 -1.28162706e+00 6.61408246e-01 1.18841887e+00 3.58149141e-01 5.94605088e-01 1.05794156e+00 -3.81875932e-01 -1.46373940e+00 -7.56022692e-01 4.73578781e-01 -1.85461611e-01 8.78569961e-01 -7.45010495e-01 -8.05232823e-01 4.41326261e-01 1.92435175e-01 3.96632791e-01 3.72355282e-01 1.65984556e-01 -2.51020432e-01 -1.31366760e-01 -1.07109928e+00 1.10001013e-01 1.01125062e+00 -3.20393711e-01 -8.43412340e-01 4.06503230e-01 4.52818274e-01 -4.09148037e-01 -4.79100853e-01 7.25982487e-01 5.12345135e-01 -1.37631798e+00 9.97000158e-01 -4.71958518e-01 -2.19725445e-01 -2.81526238e-01 -2.45636269e-01 -1.30454350e+00 -4.76947755e-01 -1.00904453e+00 -9.87458602e-02 7.42917776e-01 1.97593078e-01 -1.01742303e+00 1.12896430e+00 4.61308248e-02 -4.49753374e-01 -6.31640017e-01 -1.15611124e+00 -1.13476408e+00 5.85920691e-01 -6.98563516e-01 6.44457161e-01 6.40657961e-01 -3.54285359e-01 4.48049195e-02 -4.12885964e-01 6.96264088e-01 8.80463600e-01 3.61246973e-01 6.11596286e-01 -1.61586106e+00 -8.56148303e-01 -7.11452127e-01 -4.08128053e-01 -1.22062886e+00 -1.10169925e-01 -1.02917933e+00 -2.25174353e-02 -1.09831119e+00 -4.50131446e-01 -5.38937867e-01 -2.49730036e-01 1.78290769e-01 1.59587711e-01 5.27681231e-01 -4.74006832e-02 2.72276066e-02 2.83171665e-02 4.93908376e-01 8.23544443e-01 -8.30397531e-02 -1.12296687e-02 3.81982148e-01 -4.78451461e-01 1.04017270e+00 7.65065253e-01 -3.18063766e-01 -7.66715929e-02 -4.58302408e-01 7.22686410e-01 -9.77752730e-02 5.03580034e-01 -1.39690030e+00 4.96829569e-01 1.30023181e-01 5.12304962e-01 -5.22638381e-01 2.10204408e-01 -3.70182872e-01 -9.85516831e-02 6.00342095e-01 1.10839657e-01 -1.07013419e-01 1.46436676e-01 2.61238575e-01 -2.69085243e-02 -6.65330291e-01 1.00550854e+00 -2.22854182e-01 -6.29916489e-01 2.51717269e-01 -2.43132785e-01 2.86702514e-01 4.98982251e-01 4.43897285e-02 -3.78862143e-01 -2.61592478e-01 -1.01202655e+00 -5.32343015e-02 -5.52487634e-02 7.59779215e-02 2.60530442e-01 -1.24140573e+00 -7.42220938e-01 7.19472528e-01 -2.43939370e-01 -3.84386539e-01 4.61319461e-02 7.06948400e-01 -5.10910034e-01 4.34373230e-01 -4.28699180e-02 -5.37111342e-01 -6.37344241e-01 3.52729708e-01 8.60701501e-01 -4.50968882e-03 -6.50483608e-01 1.06644166e+00 1.53432623e-01 -4.41753507e-01 1.97929233e-01 -5.35518587e-01 1.63596496e-01 -8.43431577e-02 6.09992802e-01 1.63417563e-01 4.54726279e-01 -2.98794568e-01 -2.22846866e-01 6.54975176e-01 1.88221619e-01 -1.96524821e-02 1.42326701e+00 2.09752768e-01 4.24980186e-02 -5.53975776e-02 7.09143102e-01 1.76701427e-01 -1.06383002e+00 -4.75207686e-01 9.40171331e-02 -3.91111642e-01 3.20358098e-01 -8.99143696e-01 -1.22237027e+00 7.77129710e-01 7.94256330e-01 4.89162862e-01 1.03037047e+00 1.79999739e-01 5.28709650e-01 6.16144955e-01 4.69662577e-01 -7.30299473e-01 -6.66867569e-03 6.64172590e-01 7.65152335e-01 -1.06184471e+00 -1.04592755e-01 2.73490455e-02 3.05303931e-01 1.22871614e+00 3.81716669e-01 -4.89552170e-01 8.13041747e-01 6.06776178e-01 -7.85168856e-02 -4.53631043e-01 -4.12733853e-01 -5.01512825e-01 4.25912410e-01 3.76130611e-01 2.46892840e-01 -2.95493454e-01 -4.33633566e-01 9.72089231e-01 -6.36767566e-01 -2.35299110e-01 3.94238025e-01 5.92425406e-01 -8.10242951e-01 -1.13576436e+00 -4.46543545e-01 4.02673095e-01 -5.05577981e-01 -1.16290726e-01 -5.37289940e-02 8.98171008e-01 4.34124529e-01 7.24122822e-01 1.57039732e-01 -4.50252742e-01 2.17231706e-01 1.93677679e-01 7.20379174e-01 -2.66841561e-01 -6.31984413e-01 1.27247974e-01 -1.17070548e-01 -3.29744607e-01 -3.61492753e-01 -4.36214477e-01 -9.69246149e-01 -2.68133074e-01 -5.96465290e-01 4.94812816e-01 6.54280365e-01 1.24173975e+00 5.87094277e-02 4.74168390e-01 4.43726152e-01 -1.09622872e+00 -7.22207844e-01 -9.56278086e-01 -5.62469125e-01 3.40208113e-02 5.10516167e-01 -9.11666095e-01 -9.01102304e-01 -4.93543267e-01]

[7.589334487915039, 3.1244821548461914]

f69739d7-fab2-49ac-ba65-3411a3fe22ac

how-well-do-self-supervised-models-transfer

2011.13377

https://arxiv.org/abs/2011.13377v2

https://arxiv.org/pdf/2011.13377v2.pdf

How Well Do Self-Supervised Models Transfer?

Self-supervised visual representation learning has seen huge progress recently, but no large scale evaluation has compared the many models now available. We evaluate the transfer performance of 13 top self-supervised models on 40 downstream tasks, including many-shot and few-shot recognition, object detection, and dense prediction. We compare their performance to a supervised baseline and show that on most tasks the best self-supervised models outperform supervision, confirming the recently observed trend in the literature. We find ImageNet Top-1 accuracy to be highly correlated with transfer to many-shot recognition, but increasingly less so for few-shot, object detection and dense prediction. No single self-supervised method dominates overall, suggesting that universal pre-training is still unsolved. Our analysis of features suggests that top self-supervised learners fail to preserve colour information as well as supervised alternatives, but tend to induce better classifier calibration, and less attentive overfitting than supervised learners.

['Timothy M. Hospedales', 'Henry Gouk', 'Linus Ericsson']

2020-11-26

http://openaccess.thecvf.com//content/CVPR2021/html/Ericsson_How_Well_Do_Self-Supervised_Models_Transfer_CVPR_2021_paper.html

http://openaccess.thecvf.com//content/CVPR2021/papers/Ericsson_How_Well_Do_Self-Supervised_Models_Transfer_CVPR_2021_paper.pdf

cvpr-2021-1

['fine-grained-image-recognition', 'classifier-calibration', 'surface-normals-estimation', 'classifier-calibration']

['computer-vision', 'computer-vision', 'computer-vision', 'miscellaneous']

[ 4.81756330e-01 2.05079079e-01 -3.54068339e-01 -4.83580649e-01 -7.31423259e-01 -2.29902148e-01 1.02910483e+00 2.31354758e-01 -5.20784676e-01 6.06601954e-01 4.21182722e-01 6.12266501e-03 -1.10975794e-01 -4.18391198e-01 -7.45148838e-01 -7.56247818e-01 -4.61480319e-02 5.58265150e-01 5.48074007e-01 -1.50529742e-01 1.09356888e-01 7.93071687e-02 -1.87357879e+00 6.65466487e-01 3.81027251e-01 8.80715907e-01 1.15695864e-01 8.38432789e-01 -5.98822460e-02 1.30288994e+00 -3.74559313e-01 -4.03552324e-01 9.64524522e-02 -6.79796457e-01 -8.16320479e-01 1.25648111e-01 7.46531487e-01 3.79207148e-03 -1.59003183e-01 7.90302873e-01 6.69430256e-01 1.37676194e-01 1.04461396e+00 -1.13864684e+00 -8.92104447e-01 4.66362566e-01 -6.53425932e-01 5.87823927e-01 -1.55032044e-02 5.57213426e-01 1.19275928e+00 -1.01950657e+00 8.84846568e-01 1.17270982e+00 9.92422402e-01 5.18976331e-01 -1.49972665e+00 -4.72604454e-01 -3.73206101e-02 3.47486943e-01 -1.01270700e+00 -7.59535015e-01 3.07458282e-01 -5.37127793e-01 1.23807645e+00 -4.98479083e-02 5.95285118e-01 1.23167884e+00 2.19485283e-01 8.60858202e-01 1.27877510e+00 -6.16353929e-01 1.89021558e-01 3.58807951e-01 3.42789233e-01 7.36841321e-01 2.53551394e-01 5.14382601e-01 -7.56846309e-01 2.21418828e-01 3.38827103e-01 1.23356685e-01 -5.06280288e-02 -6.65356636e-01 -1.12102830e+00 6.71759129e-01 6.33571029e-01 4.30343390e-01 -8.96681696e-02 2.22597241e-01 6.07072532e-01 6.19081259e-01 7.12323070e-01 5.84542453e-01 -3.77764344e-01 -6.75207600e-02 -1.25886059e+00 -1.91745654e-01 5.69539487e-01 8.80007982e-01 1.14508235e+00 2.70275682e-01 -4.23409760e-01 9.24382091e-01 -4.97649387e-02 2.63879836e-01 6.93146169e-01 -8.75519097e-01 8.74736086e-02 3.89045388e-01 -1.90473855e-01 -5.10969102e-01 -5.13196349e-01 -5.60225964e-01 -7.47848213e-01 4.84442681e-01 6.13040984e-01 -9.05823782e-02 -1.12870216e+00 1.36585987e+00 -3.64679128e-01 1.52071446e-01 1.39717206e-01 6.96773589e-01 9.40731645e-01 4.85715747e-01 4.99411404e-01 -2.56363809e-01 9.72974598e-01 -1.10169518e+00 -1.63444951e-01 -6.71342671e-01 9.10910726e-01 -5.81638277e-01 1.05245531e+00 1.86324358e-01 -8.20952058e-01 -8.36781263e-01 -1.10877979e+00 -4.26179990e-02 -5.60417414e-01 2.28629727e-02 7.82157302e-01 6.13504946e-01 -9.87323523e-01 9.20822442e-01 -5.72089434e-01 -9.60121453e-01 7.89529681e-01 1.26653463e-01 -5.13821065e-01 -3.11768144e-01 -8.37873578e-01 1.31496036e+00 4.69396800e-01 -4.66891587e-01 -1.20802784e+00 -9.21363354e-01 -8.44365239e-01 -1.10486113e-02 5.73120117e-02 -4.54555184e-01 1.22479677e+00 -1.35561359e+00 -1.16623342e+00 1.42467678e+00 7.84576014e-02 -8.26640129e-01 3.79096240e-01 2.07568035e-02 -3.04802626e-01 1.48567528e-01 5.66209806e-03 1.19737053e+00 1.02344370e+00 -1.33090878e+00 -7.07264662e-01 -3.27816814e-01 -1.47291034e-01 3.22389603e-01 -3.61404419e-01 9.04369727e-02 -2.55623423e-02 -3.04192603e-01 -2.62525767e-01 -6.27014995e-01 -1.15421131e-01 1.69551834e-01 -8.89764447e-03 -2.58950204e-01 6.52354240e-01 -1.43582389e-01 7.17988014e-01 -2.39981675e+00 -1.61811173e-01 -3.31798077e-01 1.08219698e-01 3.82144690e-01 -4.82257694e-01 4.18909818e-01 -4.30156857e-01 -1.51643485e-01 -2.44287387e-01 -4.50268149e-01 -2.47084405e-02 3.24898928e-01 -3.81005138e-01 6.27424538e-01 5.96682668e-01 1.23083150e+00 -1.00775075e+00 -6.36149406e-01 4.48462993e-01 2.34764248e-01 -1.59866512e-01 1.07672684e-01 -1.09306395e-01 -6.09321781e-02 1.95679083e-01 5.80373347e-01 2.65273362e-01 -5.26984930e-01 -1.49961561e-01 4.50908206e-02 -6.31396621e-02 1.35967731e-02 -6.64918542e-01 1.64770091e+00 -2.47796401e-01 9.25167024e-01 -2.87106723e-01 -1.23144960e+00 7.97555327e-01 6.38126135e-02 1.77072152e-01 -9.78530228e-01 -9.89293773e-03 3.42012793e-02 2.84585714e-01 -4.16938305e-01 1.67367354e-01 -4.63889986e-01 2.58562773e-01 3.37636977e-01 8.45425844e-01 9.70851630e-02 2.23317981e-01 4.36111689e-01 1.15033579e+00 4.23397750e-01 4.57695097e-01 -3.26774746e-01 5.65483421e-02 4.79369581e-01 1.92938820e-01 1.09826243e+00 -4.07597244e-01 9.00446475e-01 4.67607826e-01 -2.09366694e-01 -9.90561485e-01 -7.99649477e-01 -1.35448441e-01 1.79573810e+00 -3.47942337e-02 -3.26135308e-01 -4.91991937e-01 -7.21883237e-01 8.93319324e-02 7.32191503e-01 -8.99773300e-01 -4.63072479e-01 -8.75873640e-02 -9.01873469e-01 3.60555112e-01 7.76336074e-01 9.71838981e-02 -1.18287468e+00 -7.41603196e-01 4.39870618e-02 5.95455885e-01 -8.64303768e-01 -2.03799140e-02 9.06874001e-01 -9.50180233e-01 -1.01393628e+00 -1.00924087e+00 -8.18569362e-01 4.79903579e-01 6.32652104e-01 1.31483066e+00 1.02463491e-01 -6.34398699e-01 6.04625881e-01 -5.69992006e-01 -6.20683014e-01 -3.35435718e-01 1.67265162e-02 -8.07169452e-02 -8.17013010e-02 5.22973895e-01 -3.37162644e-01 -5.28348863e-01 3.37210566e-01 -4.93303895e-01 -1.59857541e-01 8.88136983e-01 1.08853686e+00 2.33770519e-01 -4.80862468e-01 6.13764167e-01 -1.11581230e+00 1.34569451e-01 -4.35411990e-01 -7.33799711e-02 2.72556514e-01 -9.41713691e-01 -6.49307519e-02 5.81088185e-01 -2.26096705e-01 -1.07981491e+00 3.26063931e-01 5.66339903e-02 -6.56225502e-01 -5.97305477e-01 1.96693569e-01 3.05256844e-01 -1.88239753e-01 1.30178881e+00 2.94200152e-01 2.80913591e-01 -4.04324353e-01 5.71271539e-01 3.93554360e-01 4.30165082e-01 -1.66143820e-01 7.82318115e-01 5.73974669e-01 -2.59595901e-01 -1.05365205e+00 -1.29386783e+00 -8.37627232e-01 -8.70193779e-01 -1.58557743e-01 9.27638769e-01 -1.13609838e+00 -7.59641305e-02 2.76216835e-01 -8.88463914e-01 -6.79441512e-01 -9.28928137e-01 3.09188753e-01 -7.80358434e-01 1.63404599e-01 -4.12023842e-01 -7.92469919e-01 -1.54146716e-01 -6.88009858e-01 9.31188047e-01 6.15725815e-02 -2.71824092e-01 -1.01902783e+00 1.91341802e-01 2.05420226e-01 4.65260088e-01 -1.88246951e-01 7.11296082e-01 -8.89268935e-01 -2.37171337e-01 -3.40621710e-01 -5.64149261e-01 3.46932888e-01 -1.47850841e-01 -1.84745163e-01 -1.43740845e+00 -3.04739714e-01 -4.28395003e-01 -1.04543614e+00 1.57117772e+00 2.55735129e-01 6.99490666e-01 2.61408836e-01 -3.16217452e-01 6.32836998e-01 1.41611743e+00 -2.00683609e-01 7.30121195e-01 2.25006759e-01 4.64501411e-01 6.91642761e-01 6.05185628e-01 -1.69354677e-02 2.47563310e-02 3.57288033e-01 2.18074232e-01 -2.43056297e-01 -6.28581941e-01 -2.62318969e-01 4.78711098e-01 3.28092098e-01 -1.81902885e-01 4.60177697e-02 -1.00732780e+00 5.34888625e-01 -1.83104002e+00 -1.28333831e+00 -3.26645672e-02 2.03621292e+00 5.32641768e-01 4.04762834e-01 4.41790283e-01 9.19216722e-02 5.19896626e-01 4.06997353e-01 -5.88448167e-01 -1.29661739e-01 -2.89413750e-01 2.51435667e-01 5.40831268e-01 1.29844710e-01 -1.18130171e+00 1.15741527e+00 7.80271769e+00 7.17178285e-01 -9.10578549e-01 3.14698368e-01 7.81207621e-01 -1.28784955e-01 1.99551120e-01 1.18587859e-01 -7.66471565e-01 2.21095026e-01 1.08052659e+00 -1.16259173e-01 1.10759288e-02 1.03574228e+00 -2.48310164e-01 -2.16490567e-01 -1.29873419e+00 9.51074660e-01 3.85085046e-01 -1.44886386e+00 -6.17666990e-02 -1.25134215e-01 8.36687863e-01 5.46079218e-01 -1.41635075e-01 7.48183370e-01 5.49604833e-01 -1.10355365e+00 5.01954317e-01 6.28318548e-01 7.93628097e-01 -3.67738426e-01 5.94796181e-01 4.35662568e-01 -8.20215285e-01 -2.71083742e-01 -6.85678065e-01 -2.78666645e-01 -1.95032731e-01 2.16024950e-01 -5.58366179e-01 1.51204214e-01 6.26873851e-01 1.25449681e+00 -1.00451469e+00 1.21692348e+00 -2.00257242e-01 8.30865085e-01 2.17539500e-02 -6.86397254e-02 3.57684672e-01 2.53410786e-01 2.23784715e-01 1.65159023e+00 -7.01718628e-02 -1.29648164e-01 9.97678041e-02 3.57301682e-01 7.42801428e-02 1.53981358e-01 -8.33017766e-01 -8.35444704e-02 -1.02860279e-01 1.23460555e+00 -8.64879310e-01 -5.64845085e-01 -7.43167460e-01 1.19967997e+00 5.65298378e-01 3.16040814e-01 -3.97175640e-01 -2.53925353e-01 2.96136647e-01 1.45968199e-01 4.49082345e-01 1.41299322e-01 -2.71923393e-01 -1.06863451e+00 -5.19552052e-01 -6.32405698e-01 7.76375294e-01 -1.02418149e+00 -1.56717646e+00 2.91395217e-01 -1.64891124e-01 -1.26252031e+00 -2.26586610e-01 -9.49973226e-01 -8.89256358e-01 2.87658334e-01 -1.69800651e+00 -1.06444943e+00 -2.74728656e-01 4.98801440e-01 6.47935688e-01 -5.16205907e-01 9.96336401e-01 -7.70351966e-04 -4.58967865e-01 5.23647130e-01 1.46788135e-01 7.11196661e-02 1.07160056e+00 -1.34647095e+00 9.27092358e-02 5.93466163e-01 6.26611114e-01 2.56947696e-01 6.32659674e-01 -3.66187274e-01 -1.31579483e+00 -1.03087556e+00 5.06489992e-01 -6.33948445e-01 8.63625646e-01 -2.45145455e-01 -1.03367794e+00 6.76064372e-01 5.09141326e-01 4.91285920e-01 7.18991876e-01 3.46981168e-01 -7.82215834e-01 -2.15832278e-01 -7.31046617e-01 4.08488214e-02 1.32587433e+00 -5.27587414e-01 -7.34621406e-01 5.59985936e-01 3.90993983e-01 1.36730418e-01 -4.99469161e-01 1.78898871e-01 3.55427027e-01 -1.27759778e+00 9.82355475e-01 -1.01140213e+00 7.79376805e-01 1.38914183e-01 -1.17524818e-01 -1.41540885e+00 -6.84071481e-01 -3.43819708e-01 -2.22715423e-01 1.02926338e+00 6.34927452e-01 -3.24275345e-01 1.06678510e+00 1.52182922e-01 -2.14963019e-01 -4.07702655e-01 -7.39436626e-01 -1.05830276e+00 1.15385242e-01 -3.00924063e-01 -3.12520564e-01 9.51231182e-01 1.57209605e-01 9.34535384e-01 -5.09672761e-01 -4.50805098e-01 8.37611914e-01 1.19955704e-01 8.34989011e-01 -1.40388215e+00 -4.03956443e-01 -5.89005828e-01 -8.12276483e-01 -6.63769960e-01 1.46674782e-01 -1.28067017e+00 1.89235508e-01 -1.77767777e+00 5.71087837e-01 -3.45675200e-02 -5.99776208e-01 7.08362818e-01 -7.37065524e-02 7.65748858e-01 1.64364159e-01 4.07833546e-01 -1.15681303e+00 3.65954578e-01 1.04162323e+00 -2.84614980e-01 4.35321182e-02 -1.60427019e-01 -8.31478179e-01 7.15838969e-01 6.00942910e-01 -3.51453424e-01 -3.86799544e-01 -3.65358472e-01 -1.86102912e-01 -4.17120188e-01 5.15979588e-01 -1.37025809e+00 2.39495173e-01 7.88288862e-02 8.32336724e-01 -1.64314240e-01 3.30038548e-01 -5.47228515e-01 -4.70775455e-01 4.62962002e-01 -5.42062044e-01 -5.88172317e-01 1.56317428e-01 8.15481365e-01 -1.09324872e-01 -4.11899596e-01 1.31537843e+00 -3.94129336e-01 -1.37520564e+00 1.41791284e-01 -4.20478284e-01 3.24760020e-01 1.09192085e+00 -5.26434958e-01 -4.42948371e-01 -3.75154465e-01 -1.00531507e+00 6.28709272e-02 3.58147085e-01 4.50481981e-01 3.93938988e-01 -9.55186069e-01 -7.46133029e-01 -4.88325991e-02 7.32269943e-01 -6.26187563e-01 1.48387343e-01 8.90127003e-01 -1.41374409e-01 3.53593260e-01 -4.08940256e-01 -7.58344591e-01 -1.09951162e+00 6.97244525e-01 2.96601146e-01 1.03051411e-02 -9.03661907e-01 1.03551269e+00 1.00380130e-01 -1.42232120e-01 6.06857479e-01 1.74073204e-01 -2.41826043e-01 5.45328319e-01 7.53554821e-01 3.89209509e-01 -9.19628963e-02 -5.41978478e-01 -2.26457104e-01 5.56435108e-01 -2.76783913e-01 1.75725102e-01 1.47996032e+00 1.74082085e-01 3.08632433e-01 8.65540445e-01 1.20150268e+00 -8.84526432e-01 -1.41776621e+00 -3.61146480e-01 2.83794433e-01 -2.06868052e-01 8.66244584e-02 -7.09527254e-01 -8.16348612e-01 1.33544123e+00 8.10014367e-01 1.37287304e-01 6.44508839e-01 3.15749705e-01 1.82075217e-01 7.24274635e-01 2.79285133e-01 -1.21335638e+00 3.40811938e-01 6.52531087e-01 6.52240336e-01 -1.70481420e+00 2.19331354e-01 -1.19445272e-01 -9.57915127e-01 8.42983305e-01 7.92438030e-01 -3.22523773e-01 5.80942869e-01 1.56482399e-01 8.02667066e-02 -2.72401065e-01 -9.83047664e-01 -8.78311753e-01 2.59196997e-01 8.96178246e-01 4.33694303e-01 -3.50533992e-01 1.85847506e-01 2.06961930e-01 2.08136901e-01 1.50959522e-01 1.77101031e-01 7.96232164e-01 -8.64927411e-01 -6.42911971e-01 -1.84309743e-02 8.07727218e-01 -6.18979521e-02 -2.49977395e-01 -5.42133331e-01 7.98356712e-01 1.22976705e-01 6.80159688e-01 3.66471827e-01 -2.45315924e-01 2.96705127e-01 6.39413178e-01 5.83678365e-01 -1.15336144e+00 -6.13239348e-01 -8.09218585e-02 8.66190717e-02 -4.01432127e-01 -5.49573362e-01 -6.35268509e-01 -8.90407622e-01 1.90587699e-01 -2.62973934e-01 -1.22922309e-01 3.43620956e-01 9.34678257e-01 1.91589564e-01 3.56963545e-01 4.35559332e-01 -1.15757895e+00 -6.57171071e-01 -1.14166009e+00 -6.52189672e-01 6.07866406e-01 2.45395139e-01 -7.17260778e-01 -4.81905967e-01 1.51842162e-01]

[9.840818405151367, 2.600034713745117]

1f404b4b-8048-419d-bcea-5af080725d06

self-supervised-rf-signal-representation

2207.03046

https://arxiv.org/abs/2207.03046v2

https://arxiv.org/pdf/2207.03046v2.pdf

Self-Supervised RF Signal Representation Learning for NextG Signal Classification with Deep Learning

Deep learning (DL) finds rich applications in the wireless domain to improve spectrum awareness. Typically, DL models are either randomly initialized following a statistical distribution or pretrained on tasks from other domains in the form of transfer learning without accounting for the unique characteristics of wireless signals. Self-supervised learning (SSL) enables the learning of useful representations from Radio Frequency (RF) signals themselves even when only limited training data samples with labels are available. We present a self-supervised RF signal representation learning method and apply it to the automatic modulation recognition (AMR) task by specifically formulating a set of transformations to capture the wireless signal characteristics. We show that the sample efficiency (the number of labeled samples needed to achieve a certain performance) of AMR can be significantly increased (almost an order of magnitude) by learning signal representations with SSL. This translates to substantial time and cost savings. Furthermore, SSL increases the model accuracy compared to the state-of-the-art DL methods and maintains high accuracy when limited training data is available.

['Ender Ayanoglu', 'Yalin E. Sagduyu', 'Mehmet Can Ertem', 'Serdar Boztas', 'Kemal Davaslioglu']

2022-07-07

['automatic-modulation-recognition']

['time-series']

[ 8.62515509e-01 1.57505661e-01 -6.22562408e-01 -5.72559357e-01 -1.01966941e+00 -3.46656770e-01 5.48963070e-01 -2.58186519e-01 -2.77786344e-01 8.63811731e-01 -3.01089045e-02 -4.29486275e-01 -3.72386664e-01 -7.62735605e-01 -7.46826768e-01 -6.37740731e-01 -3.50773275e-01 1.41345069e-01 -1.76624238e-01 -1.51948169e-01 -2.34048396e-01 4.69184846e-01 -1.32495296e+00 2.18970433e-01 6.11320078e-01 1.48821902e+00 2.21570268e-01 6.04790688e-01 -9.59473178e-02 1.22668779e+00 -7.66197264e-01 3.38358879e-01 2.80299217e-01 -5.90508699e-01 -6.04282379e-01 -2.34110638e-01 -5.70556708e-02 -1.20305479e-01 -5.83508968e-01 8.92330468e-01 7.17175841e-01 -8.72386023e-02 7.46901274e-01 -1.02179062e+00 -2.55033016e-01 8.62435877e-01 -4.16099519e-01 1.87013313e-01 1.68417349e-01 -3.88441682e-01 8.78028393e-01 -4.95843709e-01 1.56259567e-01 8.66489530e-01 7.80659795e-01 3.81147891e-01 -1.43415391e+00 -1.04109883e+00 -2.32022963e-02 1.95330113e-01 -1.41981387e+00 -6.15694463e-01 8.80636275e-01 -1.38865724e-01 5.38072705e-01 2.14733288e-01 3.40954214e-01 1.37284684e+00 -9.22109038e-02 5.53456843e-01 1.06075644e+00 -6.99677527e-01 4.33516294e-01 1.65721729e-01 -2.38942821e-02 4.22514588e-01 -3.93639505e-02 1.65869936e-01 -4.98036444e-01 -1.42058864e-01 7.04374611e-01 -6.54467568e-02 -1.67469010e-01 -4.63111848e-01 -1.11477089e+00 7.92118788e-01 4.10867929e-01 5.85646808e-01 -3.75673026e-01 5.61204910e-01 1.64775848e-01 8.21741581e-01 4.31407958e-01 5.53652287e-01 -5.75510800e-01 -2.56889011e-03 -1.10375524e+00 -2.68674999e-01 7.01605737e-01 8.79105568e-01 9.50382054e-01 5.97215533e-01 -8.28363597e-02 1.00811768e+00 7.46072307e-02 8.43214750e-01 6.66145444e-01 -7.46399045e-01 2.87626058e-01 -3.92679088e-02 1.74570084e-02 -6.61401033e-01 -6.23554230e-01 -1.22024429e+00 -9.51376081e-01 -1.99411169e-01 2.02798173e-01 -5.35992205e-01 -9.21531796e-01 2.01622534e+00 -3.22923064e-01 3.91131490e-01 2.78529912e-01 4.62780625e-01 4.78476465e-01 8.48354816e-01 -1.99517727e-01 -3.39678377e-01 9.59798574e-01 -6.20103180e-01 -6.01727962e-01 -3.68983597e-01 6.40379012e-01 -6.21392310e-01 8.61304164e-01 5.28266132e-01 -5.92260718e-01 -6.66749060e-01 -1.42871833e+00 4.50919122e-01 -1.58184320e-01 1.15055889e-01 7.51480699e-01 1.19250095e+00 -7.83349454e-01 4.34444487e-01 -4.67845410e-01 -1.05511360e-01 6.20343506e-01 6.89080536e-01 8.90460331e-04 -1.79139972e-01 -1.30121160e+00 5.51262140e-01 4.46351990e-02 -1.51820228e-01 -1.06075740e+00 -9.26434577e-01 -6.81778908e-01 2.55180955e-01 2.95753717e-01 -2.99322605e-01 1.38741219e+00 -1.32367253e+00 -1.89708090e+00 4.43920732e-01 1.55733570e-01 -8.46964478e-01 2.80676544e-01 -5.97868562e-02 -1.00754404e+00 1.38304010e-01 -1.47807106e-01 1.79435343e-01 1.37136519e+00 -1.12211692e+00 -5.45185447e-01 8.89478326e-02 9.08743665e-02 -2.46442184e-01 -3.34725499e-01 -2.70059079e-01 1.95252955e-01 -8.72084200e-01 1.43245667e-01 -9.80672300e-01 -2.07893103e-01 -3.15257937e-01 -6.94979876e-02 2.57654965e-01 7.79619217e-01 -4.30650949e-01 9.57567453e-01 -2.21754956e+00 -2.20333502e-01 6.88828051e-01 5.24326414e-02 2.61996299e-01 -3.27311963e-01 2.47297406e-01 -2.83566624e-01 -1.40782490e-01 -1.52101174e-01 1.70365468e-01 -1.19974734e-02 -3.08245011e-02 -5.17106950e-01 4.55538362e-01 4.14890721e-02 7.39631891e-01 -8.31019819e-01 1.48354456e-01 1.78582773e-01 3.75858545e-01 -4.09287483e-01 1.51011556e-01 -2.07983837e-01 7.24838316e-01 -4.06008422e-01 3.25723350e-01 4.68860984e-01 -4.11436498e-01 5.86465836e-01 -4.21685398e-01 3.42668712e-01 5.63802123e-01 -1.02687347e+00 1.65731037e+00 -1.23374403e+00 8.86934102e-01 -9.42355096e-02 -1.62211788e+00 1.02878106e+00 3.99130911e-01 8.34712565e-01 -9.86484051e-01 2.60462701e-01 2.82099098e-01 1.67069286e-01 -1.49202392e-01 -5.51306196e-02 -2.74803281e-01 -2.73815334e-01 7.71560550e-01 4.57774401e-01 1.23285703e-01 -2.36910582e-01 -1.84799746e-01 1.28091753e+00 -2.63721585e-01 3.44772935e-01 -2.35458478e-01 5.42475879e-01 -5.13900697e-01 3.12547445e-01 1.10296023e+00 3.26934665e-01 1.01389945e-01 1.08296499e-01 -2.28127122e-01 -8.24734747e-01 -1.08302557e+00 -3.63876551e-01 1.32553279e+00 7.20661774e-04 -1.84347302e-01 -3.96211475e-01 -4.56949979e-01 -1.31980121e-01 6.63886189e-01 -3.64802897e-01 -4.54742759e-01 -3.81421983e-01 -8.89535129e-01 5.26333630e-01 4.27505940e-01 5.52723706e-01 -9.47448373e-01 -2.61697829e-01 4.02133256e-01 8.94625112e-03 -1.39933705e+00 9.36840251e-02 8.05627823e-01 -5.69143713e-01 -5.08661926e-01 -7.51211524e-01 -8.30358326e-01 4.78729188e-01 1.70801029e-01 7.61863291e-01 -2.63871670e-01 -2.38453418e-01 5.46957374e-01 -4.59983349e-01 -4.51313943e-01 -5.30984819e-01 4.46437061e-01 3.43525141e-01 3.80108058e-01 1.57929257e-01 -1.06664240e+00 -3.31730336e-01 3.13169897e-01 -6.18339300e-01 -3.15081000e-01 8.96562219e-01 8.00312221e-01 2.03456119e-01 2.80382425e-01 1.22010756e+00 -1.12320936e+00 4.18608099e-01 -5.50648868e-01 -4.06656295e-01 6.41161501e-02 -5.04075885e-01 3.77778202e-01 8.55813503e-01 -6.96047246e-01 -7.72528827e-01 -1.14636861e-01 -1.47910388e-02 -2.89901160e-02 1.19089119e-01 3.78703296e-01 -1.24792293e-01 -4.34278458e-01 8.86232793e-01 3.06211203e-01 -4.46830280e-02 -3.87822360e-01 4.07570720e-01 1.04829395e+00 4.09878671e-01 -5.36153853e-01 1.05534387e+00 2.88352698e-01 4.91437353e-02 -1.03688681e+00 -1.14585125e+00 -1.78048402e-01 -3.18335414e-01 -7.77057037e-02 2.25665331e-01 -9.95242238e-01 -4.03707504e-01 1.10782366e-02 -3.91701519e-01 -6.57205224e-01 -5.24729013e-01 7.53551722e-01 -7.69268990e-01 8.31938386e-02 -2.38173127e-01 -9.04440582e-01 -1.55649334e-01 -6.44147635e-01 6.11315310e-01 -9.71600860e-02 -2.48836294e-01 -8.26133311e-01 -2.16424271e-01 -7.95139968e-02 8.20024729e-01 -3.21161747e-02 1.20804453e+00 -5.62445223e-01 -3.18324447e-01 -2.40434021e-01 -5.53361736e-02 4.60759103e-01 4.32975233e-01 -8.76553595e-01 -1.28235984e+00 -4.80496943e-01 1.11438952e-01 -2.83078462e-01 8.34894478e-01 4.92413104e-01 1.48069715e+00 -1.95325300e-01 -3.30124617e-01 5.60113192e-01 1.18625069e+00 3.07900667e-01 4.85459715e-01 1.30045369e-01 2.46069327e-01 1.40976980e-01 3.47087145e-01 4.33655441e-01 -1.51753992e-01 8.03568125e-01 1.22528039e-01 -3.94498646e-01 -1.75086409e-01 -2.51684755e-01 3.07615131e-01 5.04749000e-01 1.63812667e-01 -2.66455058e-02 -6.19758666e-01 1.88133299e-01 -1.60903275e+00 -9.69920814e-01 4.66517180e-01 2.35548425e+00 7.49090314e-01 3.04046512e-01 1.88519344e-01 7.63266981e-01 6.02859855e-01 1.81593865e-01 -6.55175924e-01 -7.27653801e-02 4.28359471e-02 7.18003869e-01 8.06946993e-01 3.63649756e-01 -1.14895189e+00 6.90262616e-01 6.94220781e+00 9.50949609e-01 -1.49340940e+00 6.29936904e-02 4.56390351e-01 9.95276570e-02 -1.67013675e-01 -3.47440869e-01 -2.70179063e-01 2.56285757e-01 1.39929545e+00 8.93629249e-03 8.37337792e-01 7.95991600e-01 -1.00209948e-03 3.11631352e-01 -1.22477806e+00 1.50262654e+00 -1.29220739e-01 -1.30079436e+00 -1.64197415e-01 1.11984365e-01 6.06165051e-01 -1.82034895e-02 4.23094064e-01 6.49046421e-01 1.93227902e-01 -1.21222842e+00 5.45789063e-01 3.63606423e-01 1.21448684e+00 -9.47270155e-01 6.77942157e-01 3.40544254e-01 -1.05723476e+00 -6.35179996e-01 -4.38376069e-01 -5.10255657e-02 -2.85241585e-02 8.06012452e-01 -9.45788682e-01 5.91112792e-01 4.04110402e-01 6.45379841e-01 -3.07141453e-01 7.18797743e-01 -6.84179589e-02 1.09930062e+00 -3.43054712e-01 2.59548277e-02 1.30411506e-01 8.16164911e-02 2.74279147e-01 1.08454800e+00 5.25878549e-01 -2.01906592e-01 1.43269002e-01 4.72938895e-01 -4.04837430e-01 5.23883067e-02 -4.12903458e-01 -1.32116616e-01 7.34380901e-01 1.07241952e+00 -4.01437491e-01 -1.54735148e-01 -4.00691211e-01 6.90268040e-01 -3.11025791e-02 4.03305709e-01 -7.77945220e-01 -5.12103379e-01 3.43250036e-01 3.45467746e-01 4.93391186e-01 -2.57096589e-01 -1.35409877e-01 -7.10246086e-01 -3.31239849e-01 -8.95949304e-01 -4.26186360e-02 -4.77968216e-01 -1.30677056e+00 5.41877568e-01 -2.21671820e-01 -1.54461491e+00 -5.67837715e-01 -5.38765311e-01 -1.47285521e-01 4.64407772e-01 -1.68481505e+00 -9.54741359e-01 -2.25700885e-01 5.84061563e-01 2.96503484e-01 -7.90819943e-01 1.20818377e+00 4.97837931e-01 -1.73328221e-01 9.57074106e-01 3.73357207e-01 1.13613091e-01 5.87181687e-01 -1.03343260e+00 1.20679826e-01 4.01895553e-01 4.56260294e-01 4.36542064e-01 5.82244396e-01 1.22747958e-01 -1.14149547e+00 -1.26288438e+00 3.15626591e-01 4.88615483e-02 6.72832847e-01 -7.09211290e-01 -4.84093487e-01 5.77169299e-01 -1.94400519e-01 1.74329087e-01 1.22806251e+00 3.23648930e-01 -4.59088862e-01 -6.86177075e-01 -1.06670964e+00 3.74846727e-01 1.05318165e+00 -7.90036440e-01 -2.43037194e-01 2.61659890e-01 4.88183588e-01 -4.56039943e-02 -8.67515802e-01 4.65008676e-01 6.63017750e-01 -4.49161083e-01 1.13033390e+00 -4.35095131e-01 -1.80117592e-01 -2.98633933e-01 -6.25119746e-01 -1.25804031e+00 -3.30423594e-01 -9.22972381e-01 -4.40987408e-01 9.66755629e-01 6.36034667e-01 -6.13289475e-01 6.51985586e-01 -1.48435399e-01 1.65812224e-01 -4.10857081e-01 -8.35965812e-01 -9.60781217e-01 -3.14479619e-01 -8.52692962e-01 6.13251626e-01 9.71976161e-01 -1.54070193e-02 6.39385700e-01 -6.76504910e-01 2.51157761e-01 6.91768825e-01 1.09411813e-01 7.30341852e-01 -1.47818553e+00 -6.35563731e-01 -8.25532246e-03 -5.68978727e-01 -1.40157771e+00 3.46827209e-01 -1.22922611e+00 -1.18172750e-01 -1.09977460e+00 -2.45392278e-01 -9.14230466e-01 -7.53435075e-01 4.19220954e-01 4.76960719e-01 3.66979927e-01 4.33725398e-03 2.62292195e-02 -4.42634225e-01 4.56540495e-01 5.70328176e-01 -3.02689761e-01 -3.09452325e-01 5.22761405e-01 -9.17762280e-01 6.04524255e-01 1.09473050e+00 -5.53530991e-01 -6.54207826e-01 -2.25770716e-02 1.05380103e-01 -4.39140238e-02 1.13186829e-01 -1.53515887e+00 -3.28399502e-02 1.94933519e-01 6.61776602e-01 -1.40382901e-01 3.92400205e-01 -9.68758345e-01 6.68171793e-02 4.70670700e-01 -6.65238798e-01 -8.37450325e-01 -1.27864042e-02 7.98022091e-01 8.45791921e-02 -2.05725864e-01 9.29638565e-01 2.35994875e-01 -5.01208246e-01 4.76585552e-02 -7.47495890e-01 -1.03479512e-01 6.81561470e-01 -4.50629741e-02 9.37942639e-02 -1.00213122e+00 -6.95887744e-01 -3.19121391e-01 -2.46623412e-01 3.90658945e-01 2.09549502e-01 -1.50669765e+00 -3.46902430e-01 3.96609664e-01 4.71831858e-02 -3.89313370e-01 4.84069698e-02 5.93348801e-01 -1.94355957e-02 5.54501593e-01 -1.68085784e-01 -5.99865735e-01 -6.86268032e-01 3.28787029e-01 4.88604903e-01 -3.06903869e-01 -7.53181577e-01 4.79825288e-01 -7.34387385e-03 -4.36642528e-01 3.17989439e-01 -3.49501491e-01 -1.27700955e-01 -2.24441752e-01 5.31569779e-01 2.50170648e-01 2.44776681e-01 -1.62674651e-01 -3.60354751e-01 5.93958795e-01 5.13228530e-04 -1.24672987e-01 1.48362219e+00 7.67255872e-02 4.37959433e-01 4.85871524e-01 1.42322862e+00 2.22040683e-01 -1.16701710e+00 -7.25813210e-01 1.42350435e-01 -6.92023560e-02 4.25178707e-01 -8.07757437e-01 -8.61336768e-01 7.45541751e-01 1.14427567e+00 3.18415582e-01 1.14654696e+00 -3.77950110e-02 7.70928919e-01 7.70727158e-01 9.20556486e-01 -1.08474672e+00 1.89780027e-01 3.19285929e-01 5.05542338e-01 -9.08841968e-01 -6.00261539e-02 -3.58469546e-01 -2.61806190e-01 1.08015847e+00 -2.48076990e-01 -1.85940668e-01 9.16803956e-01 5.40494084e-01 1.04480870e-01 2.55647928e-01 -2.71691054e-01 -3.00158888e-01 3.01230639e-01 1.11832750e+00 4.26108778e-01 1.02731332e-01 2.08989009e-01 9.55933928e-01 -4.72813517e-01 -1.19225770e-01 3.98613095e-01 8.63568842e-01 -7.05745161e-01 -1.24882138e+00 -5.25515489e-02 8.25127840e-01 -5.33659220e-01 3.97961065e-02 3.48260701e-02 4.69950289e-01 -1.73118621e-01 1.17409360e+00 -1.11196890e-01 -5.14391005e-01 1.19736232e-01 2.17244029e-02 5.98378003e-01 -7.15310633e-01 7.15129524e-02 3.42692770e-02 1.13388725e-01 -3.13101321e-01 -5.78076959e-01 -2.09823340e-01 -1.18305075e+00 6.18218966e-02 -2.46080756e-01 1.49105042e-01 5.84518492e-01 1.07989478e+00 1.99085221e-01 1.04597843e+00 1.15642369e+00 -9.12999511e-01 -6.83075249e-01 -9.78755891e-01 -7.94618428e-01 -8.33011642e-02 5.79532981e-01 -8.24238777e-01 -3.70314747e-01 -1.16557643e-01]

[6.557858943939209, 1.5024254322052002]

c06a4d97-c714-450b-8d8a-35dff77a0bd9

a-morphological-analyzer-for-gulf-arabic

https://aclanthology.org/W17-1305

https://aclanthology.org/W17-1305.pdf

A Morphological Analyzer for Gulf Arabic Verbs

We present CALIMAGLF, a Gulf Arabic morphological analyzer currently covering over 2,600 verbal lemmas. We describe in detail the process of building the analyzer starting from phonetic dictionary entries to fully inflected orthographic paradigms and associated lexicon and orthographic variants. We evaluate the coverage of CALIMA-GLF against Modern Standard Arabic and Egyptian Arabic analyzers on part of a Gulf Arabic novel. CALIMA-GLF verb analysis token recall for identifying correct POS tag outperforms both the Modern Standard Arabic and Egyptian Arabic analyzers by over 27.4{\%} and 16.9{\%} absolute, respectively.

['Salam Khalifa', 'Sara Hassan', 'Nizar Habash']

2017-04-01

ws-2017-4

['morphological-tagging']

['natural-language-processing']

[-2.93904305e-01 -2.87491500e-01 4.58955318e-02 -2.60030001e-01 -7.82886386e-01 -1.56951928e+00 2.53852516e-01 6.27669990e-01 -8.07337701e-01 8.77110660e-01 1.45777449e-01 -8.87605846e-01 -7.87962973e-02 -8.25740337e-01 -2.67443389e-01 -3.68735909e-01 3.13737720e-01 1.00887215e+00 9.74409357e-02 -9.19456959e-01 2.92220563e-01 5.49467325e-01 -5.96443117e-01 2.16817021e-01 1.08466470e+00 2.54445434e-01 -1.11111971e-02 4.72826749e-01 -1.38981342e-01 7.22866207e-02 -8.04546237e-01 -1.00563824e+00 2.11849421e-01 -3.83109093e-01 -8.71212423e-01 -4.13135380e-01 1.74374133e-01 -1.29982352e-01 2.00719178e-01 1.30246925e+00 2.90445149e-01 -1.81164443e-02 1.02416217e+00 -5.90587914e-01 -8.64877164e-01 1.48707461e+00 -4.40299630e-01 8.23672891e-01 5.92056692e-01 -1.10806204e-01 1.00604296e+00 -1.43960881e+00 7.26522446e-01 1.54203510e+00 5.30937672e-01 -6.32720217e-02 -6.97304308e-01 -6.28929019e-01 8.56770501e-02 3.68707962e-02 -1.58615625e+00 -6.12362325e-01 1.78976711e-02 -2.14434087e-01 1.61455166e+00 -6.64633363e-02 8.10689509e-01 3.51396888e-01 3.87689590e-01 1.84834614e-01 1.30729723e+00 -1.07564890e+00 -1.05404094e-01 -5.03876090e-01 5.39774060e-01 6.66023970e-01 9.04145956e-01 -2.29793280e-01 -3.82904023e-01 1.13160938e-01 4.28137928e-01 -1.00443196e+00 1.16520867e-01 9.74670649e-01 -1.13402927e+00 9.45466459e-01 -4.32862163e-01 3.56843799e-01 -3.64205420e-01 -2.85197169e-01 4.62175161e-01 3.67730111e-01 -1.89245954e-01 5.40902615e-01 -5.99609613e-01 -2.84480155e-01 -6.16779327e-01 9.45318937e-02 5.73639870e-01 1.05963230e+00 4.09740686e-01 7.74175942e-01 6.03452921e-01 9.36430454e-01 7.10238576e-01 1.51695418e+00 7.12816536e-01 -8.51857722e-01 3.35319459e-01 2.24111438e-01 -5.65033406e-02 -5.94611883e-01 -3.68157327e-01 1.07517362e-01 2.05306530e-01 -5.84575124e-02 1.05210662e+00 -2.52674282e-01 -1.11761284e+00 1.45423448e+00 1.97818696e-01 -1.13230038e+00 3.84560376e-01 3.85901868e-01 5.83452880e-01 1.01564753e+00 3.59115541e-01 -4.20957446e-01 1.86364686e+00 -7.11332798e-01 -1.07056999e+00 -5.53118765e-01 5.15870810e-01 -1.63108683e+00 1.46616924e+00 5.21461487e-01 -1.78831625e+00 -1.02524385e-01 -1.33858263e+00 -1.11451000e-01 -8.73888552e-01 1.19213291e-01 2.76062727e-01 1.41707015e+00 -8.96067977e-01 -8.98457915e-02 -9.05174255e-01 -2.33764872e-01 -3.25285643e-01 3.12761098e-01 -5.33879697e-01 -7.61545151e-02 -1.31531227e+00 1.30540574e+00 9.53427851e-01 2.86679529e-02 -5.09569764e-01 8.79525021e-02 -1.18050957e+00 -4.10995960e-01 3.64711694e-02 3.33554029e-01 1.22802377e+00 -6.80900455e-01 -1.55229473e+00 1.19073379e+00 -5.09332754e-02 -9.88696888e-02 -3.66415262e-01 -3.64007264e-01 -1.03210807e+00 1.82150215e-01 2.41653889e-01 3.59930158e-01 4.71739680e-01 -8.43364596e-01 -9.87515390e-01 -5.72764993e-01 -2.46872500e-01 3.54199886e-01 1.31806701e-01 9.30115104e-01 -1.05655633e-01 -1.06148398e+00 2.04776227e-01 -1.07493210e+00 3.23316842e-01 -1.05575812e+00 1.49669170e-01 -1.27180263e-01 1.76061302e-01 -1.57246196e+00 1.62047970e+00 -1.65950739e+00 -2.42849424e-01 7.75522292e-01 -5.31513870e-01 5.27509809e-01 -1.94768816e-01 5.07589936e-01 7.49192685e-02 2.20492557e-02 -3.56364191e-01 3.75030369e-01 1.44594476e-01 6.92253113e-01 -4.22288179e-02 6.28669500e-01 4.07147139e-01 1.14470053e+00 -9.89095688e-01 -2.66141504e-01 -1.16918162e-01 3.02496731e-01 -2.89105415e-01 -5.52434921e-01 5.38777485e-02 -1.66132659e-01 3.61047119e-01 1.23306179e+00 6.61163688e-01 8.73660207e-01 7.31603086e-01 3.44573826e-01 -4.05182421e-01 9.85694170e-01 -1.06509912e+00 1.46737909e+00 -3.60778093e-01 6.43860698e-02 -1.09377839e-02 -3.65500420e-01 6.87709808e-01 1.11543849e-01 -4.30250198e-01 -7.89657712e-01 5.83855867e-01 1.17763209e+00 8.13692153e-01 1.46615654e-01 9.45003152e-01 -2.22967699e-01 -6.81505561e-01 5.39225280e-01 4.79759932e-01 -2.41744384e-01 8.04636240e-01 6.80213720e-02 5.14860988e-01 2.34312207e-01 1.14944923e+00 -9.13969636e-01 6.60955906e-01 3.15360576e-01 4.30107147e-01 3.62174183e-01 -2.70462513e-01 2.77137458e-01 -2.19841544e-02 7.19713420e-03 -5.75514972e-01 -1.72541010e+00 -3.18547934e-01 1.40791678e+00 -2.88851708e-01 -8.48702133e-01 -9.87983346e-01 -6.24697864e-01 -2.51033008e-01 1.19851303e+00 -3.33036691e-01 8.59259218e-02 -1.42362952e+00 -9.84165370e-01 1.35147989e+00 6.55474246e-01 2.13587247e-02 -1.36787689e+00 -6.12224758e-01 6.18859231e-01 -4.52712297e-01 -9.95230317e-01 -4.47130114e-01 2.43265063e-01 -2.66468257e-01 -9.72889364e-01 -1.25068471e-01 -1.07812190e+00 7.80642629e-02 -1.28939390e-01 1.08198929e+00 -1.29360825e-01 3.75642985e-01 1.19723670e-01 -7.46385634e-01 -8.92246187e-01 -9.34128881e-01 -2.36210711e-02 2.79239595e-01 -7.97336161e-01 9.27128673e-01 3.02196387e-02 1.80457622e-01 -2.81165987e-01 -6.58472061e-01 -1.08541989e+00 3.60394001e-01 4.93553698e-01 6.99796259e-01 -4.46941942e-01 4.98272717e-01 -9.57487643e-01 4.47156936e-01 -3.77764523e-01 -5.48641086e-01 1.38701797e-01 -7.09302008e-01 -2.33640909e-01 7.03380406e-01 -7.00160787e-02 -1.12128925e+00 -4.07094091e-01 -8.05118144e-01 7.63075709e-01 3.25033441e-02 9.70956147e-01 -7.66243756e-01 1.49721697e-01 7.76851535e-01 1.24029340e-02 -1.05587222e-01 -2.83592790e-01 6.22432530e-01 6.30630970e-01 1.06434727e+00 -7.10784197e-01 7.26462483e-01 1.19071141e-01 -3.57607841e-01 -8.83925855e-01 -2.47615814e-01 -3.18301678e-01 -9.18034852e-01 -4.55911569e-02 7.43695378e-01 -6.82975292e-01 -2.92894155e-01 7.10143089e-01 -1.19816566e+00 -3.43515426e-01 -4.35048938e-01 3.35912466e-01 -2.60869354e-01 5.48493385e-01 -1.09229183e+00 -3.52748454e-01 -5.13127089e-01 -1.15263009e+00 7.19390154e-01 -1.05814196e-01 -6.36727870e-01 -1.05456007e+00 2.62429744e-01 -2.08684817e-01 1.46826282e-01 -2.29279068e-03 1.34325111e+00 -8.72205555e-01 6.16895556e-01 6.85515404e-02 5.36055304e-02 2.93761760e-01 4.42040175e-01 1.17667638e-01 -2.11068854e-01 -2.21148804e-01 -9.82638150e-02 -1.23667791e-01 2.13590965e-01 1.04580738e-01 -4.70556170e-01 -2.85869330e-01 4.68622178e-01 2.91167915e-01 1.41199744e+00 9.05300140e-01 7.29395211e-01 7.54530311e-01 3.21180195e-01 3.79877836e-01 1.00390768e+00 2.38617092e-01 5.84510326e-01 1.23454578e-01 -2.65728980e-02 6.68776393e-01 -3.41831356e-01 1.85369980e-02 1.22398400e+00 1.75492847e+00 5.34210205e-02 -2.57665753e-01 -1.49439692e+00 9.26119506e-01 -9.00992513e-01 -2.01571018e-01 -3.49713415e-01 1.87837017e+00 1.14321899e+00 -1.10775623e-02 3.88660431e-01 5.90002179e-01 5.26233613e-01 -2.88687199e-01 4.95247871e-01 -1.20079958e+00 -7.35776186e-01 1.32132030e+00 4.82058078e-01 1.19391799e+00 -8.92804146e-01 2.06179905e+00 6.15577936e+00 8.79465818e-01 -8.11877906e-01 3.31273884e-01 -3.45224231e-01 3.78613144e-01 -4.32799309e-01 1.61670804e-01 -1.04775214e+00 2.10884094e-01 1.54186428e+00 -2.79075861e-01 4.96458441e-01 1.96289033e-01 -1.19337678e-01 -2.63266921e-01 -4.10551459e-01 7.83119798e-01 3.67178559e-01 -9.96076584e-01 7.52347589e-01 -2.75808722e-01 5.16463459e-01 5.69225997e-02 -1.77222583e-02 8.27766806e-02 7.38218069e-01 -1.14075983e+00 1.41539097e+00 -1.29513085e-01 1.21888340e+00 -1.18948853e+00 8.44400585e-01 -3.19614172e-01 -1.09972084e+00 3.36978912e-01 -3.63917083e-01 -2.73655891e-01 3.94200981e-01 -1.88670725e-01 -6.06206715e-01 4.50626284e-01 4.75895703e-01 1.57685474e-01 -1.01474690e+00 5.71738005e-01 -8.14311326e-01 1.39403689e+00 -6.21975720e-01 1.24666184e-01 7.18998373e-01 -7.76805878e-01 7.08088100e-01 1.94488907e+00 3.64366353e-01 3.52739811e-01 1.50134582e-02 -3.93277794e-01 3.62709522e-01 1.16002059e+00 -1.06919482e-01 -2.38726735e-01 8.98464501e-01 7.41704941e-01 -1.48612010e+00 -2.58468807e-01 -2.60775447e-01 8.73051465e-01 2.36100093e-01 8.33476707e-02 -6.81512296e-01 -8.37619364e-01 4.96817559e-01 -3.87441181e-02 1.94200918e-01 -6.98514640e-01 -4.32826340e-01 -6.21514201e-01 -2.54576355e-01 -1.50396514e+00 9.42753315e-01 -4.51042354e-01 -9.97791588e-01 9.77141976e-01 3.84568311e-02 -6.30372584e-01 -4.26095039e-01 -1.19358850e+00 -2.81015281e-02 1.17130756e+00 -1.41265607e+00 -1.36581385e+00 3.15823823e-01 7.42040753e-01 3.20444793e-01 -6.34375572e-01 1.50962305e+00 2.73032308e-01 -2.88606018e-01 1.00563514e+00 3.07375342e-01 5.35612226e-01 9.54777300e-01 -1.40296423e+00 5.66550136e-01 1.49518478e+00 3.61029476e-01 7.65057623e-01 4.43311512e-01 -7.93538094e-01 -9.58367407e-01 -1.02691817e+00 1.24586129e+00 -5.95759988e-01 9.02929366e-01 -3.77423945e-03 -7.09138930e-01 1.19954336e+00 9.31418955e-01 -6.54889286e-01 9.96801496e-01 -2.56063938e-01 -5.21976292e-01 3.26951444e-01 -1.18535626e+00 1.02894521e+00 8.18673968e-01 -2.87389338e-01 -1.15705049e+00 -6.66643977e-02 4.15762275e-01 -3.99367481e-01 -8.34634364e-01 1.05666809e-01 6.35211408e-01 -3.28273416e-01 5.23321629e-01 -7.55550623e-01 -4.83966768e-02 -5.43630600e-01 -5.51002383e-01 -1.66309249e+00 -4.08315629e-01 -9.43924844e-01 2.58706361e-01 1.15182316e+00 6.46453917e-01 -8.54451358e-01 -1.28639594e-01 -6.48260355e-01 -6.21599615e-01 6.49161115e-02 -9.41190183e-01 -7.93826878e-01 8.62873495e-01 -6.46740913e-01 3.64162236e-01 9.89322126e-01 2.59282857e-01 4.87916350e-01 5.70066452e-01 1.31591782e-01 1.73731297e-01 -3.91657382e-01 -1.19024247e-01 -8.21502924e-01 -9.11241025e-02 -5.70847094e-01 -4.65974689e-01 -2.59965688e-01 2.65513778e-01 -1.35257542e+00 -8.22280645e-02 -1.25930285e+00 -7.41062164e-01 -3.53058010e-01 1.22505417e-02 7.81720340e-01 -1.09424457e-01 9.89561796e-01 3.19670051e-01 -1.25338688e-01 1.80116529e-03 2.92759966e-02 5.59359610e-01 1.07597537e-01 -3.89269799e-01 -7.04522729e-01 -6.64723158e-01 1.02103245e+00 1.01970100e+00 -6.49098456e-01 3.76689471e-02 -7.76986599e-01 6.47003055e-01 -4.66392905e-01 -6.68020666e-01 -6.07085407e-01 -2.78335452e-01 -3.21669668e-01 3.47388685e-01 -4.77312177e-01 -6.89469576e-02 -2.47714356e-01 -1.55851781e-01 5.37233472e-01 4.90010291e-01 1.41331339e+00 8.50905895e-01 -4.80262935e-01 -1.97071701e-01 -6.66056633e-01 1.13364756e+00 -1.74818769e-01 -9.15750980e-01 -2.87066817e-01 -1.40895927e+00 7.12641895e-01 8.25775504e-01 -3.51479203e-01 -3.55610073e-01 3.43470946e-02 -6.71405077e-01 -1.93108514e-01 4.83319610e-01 1.75447732e-01 6.07987046e-01 -1.09208560e+00 -1.02830791e+00 5.11001587e-01 1.11382537e-01 -6.92502081e-01 -5.86618781e-01 5.66863477e-01 -1.59772432e+00 3.49111795e-01 -7.63803244e-01 2.06377119e-01 -9.65417385e-01 4.46354240e-01 8.42165127e-02 1.42277023e-02 -1.00430988e-01 7.63427496e-01 -4.94263738e-01 -4.93018061e-01 -4.32428479e-01 -1.77580845e-02 -2.97288716e-01 4.11927998e-01 5.13895571e-01 6.97109818e-01 5.09585321e-01 -1.63244212e+00 -8.14247131e-01 3.41214508e-01 -9.72471237e-02 -1.06840014e+00 7.14464724e-01 -4.02868301e-01 -7.96981871e-01 4.87992585e-01 4.31978494e-01 1.47962713e+00 3.86805125e-02 3.43312830e-01 4.29067522e-01 -3.10990196e-02 -4.86747384e-01 -1.36193776e+00 -3.54005694e-01 7.32811809e-01 1.84513032e-01 -4.31370139e-01 9.06553388e-01 -4.86865938e-01 1.00647509e+00 4.40471768e-01 2.95305878e-01 -1.57722998e+00 -4.82368976e-01 1.74013579e+00 7.51696348e-01 -3.24518919e-01 -3.89751554e-01 -6.05567276e-01 -6.55819476e-01 1.31577432e+00 2.35223278e-01 -3.35429490e-01 9.07691598e-01 5.20547152e-01 8.33946407e-01 -9.42828655e-02 1.99183747e-01 -4.42638457e-01 -9.98823196e-02 8.44463527e-01 6.51119292e-01 4.32816714e-01 -1.38190067e+00 1.02718210e+00 -1.17263889e+00 -1.00044847e+00 9.43515182e-01 1.03457153e+00 -3.49191040e-01 -1.48732889e+00 -8.66901815e-01 2.25539863e-01 -1.07376027e+00 -8.05342972e-01 -4.01628882e-01 1.37637210e+00 2.15811625e-01 1.07353044e+00 9.89901349e-02 1.26017228e-01 2.36488491e-01 7.18427896e-01 9.39161003e-01 -8.71176898e-01 -8.64352524e-01 6.54089928e-01 4.52707738e-01 -6.50680363e-02 -2.19110191e-01 -1.15047872e+00 -1.88364959e+00 -4.10087734e-01 -2.85444766e-01 5.28626323e-01 2.74588794e-01 1.10623801e+00 -5.70713758e-01 2.66838279e-02 -2.01564938e-01 -4.96953636e-01 -2.57651508e-01 -1.03612435e+00 -8.11472356e-01 -1.07649811e-01 -4.42031056e-01 -5.21629870e-01 -1.34276167e-01 1.55484259e-01]

[10.37196159362793, 10.423514366149902]

f2c05455-05d9-4108-8f3c-3783db4a482f

needle-match-reliable-patch-matching-under

http://openaccess.thecvf.com/content_cvpr_2016/html/Lotan_Needle-Match_Reliable_Patch_CVPR_2016_paper.html

http://openaccess.thecvf.com/content_cvpr_2016/papers/Lotan_Needle-Match_Reliable_Patch_CVPR_2016_paper.pdf

Needle-Match: Reliable Patch Matching Under High Uncertainty

Reliable patch-matching forms the basis for many algorithms (super-resolution, denoising, inpainting, etc.) However, when the image quality deteriorates (by noise, blur or geometric distortions), the reliability of patch-matching deteriorates as well. Matched patches in the degraded image, do not necessarily imply similarity of the underlying patches in the (unknown) high-quality image. This restricts the applicability of patch-based methods. In this paper we present a patch representation called "Needle", which consists of small multi-scale versions of the patch and its immediate surrounding region. While the patch at the finest image scale is severely degraded, the degradation decreases dramatically in coarser needle scales, revealing reliable information for matching. We show that the Needle is robust to many types of image degradations, leads to matches faithful to the underlying high-quality patches, and to improvement in existing patch-based methods.

['Michal Irani', 'Or Lotan']

2016-06-01

cvpr-2016-6

['patch-matching']

['computer-vision']

[ 5.57692885e-01 -3.11446488e-02 3.32453161e-01 2.82685369e-01 -7.76582301e-01 -4.78229940e-01 2.95486212e-01 -1.35370836e-01 3.22977811e-01 9.79864240e-01 3.06633145e-01 3.94855142e-01 -2.59921551e-01 -8.42443407e-01 -6.96970522e-01 -9.93645847e-01 4.75646704e-02 8.41401294e-02 6.95061386e-01 -4.54072595e-01 2.30501801e-01 6.25542283e-01 -1.63686109e+00 5.68686306e-01 7.93146908e-01 8.37802410e-01 3.53468806e-01 7.67333806e-01 1.79646194e-01 3.64299774e-01 -6.01742685e-01 -1.05472960e-01 5.02859771e-01 -6.83911324e-01 -6.80632353e-01 5.54235160e-01 9.40571547e-01 -2.13951916e-01 -3.34652007e-01 1.45105863e+00 6.79622814e-02 4.16006036e-02 4.39753026e-01 -6.71893775e-01 -6.41118169e-01 -2.00705767e-01 -7.72015810e-01 1.33499473e-01 6.67607248e-01 -1.97171509e-01 5.79445481e-01 -8.61728728e-01 9.14899886e-01 1.27727163e+00 1.07859206e+00 1.46397501e-01 -1.67617834e+00 3.42272371e-02 -3.79215896e-01 1.68862805e-01 -1.37229598e+00 -4.92960423e-01 9.25360978e-01 -2.11834684e-01 3.30860466e-01 5.51184297e-01 5.13012528e-01 5.92955172e-01 6.26463413e-01 2.24095970e-01 1.53004229e+00 -5.28070033e-01 1.82454556e-01 -1.42144151e-02 -3.70958328e-01 4.15592551e-01 2.15378821e-01 4.95869100e-01 -3.46132398e-01 -4.13491517e-01 1.31705832e+00 1.17948331e-01 -7.47446179e-01 -3.90787572e-01 -1.16148007e+00 1.86278969e-01 4.16189045e-01 7.98203111e-01 -8.13833296e-01 -1.47432119e-01 -9.83308032e-02 6.81739628e-01 6.92686021e-01 2.65763104e-01 -1.12658553e-01 1.48677692e-01 -1.20016515e+00 4.41795826e-01 3.92426431e-01 6.78571224e-01 1.12111354e+00 -3.80140729e-02 2.92855371e-02 8.45125377e-01 -1.52540371e-01 5.16871035e-01 2.00741991e-01 -1.37759590e+00 1.47431925e-01 2.60140985e-01 5.32100081e-01 -1.36242867e+00 -7.18804523e-02 -6.12743020e-01 -1.16471207e+00 6.76166236e-01 5.07085204e-01 6.02898717e-01 -6.58616722e-01 1.23695385e+00 4.25189823e-01 2.12940931e-01 -1.52156919e-01 8.06870639e-01 3.55354160e-01 7.73397386e-01 -7.11485982e-01 -5.71239412e-01 1.07723033e+00 -7.51612246e-01 -9.49459374e-01 -1.78695962e-01 -2.78482676e-01 -1.24009752e+00 9.22845781e-01 6.10556006e-01 -1.53465915e+00 -9.20319557e-01 -9.51131761e-01 1.65118605e-01 1.50572613e-01 -5.13786256e-01 -3.00314695e-01 2.83358932e-01 -1.41983092e+00 1.00938761e+00 -2.25127906e-01 -4.10028428e-01 2.24625215e-01 -1.41145304e-01 -7.77027190e-01 -6.54170990e-01 -7.27037847e-01 9.45151150e-01 -1.62101984e-01 -7.18329921e-02 -6.72308803e-01 -6.53376818e-01 -5.31019926e-01 5.20679122e-03 4.42414433e-02 -6.07455909e-01 7.34764278e-01 -1.21291029e+00 -1.02024496e+00 8.79228294e-01 -2.82201618e-01 2.28814874e-03 5.43868065e-01 7.44283125e-02 -5.60828984e-01 5.87264955e-01 2.51701951e-01 8.41075853e-02 1.54862750e+00 -1.86169803e+00 -4.07918513e-01 -3.89938027e-01 -2.53898174e-01 5.06430194e-02 3.05627286e-01 -1.12269424e-01 -8.48377794e-02 -9.66698825e-01 6.02094352e-01 -4.93394494e-01 -2.43582621e-01 3.57550442e-01 -6.39353022e-02 5.83069026e-01 9.33984399e-01 -9.93362904e-01 9.77176249e-01 -2.58206773e+00 8.73494893e-02 1.49964213e-01 2.28559062e-01 1.42829604e-02 -4.48918134e-01 6.53759897e-01 -3.36627543e-01 -1.85464635e-01 -6.00499928e-01 -1.59811392e-01 -5.27365029e-01 4.25600022e-01 -2.19117716e-01 8.61190140e-01 1.79503292e-01 6.18428886e-01 -7.19609201e-01 -4.18226928e-01 1.26559913e-01 5.82647204e-01 -2.33339638e-01 2.11520359e-01 3.44820827e-01 6.81909621e-01 -2.02859551e-01 7.23961174e-01 1.17667782e+00 1.29377738e-01 -3.92924368e-01 -5.20070672e-01 -2.47544438e-01 -3.13418299e-01 -1.31387162e+00 1.53527713e+00 -3.42630923e-01 5.43369174e-01 8.63601208e-01 -1.02837002e+00 9.79837120e-01 5.11971474e-01 4.87619519e-01 -9.01083112e-01 -4.09050643e-01 3.92214417e-01 -3.51948023e-01 -5.61213553e-01 4.97541010e-01 -5.97140849e-01 3.97824079e-01 2.18759999e-01 -4.53108817e-01 -3.72410268e-01 -3.61584723e-02 9.94823501e-03 1.21944654e+00 -4.55150567e-02 3.61794651e-01 -4.62728590e-01 5.87797165e-01 5.04664853e-02 4.61379200e-01 6.38804555e-01 -1.83321238e-01 1.38604331e+00 2.21972406e-01 -3.45542818e-01 -1.58399391e+00 -1.10165393e+00 -3.96827787e-01 2.68280804e-01 3.74304414e-01 -4.51933630e-02 -9.13461864e-01 -1.45246655e-01 -2.14524623e-02 2.07301211e-02 -8.01323175e-01 -1.06122054e-01 -5.95111370e-01 -4.34652984e-01 2.67913342e-02 7.90950805e-02 6.41808629e-01 -8.46080303e-01 -3.07254344e-01 5.07362425e-01 -4.43974137e-01 -9.55783188e-01 -4.36955631e-01 -3.44214648e-01 -1.41031241e+00 -1.07127285e+00 -1.19111168e+00 -6.85718060e-01 8.37696373e-01 6.96269751e-01 1.35186267e+00 4.10051048e-01 -2.68476725e-01 3.52120131e-01 -3.75410855e-01 5.96728206e-01 -6.82709336e-01 -9.32224035e-01 -2.03194514e-01 3.73537362e-01 -7.15576589e-01 -1.03356528e+00 -6.46032155e-01 5.49192548e-01 -1.21175385e+00 -1.71985060e-01 6.20049715e-01 1.13089895e+00 9.30986047e-01 7.49215066e-01 1.84476778e-01 -6.98311031e-01 4.28719223e-01 -1.76284298e-01 -2.59788424e-01 1.71018839e-02 -5.47006786e-01 -2.81619877e-01 6.91197813e-01 -3.46728593e-01 -1.02677095e+00 -2.98248857e-01 -1.11259678e-02 -4.65673357e-01 -2.98553616e-01 1.29207209e-01 -1.86237127e-01 -5.20229936e-01 9.39283967e-01 5.30828178e-01 1.72918543e-01 -9.68352973e-01 2.07436338e-01 2.77121902e-01 9.28745687e-01 -2.79513747e-01 1.09563279e+00 8.50705385e-01 1.74184397e-01 -9.20101166e-01 -5.13056636e-01 -4.94683683e-01 -6.57849550e-01 -2.29349717e-01 3.94674540e-01 -5.98854780e-01 4.35966216e-02 3.41949105e-01 -1.09298301e+00 -1.73959777e-01 -7.36470163e-01 7.63552710e-02 -7.31327653e-01 7.54380345e-01 -6.29086375e-01 -5.65562308e-01 1.06840491e-01 -8.15267682e-01 1.04626739e+00 1.11853115e-01 1.20806694e-02 -9.30375636e-01 1.74953014e-01 3.72405678e-01 7.46264696e-01 4.18113500e-01 9.11270976e-01 5.01998365e-01 -6.86060786e-01 -4.24464434e-01 -2.87570804e-01 5.91177046e-01 3.94557148e-01 -2.25952834e-01 -8.31536889e-01 -4.98200685e-01 5.84546506e-01 2.57084340e-01 5.69826424e-01 6.66045904e-01 6.19981050e-01 -5.02165020e-01 -1.81743056e-01 4.47096467e-01 1.86469901e+00 -2.15707403e-02 1.23788393e+00 3.52740556e-01 2.69580543e-01 7.45694041e-01 7.13810444e-01 1.01805441e-01 -3.56044620e-01 9.47677314e-01 4.55409437e-01 -4.82196242e-01 -7.42683887e-01 -1.71740516e-03 2.93129414e-01 4.52007294e-01 -2.14201808e-01 1.58157453e-01 -3.64681542e-01 7.00579405e-01 -1.62045872e+00 -1.08998251e+00 -4.58327442e-01 2.39979219e+00 8.37822020e-01 -2.46023610e-01 5.35678044e-02 5.55378377e-01 9.13330495e-01 1.56973511e-01 -2.27199927e-01 -2.76792914e-01 -5.87686598e-01 2.31857032e-01 1.85070246e-01 8.97900522e-01 -6.39148414e-01 4.36634809e-01 7.53151083e+00 1.08058405e+00 -6.99188709e-01 3.41883868e-01 5.70170522e-01 3.73859406e-01 -4.29957032e-01 1.19277306e-01 -8.57109651e-02 4.54842418e-01 5.07902324e-01 -8.04031789e-02 5.24275899e-01 2.74460495e-01 4.70163643e-01 -4.79517639e-01 -7.18648970e-01 1.02322578e+00 8.51349309e-02 -1.19908786e+00 -1.50681064e-01 1.39891386e-01 1.09070528e+00 -4.55884963e-01 6.37234822e-02 -4.70138788e-01 -3.72159213e-01 -9.37966883e-01 6.24705315e-01 8.17596614e-01 8.64594579e-01 -6.52118921e-01 7.91977167e-01 3.03020269e-01 -1.05202532e+00 1.90344751e-01 -6.38311744e-01 -4.01376374e-02 3.25246513e-01 1.22858810e+00 2.42847018e-02 6.38819695e-01 8.62341881e-01 7.31133342e-01 -5.07319212e-01 1.15741646e+00 1.22245461e-01 1.46167472e-01 -7.81316832e-02 9.37176824e-01 -6.10856861e-02 -6.44425154e-01 9.92046177e-01 8.29666853e-01 5.06806374e-01 2.89379209e-01 -2.13430338e-02 9.49457467e-01 3.54332566e-01 1.99824627e-02 -6.67659879e-01 4.20798689e-01 1.53397515e-01 1.10761428e+00 -5.44136703e-01 -5.59331536e-01 -4.18865204e-01 1.42959201e+00 -2.63966382e-01 7.21406996e-01 -1.89329252e-01 -1.09248914e-01 3.94092232e-01 6.61028206e-01 2.85620838e-01 3.58805247e-03 -3.53706479e-01 -9.73963201e-01 2.79427230e-01 -1.11196578e+00 -1.77392233e-02 -1.08266985e+00 -1.32969058e+00 7.24234581e-01 -2.05319583e-01 -1.65040326e+00 3.21365409e-02 -7.90977553e-02 -6.14228368e-01 1.15088463e+00 -1.35358417e+00 -8.90686214e-01 -3.64207447e-01 7.65061736e-01 3.74857485e-01 2.26646751e-01 6.20862544e-01 3.36863846e-01 7.65799284e-02 -3.79213095e-02 6.33512318e-01 -5.34925282e-01 6.77503943e-01 -1.00109696e+00 5.83753698e-02 1.02545905e+00 -1.11317120e-01 1.32698670e-01 1.23960733e+00 -7.20020533e-01 -1.13360131e+00 -7.86872029e-01 8.31431687e-01 -3.79216336e-02 3.29842925e-01 2.53961354e-01 -1.35730398e+00 1.08044259e-01 2.42273569e-01 3.21877152e-01 8.43046159e-02 -4.89135742e-01 -3.79337549e-01 -3.83201510e-01 -1.66574264e+00 2.09609374e-01 7.53494442e-01 -6.25253975e-01 -5.73895872e-01 2.01903462e-01 1.37485385e-01 -9.79035795e-02 -1.05679512e+00 3.04911554e-01 5.46758771e-01 -1.68695605e+00 1.25467002e+00 -7.47914389e-02 4.76051927e-01 -7.30641305e-01 -2.50527233e-01 -1.35485196e+00 -8.23374629e-01 -3.85932595e-01 6.33321181e-02 9.73672450e-01 1.08368918e-02 -5.33670723e-01 6.79235399e-01 5.51399514e-02 -1.40681446e-01 -4.76944387e-01 -1.11047125e+00 -1.04948056e+00 -2.20542863e-01 1.94998637e-01 3.58886570e-01 9.19311345e-01 -2.62422115e-01 -1.53848916e-01 -3.66077423e-01 3.67428303e-01 1.13779974e+00 4.18562949e-01 4.28758979e-01 -1.11188495e+00 -4.55200195e-01 -4.40368861e-01 -5.38914859e-01 -9.17588770e-01 -5.80267668e-01 -2.18514293e-01 1.56391665e-01 -1.58568454e+00 1.93876430e-01 -4.15609717e-01 1.33129638e-02 7.26625696e-02 -1.73758566e-01 9.02247787e-01 -2.84002740e-02 5.99635363e-01 -9.22142267e-02 2.97874391e-01 1.63080418e+00 -3.53399999e-02 5.82766496e-02 -1.36460476e-02 -3.42763692e-01 4.54609156e-01 4.34837997e-01 -4.46134686e-01 -3.90936919e-02 -6.17003404e-02 1.32910728e-01 3.81007344e-01 5.32810032e-01 -1.13882709e+00 -2.00024992e-01 -9.97210480e-03 5.00473797e-01 -3.63436192e-01 4.82241422e-01 -9.94016290e-01 8.54036927e-01 5.21104991e-01 6.45980611e-03 4.99391370e-02 -3.92723735e-03 6.92564845e-01 -6.86307549e-01 -3.58234525e-01 1.35637915e+00 -3.51412028e-01 -2.92705059e-01 9.60313305e-02 -2.99598306e-01 -2.99882412e-01 6.28462076e-01 -8.24937761e-01 -8.97818729e-02 -5.88829875e-01 -9.71891761e-01 -6.86121166e-01 1.21479058e+00 -1.65934578e-01 8.03790808e-01 -1.50803292e+00 -8.98229063e-01 3.33770365e-01 -1.28528878e-01 -7.29630515e-02 5.85490584e-01 9.78701591e-01 -7.32227266e-01 -2.36628473e-01 -3.99163663e-01 -5.92783093e-01 -1.42760646e+00 6.71651781e-01 4.19473797e-01 -1.21313237e-01 -1.00275934e+00 7.31274843e-01 3.87465686e-01 3.58475059e-01 -7.27184191e-02 -4.37951088e-02 7.10931048e-02 -2.79042900e-01 9.91082072e-01 4.68680024e-01 1.83589563e-01 -8.84125590e-01 -1.03563117e-02 1.11899531e+00 1.62850991e-01 -1.37930647e-01 1.29646337e+00 -6.32579207e-01 -6.38907313e-01 3.16379070e-01 1.02198672e+00 2.85759956e-01 -1.40233207e+00 -3.93139809e-01 -3.39738458e-01 -1.14178085e+00 3.10620725e-01 -5.46603084e-01 -1.24420905e+00 6.97321117e-01 7.62090743e-01 6.88892961e-01 1.63460779e+00 4.97957952e-02 7.28551507e-01 -3.43791306e-01 4.81392384e-01 -7.36254752e-01 1.24498695e-01 4.34974059e-02 1.37344682e+00 -8.03269088e-01 2.40863681e-01 -6.86720967e-01 2.71335766e-02 1.05064738e+00 6.29151124e-04 -4.74267036e-01 5.34955204e-01 3.30101341e-01 5.37144393e-02 -1.66471750e-01 -5.05965710e-01 -3.07530928e-02 3.08977246e-01 1.01682162e+00 5.83769605e-02 -2.44300112e-01 -2.65316486e-01 -4.22959030e-02 2.12611854e-01 -2.41966620e-01 6.70876145e-01 7.34042168e-01 -7.14483380e-01 -1.09804666e+00 -1.35962915e+00 1.27185777e-01 -2.94685602e-01 -6.95461184e-02 -1.49020568e-01 4.23159987e-01 3.16201866e-01 8.68999302e-01 -4.79904301e-02 -1.36915930e-02 5.53473353e-01 -3.26458454e-01 8.72210264e-01 -3.05384487e-01 -2.88995534e-01 5.09129882e-01 -7.09509328e-02 -8.80861461e-01 -6.06798351e-01 -7.01567054e-01 -6.61485732e-01 -4.00763363e-01 -2.17452481e-01 2.36762711e-03 2.17811376e-01 6.34683311e-01 1.53996035e-01 2.23579690e-01 7.65893817e-01 -1.16087830e+00 -1.28059059e-01 -9.20649290e-01 -1.22518694e+00 7.14362264e-01 7.76833773e-01 -5.36421537e-01 -7.05392778e-01 3.45483303e-01]

[11.287572860717773, -2.2264842987060547]

9365008f-f179-401c-804a-4cb108fa2c72

dueqnet-dual-equivariance-network-in-outdoor

2302.13577

https://arxiv.org/abs/2302.13577v1

https://arxiv.org/pdf/2302.13577v1.pdf

DuEqNet: Dual-Equivariance Network in Outdoor 3D Object Detection for Autonomous Driving

Outdoor 3D object detection has played an essential role in the environment perception of autonomous driving. In complicated traffic situations, precise object recognition provides indispensable information for prediction and planning in the dynamic system, improving self-driving safety and reliability. However, with the vehicle's veering, the constant rotation of the surrounding scenario makes a challenge for the perception systems. Yet most existing methods have not focused on alleviating the detection accuracy impairment brought by the vehicle's rotation, especially in outdoor 3D detection. In this paper, we propose DuEqNet, which first introduces the concept of equivariance into 3D object detection network by leveraging a hierarchical embedded framework. The dual-equivariance of our model can extract the equivariant features at both local and global levels, respectively. For the local feature, we utilize the graph-based strategy to guarantee the equivariance of the feature in point cloud pillars. In terms of the global feature, the group equivariant convolution layers are adopted to aggregate the local feature to achieve the global equivariance. In the experiment part, we evaluate our approach with different baselines in 3D object detection tasks and obtain State-Of-The-Art performance. According to the results, our model presents higher accuracy on orientation and better prediction efficiency. Moreover, our dual-equivariance strategy exhibits the satisfied plug-and-play ability on various popular object detection frameworks to improve their performance.

['Xian Wei', 'Arafat Al-Jawari', 'Hai Lan', 'JiaMing Lei', 'Xihao Wang']

2023-02-27

['object-recognition']

['computer-vision']

[-3.81931007e-01 -2.85481185e-01 1.21004656e-01 -3.64713252e-01 -1.01999350e-01 -3.01211476e-01 4.96127635e-01 -2.05831498e-01 -3.72468054e-01 -8.41146186e-02 -3.01951349e-01 -4.14234161e-01 -1.02429770e-01 -8.87551665e-01 -7.55855918e-01 -7.84409642e-01 1.09030768e-01 3.80548164e-02 7.91826665e-01 -7.02275336e-01 2.21354038e-01 8.83416772e-01 -1.79220414e+00 -1.43658847e-01 8.01568270e-01 1.21147478e+00 6.09063923e-01 3.01222384e-01 8.26086011e-03 2.60056883e-01 -3.48164469e-01 -2.20597163e-01 5.64097941e-01 2.92719930e-01 1.96614474e-01 1.47979587e-01 3.66002291e-01 -4.54842716e-01 -7.10133195e-01 1.22434676e+00 4.37018126e-01 3.02486084e-02 5.73803008e-01 -1.49828374e+00 -2.20444635e-01 -8.09211060e-02 -6.33004487e-01 4.39702183e-01 -1.74685102e-02 4.47546184e-01 6.77577436e-01 -1.13585603e+00 3.90320688e-01 1.47189367e+00 4.46054131e-01 1.39768377e-01 -6.78791761e-01 -9.27628696e-01 5.48289418e-01 7.10036755e-01 -1.53515053e+00 -3.15739661e-01 1.02899551e+00 -2.72283524e-01 8.67904902e-01 1.84281111e-01 6.40965521e-01 6.59732342e-01 5.27138233e-01 7.16379344e-01 8.19636405e-01 9.26173255e-02 -1.19130261e-01 1.56623587e-01 1.20687038e-01 7.24256396e-01 4.40997958e-01 2.22423613e-01 -1.65269017e-01 5.37946761e-01 6.58792675e-01 2.85028517e-01 9.64242816e-02 -5.10238349e-01 -1.13008869e+00 5.19181848e-01 8.10334623e-01 -4.27089371e-02 -8.22161585e-02 -6.14356063e-02 2.26072460e-01 1.45151764e-01 2.45893970e-01 -1.25872821e-01 -1.06512122e-01 2.15470791e-01 -3.02168965e-01 3.32445592e-01 2.70335197e-01 1.31543612e+00 8.06639135e-01 1.25176936e-01 -8.60254914e-02 4.70686197e-01 5.24464369e-01 8.49712253e-01 1.44226655e-01 -6.86602771e-01 6.85538650e-01 8.35537493e-01 1.26525939e-01 -1.56462467e+00 -6.30285442e-01 -9.22807455e-01 -8.94287944e-01 5.68033338e-01 -1.89403445e-02 3.48081350e-01 -8.44369709e-01 1.51648402e+00 7.34650314e-01 3.39483052e-01 1.14797369e-01 1.20897007e+00 8.67254555e-01 6.24222517e-01 -1.26457155e-01 -3.32697295e-02 1.47956395e+00 -7.92632818e-01 -4.82654601e-01 -2.07490355e-01 4.16427732e-01 -8.92735779e-01 7.20625043e-01 1.11966863e-01 -5.64987481e-01 -1.03588736e+00 -1.39126587e+00 -1.18035294e-01 -4.34179097e-01 2.36925721e-01 3.83398384e-01 5.06513476e-01 -6.55995846e-01 -2.94537153e-02 -8.31659377e-01 -2.26920322e-01 3.59715074e-01 2.34221935e-01 -4.01664466e-01 -2.13404909e-01 -1.03730035e+00 1.03676927e+00 2.92451560e-01 4.57911253e-01 -8.47677767e-01 -5.68098307e-01 -7.30687618e-01 -5.28112724e-02 5.15808284e-01 -5.48040450e-01 9.45277989e-01 -1.75059468e-01 -1.10828316e+00 6.69379234e-01 -2.65925556e-01 -4.61115777e-01 8.56319726e-01 -1.61757454e-01 -5.49137235e-01 -1.05458103e-01 1.69755816e-01 5.34047246e-01 8.28114629e-01 -1.16748512e+00 -1.02938902e+00 -5.24289012e-01 2.18324900e-01 6.23372495e-01 -1.12238325e-01 -3.20681542e-01 -7.21276879e-01 -2.79406816e-01 6.60172999e-01 -7.70350397e-01 -2.16860592e-01 1.31018415e-01 -1.98258400e-01 -6.01563752e-01 1.36896026e+00 -1.57911241e-01 8.45153809e-01 -2.63986635e+00 -1.59375936e-01 3.45010459e-02 4.86139864e-01 7.27579594e-02 5.69064841e-02 -4.09679161e-03 2.27634296e-01 -2.77593046e-01 1.45793632e-01 -2.25634128e-01 2.01255217e-01 2.07521468e-01 -2.97341645e-01 6.36826992e-01 4.81767863e-01 7.24345267e-01 -7.17736304e-01 -6.58249617e-01 6.53121710e-01 3.30807686e-01 -5.96482813e-01 1.91940650e-01 1.34393990e-01 1.45516410e-01 -8.37126434e-01 6.69310868e-01 1.22580755e+00 3.07294011e-01 -5.10555983e-01 -4.29930538e-01 -5.05220473e-01 -1.23679452e-02 -1.51968551e+00 1.26756799e+00 -3.03795636e-01 4.94338155e-01 2.20891014e-01 -9.90265131e-01 1.19532239e+00 -1.76720694e-01 3.91840965e-01 -8.45443308e-01 2.20994487e-01 3.75722110e-01 1.25270277e-01 -5.36102057e-01 4.81324077e-01 2.04883084e-01 -2.85724141e-02 -2.79592574e-01 -3.44316363e-01 -2.26457447e-01 -8.10937211e-02 1.35130331e-01 7.97611713e-01 4.46737446e-02 -1.12847693e-01 -2.27287233e-01 7.83200920e-01 -3.47669865e-03 5.85926771e-01 6.50622547e-01 -4.51182723e-01 3.19108725e-01 1.98374406e-01 -4.01590109e-01 -9.11893725e-01 -1.13958549e+00 -4.51227903e-01 5.27145088e-01 9.80292380e-01 -5.48858754e-02 -4.84076768e-01 -6.33881450e-01 2.91310281e-01 5.21322548e-01 -2.16808319e-01 -5.47958255e-01 -5.63551366e-01 -6.09335244e-01 1.95062354e-01 4.15822685e-01 1.11906254e+00 -6.02289021e-01 -7.09431827e-01 1.77856550e-01 -4.54239286e-02 -1.46587133e+00 -3.41157258e-01 -8.40026364e-02 -5.84531844e-01 -8.26414526e-01 -2.13420942e-01 -7.96476781e-01 5.16685307e-01 1.09838867e+00 5.05125701e-01 -6.22749068e-02 -1.74239382e-01 1.78738087e-01 -2.19856679e-01 -7.35049605e-01 -2.39814501e-02 -8.64201132e-03 3.70563895e-01 2.07387015e-01 6.27562344e-01 -5.69214702e-01 -8.65687490e-01 7.74077535e-01 -5.29833078e-01 1.34781092e-01 8.25886428e-01 3.28968376e-01 6.02891088e-01 3.39461029e-01 2.26606280e-01 1.42298406e-03 1.43601790e-01 -3.86067122e-01 -8.76999617e-01 -2.52722681e-01 -4.90551114e-01 -2.44596869e-01 3.85000825e-01 -2.83391923e-01 -8.23686719e-01 1.32797360e-01 -1.82200193e-01 -6.22905612e-01 -1.17851622e-01 7.12010711e-02 -6.10203385e-01 -3.63404989e-01 3.00701946e-01 3.95796508e-01 -1.86979026e-02 -4.39694881e-01 1.44111410e-01 6.63415074e-01 5.03836870e-01 -2.15859979e-01 1.21077776e+00 6.36331081e-01 4.61535215e-01 -8.72271478e-01 -5.60969710e-01 -5.01168489e-01 -5.43369055e-01 -6.09153986e-01 9.37542439e-01 -1.21849239e+00 -1.09078753e+00 5.66757202e-01 -1.24893010e+00 3.42162639e-01 1.10428162e-01 7.26829529e-01 -3.06413203e-01 3.73508215e-01 -4.77580428e-02 -8.03610682e-01 5.43186441e-02 -1.47239125e+00 1.21161270e+00 2.99558967e-01 7.36287594e-01 -3.13811332e-01 -4.58799332e-01 1.45112246e-01 2.78405070e-01 1.47545755e-01 6.23242795e-01 -3.04623663e-01 -1.07104445e+00 -3.93555731e-01 -5.70856631e-01 1.95757478e-01 -1.55312672e-01 -1.19004928e-01 -9.22857583e-01 -1.29500255e-01 1.99381664e-01 3.27206314e-01 8.57759118e-01 1.02230862e-01 1.10615373e+00 -2.01928969e-02 -4.06120181e-01 7.17008412e-01 1.15954065e+00 1.23030156e-01 3.99821311e-01 4.35855836e-01 8.49374652e-01 5.60924113e-01 1.03990400e+00 2.38752663e-01 6.74740374e-01 7.27475286e-01 9.27483916e-01 -1.41733989e-01 -1.72708720e-01 -2.98485249e-01 4.05580401e-01 6.88212872e-01 -2.60844678e-01 -3.42684798e-02 -6.50022566e-01 3.40723634e-01 -1.88874173e+00 -8.10289562e-01 -4.43618029e-01 1.97671735e+00 7.52754584e-02 7.41218925e-01 -1.30132154e-01 -2.16992162e-02 7.12806463e-01 1.78342402e-01 -7.38057733e-01 -5.01119792e-02 -2.26411492e-01 -4.66921151e-01 7.15209305e-01 3.69690329e-01 -1.10400522e+00 9.58113909e-01 4.75632334e+00 1.08412015e+00 -1.18327761e+00 1.02373883e-01 2.37985909e-01 1.41983345e-01 -5.53499684e-02 -1.56178445e-01 -1.29299617e+00 3.19140226e-01 2.62779057e-01 -6.19547367e-02 5.07337712e-02 1.16847348e+00 4.94642496e-01 1.83979079e-01 -7.49278545e-01 1.01572287e+00 -7.46897981e-02 -1.07061625e+00 7.88691789e-02 1.40802532e-01 3.45052510e-01 2.21774876e-01 2.18535855e-01 3.91131788e-01 -1.66435957e-01 -6.32266641e-01 1.08075500e+00 4.73385096e-01 3.72590750e-01 -8.08874905e-01 8.32722425e-01 6.05252624e-01 -1.58489525e+00 -7.42347464e-02 -7.30500519e-01 -8.94252509e-02 1.08354844e-01 5.23607373e-01 -7.86327958e-01 7.38206387e-01 8.65964174e-01 8.32190037e-01 -6.24519408e-01 1.08552921e+00 -4.03551646e-02 1.86274618e-01 -5.79603016e-01 -1.69082701e-01 3.88942957e-01 -3.63891035e-01 1.08196974e+00 1.02535248e+00 4.21330959e-01 1.90405801e-01 3.85143727e-01 8.02638173e-01 2.23933518e-01 1.17589816e-01 -6.81985378e-01 5.07155776e-01 3.96617770e-01 1.40328932e+00 -4.93995309e-01 -2.30030194e-01 -5.04201055e-01 4.77135122e-01 6.07052771e-03 4.07342017e-01 -1.01897347e+00 -4.70633149e-01 9.50504780e-01 2.72747636e-01 4.81206745e-01 -7.46479928e-01 -1.65429920e-01 -1.04947484e+00 4.09851104e-01 -5.00503540e-01 -1.44026689e-02 -7.28959203e-01 -9.21057999e-01 4.61409271e-01 1.13974825e-01 -1.65120065e+00 3.21524143e-01 -9.97627079e-01 -6.57673419e-01 7.50635624e-01 -1.77767205e+00 -1.17827427e+00 -7.44023263e-01 6.82746112e-01 5.29964983e-01 -1.37022614e-01 5.50733842e-02 4.76771802e-01 -6.95904851e-01 4.59994555e-01 -1.92284837e-01 -5.48866093e-02 5.34539104e-01 -7.85063982e-01 2.76681900e-01 1.07652843e+00 -2.27640033e-01 3.21639806e-01 7.37304568e-01 -5.04510164e-01 -1.71135032e+00 -1.21468282e+00 4.16466504e-01 -4.90802318e-01 5.16630173e-01 -5.95626593e-01 -7.47163951e-01 4.52597797e-01 -3.42365026e-01 2.00233430e-01 -2.49286622e-01 -2.35573500e-01 -3.34800988e-01 -7.29233325e-01 -9.32688653e-01 7.09054232e-01 1.37894177e+00 -1.81175396e-01 -4.41856712e-01 2.36966491e-01 1.08022380e+00 -7.18140721e-01 -5.92447817e-01 8.23083997e-01 5.25950253e-01 -9.35793340e-01 1.10174692e+00 -4.03589904e-01 8.25381055e-02 -9.68328416e-01 -4.22652870e-01 -6.42882526e-01 -5.51387727e-01 -2.10285634e-01 5.61551051e-03 1.00885558e+00 1.70152262e-01 -9.16433096e-01 6.79671705e-01 7.22526386e-02 -7.21804082e-01 -6.73750043e-01 -1.25290596e+00 -9.00340974e-01 -2.85479516e-01 -8.05768430e-01 8.52431297e-01 4.43469673e-01 -5.66836238e-01 2.77397901e-01 -7.60504231e-02 8.53047550e-01 7.43254483e-01 1.67709485e-01 1.10078108e+00 -1.21225810e+00 1.21889867e-01 -6.08020842e-01 -1.12671435e+00 -1.48699486e+00 -8.71079415e-02 -9.09649968e-01 7.76210278e-02 -1.19772005e+00 -3.57691795e-02 -5.14983654e-01 -4.72990692e-01 7.53360838e-02 -7.59755522e-02 2.50616521e-01 3.00148875e-01 2.18396813e-01 -6.95855677e-01 7.41962254e-01 1.53127420e+00 -2.98517078e-01 -1.27793968e-01 1.43144190e-01 -4.18517470e-01 7.15117693e-01 6.22917950e-01 -2.31507346e-01 -2.69544929e-01 -5.15390396e-01 5.85311092e-02 -3.55614275e-01 7.76081085e-01 -1.25860643e+00 3.87735665e-01 -2.35364243e-01 3.49904269e-01 -9.93376613e-01 5.18944800e-01 -1.08985376e+00 -1.90053090e-01 6.31298661e-01 3.96436810e-01 1.17039479e-01 2.91433930e-01 8.20794046e-01 -2.09077984e-01 1.24609970e-01 7.00717449e-01 1.74837798e-01 -1.15207541e+00 7.22433925e-01 -1.64589629e-01 -3.30718428e-01 1.15484607e+00 -4.16382432e-01 -2.81107157e-01 -6.81514069e-02 -3.19069952e-01 5.51406205e-01 2.69941270e-01 8.03141415e-01 8.34930658e-01 -1.49423826e+00 -8.28725040e-01 5.16440213e-01 4.57947135e-01 3.35330397e-01 3.68758768e-01 1.01541328e+00 -4.32674617e-01 3.76736492e-01 -1.51609853e-01 -1.14571428e+00 -1.11460114e+00 6.50508523e-01 5.11353910e-01 2.00078592e-01 -6.50133610e-01 4.11472440e-01 7.20510364e-01 -2.90986896e-01 7.80140460e-02 -3.94728661e-01 -3.53550047e-01 -2.29109347e-01 2.95529604e-01 3.62100601e-01 1.69539273e-01 -9.30365384e-01 -4.94405925e-01 9.81672645e-01 -7.57937431e-02 1.88375995e-01 1.01158428e+00 -3.18473279e-01 1.19810343e-01 2.23687530e-01 1.15653408e+00 1.21938117e-01 -1.38366568e+00 -1.09841108e-01 -5.23744583e-01 -5.53602457e-01 2.85943747e-01 -7.43311942e-02 -8.48431826e-01 1.04995680e+00 9.20235276e-01 1.80714786e-01 8.49027991e-01 -4.58623692e-02 5.60251892e-01 5.87488234e-01 4.86602783e-01 -7.40632415e-01 -6.13670684e-02 5.76273918e-01 8.19899857e-01 -1.38492334e+00 -1.00351050e-01 -7.37386823e-01 -2.62103796e-01 1.02970958e+00 9.34440255e-01 -3.14453602e-01 7.68526495e-01 -1.35342538e-01 -1.21377641e-02 -2.02691972e-01 -3.30777079e-01 -3.00625503e-01 4.94167417e-01 5.23458660e-01 -3.22146088e-01 7.00906962e-02 -2.20882595e-01 5.28410733e-01 -3.25814903e-01 -6.47126198e-01 6.80377558e-02 5.88762403e-01 -8.57564867e-01 -5.97548902e-01 -3.57422292e-01 1.03822365e-01 1.53674498e-01 1.89009219e-01 5.47900200e-02 1.01965463e+00 5.68999112e-01 9.78825033e-01 2.05440462e-01 -6.44498229e-01 8.61102164e-01 -3.67528021e-01 2.07183108e-01 -2.59328961e-01 1.84918106e-01 4.56977040e-02 -2.54234642e-01 -6.33128524e-01 -2.25197256e-01 -7.66145706e-01 -1.31132329e+00 -2.48550653e-01 -3.27983260e-01 -1.32420495e-01 8.54521751e-01 9.77612853e-01 5.37971139e-01 6.13333881e-01 1.01008546e+00 -8.19707990e-01 -5.39474070e-01 -7.33325601e-01 -4.69163418e-01 1.86653644e-01 4.23535645e-01 -1.01843011e+00 -4.06191409e-01 -4.66637552e-01]

[7.965858459472656, -2.008829355239868]

b2684b3b-827f-45ab-9a6f-35be6f876286

parameter-level-soft-masking-for-continual

2306.14775

https://arxiv.org/abs/2306.14775v1

https://arxiv.org/pdf/2306.14775v1.pdf

Parameter-Level Soft-Masking for Continual Learning

Existing research on task incremental learning in continual learning has primarily focused on preventing catastrophic forgetting (CF). Although several techniques have achieved learning with no CF, they attain it by letting each task monopolize a sub-network in a shared network, which seriously limits knowledge transfer (KT) and causes over-consumption of the network capacity, i.e., as more tasks are learned, the performance deteriorates. The goal of this paper is threefold: (1) overcoming CF, (2) encouraging KT, and (3) tackling the capacity problem. A novel technique (called SPG) is proposed that soft-masks (partially blocks) parameter updating in training based on the importance of each parameter to old tasks. Each task still uses the full network, i.e., no monopoly of any part of the network by any task, which enables maximum KT and reduction in capacity usage. To our knowledge, this is the first work that soft-masks a model at the parameter-level for continual learning. Extensive experiments demonstrate the effectiveness of SPG in achieving all three objectives. More notably, it attains significant transfer of knowledge not only among similar tasks (with shared knowledge) but also among dissimilar tasks (with little shared knowledge) while mitigating CF.

['Bing Liu', 'Gyuhak Kim', 'Zixuan Ke', 'Chihiro Ono', 'Mori Kurokawa', 'Tatsuya Konishi']

2023-06-26

['incremental-learning', 'transfer-learning']

['methodology', 'miscellaneous']

[ 3.08638304e-01 1.69481024e-01 -1.76185414e-01 1.90331116e-01 1.28822150e-02 -4.82525617e-01 1.92813680e-01 -3.53413261e-02 -7.36687899e-01 1.14728856e+00 -1.36500970e-01 -4.41582620e-01 -4.51106906e-01 -6.36630237e-01 -7.75943637e-01 -8.12338531e-01 -9.62232724e-02 2.42666900e-01 7.21716762e-01 -8.39317590e-02 1.94029734e-01 4.43225265e-01 -1.32955086e+00 -3.43318209e-02 1.17120779e+00 5.89273095e-01 6.85476482e-01 2.61052698e-01 -5.07396609e-02 9.18934822e-01 -6.86144650e-01 -3.40682834e-01 2.69398332e-01 -2.38192558e-01 -9.55952406e-01 -8.15947726e-02 3.03127140e-01 -3.93255800e-01 -4.14495558e-01 8.38374913e-01 4.24540550e-01 4.75688368e-01 4.72890645e-01 -1.33208489e+00 -7.67805099e-01 8.13725710e-01 -5.99048138e-01 4.45199430e-01 -4.02950287e-01 6.96479827e-02 5.68400621e-01 -7.24082053e-01 1.51633993e-01 1.04281390e+00 6.35435462e-01 7.34926403e-01 -1.20707309e+00 -8.28789830e-01 5.50782323e-01 6.37464747e-02 -1.32887328e+00 -4.55998987e-01 4.90173727e-01 -3.74206781e-01 1.11776423e+00 -3.63653228e-02 6.72250330e-01 6.89939022e-01 3.07255358e-01 8.23872685e-01 1.00906527e+00 -5.02131104e-01 2.35969022e-01 2.98271716e-01 2.39181131e-01 5.64579785e-01 4.80364174e-01 -4.60223332e-02 -9.24054861e-01 7.60247037e-02 9.59524453e-01 9.44081694e-02 -4.31554735e-01 -5.74781835e-01 -9.91701543e-01 3.94071817e-01 2.96956241e-01 4.20769393e-01 -3.80228788e-01 2.30908990e-01 3.69748682e-01 9.00592685e-01 3.96293133e-01 4.52507734e-01 -7.82172441e-01 1.59750395e-02 -7.59023368e-01 -1.38135403e-01 7.20195889e-01 9.28244770e-01 1.04266477e+00 3.64271343e-01 -2.40527049e-01 9.17559743e-01 -1.71966285e-01 2.48256579e-01 7.82054424e-01 -8.13346863e-01 6.61423862e-01 5.82198679e-01 6.97243139e-02 -5.59381962e-01 -4.16590780e-01 -7.81142056e-01 -8.51969063e-01 2.26248980e-01 3.94234002e-01 -4.84889299e-01 -8.70544255e-01 2.14325905e+00 1.25971183e-01 1.84918433e-01 5.82781360e-02 4.10097271e-01 1.02141082e-01 3.37433040e-01 2.07804382e-01 -4.83533263e-01 1.05947077e+00 -1.01600242e+00 -7.37793446e-01 -5.39753139e-01 5.91603160e-01 -5.12170374e-01 1.14764059e+00 3.97905678e-01 -1.19569409e+00 -7.08148062e-01 -1.10121918e+00 3.65919739e-01 -2.35859200e-01 -2.86584228e-01 5.61424375e-01 7.19038546e-01 -1.32357001e+00 6.15347207e-01 -7.27279842e-01 -1.18584134e-01 5.55482149e-01 5.95557690e-01 -8.49742964e-02 -1.65106341e-01 -1.28665078e+00 1.02177143e+00 7.25830317e-01 -2.22668082e-01 -1.04129016e+00 -1.03914344e+00 -3.52236480e-01 4.04714257e-01 9.04140711e-01 -9.32403803e-01 1.21527314e+00 -1.25230813e+00 -1.35362685e+00 1.03135034e-01 -2.33245362e-03 -5.11017323e-01 5.68785071e-01 -5.52826464e-01 -7.46322200e-02 -1.61641493e-01 -1.51135787e-01 7.32554853e-01 1.16500020e+00 -1.36734581e+00 -7.41434813e-01 -3.47725987e-01 2.96355605e-01 6.05415106e-01 -1.23774803e+00 -4.90470320e-01 -4.41989183e-01 -7.83625603e-01 -6.15446158e-02 -1.05709302e+00 1.84030026e-01 -1.57918379e-01 -9.16719437e-02 -3.46716911e-01 7.74155319e-01 -5.57399571e-01 1.47764671e+00 -2.26630235e+00 7.51016065e-02 2.20512412e-02 4.88126755e-01 6.31906807e-01 -2.65441269e-01 3.37550372e-01 5.06895892e-02 1.87318772e-01 -1.00363299e-01 -4.23312366e-01 -4.52100158e-01 4.94406879e-01 -1.46867871e-01 5.66695668e-02 -1.07990399e-01 9.75549579e-01 -9.35766399e-01 -2.35981330e-01 -1.84732199e-01 2.28831872e-01 -4.20766473e-01 -1.44815266e-01 -3.45387831e-02 2.17203543e-01 -2.66859829e-01 2.77667224e-01 6.45053387e-01 -2.98039228e-01 2.95902193e-01 1.30716085e-01 -1.15455864e-02 5.27439378e-02 -1.16429722e+00 1.27958179e+00 -6.23715997e-01 2.97948390e-01 5.30209951e-02 -9.98315096e-01 6.63919926e-01 4.01329190e-01 2.82841325e-01 -8.37262034e-01 -3.45476061e-01 8.85645598e-02 1.41844451e-01 -1.88302279e-01 4.06216681e-01 -2.63220463e-02 3.49801779e-01 6.73002958e-01 1.08758256e-01 3.73012453e-01 4.43792976e-02 4.37775612e-01 1.05499578e+00 -2.23465398e-01 1.31527171e-01 -3.76520783e-01 3.83391619e-01 -3.18136632e-01 6.40073478e-01 1.07448554e+00 -2.78949320e-01 -8.11589584e-02 3.89649242e-01 -1.21739835e-01 -9.78351235e-01 -1.05610061e+00 3.14563990e-01 1.52859974e+00 6.23361282e-02 1.75802186e-02 -5.84578931e-01 -8.11957657e-01 4.62315410e-01 7.58958399e-01 -6.62582755e-01 -6.25177383e-01 -6.77290976e-01 -4.55237418e-01 3.55420977e-01 5.51432192e-01 8.03614795e-01 -1.11858428e+00 -4.88832384e-01 2.18676448e-01 -1.06232904e-01 -5.06748617e-01 -8.24557602e-01 4.84667659e-01 -1.37841725e+00 -1.03129411e+00 -8.84313107e-01 -8.36838365e-01 8.57069254e-01 9.86949444e-01 8.45952690e-01 2.40226358e-01 1.59795880e-01 4.01509732e-01 -3.70888822e-02 -3.94004971e-01 -1.45533755e-01 4.53969955e-01 3.21384639e-01 -1.72834203e-01 1.10399514e-03 -6.92751944e-01 -4.86358494e-01 5.78309894e-01 -9.77468908e-01 -4.26448062e-02 8.47846925e-01 8.89250100e-01 3.09059471e-01 5.53948760e-01 1.27900612e+00 -1.04000556e+00 9.18981791e-01 -5.23327947e-01 -3.20819020e-01 5.52055299e-01 -1.13554597e+00 -2.95981914e-02 6.45164013e-01 -8.37443948e-01 -1.50658369e+00 -1.84987396e-01 5.90833366e-01 -5.14751434e-01 4.35376316e-01 4.83025700e-01 1.24471292e-01 -1.68201402e-01 6.95394099e-01 4.74872440e-01 2.19545946e-01 -4.15603518e-01 2.24207029e-01 3.56407732e-01 3.56204659e-01 -4.02553678e-01 8.82676959e-01 2.59205997e-01 -1.77426085e-01 -5.17954648e-01 -9.47024941e-01 -3.81497920e-01 -7.12689102e-01 -2.13506937e-01 2.11224273e-01 -1.01534271e+00 -6.53494120e-01 7.60484278e-01 -6.41486764e-01 -8.09487045e-01 -4.01346654e-01 4.15163934e-01 -3.01354587e-01 3.22554767e-01 -4.54848826e-01 -8.91993165e-01 -3.03063720e-01 -6.81619704e-01 1.90508068e-02 3.45730305e-01 -2.20673922e-02 -1.10658872e+00 -2.94184893e-01 2.52051204e-01 9.03715432e-01 -5.08716106e-01 1.09793711e+00 -4.98870373e-01 -5.48173904e-01 1.66986316e-01 -2.08040684e-01 6.91135585e-01 4.08019543e-01 -6.56014740e-01 -7.87129700e-01 -8.84078801e-01 9.52405259e-02 -5.28106391e-01 1.07791102e+00 2.50046343e-01 1.07550836e+00 -4.16153729e-01 -3.48852515e-01 1.96469739e-01 1.19356847e+00 3.87861222e-01 5.32877505e-01 4.97192025e-01 5.85753918e-01 4.67450947e-01 4.59936678e-01 3.61888349e-01 2.22576767e-01 3.42581898e-01 3.35048258e-01 5.55310696e-02 -4.63989198e-01 -2.24690869e-01 4.10205871e-01 8.93618524e-01 3.05422749e-02 -2.09949419e-01 -8.11561704e-01 7.61662960e-01 -2.04891348e+00 -6.09060287e-01 2.91396379e-01 2.35199022e+00 1.12678456e+00 3.63122880e-01 -9.61479694e-02 4.59281430e-02 7.34273434e-01 -1.03707924e-01 -1.14363122e+00 -1.14663854e-01 3.53387534e-03 -1.08085357e-01 5.12073517e-01 4.05543715e-01 -6.45045638e-01 8.71191084e-01 6.04024839e+00 9.64432657e-01 -7.74366379e-01 3.92519474e-01 4.12599474e-01 -3.25792700e-01 -1.25814214e-01 -2.97561493e-02 -9.71470296e-01 4.53554749e-01 6.87528908e-01 -5.75751662e-01 7.65783429e-01 8.63536477e-01 -1.56762049e-01 -3.36196423e-01 -8.11630189e-01 5.26471317e-01 5.28652482e-02 -9.04030740e-01 1.99880838e-01 -5.94377033e-02 1.04385555e+00 -8.03558603e-02 1.03750162e-01 5.89579761e-01 4.40281183e-01 -5.05919755e-01 6.91170394e-01 6.66528463e-01 6.44611061e-01 -9.93048668e-01 5.69984496e-01 7.25603044e-01 -9.60449874e-01 -5.79746008e-01 -4.78533059e-01 -8.83925706e-03 -5.07463291e-02 8.38025689e-01 -7.70718753e-01 3.76359344e-01 5.66174507e-01 2.84881175e-01 -6.63843155e-01 1.13961518e+00 -2.78110534e-01 7.45654285e-01 -1.57142282e-01 2.26907492e-01 -3.05430014e-02 1.70559525e-01 4.09883231e-01 8.97433937e-01 2.17070103e-01 -1.73969522e-01 2.72360593e-01 4.21506852e-01 -4.28957194e-01 -1.65347710e-01 -5.05864084e-01 1.24094583e-01 9.82253015e-01 9.15549934e-01 -6.57038331e-01 -3.99753213e-01 -2.61878401e-01 9.60611105e-01 7.20690250e-01 4.58299071e-01 -5.26663661e-01 -4.21018004e-01 4.21129256e-01 1.74445614e-01 2.37741038e-01 -3.62615824e-01 -3.62280935e-01 -6.87542439e-01 1.02731876e-01 -4.78046685e-01 3.74832690e-01 -5.11354268e-01 -1.29688203e+00 1.89776838e-01 5.97252473e-02 -6.64289415e-01 1.50103062e-01 -5.17772809e-02 -4.60632980e-01 1.00846565e+00 -1.76862907e+00 -6.77330554e-01 -3.18092823e-01 9.00135458e-01 5.85659444e-01 -3.17478061e-01 3.32937181e-01 3.73989373e-01 -6.87764406e-01 9.17065620e-01 3.60958338e-01 -5.20786822e-01 9.67598498e-01 -1.09963894e+00 1.27311587e-01 5.95679939e-01 -3.67457777e-01 9.33631659e-01 1.99735731e-01 -9.73043203e-01 -1.35178781e+00 -1.17545474e+00 8.08976650e-01 -1.34531870e-01 5.80980182e-01 -3.02266657e-01 -1.22056556e+00 5.97404659e-01 -1.20121285e-01 -5.14728546e-01 5.43266296e-01 3.31544608e-01 -1.98187560e-01 -3.28351825e-01 -9.52476859e-01 5.70798755e-01 1.27275705e+00 -3.29757750e-01 -4.18640882e-01 3.88677090e-01 9.98293459e-01 -2.54789535e-02 -6.39940917e-01 1.94864288e-01 4.03032154e-01 -7.32722402e-01 9.01163042e-01 -4.26953197e-01 -9.83272027e-03 -1.39732108e-01 3.87845099e-01 -1.53639781e+00 -7.08365500e-01 -5.61922669e-01 -6.66819155e-01 1.26804090e+00 4.00220215e-01 -1.03667581e+00 7.43700445e-01 5.36459863e-01 -2.75295675e-01 -6.94872379e-01 -9.05524433e-01 -1.34048963e+00 9.31289420e-02 -2.18617022e-02 3.24044168e-01 1.16624320e+00 -2.63713032e-01 4.55626398e-01 -6.53945029e-01 -3.15398872e-02 5.06004751e-01 -4.77717608e-01 5.34998536e-01 -1.39054692e+00 -2.86713600e-01 -1.70431644e-01 3.73907030e-01 -1.16078818e+00 -8.45995098e-02 -7.66556203e-01 -9.40283537e-02 -1.38633692e+00 2.52230406e-01 -7.80589581e-01 -6.13627911e-01 1.03231680e+00 -4.98722225e-01 -1.01720616e-01 3.77693892e-01 6.87709630e-01 -7.05744684e-01 4.25633460e-01 1.50945151e+00 1.35734856e-01 -6.02152288e-01 5.39803207e-02 -1.00725996e+00 4.74084139e-01 1.18997037e+00 -6.13348305e-01 -1.07749319e+00 -6.38371050e-01 3.18585724e-01 -1.32249981e-01 1.43467650e-01 -1.10873234e+00 6.07568622e-01 -8.02319720e-02 4.33001071e-01 -2.57173598e-01 4.66425680e-02 -8.02400291e-01 1.02187417e-01 9.18623865e-01 -2.03920364e-01 1.38903679e-02 4.39694405e-01 8.76542687e-01 2.66878158e-01 -5.37703335e-01 7.45437503e-01 -1.34013683e-01 -6.30694687e-01 1.34850502e-01 -3.71172041e-01 -7.00608194e-02 1.06562984e+00 -3.00520241e-01 -6.36580765e-01 -1.91767395e-01 -7.32850730e-01 7.68115044e-01 2.08822802e-01 4.12931532e-01 5.18459558e-01 -1.08156657e+00 -4.84813839e-01 1.04952022e-01 -2.36722022e-01 2.95672081e-02 6.58491671e-01 8.55377018e-01 1.27089977e-01 4.89465564e-01 -2.54677743e-01 -1.74458891e-01 -1.17881894e+00 6.32964969e-01 1.28338292e-01 -5.78498125e-01 -4.51213092e-01 9.20412004e-01 4.13940132e-01 -1.10191107e-01 5.25079072e-01 3.48481119e-01 -2.26264641e-01 4.23385352e-01 5.31022251e-01 5.91751397e-01 1.33687556e-01 2.33001813e-01 -1.75385308e-02 1.56896636e-01 -7.72722304e-01 1.25981957e-01 1.21721590e+00 -3.76868635e-01 -1.79783516e-02 4.50823963e-01 7.07561970e-01 -2.58130312e-01 -1.76086915e+00 -7.70651579e-01 -1.19494200e-01 -4.57412660e-01 2.18415096e-01 -1.17647111e+00 -1.18637896e+00 5.99625289e-01 5.18792272e-01 1.91428199e-01 1.20658422e+00 -3.34134281e-01 6.64732039e-01 7.72599757e-01 5.29503345e-01 -1.39951682e+00 7.92169213e-01 8.69976938e-01 6.50858879e-01 -7.07012117e-01 9.45521221e-02 -1.39901489e-01 -8.05980980e-01 8.54725480e-01 1.00335717e+00 2.65555888e-01 5.57110786e-01 3.20088193e-02 -4.20604080e-01 -1.79844275e-02 -8.32368195e-01 -5.00390790e-02 -1.65974706e-01 6.22488260e-01 1.28317893e-01 -1.38169006e-01 -2.73884147e-01 5.52968502e-01 2.60553926e-01 3.70496720e-01 4.10425991e-01 1.31484342e+00 -9.74424541e-01 -1.00329304e+00 -1.00591376e-01 7.65748382e-01 -1.92040488e-01 -3.87031257e-01 -1.44180924e-01 7.43212521e-01 2.32616112e-01 7.94640958e-01 -1.09162576e-01 -2.30549112e-01 4.58159745e-01 3.60648602e-01 3.35193783e-01 -5.96626282e-01 -7.34803855e-01 -1.87700972e-01 -9.51807424e-02 -1.03937194e-01 -1.56205237e-01 -4.24381047e-01 -9.47510600e-01 -5.45180857e-01 -6.59969211e-01 8.34088847e-02 4.56783801e-01 6.12107098e-01 4.82688636e-01 9.41488266e-01 6.32377207e-01 -4.87259030e-01 -8.39106500e-01 -9.76831555e-01 -8.01381171e-01 1.06334142e-01 1.57956362e-01 -9.15002048e-01 -4.30272490e-01 -5.42769097e-02]

[9.740456581115723, 3.478543519973755]

2553d820-9046-4c3a-8289-f9c037da8324

qcri-at-semeval-2023-task-3-news-genre

2305.03336

https://arxiv.org/abs/2305.03336v1

https://arxiv.org/pdf/2305.03336v1.pdf

QCRI at SemEval-2023 Task 3: News Genre, Framing and Persuasion Techniques Detection using Multilingual Models

Misinformation spreading in mainstream and social media has been misleading users in different ways. Manual detection and verification efforts by journalists and fact-checkers can no longer cope with the great scale and quick spread of misleading information. This motivated research and industry efforts to develop systems for analyzing and verifying news spreading online. The SemEval-2023 Task 3 is an attempt to address several subtasks under this overarching problem, targeting writing techniques used in news articles to affect readers' opinions. The task addressed three subtasks with six languages, in addition to three ``surprise'' test languages, resulting in 27 different test setups. This paper describes our participating system to this task. Our team is one of the 6 teams that successfully submitted runs for all setups. The official results show that our system is ranked among the top 3 systems for 10 out of the 27 setups.

['Firoj Alam', 'Preslav Nakov', 'Rabindra Nath Nandi', 'Ahmed Oumar El-Shangiti', 'Maram Hasanain']

2023-05-05

['misinformation']

['miscellaneous']

[-7.19207749e-02 1.39520779e-01 -6.08722642e-02 -3.23000729e-01 -1.15190148e+00 -1.00029707e+00 1.08038688e+00 4.39019829e-01 -4.90447909e-01 9.62273479e-01 3.25507849e-01 -5.67851841e-01 2.80098498e-01 -3.93549889e-01 -6.43704832e-01 7.93429464e-02 1.78093374e-01 7.13129163e-01 6.87351584e-01 -7.33774483e-01 8.44164729e-01 -1.01458445e-01 -9.82899785e-01 9.18476343e-01 7.99998462e-01 5.81084132e-01 -4.99991179e-01 8.33078504e-01 -1.87964171e-01 1.28953803e+00 -1.21091521e+00 -1.15164983e+00 -1.23622194e-01 -4.33410674e-01 -9.53544438e-01 -1.29598275e-01 6.88264310e-01 -8.94480497e-02 1.91578925e-01 1.36631334e+00 4.19131905e-01 -3.97185773e-01 3.74964863e-01 -1.42732275e+00 -9.33669865e-01 1.29328847e+00 -7.57461727e-01 6.20845437e-01 8.31776500e-01 -7.84286559e-02 9.99592185e-01 -6.62006378e-01 9.20218050e-01 1.15606999e+00 7.73436248e-01 1.90331459e-01 -1.02341163e+00 -6.50466025e-01 4.71181609e-03 2.56984290e-02 -1.26097500e+00 -5.08414268e-01 2.47451723e-01 -8.21727395e-01 1.01989698e+00 4.04038280e-01 3.61950517e-01 1.66614246e+00 4.24790949e-01 6.69798791e-01 1.62926030e+00 -2.71349341e-01 1.05784610e-01 9.21258450e-01 4.64179218e-01 4.49436903e-01 7.12689221e-01 -3.88524115e-01 -7.66688228e-01 -6.08147502e-01 2.65746773e-03 -6.75160885e-01 -3.21244836e-01 4.83629346e-01 -1.29164612e+00 9.17910278e-01 -2.16598995e-02 4.82387125e-01 -2.20843464e-01 -1.74032256e-01 5.47934055e-01 7.97078609e-01 1.10266459e+00 9.28916693e-01 -5.57941675e-01 -4.25187767e-01 -1.02930462e+00 8.10078442e-01 1.55005193e+00 9.02222037e-01 1.05794460e-01 -2.35779747e-01 -3.13672334e-01 7.58425593e-01 1.32086873e-01 6.84532702e-01 4.42865163e-01 -6.36580646e-01 7.74738371e-01 5.30580342e-01 6.46956146e-01 -1.56514978e+00 -4.46757734e-01 -8.42145205e-01 -4.79290336e-01 1.03207357e-01 6.30482316e-01 -5.13268352e-01 -4.92635131e-01 1.20693898e+00 -1.54581383e-01 -5.93743026e-01 -4.02593136e-01 6.04673088e-01 6.25132680e-01 5.56411684e-01 -3.40360105e-01 -2.32312262e-01 1.36972558e+00 -7.91636765e-01 -8.96115899e-01 -4.43149358e-01 7.39598989e-01 -1.30984604e+00 9.23026860e-01 9.23200130e-01 -1.17314041e+00 2.89958697e-02 -1.23689222e+00 1.69532076e-01 -5.77461064e-01 -1.18765727e-01 8.03524628e-02 5.69639742e-01 -1.05988419e+00 5.43214738e-01 -3.51352096e-01 -2.75688350e-01 2.77677834e-01 -3.58438790e-01 -4.41842861e-02 1.53983831e-01 -1.58595312e+00 1.41667950e+00 -1.55498907e-01 -2.21231818e-01 -5.19675672e-01 -5.87506473e-01 -1.85261220e-01 -4.59692627e-01 5.46545565e-01 -3.08054000e-01 1.87516618e+00 -7.41764307e-01 -1.12486398e+00 1.11732721e+00 8.32523853e-02 -5.62915742e-01 1.20602870e+00 -5.12123048e-01 -8.62754285e-01 -4.85227287e-01 7.03856170e-01 -2.47951806e-01 6.82706356e-01 -7.47593701e-01 -6.04378521e-01 -1.64101437e-01 -2.11387426e-01 -2.33373493e-01 -2.61234790e-02 8.24987352e-01 4.30260114e-02 -7.63976276e-01 -1.40596613e-01 -7.29968667e-01 5.67134209e-02 -5.28201461e-01 -8.24588597e-01 -1.02919921e-01 4.04934734e-01 -9.54408944e-01 1.55781758e+00 -1.52119005e+00 -1.51929662e-01 2.37708285e-01 6.33785844e-01 6.39252514e-02 4.83247370e-01 6.09379172e-01 2.03678846e-01 7.33781576e-01 2.00956598e-01 -3.31084430e-01 1.97939113e-01 -4.92579490e-01 -4.57752913e-01 5.87518334e-01 -6.39080703e-02 7.71854639e-01 -9.18901920e-01 -1.53563678e-01 -5.71336269e-01 2.80749369e-02 -4.64660153e-02 -4.63269860e-01 -6.09595656e-01 4.03177068e-02 -3.33248228e-01 3.74605209e-01 3.85080963e-01 -5.76278627e-01 -1.59190834e-01 3.96448910e-01 -3.80143523e-01 8.13741922e-01 -9.86214101e-01 1.04387784e+00 -1.68717019e-02 1.05236399e+00 1.68524787e-01 6.69506798e-03 7.57516980e-01 1.88633576e-01 -2.42422789e-01 -6.97777748e-01 2.24072158e-01 6.10382438e-01 -1.29885273e-02 -4.37533647e-01 6.95138395e-01 3.33559103e-02 -4.48477268e-01 1.00073314e+00 -3.30495954e-01 -1.41059563e-01 4.07867044e-01 8.05092394e-01 1.38601196e+00 -5.35569787e-01 3.53788108e-01 -5.25903165e-01 3.43433976e-01 5.38152575e-01 7.44643807e-02 1.14549255e+00 -2.35216439e-01 4.49754894e-01 8.75108659e-01 -5.35195053e-01 -1.03975284e+00 -5.12322664e-01 1.67275593e-01 1.15722001e+00 -1.50005758e-01 -6.94939315e-01 -7.00584471e-01 -8.20649743e-01 -8.77457187e-02 1.30371666e+00 -7.17481136e-01 2.53463596e-01 -1.85348973e-01 -5.71904182e-01 9.14942622e-01 5.43619916e-02 3.74088883e-01 -7.59903729e-01 -3.21097881e-01 3.87479573e-01 -5.51544487e-01 -9.71190393e-01 -5.19350708e-01 -1.64582014e-01 -1.60217494e-01 -1.12285376e+00 -6.41352952e-01 -3.25791180e-01 2.36252576e-01 1.87817305e-01 1.38186967e+00 1.90082926e-03 -6.63711084e-03 -2.48101652e-01 -5.67829430e-01 -9.05963063e-01 -1.07708836e+00 1.29942045e-01 -1.00650832e-01 -1.93472311e-01 5.61417043e-01 -1.15994833e-01 2.06003711e-02 4.58745629e-01 -6.77447438e-01 -5.08104861e-02 4.66673791e-01 4.43956256e-01 -3.61256480e-01 4.84170280e-02 5.86564720e-01 -1.66801620e+00 1.45809603e+00 -6.24341667e-01 -6.73427820e-01 1.71096906e-01 -8.42499435e-01 -6.65311515e-02 3.48241806e-01 -1.06050961e-01 -7.92869925e-01 -7.44292855e-01 -1.00002438e-01 5.20957589e-01 8.10575858e-03 8.63121390e-01 4.62825775e-01 1.50550827e-01 1.56934512e+00 -1.82163388e-01 -1.01734027e-01 -4.16956842e-01 2.47883666e-02 7.73153663e-01 2.11064190e-01 -1.26850814e-01 6.89605474e-01 -3.05120423e-02 -8.36747229e-01 -6.83043122e-01 -1.25544739e+00 -4.72983629e-01 8.67611989e-02 -2.52796471e-01 3.44681203e-01 -8.17090511e-01 -1.45520911e-01 9.03479755e-01 -1.61103773e+00 -8.35830718e-02 1.91877663e-01 1.34651050e-01 2.26504616e-02 1.65504754e-01 -9.78269100e-01 -7.65193522e-01 -2.54328936e-01 -8.47616494e-01 5.59546709e-01 -2.46274784e-01 -9.13329065e-01 -8.38657320e-01 3.97906870e-01 4.26436841e-01 1.05241430e+00 3.97661746e-01 3.90388310e-01 -1.13842964e+00 -2.09281147e-02 -8.55318546e-01 -1.85435340e-01 2.15288460e-01 -1.84403017e-01 3.15788954e-01 -8.12786162e-01 -3.49397138e-02 7.40978792e-02 -5.67260087e-01 6.99372351e-01 1.19782425e-01 2.37366691e-01 -7.48376310e-01 -2.82956570e-01 -4.70417798e-01 8.71098638e-01 -2.21081018e-01 4.67971772e-01 6.66885316e-01 2.70619869e-01 5.47733426e-01 4.24641073e-01 4.51360106e-01 3.15164536e-01 5.55625319e-01 -1.42246587e-02 3.52163047e-01 2.17629999e-01 -2.80098081e-01 6.72442019e-01 7.35549331e-01 1.25483751e-01 -6.76322401e-01 -1.14665651e+00 3.89637828e-01 -1.75460827e+00 -1.04836881e+00 -7.24215865e-01 1.82488263e+00 1.05402946e+00 9.42360580e-01 3.22337717e-01 -1.29006281e-02 7.91171789e-01 2.29561016e-01 6.81164861e-03 -5.26683807e-01 -1.43739641e-01 -2.97507823e-01 5.34104526e-01 7.59105086e-01 -1.00581229e+00 7.36364424e-01 6.44718838e+00 5.88289976e-01 -9.72349226e-01 3.67504716e-01 6.11755729e-01 5.73091395e-02 -4.75184888e-01 -2.00055584e-01 -1.03509188e+00 7.09980726e-01 1.17868352e+00 -6.17010295e-01 1.55272692e-01 8.24020922e-01 1.46371424e-01 -1.41475260e-01 -8.49088192e-01 4.23759371e-01 3.92749250e-01 -1.43644071e+00 -3.20644349e-01 -1.18373938e-01 9.98638928e-01 5.66090167e-01 -1.12156868e-01 3.54594439e-01 8.86866093e-01 -8.78322005e-01 1.26259434e+00 4.90263671e-01 3.78169686e-01 -3.09101522e-01 1.08992565e+00 8.62790823e-01 -1.54631082e-02 3.29445511e-01 1.51565403e-01 -3.28888476e-01 4.50394750e-01 9.86827195e-01 -1.10998261e+00 4.29172106e-02 8.31841350e-01 5.81464589e-01 -1.02318180e+00 1.05772233e+00 -4.96804774e-01 8.46001506e-01 -2.31793717e-01 -7.56326318e-01 7.07797036e-02 4.10382152e-01 9.54484582e-01 1.67333817e+00 3.28544043e-02 -4.29871112e-01 -4.43443209e-02 8.35283220e-01 -2.95977116e-01 -2.15337202e-02 -6.15623057e-01 -5.68898559e-01 3.60079914e-01 9.49298799e-01 -7.89774656e-01 -4.93543595e-01 -1.05751440e-01 8.14066589e-01 2.40247563e-01 1.42667696e-01 -9.03656662e-01 -2.76039004e-01 5.37704527e-02 6.16889298e-01 -4.18784842e-02 -1.70249701e-01 -3.64447057e-01 -1.18818665e+00 1.18518725e-01 -1.43306696e+00 2.89108157e-01 -7.06117690e-01 -1.59389853e+00 1.00931919e+00 -1.81812629e-01 -7.63210833e-01 -2.20094413e-01 -4.08835173e-01 -3.48484576e-01 7.96422005e-01 -1.09934711e+00 -6.28819883e-01 -2.25924388e-01 2.30496392e-01 5.65041184e-01 -1.93627894e-01 6.54520333e-01 3.48374099e-01 -3.61497283e-01 5.16562343e-01 -1.27570510e-01 -4.85503627e-03 1.35851204e+00 -1.34396362e+00 9.84983683e-01 8.31847966e-01 7.08582848e-02 9.29613948e-01 1.31565130e+00 -8.97349060e-01 -6.57103419e-01 -7.97323525e-01 1.84488833e+00 -1.15611899e+00 1.57803392e+00 -4.73577201e-01 -8.74031603e-01 7.55406499e-01 5.98072588e-01 -7.95599222e-01 4.46077555e-01 2.72675484e-01 -8.32903028e-01 3.65129590e-01 -7.03487933e-01 2.89341837e-01 5.33043981e-01 -5.74458838e-01 -7.25873113e-01 1.02113307e+00 5.75763285e-01 -6.81224585e-01 -3.22725028e-01 -1.43751010e-01 2.88815349e-01 -9.10699129e-01 1.82517067e-01 -8.34039152e-01 1.14545441e+00 -2.48983636e-01 1.76037103e-01 -1.48390436e+00 -3.70084167e-01 -8.71174455e-01 1.89819977e-01 1.19972789e+00 1.34048605e+00 -8.19732845e-01 3.57944757e-01 5.17409503e-01 4.23788242e-02 -2.85517752e-01 -3.74604642e-01 -4.04512137e-01 -8.43526945e-02 -5.70819795e-01 -9.50695053e-02 1.13652444e+00 1.25393227e-01 8.71491134e-01 -4.49131340e-01 -1.64117008e-01 2.98726648e-01 -2.20375776e-01 7.03970492e-01 -1.12089205e+00 -4.68420655e-01 -6.87506974e-01 -2.94949263e-01 -6.68818772e-01 -5.39405644e-01 -6.34848833e-01 2.54031837e-01 -1.48457932e+00 2.74225682e-01 6.91744909e-02 1.77459911e-01 3.65409732e-01 -1.03436276e-01 2.51322627e-01 -2.00320497e-01 3.83188546e-01 -1.08012486e+00 -2.66980618e-01 8.71841788e-01 -2.95800507e-01 1.03868552e-01 4.03672904e-01 -1.40931535e+00 8.18194568e-01 8.44945788e-01 -7.58497715e-01 -7.48305693e-02 -4.43769187e-01 1.30522597e+00 -4.17498916e-01 4.91920054e-01 -8.55050147e-01 6.48509741e-01 6.54358119e-02 1.57093210e-03 -4.92642939e-01 -2.25595430e-01 -2.84365982e-01 3.95579040e-02 3.18593293e-01 -6.29579127e-01 4.23795074e-01 1.54998437e-01 5.29269874e-01 -2.27873340e-01 -1.06437489e-01 5.76642632e-01 -3.67803097e-01 -2.77581573e-01 -4.25839454e-01 -7.90535927e-01 7.94876039e-01 8.40556324e-01 2.61661023e-01 -1.26573896e+00 -8.93634439e-01 -5.22594452e-01 1.93678200e-01 3.29989374e-01 5.33389866e-01 1.30830020e-01 -7.68586993e-01 -1.41641486e+00 -2.19814643e-01 1.54204309e-01 -6.22370481e-01 -1.04623072e-01 1.15705681e+00 -9.42079842e-01 3.36865246e-01 5.34723047e-03 -2.47207090e-01 -1.07210696e+00 2.13834345e-01 1.35971323e-01 -4.36860204e-01 -2.32901469e-01 1.28032112e+00 -7.48094201e-01 -2.58600712e-01 1.81314468e-01 -1.78633526e-01 -1.70178890e-01 3.16740811e-01 1.05170166e+00 6.60806000e-01 6.12601638e-01 -4.13396537e-01 -3.78593355e-01 -1.67746007e-01 -7.25993454e-01 -4.38304812e-01 1.05064011e+00 -3.59357037e-02 -3.27541560e-01 7.55610466e-01 9.92544711e-01 6.29585803e-01 -2.75906593e-01 -2.17215985e-01 4.02860224e-01 -3.14508229e-01 1.27374649e-01 -1.60946548e+00 -3.79642308e-01 3.19791138e-01 -3.73271435e-01 1.11157107e+00 2.37662166e-01 9.16794986e-02 5.78462124e-01 3.00893515e-01 4.57474977e-01 -1.25902879e+00 -5.02112918e-02 7.66120374e-01 1.41664648e+00 -1.22718728e+00 1.93878680e-01 -3.56155783e-01 -8.72664034e-01 8.00209582e-01 4.99101043e-01 1.42006129e-01 5.34713566e-01 3.90053064e-01 3.57999682e-01 -6.31303251e-01 -1.01134920e+00 4.74789590e-01 3.35877597e-01 -1.45115890e-02 7.10064471e-01 1.99392699e-02 -6.67309165e-01 7.22980917e-01 -4.47811514e-01 -1.10170513e-01 8.88663054e-01 1.00084472e+00 -5.97584307e-01 -7.41851449e-01 -3.40487301e-01 6.11432254e-01 -7.40594804e-01 -1.95372716e-01 -1.45584714e+00 7.51017451e-01 -1.83610409e-01 1.23842132e+00 -4.21628565e-01 -6.68477118e-01 6.62947953e-01 -6.32419586e-02 -4.79431264e-02 -6.49337709e-01 -9.40281510e-01 -5.55598363e-02 9.06253040e-01 -6.25802457e-01 -1.44087434e-01 -8.32385242e-01 -5.74131429e-01 -9.66205418e-01 -5.68125129e-01 4.13677216e-01 6.04246676e-01 8.86401176e-01 2.89680481e-01 3.07555616e-01 2.28801042e-01 -1.54788839e-02 -1.05705559e+00 -1.34942293e+00 -4.56228942e-01 3.85739863e-01 2.47119144e-01 -4.35060233e-01 -7.68300653e-01 -1.78058017e-02]

[8.31370735168457, 10.097750663757324]

d8445cdd-97e4-42d0-a009-21961820d9dd

robust-general-and-low-complexity-acoustic

2210.08610

https://arxiv.org/abs/2210.08610v1

https://arxiv.org/pdf/2210.08610v1.pdf

Robust, General, and Low Complexity Acoustic Scene Classification Systems and An Effective Visualization for Presenting a Sound Scene Context

In this paper, we present a comprehensive analysis of Acoustic Scene Classification (ASC), the task of identifying the scene of an audio recording from its acoustic signature. In particular, we firstly propose an inception-based and low footprint ASC model, referred to as the ASC baseline. The proposed ASC baseline is then compared with benchmark and high-complexity network architectures of MobileNetV1, MobileNetV2, VGG16, VGG19, ResNet50V2, ResNet152V2, DenseNet121, DenseNet201, and Xception. Next, we improve the ASC baseline by proposing a novel deep neural network architecture which leverages residual-inception architectures and multiple kernels. Given the novel residual-inception (NRI) model, we further evaluate the trade off between the model complexity and the model accuracy performance. Finally, we evaluate whether sound events occurring in a sound scene recording can help to improve ASC accuracy, then indicate how a sound scene context is well presented by combining both sound scene and sound event information. We conduct extensive experiments on various ASC datasets, including Crowded Scenes, IEEE AASP Challenge on Detection and Classification of Acoustic Scenes and Events (DCASE) 2018 Task 1A and 1B, 2019 Task 1A and 1B, 2020 Task 1A, 2021 Task 1A, 2022 Task 1. The experimental results on several different ASC challenges highlight two main achievements; the first is to propose robust, general, and low complexity ASC systems which are suitable for real-life applications on a wide range of edge devices and mobiles; the second is to propose an effective visualization method for comprehensively presenting a sound scene context.

['Phu X. Nguyen', 'Canh Vu', 'Khoa Tran', 'Alexander Schindler', 'Anahid Jalali', 'Dusan Salovic', 'Lam Pham']

2022-10-16

['scene-classification']

['computer-vision']

[ 2.76236832e-01 -5.97407401e-01 7.98009157e-01 -8.46711919e-02 -7.91584373e-01 -2.65234411e-01 3.18456262e-01 9.22047794e-02 -2.80432552e-01 1.26264781e-01 2.84747511e-01 -2.94920772e-01 -1.53145581e-01 -3.67660493e-01 -6.16384983e-01 -6.60268724e-01 -4.08508569e-01 -4.43170369e-01 4.20290828e-01 -7.73081481e-02 1.43799439e-01 3.01135927e-01 -1.91232455e+00 4.57679272e-01 3.42054069e-01 1.55537057e+00 4.05210286e-01 1.05175054e+00 1.17830902e-01 8.28426361e-01 -9.01077986e-01 5.35445288e-03 -2.81721354e-01 -1.97898313e-01 -4.98289704e-01 -4.17522907e-01 6.93714499e-01 4.46043797e-02 -2.41558954e-01 1.05589020e+00 1.28543448e+00 3.08682531e-01 2.31875777e-01 -1.60435176e+00 1.53327249e-02 6.37698948e-01 -3.27728122e-01 6.18903279e-01 3.80383164e-01 3.05992007e-01 1.06025410e+00 -1.22577238e+00 1.82246208e-01 1.02480483e+00 9.89331663e-01 3.79844636e-01 -5.96102834e-01 -9.57348526e-01 2.94354707e-01 6.94619477e-01 -1.45160687e+00 -5.78927696e-01 1.21639478e+00 -2.89284676e-01 9.10704672e-01 7.59339809e-01 6.11348391e-01 1.23855746e+00 -5.03472835e-02 8.37500870e-01 9.39556837e-01 -3.72589737e-01 4.85115975e-01 -2.34245121e-01 4.33214128e-01 2.73685157e-01 -2.04630256e-01 -1.67627096e-01 -1.05890930e+00 -2.79317021e-01 2.45739400e-01 -3.85803252e-01 -5.97160459e-01 2.51986742e-01 -9.54846442e-01 2.95613527e-01 3.55761766e-01 2.32007071e-01 -2.87921399e-01 4.00804609e-01 8.04801106e-01 3.00506391e-02 5.52318513e-01 8.47751126e-02 -1.61788583e-01 -4.49447721e-01 -8.22618067e-01 1.52616516e-01 6.60850585e-01 7.88400650e-01 3.81173193e-01 6.31766319e-01 -2.07530916e-01 1.30772436e+00 2.27227464e-01 5.58186889e-01 5.06092608e-01 -5.32433987e-01 5.45408070e-01 7.28906319e-02 -1.02485724e-01 -1.22142518e+00 -7.31401324e-01 -6.59100890e-01 -8.55734766e-01 -2.47008920e-01 -1.74427837e-01 -2.31536031e-01 -8.27011347e-01 1.67117119e+00 2.90998995e-01 1.09408116e+00 -1.11622751e-01 7.98457861e-01 1.57915902e+00 7.93170452e-01 1.52025759e-01 1.19990915e-01 1.40251589e+00 -9.20558155e-01 -7.56301284e-01 -8.35155621e-02 3.68950784e-01 -6.24714077e-01 1.42871737e+00 5.29208124e-01 -9.32717025e-01 -9.33947802e-01 -1.03216040e+00 2.56868035e-01 -3.92310083e-01 4.47157711e-01 3.10028583e-01 5.78242183e-01 -1.26592517e+00 1.76914200e-01 -7.18886077e-01 -1.91632986e-01 5.18272102e-01 7.19272941e-02 1.26983896e-01 3.51105481e-01 -1.23746014e+00 1.09604783e-01 1.37105063e-01 4.50781137e-01 -1.28429735e+00 -1.04418027e+00 -8.37828696e-01 1.78871289e-01 3.05773288e-01 -3.31519485e-01 1.25095046e+00 -4.89954501e-01 -1.36089635e+00 4.58196223e-01 -9.35123637e-02 -5.88767588e-01 2.32796907e-01 -4.55360293e-01 -1.11474931e+00 2.41566822e-01 -3.17096002e-02 2.62047946e-01 7.16871440e-01 -1.26533592e+00 -9.10129905e-01 3.27870878e-03 -2.96469592e-02 3.10897708e-01 -4.65428710e-01 4.56546187e-01 -6.30702615e-01 -7.40140796e-01 -6.30367994e-02 -7.11455703e-01 -4.19905111e-02 -2.87548393e-01 -9.64953482e-01 -2.27564171e-01 1.15417886e+00 -6.39274180e-01 1.58635628e+00 -2.74623847e+00 -4.74156380e-01 1.39852926e-01 2.80548364e-01 4.48263943e-01 -2.67262131e-01 1.95381731e-01 -2.02965483e-01 9.83621646e-03 -2.53209889e-01 -4.70255315e-01 5.59863821e-02 -1.70452416e-01 -6.34467423e-01 2.61899710e-01 -1.20981015e-01 5.95746756e-01 -7.25198269e-01 -2.34332338e-01 3.36087435e-01 5.34366429e-01 -3.92058045e-01 2.61655867e-01 2.86103040e-01 1.43439040e-01 -2.17573687e-01 6.96950674e-01 8.42206657e-01 -2.46611144e-02 -2.92183816e-01 -4.91449356e-01 -1.41704544e-01 4.65762615e-01 -1.52144694e+00 1.51669061e+00 -7.94650555e-01 9.52474535e-01 4.74327207e-01 -8.89989316e-01 8.23561311e-01 4.36347693e-01 4.29150790e-01 -6.66232944e-01 -1.02317460e-01 1.56002358e-01 -1.73621356e-01 -6.21661484e-01 5.47713459e-01 3.28429312e-01 -6.02150857e-02 1.21159814e-01 5.24879806e-02 -4.14566584e-02 -1.10002264e-01 3.20106149e-01 1.15418291e+00 -3.32916081e-01 -1.16018960e-02 -3.71873021e-01 5.45128703e-01 -7.27132976e-01 4.05721724e-01 1.06925249e+00 -4.76125717e-01 9.24180388e-01 1.88027546e-01 -3.62738162e-01 -2.75399357e-01 -1.01601410e+00 -2.06939921e-01 1.33140564e+00 3.41225326e-01 -6.51878953e-01 -7.86940217e-01 -4.38860059e-01 -2.68009216e-01 7.70723879e-01 -5.40007591e-01 -1.61873415e-01 -7.22882330e-01 -9.44338739e-01 1.20092416e+00 7.36977339e-01 8.62848639e-01 -1.36713779e+00 -8.42091262e-01 1.81655228e-01 -5.05299449e-01 -1.39890504e+00 -3.61310035e-01 3.07376802e-01 -1.46804810e-01 -8.29883218e-01 -4.78094518e-01 -7.77670264e-01 4.89262529e-02 3.70209724e-01 1.18826282e+00 -2.48277579e-02 -3.96362394e-01 7.72004962e-01 -4.10075366e-01 -9.80654418e-01 1.14539132e-01 -1.83293745e-01 3.89182240e-01 4.42734331e-01 1.43718258e-01 -7.77993202e-01 -6.90379381e-01 2.46962756e-01 -5.93931615e-01 -3.67208049e-02 5.90235330e-02 5.56618333e-01 5.00779390e-01 2.65510529e-02 7.76908398e-01 -5.10795534e-01 6.45478725e-01 -5.95507979e-01 -1.95858687e-01 3.44892889e-02 -1.74295649e-01 -9.15860653e-01 6.43752396e-01 -4.64122534e-01 -1.05487049e+00 -8.24083835e-02 -6.96528196e-01 -5.62503219e-01 -5.23778915e-01 3.51235896e-01 -4.11104232e-01 -2.66844891e-02 6.06216431e-01 4.84943539e-01 -7.50849009e-01 -7.21034050e-01 9.94312540e-02 9.36001837e-01 8.98412824e-01 -4.16011035e-01 4.47494417e-01 5.28532147e-01 -1.89628735e-01 -1.25496984e+00 -6.27936184e-01 -6.72813416e-01 -3.02889913e-01 -4.77932930e-01 8.32649350e-01 -1.13714719e+00 -6.30982220e-01 8.56479347e-01 -1.15220833e+00 -4.89457309e-01 -2.49094799e-01 4.23457682e-01 -2.46181488e-01 2.15167671e-01 -4.54350412e-01 -1.08050716e+00 -4.19852406e-01 -9.98948455e-01 1.27775860e+00 1.56599864e-01 -1.79540813e-02 -7.27449775e-01 2.50967834e-02 -2.08285358e-02 4.70344931e-01 1.67945951e-01 6.91509664e-01 -7.53030658e-01 -2.03304663e-01 2.25812092e-01 -1.84654146e-01 4.22357500e-01 -5.78340255e-02 -1.29277706e-01 -1.69174719e+00 -1.34767592e-01 5.11313863e-02 -2.36442104e-01 1.01153696e+00 6.84003234e-01 1.77773571e+00 -1.10139288e-01 -2.09177136e-01 9.25624669e-01 1.04003394e+00 6.15403116e-01 6.69831634e-01 1.91534981e-01 9.88865137e-01 3.96117508e-01 5.07254004e-01 4.12935913e-01 3.44076067e-01 8.56817782e-01 5.94976187e-01 -5.10731697e-01 -6.22324169e-01 -1.35461509e-01 3.26828450e-01 1.21919084e+00 1.00805975e-01 -4.71839190e-01 -9.70376432e-01 7.60908425e-01 -1.57062018e+00 -6.68166876e-01 -3.81926507e-01 1.85838914e+00 5.08049548e-01 -3.22455652e-02 9.13853869e-02 5.63774168e-01 7.36172795e-01 4.68818307e-01 -3.68631303e-01 -2.49630034e-01 -3.11587244e-01 3.49367946e-01 -1.76957607e-01 9.70613584e-02 -1.40447676e+00 6.73856080e-01 5.80907631e+00 1.26305223e+00 -1.36777306e+00 3.09327334e-01 6.63577020e-01 -2.98050672e-01 1.44470215e-01 -4.41964477e-01 -7.74748981e-01 6.89338028e-01 1.12738669e+00 1.58833802e-01 1.23497680e-01 8.27584147e-01 2.22626567e-01 1.26768634e-01 -9.46553826e-01 1.51767623e+00 2.05784440e-01 -1.28857517e+00 -7.11862072e-02 -3.81407470e-01 3.97969604e-01 1.87072515e-01 2.68045336e-01 3.87948185e-01 -1.60274237e-01 -7.60162592e-01 1.00089955e+00 1.78125396e-01 9.55345571e-01 -7.21161306e-01 5.64101815e-01 -1.87639982e-01 -1.80530918e+00 -1.87909231e-01 -1.25501573e-01 1.39181316e-01 2.61158109e-01 5.34472048e-01 -5.56936204e-01 6.66841030e-01 1.52583230e+00 6.97277486e-01 -6.94567084e-01 1.28023839e+00 4.00064103e-02 1.38382697e+00 -2.87995040e-01 -1.47161469e-01 1.08107559e-01 4.72540706e-01 1.01545727e+00 1.86613703e+00 3.23256969e-01 -1.61186885e-02 9.89438221e-02 6.48031771e-01 -9.71067920e-02 1.56948701e-01 -5.21134198e-01 4.02197242e-01 7.14601636e-01 1.32033575e+00 -7.01849818e-01 -2.57137239e-01 -2.56812662e-01 5.62344730e-01 -2.09822461e-01 6.55088127e-01 -1.05317187e+00 -8.41633677e-01 7.04930007e-01 -2.54101753e-01 1.99976131e-01 7.34149814e-02 -3.16669106e-01 -7.96613038e-01 1.62907392e-01 -8.20029795e-01 3.79607052e-01 -9.89883184e-01 -1.02864730e+00 9.85617578e-01 4.28173542e-02 -1.54946780e+00 1.42276347e-01 -5.21724522e-01 -1.13883352e+00 6.48614943e-01 -1.34500206e+00 -1.04464805e+00 -5.96277416e-01 7.81567514e-01 7.90873468e-01 -2.94079244e-01 5.96476197e-01 7.21937537e-01 -8.14200878e-01 9.50913250e-01 -8.49107429e-02 6.33408353e-02 3.53896350e-01 -1.21771097e+00 6.84251964e-01 9.66206610e-01 5.93266547e-01 2.76157141e-01 4.71623600e-01 -4.03783977e-01 -1.06437194e+00 -1.45968258e+00 4.21153426e-01 -1.99780062e-01 7.23328650e-01 -8.41687441e-01 -9.39570606e-01 2.69033194e-01 -3.04438043e-02 2.15421245e-01 5.77833474e-01 1.49566084e-01 -1.86833978e-01 -4.37719256e-01 -5.91719627e-01 4.76227432e-01 1.10041499e+00 -8.28411639e-01 -3.18520032e-02 2.44977787e-01 1.01913369e+00 -5.32525480e-01 -4.54883516e-01 5.22579551e-01 3.72224391e-01 -8.25444996e-01 1.28672671e+00 -2.50975251e-01 1.35755107e-01 -2.96261102e-01 -4.82032239e-01 -1.11950171e+00 -9.22757760e-02 -7.82763064e-01 -2.58775800e-01 1.48080778e+00 2.70361304e-01 -4.08126444e-01 2.47531950e-01 -5.85941076e-02 -7.46636033e-01 -8.16897869e-01 -1.27912855e+00 -8.92550409e-01 -4.05441642e-01 -1.35792089e+00 6.75464869e-01 7.32967019e-01 -5.42727768e-01 1.32653072e-01 -6.18625522e-01 4.70980108e-01 3.61434221e-01 -1.31786853e-01 8.63778889e-01 -1.05753696e+00 -2.21125647e-01 -4.02503788e-01 -6.05981886e-01 -1.07873523e+00 -2.02716455e-01 -4.84636962e-01 3.06569129e-01 -1.19158089e+00 -2.61318032e-02 -4.27639216e-01 -8.79445493e-01 3.49447161e-01 -2.27987692e-01 5.48315406e-01 1.91135466e-01 1.33893251e-01 -9.26807940e-01 4.65673327e-01 6.26132131e-01 -5.43881394e-02 -3.65043730e-01 2.18884632e-01 -5.51567554e-01 9.07882452e-01 4.96072501e-01 -2.46481135e-01 -4.37928736e-01 -4.51474637e-01 4.88071479e-02 -2.53737628e-01 9.58191693e-01 -1.48638666e+00 5.55183709e-01 2.63932079e-01 -4.59737182e-02 -9.05893505e-01 6.00478292e-01 -3.99364710e-01 -2.83271866e-03 1.49537977e-02 -3.54914218e-01 -2.35636070e-01 7.34202027e-01 7.28628278e-01 -4.28514808e-01 -4.54696529e-02 6.20631635e-01 2.96585679e-01 -1.11424351e+00 1.45435095e-01 -4.52889681e-01 2.10894153e-01 5.14612913e-01 2.34010145e-02 -3.44144851e-01 -5.18746853e-01 -9.00910020e-01 -2.16221623e-02 -5.45188963e-01 5.36066294e-01 7.82250285e-01 -1.46604288e+00 -7.20897794e-01 6.51153401e-02 1.43434182e-01 -1.77937895e-01 8.94568324e-01 8.02690089e-01 -4.07704264e-01 1.93078592e-01 1.20947435e-01 -1.06564629e+00 -1.45199227e+00 1.99927520e-02 4.23367888e-01 7.78959617e-02 -9.50282454e-01 1.31524980e+00 7.88733423e-01 -1.96154729e-01 8.13977361e-01 -5.15921354e-01 -4.56447273e-01 -2.22701296e-01 7.86810040e-01 6.69840097e-01 3.80518228e-01 -8.73261392e-01 -7.22181082e-01 5.06248355e-01 3.24008733e-01 -6.07585683e-02 1.14014065e+00 -2.59153605e-01 3.15662950e-01 6.81031346e-01 1.20887399e+00 3.26073281e-02 -9.76737499e-01 -2.50167876e-01 -2.35146672e-01 -2.06239209e-01 2.61332572e-01 -9.09332275e-01 -1.15910912e+00 1.39206183e+00 1.18590176e+00 4.60239112e-01 1.67908084e+00 -7.89235085e-02 8.64488721e-01 2.23022729e-01 1.17016084e-01 -1.09697485e+00 3.01106751e-01 4.98234570e-01 1.10601139e+00 -8.79507959e-01 -4.43529040e-01 -4.40086842e-01 -5.97695231e-01 9.62580144e-01 6.43200994e-01 2.30670437e-01 9.70764875e-01 5.82486928e-01 1.90795586e-01 -4.61000115e-01 -6.10171378e-01 -1.47211343e-01 4.96502608e-01 6.90970361e-01 -7.93814752e-03 6.16316497e-02 3.65176767e-01 1.13176560e+00 -5.64675927e-01 -3.95852238e-01 2.32116520e-01 7.89975286e-01 -1.62117362e-01 -2.43032470e-01 -2.33745679e-01 3.32206070e-01 -6.10400140e-01 -2.98463553e-01 -3.66912961e-01 5.27333558e-01 2.94685811e-01 1.29764986e+00 1.46203384e-01 -9.32402790e-01 5.43119073e-01 -1.00672901e-01 -2.17145294e-01 -4.31058228e-01 -7.90588796e-01 4.09037948e-01 1.09511495e-01 -6.69316947e-01 -2.37463802e-01 -4.11772460e-01 -1.14627969e+00 2.78136104e-01 -4.72608179e-01 1.39892638e-01 8.16774487e-01 6.84363127e-01 6.74058795e-01 1.07069075e+00 6.20492756e-01 -9.04538035e-01 -1.05179273e-01 -9.91243899e-01 -7.01187313e-01 2.53175229e-01 5.47513068e-01 -6.23105943e-01 -5.40045559e-01 1.33995920e-01]

[15.164155006408691, 5.206484317779541]

42091ca1-e52e-4b7d-b840-b18c1e91f8f5

rethinking-deep-policy-gradients-via-state

https://openreview.net/forum?id=USDowdRKXmN

https://openreview.net/pdf?id=USDowdRKXmN

Rethinking Deep Policy Gradients via State-Wise Policy Improvement

Deep policy gradient is one of the major frameworks in reinforcement learning, and it has been shown to improve parameterized policies across various tasks and environments. However, recent studies show that the key components of the deep policy gradient methods, such as gradient estimation, value prediction, and optimization landscapes, fail to reflect the conceptual framework. This paper aims to investigate the mechanism behind the deep policy gradient methods through the lens of state-wise policy improvement. Based on the fundamental properties of policy improvement, we propose an alternative theoretical framework to reinterpret the deep policy gradient update as training a binary classifier, with labels provided by the advantage function. This framework obviates the statistical difficulties in the gradient estimates and predicted values of the deep policy gradient update. Experimental results are included to corroborate the proposed framework.

['I-Chen Wu', 'Ting Han Wei', 'Ping-Chun Hsieh', 'Kai-Chun Hu']

2020-10-19

neurips-workshop-icbinb-2020-12

['value-prediction', 'policy-gradient-methods']

['computer-code', 'methodology']

[-1.00714266e-01 -1.27379060e-01 -8.78930032e-01 -1.84751660e-01 -1.49417460e-01 -4.38993573e-01 7.90902734e-01 -3.75919119e-02 -6.73006058e-01 1.38224936e+00 2.57768154e-01 -7.24697888e-01 -3.92590553e-01 -6.59157395e-01 -6.69998825e-01 -8.79953742e-01 -4.53685261e-02 -7.61998519e-02 1.00169778e-02 -2.96939284e-01 7.75408685e-01 3.59136492e-01 -1.51125634e+00 -5.46564795e-02 1.17720699e+00 1.25036180e+00 9.07001719e-02 6.48312509e-01 -2.12196186e-01 6.62267029e-01 -6.98070288e-01 -1.92631438e-01 4.06995714e-01 -5.49943328e-01 -5.34600139e-01 -3.15495521e-01 2.14793175e-01 -7.78361022e-01 -4.33128506e-01 1.28624356e+00 6.81485772e-01 2.91375250e-01 5.55793822e-01 -1.27275944e+00 -7.21330702e-01 3.67724985e-01 -3.27735722e-01 3.02510709e-01 7.65690953e-02 2.46537790e-01 9.18743372e-01 -4.17349577e-01 4.73258168e-01 1.40534425e+00 5.41434109e-01 6.91642344e-01 -7.88126111e-01 -4.19458032e-01 5.51259637e-01 4.35377866e-01 -5.69379330e-01 -8.76153037e-02 7.78928280e-01 -2.49623373e-01 6.62695527e-01 4.88172621e-02 9.81620550e-01 1.13991153e+00 4.49040204e-01 1.09235823e+00 1.68286824e+00 -3.54320168e-01 5.02856672e-01 2.64767766e-01 -8.31687823e-02 7.24172115e-01 3.89884740e-01 9.74651694e-01 -2.34414950e-01 -2.20799908e-01 9.85184550e-01 -1.66507795e-01 -2.62568146e-01 -6.69585347e-01 -5.77642739e-01 9.24503326e-01 5.17592013e-01 9.59086120e-02 -5.09238958e-01 3.80776256e-01 5.20792902e-01 4.30855483e-01 5.15049160e-01 6.30932093e-01 -5.61559975e-01 -5.71977735e-01 -4.14646566e-01 5.56602478e-01 5.87073088e-01 4.69391972e-01 6.26843393e-01 4.78892982e-01 -6.36797428e-01 5.41668117e-01 2.91758001e-01 5.85507452e-01 7.18450606e-01 -1.11620247e+00 3.20546061e-01 3.97958696e-01 5.79209685e-01 -7.59405613e-01 -2.82725453e-01 -7.61913657e-01 -3.39441121e-01 8.20171595e-01 5.09728789e-01 -5.14665961e-01 -7.95837164e-01 1.69573784e+00 5.00685990e-01 -1.13739213e-02 -6.80923909e-02 9.25891519e-01 5.54896966e-02 3.86582077e-01 1.79563478e-01 -2.67576456e-01 5.81710756e-01 -9.70137775e-01 -7.77555585e-01 3.27532552e-02 4.76858497e-01 -2.57551372e-01 1.35693800e+00 2.02377230e-01 -9.86444890e-01 -5.80159009e-01 -1.08192623e+00 4.09247726e-01 -2.20636234e-01 -9.46928188e-03 7.00866640e-01 7.80039966e-01 -1.01643074e+00 1.12774777e+00 -6.69473886e-01 -3.48237902e-01 4.43391055e-01 1.66146949e-01 3.82691890e-01 3.60774815e-01 -1.18589103e+00 1.33702588e+00 6.39414966e-01 -1.32456515e-03 -1.01297510e+00 -5.09314239e-01 -5.07404566e-01 3.00611332e-02 4.89038289e-01 -5.57439685e-01 1.67320395e+00 -1.00296366e+00 -2.27491999e+00 1.00749709e-01 -6.91458881e-02 -5.83811224e-01 8.37875962e-01 -5.78943968e-01 1.95055958e-02 -1.41030058e-01 -3.42231482e-01 2.64113843e-01 9.89346981e-01 -1.22288787e+00 -1.25889421e+00 -3.86579514e-01 1.13180652e-01 5.05058825e-01 -3.70663702e-01 -3.92858863e-01 4.06257004e-01 -3.11310053e-01 -5.13374269e-01 -6.85402632e-01 -4.97776449e-01 -2.31752291e-01 -6.11599497e-02 -4.34757352e-01 9.00797427e-01 -4.83053446e-01 1.27427185e+00 -1.84241855e+00 -1.14980437e-01 1.90128297e-01 7.60533288e-02 5.60373008e-01 -4.61854748e-02 1.95594579e-01 2.62458742e-01 6.57559037e-02 -6.72938675e-02 1.07232913e-01 3.85705113e-01 3.79097104e-01 -6.33512914e-01 5.83825946e-01 6.54522190e-03 9.83345747e-01 -1.34424233e+00 -4.47110049e-02 4.52624530e-01 -9.65115521e-03 -5.00251412e-01 3.04463744e-01 -4.12467152e-01 4.57843512e-01 -9.01551247e-01 4.87839222e-01 5.41261554e-01 2.77030855e-01 2.25888237e-01 2.07360417e-01 -5.61696708e-01 4.78547156e-01 -1.00051892e+00 1.30262148e+00 -1.23644032e-01 5.19169629e-01 7.89222941e-02 -1.49540484e+00 8.61699164e-01 3.34921442e-02 5.38087666e-01 -9.98608112e-01 1.25838965e-01 3.30805838e-01 2.70858072e-02 -3.97892892e-01 5.60099542e-01 -1.33499056e-01 2.47308150e-01 3.65997612e-01 -1.34527653e-01 -2.55045176e-01 1.70180574e-01 -3.37356269e-01 7.23629236e-01 6.00899100e-01 5.45665622e-01 -3.38855714e-01 2.91116148e-01 3.13016982e-03 5.71867526e-01 1.18546069e+00 -7.91010797e-01 -2.36191660e-01 7.29278505e-01 -5.45470774e-01 -1.16525102e+00 -9.45449352e-01 -1.33200034e-01 1.28620279e+00 1.12793304e-01 1.75202321e-02 -6.21547937e-01 -1.02207625e+00 4.80789959e-01 7.88176656e-01 -7.20421433e-01 -3.86561245e-01 -5.09200931e-01 -7.11074233e-01 4.57071453e-01 4.64360982e-01 8.24720085e-01 -9.65383172e-01 -8.39161336e-01 3.82433951e-01 3.94873887e-01 -5.46364188e-01 -4.77538630e-03 1.07462279e-01 -1.15128386e+00 -1.07131636e+00 -7.17110336e-01 -2.75495499e-01 2.93210655e-01 -3.44671831e-02 8.64830434e-01 2.07271613e-02 1.32741362e-01 6.11995459e-01 -2.18459293e-01 -5.94557285e-01 -3.91539574e-01 3.61954048e-02 4.34334338e-01 -3.68291557e-01 -1.38974190e-02 -3.80247116e-01 -8.58423412e-01 1.15259483e-01 -4.94748384e-01 -2.94240683e-01 7.21331954e-01 1.19388545e+00 2.92930275e-01 -1.85098156e-01 8.88713360e-01 -5.89558780e-01 1.40270293e+00 -4.38676357e-01 -8.90531540e-01 3.31278235e-01 -1.39180350e+00 4.57099199e-01 5.32716930e-01 -4.50413495e-01 -1.29168975e+00 -4.31645125e-01 -2.00806081e-01 -8.39021280e-02 -1.37439615e-03 4.03527409e-01 2.44916469e-01 -2.23782912e-01 6.76501572e-01 3.04063797e-01 1.30981237e-01 -4.53193128e-01 5.94908834e-01 4.71826047e-01 2.68059731e-01 -9.59279299e-01 3.42701584e-01 3.22459519e-01 -2.32586600e-02 -4.68995690e-01 -7.43743300e-01 -7.03675374e-02 -1.90845117e-01 -3.24355900e-01 5.40935993e-01 -3.37488681e-01 -9.27936077e-01 4.42968488e-01 -7.84719884e-01 -6.97738886e-01 -6.54783428e-01 5.69399178e-01 -9.22621608e-01 2.29293153e-01 -5.26843131e-01 -1.25065005e+00 -2.81515360e-01 -1.05372453e+00 5.02983034e-01 6.96302235e-01 2.78258651e-01 -1.30504262e+00 2.50673831e-01 -4.31724727e-01 7.42485046e-01 1.09543294e-01 1.06858265e+00 -3.49830568e-01 -2.26726696e-01 9.62224752e-02 -2.21320093e-02 5.28256476e-01 1.34851858e-01 -1.19012669e-02 -8.35467756e-01 -4.14349884e-01 2.15268448e-01 -3.42723101e-01 7.53806949e-01 6.48291230e-01 1.23350942e+00 -3.15105945e-01 -2.24765658e-01 6.15764081e-01 1.30494928e+00 6.82814360e-01 5.29464602e-01 6.73570335e-01 3.46030951e-01 1.67468086e-01 6.60420835e-01 7.86184192e-01 3.22015211e-02 3.16107512e-01 6.47423148e-01 1.49521017e-02 1.07480250e-01 -6.93200827e-01 5.44923544e-01 2.78841764e-01 -1.44663751e-01 2.53423899e-02 -5.11384428e-01 2.02140898e-01 -2.02912259e+00 -9.95382845e-01 4.19982523e-01 2.17947626e+00 8.03578138e-01 1.85710162e-01 2.22003222e-01 -1.45882189e-01 6.04006886e-01 1.67438507e-01 -1.21483469e+00 -7.19535947e-01 4.01850641e-02 2.39540309e-01 6.33664370e-01 5.15033960e-01 -9.34686601e-01 1.15544987e+00 7.98274565e+00 7.05917299e-01 -1.25505877e+00 -1.58646330e-01 6.41570091e-01 8.75720158e-02 -2.08379686e-01 -1.22109346e-01 -8.12583804e-01 4.71279114e-01 6.85243547e-01 -4.19565737e-01 6.70994997e-01 1.09098029e+00 4.99976635e-01 -1.93347707e-01 -7.50798225e-01 6.07233763e-01 -4.49196398e-01 -1.23146617e+00 -6.85005412e-02 1.28390834e-01 1.00348127e+00 6.19880036e-02 5.39531708e-01 9.42842305e-01 7.53575563e-01 -8.08737516e-01 6.90569699e-01 3.82450491e-01 4.57128853e-01 -7.08488822e-01 5.62427282e-01 3.23808551e-01 -7.19656110e-01 -8.02399874e-01 -5.81014693e-01 -6.52713537e-01 -1.07726812e-01 2.93717682e-01 -1.00030136e+00 3.20225328e-01 4.96278763e-01 6.29536748e-01 -1.94993362e-01 1.23631990e+00 -5.80822527e-01 7.80978143e-01 -6.74383566e-02 -5.73849201e-01 6.61684573e-01 -4.86590028e-01 6.87783659e-01 8.31273317e-01 3.95617753e-01 -3.74266714e-01 1.75087795e-01 8.63002539e-01 1.66659355e-01 1.14483431e-01 -6.42575920e-01 -2.09122717e-01 2.71281600e-01 8.01402569e-01 -3.02219570e-01 -4.27273542e-01 -9.01159495e-02 6.56854808e-01 5.92657030e-01 8.21793377e-01 -7.33772814e-01 -2.01465636e-01 1.10176265e+00 -2.59242535e-01 1.94720775e-01 -3.14540267e-01 -5.96449673e-01 -8.61320853e-01 -1.43465698e-02 -8.46241713e-01 1.25424922e-01 -2.94784486e-01 -1.04237545e+00 -3.93913984e-02 1.22279115e-01 -9.32506263e-01 -5.57668686e-01 -9.41707611e-01 -6.80012167e-01 7.54922509e-01 -1.79545832e+00 -3.41047525e-01 1.57343298e-01 4.30558980e-01 4.51615542e-01 -2.26908371e-01 5.40793538e-01 -7.26452395e-02 -5.79016209e-01 5.58374941e-01 9.84652758e-01 -3.34362388e-01 4.01019871e-01 -1.59656298e+00 1.91930570e-02 5.77055633e-01 -4.18140531e-01 4.52632219e-01 8.08699965e-01 -5.81767559e-01 -1.24327552e+00 -5.96081734e-01 5.84025420e-02 -9.30883139e-02 7.23462999e-01 9.39735994e-02 -6.29478872e-01 1.87100619e-01 2.02152103e-01 -3.14178169e-01 3.03972624e-02 1.80999905e-01 1.35362461e-01 -1.67058319e-01 -1.06939685e+00 8.18239391e-01 1.05269349e+00 -4.10243481e-01 -4.84157056e-01 7.55625516e-02 5.63920200e-01 -4.16645497e-01 -5.17522216e-01 5.59312820e-01 7.50122309e-01 -9.82320249e-01 6.66207135e-01 -1.22956073e+00 2.09658071e-01 6.63382187e-02 3.41934860e-02 -1.96141613e+00 -2.11075574e-01 -6.49846494e-01 -5.51309884e-01 6.53525293e-01 1.37982741e-01 -9.66255248e-01 8.59789312e-01 4.83144939e-01 -2.84835875e-01 -1.11493814e+00 -1.08817029e+00 -9.89407718e-01 4.67875630e-01 -2.70426273e-01 5.79287291e-01 6.97715640e-01 8.45746472e-02 1.74256247e-02 -3.77303660e-01 -2.90964872e-01 3.80834699e-01 4.16607074e-02 7.10363448e-01 -8.97824824e-01 -2.46417210e-01 -1.11526871e+00 -1.31052345e-01 -1.38062978e+00 3.44526857e-01 -4.44513559e-01 8.27905461e-02 -1.67278981e+00 -8.83903503e-02 -4.67427850e-01 -8.26343298e-01 1.22937061e-01 -6.00667238e-01 -6.99605703e-01 2.24352032e-01 8.19800720e-02 -4.97405708e-01 1.09831166e+00 1.56647050e+00 -4.56916681e-03 -3.65857184e-01 2.65913516e-01 -6.71036839e-01 7.13534832e-01 1.29669452e+00 -3.35708618e-01 -6.26693726e-01 -2.20073536e-01 2.80549712e-02 -1.23968065e-01 3.81876305e-02 -7.64760911e-01 -1.72446609e-01 -7.10624456e-01 5.07493794e-01 -1.33659229e-01 -9.14907977e-02 -5.54213464e-01 -7.21717298e-01 8.91783535e-01 -5.98539710e-01 2.83583671e-01 2.25091353e-01 6.92699432e-01 7.49141797e-02 -2.42413118e-01 7.58937955e-01 -1.15150802e-01 -9.83476758e-01 3.00298929e-01 -4.39253569e-01 1.56688213e-01 8.70660365e-01 -8.82501453e-02 -4.18521285e-01 -4.55981374e-01 -2.64400303e-01 3.04620177e-01 1.99582994e-01 3.21397036e-01 5.65226138e-01 -1.38939559e+00 -4.59602416e-01 -9.18508321e-03 -4.99601156e-01 -5.37390590e-01 -1.93423703e-01 5.66717684e-01 -1.13587759e-01 6.60261333e-01 -4.50702876e-01 -3.41374248e-01 -6.24507546e-01 4.96308804e-01 8.17178309e-01 -6.01024032e-01 -3.64211053e-01 5.57848096e-01 -1.63743570e-02 -4.71653253e-01 4.60956126e-01 -6.44777298e-01 -2.02220991e-01 -2.62691557e-01 4.37098771e-01 5.87287545e-01 -3.20180029e-01 1.68137580e-01 1.50459170e-01 2.18126401e-01 1.30812660e-01 -4.83113706e-01 9.80703473e-01 -1.23600885e-01 2.98293591e-01 3.34109962e-01 8.77305329e-01 -4.65114713e-01 -1.89052188e+00 -1.74715057e-01 3.10947508e-01 -5.80055952e-01 2.02490583e-01 -9.14577484e-01 -6.77587450e-01 8.56999815e-01 1.11101794e+00 1.52368456e-01 8.15210104e-01 -6.35122299e-01 5.95741332e-01 5.95654964e-01 4.32274789e-01 -1.57593429e+00 8.52130055e-02 8.45256686e-01 7.83069253e-01 -1.26100385e+00 -5.61349839e-02 3.98321003e-01 -5.90754688e-01 9.28140223e-01 8.29633057e-01 -1.49299905e-01 4.55583930e-01 -7.65826479e-02 1.05719730e-01 1.88764527e-01 -7.57860839e-01 -4.01654840e-01 1.02645673e-01 8.04531336e-01 2.99878061e-01 2.46472195e-01 -7.79701233e-01 1.45877570e-01 2.59513948e-02 2.55888462e-01 1.66121274e-02 1.05994999e+00 -8.31990838e-01 -1.14536762e+00 -2.59061754e-01 4.34071094e-01 -5.15310109e-01 1.04219675e-01 -2.84241021e-01 8.17644298e-01 -1.57346442e-01 7.65672147e-01 -9.29791182e-02 -4.01352972e-01 3.64291102e-01 8.57288837e-02 6.08195841e-01 -2.11950229e-03 -5.13293326e-01 -3.05939108e-01 -1.40537024e-01 -6.76490605e-01 -8.00943151e-02 -3.35882276e-01 -1.21837711e+00 -1.70638442e-01 -5.94630986e-02 4.51272726e-01 9.39433753e-01 1.05936873e+00 3.39002341e-01 3.82248282e-01 8.90504062e-01 -7.71654963e-01 -1.51568735e+00 -1.01121593e+00 -4.92152661e-01 3.98494035e-01 4.84035552e-01 -1.16604590e+00 -3.17300439e-01 -6.69726729e-01]

[4.125992298126221, 2.3500053882598877]

71572295-416a-4ffc-ad13-2df1a5053d16

dont-throw-those-morphological-analyzers-away

https://aclanthology.org/D17-1073

https://aclanthology.org/D17-1073.pdf

Don't Throw Those Morphological Analyzers Away Just Yet: Neural Morphological Disambiguation for Arabic

This paper presents a model for Arabic morphological disambiguation based on Recurrent Neural Networks (RNN). We train Long Short-Term Memory (LSTM) cells in several configurations and embedding levels to model the various morphological features. Our experiments show that these models outperform state-of-the-art systems without explicit use of feature engineering. However, adding learning features from a morphological analyzer to model the space of possible analyses provides additional improvement. We make use of the resulting morphological models for scoring and ranking the analyses of the morphological analyzer for morphological disambiguation. The results show significant gains in accuracy across several evaluation metrics. Our system results in 4.4{\%} absolute increase over the state-of-the-art in full morphological analysis accuracy (30.6{\%} relative error reduction), and 10.6{\%} (31.5{\%} relative error reduction) for out-of-vocabulary words.

['Nizar Habash', 'Nasser Zalmout']

2017-09-01

emnlp-2017-9

['morphological-disambiguation', 'morphological-tagging']

['natural-language-processing', 'natural-language-processing']

[ 1.15592539e-01 -5.98821826e-02 2.41507009e-01 -2.63531953e-01 -1.02289629e+00 -7.38563716e-01 2.68942863e-01 7.03038752e-01 -8.58043492e-01 3.36007386e-01 7.46637508e-02 -6.33818328e-01 5.11068590e-02 -9.84142542e-01 -4.39907879e-01 -5.52457392e-01 -3.50612730e-01 5.01690388e-01 -4.55903858e-02 -7.23499596e-01 2.99703956e-01 7.16091394e-01 -1.17087948e+00 5.24841011e-01 9.04702127e-01 8.01216424e-01 1.56661034e-01 5.79969466e-01 -2.74435878e-01 3.42967123e-01 -1.01353681e+00 -6.52978003e-01 -7.34069273e-02 -1.42932132e-01 -8.46192837e-01 -2.39416420e-01 1.88543752e-01 -6.09926172e-02 4.05100808e-02 1.23829162e+00 6.30912840e-01 3.57527375e-01 4.57019001e-01 -2.73996323e-01 -1.19852197e+00 1.16063738e+00 -4.54241186e-01 3.14757228e-01 3.81050020e-01 -2.68734366e-01 1.30997860e+00 -1.25286663e+00 5.71235001e-01 1.11447573e+00 7.58978069e-01 3.43086988e-01 -1.03636122e+00 -4.93896484e-01 2.87309259e-01 1.99536279e-01 -1.40506172e+00 -4.87092108e-01 3.60609442e-01 4.27280664e-02 1.82348692e+00 9.32426751e-02 3.81458253e-01 4.80064243e-01 1.22845590e-01 7.11800754e-01 1.00779390e+00 -9.58937109e-01 1.77200679e-02 -1.51004717e-01 5.74007332e-01 9.71250236e-01 1.50103256e-01 -3.28665525e-01 -3.32623154e-01 3.54071744e-02 5.41889608e-01 -4.98178720e-01 1.31140068e-01 5.52213669e-01 -8.39996994e-01 9.01086330e-01 4.75104600e-01 4.94052917e-01 -3.29020947e-01 -2.68441886e-01 3.02620381e-01 3.61869484e-01 2.83619970e-01 7.52616942e-01 -6.82615459e-01 4.16917317e-02 -1.00569618e+00 -1.47746980e-01 7.43403018e-01 5.43574691e-01 4.87903863e-01 6.00585818e-01 1.59553126e-01 1.26973009e+00 2.56735831e-01 5.43941021e-01 7.89930344e-01 -5.36242247e-01 6.15397871e-01 7.30306923e-01 -1.24641240e-01 -8.29331636e-01 -7.16877222e-01 -6.15363419e-01 -4.85964924e-01 1.36870369e-01 5.55505991e-01 -1.08050115e-01 -1.22524655e+00 1.54155028e+00 -1.47898495e-01 -5.08303463e-01 3.85571152e-01 3.26710641e-01 7.93499410e-01 8.61489594e-01 2.55267154e-02 -3.38218063e-01 1.55130327e+00 -7.22666085e-01 -8.13100994e-01 -3.41287524e-01 9.24202442e-01 -1.17184007e+00 1.26671386e+00 5.76927245e-01 -1.32677996e+00 -3.49445313e-01 -1.38680077e+00 6.04143888e-02 -8.61742795e-01 4.66922730e-01 2.35093355e-01 9.31210935e-01 -1.29634583e+00 7.68017709e-01 -8.57120991e-01 -4.92285997e-01 -8.38230848e-02 8.60931575e-01 -2.31355071e-01 3.84222358e-01 -1.21937668e+00 1.15533388e+00 6.45575166e-01 4.43025231e-01 -3.11672539e-01 -2.53991485e-01 -1.07034302e+00 8.21557269e-02 1.70314357e-01 5.75876944e-02 1.20087719e+00 -5.84606111e-01 -1.50045621e+00 8.44283104e-01 -1.54027805e-01 -5.33270061e-01 -4.60983902e-01 -4.10558850e-01 -7.89133251e-01 -6.37409762e-02 -2.80268461e-01 5.01381457e-01 1.92929193e-01 -9.67891991e-01 -6.20429337e-01 -3.44616562e-01 -1.37369528e-01 1.73544765e-01 -6.69176280e-01 6.76271260e-01 -1.94845632e-01 -1.02952588e+00 3.15346330e-01 -1.10451615e+00 -1.29683018e-01 -8.00775290e-01 -1.87885121e-01 -3.22897047e-01 5.45161963e-01 -1.48729074e+00 1.46394360e+00 -1.81854010e+00 2.41237041e-02 1.65717125e-01 -2.31445312e-01 7.98349202e-01 -3.92408133e-01 3.37050676e-01 -8.07211623e-02 5.12068033e-01 -2.75114506e-01 -3.15025002e-01 1.94661710e-02 1.26993880e-01 -1.61190137e-01 1.40786633e-01 5.07842481e-01 9.63168144e-01 -6.42202675e-01 -1.13205470e-01 9.64652523e-02 5.73643029e-01 -1.69654652e-01 -2.58946061e-01 -1.09527605e-02 -3.30420047e-01 2.63346117e-02 8.81968319e-01 3.55544806e-01 1.54315338e-01 5.51031053e-01 -2.25994084e-02 2.10704785e-02 6.46197379e-01 -1.17337978e+00 1.60358143e+00 -6.19165719e-01 4.92209166e-01 -1.38064861e-01 -5.34550726e-01 1.21140730e+00 3.30363393e-01 -2.25695327e-01 -7.25211442e-01 3.12822700e-01 6.83725834e-01 4.63335603e-01 1.11190297e-01 8.31921816e-01 4.37878445e-03 -4.00408953e-01 6.09664440e-01 3.62655759e-01 2.49502018e-01 3.90606314e-01 -8.57774541e-02 9.83180225e-01 5.74301817e-02 3.42519671e-01 -2.48227552e-01 6.26888216e-01 -2.76760429e-01 4.67955112e-01 4.92691070e-01 4.67545427e-02 3.33588690e-01 1.93812713e-01 -5.61713874e-01 -7.63964951e-01 -9.33387280e-01 1.34125099e-01 1.57529402e+00 -4.11490679e-01 -5.80407441e-01 -9.36817467e-01 -5.64187825e-01 -3.57048035e-01 9.73574579e-01 -5.79249442e-01 3.73689122e-02 -1.36238861e+00 -1.31072915e+00 1.03082550e+00 9.54666793e-01 2.01377645e-01 -1.63758278e+00 -5.17698586e-01 4.68235582e-01 -1.38578415e-01 -9.32497501e-01 -1.02093577e-01 6.33666754e-01 -8.69662464e-01 -5.47596872e-01 -3.49781662e-01 -1.19494760e+00 4.10690069e-01 -2.35929295e-01 9.84803259e-01 1.23578846e-01 2.41761506e-02 -2.58450031e-01 -6.12537205e-01 -3.99788976e-01 -6.72813594e-01 3.50650787e-01 1.49979219e-01 -4.21311855e-01 6.76571071e-01 -2.01040193e-01 4.73119086e-03 -8.93863738e-02 -8.24118912e-01 -6.48057103e-01 6.72892809e-01 9.39495265e-01 6.10376179e-01 -3.06522455e-02 5.71628273e-01 -7.93360949e-01 7.94220626e-01 2.62083709e-02 -5.43407142e-01 4.78480846e-01 -7.86400139e-01 5.05326390e-01 7.97847450e-01 -2.85723031e-01 -1.05736983e+00 -1.11154050e-01 -4.05982494e-01 7.20992312e-02 -2.44690869e-02 1.01348794e+00 -2.48109445e-01 2.13295594e-02 5.43448448e-01 9.67298672e-02 -1.91549301e-01 -6.84170067e-01 4.25840050e-01 6.53114736e-01 4.98342842e-01 -3.30708504e-01 3.45741868e-01 -6.62617758e-02 -2.49627948e-01 -6.79342270e-01 -5.94991505e-01 -1.81479380e-01 -9.06184137e-01 2.28898406e-01 6.56184435e-01 -4.16940540e-01 -5.33580601e-01 5.02976239e-01 -1.04031372e+00 -4.17859763e-01 -1.05476916e-01 2.26657018e-01 -1.28688127e-01 3.82598519e-01 -1.21955812e+00 -6.78714275e-01 -8.56984794e-01 -1.25343323e+00 8.45833480e-01 2.49297038e-01 -4.44380552e-01 -1.10276794e+00 -2.80257970e-01 4.91070859e-02 2.98511714e-01 2.59796176e-02 1.32864797e+00 -1.02418983e+00 3.33348401e-02 -3.22357982e-01 -7.26468191e-02 2.54627138e-01 2.35483691e-01 9.36050043e-02 -9.78838265e-01 -2.31703952e-01 -5.52472360e-02 -5.85325472e-02 9.15689409e-01 3.30455214e-01 6.84555829e-01 -2.10126936e-01 -2.35638823e-02 3.47257853e-01 1.31814396e+00 5.53027093e-01 5.10892689e-01 6.71842813e-01 5.94180465e-01 6.48159444e-01 5.83771884e-01 3.71526629e-02 3.32598895e-01 5.84448099e-01 2.19615981e-01 1.07590653e-01 -3.56608063e-01 1.81913257e-01 8.66556942e-01 1.35252678e+00 8.72377530e-02 -4.04639095e-01 -1.24970210e+00 7.18856871e-01 -1.49856722e+00 -5.37511051e-01 -2.09102686e-02 2.16660619e+00 8.75727236e-01 2.08408102e-01 1.15642369e-01 5.32135725e-01 8.21792781e-01 7.30671212e-02 -1.28106177e-01 -1.29805279e+00 -4.13631678e-01 8.32465947e-01 4.05781180e-01 7.05075502e-01 -1.22841442e+00 1.55996478e+00 5.92569923e+00 9.24619496e-01 -1.03372049e+00 -1.00791357e-01 5.20754516e-01 3.00142486e-02 -3.86297889e-02 -1.73965409e-01 -1.12338519e+00 -2.42204089e-02 1.47108519e+00 2.07633257e-01 2.40143657e-01 2.59015173e-01 -2.15704203e-01 6.94034621e-02 -7.18799591e-01 5.94008625e-01 4.26218867e-01 -1.10647047e+00 2.68324614e-01 -4.03063931e-02 4.33380038e-01 2.68828478e-02 1.75263837e-01 2.63853550e-01 6.25923216e-01 -1.21792865e+00 6.55526221e-01 1.93439499e-01 8.32608521e-01 -1.17483318e+00 1.04407775e+00 -1.54701965e-02 -1.28261137e+00 -1.17087103e-01 -3.15136999e-01 4.78793196e-02 1.53715432e-01 2.01825932e-01 -1.03613412e+00 4.60373074e-01 6.61256373e-01 4.47598934e-01 -1.02195179e+00 5.77303290e-01 -6.16088569e-01 7.53631890e-01 -3.01412553e-01 -4.38599735e-01 4.86477375e-01 -1.11744463e-01 4.47915167e-01 1.66469896e+00 2.88409084e-01 1.51380198e-02 3.44825946e-02 2.38543928e-01 -1.28303692e-01 5.22322357e-01 2.93203648e-02 -8.05171952e-02 5.83625138e-01 1.11585784e+00 -1.25266206e+00 -2.81547308e-01 -3.62606836e-03 7.95092821e-01 6.57786191e-01 1.42141268e-01 -4.35366780e-01 -1.10587633e+00 2.96806931e-01 -4.99961287e-01 7.06392705e-01 -3.28146040e-01 -5.63759983e-01 -8.40265095e-01 1.79632723e-01 -1.09219515e+00 7.54652321e-01 -5.77001989e-01 -1.19299054e+00 1.07111323e+00 -4.08126444e-01 -6.70369625e-01 -3.63099933e-01 -1.04236305e+00 -1.83146521e-01 9.80538547e-01 -1.15699589e+00 -1.31597698e+00 3.08915287e-01 1.76851917e-02 6.29479349e-01 -4.99485880e-01 1.46715200e+00 4.64989729e-02 -6.74245656e-01 1.13560295e+00 9.11705121e-02 5.27530849e-01 7.04985380e-01 -1.44247437e+00 7.52046585e-01 1.28480291e+00 6.25443161e-01 8.27305138e-01 4.16264325e-01 -5.46499074e-01 -1.05493510e+00 -1.05916727e+00 1.51342499e+00 -5.42613328e-01 9.03539836e-01 -3.72878402e-01 -1.12930572e+00 6.67185485e-01 4.79621500e-01 -3.33583117e-01 8.65881979e-01 4.46371168e-01 -6.81163609e-01 -4.84188609e-02 -1.07591403e+00 8.39879334e-01 7.61332810e-01 -5.26340067e-01 -9.15286124e-01 2.24470552e-02 7.28628814e-01 -3.30642790e-01 -9.94105279e-01 3.23363215e-01 7.06481457e-01 -4.68121648e-01 7.95484245e-01 -5.44480026e-01 1.34241432e-01 -1.24323785e-01 -3.89682680e-01 -1.36573088e+00 -4.31230932e-01 -6.72388613e-01 -8.53792652e-02 1.17003024e+00 9.46834147e-01 -5.76347411e-01 5.41245043e-01 1.63703009e-01 -3.50473136e-01 -7.59766757e-01 -6.67138159e-01 -6.82164729e-01 5.47850430e-01 -4.30508524e-01 6.54229820e-01 9.63582993e-01 2.37720236e-01 4.28239107e-01 1.99016169e-01 3.24155122e-01 8.44736099e-02 2.78917253e-02 -1.80836812e-01 -9.28120792e-01 -2.21103594e-01 -9.11372006e-01 -5.68941355e-01 -5.08928835e-01 4.28576112e-01 -1.01031053e+00 -9.24721658e-02 -1.66553462e+00 -2.60534912e-01 -2.50344098e-01 -6.74622715e-01 1.08610868e+00 -2.41496027e-01 5.09931505e-01 4.25997168e-01 -1.19576491e-01 -3.03707600e-01 1.00955389e-01 2.87471563e-01 -7.62154832e-02 -4.36596125e-01 -3.60959858e-01 -6.96589172e-01 8.71841908e-01 1.25027335e+00 -4.66838390e-01 6.73704967e-02 -9.46991980e-01 5.10889351e-01 -1.59136429e-01 -2.10209131e-01 -9.19437230e-01 1.78488359e-01 2.33133867e-01 5.76926172e-01 -6.57141745e-01 3.57061714e-01 -9.06783566e-02 -4.67044741e-01 5.88915169e-01 -3.13674659e-01 9.69466627e-01 7.60007918e-01 4.91650701e-02 -1.50488272e-01 -5.60162604e-01 5.54849207e-01 -8.96383300e-02 -7.28800058e-01 -3.01793605e-01 -8.29410493e-01 -1.53311983e-01 5.06185949e-01 -3.40841383e-01 -4.35379416e-01 -3.02396324e-02 -1.12690771e+00 -1.99346185e-01 2.81242758e-01 4.34976816e-01 6.11997604e-01 -9.29792225e-01 -7.95734584e-01 2.54167944e-01 -1.95036575e-01 -4.83825684e-01 -3.66928428e-01 3.99550349e-01 -7.54117668e-01 4.40175682e-01 -2.51935244e-01 1.95743702e-02 -1.50040662e+00 3.28975320e-01 2.01130569e-01 -7.11856663e-01 -2.58378714e-01 1.27924299e+00 -5.88682115e-01 -7.75625944e-01 1.54794514e-01 -3.47019106e-01 -5.37701309e-01 4.04058576e-01 3.99268597e-01 5.84181368e-01 8.37129056e-01 -8.84569824e-01 -4.61273432e-01 3.71540874e-01 -4.59386319e-01 -5.17084062e-01 1.40590703e+00 1.06255315e-01 -4.66206938e-01 5.63895226e-01 7.99326420e-01 4.21824664e-01 -3.66015434e-01 -2.15007260e-01 5.23409903e-01 3.13976884e-01 -9.28547755e-02 -1.38647521e+00 -7.33417809e-01 1.16270137e+00 6.55413151e-01 1.59269571e-01 1.16721284e+00 -3.88070732e-01 1.05283427e+00 7.96588600e-01 2.89505601e-01 -1.21382034e+00 -2.43219510e-01 1.19647539e+00 6.36510551e-01 -7.27150381e-01 -2.43587792e-02 -2.99895197e-01 -5.67529976e-01 1.39668274e+00 3.07527393e-01 -3.56164426e-01 5.69075644e-01 3.79234880e-01 4.53241587e-01 -6.21543415e-02 -5.76816618e-01 -1.71171471e-01 2.89722174e-01 4.28711414e-01 8.86507690e-01 2.97667801e-01 -4.51263577e-01 7.78111756e-01 -6.61157787e-01 -8.64591599e-01 6.65015519e-01 9.86864865e-01 -4.08249497e-01 -1.52517188e+00 -3.11554521e-01 5.30093014e-01 -7.54377782e-01 -6.02877855e-01 -5.90951204e-01 8.34553719e-01 -1.16458535e-01 1.10610271e+00 -1.49633056e-02 -3.36081892e-01 2.85553813e-01 5.37120521e-01 5.44228613e-01 -8.42740715e-01 -1.26788676e+00 3.37367445e-01 3.56146485e-01 -2.17579618e-01 -1.30784675e-01 -8.90439987e-01 -1.59782314e+00 6.44306540e-02 -6.95644557e-01 1.37410656e-01 5.96073449e-01 9.14911747e-01 3.29831600e-01 4.74048227e-01 1.42176315e-01 -5.08618176e-01 -6.58460796e-01 -1.23403835e+00 -4.62213188e-01 3.14902775e-02 -6.12070449e-02 -3.38619590e-01 -6.00766540e-02 5.42400368e-02]

[10.471071243286133, 10.174099922180176]

269d74f4-6efd-4bed-9443-4fd84ec37012

a-cognitive-approach-based-on-the-actionable

2011.09554

https://arxiv.org/abs/2011.09554v1

https://arxiv.org/pdf/2011.09554v1.pdf

A Cognitive Approach based on the Actionable Knowledge Graph for supporting Maintenance Operations

In the era of Industry 4.0, cognitive computing and its enabling technologies (Artificial Intelligence, Machine Learning, etc.) allow to define systems able to support maintenance by providing relevant information, at the right time, retrieved from structured companies' databases, and unstructured documents, like technical manuals, intervention reports, and so on. Moreover, contextual information plays a crucial role in tailoring the support both during the planning and the execution of interventions. Contextual information can be detected with the help of sensors, wearable devices, indoor and outdoor positioning systems, and object recognition capabilities (using fixed or wearable cameras), all of which can collect historical data for further analysis. In this work, we propose a cognitive system that learns from past interventions to generate contextual recommendations for improving maintenance practices in terms of time, budget, and scope. The system uses formal conceptual models, incremental learning, and ranking algorithms to accomplish these objectives.

['Francesco Orciuoli', 'Domenico Marino', 'Vincenzo Loia', 'Mariacristina Gallo', 'Giuseppe Fenza']

2020-11-18

['outdoor-positioning']

['miscellaneous']

[ 3.58154535e-01 3.76855768e-02 -6.54179215e-01 -2.80354470e-01 -3.95728648e-02 -5.44658780e-01 5.28103352e-01 5.87690711e-01 -5.90201244e-02 6.23933792e-01 3.22458357e-01 -3.91090512e-01 -9.72687185e-01 -9.75710988e-01 -1.97357818e-01 -2.02132925e-01 3.26462567e-01 2.43448004e-01 2.59068400e-01 -2.16682643e-01 7.28648007e-01 6.24397099e-01 -1.96121669e+00 3.38124573e-01 6.05215669e-01 1.28633988e+00 8.64945233e-01 4.04913872e-01 -3.66086543e-01 9.11068141e-01 -6.25287652e-01 -3.88572030e-02 -1.42162114e-01 4.48722206e-02 -5.91859400e-01 3.55750136e-02 -4.29216802e-01 -3.44949841e-01 2.46070966e-01 5.55889130e-01 8.79578590e-02 -1.35251224e-01 2.94352293e-01 -9.65265512e-01 -8.14551890e-01 4.33347464e-01 1.63655564e-01 1.34384902e-02 6.99442148e-01 -6.13797344e-02 4.60902959e-01 -6.32518768e-01 4.28600490e-01 8.67231131e-01 2.77933866e-01 2.11486727e-01 -6.92817867e-01 -2.77244728e-02 2.36671433e-01 5.67507863e-01 -7.96574056e-01 -3.29758406e-01 8.17547083e-01 -7.64880955e-01 6.20731115e-01 5.68772256e-01 7.54249930e-01 1.02058697e+00 3.85830998e-01 3.52686673e-01 8.80341351e-01 -9.29728448e-01 6.35970771e-01 4.35220271e-01 1.87628075e-01 3.87424856e-01 2.98555195e-01 1.03083752e-01 -4.42020714e-01 -4.38809879e-02 3.32463384e-01 7.15189576e-01 1.15102850e-01 -1.01225108e-01 -9.98024166e-01 1.70415342e-01 2.28377990e-02 7.79914498e-01 -8.14010978e-01 -2.10509345e-01 1.31372347e-01 1.78110406e-01 -3.90599705e-02 6.70433462e-01 -4.46581423e-01 -1.68702111e-01 -4.34832066e-01 -1.86469361e-01 5.43614328e-01 1.05398023e+00 7.05612838e-01 -2.04689309e-01 -4.10726964e-01 2.92985350e-01 2.24065304e-01 3.14617574e-01 6.16190791e-01 -1.16653216e+00 3.52321744e-01 1.10444808e+00 5.76717436e-01 -9.16596591e-01 -5.22409856e-01 -3.68482292e-01 -6.35571420e-01 -2.83169419e-01 -3.18976581e-01 -3.47325094e-02 -4.30881381e-01 1.08794582e+00 1.72518194e-01 2.37822924e-02 -7.42651969e-02 3.97808701e-01 1.32839635e-01 3.69998157e-01 1.26265779e-01 -5.46239257e-01 1.14385223e+00 -3.86471778e-01 -1.14301169e+00 -2.09976614e-01 3.70580703e-01 -5.29860675e-01 1.01030314e+00 6.55468345e-01 -8.29294801e-01 -8.40268493e-01 -1.13272667e+00 4.38608170e-01 -6.96583927e-01 5.06032050e-01 6.78890049e-01 5.30891478e-01 -8.61227274e-01 5.58428824e-01 -6.13864005e-01 -2.40671650e-01 -3.36564295e-02 1.23660900e-01 3.62493604e-01 -2.17371717e-01 -9.61534321e-01 8.68667126e-01 4.00321543e-01 1.48286507e-01 -6.10308468e-01 -4.90950346e-01 -3.52479100e-01 5.69747016e-02 6.58370614e-01 -7.97123790e-01 1.04004216e+00 -6.65949583e-01 -1.31725442e+00 2.91727781e-01 2.02918425e-01 -5.73538601e-01 -3.84670752e-03 -4.15784627e-01 -8.77418160e-01 -2.99979728e-02 -6.53820261e-02 -2.47842431e-01 6.99212432e-01 -8.72639060e-01 -1.04802608e+00 -8.26522589e-01 2.64053643e-01 -2.50191212e-01 -6.07516468e-01 -1.14541113e-01 -2.74474382e-01 -1.10555269e-01 -7.13017806e-02 -7.30284989e-01 -3.80438268e-01 -6.75450265e-01 -8.44703317e-02 -1.55879140e-01 8.14514339e-01 -8.38552952e-01 1.73774981e+00 -1.93040240e+00 3.87433618e-02 5.22862256e-01 -1.48867309e-01 2.62475103e-01 4.67615783e-01 4.18115854e-01 3.53632718e-01 -4.06559817e-02 3.68394345e-01 2.79991210e-01 9.41498801e-02 2.82422990e-01 -7.51260966e-02 -3.38226855e-01 -1.89183801e-01 5.75017154e-01 -7.73744524e-01 -3.36451799e-01 6.97158337e-01 7.19191972e-03 -1.96273550e-01 1.54620796e-01 -4.69051450e-01 4.16529685e-01 -8.93607736e-01 9.01506364e-01 -1.52287811e-01 -1.46021590e-01 6.21253431e-01 -1.65398002e-01 -5.82851768e-01 1.97334960e-01 -1.04144740e+00 1.60135698e+00 -1.04899335e+00 1.77588344e-01 -1.00389026e-01 -1.10908031e+00 1.07963526e+00 4.63860422e-01 7.10258484e-01 -9.78801310e-01 -2.70432476e-02 9.79723111e-02 -6.46352828e-01 -1.03810322e+00 3.78731161e-01 6.41773582e-01 -2.22683713e-01 2.49274731e-01 -3.43567938e-01 3.52801114e-01 2.10181937e-01 -2.25772798e-01 1.14456439e+00 3.50242071e-02 3.03166538e-01 6.31378442e-02 7.65716553e-01 -1.77615151e-01 4.86746550e-01 4.67893809e-01 1.69811234e-01 -9.49550122e-02 2.13987175e-02 -6.51364148e-01 -6.42759323e-01 -5.34849763e-01 3.57119352e-01 1.12832904e+00 2.05967464e-02 -3.38409871e-01 -6.23464823e-01 -3.78575951e-01 -2.57089343e-02 1.01838183e+00 -2.81426400e-01 -4.38810647e-01 -2.72789866e-01 -9.46708173e-02 -4.07178164e-01 5.76309144e-01 3.72268051e-01 -1.01651835e+00 -9.97236311e-01 6.43155754e-01 2.61679571e-02 -7.21187174e-01 -6.56615570e-02 1.70026138e-01 -1.00873482e+00 -1.19221890e+00 5.95604330e-02 -5.18309712e-01 4.60925072e-01 2.38101631e-01 9.04093266e-01 2.87734509e-01 -2.72740752e-01 6.58397615e-01 -4.06828016e-01 -6.13946855e-01 -3.59330148e-01 -6.46715472e-03 7.88251497e-03 -4.93119322e-02 2.92943299e-01 -6.97035074e-01 -4.97588277e-01 4.73444819e-01 -7.68183589e-01 -2.85760295e-02 7.75015891e-01 3.90130132e-01 7.26057649e-01 5.29986322e-01 8.48238409e-01 -7.26002812e-01 9.36950684e-01 -5.29289186e-01 -8.47065151e-01 8.79846573e-01 -1.17412961e+00 9.01632905e-02 4.20259029e-01 -7.51285627e-02 -1.22236216e+00 -5.80608882e-02 2.28048205e-01 -5.60852000e-03 -3.45962405e-01 7.62968957e-01 -5.13397574e-01 4.07561332e-01 7.23124146e-01 2.11887106e-01 -2.71599382e-01 -6.59625351e-01 2.75046557e-01 1.13760591e+00 6.00956976e-01 -5.01493633e-01 2.41863340e-01 1.90867752e-01 2.16471292e-02 -2.61624575e-01 -7.63781846e-01 -4.03037846e-01 -6.85231686e-01 -7.00843811e-01 6.37193203e-01 -3.43106925e-01 -8.15697312e-01 -2.96447337e-01 -7.12168157e-01 6.13058396e-02 -2.78386414e-01 4.14700091e-01 -5.29506207e-01 -2.56410778e-01 -1.18954323e-01 -1.12910950e+00 -3.60827833e-01 -8.02155614e-01 5.39150953e-01 3.25478047e-01 -2.80809313e-01 -6.44040287e-01 -2.84547925e-01 6.74602568e-01 5.16803145e-01 2.77772188e-01 1.14002132e+00 -3.39394182e-01 -9.10089076e-01 -4.67956960e-01 4.62120563e-01 5.65704465e-01 5.12406647e-01 -1.10310264e-01 -3.63420099e-01 2.55437404e-01 1.75057516e-01 3.20157439e-01 8.59546512e-02 2.30254337e-01 1.54353106e+00 -5.69114506e-01 -5.38759470e-01 -5.81859723e-02 1.22042418e+00 1.09254587e+00 6.61568463e-01 3.18404883e-01 2.19858870e-01 7.17339814e-01 1.23047256e+00 6.79772675e-01 4.10418302e-01 7.91321695e-01 4.53913778e-01 5.49087107e-01 7.50816092e-02 -1.62787184e-01 7.13541433e-02 5.97889304e-01 -4.79268909e-01 8.62509906e-02 -8.27579856e-01 4.49083298e-01 -2.18756223e+00 -1.16710269e+00 2.02725798e-01 2.37141943e+00 3.60183597e-01 4.48983163e-01 -2.89216619e-02 4.78592336e-01 5.57267785e-01 -5.29586017e-01 -7.02602744e-01 -3.83921415e-01 3.93953234e-01 -1.51624754e-01 5.02571344e-01 -1.26775444e-01 -5.84966183e-01 2.35057130e-01 5.51532650e+00 2.85284013e-01 -7.02594757e-01 1.80204451e-01 4.32671934e-01 -1.25312641e-01 -4.00186956e-01 -1.85697868e-01 -5.62350869e-01 6.98802471e-01 1.13348043e+00 -1.55588090e-01 7.75149167e-01 1.17682338e+00 5.71610808e-01 -1.38467595e-01 -1.06130600e+00 7.95682490e-01 -1.51812389e-01 -1.79724884e+00 -3.32563221e-01 5.20661511e-02 7.28552103e-01 -6.27303898e-01 -7.74655864e-02 2.05312639e-01 -7.44248629e-02 -1.81997672e-01 8.65255296e-01 1.33354533e+00 3.02719772e-01 -8.94750953e-01 7.72542953e-01 6.71265662e-01 -9.76061761e-01 -8.91345382e-01 -2.47055106e-02 8.60223267e-03 3.69840749e-02 8.07065904e-01 -7.98358560e-01 7.62168884e-01 7.48259366e-01 4.22952622e-01 -5.30026913e-01 7.26076484e-01 -2.49714538e-01 3.18270475e-01 2.16084808e-01 -3.03124130e-01 -3.11033010e-01 -2.40630776e-01 1.13779880e-01 6.08537853e-01 7.88258076e-01 1.05143994e-01 1.34962201e-01 5.10038555e-01 3.22333425e-01 -7.84075782e-02 -5.49799085e-01 7.65057057e-02 7.38283515e-01 9.25016642e-01 -4.17152077e-01 -3.39148551e-01 -3.70778322e-01 3.19766134e-01 -3.44173193e-01 2.12166831e-01 -4.92936671e-01 -3.58153045e-01 4.59896654e-01 6.95441306e-01 4.26650159e-02 -2.58548439e-01 -3.64542753e-01 -3.82189780e-01 1.09809302e-01 -6.81929469e-01 1.37832791e-01 -9.73993361e-01 -7.88818955e-01 1.12385668e-01 6.19448647e-02 -1.07452488e+00 -5.75641096e-01 -5.78590810e-01 -2.32977703e-01 4.77662385e-01 -9.33616698e-01 -6.65896535e-01 -3.86353463e-01 6.74733639e-01 4.61713165e-01 -4.50087368e-01 7.64733672e-01 4.11825567e-01 -4.45121437e-01 -2.87540197e-01 1.28587171e-01 -6.36455715e-01 3.17151666e-01 -1.03025126e+00 -3.14603597e-01 4.28766489e-01 -1.07498623e-01 5.32082796e-01 3.04050416e-01 -7.15958953e-01 -1.68902373e+00 -9.60026205e-01 7.89094508e-01 -3.65453392e-01 3.75863612e-01 -1.75295454e-02 -3.58407646e-01 4.28716570e-01 -1.57324821e-01 -6.42771125e-01 8.35521281e-01 3.43819827e-01 2.17968196e-01 -7.73459792e-01 -1.06001592e+00 4.73105550e-01 1.11532128e+00 -4.16640311e-01 -5.32909036e-01 4.81438249e-01 8.92610967e-01 -1.25872970e-01 -9.22470927e-01 4.04839277e-01 6.00703418e-01 -9.56471741e-01 1.02044749e+00 -6.81638122e-01 -6.55932352e-02 -2.75003344e-01 -2.71534890e-01 -8.14526081e-01 -4.77062643e-01 -3.73577207e-01 -6.17126942e-01 1.24772334e+00 3.01114917e-01 -5.00747859e-01 4.50145066e-01 7.40441382e-01 -4.89198685e-01 -6.18017435e-01 -4.27751571e-01 -4.80388463e-01 -1.24073887e+00 -7.21263468e-01 1.38885725e+00 6.63950324e-01 1.24084264e-01 1.96743220e-01 -3.11217934e-01 1.95477948e-01 1.91414878e-01 4.45184469e-01 4.04838055e-01 -1.64044189e+00 -2.28316426e-01 -1.66249901e-01 -4.38021332e-01 -6.11049473e-01 -1.16502769e-01 -4.15642709e-01 -2.46423513e-01 -1.94468391e+00 -2.73644358e-01 -4.59271401e-01 -6.63692296e-01 4.68660116e-01 3.51000071e-01 -5.85457265e-01 2.22594161e-02 1.52650744e-01 -6.68781281e-01 -2.64658481e-02 8.87296617e-01 -1.40145198e-01 -4.87281561e-01 6.85729802e-01 -9.53278065e-01 6.96246862e-01 1.09082568e+00 -5.26355803e-02 -6.05677962e-01 -3.51945907e-01 7.02803433e-01 5.02514064e-01 3.13989669e-01 -1.23314929e+00 3.38348031e-01 -6.43883169e-01 2.79205084e-01 -6.11978054e-01 1.61229596e-02 -1.51291859e+00 6.26845062e-01 5.19681513e-01 -4.33003813e-01 1.96273893e-01 -3.37946922e-01 6.00369930e-01 1.36108641e-02 -4.18583035e-01 -1.03715956e-01 3.03651206e-02 -6.52457058e-01 -3.70240696e-02 -4.75201070e-01 -7.21397877e-01 1.07814372e+00 -1.64329037e-01 -3.24722797e-01 -1.54254228e-01 -9.56826150e-01 1.33364454e-01 1.20740183e-01 7.46636689e-01 4.59820211e-01 -1.26420188e+00 1.59399286e-01 2.56551325e-01 1.33880198e-01 -4.02616918e-01 3.09038192e-01 6.51532114e-01 -6.54182062e-02 8.59044373e-01 -2.08063170e-01 -2.59637207e-01 -9.92169201e-01 9.46276069e-01 -1.51117772e-01 -2.11544991e-01 -4.38839309e-02 -2.17463851e-01 -6.90380454e-01 7.81552866e-02 5.11499882e-01 -6.34013653e-01 -5.82152009e-01 1.96214214e-01 7.05149472e-01 7.97416687e-01 4.90732372e-01 1.15970634e-01 -1.55179501e-01 3.36842507e-01 4.97432619e-01 1.97292581e-01 1.33060777e+00 -4.56813604e-01 -5.07570915e-02 4.26947206e-01 4.29830372e-01 8.04556236e-02 -7.41048157e-01 -3.15079868e-01 6.39920294e-01 -3.27634037e-01 2.21520960e-01 -1.22921360e+00 -8.70134771e-01 4.07271624e-01 8.85296524e-01 7.48227060e-01 1.58379269e+00 1.21571250e-01 4.34771568e-01 5.87804317e-01 9.53860343e-01 -1.66890121e+00 1.30982324e-01 -1.25364527e-01 8.59029531e-01 -6.71199799e-01 -7.63817877e-02 -1.84992120e-01 -3.65663141e-01 1.06078064e+00 3.91820997e-01 6.63414598e-01 8.28364491e-01 9.19423848e-02 -3.62327158e-01 -2.03237280e-01 -8.05740535e-01 -1.53051287e-01 1.64815307e-01 7.69588649e-01 7.50829950e-02 4.14448828e-01 -2.80789375e-01 9.76092458e-01 7.74791464e-02 4.03647959e-01 9.45113152e-02 1.11194921e+00 -7.88773179e-01 -1.29426277e+00 -5.69509089e-01 5.69649637e-01 -1.37179688e-01 4.35329437e-01 -2.90218949e-01 2.98702359e-01 8.67423952e-01 1.23954940e+00 -6.65730014e-02 -6.08278453e-01 9.43466663e-01 1.42455965e-01 2.64097005e-01 -5.73367298e-01 -3.54982406e-01 -2.52719641e-01 2.65782267e-01 -6.51416361e-01 -3.80571246e-01 -6.76011145e-01 -1.05229568e+00 1.99702844e-01 -1.66486889e-01 4.90843207e-02 1.21784472e+00 1.02544928e+00 7.54826903e-01 1.11365199e+00 9.27419484e-01 -4.91385370e-01 -3.61720860e-01 -8.00524354e-01 -2.94513255e-01 -2.06308849e-02 -7.53098577e-02 -7.10158288e-01 9.28477794e-02 3.18515778e-01]

[8.541095733642578, 5.942290306091309]

3f687141-e89b-4ce4-9287-c04cc71922d8

app-net-auxiliary-point-based-push-and-pull

2205.00847

https://arxiv.org/abs/2205.00847v2

https://arxiv.org/pdf/2205.00847v2.pdf

APP-Net: Auxiliary-point-based Push and Pull Operations for Efficient Point Cloud Classification

Aggregating neighbor features is essential for point cloud classification. In the existing work, each point in the cloud may inevitably be selected as the neighbors of multiple aggregation centers, as all centers will gather neighbor features from the whole point cloud independently. Thus each point has to participate in the calculation repeatedly and generates redundant duplicates in the memory, leading to intensive computation costs and memory consumption. Meanwhile, to pursue higher accuracy, previous methods often rely on a complex local aggregator to extract fine geometric representation, which further slows down the classification pipeline. To address these issues, we propose a new local aggregator of linear complexity for point cloud classification, coined as APP. Specifically, we introduce an auxiliary container as an anchor to exchange features between the source point and the aggregating center. Each source point pushes its feature to only one auxiliary container, and each center point pulls features from only one auxiliary container. This avoids the re-computation issue of each source point. To facilitate the learning of the local structure of cloud point, we use an online normal estimation module to provide the explainable geometric information to enhance our APP modeling capability. Our built network is more efficient than all the previous baselines with a clear margin while still consuming a lower memory. Experiments on both synthetic and real datasets demonstrate that APP-Net reaches comparable accuracies to other networks. It can process more than 10,000 samples per second with less than 10GB of memory on a single GPU. We will release the code in https://github.com/MCG-NJU/APP-Net.

['LiMin Wang', 'Gangshan Wu', 'Youxin Chen', 'Chunxu Liu', 'Tao Lu']

2022-05-02

['3d-point-cloud-classification', '3d-classification', 'point-cloud-classification']

['computer-vision', 'computer-vision', 'computer-vision']

[-4.19594795e-01 -3.39869589e-01 -5.00909574e-02 -2.62663871e-01 -6.25704527e-01 -5.25008261e-01 1.67419180e-01 4.52435225e-01 -1.77035436e-01 4.26265776e-01 -5.44723332e-01 -2.65726864e-01 5.53364567e-02 -1.29863775e+00 -7.85324216e-01 -6.96648836e-01 -1.29885584e-01 5.88321030e-01 5.73043108e-01 1.62322327e-01 2.67082930e-01 8.59631479e-01 -1.54330301e+00 2.07002878e-01 8.97384286e-01 1.47081256e+00 3.43364000e-01 4.00238544e-01 -3.49161416e-01 1.75221130e-01 -5.65992713e-01 -2.05265820e-01 4.62609410e-01 3.59178960e-01 -4.12917078e-01 -1.97648078e-01 5.23894370e-01 -3.97532463e-01 -2.02522814e-01 8.09756100e-01 4.26250368e-01 1.09935766e-02 3.17231655e-01 -1.52922428e+00 -3.73544425e-01 3.99541736e-01 -7.11185992e-01 -3.81052420e-02 4.18299623e-02 1.45083293e-01 7.85400510e-01 -1.21616781e+00 1.33633941e-01 7.63625979e-01 8.56975734e-01 1.35412872e-01 -8.72581899e-01 -1.02105474e+00 2.49440953e-01 -1.05546275e-02 -1.72724569e+00 -1.11153632e-01 7.83132851e-01 -9.21588838e-02 5.98822415e-01 5.45171320e-01 8.58948410e-01 4.82940346e-01 -1.00520916e-01 5.79844713e-01 4.09151971e-01 2.07655028e-01 2.78009295e-01 -9.43818912e-02 1.86331362e-01 6.71204031e-01 4.30188864e-01 -3.32290620e-01 -4.73074049e-01 -6.02328539e-01 9.77281034e-01 6.26003444e-01 -1.51290923e-01 -4.98965532e-01 -1.16632974e+00 8.40431690e-01 8.96836638e-01 -6.84150606e-02 -3.23137581e-01 3.82386059e-01 2.26060122e-01 4.88342568e-02 6.24446034e-01 1.21639550e-01 -6.12555623e-01 -1.24492161e-01 -9.43566740e-01 2.83979893e-01 7.22588360e-01 1.21548843e+00 1.28685594e+00 -4.27745581e-01 1.26413062e-01 6.43571258e-01 9.88884345e-02 6.36818886e-01 -5.68584651e-02 -9.35994744e-01 7.53116548e-01 1.09068167e+00 -5.14262840e-02 -1.42355013e+00 -3.40531707e-01 -4.92511153e-01 -1.09116673e+00 1.40229702e-01 1.93297073e-01 -7.18353093e-02 -5.13267756e-01 1.10225022e+00 7.88341641e-01 6.02899909e-01 -4.67978686e-01 8.72895420e-01 6.58459663e-01 7.06485689e-01 -3.77616316e-01 6.97538629e-02 1.18669975e+00 -1.17391300e+00 1.84655815e-01 1.24197580e-01 5.90864480e-01 -9.30182397e-01 9.89526033e-01 3.70249689e-01 -1.07670617e+00 -5.01065493e-01 -9.21147406e-01 -6.15753755e-02 -2.62367517e-01 2.56341338e-01 8.95998418e-01 1.83105618e-01 -7.61986136e-01 7.15931118e-01 -1.11317277e+00 1.18736230e-01 7.93381095e-01 6.08228505e-01 -1.85734406e-01 -8.66776258e-02 -5.24564028e-01 9.56443399e-02 -4.57598381e-02 1.68466419e-01 -2.54741043e-01 -1.22990477e+00 -6.29630864e-01 1.86932266e-01 2.45582655e-01 -8.42222750e-01 9.62218940e-01 -5.31916738e-01 -1.02772617e+00 2.06023782e-01 -5.69070637e-01 -1.53651789e-01 5.12834430e-01 -1.85126200e-01 -9.28342193e-02 1.17385194e-01 2.70976365e-01 5.27016103e-01 7.42288768e-01 -1.30775249e+00 -8.36543620e-01 -5.33077478e-01 -1.08223837e-02 3.93037274e-02 -3.13923448e-01 -3.22995991e-01 -8.56891632e-01 -5.81698835e-01 6.12026572e-01 -1.03879011e+00 -2.65089780e-01 3.37704390e-01 -3.34276229e-01 -3.85311633e-01 1.23621726e+00 5.15701286e-02 1.05008411e+00 -2.38240123e+00 -4.50885862e-01 6.03982687e-01 8.35742652e-01 1.68118432e-01 7.60700926e-02 1.70206696e-01 1.01970538e-01 9.02572796e-02 -2.13267311e-01 -6.86268985e-01 -2.67971009e-01 1.44356549e-01 -4.76331443e-01 5.42941749e-01 -4.77295406e-02 7.48868167e-01 -8.18563879e-01 -4.56752658e-01 4.03888166e-01 6.63431048e-01 -6.71496928e-01 -1.02154054e-01 6.46883249e-02 2.45316654e-01 -6.66628242e-01 7.97412694e-01 1.46531916e+00 -6.70542955e-01 -3.31016093e-01 -2.30876431e-01 -1.01487793e-01 2.99915731e-01 -1.45973849e+00 1.52677190e+00 -5.65448642e-01 2.39267007e-01 3.50303240e-02 -6.41109765e-01 9.47007835e-01 -1.32749304e-01 7.15962768e-01 -2.29590714e-01 -9.08824429e-02 3.82955909e-01 -2.30025932e-01 1.54940292e-01 4.75693375e-01 4.70222443e-01 1.24109000e-01 5.53348839e-01 -5.52947998e-01 -1.54381441e-02 -2.23278537e-01 2.01417595e-01 1.15397620e+00 -3.49792093e-01 -1.41811326e-01 1.12318538e-01 4.56679672e-01 6.82576299e-02 6.48605466e-01 7.36606300e-01 1.18773356e-01 8.66266608e-01 2.47112378e-01 -8.67514789e-01 -8.41195405e-01 -1.08273208e+00 -9.80206206e-02 7.92605639e-01 6.82437956e-01 -7.81870306e-01 -4.98262912e-01 -6.62396550e-01 3.56306165e-01 2.15530232e-01 -3.54187548e-01 1.42168269e-01 -8.67415965e-01 -5.27933896e-01 2.56695926e-01 6.87405050e-01 4.87229317e-01 -6.75963342e-01 -5.31323373e-01 1.30428836e-01 1.21147111e-02 -9.40356314e-01 -5.03824651e-01 -7.91881308e-02 -1.17975283e+00 -1.07684278e+00 -2.37561047e-01 -4.95235980e-01 9.93391573e-01 7.99206436e-01 1.01100206e+00 7.33454168e-01 -1.22399554e-02 -1.76719248e-01 -3.17209184e-01 -3.54544938e-01 2.83481479e-01 6.45538807e-01 2.31346637e-02 -1.71528578e-01 5.23244202e-01 -8.95078182e-01 -9.31638300e-01 4.92947996e-01 -6.70162618e-01 1.03746973e-01 3.32298875e-01 5.37228882e-01 8.26188326e-01 1.21519454e-01 3.96705717e-02 -6.66043878e-01 2.25504175e-01 -7.77910411e-01 -7.64073014e-01 -1.79946572e-01 -3.34902793e-01 -2.96206981e-01 8.25756013e-01 -4.54829276e-01 -4.39315349e-01 3.90731722e-01 1.43250331e-01 -7.57816732e-01 1.15464544e-02 2.60812372e-01 -1.57027096e-01 -4.23002988e-01 2.68674642e-01 2.22333670e-01 5.84430583e-02 -4.88082737e-01 1.61695510e-01 6.01196408e-01 2.38193497e-01 -6.86475217e-01 1.19412589e+00 7.46631682e-01 1.91244900e-01 -5.49867690e-01 -5.24110615e-01 -4.68861282e-01 -6.67552531e-01 1.12579860e-01 2.44783640e-01 -9.41210151e-01 -1.16182208e+00 3.78121406e-01 -1.49286509e+00 -1.25440583e-01 -1.30939543e-01 3.08260947e-01 -6.82872832e-02 1.99676886e-01 -5.31131566e-01 -4.89808291e-01 -5.07409036e-01 -1.25690532e+00 1.36857200e+00 1.41004235e-01 5.36980256e-02 -5.89296758e-01 -3.61776441e-01 1.62174582e-01 3.33444685e-01 5.87708205e-02 5.99976718e-01 -4.37015146e-01 -1.24767840e+00 -3.36796284e-01 -6.04951680e-01 5.02339490e-02 1.44464210e-01 2.38804817e-01 -7.83957362e-01 -2.71628171e-01 1.07578672e-01 2.72695154e-01 6.51224434e-01 6.96890652e-02 1.86235595e+00 -2.28327662e-01 -6.12391829e-01 9.94531393e-01 1.41567469e+00 -1.28726318e-01 4.29762483e-01 1.72008663e-01 1.07226491e+00 1.69171602e-01 5.82512677e-01 5.67540944e-01 5.69000542e-01 5.04136324e-01 6.35507882e-01 -2.28993103e-01 1.48157090e-01 -9.87483114e-02 4.90881205e-02 1.04017174e+00 -1.93602324e-01 -7.08416179e-02 -1.00169718e+00 3.55607778e-01 -1.91004992e+00 -6.13113463e-01 -4.84414697e-01 2.31816387e+00 4.33274478e-01 -1.18109363e-03 7.37111503e-03 3.78901549e-02 7.18930900e-01 1.18508838e-01 -6.55923724e-01 2.70500749e-01 1.69561580e-01 2.01911569e-01 7.35835493e-01 4.13506240e-01 -9.60298359e-01 8.10689628e-01 5.04761219e+00 1.02832949e+00 -1.29019046e+00 7.67264366e-02 7.41752148e-01 -4.80843961e-01 -4.75165769e-02 1.70392886e-01 -9.98497427e-01 8.48686934e-01 4.88422275e-01 6.97303042e-02 3.28031242e-01 1.32181656e+00 7.48512894e-02 -1.84109300e-01 -9.47802246e-01 1.25246394e+00 -7.01411143e-02 -1.62940228e+00 1.10369153e-01 3.32266450e-01 5.51680386e-01 5.61070323e-01 -1.02414511e-01 -8.37481618e-02 4.05520871e-02 -6.99518561e-01 4.86331195e-01 3.91909063e-01 7.07041919e-01 -1.11819351e+00 8.20488870e-01 4.45338935e-01 -1.69896054e+00 1.26423910e-01 -7.18514740e-01 -2.57145762e-01 5.57438806e-02 1.10309410e+00 -6.19920313e-01 6.26515031e-01 9.01015341e-01 5.30555248e-01 -6.13177180e-01 1.08492160e+00 2.15148389e-01 3.06761920e-01 -9.72239017e-01 -5.97326644e-03 9.78338718e-02 -1.84123307e-01 4.75695342e-01 7.49092638e-01 7.03590572e-01 7.83748999e-02 4.78819788e-01 8.01178753e-01 -1.83283418e-01 1.06222637e-01 -4.46791887e-01 6.27880812e-01 9.99210119e-01 1.50577676e+00 -9.78812337e-01 -3.92575532e-01 -5.22645116e-01 8.63412738e-01 5.14004469e-01 3.62068452e-02 -1.03725863e+00 -5.79976201e-01 1.04399574e+00 4.85826999e-01 4.79727000e-01 -5.09913087e-01 -6.98005378e-01 -1.14171731e+00 5.90064883e-01 -4.55534816e-01 -2.05084570e-02 -6.09414876e-01 -1.40285981e+00 7.30670035e-01 -2.51750857e-01 -1.63219786e+00 2.24731982e-01 -4.35070962e-01 -7.54545271e-01 8.87053013e-01 -1.34667647e+00 -1.12603080e+00 -9.57036853e-01 6.93143427e-01 2.42991120e-01 1.09274954e-01 5.54624379e-01 4.50169325e-01 -4.76381332e-01 6.26189828e-01 -1.75704837e-01 1.48599565e-01 5.26016533e-01 -8.84490311e-01 6.81566298e-01 5.59294045e-01 -1.13368668e-02 8.77658963e-01 9.46077257e-02 -8.33673060e-01 -1.37939858e+00 -1.27891004e+00 6.62231147e-01 -5.84133506e-01 5.47927737e-01 -6.22504175e-01 -1.06624079e+00 4.95034784e-01 -4.27178115e-01 6.09962583e-01 6.26372516e-01 1.17946379e-02 -2.26830944e-01 -4.70112234e-01 -9.94743943e-01 5.53632021e-01 1.31020033e+00 -2.34711513e-01 -1.54420105e-03 4.71520364e-01 9.07684028e-01 -6.96558535e-01 -9.07277584e-01 4.64356244e-01 2.38060698e-01 -9.80398357e-01 1.03566885e+00 2.61255354e-02 2.94537544e-01 -8.32647085e-01 -2.41623875e-02 -9.23797607e-01 -2.16247469e-01 -5.03911436e-01 -4.39421535e-01 1.15511596e+00 4.05572295e-01 -9.24523771e-01 1.13828731e+00 6.99008226e-01 -1.62462801e-01 -1.28092957e+00 -8.62077177e-01 -7.59788990e-01 -1.91946864e-01 -8.91854644e-01 1.47896433e+00 8.10060620e-01 -3.88489276e-01 8.37791041e-02 1.12757720e-01 5.62290430e-01 5.57397783e-01 4.61849183e-01 1.33078766e+00 -1.50112987e+00 4.38681012e-03 -3.85966212e-01 -4.28733081e-01 -1.39461160e+00 -8.92135799e-02 -8.62914085e-01 -3.27904731e-01 -1.36310768e+00 4.59948331e-02 -1.35527039e+00 1.41265579e-02 5.83503664e-01 -1.49265945e-01 4.50322390e-01 2.62821972e-01 6.63893044e-01 -5.38308203e-01 4.78010207e-01 1.16253793e+00 1.58923954e-01 -3.14968646e-01 2.87773818e-01 -6.68934405e-01 9.65394795e-01 8.78653228e-01 -6.96791053e-01 -2.59060055e-01 -7.92609870e-01 3.07334483e-01 -2.91994065e-01 5.40183485e-01 -1.35321522e+00 6.23156607e-01 -7.60532469e-02 6.96217179e-01 -1.13553762e+00 4.88096446e-01 -1.11121798e+00 2.20540434e-01 1.50037512e-01 3.74793947e-01 4.13193703e-01 1.27948254e-01 3.44830096e-01 -1.19819857e-01 -3.13522667e-02 4.78030413e-01 4.49574888e-02 -7.41702989e-02 1.00756347e+00 3.32740068e-01 -4.84752476e-01 1.17145956e+00 -3.93224299e-01 -4.55252290e-01 -1.11425214e-01 -3.43984663e-01 3.10346186e-01 8.42368305e-01 2.16758505e-01 6.29131913e-01 -1.47335637e+00 -4.44106191e-01 4.18337286e-01 -1.73906479e-02 9.01360929e-01 4.39761579e-01 8.16465676e-01 -8.80850494e-01 7.58877769e-02 3.45607489e-01 -9.58880901e-01 -1.06627452e+00 4.50334579e-01 7.77832270e-02 -1.43003045e-02 -7.92301118e-01 7.62136340e-01 1.84012026e-01 -4.66738641e-01 -1.03068210e-01 -4.25766557e-01 4.01390791e-01 -1.48834825e-01 6.22114956e-01 6.27383769e-01 2.53654689e-01 -5.13169706e-01 -4.46817398e-01 7.52762735e-01 -1.59969047e-01 3.14574182e-01 1.35891414e+00 -7.18327658e-03 -5.51772118e-01 2.62501001e-01 1.39458644e+00 3.67167860e-01 -1.20711088e+00 -1.23632535e-01 -3.41011345e-01 -8.45123649e-01 -7.80332740e-03 -1.07145451e-01 -1.53356159e+00 6.55640960e-01 2.89586127e-01 2.97336817e-01 9.10286009e-01 8.28934461e-02 1.17482340e+00 3.37806046e-01 6.77661598e-01 -6.82734549e-01 -2.95270950e-01 4.76483941e-01 5.67640901e-01 -1.18811500e+00 1.90390453e-01 -9.65662241e-01 -1.99847937e-01 1.15922952e+00 9.23777103e-01 -4.24684048e-01 9.23592627e-01 4.29689139e-01 -1.20586507e-01 -3.36714178e-01 -5.22491157e-01 3.14619631e-01 9.99158099e-02 3.11329842e-01 9.39255860e-03 1.74817562e-01 1.30360141e-01 5.61709404e-01 -5.43439925e-01 -8.34556371e-02 1.38428122e-01 8.73250663e-01 -2.17842728e-01 -1.21220589e+00 -4.11125988e-01 7.51108527e-01 -1.26312852e-01 -2.13852122e-01 -5.34868091e-02 6.36383057e-01 3.05602789e-01 6.13431871e-01 7.32935607e-01 -4.53251779e-01 1.50086269e-01 -3.91709238e-01 -3.07515450e-03 -5.96275151e-01 -4.72858518e-01 -2.91540008e-02 -5.31409740e-01 -9.35725451e-01 -1.33941263e-01 -5.13318479e-01 -1.53869283e+00 -8.68095875e-01 -3.85185808e-01 2.38533407e-01 7.34556317e-01 5.28523028e-01 1.01036286e+00 1.49785623e-01 9.42394853e-01 -1.16244185e+00 -1.14035420e-01 -6.65534973e-01 -2.63564378e-01 -6.42032325e-02 2.30431512e-01 -6.13381267e-01 -5.75550914e-01 -4.64589924e-01]

[7.915178298950195, -3.426178216934204]

59b38cc5-06a4-4ede-ac57-09f56333ca37

generative-adversarial-network-in-medical

1809.07294

https://arxiv.org/abs/1809.07294v4

https://arxiv.org/pdf/1809.07294v4.pdf

Generative Adversarial Network in Medical Imaging: A Review

Generative adversarial networks have gained a lot of attention in the computer vision community due to their capability of data generation without explicitly modelling the probability density function. The adversarial loss brought by the discriminator provides a clever way of incorporating unlabeled samples into training and imposing higher order consistency. This has proven to be useful in many cases, such as domain adaptation, data augmentation, and image-to-image translation. These properties have attracted researchers in the medical imaging community, and we have seen rapid adoption in many traditional and novel applications, such as image reconstruction, segmentation, detection, classification, and cross-modality synthesis. Based on our observations, this trend will continue and we therefore conducted a review of recent advances in medical imaging using the adversarial training scheme with the hope of benefiting researchers interested in this technique.

['Paul Babyn', 'Ekta Walia', 'Xin Yi']

2018-09-19

['medical-image-generation']

['medical']

[ 6.67030811e-01 3.29792947e-01 -2.84357995e-01 -2.73975551e-01 -7.15434670e-01 -5.16441464e-01 5.64145088e-01 1.21796004e-01 -4.21732843e-01 7.76180744e-01 8.69422629e-02 -2.86622852e-01 1.53639853e-01 -7.78478265e-01 -4.89713669e-01 -9.67323661e-01 -2.89387554e-02 4.38701630e-01 7.25407712e-03 3.61636057e-02 -1.26830384e-01 6.18070543e-01 -1.03694153e+00 -1.13054048e-02 9.05117333e-01 8.59702885e-01 -8.67718011e-02 3.86981696e-01 -1.85536407e-02 6.11681700e-01 -4.14310455e-01 -5.45972049e-01 2.42865607e-01 -8.82100463e-01 -5.49363375e-01 2.02236876e-01 7.94419870e-02 -1.40836269e-01 -5.78363180e-01 1.07140291e+00 5.69895864e-01 1.88134536e-01 6.84259653e-01 -1.09247315e+00 -7.23971486e-01 3.15514952e-01 -5.61636865e-01 7.35630989e-02 9.68473256e-02 -7.02957511e-02 4.94844019e-01 -4.76535052e-01 6.86089039e-01 8.77313912e-01 5.48748314e-01 8.42039645e-01 -1.22666752e+00 -5.51648855e-01 -2.46113032e-01 1.13043085e-01 -1.12160850e+00 -2.43035287e-01 1.13181484e+00 -4.61621910e-01 2.03212649e-01 3.79497588e-01 4.38496113e-01 1.05502617e+00 3.53476465e-01 9.80623186e-01 1.39590394e+00 -5.25431275e-01 2.23279446e-01 3.51907045e-01 -5.00840008e-01 7.22011924e-01 -1.44713059e-01 2.48713791e-01 -1.20074786e-01 -1.59252778e-01 1.12032187e+00 1.34841427e-01 -2.78135777e-01 -4.38488185e-01 -1.08439529e+00 1.13109732e+00 7.78703749e-01 3.69418502e-01 -3.95250231e-01 2.40470599e-02 4.77124482e-01 2.24553272e-01 5.50263703e-01 6.46049678e-01 7.27460086e-02 2.13530421e-01 -1.00063562e+00 1.39602602e-01 4.56673652e-01 5.54647684e-01 1.99591234e-01 3.59386474e-01 3.27345505e-02 9.55165148e-01 1.08342312e-01 5.04292846e-01 8.85462523e-01 -1.01007605e+00 -9.94786620e-02 4.65261787e-01 -3.11128616e-01 -9.08364713e-01 -1.53822184e-01 -5.53508401e-01 -1.32235777e+00 3.75097215e-01 2.44137675e-01 3.40759940e-02 -1.18835235e+00 1.80051661e+00 3.99166375e-01 1.38979122e-01 3.90905738e-02 7.16163814e-01 6.49472952e-01 4.18163031e-01 2.02760492e-02 -2.33026877e-01 1.02575374e+00 -6.60139322e-01 -7.32708156e-01 -1.56250745e-01 3.36533301e-02 -9.50337827e-01 6.45453215e-01 2.61683106e-01 -1.03907418e+00 -4.76600081e-01 -9.30049598e-01 1.36629030e-01 -2.76082337e-01 -4.53560442e-01 8.91609728e-01 7.14233875e-01 -7.14916527e-01 5.61098218e-01 -1.01125741e+00 -1.67030886e-01 7.92961836e-01 3.40521604e-01 -5.21355033e-01 -3.31643701e-01 -1.05960214e+00 8.93052459e-01 2.33704031e-01 -1.26243308e-02 -6.81948006e-01 -5.74285805e-01 -9.50284362e-01 -2.36780673e-01 1.46736324e-01 -9.38620210e-01 9.21077192e-01 -1.11964238e+00 -1.27303004e+00 1.15648353e+00 4.53380384e-02 -4.60462838e-01 7.01308370e-01 2.01279357e-01 -4.32878345e-01 4.03567813e-02 8.39592516e-02 7.42794991e-01 1.05938530e+00 -1.02704072e+00 -2.88202882e-01 -5.94690621e-01 -2.87600279e-01 5.09344600e-02 -7.03351274e-02 -1.00885369e-01 -2.02102423e-01 -1.11401951e+00 3.69267970e-01 -1.09632504e+00 -4.61570740e-01 2.78106451e-01 -4.10209090e-01 1.61058053e-01 8.87022078e-01 -6.25261962e-01 6.81981623e-01 -2.23485613e+00 1.48793638e-01 2.10539639e-01 2.41781369e-01 4.79079217e-01 6.92853257e-02 2.29409993e-01 -1.83541387e-01 6.13457151e-02 -7.79687822e-01 -1.81190565e-01 -3.94810885e-01 3.19707155e-01 -2.93707758e-01 5.70071936e-01 4.16195333e-01 1.19937277e+00 -9.69979525e-01 -5.85849047e-01 5.95720232e-01 7.16577232e-01 -4.14118379e-01 3.73418123e-01 -3.42231840e-02 1.00546932e+00 -5.30650139e-01 5.29091597e-01 4.96723026e-01 -1.86319485e-01 -5.31733446e-02 6.69898614e-02 3.44355315e-01 -2.18209893e-01 -8.52928340e-01 1.46335709e+00 -3.61213326e-01 6.78273857e-01 -4.86376360e-02 -1.09362650e+00 6.62961364e-01 4.12797928e-01 7.43136406e-01 -6.11697435e-01 2.32376933e-01 1.36228427e-01 2.42114887e-01 -3.50467205e-01 8.66079628e-02 -7.04448164e-01 1.88989773e-01 2.98772812e-01 -1.92319214e-01 -4.11945611e-01 -8.17518607e-02 6.46021217e-02 7.18840420e-01 -1.25171095e-02 2.83179373e-01 1.49994582e-01 5.41025341e-01 2.83843949e-02 3.22749496e-01 4.67960417e-01 -2.74598151e-01 9.67612505e-01 5.27469367e-02 -2.30307758e-01 -1.08676934e+00 -1.19226778e+00 -4.52182412e-01 3.31666917e-01 -5.10685630e-02 4.78709102e-01 -6.87310457e-01 -6.51592433e-01 -9.15623084e-02 3.32381070e-01 -8.66615653e-01 -3.70898753e-01 -6.44416511e-01 -7.77577341e-01 5.00574529e-01 6.40492857e-01 6.19096458e-01 -1.22921240e+00 -1.99703038e-01 3.85577232e-01 -1.80306301e-01 -1.18967640e+00 -5.83889723e-01 1.69327393e-01 -1.23304057e+00 -1.06022894e+00 -1.21060455e+00 -8.81622434e-01 9.18938637e-01 -8.69091153e-02 9.16772068e-01 -1.17605813e-02 -5.31279683e-01 3.84369344e-01 -3.16365510e-01 -5.43731153e-01 -1.00408828e+00 -5.70122078e-02 -8.04836228e-02 7.28530586e-02 -1.19896054e-01 -7.30318487e-01 -6.40395522e-01 2.68892884e-01 -1.29895556e+00 -7.38493204e-02 6.07399344e-01 1.16390347e+00 7.22893417e-01 -9.36039686e-02 6.41612530e-01 -1.25534391e+00 4.62401688e-01 -3.02248240e-01 -3.73624384e-01 -2.52946820e-02 -6.04184330e-01 -3.13820713e-03 6.31283641e-01 -4.51932311e-01 -8.82984817e-01 -1.37272256e-03 -5.55852771e-01 -4.97779340e-01 -2.31993392e-01 4.78627801e-01 -6.55048564e-02 -4.02247548e-01 5.86735725e-01 3.62605304e-01 4.18951929e-01 -1.33978963e-01 3.79063636e-01 4.14529115e-01 9.36976731e-01 -1.42593026e-01 9.21550095e-01 6.93502188e-01 2.77048796e-01 -8.17424417e-01 -6.15322530e-01 -4.61227059e-01 -4.36890304e-01 1.10586109e-02 8.86723101e-01 -5.46132445e-01 -2.26251319e-01 5.99685192e-01 -8.36091876e-01 -6.73371553e-02 -5.94927609e-01 4.04783756e-01 -5.46299160e-01 4.71819073e-01 -5.14868081e-01 -5.36625385e-01 -3.77861440e-01 -1.29040563e+00 6.57763898e-01 4.53508437e-01 -8.49584341e-02 -1.46443081e+00 -1.50811905e-02 3.30525875e-01 4.66743469e-01 7.53003836e-01 1.02622986e+00 -5.57293892e-01 -5.14992952e-01 -6.93559468e-01 -9.53488518e-03 7.59710550e-01 3.41710031e-01 -3.36307943e-01 -1.09761500e+00 -2.05852792e-01 2.05986157e-01 -2.28780448e-01 6.69065237e-01 6.82327092e-01 1.26888716e+00 -3.21990065e-02 -3.88650984e-01 7.90852785e-01 1.28152001e+00 2.43791327e-01 8.96918952e-01 6.54563447e-03 6.65429354e-01 3.44310492e-01 2.06560642e-01 -5.31270579e-02 -1.70591608e-01 5.80229580e-01 5.17614067e-01 -6.32500529e-01 -3.90333325e-01 -2.43387923e-01 -2.12806046e-01 5.55286288e-01 -1.05095813e-02 -1.61427885e-01 -7.10833967e-01 4.81472373e-01 -1.45720208e+00 -9.53645170e-01 1.82711095e-01 2.09172630e+00 8.58988345e-01 -3.38750221e-02 -9.50715393e-02 2.47578323e-01 6.95454836e-01 2.38413177e-02 -6.50000989e-01 -2.38249853e-01 -5.90384118e-02 4.48311538e-01 4.18282449e-01 2.37883121e-01 -1.05260444e+00 5.90533078e-01 6.79967356e+00 7.55632102e-01 -1.50786436e+00 6.00002473e-03 1.05463767e+00 6.23710811e-01 -2.41597950e-01 -3.85961056e-01 -1.58067405e-01 4.95121658e-01 5.12317955e-01 -2.66029313e-02 2.58608520e-01 7.69002974e-01 1.13114052e-01 -2.59321153e-01 -1.01304924e+00 8.88023555e-01 1.41938567e-01 -1.33195353e+00 9.71645862e-02 2.25009680e-01 1.02115214e+00 -1.94549009e-01 3.81764084e-01 1.04676257e-03 5.74523546e-02 -1.22027779e+00 1.47733614e-01 2.34280512e-01 9.15175080e-01 -7.54703462e-01 8.13370287e-01 3.32445741e-01 -5.28235853e-01 2.62120694e-01 -1.78594384e-02 3.16242784e-01 3.55697870e-01 7.07304835e-01 -1.05347002e+00 5.71122408e-01 1.31453678e-01 3.15144151e-01 -5.25760837e-02 1.15477264e+00 -1.51059091e-01 6.37933433e-01 -1.49359882e-01 3.70845616e-01 8.59648436e-02 -3.04593593e-01 6.08378053e-01 7.35035121e-01 -4.64417078e-02 7.34934658e-02 1.16778351e-01 9.31682706e-01 -2.68250585e-01 -2.16311496e-03 -5.66482723e-01 -1.65498555e-01 1.16187401e-01 1.04062581e+00 -1.00171196e+00 -2.65595675e-01 -2.08549455e-01 1.12148988e+00 -2.53823876e-01 1.52706221e-01 -8.25046659e-01 -2.67054856e-01 4.81711864e-01 1.70859888e-01 1.50362670e-01 -1.94878817e-01 -5.11768639e-01 -8.84319603e-01 -1.33852988e-01 -8.89754713e-01 1.82564348e-01 -4.89259213e-01 -1.28685272e+00 7.63818324e-01 -1.12309188e-01 -1.08751369e+00 -4.46005464e-01 -5.17250180e-01 -4.82676208e-01 1.01690567e+00 -1.50175238e+00 -1.22104621e+00 -2.05586508e-01 5.20694256e-01 5.03624260e-01 -2.03369588e-01 9.76834357e-01 4.70719934e-01 -2.10932270e-01 6.43435478e-01 2.64134020e-01 3.02341193e-01 7.22663820e-01 -1.14892507e+00 2.32469156e-01 7.50652611e-01 3.41273367e-01 5.90366721e-01 8.19372118e-01 -5.17849326e-01 -1.07787120e+00 -1.00307918e+00 4.82336193e-01 -4.36119497e-01 3.41088921e-01 5.32544702e-02 -9.17584658e-01 5.10765374e-01 2.71984213e-03 2.11731136e-01 6.72205508e-01 -3.76248300e-01 2.67761108e-02 -9.02437568e-02 -1.54933226e+00 5.25209308e-01 3.87003720e-01 -4.53599513e-01 -4.30116743e-01 3.35307449e-01 3.10348630e-01 -7.13575721e-01 -9.09832299e-01 5.24027586e-01 4.39666748e-01 -7.70300448e-01 1.09164500e+00 -3.25148940e-01 5.60505569e-01 1.01331342e-02 1.88006625e-01 -1.29418910e+00 -6.89436197e-02 -4.73168641e-01 2.11642846e-01 9.74360883e-01 1.68473542e-01 -8.19080472e-01 1.04101884e+00 6.30206168e-01 -3.09124202e-01 -7.89322853e-01 -9.05282915e-01 -5.61404169e-01 2.70398796e-01 -2.82066435e-01 1.70693159e-01 9.59507346e-01 -4.22446102e-01 1.75602473e-02 -3.89447153e-01 -8.61625001e-02 6.31266534e-01 4.14401889e-02 5.83310843e-01 -8.81731272e-01 -4.88735050e-01 -4.85124916e-01 -6.28012478e-01 -8.71860027e-01 -7.48431757e-02 -9.87609982e-01 1.17182896e-01 -1.36962569e+00 1.30051583e-01 -6.62490129e-01 -1.03771977e-01 3.49833876e-01 -4.01391506e-01 8.38142574e-01 -6.14104606e-03 2.64143974e-01 9.29230377e-02 3.65602285e-01 1.75945675e+00 -3.16279233e-01 5.22411205e-02 3.83848995e-01 -7.01841772e-01 6.94041848e-01 7.10354567e-01 -5.49603879e-01 -2.82051563e-01 -9.70882550e-02 -3.03196311e-01 7.14489222e-02 4.17918980e-01 -9.05236602e-01 -1.62491173e-01 -4.26491164e-02 6.37644589e-01 -5.14507219e-02 3.19632679e-01 -9.59244728e-01 1.95871100e-01 6.40494108e-01 -1.95375025e-01 -2.28758469e-01 1.33036524e-01 3.81133288e-01 -6.19375110e-01 -2.54091203e-01 1.34656477e+00 -3.19579393e-01 -6.67899907e-01 3.91705722e-01 -9.29009989e-02 1.25643358e-01 1.14901543e+00 -2.89163738e-01 1.62457943e-01 -4.83888477e-01 -9.22392488e-01 -1.61561713e-01 3.95581782e-01 2.00948432e-01 6.62838340e-01 -1.14543355e+00 -7.16511786e-01 4.21833515e-01 -7.70901442e-02 1.12531058e-01 2.90014029e-01 8.91590834e-01 -5.03043592e-01 1.63102269e-01 -2.96645522e-01 -7.29777992e-01 -1.00248778e+00 5.67246854e-01 4.02519792e-01 -3.43230307e-01 -6.88838720e-01 7.42577314e-01 2.74473786e-01 -9.62044299e-02 2.28952561e-02 3.58816497e-02 7.60055855e-02 -4.43057209e-01 3.17634314e-01 1.98160633e-01 8.31325278e-02 -4.73055780e-01 -2.37686366e-01 3.67770404e-01 -1.31627634e-01 3.38595733e-03 1.18288231e+00 1.78588688e-01 7.45704547e-02 2.31471345e-01 1.12525046e+00 9.63853672e-03 -1.01654780e+00 -1.57588780e-01 -2.96638370e-01 -3.76440555e-01 9.97833759e-02 -8.66584420e-01 -1.27315629e+00 8.78909230e-01 7.36511052e-01 4.25896168e-01 1.23175049e+00 1.81271225e-01 9.41012681e-01 -1.76881552e-01 3.28626275e-01 -6.06710136e-01 1.13213226e-01 1.93561897e-01 7.79488325e-01 -1.49906516e+00 4.24628891e-03 -4.65723455e-01 -4.89676952e-01 8.04080486e-01 1.61443457e-01 -3.21888804e-01 5.94971657e-01 3.41019213e-01 3.83703142e-01 1.43959150e-01 9.09316912e-02 -5.49759204e-03 5.08883715e-01 9.15110171e-01 6.62651300e-01 4.34144549e-02 -3.09491426e-01 -9.35978442e-02 1.24588057e-01 1.07110836e-01 3.25202376e-01 8.36808920e-01 -1.04297362e-01 -1.57874763e+00 -3.68681312e-01 5.04692197e-01 -6.67940676e-01 -8.69892314e-02 -9.88820717e-02 9.83725548e-01 6.00652173e-02 4.32172716e-01 -1.35612428e-01 1.34356871e-01 1.63916841e-01 -6.86500371e-02 7.35907316e-01 -4.89929289e-01 -3.65314126e-01 8.61055180e-02 -4.27018374e-01 -1.92944393e-01 -4.08442199e-01 -7.39102900e-01 -1.15050590e+00 -7.35833421e-02 -3.46038282e-01 2.40988940e-01 8.09590101e-01 8.34239423e-01 -2.07814667e-02 6.65187359e-01 7.69773841e-01 -6.39036238e-01 -4.59665358e-01 -9.49838340e-01 -5.73057294e-01 6.55051768e-01 3.65242481e-01 -4.23683584e-01 -1.55568033e-01 3.56854320e-01]

[14.113710403442383, -1.9847356081008911]

f6a6b0ed-4e08-4e4b-9ce0-a1a311032c17

improving-zero-and-few-shot-generalization-in

2205.12673

https://arxiv.org/abs/2205.12673v2

https://arxiv.org/pdf/2205.12673v2.pdf

InstructDial: Improving Zero and Few-shot Generalization in Dialogue through Instruction Tuning

Instruction tuning is an emergent paradigm in NLP wherein natural language instructions are leveraged with language models to induce zero-shot performance on unseen tasks. Instructions have been shown to enable good performance on unseen tasks and datasets in both large and small language models. Dialogue is an especially interesting area to explore instruction tuning because dialogue systems perform multiple kinds of tasks related to language (e.g., natural language understanding and generation, domain-specific interaction), yet instruction tuning has not been systematically explored for dialogue-related tasks. We introduce InstructDial, an instruction tuning framework for dialogue, which consists of a repository of 48 diverse dialogue tasks in a unified text-to-text format created from 59 openly available dialogue datasets. Next, we explore cross-task generalization ability on models tuned on InstructDial across diverse dialogue tasks. Our analysis reveals that InstructDial enables good zero-shot performance on unseen datasets and tasks such as dialogue evaluation and intent detection, and even better performance in a few-shot setting. To ensure that models adhere to instructions, we introduce novel meta-tasks. We establish benchmark zero-shot and few-shot performance of models trained using the proposed framework on multiple dialogue tasks.

['Jeffrey P. Bigham', 'Maxine Eskenazi', 'Shikib Mehri', 'Yi-Ting Yeh', 'Cathy Jiao', 'Prakhar Gupta']

2022-05-25

['dialogue-evaluation', 'open-domain-dialog']

['natural-language-processing', 'natural-language-processing']

[ 2.73922265e-01 3.83761019e-01 -1.40278280e-01 -5.55613339e-01 -6.88037217e-01 -6.12752140e-01 1.15047181e+00 1.09515272e-01 -5.07998526e-01 9.38384235e-01 6.15044773e-01 -3.11690718e-01 1.12333201e-01 -5.91389418e-01 -2.15253845e-01 -2.89065868e-01 8.25958401e-02 8.20625782e-01 5.66851906e-02 -1.02921724e+00 3.07605863e-01 -2.95428455e-01 -1.35614240e+00 6.09245241e-01 1.08738613e+00 4.49538767e-01 6.48010150e-02 1.19549680e+00 -3.37178946e-01 8.85729730e-01 -9.74594355e-01 -4.42049682e-01 -8.17555636e-02 -5.99300683e-01 -1.20878172e+00 -1.12865448e-01 3.26999635e-01 -7.60945320e-01 -2.13970795e-01 4.53517377e-01 9.13422108e-01 7.32223868e-01 9.10510600e-01 -1.10560453e+00 -9.35015798e-01 7.77820945e-01 6.95077851e-02 2.63934255e-01 7.83160925e-01 6.97157621e-01 1.11764240e+00 -1.00377762e+00 6.67220235e-01 1.59581792e+00 1.26045957e-01 1.22838950e+00 -1.26847994e+00 -3.86434227e-01 -2.70611435e-01 -2.86415309e-01 -5.02544403e-01 -8.05677652e-01 4.16485816e-01 -2.88901389e-01 1.34750199e+00 3.40689152e-01 -2.70358566e-02 1.92781353e+00 3.82359713e-01 1.04340732e+00 9.75986838e-01 -5.57047427e-01 2.20578536e-01 3.42401177e-01 7.61377454e-01 6.32155716e-01 -2.94708967e-01 7.38941133e-03 -8.00750136e-01 -1.29315704e-01 1.70168415e-01 -3.79259378e-01 -4.76420373e-01 7.09863529e-02 -1.29317153e+00 1.31153297e+00 -6.76695853e-02 3.36972445e-01 1.57961488e-01 -2.61071831e-01 9.82965887e-01 7.55068481e-01 5.12364745e-01 1.17657626e+00 -6.80701017e-01 -6.90015137e-01 -7.03458339e-02 3.20699036e-01 1.45859170e+00 1.02003002e+00 6.98437631e-01 2.55012978e-02 -1.13663602e+00 1.17431569e+00 -1.08958252e-01 1.82024777e-01 9.76796448e-01 -7.20093191e-01 8.34981799e-01 6.75321221e-01 7.70642310e-02 -5.06716132e-01 -5.60809076e-01 6.58538491e-02 -8.41428399e-01 -3.34326744e-01 4.55282986e-01 -6.58396304e-01 -5.19072652e-01 1.87339902e+00 1.89863056e-01 -1.48978025e-01 5.88623822e-01 5.93912721e-01 1.72393596e+00 7.99505353e-01 1.94252357e-01 -1.75964847e-01 1.61326754e+00 -1.21441221e+00 -8.65871310e-01 -4.10313070e-01 1.17865944e+00 -4.83121604e-01 1.96594667e+00 1.19627133e-01 -6.17427409e-01 -6.14028156e-01 -8.31707954e-01 -5.27758062e-01 -6.04062557e-01 -6.09746724e-02 5.52455902e-01 6.46032393e-01 -7.63048947e-01 2.93319196e-01 -2.20792711e-01 -5.62690794e-01 -1.35827661e-01 1.12928964e-01 -1.71895653e-01 3.29871118e-01 -1.66000009e+00 1.09956384e+00 5.17505407e-01 -4.30985481e-01 -1.19272089e+00 -7.48322845e-01 -1.13405836e+00 6.31537214e-02 6.60518467e-01 -6.66229725e-01 1.85498714e+00 -1.39086455e-01 -2.10188532e+00 9.60000813e-01 7.57706314e-02 -7.72220492e-01 4.21055436e-01 -4.37217623e-01 -1.71999186e-01 -2.64153868e-01 6.37058392e-02 7.64135361e-01 6.46763384e-01 -8.68616462e-01 -3.30239058e-01 3.41773182e-02 5.96326590e-01 4.04332578e-01 -6.23180926e-01 -2.59232193e-01 1.13477208e-01 -2.40513951e-01 -7.97816038e-01 -8.10281515e-01 1.08443230e-01 -7.26463556e-01 -7.67479181e-01 -8.22308362e-01 8.59253883e-01 -2.49592915e-01 1.32011318e+00 -1.72688282e+00 8.39361176e-02 -8.03053141e-01 3.76622200e-01 4.65759337e-01 -3.28162521e-01 7.26010561e-01 3.43218923e-01 1.26601800e-01 -1.08659670e-01 -3.64548564e-01 4.50966507e-01 3.69859457e-01 -4.64564979e-01 -1.27414659e-01 2.18302041e-01 1.22676837e+00 -1.15188336e+00 -3.34512770e-01 4.14216429e-01 -6.84730634e-02 -6.67056859e-01 8.74735594e-01 -6.93614244e-01 6.00063264e-01 -4.49746341e-01 2.90886581e-01 -1.50794744e-01 -3.71789545e-01 -3.96432489e-01 1.01800255e-01 1.52857065e-01 5.58730304e-01 -3.78788143e-01 1.95012403e+00 -9.02020037e-01 6.03690982e-01 -2.61881590e-01 -5.78846931e-01 9.50524509e-01 6.77991450e-01 -3.08522973e-02 -6.13621116e-01 3.65404576e-01 -1.50714874e-01 2.15689123e-01 -8.62966239e-01 1.08127153e+00 -1.09036364e-01 -6.88829303e-01 9.34599340e-01 6.64988637e-01 -3.43167990e-01 4.08752948e-01 4.97197747e-01 1.12185454e+00 -3.55076939e-01 6.86648726e-01 -2.45210096e-01 5.91909587e-01 1.72946244e-01 -1.20482840e-01 1.06650162e+00 -4.26537275e-01 8.41331780e-02 4.42649543e-01 -3.92494708e-01 -6.55438602e-01 -6.88085616e-01 -1.49940938e-01 2.21891952e+00 -1.49293453e-01 -4.96519387e-01 -8.39324832e-01 -9.20632660e-01 -9.85094756e-02 1.30888844e+00 -9.11810517e-01 -5.26149273e-01 -2.20466018e-01 -6.94144726e-01 7.21624255e-01 5.00266440e-02 7.50161529e-01 -1.41141319e+00 -5.34636557e-01 2.74523199e-01 -2.24590182e-01 -1.18977070e+00 -7.28967786e-01 3.42801839e-01 -4.06059086e-01 -9.78557646e-01 -5.03910303e-01 -6.44705653e-01 1.20679393e-01 2.18995973e-01 1.76172638e+00 -3.04932762e-02 -2.36296549e-01 7.96829641e-01 -5.43212652e-01 -4.14031744e-01 -1.02091098e+00 6.58504665e-01 2.33065635e-01 -2.61574298e-01 7.17779458e-01 -7.38001317e-02 -2.34619692e-01 4.68792319e-01 -7.04288423e-01 1.23009227e-01 1.61444530e-01 1.39168477e+00 -3.67350221e-01 -7.18225598e-01 1.00295007e+00 -1.19207764e+00 1.80011785e+00 -6.00045741e-01 -9.45089534e-02 4.30197328e-01 -3.07251185e-01 3.64312559e-01 7.06249416e-01 -5.22733748e-01 -1.30733120e+00 -5.41717768e-01 3.24693285e-02 6.04715087e-02 -4.68730569e-01 1.53628483e-01 3.56727988e-02 2.67358452e-01 1.29209244e+00 2.30218783e-01 -1.27614588e-02 -1.89019293e-01 8.16464603e-01 1.05079031e+00 4.25241590e-01 -1.08045435e+00 3.04225326e-01 -2.05935672e-01 -7.02336252e-01 -1.14478838e+00 -1.00426388e+00 -5.14753997e-01 -6.04475319e-01 -3.24823000e-02 1.12423491e+00 -5.77166021e-01 -9.68150973e-01 9.29442719e-02 -1.29046118e+00 -7.57374048e-01 -2.16685921e-01 1.28162026e-01 -6.40515745e-01 3.97853702e-02 -7.46155679e-01 -7.90566206e-01 -8.45568955e-01 -1.29613197e+00 1.18124652e+00 4.12338555e-01 -8.64948809e-01 -1.37695551e+00 5.13362527e-01 4.94725466e-01 4.33405161e-01 -5.37855998e-02 1.05475605e+00 -1.58552051e+00 8.72847810e-02 2.17231616e-01 8.53480920e-02 1.17797136e-01 3.82166833e-01 -2.52793044e-01 -1.33011568e+00 -3.63996655e-01 2.25039169e-01 -1.31606591e+00 4.99140590e-01 -1.11704282e-01 6.73451066e-01 -5.91619253e-01 -6.82152342e-03 1.88386381e-01 6.45048976e-01 1.93551213e-01 1.33432105e-01 1.79105297e-01 4.98772532e-01 7.96200633e-01 7.02085137e-01 5.67537129e-01 3.86649609e-01 6.30677223e-01 7.44631588e-02 3.57348233e-01 1.10652082e-01 -2.42222473e-01 4.34209257e-01 9.51564610e-01 4.37819868e-01 -5.61242938e-01 -9.94126737e-01 4.58556682e-01 -1.63040674e+00 -7.60789394e-01 2.13497773e-01 1.88215578e+00 1.31931078e+00 9.89154726e-02 1.07296415e-01 -4.49278355e-01 4.36763585e-01 4.59103942e-01 -7.27317452e-01 -9.90472436e-01 1.67599276e-01 1.00188322e-01 -2.99363822e-01 5.47714055e-01 -1.01914263e+00 1.16690063e+00 5.87278128e+00 7.01656818e-01 -7.98518717e-01 1.78036407e-01 7.11698055e-01 8.56477916e-02 -1.41974851e-01 -2.72609413e-01 -1.07159185e+00 3.78594160e-01 1.14329040e+00 -7.75925636e-01 1.84913144e-01 9.30036545e-01 1.09681495e-01 9.11977366e-02 -1.63633609e+00 6.42292082e-01 1.33701742e-01 -1.14640570e+00 2.70045161e-01 -1.09233327e-01 7.22266257e-01 -1.30315185e-01 -2.28387378e-02 1.34612215e+00 7.78309107e-01 -1.24005628e+00 -2.86363930e-01 5.79994358e-02 8.54080856e-01 -4.07056838e-01 5.82361221e-01 7.46884227e-01 -6.08388186e-01 8.80482122e-02 -4.83075202e-01 -1.16214961e-01 6.41436800e-02 -1.61492854e-01 -1.70516193e+00 1.65465072e-01 2.17751443e-01 4.17510003e-01 -5.12934804e-01 2.47169808e-01 -1.71244174e-01 4.35733527e-01 1.94027111e-01 -5.66972136e-01 5.13045132e-01 -3.75459678e-02 5.97596645e-01 1.36984229e+00 -1.12088948e-01 3.73585910e-01 4.30561185e-01 7.62823701e-01 -3.79984707e-01 2.82628149e-01 -1.08834600e+00 -2.52045900e-01 5.52727640e-01 1.35345650e+00 1.43113416e-02 -5.09086668e-01 -4.44686681e-01 8.71640801e-01 4.28135902e-01 1.74885496e-01 -5.92535019e-01 -4.33771610e-01 6.54826045e-01 -3.35159749e-01 -4.87986177e-01 -6.21443316e-02 -8.10006931e-02 -1.23976874e+00 -5.53680658e-01 -1.38714719e+00 3.98798406e-01 -4.83783215e-01 -1.41543496e+00 6.74182475e-01 1.78388022e-02 -1.00685143e+00 -1.07986963e+00 -7.34001458e-01 -1.07087648e+00 6.51339591e-01 -9.37869430e-01 -6.95663214e-01 -4.55572069e-01 6.39140368e-01 1.48588538e+00 -6.76983714e-01 1.34847307e+00 -3.35109442e-01 -7.31738806e-01 8.13491881e-01 -1.51945561e-01 3.29759002e-01 1.29471719e+00 -1.52923238e+00 5.67773700e-01 2.77502030e-01 1.78503599e-02 8.01087081e-01 9.38918233e-01 -4.59727794e-01 -1.49597466e+00 -9.69728172e-01 4.93082762e-01 -9.37821984e-01 7.86930978e-01 -5.75669765e-01 -1.12753177e+00 6.28307462e-01 8.35681021e-01 -4.76472735e-01 9.28088367e-01 3.14774990e-01 -5.16015179e-02 4.65175718e-01 -1.00955963e+00 7.95713007e-01 8.83829176e-01 -7.67704844e-01 -1.24734855e+00 5.95930576e-01 1.28604054e+00 -6.85356438e-01 -1.00407708e+00 1.52426705e-01 2.78810591e-01 -9.23304617e-01 7.63741732e-01 -1.15068233e+00 7.55315781e-01 6.20834231e-01 2.47531068e-02 -1.76448131e+00 1.44314766e-01 -1.14368594e+00 -2.83219308e-01 1.30771518e+00 5.67628384e-01 -5.60823321e-01 4.49705034e-01 8.88117909e-01 -3.48281264e-01 -6.66358113e-01 -6.93267465e-01 -5.30859351e-01 2.03400716e-01 9.54611301e-02 4.95098144e-01 1.07218742e+00 4.07146990e-01 1.48027813e+00 -5.97661614e-01 -6.14928842e-01 2.03112423e-01 -4.12604026e-02 1.55940378e+00 -1.07451546e+00 -3.19982797e-01 -3.27204734e-01 2.79295444e-01 -1.32577360e+00 5.01007497e-01 -7.84181297e-01 2.86800712e-01 -1.10032177e+00 9.90764052e-02 -2.56773923e-02 3.15092176e-01 3.95611525e-01 -6.15852416e-01 -4.21610892e-01 -6.83308914e-02 1.08861725e-03 -8.30764413e-01 1.06162250e+00 1.49233437e+00 -3.03473949e-01 -5.72151363e-01 -2.01999433e-02 -6.67358100e-01 4.92537439e-01 7.00418413e-01 -8.06276873e-02 -8.63679111e-01 -3.80084962e-01 -3.27042699e-01 3.60890955e-01 -1.44095436e-01 -5.67310810e-01 2.71472543e-01 -1.83995113e-01 -2.40438506e-01 -2.53942817e-01 5.03259242e-01 -6.10567220e-02 -9.23664808e-01 3.43395948e-01 -8.80556583e-01 2.43904684e-02 1.41497910e-01 5.25757849e-01 -1.71968013e-01 -4.17552471e-01 6.35458946e-01 -2.89440781e-01 -9.93109167e-01 1.89192951e-01 -2.61643171e-01 8.95205140e-01 9.32449162e-01 -5.18121533e-02 -1.04708695e+00 -6.61762059e-01 -5.66427708e-01 6.53571784e-01 1.33825123e-01 9.95810568e-01 5.13869703e-01 -1.00018251e+00 -1.00684571e+00 8.26511607e-02 6.53599322e-01 -1.07440129e-01 1.68573499e-01 4.89638776e-01 -5.38830720e-02 9.14188087e-01 -9.57977474e-02 -8.99860203e-01 -1.16679668e+00 3.49573195e-01 2.37469375e-01 -6.31418884e-01 -5.41301012e-01 1.03420258e+00 5.17277539e-01 -1.12702584e+00 3.12646031e-01 -3.73767763e-01 -3.71335894e-01 2.51002848e-01 7.02081203e-01 2.49696299e-01 -1.02462023e-01 -3.48713279e-01 1.39748618e-01 -3.44201014e-03 -4.26262200e-01 -7.55395666e-02 9.26063001e-01 -1.47395357e-01 2.13207290e-01 8.81893694e-01 1.17536020e+00 -3.74926507e-01 -1.15370810e+00 -4.03102458e-01 1.38359949e-01 -3.30542534e-04 -4.19848531e-01 -8.35033298e-01 -1.59858391e-01 1.05462921e+00 2.13204265e-01 4.85842347e-01 4.38094437e-01 -1.58366963e-01 8.78218234e-01 1.02541602e+00 4.33539003e-01 -1.07298684e+00 6.46855235e-01 1.25979853e+00 1.08195829e+00 -1.93397403e+00 -4.63102162e-01 4.14956622e-02 -1.08383584e+00 9.08075333e-01 1.24405575e+00 2.54323840e-01 1.35104313e-01 -8.90920237e-02 1.20002307e-01 -3.23776007e-01 -1.47655332e+00 -1.20724970e-02 3.82833987e-01 4.64605629e-01 8.56856942e-01 9.83935408e-03 -1.61088586e-01 7.06651628e-01 -5.48882127e-01 -3.01855624e-01 5.22018969e-01 7.38488436e-01 -5.79618871e-01 -8.98069501e-01 -8.23147967e-02 6.11055195e-01 -1.79967493e-01 -2.89183766e-01 -7.63469100e-01 8.85468423e-01 -6.52417481e-01 1.34617436e+00 -1.62779585e-01 -5.46261787e-01 4.16521847e-01 5.21713734e-01 6.09648451e-02 -1.36936378e+00 -1.11221898e+00 -6.99490964e-01 6.42744362e-01 -3.36671054e-01 1.51956618e-01 8.22718740e-02 -9.76320922e-01 -2.78513759e-01 -4.34683204e-01 4.99133050e-01 1.88054100e-01 1.12183607e+00 1.99318647e-01 7.18032598e-01 5.63877702e-01 -6.96842849e-01 -1.18926001e+00 -1.64275777e+00 -1.05496906e-01 5.46089590e-01 3.42545718e-01 -7.29285181e-01 -5.64743161e-01 -3.66194487e-01]

[12.726268768310547, 7.943490505218506]

d9e4f663-2e3d-4e69-ad42-1cb2fb77a06a

nfts-to-mars-multi-attention-recommender

2306.10053

https://arxiv.org/abs/2306.10053v1

https://arxiv.org/pdf/2306.10053v1.pdf

NFTs to MARS: Multi-Attention Recommender System for NFTs

Recommender systems have become essential tools for enhancing user experiences across various domains. While extensive research has been conducted on recommender systems for movies, music, and e-commerce, the rapidly growing and economically significant Non-Fungible Token (NFT) market remains underexplored. The unique characteristics and increasing prominence of the NFT market highlight the importance of developing tailored recommender systems to cater to its specific needs and unlock its full potential. In this paper, we examine the distinctive characteristics of NFTs and propose the first recommender system specifically designed to address NFT market challenges. In specific, we develop a Multi-Attention Recommender System for NFTs (NFT-MARS) with three key characteristics: (1) graph attention to handle sparse user-item interactions, (2) multi-modal attention to incorporate feature preference of users, and (3) multi-task learning to consider the dual nature of NFTs as both artwork and financial assets. We demonstrate the effectiveness of NFT-MARS compared to various baseline models using the actual transaction data of NFTs collected directly from blockchain for four of the most popular NFT collections. The source code and data are available at https://anonymous.4open.science/r/RecSys2023-93ED.

['YongJae lee', 'Joohwan Hong', 'Yejin Kim', 'Youngbin Lee', 'Seonmi Kim']

2023-06-13

['graph-attention', 'multi-task-learning']

['graphs', 'methodology']

[-3.07740986e-01 -4.11687791e-01 -6.14777505e-01 -1.88823298e-01 -5.89436173e-01 -6.19138360e-01 7.65570641e-01 -1.12203501e-01 -1.45795643e-01 2.76960969e-01 7.61674106e-01 -6.42497420e-01 -3.55337888e-01 -5.51041722e-01 -6.47254944e-01 -3.18739504e-01 -1.02663852e-01 6.73594356e-01 -3.66370410e-01 -6.58737004e-01 6.91461086e-01 1.67805359e-01 -1.29094279e+00 2.79505193e-01 1.08942544e+00 1.17589355e+00 2.10760117e-01 3.37669820e-01 -2.04391152e-01 1.00903630e+00 -2.82927275e-01 -8.00099492e-01 6.84699237e-01 1.40264213e-01 -6.48955762e-01 -4.21119660e-01 3.08947325e-01 -6.14241540e-01 -6.78841352e-01 7.16884375e-01 4.63547111e-01 3.57912809e-01 8.57073665e-01 -1.27173531e+00 -1.55228066e+00 1.23046219e+00 -7.85378873e-01 4.55193371e-01 2.51735359e-01 8.89822468e-02 1.60671043e+00 -1.09048235e+00 4.78873223e-01 1.13727462e+00 3.53423655e-01 2.12743297e-01 -1.06226242e+00 -1.03387964e+00 2.57236362e-01 3.48227173e-01 -9.36852098e-01 -3.77449721e-01 4.64767158e-01 -5.29687047e-01 1.21795154e+00 1.97397456e-01 8.78316462e-01 1.42875576e+00 1.34082809e-01 1.20126593e+00 9.16929245e-01 -4.36172411e-02 -2.28471175e-01 1.14395700e-01 3.76002014e-01 -1.70486476e-02 6.23882115e-02 6.35298267e-02 -5.18067598e-01 -2.12851495e-01 8.70781839e-01 4.66262102e-01 3.64625938e-02 -1.71724975e-01 -1.23273110e+00 1.16800129e+00 1.98476374e-01 5.28624475e-01 -7.64671862e-01 3.50901783e-01 4.87999827e-01 5.20834088e-01 5.09692311e-01 6.89110935e-01 -4.62030530e-01 -5.29589534e-01 -6.80545747e-01 4.45258170e-01 9.48260903e-01 1.08250594e+00 2.05454320e-01 1.57790080e-01 -3.24355245e-01 8.62622917e-01 3.41712773e-01 5.21355391e-01 3.85096312e-01 -9.68331695e-01 5.16492307e-01 4.31571603e-01 2.38357291e-01 -8.94090354e-01 -1.01251744e-01 -7.50824153e-01 -4.94220912e-01 -4.72487688e-01 3.31837445e-01 9.33291316e-02 -3.80599171e-01 1.58380258e+00 1.73926607e-01 -2.97646243e-02 -3.19856882e-01 9.66513634e-01 8.31887007e-01 8.55769217e-01 5.88941649e-02 3.84901352e-02 1.20200491e+00 -1.02165008e+00 -7.11077154e-01 6.54859915e-02 4.32690114e-01 -9.34505999e-01 1.11132360e+00 5.32623887e-01 -1.04022813e+00 -1.97110310e-01 -4.86791432e-01 -2.06728831e-01 -3.21994215e-01 -3.19736272e-01 1.06602383e+00 5.76952398e-01 -8.39306712e-01 4.48135257e-01 -2.15074658e-01 -2.80742079e-01 5.63116670e-01 4.61848497e-01 -2.68291477e-02 -1.52242973e-01 -1.48496485e+00 9.19367552e-01 -1.94290608e-01 -1.05555505e-02 -8.75492334e-01 -8.70755911e-01 -3.56241882e-01 3.19274247e-01 7.71804392e-01 -6.16145432e-01 1.32711506e+00 -1.00555551e+00 -1.44554114e+00 1.82791844e-01 5.67668557e-01 -3.02519172e-01 3.54848325e-01 -6.29823446e-01 -4.93077576e-01 -2.98949361e-01 8.73964950e-02 2.31625676e-01 8.39983821e-01 -8.15800846e-01 -5.72004616e-01 -9.90939364e-02 2.53337950e-01 2.35362440e-01 -6.68048263e-01 3.85528445e-01 -2.97271848e-01 -1.07039332e+00 -5.78323543e-01 -9.35338974e-01 -1.41975377e-02 -6.12147152e-01 -3.33136827e-01 -5.65198541e-01 5.31205356e-01 -7.55478382e-01 1.30270624e+00 -1.94387949e+00 1.01991579e-01 2.15594605e-01 2.24211797e-01 9.26454738e-02 -4.74979788e-01 7.64925182e-01 5.90921789e-02 2.20027760e-01 4.80653673e-01 -3.55136953e-02 3.59903514e-01 -1.61889896e-01 -2.63367951e-01 3.44807118e-01 -1.55656248e-01 1.04621160e+00 -9.74934697e-01 7.83885345e-02 -2.45580939e-03 5.93958676e-01 -7.99655914e-01 -6.78277463e-02 -2.21111432e-01 4.05073851e-01 -8.49094570e-01 8.52276206e-01 3.98676395e-01 -5.28768897e-01 1.08294943e-02 -1.45957813e-01 -2.54842162e-01 4.84397024e-01 -1.03581679e+00 1.38625228e+00 -4.12173599e-01 3.08417916e-01 8.58803745e-03 -6.64694965e-01 5.27349889e-01 2.76048392e-01 8.02591264e-01 -9.77619410e-01 4.49327499e-01 3.01764071e-01 1.58529073e-01 -5.66705883e-01 1.00340569e+00 -6.53956458e-02 -2.15417534e-01 1.08215809e+00 -2.88494583e-03 4.03729111e-01 1.59629062e-02 6.80794597e-01 1.10323703e+00 7.96829462e-02 1.58315763e-01 -3.47152978e-01 -1.87161356e-01 -3.08016419e-01 4.55861032e-01 8.10549140e-01 2.09071636e-02 1.11750335e-01 2.49665424e-01 -2.74681330e-01 -1.09820354e+00 -6.20784283e-01 3.83639634e-02 1.63908243e+00 1.61420349e-02 -3.40028286e-01 -2.00376719e-01 -7.76158154e-01 7.05164015e-01 1.01191008e+00 -6.06254339e-01 1.63084641e-01 -3.89667004e-01 -4.55714345e-01 2.72189956e-02 5.91906250e-01 1.01324178e-01 -1.12936032e+00 -1.79496288e-01 2.55030781e-01 -2.66870737e-01 -4.99297261e-01 -1.24803531e+00 -7.55332559e-02 -5.01354694e-01 -8.51174831e-01 -7.64745891e-01 -2.58342177e-01 6.58727884e-02 6.70360804e-01 1.09926009e+00 -3.27138603e-02 3.96791622e-02 7.66800046e-01 -5.15503407e-01 -2.26244986e-01 -1.81768060e-01 1.67703778e-01 4.19951588e-01 1.76620960e-01 3.90040874e-01 -5.35173833e-01 -8.63476276e-01 3.96266311e-01 -6.66848183e-01 -2.01105952e-01 6.87290490e-01 7.60228336e-01 3.81801464e-02 -3.34633619e-01 9.63687181e-01 -1.14320159e+00 8.92498553e-01 -1.20308959e+00 -2.65404820e-01 1.61916077e-01 -1.04955626e+00 -3.34409326e-01 3.45041513e-01 -7.30986238e-01 -1.23118842e+00 -8.31018388e-01 9.68509093e-02 -5.49657643e-01 5.67819297e-01 7.44380236e-01 1.07232429e-01 1.25667438e-01 4.09719169e-01 -6.96015283e-02 -1.62962779e-01 -5.75608075e-01 6.44164324e-01 8.55958581e-01 1.38225898e-01 -8.00497174e-01 6.03206575e-01 -7.59800430e-04 -5.14196217e-01 -4.15591508e-01 -4.41434532e-01 -5.95061064e-01 1.61772683e-01 -1.47502422e-01 3.32598954e-01 -8.44843149e-01 -1.19572937e+00 3.64494294e-01 -7.33703673e-01 -3.04897040e-01 -3.15118849e-01 6.76305175e-01 -4.07685399e-01 4.59225267e-01 -1.15458584e+00 -1.00254941e+00 -8.15149128e-01 -1.02777147e+00 4.27371413e-01 7.47532696e-02 -2.61225283e-01 -7.79114783e-01 1.37643084e-01 9.47626054e-01 8.87157083e-01 -4.16966379e-01 8.34973872e-01 -1.15914536e+00 -9.05889928e-01 -1.69593111e-01 -3.78769785e-01 3.93176824e-01 6.02554996e-03 -7.30207264e-02 -5.26531935e-01 -3.82262856e-01 -1.24152139e-01 -3.97334218e-01 8.39523017e-01 5.17343462e-01 7.55230308e-01 -4.83814746e-01 -4.57558595e-02 2.77916074e-01 1.21367371e+00 3.38408351e-01 3.23394179e-01 4.55339879e-01 8.16705763e-01 6.23680115e-01 6.62798703e-01 7.85848081e-01 7.36488283e-01 6.25073969e-01 4.50174242e-01 3.21280330e-01 2.69709468e-01 -4.42552239e-01 4.81551528e-01 1.02999449e+00 -4.35684711e-01 -4.96712834e-01 -7.53955960e-01 5.15698969e-01 -2.04012799e+00 -1.19622087e+00 -3.81491408e-02 2.00243163e+00 3.45279694e-01 -1.16757892e-01 5.44563234e-01 -2.95435131e-01 6.36612058e-01 -5.10236137e-02 -7.59565771e-01 -4.62213695e-01 -2.10066125e-01 -5.94770685e-02 6.07796669e-01 -1.72288164e-01 -7.58546233e-01 7.37865925e-01 5.88240385e+00 1.05745721e+00 -7.58494496e-01 2.53334999e-01 6.11863434e-01 -4.67660546e-01 -8.03905070e-01 5.29634692e-02 -4.38218623e-01 7.23130703e-01 7.57209182e-01 -2.68269122e-01 9.22679603e-01 5.85697472e-01 1.84632882e-01 3.42700928e-01 -9.17431593e-01 8.20123971e-01 -5.69384024e-02 -1.34445775e+00 2.58329487e-03 6.35794699e-01 7.81751871e-01 4.24444020e-01 6.49323106e-01 7.35262692e-01 6.21749461e-01 -7.62646794e-01 1.06377625e+00 5.33259511e-01 5.56647360e-01 -6.73633933e-01 5.48947573e-01 8.37880149e-02 -8.17403436e-01 -6.62744164e-01 4.67946939e-02 6.83771074e-02 1.78989381e-01 4.51723009e-01 -4.66676682e-01 5.67226112e-01 8.24826121e-01 1.06991148e+00 -2.55683698e-02 9.58607197e-01 2.36753777e-01 6.51173830e-01 -2.28165239e-01 -7.26566389e-02 3.08089852e-01 -5.01395941e-01 6.14520609e-01 1.01341975e+00 7.92096138e-01 1.61580145e-01 -1.38083234e-01 7.46625125e-01 -5.09644508e-01 5.74305892e-01 -5.51293492e-01 -5.81816792e-01 7.55288243e-01 1.26369572e+00 -4.62239891e-01 -6.90431371e-02 -8.86419475e-01 4.79828477e-01 2.49147281e-01 3.33486646e-01 -8.24084401e-01 -2.42417376e-03 7.80907571e-01 3.40799540e-01 5.37602663e-01 1.44867808e-01 5.88481277e-02 -1.42399168e+00 -3.07362974e-01 -1.18098676e+00 4.49613303e-01 -6.38275146e-01 -1.89264154e+00 3.61829668e-01 -2.29995787e-01 -1.04359019e+00 8.09846520e-02 -2.52645552e-01 -4.91892844e-01 6.51615560e-01 -1.36197031e+00 -1.37533200e+00 4.22005981e-01 5.93514800e-01 5.19209504e-01 -4.44205672e-01 3.33752215e-01 5.78325629e-01 -5.87887585e-01 6.72450006e-01 6.72974706e-01 -2.83720016e-01 9.42266524e-01 -1.03601658e+00 6.20717049e-01 3.20375234e-01 -2.74176244e-02 1.10874677e+00 5.11479616e-01 -9.71206963e-01 -1.93221414e+00 -6.14666879e-01 8.05202425e-01 -6.73380017e-01 9.82423246e-01 -3.65911007e-01 -7.46291876e-01 9.09919560e-01 3.75143558e-01 -5.58070540e-01 9.27836955e-01 8.25301826e-01 -7.44237006e-01 -1.61150675e-02 -1.09263957e+00 4.68408287e-01 1.13430059e+00 -4.20463383e-01 -2.37601146e-01 2.07131013e-01 5.87313712e-01 -1.22163687e-02 -1.25499129e+00 -4.76418436e-02 9.93029714e-01 -6.95861697e-01 1.04093325e+00 -5.87226212e-01 4.19069886e-01 2.21954793e-01 -4.01461452e-01 -1.15367830e+00 -8.92250776e-01 -1.08611178e+00 -3.78580838e-01 1.47954202e+00 3.94211143e-01 -7.63272047e-01 6.04754031e-01 8.23659897e-01 -1.08238146e-01 -8.46054971e-01 -7.35988081e-01 -5.69610476e-01 9.50086564e-02 -4.77556884e-01 8.91515315e-01 1.35799837e+00 3.34799230e-01 5.06415188e-01 -9.90824461e-01 -5.31981766e-01 5.52788436e-01 3.07416558e-01 6.49189591e-01 -1.23595715e+00 -6.46465480e-01 -7.49735117e-01 1.98477089e-01 -1.14167392e+00 8.17596819e-03 -1.08531296e+00 -5.72475493e-01 -1.34180343e+00 7.09069669e-01 -7.40231812e-01 -6.73233867e-01 3.98471892e-01 3.60324979e-02 4.78390278e-03 4.40982640e-01 4.63645011e-01 -7.93851852e-01 6.70230329e-01 1.27814722e+00 -2.47628793e-01 6.02904148e-02 1.66213796e-01 -1.42564297e+00 1.34305596e-01 4.87248540e-01 -3.43093783e-01 -3.26924920e-01 -5.02208710e-01 7.46710956e-01 2.30912372e-01 -6.83434084e-02 -2.32050959e-02 1.91613570e-01 -4.19809401e-01 2.49792598e-02 -5.92716932e-01 2.50383407e-01 -7.89847016e-01 3.82418424e-01 1.74102709e-02 -4.62172925e-01 2.96081245e-01 -2.05092341e-01 8.39968979e-01 3.10893536e-01 -1.34704560e-01 7.51466453e-02 -7.12186247e-02 -3.57728004e-01 6.58906817e-01 -5.56305647e-01 6.34267032e-02 7.25149632e-01 -1.30555898e-01 -6.67412937e-01 -7.57163346e-01 -2.33543396e-01 5.85590959e-01 4.19539064e-01 8.38779509e-01 4.00444269e-01 -1.35237384e+00 -9.10429060e-01 -3.98182310e-02 5.97277954e-02 -6.55382633e-01 6.58397436e-01 9.52279389e-01 1.60410106e-01 5.04590154e-01 -3.29656959e-01 6.31266385e-02 -8.13764751e-01 8.88934612e-01 -2.45290577e-01 -3.93488586e-01 -5.03769219e-01 6.88327968e-01 2.78668553e-01 -3.51515830e-01 1.21432647e-01 -2.00657696e-01 -4.51527953e-01 2.65590519e-01 2.35202357e-01 8.21331024e-01 -2.42306575e-01 -6.91679120e-01 -3.95004191e-02 1.23454876e-01 -5.39166629e-01 2.28265658e-01 1.83597505e+00 -1.28250450e-01 -7.25458050e-03 1.76036522e-01 9.09666121e-01 1.78535670e-01 -9.86792982e-01 -5.52245259e-01 2.48150323e-02 -7.60279477e-01 1.91332117e-01 -9.05423224e-01 -1.32156718e+00 3.80968064e-01 2.11298898e-01 5.45509934e-01 7.83142090e-01 -7.61219561e-02 1.14703250e+00 1.65044799e-01 5.35348654e-01 -1.11178207e+00 1.41494155e-01 4.26149398e-01 8.31583738e-01 -1.06076133e+00 -2.83218473e-02 9.07074288e-02 -9.82286036e-01 7.31561482e-01 3.47138882e-01 -2.19942606e-03 7.95791090e-01 -6.40342012e-02 -2.81077862e-01 -2.62553155e-01 -1.04659200e+00 2.13204339e-01 9.98768881e-02 4.07789111e-01 6.62534177e-01 1.37563378e-01 -2.10107848e-01 1.15524423e+00 7.60031044e-02 1.85806174e-02 4.49039996e-01 6.85694396e-01 -1.07579798e-01 -1.24798870e+00 -8.38187635e-02 1.14817309e+00 -8.77287269e-01 -3.56577128e-01 -1.81822643e-01 7.05870748e-01 -2.02711627e-01 9.38598156e-01 -2.21948832e-01 -5.59277296e-01 4.33507800e-01 -2.12632880e-01 3.74489397e-01 -5.65777898e-01 -1.36593008e+00 3.32363456e-01 2.28083402e-01 -4.36621308e-01 -2.06996664e-01 -1.16938674e+00 -6.39345348e-01 -9.34959590e-01 -5.96080184e-01 2.20556334e-01 6.48496568e-01 6.75801396e-01 5.62399924e-01 2.55217999e-01 7.52845466e-01 -8.55722547e-01 -8.27722132e-01 -1.11599267e+00 -9.44054186e-01 4.26819980e-01 1.26949757e-01 -8.41700256e-01 -2.01832399e-01 -5.80845416e-01]

[10.086334228515625, 5.617682456970215]

7edd5daa-ff07-41f9-853b-b60d2991368f

potter-pooling-attention-transformer-for

2303.13357

https://arxiv.org/abs/2303.13357v1

https://arxiv.org/pdf/2303.13357v1.pdf

POTTER: Pooling Attention Transformer for Efficient Human Mesh Recovery

Transformer architectures have achieved SOTA performance on the human mesh recovery (HMR) from monocular images. However, the performance gain has come at the cost of substantial memory and computational overhead. A lightweight and efficient model to reconstruct accurate human mesh is needed for real-world applications. In this paper, we propose a pure transformer architecture named POoling aTtention TransformER (POTTER) for the HMR task from single images. Observing that the conventional attention module is memory and computationally expensive, we propose an efficient pooling attention module, which significantly reduces the memory and computational cost without sacrificing performance. Furthermore, we design a new transformer architecture by integrating a High-Resolution (HR) stream for the HMR task. The high-resolution local and global features from the HR stream can be utilized for recovering more accurate human mesh. Our POTTER outperforms the SOTA method METRO by only requiring 7% of total parameters and 14% of the Multiply-Accumulate Operations on the Human3.6M (PA-MPJPE metric) and 3DPW (all three metrics) datasets. The project webpage is https://zczcwh.github.io/potter_page.

['Chen Chen', 'Guo-Jun Qi', 'Xianpeng Liu', 'Ce Zheng']

2023-03-23

http://openaccess.thecvf.com//content/CVPR2023/html/Zheng_POTTER_Pooling_Attention_Transformer_for_Efficient_Human_Mesh_Recovery_CVPR_2023_paper.html

http://openaccess.thecvf.com//content/CVPR2023/papers/Zheng_POTTER_Pooling_Attention_Transformer_for_Efficient_Human_Mesh_Recovery_CVPR_2023_paper.pdf

cvpr-2023-1

['human-mesh-recovery']

['computer-vision']

[-1.02377027e-01 3.01181134e-02 5.86086214e-02 1.65327042e-02 -9.90406394e-01 2.91765839e-01 7.55354539e-02 -4.02595639e-01 -3.97961527e-01 5.03623009e-01 2.72427946e-01 2.11919069e-01 4.22733091e-02 -1.08469534e+00 -8.59590113e-01 -4.79651004e-01 1.55008301e-01 3.78419042e-01 2.17943519e-01 -1.05098516e-01 -8.11755471e-03 5.75723588e-01 -1.83814335e+00 3.93783897e-01 6.07293904e-01 1.36149108e+00 4.36408430e-01 6.34077847e-01 1.66200653e-01 8.40120792e-01 -1.63541988e-01 -3.52280647e-01 2.70169973e-01 -9.05656591e-02 -9.59889948e-01 -1.69848263e-01 6.13529682e-01 -7.55382359e-01 -6.52071595e-01 7.55604148e-01 9.08460140e-01 -2.26834640e-02 2.38317132e-01 -9.89686012e-01 -5.97238481e-01 1.30284563e-01 -7.42727280e-01 -9.84015130e-03 4.05798644e-01 1.64530165e-02 8.32761049e-01 -1.59045506e+00 7.24247694e-01 1.26565146e+00 7.64087379e-01 3.30174953e-01 -8.80543590e-01 -8.85593355e-01 -2.47794241e-01 5.24288356e-01 -1.65787172e+00 -5.59809029e-01 5.54920912e-01 -1.78643763e-01 1.14558828e+00 2.47994840e-01 9.02363658e-01 7.12467432e-01 3.75926882e-01 7.67999232e-01 8.20693851e-01 -3.44183482e-02 -1.76443845e-01 -2.92765349e-01 -1.99254915e-01 9.90969062e-01 1.61374435e-01 8.28640684e-02 -8.53465319e-01 6.64005578e-02 1.19111097e+00 2.44140297e-01 -6.08927846e-01 -1.17162973e-01 -1.22456157e+00 4.54546124e-01 7.81059325e-01 4.28475738e-02 -6.72278404e-01 5.26122034e-01 2.10394636e-01 2.22722933e-01 5.89962542e-01 1.65775150e-01 -3.94051701e-01 -9.09424946e-02 -8.68522406e-01 2.02960849e-01 4.04789031e-01 1.05686653e+00 7.71940470e-01 -1.14074737e-01 -4.70105410e-02 8.76755118e-01 1.13586925e-01 7.63413966e-01 2.63834685e-01 -1.20903909e+00 3.62597972e-01 4.75366652e-01 1.13803282e-01 -1.22185922e+00 -2.77204961e-01 -2.68272847e-01 -1.08156419e+00 1.30555108e-01 6.26614466e-02 3.44766974e-01 -9.97739315e-01 1.30767798e+00 3.80170852e-01 1.19047150e-01 -4.64345574e-01 1.15892994e+00 1.04177094e+00 7.14368880e-01 -1.49320677e-01 6.77261502e-02 1.39658952e+00 -1.20204675e+00 -6.48542643e-01 2.82276720e-02 1.73296332e-01 -9.56507623e-01 1.24105024e+00 2.51676798e-01 -1.52577305e+00 -5.15653074e-01 -1.04986906e+00 -8.46793413e-01 -1.02858409e-01 1.74647838e-01 4.51206833e-01 -1.62811242e-02 -1.09021449e+00 8.27303886e-01 -1.00769269e+00 -2.38101661e-01 4.96335149e-01 4.68666404e-01 -5.41577220e-01 -3.89441252e-01 -9.78866339e-01 6.92292511e-01 -2.91711092e-01 2.24033237e-01 -6.66829884e-01 -9.95306194e-01 -8.91309261e-01 3.31305504e-01 2.06696868e-01 -8.92868161e-01 1.04593527e+00 -5.47309279e-01 -1.66313899e+00 8.39395463e-01 -3.08423042e-01 -2.09765673e-01 5.29740095e-01 -5.86793900e-01 1.63713485e-01 4.40862864e-01 1.43946543e-01 7.58850813e-01 8.30388069e-01 -9.14816856e-01 -3.48916680e-01 -3.51580083e-01 -1.31494522e-01 2.68736720e-01 -1.66531280e-01 -3.60124595e-02 -6.45434618e-01 -7.80478001e-01 3.40456069e-01 -8.41165721e-01 -1.54102193e-02 4.48823839e-01 -1.15884483e-01 -1.46819483e-02 7.12097883e-01 -1.02366710e+00 1.01060021e+00 -2.08529782e+00 3.61743540e-01 -8.88878852e-02 4.61071998e-01 7.01765493e-02 -2.06062868e-02 1.19582854e-01 7.33748600e-02 -3.69939357e-02 -1.50667146e-01 -6.83375299e-01 -1.97323024e-01 8.44167396e-02 -3.65666479e-01 6.03977025e-01 1.41099319e-01 1.08743763e+00 -5.34780681e-01 -7.10323215e-01 3.53783339e-01 1.01414621e+00 -8.99739921e-01 3.18747282e-01 1.82663381e-01 3.50778550e-01 -1.69107586e-01 9.81058180e-01 7.56341755e-01 -6.71801150e-01 -1.52298078e-01 -5.98669887e-01 -3.48592438e-02 2.43434682e-01 -9.88651097e-01 1.81352568e+00 -7.39047050e-01 4.94697124e-01 2.09719509e-01 -5.24039030e-01 5.17417192e-01 3.85567099e-01 6.55915678e-01 -1.01808441e+00 2.96994925e-01 3.52851897e-01 -4.45883960e-01 -2.91146904e-01 7.08169580e-01 1.23424545e-01 1.50062844e-01 3.50058883e-01 4.76680659e-02 1.64132103e-01 -1.59269124e-01 -9.40427706e-02 1.07883990e+00 2.78417706e-01 2.27502227e-01 -2.46453971e-01 4.06472772e-01 -1.99638531e-01 7.17276335e-01 1.34487182e-01 -1.45662418e-02 1.04186964e+00 1.30110160e-01 -7.04071045e-01 -1.19518137e+00 -1.03864336e+00 -7.11850747e-02 9.44074631e-01 3.39358807e-01 -5.95887303e-01 -5.43384433e-01 -5.75284176e-02 -1.05843082e-01 -1.07545868e-01 -5.51040053e-01 1.40396878e-01 -1.00344908e+00 -3.92830253e-01 2.49108493e-01 6.94006383e-01 8.30726683e-01 -1.06147265e+00 -1.06155217e+00 9.81534496e-02 -6.10233545e-01 -1.12642598e+00 -7.44505465e-01 -4.22696203e-01 -8.21732640e-01 -1.04135370e+00 -1.25757420e+00 -6.14602447e-01 5.70415020e-01 3.81119788e-01 1.19023931e+00 3.96675617e-01 -5.78541696e-01 1.57817081e-01 -1.03703573e-01 -1.60877071e-02 3.93496573e-01 1.47787169e-01 -9.51370075e-02 -1.81514576e-01 -1.18812829e-01 -7.32913435e-01 -9.80631530e-01 3.89543653e-01 -4.73201722e-01 4.92704421e-01 7.55744338e-01 1.05672324e+00 9.68888342e-01 -2.81845599e-01 2.18613446e-01 -5.20986021e-01 1.46357054e-02 -3.79491150e-01 -4.66760457e-01 -2.13159807e-02 -4.60777521e-01 -6.26172870e-02 3.91961038e-01 -3.61281216e-01 -1.02443027e+00 -4.86604348e-02 -2.99147964e-01 -9.12202954e-01 4.14193958e-01 1.66588485e-01 -1.30421698e-01 -1.42290384e-01 1.88589409e-01 1.82094574e-01 -1.21617295e-01 -7.74368525e-01 5.99314421e-02 4.21768069e-01 5.21658301e-01 -5.21037757e-01 5.27405977e-01 7.17039943e-01 -7.10046012e-03 -8.55408490e-01 -7.97132134e-01 -1.89523637e-01 -3.34648818e-01 -1.20336249e-01 8.91055048e-01 -1.21957266e+00 -8.42307031e-01 5.54543138e-01 -1.25629818e+00 -3.37774932e-01 -2.94872135e-01 4.44397032e-01 -7.08293319e-01 2.61844128e-01 -1.06648695e+00 -4.41965550e-01 -9.15341973e-01 -1.24475002e+00 1.36333513e+00 3.31388265e-02 -8.74906480e-02 -2.01120540e-01 -2.43855968e-01 5.06511927e-01 6.08408689e-01 1.61183819e-01 7.07758367e-01 3.37308139e-01 -1.14615571e+00 -8.47797766e-02 -7.04100311e-01 -4.68956446e-03 -2.11402386e-01 -3.42581809e-01 -1.00563824e+00 -3.00240427e-01 -6.05565868e-03 -1.85524464e-01 6.95262253e-01 4.77314562e-01 1.43916309e+00 -2.85556018e-01 -2.34050244e-01 9.99382496e-01 1.39965868e+00 -2.17588082e-01 9.35907006e-01 8.40165019e-02 9.98876512e-01 2.85652727e-01 6.78608894e-01 3.28724563e-01 5.46281159e-01 8.72514784e-01 1.62562579e-01 -3.86866301e-01 -7.02284873e-01 -3.41830105e-01 2.94445276e-01 1.05972242e+00 -5.13477027e-01 3.79708745e-02 -8.99103463e-01 7.22155392e-01 -1.83437014e+00 -7.84321666e-01 7.69642889e-02 2.22009945e+00 7.90234149e-01 -3.15930545e-01 -2.49687687e-01 1.56489223e-01 4.33478862e-01 5.17559536e-02 -4.07560110e-01 -7.10721463e-02 1.32930264e-01 7.37520814e-01 4.95261967e-01 5.88341713e-01 -7.49849558e-01 8.20087194e-01 5.68522024e+00 8.02243412e-01 -1.06836748e+00 3.43674481e-01 4.37680066e-01 -4.25014019e-01 6.06368892e-02 -3.43268514e-01 -5.86618304e-01 3.58959019e-01 7.52219439e-01 -1.62615046e-01 4.34603989e-01 8.10453653e-01 -8.15059543e-02 -1.74367800e-01 -9.27170455e-01 1.39933991e+00 1.22195050e-01 -1.32449722e+00 9.58792493e-02 3.05213064e-01 4.76207376e-01 1.63903072e-01 3.64711322e-02 3.99674065e-02 -1.25644594e-01 -1.16723514e+00 6.28175855e-01 6.79655194e-01 1.23012173e+00 -8.73805165e-01 8.68331015e-01 -1.02257259e-01 -1.62551177e+00 1.39564067e-01 -4.35938507e-01 9.81868827e-04 2.40898356e-01 6.72439754e-01 -2.85911232e-01 5.56150913e-01 1.28777862e+00 7.74204195e-01 -2.72263259e-01 7.67194748e-01 2.15653665e-02 4.58974205e-02 -4.35247362e-01 5.20017207e-01 -3.14814031e-01 1.48388341e-01 4.66069490e-01 8.37468684e-01 6.10955477e-01 2.71731108e-01 -5.97087033e-02 7.38295496e-01 -3.69629294e-01 9.23604742e-02 -3.87766302e-01 2.90651232e-01 4.15194690e-01 1.14233828e+00 -4.95299101e-01 -3.74078989e-01 -3.12925547e-01 1.26329017e+00 3.80995393e-01 1.09438740e-01 -9.94606197e-01 -5.02133727e-01 8.57244492e-01 6.61372006e-01 4.50487077e-01 -1.91759691e-01 -3.95109802e-01 -1.18159831e+00 2.87127703e-01 -6.66277885e-01 1.10944256e-01 -9.33462322e-01 -1.00705945e+00 7.12223768e-01 -2.41821334e-01 -1.23948550e+00 1.88536018e-01 -4.17682707e-01 -1.56372696e-01 8.37531865e-01 -1.59886634e+00 -1.31386805e+00 -8.61010015e-01 6.86788976e-01 5.32341599e-01 2.83016562e-01 7.38410771e-01 9.23770308e-01 -4.80265528e-01 6.86891139e-01 -6.12881720e-01 1.12732522e-01 7.65411973e-01 -7.98231542e-01 4.15093035e-01 5.78190565e-01 -3.84822637e-01 4.86764640e-01 4.33795035e-01 -6.19326651e-01 -1.57340348e+00 -1.03280890e+00 1.00726128e+00 -2.09032133e-01 1.22601494e-01 -1.17719956e-01 -8.53777885e-01 6.70091510e-01 2.96223760e-02 2.28587285e-01 3.38954955e-01 -2.02041566e-01 -2.93419391e-01 -9.26734284e-02 -1.15384185e+00 6.11334801e-01 1.33968818e+00 -5.58997035e-01 -2.76384652e-01 5.89692853e-02 8.87079418e-01 -7.09505975e-01 -1.36916125e+00 5.98243058e-01 7.92194009e-01 -9.65404570e-01 1.22606516e+00 1.52876690e-01 8.36721480e-01 -3.43177229e-01 -4.15748358e-01 -7.11795628e-01 -3.82290244e-01 -4.53291029e-01 -4.99801755e-01 7.66667247e-01 1.88387502e-02 -5.77836752e-01 6.96035802e-01 5.67378283e-01 -1.96005344e-01 -1.12509882e+00 -1.06889379e+00 -4.34489131e-01 -1.11879930e-01 -2.14574635e-01 7.09614813e-01 5.85501492e-01 -8.98853764e-02 1.68980315e-01 -6.62564278e-01 1.04791671e-01 8.13310385e-01 2.65937686e-01 8.48854482e-01 -9.65501904e-01 -1.59189403e-01 -5.66674918e-02 -3.64486694e-01 -1.13223767e+00 -3.01544547e-01 -7.55368829e-01 2.55162660e-02 -1.56126750e+00 2.77329892e-01 -3.14512104e-01 -3.76503123e-03 4.51537102e-01 1.03307486e-01 6.72184229e-01 2.71257222e-01 3.79354089e-01 -4.47371185e-01 9.02687907e-01 1.42980373e+00 6.35714009e-02 1.08220950e-01 -4.65826243e-01 -3.30811679e-01 7.56673753e-01 6.43275559e-01 -3.97328585e-01 -1.44039527e-01 -8.20022523e-01 1.08848475e-01 1.59687191e-01 6.83121204e-01 -1.30339694e+00 3.61310482e-01 1.87419176e-01 5.30993938e-01 -8.97142649e-01 8.16271126e-01 -7.19777465e-01 4.64985996e-01 5.85513294e-01 2.14219794e-01 3.38271648e-01 -2.35353434e-03 3.52235138e-01 -1.86340392e-01 5.27404904e-01 1.10726571e+00 -1.61835700e-01 -4.95282441e-01 7.32355654e-01 -2.15161941e-03 -1.15106665e-01 8.37231100e-01 -1.26122504e-01 -1.86204389e-01 -1.68709457e-01 -6.50663733e-01 1.06451809e-01 7.76246190e-01 3.09057444e-01 1.06886482e+00 -1.48040867e+00 -6.34528399e-01 3.70605141e-01 -9.91254449e-02 3.54146838e-01 6.68651342e-01 1.00615728e+00 -8.99993777e-01 2.91633606e-01 -4.40726787e-01 -5.35148203e-01 -1.34702575e+00 3.50355595e-01 3.90666753e-01 -4.17548418e-01 -1.25002599e+00 7.91797698e-01 2.28210166e-01 -1.52538612e-01 2.02206194e-01 -3.00602883e-01 2.15457663e-01 -1.30425259e-01 6.81442261e-01 7.73559511e-01 7.39871711e-02 -6.04472756e-01 -2.65845299e-01 1.07152057e+00 4.47270647e-02 5.10147735e-02 1.41972768e+00 -1.86210632e-01 -1.87597796e-01 1.92662440e-02 1.23530591e+00 -6.98975921e-02 -1.08813298e+00 -2.49199137e-01 -4.04985309e-01 -7.62344182e-01 2.78981239e-01 -3.30801368e-01 -1.50335288e+00 8.86654258e-01 7.12010682e-01 -4.91676807e-01 1.27773428e+00 -2.75261849e-02 1.47978199e+00 -2.22514253e-02 7.63038814e-01 -8.53617728e-01 -8.29924457e-03 2.76806206e-01 1.36299551e+00 -1.01024234e+00 3.24034721e-01 -6.33287847e-01 -2.61670381e-01 7.26967156e-01 7.00252593e-01 -4.11661327e-01 8.63929093e-01 4.11283791e-01 -2.86692739e-01 -4.42283988e-01 -6.81914091e-01 -2.52025034e-02 2.12543949e-01 3.40071470e-01 4.46711153e-01 7.40470588e-02 -2.70907730e-01 5.19950628e-01 -2.85135090e-01 3.72228384e-01 3.07355314e-01 8.24807763e-01 -1.95425868e-01 -7.18862712e-01 -2.14259565e-01 3.42328787e-01 -4.52897668e-01 -1.56723559e-01 1.71774238e-01 6.40069485e-01 1.25607118e-01 5.41440785e-01 2.15275019e-01 -4.43767607e-01 4.25471097e-01 -2.93835998e-01 5.33391535e-01 -4.01204646e-01 -3.42896700e-01 3.17147113e-02 -3.85441966e-02 -1.27826011e+00 -3.15262914e-01 -1.20597631e-01 -1.18405199e+00 -9.08029199e-01 5.52250724e-03 -3.29171807e-01 2.95584381e-01 4.26356882e-01 7.63619900e-01 5.90082347e-01 3.01885694e-01 -1.34554017e+00 2.05239374e-03 -8.77782762e-01 -2.68355697e-01 1.45702049e-01 2.43765399e-01 -9.61247146e-01 -6.17801771e-02 -6.47507384e-02]

[7.2000532150268555, -1.1487627029418945]

f304f7db-2ae4-482d-95d2-520a7d895e86

adversarial-alignment-of-multilingual-models

2005.09392

https://arxiv.org/abs/2005.09392v1

https://arxiv.org/pdf/2005.09392v1.pdf

Adversarial Alignment of Multilingual Models for Extracting Temporal Expressions from Text

Although temporal tagging is still dominated by rule-based systems, there have been recent attempts at neural temporal taggers. However, all of them focus on monolingual settings. In this paper, we explore multilingual methods for the extraction of temporal expressions from text and investigate adversarial training for aligning embedding spaces to one common space. With this, we create a single multilingual model that can also be transferred to unseen languages and set the new state of the art in those cross-lingual transfer experiments.

['Jannik Strötgen', 'Heike Adel', 'Anastasiia Iurshina', 'Lukas Lange']

2020-05-19

adversarial-alignment-of-multilingual-models-1

https://aclanthology.org/2020.repl4nlp-1.14

https://aclanthology.org/2020.repl4nlp-1.14.pdf

ws-2020-7

['temporal-tagging']

['natural-language-processing']

[-2.00731099e-01 6.00945130e-02 -4.66855198e-01 -3.16968173e-01 -7.71543860e-01 -1.03521872e+00 9.44052756e-01 -4.10509080e-01 -6.93376243e-01 1.04934299e+00 1.53900325e-01 -4.66860563e-01 3.30994606e-01 -3.93682331e-01 -5.67343831e-01 -5.19315958e-01 -3.50991040e-01 3.90036136e-01 4.41514343e-01 -3.70438933e-01 -5.02179503e-01 3.08297217e-01 -8.17195475e-01 1.53469384e-01 5.26634753e-01 5.66995859e-01 -3.91221702e-01 5.09638429e-01 -1.25172004e-01 9.03384745e-01 -3.85342300e-01 -8.02934945e-01 1.65588140e-01 -3.61815125e-01 -8.79695177e-01 -6.93282485e-01 3.51700753e-01 -8.97485688e-02 -7.90755332e-01 9.43042994e-01 5.56847334e-01 2.21804440e-01 4.30998683e-01 -1.19933140e+00 -1.07603335e+00 8.33280981e-01 -1.71243161e-01 3.86722863e-01 4.04463559e-01 -1.50240093e-01 9.66491461e-01 -5.56470811e-01 1.10648894e+00 1.25290084e+00 8.56420517e-01 8.95274103e-01 -1.19747424e+00 -7.87455320e-01 4.51659352e-01 3.69723469e-01 -1.29513752e+00 -5.22044241e-01 8.85812163e-01 -5.02426088e-01 1.32440293e+00 -1.42997250e-01 7.21978724e-01 1.80420446e+00 3.38637471e-01 7.48086572e-01 1.38164091e+00 -6.33176684e-01 -2.16898918e-01 8.20472911e-02 2.14069299e-02 6.45969987e-01 -3.80989671e-01 5.11639714e-01 -7.76633978e-01 3.13717782e-01 5.45189440e-01 -6.35043681e-01 6.63457718e-03 -1.46815956e-01 -1.24017358e+00 8.11759830e-01 1.43308491e-01 9.99182165e-01 9.74206030e-02 3.33096832e-01 7.71308005e-01 6.42980695e-01 8.70331705e-01 3.34799558e-01 -6.89545989e-01 -3.07313323e-01 -7.94073522e-01 -2.49584690e-02 6.66642070e-01 9.33790088e-01 4.32139456e-01 2.60528833e-01 -3.37507367e-01 6.88131213e-01 6.32782653e-02 3.46572548e-01 6.57371283e-01 -5.60026944e-01 3.78308475e-01 1.03797272e-01 -1.92040890e-01 -5.15361369e-01 -3.12367082e-01 -5.53395180e-03 -3.46603334e-01 -1.82769299e-01 4.30781662e-01 -4.36630368e-01 -8.08787048e-01 2.31578302e+00 1.73475772e-01 4.43771690e-01 2.77931690e-01 5.07264078e-01 3.29087794e-01 6.50718331e-01 5.42406201e-01 -2.54354835e-01 1.17860484e+00 -1.10002053e+00 -1.17884135e+00 -2.16404036e-01 9.44105268e-01 -7.47605741e-01 7.08781302e-01 2.32734606e-01 -8.17559302e-01 -3.40327412e-01 -1.12042403e+00 -4.00066823e-01 -9.89506841e-01 -4.18742262e-02 8.64430487e-01 4.28429455e-01 -1.34771168e+00 6.11422181e-01 -1.21514964e+00 -9.20670092e-01 -3.19200456e-01 3.40989769e-01 -6.44992411e-01 3.37253898e-01 -2.03526473e+00 1.55920553e+00 6.36835694e-01 1.79927930e-01 -8.95074904e-01 -4.38057303e-01 -1.00754786e+00 -7.48593688e-01 -7.51002505e-02 -4.31904614e-01 1.47448778e+00 -8.17243218e-01 -1.70722640e+00 1.15840685e+00 -1.37360513e-01 -5.92691839e-01 4.75672573e-01 -4.23112184e-01 -9.78515327e-01 -2.54595459e-01 1.34909183e-01 7.85015047e-01 5.20712137e-01 -6.29892409e-01 -4.63882387e-01 -1.04976378e-01 1.58537611e-01 -7.48320147e-02 -7.54018366e-01 6.06629252e-01 -5.23725927e-01 -9.24429834e-01 -4.43305910e-01 -1.22889674e+00 -1.18777387e-01 -1.25852272e-01 -6.60137013e-02 -5.57575285e-01 8.10381234e-01 -8.32724512e-01 1.52848268e+00 -2.15722299e+00 3.39385092e-01 -3.42736751e-01 -5.24359941e-01 2.04340965e-01 -2.39340201e-01 5.11334479e-01 -1.68943658e-01 2.69658536e-01 -5.37541434e-02 -5.82428873e-01 2.76632607e-01 6.37253821e-01 -6.18939757e-01 5.31739056e-01 3.12487602e-01 1.08820271e+00 -1.15508592e+00 -9.43768442e-01 6.11345582e-02 6.24186575e-01 -2.48905733e-01 2.11581081e-01 -9.27915424e-02 6.09556139e-01 -3.82306844e-01 4.45499718e-01 1.55496433e-01 4.17747438e-01 3.24373096e-01 -7.17300922e-02 -4.13069874e-01 5.32878578e-01 -5.06701767e-01 2.37828040e+00 -6.01008773e-01 7.45346725e-01 -4.15847927e-01 -7.88920760e-01 6.14301383e-01 9.50077951e-01 7.48261809e-01 -7.49165416e-01 4.11919281e-02 4.04185653e-01 -1.38605118e-01 -4.73301440e-01 4.50997204e-01 -4.23588693e-01 -4.99541581e-01 4.06444132e-01 6.41142249e-01 6.82713464e-02 3.12696010e-01 -1.78525075e-01 1.04944801e+00 9.17985082e-01 2.78189927e-01 -4.61463593e-02 3.22530091e-01 -1.59758832e-02 6.73807859e-01 2.37291470e-01 -5.32244802e-01 2.00289905e-01 1.09797992e-01 -7.22050905e-01 -9.76353288e-01 -1.26088905e+00 -3.22003365e-01 1.33044076e+00 -2.51220495e-01 -5.27223647e-01 -5.31858504e-01 -1.14980865e+00 -2.93975383e-01 7.54676521e-01 -8.14039648e-01 -1.34980632e-02 -9.74285483e-01 -5.61813414e-01 1.18336689e+00 6.26768529e-01 1.41147584e-01 -1.15724051e+00 -3.18462327e-02 5.13301015e-01 -2.39223629e-01 -1.54079843e+00 -7.14205205e-01 6.50086462e-01 -7.04070926e-01 -7.15364575e-01 -4.80234474e-01 -1.04114401e+00 1.31465197e-01 -5.73769569e-01 1.04894888e+00 -6.61176682e-01 2.41884395e-01 2.51261592e-01 -4.29296911e-01 -3.94084007e-01 -3.69874179e-01 5.15671670e-01 4.57882434e-01 -1.09929591e-01 6.84189558e-01 -8.25386465e-01 -5.80166765e-02 1.23218983e-01 -6.56865716e-01 -2.32960284e-01 2.42653757e-01 5.16417086e-01 4.61510658e-01 -4.50093687e-01 4.21112746e-01 -7.62131333e-01 5.07557273e-01 -3.45120221e-01 -6.37843013e-01 3.24871868e-01 -6.05163157e-01 4.71240342e-01 5.66554606e-01 -8.94161642e-01 -1.17228091e+00 7.78350532e-02 -2.29311138e-01 -5.13635218e-01 6.21094964e-02 6.70372009e-01 1.98179245e-01 -9.22814459e-02 6.01154923e-01 -5.66814968e-04 -6.15528226e-01 -4.21985328e-01 8.31289589e-01 1.85033664e-01 6.36721909e-01 -8.30490232e-01 8.06742609e-01 2.66356021e-01 -3.47962439e-01 -3.70236695e-01 -1.09771478e+00 -2.59756058e-01 -1.08668101e+00 -2.10128859e-01 1.32396233e+00 -9.01634812e-01 -4.17340174e-02 4.40337121e-01 -1.52775133e+00 -5.26467323e-01 -1.89737067e-01 5.85991323e-01 -5.45159519e-01 4.12765667e-02 -9.55038786e-01 -5.24320185e-01 -1.69458628e-01 -7.51775026e-01 8.57173383e-01 -2.35032469e-01 -5.46340346e-01 -1.54299307e+00 9.56306338e-01 -2.70943701e-01 2.50027597e-01 3.03804815e-01 6.07376039e-01 -6.90718770e-01 -1.92319214e-01 -7.76504204e-02 3.00852597e-01 2.38309205e-01 2.11109910e-02 5.04942760e-02 -1.13707030e+00 -2.49851316e-01 -3.42601746e-01 -4.63455200e-01 4.19594109e-01 5.08380793e-02 2.93956488e-01 -2.47229576e-01 -6.63935661e-01 7.91513443e-01 1.14319122e+00 3.96164149e-01 4.76461381e-01 5.31898141e-01 6.57388270e-01 5.24936557e-01 6.18636250e-01 -2.27105871e-01 4.80413646e-01 1.02253997e+00 -1.56263679e-01 -2.62307152e-02 -9.38926265e-02 -4.44308937e-01 9.13210273e-01 1.59287119e+00 -2.34603748e-01 -1.86065808e-01 -1.02221894e+00 9.79573786e-01 -1.82607675e+00 -8.75930071e-01 2.21167058e-01 1.61831427e+00 1.35098386e+00 2.66441107e-02 -1.95799023e-01 -2.81054497e-01 4.67470765e-01 4.01517212e-01 -1.74959034e-01 -7.04702973e-01 -2.79404998e-01 5.41616440e-01 4.12187368e-01 5.49420595e-01 -1.45930731e+00 1.72669899e+00 7.21964169e+00 5.87040305e-01 -1.46157849e+00 6.22554719e-01 9.56638902e-02 1.76313773e-01 -7.81743154e-02 2.36919001e-01 -9.00847673e-01 1.02680683e-01 1.39256275e+00 -3.96226346e-01 4.91395652e-01 5.15746593e-01 -1.65525638e-02 4.18602049e-01 -1.44896448e+00 8.33374739e-01 -5.80826849e-02 -7.03540206e-01 -1.85577467e-01 -1.59943238e-01 9.69508708e-01 2.81241298e-01 3.63444984e-02 7.83244789e-01 8.19966078e-01 -9.12623703e-01 7.84117460e-01 5.60503960e-01 9.43119347e-01 -5.10688424e-01 5.71264148e-01 -3.09019778e-02 -1.54932380e+00 2.72751153e-01 1.39500812e-01 1.27636269e-01 4.74219024e-01 7.91751146e-02 -6.61591470e-01 7.60452509e-01 6.94728851e-01 9.98090863e-01 -5.03142595e-01 4.16003019e-01 -7.28403807e-01 6.50329173e-01 -2.65747994e-01 8.53133649e-02 3.91990870e-01 -1.15405612e-01 3.68020356e-01 1.51384115e+00 5.90645373e-01 -3.00426930e-01 1.99489966e-01 4.90358651e-01 -8.09893608e-02 2.56792217e-01 -1.06554627e+00 -4.64635015e-01 3.15921485e-01 1.05543017e+00 -2.49568805e-01 -7.08640739e-02 -7.45352387e-01 1.30400002e+00 8.35425615e-01 4.12952632e-01 -1.15249109e+00 -1.23256683e-01 6.73095822e-01 -2.84132630e-01 4.75396588e-02 -7.76328027e-01 1.30727991e-01 -1.55430281e+00 -6.03878237e-02 -7.01924860e-01 6.38575613e-01 -7.37295866e-01 -1.37903214e+00 9.46353614e-01 1.74930871e-01 -1.15831017e+00 -7.07091391e-01 -7.49893129e-01 -3.31715137e-01 8.80714118e-01 -1.47645187e+00 -1.65881646e+00 4.54390228e-01 8.00865591e-01 1.99198097e-01 -1.21446863e-01 1.15162635e+00 7.29454637e-01 -4.32588935e-01 8.43864381e-01 -2.54335105e-01 5.12021184e-01 1.29479861e+00 -1.25458848e+00 6.22170508e-01 1.25001895e+00 5.19750774e-01 8.16805542e-01 6.62899315e-01 -4.63704258e-01 -1.01107907e+00 -1.43570209e+00 1.62125778e+00 -8.79058540e-01 1.51331711e+00 -6.80170178e-01 -7.90701509e-01 1.35483229e+00 7.09550261e-01 2.53418624e-01 4.15645748e-01 3.74947071e-01 -5.05168855e-01 -7.08037093e-02 -5.33060372e-01 8.40608060e-01 1.24047983e+00 -1.28010166e+00 -5.68256319e-01 1.59946665e-01 9.17520523e-01 -4.04885620e-01 -1.36075890e+00 4.39624965e-01 5.26143372e-01 -2.27549046e-01 7.82771528e-01 -8.61506402e-01 5.11698425e-03 -3.74158859e-01 -2.87288893e-02 -1.43260562e+00 -2.23598540e-01 -9.05978501e-01 -1.28166497e-01 1.67932749e+00 5.70695043e-01 -6.18190706e-01 3.28142285e-01 2.36751556e-01 -3.67327839e-01 -2.82265306e-01 -1.36983430e+00 -1.20045996e+00 6.30528986e-01 -6.21142387e-01 1.55062124e-01 1.57433259e+00 2.53374666e-01 5.33233285e-01 -6.87916398e-01 -2.76749767e-02 2.04461649e-01 -2.88473010e-01 2.85639971e-01 -9.24433827e-01 -5.39944507e-02 -2.90673584e-01 -2.80332804e-01 -6.74846530e-01 8.98263216e-01 -1.36858010e+00 1.68500662e-01 -1.18190205e+00 -2.28606597e-01 -3.16382259e-01 -6.01888120e-01 1.07826734e+00 7.41482228e-02 2.17821822e-01 -1.29989624e-01 -5.37166325e-03 -5.39417386e-01 6.47352338e-01 9.86861110e-01 -2.19597429e-01 -3.11466474e-02 -6.31395340e-01 -6.45901114e-02 4.37031955e-01 6.62687719e-01 -7.62395799e-01 -4.03496385e-01 -6.29224181e-01 3.59732002e-01 -7.27439821e-02 -3.25380154e-02 -8.92486274e-01 2.03631371e-01 -8.04032907e-02 -1.97225660e-01 -1.82228416e-01 1.91196620e-01 -7.67295778e-01 3.18541586e-01 4.04006653e-02 -3.03734004e-01 5.53380251e-01 5.90919852e-01 3.19759727e-01 -6.26816213e-01 1.19083866e-01 7.35394180e-01 8.03323910e-02 -9.48155105e-01 4.88804013e-01 -3.79500747e-01 2.38276333e-01 1.04553771e+00 1.05086856e-01 -3.41550022e-01 6.88648373e-02 -1.02650857e+00 7.45579749e-02 2.81894237e-01 8.46105278e-01 -7.02490881e-02 -1.84802759e+00 -5.70017338e-01 -2.48436227e-01 2.38450348e-01 -6.51598692e-01 -1.92074925e-01 9.87093687e-01 -1.51650637e-01 8.91087651e-01 -2.72826433e-01 -2.88521945e-01 -9.35147226e-01 9.32646394e-01 5.02270401e-01 -7.27103770e-01 -2.49504596e-01 6.07218683e-01 -5.54826707e-02 -6.42391562e-01 1.65059250e-02 -2.05114439e-01 -2.48182058e-01 2.45632887e-01 -1.17568262e-02 -1.83047548e-01 3.81905362e-02 -7.36452103e-01 -5.21936059e-01 6.17830157e-01 6.66921511e-02 -6.50459349e-01 1.24792373e+00 7.56527856e-02 -2.80802459e-01 1.15069962e+00 1.30101764e+00 1.98848233e-01 -1.00762796e+00 -3.97875160e-01 3.97555500e-01 2.18358040e-01 -7.90912732e-02 -5.49144030e-01 -9.06152964e-01 8.13743711e-01 6.87499404e-01 8.04744195e-03 1.06314027e+00 6.00880124e-02 7.41923571e-01 4.06088144e-01 6.19140983e-01 -1.24044335e+00 -1.68511063e-01 1.13597238e+00 7.92266250e-01 -9.95556772e-01 -5.15688956e-01 -6.10968620e-02 -3.42407525e-01 1.02755463e+00 6.62220001e-01 6.07423596e-02 7.27769732e-01 5.40173471e-01 4.96953458e-01 6.24332651e-02 -1.00789070e+00 -2.78610587e-01 3.71167988e-01 6.44699216e-01 1.04137158e+00 -7.64197111e-02 -6.45771921e-01 5.04969001e-01 -2.67503440e-01 1.72600582e-01 -5.48777208e-02 9.94460642e-01 4.33178306e-01 -1.74825728e+00 6.79042786e-02 -4.13848817e-01 -7.93441653e-01 -4.88092959e-01 -3.10452878e-01 1.09330332e+00 3.20925206e-01 6.83758378e-01 -2.28446111e-01 -5.39646626e-01 3.96132112e-01 5.60196579e-01 7.76622772e-01 -4.34522271e-01 -6.59891665e-01 -1.39811128e-01 5.37330449e-01 -4.64166433e-01 -7.16143966e-01 -9.47058618e-01 -9.45369720e-01 5.51857911e-02 4.57465425e-02 2.19778940e-01 3.61847550e-01 9.94644463e-01 -8.83232430e-02 5.56567729e-01 2.29968637e-01 -4.85305965e-01 -5.72619252e-02 -1.13475382e+00 -2.02477381e-01 4.88912076e-01 1.74562156e-01 -6.79223120e-01 -1.96190536e-01 4.55018491e-01]

[10.044873237609863, 9.352069854736328]

7e4fabd5-f91f-4130-93ea-3ed3fed2aa37

damo-yolo-a-report-on-real-time-object

2211.15444

https://arxiv.org/abs/2211.15444v4

https://arxiv.org/pdf/2211.15444v4.pdf

DAMO-YOLO : A Report on Real-Time Object Detection Design

In this report, we present a fast and accurate object detection method dubbed DAMO-YOLO, which achieves higher performance than the state-of-the-art YOLO series. DAMO-YOLO is extended from YOLO with some new technologies, including Neural Architecture Search (NAS), efficient Reparameterized Generalized-FPN (RepGFPN), a lightweight head with AlignedOTA label assignment, and distillation enhancement. In particular, we use MAE-NAS, a method guided by the principle of maximum entropy, to search our detection backbone under the constraints of low latency and high performance, producing ResNet/CSP-like structures with spatial pyramid pooling and focus modules. In the design of necks and heads, we follow the rule of ``large neck, small head''.We import Generalized-FPN with accelerated queen-fusion to build the detector neck and upgrade its CSPNet with efficient layer aggregation networks (ELAN) and reparameterization. Then we investigate how detector head size affects detection performance and find that a heavy neck with only one task projection layer would yield better results.In addition, AlignedOTA is proposed to solve the misalignment problem in label assignment. And a distillation schema is introduced to improve performance to a higher level. Based on these new techs, we build a suite of models at various scales to meet the needs of different scenarios. For general industry requirements, we propose DAMO-YOLO-T/S/M/L. They can achieve 43.6/47.7/50.2/51.9 mAPs on COCO with the latency of 2.78/3.83/5.62/7.95 ms on T4 GPUs respectively. Additionally, for edge devices with limited computing power, we have also proposed DAMO-YOLO-Ns/Nm/Nl lightweight models. They can achieve 32.3/38.2/40.5 mAPs on COCO with the latency of 4.08/5.05/6.69 ms on X86-CPU. Our proposed general and lightweight models have outperformed other YOLO series models in their respective application scenarios.

['Xiuyu Sun', 'Yuan Zhang', 'Yilun Huang', 'Weihua Chen', 'Yiqi Jiang', 'Xianzhe Xu']

2022-11-23

['real-time-object-detection']

['computer-vision']

[-2.51289785e-01 -3.67343515e-01 2.17870280e-01 -1.66366532e-01 -3.82478654e-01 -2.40750283e-01 -2.66612368e-03 -2.80606151e-01 -6.49450481e-01 6.38112009e-01 -1.83516830e-01 -1.22237630e-01 -4.45674872e-03 -6.81044638e-01 -6.59198284e-01 -7.29298949e-01 -2.07372591e-01 1.71210244e-01 7.95128644e-01 -1.55517086e-01 2.15033695e-01 6.37565851e-01 -1.49882388e+00 3.33128095e-01 5.57966650e-01 1.45825112e+00 6.49879515e-01 8.04608941e-01 2.92579830e-01 6.47400558e-01 -4.94935393e-01 -8.81808326e-02 7.51853168e-01 8.97702798e-02 -2.87026227e-01 -2.23398283e-01 6.28264725e-01 -4.42749709e-01 -3.09838593e-01 9.79741514e-01 9.84536052e-01 -1.05790585e-01 5.44134498e-01 -1.43770802e+00 -4.01809841e-01 5.15693486e-01 -1.23889768e+00 6.51874125e-01 -3.33191782e-01 2.92642340e-02 6.27905011e-01 -9.33268607e-01 5.08090854e-01 1.09374344e+00 9.21501398e-01 2.41181269e-01 -8.09140623e-01 -1.41088200e+00 5.91705330e-02 2.90304542e-01 -1.56938577e+00 -3.58283550e-01 2.19169408e-01 -6.66344836e-02 1.19466817e+00 1.53537154e-01 2.71842748e-01 5.95682442e-01 4.73603487e-01 7.12358713e-01 1.22675765e+00 -3.33128065e-01 8.59606415e-02 1.95142329e-01 4.77964044e-01 8.89943600e-01 4.72923934e-01 -2.12347284e-01 -7.20821619e-01 7.34256878e-02 7.54166603e-01 -1.08786799e-01 -5.42359613e-02 5.23173399e-02 -8.50618362e-01 5.41339159e-01 8.27261090e-01 5.74633852e-02 -3.87993634e-01 3.21332544e-01 6.25070333e-01 -7.61636868e-02 3.32598500e-02 2.33369261e-01 -3.13786179e-01 2.96546161e-01 -1.25447440e+00 1.54871538e-01 5.32055199e-01 1.46598887e+00 6.00304961e-01 1.09360769e-01 -2.47474298e-01 6.95469201e-01 2.60034442e-01 5.20353973e-01 4.95416403e-01 -1.00898850e+00 4.72915858e-01 4.93008912e-01 -1.64054200e-01 -6.42036736e-01 -1.01557565e+00 -1.06803739e+00 -1.00528193e+00 2.84743488e-01 -1.08171180e-01 -3.46532106e-01 -1.22291982e+00 1.53348172e+00 1.61054164e-01 1.62604496e-01 4.99329232e-02 8.02261293e-01 7.56063879e-01 9.46154654e-01 -2.23541465e-02 2.47844607e-02 1.95751822e+00 -1.43579459e+00 -4.21140432e-01 -2.06116393e-01 7.29967713e-01 -9.57996130e-01 6.71588182e-01 4.72798258e-01 -9.90837693e-01 -8.16428423e-01 -1.42435884e+00 -8.16091001e-02 -2.87112802e-01 7.69105673e-01 5.78978658e-01 8.69515240e-01 -1.30133271e+00 1.01316832e-01 -8.43695819e-01 -5.61772704e-01 3.80756438e-01 7.49839365e-01 -4.12195213e-02 1.12274423e-01 -8.07891548e-01 5.05822837e-01 7.70248473e-01 1.94308355e-01 -8.43121529e-01 -6.04551494e-01 -2.80690998e-01 2.86075473e-01 3.67524266e-01 -6.30806625e-01 1.07856905e+00 -4.63740319e-01 -1.07226384e+00 5.34173846e-01 -8.82999797e-04 -8.96643996e-01 1.70009539e-01 -2.28430212e-01 -3.54968071e-01 1.32123545e-01 3.20865899e-01 1.18401945e+00 3.80298048e-01 -7.45814025e-01 -1.25956953e+00 -5.72896600e-01 -5.09429723e-02 5.57460845e-01 -3.15722138e-01 2.95904934e-01 -6.60556078e-01 -4.87857789e-01 2.00810298e-01 -1.02336454e+00 -1.16450682e-01 -1.36167765e-01 -4.28456336e-01 -1.37707889e-01 1.04079866e+00 -3.77356499e-01 1.39984679e+00 -2.17808437e+00 -4.49842513e-01 1.85282484e-01 4.84824896e-01 4.05816108e-01 3.06460947e-01 5.57331517e-02 2.54463136e-01 -1.73707530e-01 9.92144421e-02 -4.28373933e-01 -1.84002295e-01 -1.60222650e-01 6.96794838e-02 4.47373211e-01 -2.64567822e-01 5.48469365e-01 -2.74806023e-01 -4.86741096e-01 2.83210222e-02 1.84059381e-01 -6.10217869e-01 -1.44603685e-01 3.76725256e-01 -2.25054070e-01 -3.39199185e-01 9.11091924e-01 1.02149677e+00 -2.48520285e-01 -3.26030821e-01 -5.64634740e-01 -6.21332943e-01 -2.10503653e-01 -1.43571079e+00 1.41566551e+00 -3.20568681e-01 8.24828565e-01 4.70028192e-01 -6.38289511e-01 9.80053484e-01 2.16240305e-02 1.32830590e-01 -7.70384669e-01 4.23732549e-01 3.37682962e-01 2.02054784e-01 -2.05736354e-01 7.33350396e-01 3.35044712e-01 1.03251606e-01 9.79901850e-02 4.80717383e-02 4.24402773e-01 2.85567969e-01 3.28250349e-01 1.22237706e+00 -8.98005590e-02 1.99682079e-03 -7.57158399e-01 4.55130398e-01 6.05771430e-02 8.14995706e-01 1.04444754e+00 -4.59763408e-01 5.08860111e-01 2.84946144e-01 -2.54938930e-01 -1.04927504e+00 -9.09754515e-01 -4.51066703e-01 1.36228192e+00 2.84739852e-01 -5.13873875e-01 -9.25348818e-01 -3.31919380e-02 -2.85022080e-01 3.16618025e-01 -1.92500800e-01 1.47288248e-01 -7.80702055e-01 -1.46537995e+00 9.44393337e-01 7.33685553e-01 1.29222822e+00 -9.50669885e-01 -1.09596658e+00 3.01286966e-01 4.98290539e-01 -1.23992419e+00 -4.04222041e-01 5.31574965e-01 -6.31859362e-01 -7.09724605e-01 -6.78631127e-01 -1.04331434e+00 5.29907286e-01 4.56644386e-01 7.21315384e-01 -3.53820562e-01 -5.13137162e-01 -3.66936833e-01 -3.57358813e-01 -3.81374568e-01 3.15169811e-01 3.47537875e-01 3.88878584e-01 -2.00372562e-01 2.39819661e-01 -5.01180768e-01 -9.85641539e-01 4.35369045e-01 -5.80490470e-01 2.95216918e-01 9.52742159e-01 5.92013657e-01 4.41912472e-01 -1.48837879e-01 4.88999099e-01 -6.74538910e-01 3.60552460e-01 -3.89907181e-01 -8.36182654e-01 1.99546814e-01 -7.64668047e-01 2.78677158e-02 7.16528535e-01 -2.20172286e-01 -9.92332399e-01 5.38661182e-02 -4.24541160e-02 -3.62963855e-01 9.82197672e-02 -2.04243846e-02 1.34991631e-01 -2.67162114e-01 8.17201018e-01 8.50317553e-02 -5.00340044e-01 -4.26415205e-01 6.28578067e-02 9.48349476e-01 7.65355945e-01 -4.07086164e-01 3.60506147e-01 6.25419438e-01 6.63287789e-02 -8.02090228e-01 -4.83667761e-01 -5.38554192e-01 -2.89569765e-01 6.16509207e-02 1.06440794e+00 -1.30974066e+00 -8.83658469e-01 6.88949883e-01 -9.97548878e-01 -6.84646443e-02 3.47755641e-01 5.32174349e-01 -8.97639319e-02 1.45879254e-01 -9.18299794e-01 -6.93392217e-01 -1.00474381e+00 -1.44412804e+00 9.37319934e-01 8.29093575e-01 3.34326714e-01 -1.45890310e-01 -5.17875016e-01 5.41625731e-02 6.40365124e-01 -1.32902682e-01 3.58082831e-01 -4.52495456e-01 -7.31986642e-01 8.06616619e-02 -1.10251224e+00 3.09414297e-01 -3.57863396e-01 -1.95539370e-01 -8.96569431e-01 -5.98582566e-01 3.42719443e-02 -1.64283305e-01 9.22423780e-01 6.02087855e-01 1.01102769e+00 1.78962678e-01 -6.03865743e-01 1.12050784e+00 1.51906157e+00 5.21572173e-01 2.96584755e-01 6.00936949e-01 5.77218294e-01 9.41005275e-02 5.20070434e-01 4.37492192e-01 3.74148875e-01 6.73748195e-01 4.43536073e-01 1.20944399e-02 -3.08826059e-01 2.26060957e-01 4.81547117e-01 8.16194952e-01 1.96855627e-02 -3.62258285e-01 -8.97503793e-01 2.77208537e-01 -1.81562042e+00 -4.05612767e-01 -3.69842917e-01 1.89202523e+00 3.31037372e-01 5.96480548e-01 1.13805654e-02 -3.55949223e-01 9.37919497e-01 1.80072896e-02 -5.49405396e-01 -4.63820964e-01 -1.63288921e-01 2.08632469e-01 1.45021808e+00 1.48467854e-01 -1.09833467e+00 1.00990641e+00 4.79548264e+00 1.21292734e+00 -1.07102168e+00 4.55590546e-01 7.76525676e-01 -3.27060044e-01 7.81068623e-01 -8.69498476e-02 -1.78509140e+00 4.51713711e-01 9.64565098e-01 1.28885046e-01 3.11160758e-02 1.02095747e+00 5.19627742e-02 -5.31023502e-01 -7.39858985e-01 1.21815658e+00 2.08716422e-01 -1.43195677e+00 -3.62233937e-01 1.32400729e-02 7.27891207e-01 4.97218341e-01 -9.90468562e-02 4.82080430e-01 1.52851343e-01 -8.17822158e-01 8.33359778e-01 7.35602304e-02 7.06568360e-01 -7.44660676e-01 1.00841856e+00 3.74415755e-01 -1.60298884e+00 -3.50356460e-01 -6.58201098e-01 -3.35718431e-02 2.15657204e-01 3.01792592e-01 -6.56751573e-01 3.92653048e-01 1.19647765e+00 1.76810205e-01 -6.59801006e-01 1.21551490e+00 2.46057868e-01 3.61393631e-01 -1.00626588e+00 -1.73771024e-01 6.73071265e-01 2.05372110e-01 4.12460804e-01 1.33593619e+00 5.03466010e-01 -4.75380421e-02 8.46643671e-02 5.75236201e-01 -2.10912883e-01 -2.42239982e-02 -1.31280869e-01 7.86230266e-01 8.76742303e-01 1.57703543e+00 -1.07513046e+00 -4.29227889e-01 -3.83411229e-01 8.45095813e-01 1.75455987e-01 2.81244181e-02 -1.14122427e+00 -5.78488588e-01 2.38765553e-01 3.20060290e-02 3.10685158e-01 -1.86371401e-01 -2.35739440e-01 -7.21567452e-01 3.98210585e-02 -5.81231833e-01 3.50989103e-01 -8.85438561e-01 -7.67824590e-01 1.01243103e+00 5.96633889e-02 -8.44262958e-01 3.82742852e-01 -8.97993326e-01 -5.24816811e-01 8.83114278e-01 -1.28914475e+00 -1.08099759e+00 -5.47837436e-01 3.45323652e-01 7.77486920e-01 -3.02242190e-01 2.20518202e-01 6.31707549e-01 -1.02933609e+00 8.83113384e-01 1.83349878e-01 -5.10958582e-02 6.88968539e-01 -9.76894915e-01 4.28121388e-01 1.00582993e+00 -2.19700664e-01 4.98574078e-01 5.25856793e-01 -5.49607098e-01 -1.23184192e+00 -1.13883853e+00 1.37367055e-01 6.12290129e-02 5.48803866e-01 -5.48088253e-01 -6.81709230e-01 6.22820735e-01 3.26056242e-01 2.43502632e-01 8.92653167e-02 -3.44551980e-01 1.27358884e-01 -5.16335189e-01 -1.10494637e+00 5.69044054e-01 1.10728467e+00 -6.25524223e-02 -7.72394761e-02 3.69365066e-01 5.80179393e-01 -7.02199459e-01 -5.85259676e-01 4.45207804e-01 5.46731353e-01 -1.13138545e+00 7.69833684e-01 1.45102113e-01 -2.71221191e-01 -5.58596969e-01 -3.55059117e-01 -4.32263285e-01 -6.39489949e-01 -6.02964759e-01 9.54942554e-02 1.16365337e+00 4.35652822e-01 -7.41473615e-01 1.12734866e+00 2.22847596e-01 -4.67512608e-01 -8.89224589e-01 -1.11478364e+00 -7.47555852e-01 -3.90944690e-01 -3.19812834e-01 3.16654354e-01 3.40024740e-01 -5.80256701e-01 3.60465556e-01 -4.05075908e-01 6.05775595e-01 6.65627241e-01 -2.49562666e-01 5.45213699e-01 -7.49178827e-01 -3.00663382e-01 -2.93078840e-01 -4.99241263e-01 -1.22405827e+00 -4.68751431e-01 -7.61062562e-01 -1.64503917e-01 -1.07868659e+00 2.16071367e-01 -6.65392041e-01 -3.89657885e-01 3.68811697e-01 1.77804470e-01 6.46797776e-01 4.08091456e-01 3.90812576e-01 -6.73828781e-01 1.71612248e-01 7.16211438e-01 4.49615926e-01 -3.26277554e-01 -2.05888391e-01 -3.61609757e-01 7.94033647e-01 1.07803237e+00 -5.13696432e-01 -2.26255238e-01 -4.27490234e-01 2.37314910e-01 -2.37268612e-01 1.90428600e-01 -1.67059708e+00 7.98138797e-01 6.36491179e-01 4.51295048e-01 -9.20847476e-01 4.49132025e-01 -6.82017922e-01 1.54910132e-01 6.75519347e-01 2.51047671e-01 6.88078761e-01 3.75196099e-01 2.28140607e-01 3.43948677e-02 -4.03162599e-01 8.66153777e-01 -6.84387609e-02 -1.15451550e+00 1.38478875e-01 -2.72601634e-01 -3.22414339e-01 1.33899903e+00 -4.93597299e-01 -8.10738206e-01 2.48482212e-01 -5.23134708e-01 7.69669890e-01 7.60552213e-02 1.27640650e-01 4.28268433e-01 -1.17541885e+00 -6.19688570e-01 2.05176115e-01 -1.62316844e-01 1.50479957e-01 3.99887294e-01 1.04607451e+00 -1.21192753e+00 7.05269396e-01 -5.98533988e-01 -5.77760696e-01 -1.37181199e+00 2.26166308e-01 3.20895851e-01 -3.99697781e-01 -5.96139133e-01 1.13092697e+00 4.98121232e-01 3.25636230e-02 2.60140836e-01 -3.49172205e-01 -4.84478287e-02 4.48070234e-03 4.71313268e-01 7.51453757e-01 8.28746110e-02 -5.04553556e-01 -4.87996787e-01 5.45388341e-01 -4.20009106e-01 8.46495107e-02 1.08226693e+00 -1.78053156e-01 -1.37437031e-01 -1.56179607e-01 9.76222456e-01 -1.73701465e-01 -1.19178748e+00 -4.72187176e-02 -9.19456929e-02 4.91906032e-02 3.82804692e-01 -5.55546582e-01 -9.69441772e-01 6.96477652e-01 1.20322359e+00 -8.99634585e-02 1.39029276e+00 -5.63391186e-02 9.55232084e-01 3.17329824e-01 3.84086847e-01 -1.29071593e+00 -2.39852056e-01 5.55747628e-01 5.83204329e-01 -9.32973504e-01 1.35027722e-01 -4.66824383e-01 -3.64574492e-01 1.07421219e+00 1.21818674e+00 -3.54678988e-01 5.38469672e-01 7.34106839e-01 -3.27923477e-01 -3.04254919e-01 -6.59026444e-01 -1.72940224e-01 -6.80567324e-02 1.41093269e-01 9.92508084e-02 1.58136919e-01 -3.16785902e-01 5.15854537e-01 -2.78979272e-01 -7.20887259e-02 4.66178924e-01 9.60374475e-01 -7.50217497e-01 -7.30004430e-01 -6.03327394e-01 7.43794143e-01 -5.69520831e-01 -4.03256953e-01 3.28494489e-01 8.82230878e-01 5.83674252e-01 5.12451231e-01 2.44093612e-01 -4.55490023e-01 8.84466618e-02 -1.99130788e-01 9.77284387e-02 -5.43192565e-01 -6.93507195e-01 2.79255062e-01 -3.22744660e-02 -4.87435818e-01 -4.38640732e-03 -3.39008033e-01 -1.40630710e+00 -4.10499185e-01 -5.39576173e-01 -6.11765832e-02 7.79002070e-01 5.99083841e-01 7.12054729e-01 7.19782174e-01 9.60836038e-02 -9.30983305e-01 -3.60130847e-01 -1.00060678e+00 -7.18458772e-01 -4.71640557e-01 -2.61110589e-02 -6.07448161e-01 -1.39954910e-01 -4.05647606e-01]

[8.704943656921387, -0.2918030619621277]

addbf691-3d3d-4b3d-ac76-efde7e386c5e

m-2-dar-multi-view-multi-scale-driver-action

2305.08877

https://arxiv.org/abs/2305.08877v1

https://arxiv.org/pdf/2305.08877v1.pdf

M$^2$DAR: Multi-View Multi-Scale Driver Action Recognition with Vision Transformer

Ensuring traffic safety and preventing accidents is a critical goal in daily driving, where the advancement of computer vision technologies can be leveraged to achieve this goal. In this paper, we present a multi-view, multi-scale framework for naturalistic driving action recognition and localization in untrimmed videos, namely M$^2$DAR, with a particular focus on detecting distracted driving behaviors. Our system features a weight-sharing, multi-scale Transformer-based action recognition network that learns robust hierarchical representations. Furthermore, we propose a new election algorithm consisting of aggregation, filtering, merging, and selection processes to refine the preliminary results from the action recognition module across multiple views. Extensive experiments conducted on the 7th AI City Challenge Track 3 dataset demonstrate the effectiveness of our approach, where we achieved an overlap score of 0.5921 on the A2 test set. Our source code is available at \url{https://github.com/PurdueDigitalTwin/M2DAR}.

['Ziran Wang', 'Zihao Li', 'Rohit Gupta', 'Kyungtae Han', 'Amr Abdelraouf', 'Liangqi Yuan', 'Yunsheng Ma']

2023-05-13

['action-recognition-in-videos']

['computer-vision']

[ 1.47416353e-01 -1.69570178e-01 -2.92590767e-01 -4.11725372e-01 -1.19444346e+00 -3.32313567e-01 5.35375834e-01 -4.45042044e-01 -2.59510159e-01 4.54868019e-01 5.73574126e-01 -1.27931386e-01 -1.90577790e-01 -3.58603686e-01 -5.39788961e-01 -6.84725285e-01 1.48370132e-01 3.70932673e-03 5.12467086e-01 -3.19679409e-01 2.88723916e-01 5.31554401e-01 -1.85843623e+00 4.44578767e-01 5.36772966e-01 1.03352022e+00 -1.27557114e-01 7.06403017e-01 6.79674387e-01 1.10719287e+00 -2.07449496e-01 -5.25077045e-01 5.62261641e-01 -8.76004025e-02 -5.68143964e-01 3.06079179e-01 7.09767163e-01 -4.27323282e-01 -8.74850094e-01 8.18173826e-01 4.45255280e-01 3.07874441e-01 2.90732473e-01 -1.69485378e+00 -8.99012908e-02 1.55745223e-01 -6.07576728e-01 6.59690380e-01 2.13444427e-01 4.69253242e-01 1.02870107e+00 -8.80535066e-01 4.12490845e-01 1.06984067e+00 3.53024751e-01 4.07766253e-01 -7.81965077e-01 -1.00793910e+00 2.56602168e-01 8.93736124e-01 -1.39103854e+00 -9.40784156e-01 7.68077314e-01 -4.81009543e-01 1.02497411e+00 1.88633606e-01 5.65717816e-01 1.22425413e+00 4.46925879e-01 1.27256477e+00 1.06233633e+00 1.06195189e-01 -1.07769168e-03 -2.10333660e-01 2.33947694e-01 7.77150035e-01 2.38875002e-01 1.91601321e-01 -8.32378447e-01 6.43769354e-02 4.94113445e-01 1.87885702e-01 2.04386473e-01 -5.20245790e-01 -1.10007572e+00 7.67508388e-01 1.55675933e-01 7.27280900e-02 -4.35017377e-01 3.48427296e-01 4.86887902e-01 1.42995268e-01 3.80576789e-01 1.33698076e-01 -7.85905272e-02 -6.75316334e-01 -5.66306353e-01 1.31389096e-01 1.56105831e-01 8.81097555e-01 5.30859411e-01 1.33063197e-01 -3.05883735e-01 7.96363354e-01 2.65698489e-02 5.77932477e-01 3.37768942e-01 -1.37987590e+00 5.42827308e-01 7.04823136e-01 3.46032088e-03 -8.84337008e-01 -5.30005157e-01 -5.99347092e-02 -5.57139874e-01 4.80300814e-01 1.85976177e-01 -8.42018705e-03 -7.45409667e-01 1.58811116e+00 3.47461104e-01 3.85004222e-01 -6.20217528e-03 7.90441930e-01 5.21676540e-01 7.65452022e-03 1.83018073e-01 1.30801857e-01 1.23204315e+00 -1.14041626e+00 -5.15907764e-01 -4.24428076e-01 5.91251910e-01 -4.72728342e-01 5.83093762e-01 4.38259721e-01 -1.03262317e+00 -7.58264482e-01 -1.00551558e+00 1.12264864e-01 -3.00697833e-01 1.93740174e-01 5.31754017e-01 4.75277692e-01 -8.85656178e-01 1.20410152e-01 -1.00636327e+00 -2.70737529e-01 7.81052113e-01 2.41168082e-01 -7.20326424e-01 -3.58512998e-01 -1.02367854e+00 1.08640754e+00 1.86323732e-01 2.96821166e-02 -1.08750772e+00 -2.31623441e-01 -8.22911799e-01 -3.47956419e-01 6.10924423e-01 -3.50419581e-01 1.11073911e+00 -5.61478913e-01 -9.29543138e-01 7.10430503e-01 -3.78143400e-01 -6.18215382e-01 5.32400250e-01 -3.99493009e-01 -7.72339761e-01 3.05396497e-01 3.83470565e-01 6.62389398e-01 9.79808629e-01 -9.74340022e-01 -1.24166262e+00 -5.24714649e-01 -6.61938830e-05 3.50796968e-01 -1.62092209e-01 8.05996135e-02 -4.77665246e-01 -2.99254566e-01 -5.74169904e-02 -1.06249809e+00 -1.68213934e-01 -3.68132830e-01 -1.67582065e-01 -4.73518372e-01 7.75807321e-01 -6.41745567e-01 1.16923308e+00 -2.40662122e+00 8.42127874e-02 1.57669932e-01 4.09657508e-01 1.39666632e-01 -1.38303339e-01 2.14956209e-01 -9.10548568e-02 -3.14672470e-01 -1.05544052e-03 -2.41854727e-01 4.26432751e-02 1.06421173e-01 -9.25511718e-02 6.51167929e-01 3.32580715e-01 9.83193696e-01 -8.67932916e-01 -3.99197251e-01 4.99629527e-01 2.45394647e-01 -2.96082377e-01 8.61006826e-02 4.20975655e-01 3.21725577e-01 -5.84564865e-01 8.80044997e-01 3.08986783e-01 -2.03765612e-02 -1.41231745e-01 -8.23037773e-02 -1.45079896e-01 -1.00932851e-01 -1.20251107e+00 1.58857131e+00 -9.79430079e-02 6.73067927e-01 5.24001978e-02 -9.96820092e-01 5.73181689e-01 2.25539893e-01 9.27165449e-01 -1.08875215e+00 2.65820831e-01 -6.41710684e-02 -2.06987886e-03 -7.75630713e-01 4.84617472e-01 2.71962941e-01 -3.11613113e-01 3.13197017e-01 -8.11505616e-02 1.24739125e-01 4.80057895e-01 2.78119653e-01 1.57814026e+00 1.81605548e-01 3.31068903e-01 1.02679588e-01 5.87381124e-01 1.05476134e-01 6.06390178e-01 6.03098810e-01 -8.90317440e-01 3.87796223e-01 2.92661041e-01 -4.81314003e-01 -7.84184456e-01 -9.97555017e-01 2.54124752e-03 1.13516641e+00 2.56230325e-01 -2.95028001e-01 -4.77222890e-01 -9.47209179e-01 2.05158070e-01 8.80557358e-01 -6.36349559e-01 -5.19276261e-01 -6.74091995e-01 -3.66442651e-01 5.25573671e-01 8.74866426e-01 6.52847767e-01 -1.06777799e+00 -9.80058491e-01 -6.38237447e-02 -4.75931615e-01 -1.35083759e+00 -4.74163055e-01 -1.51408285e-01 -4.62547839e-01 -1.33118200e+00 -1.58209220e-01 -4.35102552e-01 4.20753151e-01 7.86842346e-01 8.44500601e-01 -1.02066495e-01 -3.95184845e-01 7.51492977e-01 -3.22038233e-01 -4.73552316e-01 -1.89525008e-01 -1.53332427e-01 3.33117902e-01 4.00403857e-01 9.26252782e-01 -4.59464997e-01 -7.67523885e-01 6.94748819e-01 -6.36887133e-01 -1.06225293e-02 8.44951689e-01 5.21368146e-01 4.67745751e-01 -5.20028546e-02 7.39922345e-01 -1.95300370e-01 3.75244409e-01 -4.96621221e-01 -5.11274099e-01 -6.23160713e-06 -4.39049184e-01 -3.16203624e-01 1.38453200e-01 -1.02599077e-01 -9.50581253e-01 3.93168390e-01 -1.37648150e-01 -7.75856316e-01 -6.29816830e-01 -1.24719189e-02 -1.14284292e-01 3.30955833e-02 4.99378055e-01 3.54620993e-01 1.79847494e-01 -5.24381474e-02 4.61488456e-01 7.47258186e-01 4.67528462e-01 -7.76956305e-02 7.22714722e-01 7.23371327e-01 -1.51583329e-01 -6.26304209e-01 -8.15006196e-01 -9.07878101e-01 -7.97078252e-01 -6.66705430e-01 1.01087534e+00 -1.34588981e+00 -7.12765098e-01 4.46771950e-01 -5.15357018e-01 -1.83832824e-01 -2.38279611e-01 6.93842351e-01 -8.28633547e-01 2.26903021e-01 -1.79444522e-01 -7.90199637e-01 4.80170809e-02 -1.18685448e+00 1.13965178e+00 -3.39318835e-03 -1.86662629e-01 -4.06777114e-01 3.53750251e-02 1.29165888e+00 2.46141508e-01 1.83194429e-01 3.09703797e-01 -6.86600089e-01 -6.67294502e-01 -4.37688947e-01 -1.15139961e-01 4.47992325e-01 1.36745870e-01 -3.00777793e-01 -1.05141115e+00 -2.88024366e-01 -3.02719682e-01 -5.25661230e-01 1.12620473e+00 3.65310192e-01 1.11218524e+00 2.18397960e-01 -4.20838535e-01 1.83002695e-01 8.45532298e-01 3.33558589e-01 7.87486434e-01 2.60444075e-01 6.89815342e-01 4.78180259e-01 8.65284026e-01 5.90285659e-01 5.96922517e-01 7.36520946e-01 5.10272503e-01 1.35846967e-02 -3.46088707e-01 -5.39134964e-02 5.67916512e-01 5.61384141e-01 -2.46115610e-01 7.77148977e-02 -9.13405776e-01 8.55116725e-01 -2.12683606e+00 -1.51363587e+00 1.23160176e-01 2.08367348e+00 2.26332396e-01 4.34107304e-01 4.48996305e-01 8.98862705e-02 3.92151773e-01 2.84426332e-01 -8.11953843e-01 -1.26601979e-01 2.05931351e-01 -1.43834621e-01 6.09834135e-01 2.79066950e-01 -1.49338007e+00 8.68528247e-01 5.81057692e+00 8.83596957e-01 -8.01126838e-01 2.47124061e-01 6.33555114e-01 -5.54377556e-01 6.99783564e-02 -5.36289573e-01 -8.96695554e-01 2.26041749e-01 1.07950521e+00 -2.29770169e-01 2.98722237e-01 8.45617533e-01 5.16710460e-01 -2.91378021e-01 -8.18353772e-01 1.08536589e+00 4.21132654e-01 -1.11365473e+00 -3.61913115e-01 8.79813954e-02 5.06034076e-01 5.02136946e-01 9.14516076e-02 5.66723943e-01 4.22411144e-01 -8.41800034e-01 6.46153092e-01 4.97240335e-01 6.67909503e-01 -8.27698231e-01 4.24156934e-01 2.47748211e-01 -1.48945546e+00 -5.69745302e-01 7.23920912e-02 4.48000133e-02 2.65463799e-01 9.86098871e-02 -5.32673061e-01 5.75819492e-01 8.79429817e-01 1.11286867e+00 -7.74329364e-01 9.51779544e-01 -1.03365302e-01 5.58058202e-01 -3.55720753e-03 2.09889844e-01 1.67329416e-01 -1.33757874e-01 6.19035244e-01 8.87849450e-01 1.24914676e-01 2.80302644e-01 2.79072344e-01 2.99194574e-01 1.77842900e-02 -2.29248285e-01 -1.05146050e+00 1.86021432e-01 1.71042293e-01 1.52074778e+00 -4.84251291e-01 -2.50897765e-01 -8.43676984e-01 6.50358319e-01 3.15488487e-01 2.31074557e-01 -1.24386883e+00 -1.48361519e-01 1.08974445e+00 1.25297299e-02 5.88764250e-01 -3.95517796e-01 -9.59497988e-02 -1.10671222e+00 9.81213152e-02 -1.00121605e+00 5.46499014e-01 -6.15417242e-01 -9.74698544e-01 3.94236773e-01 2.36103415e-01 -1.75250399e+00 -1.28203854e-01 -5.00809371e-01 -3.72736901e-01 4.08553809e-01 -1.48791516e+00 -1.16298366e+00 -4.37114537e-01 8.63985598e-01 9.74872351e-01 -4.79487479e-01 4.44152236e-01 5.38548946e-01 -8.80746663e-01 5.52288353e-01 -2.16442063e-01 1.95564687e-01 7.04652369e-01 -8.02596569e-01 3.43552798e-01 1.06565487e+00 1.60430625e-01 -3.47563140e-02 4.03463960e-01 -6.29664779e-01 -1.51309109e+00 -1.34288275e+00 5.00847399e-01 -9.96222854e-01 7.99093366e-01 -2.71911383e-01 -5.36048651e-01 8.26316655e-01 2.93248415e-01 6.30509630e-02 7.66503513e-01 -4.80299927e-02 -3.21047902e-01 -2.27262914e-01 -9.76972044e-01 5.91995180e-01 1.33029830e+00 -3.84044319e-01 -5.19753337e-01 3.62669975e-01 2.82200873e-01 -4.03358996e-01 -8.38684380e-01 4.19403821e-01 7.60739028e-01 -1.00216234e+00 1.05334640e+00 -8.03421915e-01 3.13814729e-01 -3.51401657e-01 -2.92937368e-01 -1.02254677e+00 -6.80490851e-01 -2.80328900e-01 -2.13798568e-01 6.35950863e-01 4.50741112e-01 -6.28668010e-01 7.42355824e-01 6.36443615e-01 -3.58269036e-01 -7.61891365e-01 -1.34345138e+00 -7.11632669e-01 -2.30590209e-01 -7.54864097e-01 2.30543882e-01 4.41972107e-01 8.24604258e-02 1.88008770e-01 -5.71522236e-01 2.49584571e-01 6.74194574e-01 2.99156234e-02 9.00071383e-01 -9.87578690e-01 -9.34994221e-03 -3.82939219e-01 -8.19555759e-01 -8.25805962e-01 2.33076051e-01 -6.88721955e-01 7.78809935e-02 -1.38991797e+00 3.01923960e-01 1.06123999e-01 -7.19242990e-01 7.03215659e-01 2.42105909e-02 4.09020901e-01 1.78632438e-01 9.72543955e-02 -1.15700161e+00 6.06474757e-01 1.03459573e+00 -2.16758683e-01 2.19238520e-01 1.44776985e-01 -8.67080331e-01 6.98855758e-01 9.34636712e-01 -2.33593181e-01 -4.40426499e-01 -3.30859870e-01 -3.47074240e-01 -7.14806989e-02 4.36856747e-01 -1.23518324e+00 2.46203855e-01 -3.56025666e-01 3.00250113e-01 -6.20384812e-01 7.93837607e-01 -8.83521676e-01 -1.27535626e-01 4.75300282e-01 -2.64949530e-01 2.20738038e-01 3.05502623e-01 7.14042306e-01 -2.23482653e-01 4.13066626e-01 6.79351747e-01 -8.68573599e-03 -1.40100193e+00 4.01792526e-01 -5.35230458e-01 6.19593710e-02 1.52468276e+00 -2.79675752e-01 -5.12252927e-01 -4.10052210e-01 -6.07894123e-01 6.83306694e-01 5.47270961e-02 1.01872587e+00 9.02251124e-01 -1.59355557e+00 -9.38047588e-01 3.43939304e-01 6.16189897e-01 -6.11178160e-01 4.34497297e-01 1.29104888e+00 1.15424626e-01 4.79887038e-01 -3.92512858e-01 -5.67299724e-01 -1.55783081e+00 3.55699509e-01 2.53770858e-01 -7.38798678e-02 -7.74718463e-01 6.59185946e-01 -4.51827906e-02 -1.56052172e-01 1.32791057e-01 5.55402674e-02 -4.78286773e-01 5.76878637e-02 7.36648262e-01 7.27750421e-01 9.71600935e-02 -1.13553786e+00 -6.47006452e-01 4.07182217e-01 -1.08322330e-01 5.80136571e-03 1.20399153e+00 -2.19385058e-01 5.03802299e-01 2.33774871e-01 1.02643859e+00 -2.83649325e-01 -1.42491806e+00 -1.40518755e-01 -9.10311714e-02 -7.24270225e-01 3.51757705e-02 -4.84341353e-01 -1.13694680e+00 5.27020097e-01 8.06329548e-01 4.64641117e-02 1.05899799e+00 1.90887079e-01 8.27128410e-01 4.42962795e-01 3.16730082e-01 -1.33061373e+00 1.62423342e-01 4.05791253e-01 9.97073174e-01 -1.65299511e+00 -1.31531343e-01 -5.71126752e-02 -1.10817683e+00 7.62942672e-01 9.21542823e-01 -3.87606062e-02 5.36014795e-01 7.99036250e-02 1.63348064e-01 -4.72180873e-01 -9.38619554e-01 -5.49717546e-01 2.79180348e-01 4.00298983e-01 -4.87293787e-02 1.21145457e-01 -6.20937347e-02 1.33824408e-01 9.80386958e-02 1.19717561e-01 4.12943900e-01 1.11679530e+00 -6.02988660e-01 -7.21615374e-01 -1.13107309e-01 7.03366756e-01 -1.26881108e-01 3.12737435e-01 -2.36166298e-01 6.93597972e-01 1.55833140e-01 1.30278409e+00 -2.48590764e-03 -8.30047071e-01 8.04103196e-01 2.90494442e-01 1.22545876e-01 -3.06220382e-01 -2.22663611e-01 -1.14267603e-01 3.89366060e-01 -1.30537748e+00 -6.64635599e-01 -1.14721596e+00 -1.02795649e+00 -2.07145318e-01 1.56927973e-01 -2.50380725e-01 2.51221329e-01 9.00462925e-01 7.50123739e-01 5.73278666e-01 8.39446425e-01 -7.26942122e-01 -4.79641557e-01 -7.88768709e-01 -4.89906996e-01 6.14157140e-01 3.73553842e-01 -1.00870764e+00 -2.75262207e-01 9.49421898e-03]

[7.995996475219727, 0.3267695903778076]

5a32c5a8-660e-422f-96f4-b09479d19bdd

lmr-a-large-scale-multi-reference-dataset-for

2303.04970

https://arxiv.org/abs/2303.04970v1

https://arxiv.org/pdf/2303.04970v1.pdf

LMR: A Large-Scale Multi-Reference Dataset for Reference-based Super-Resolution

It is widely agreed that reference-based super-resolution (RefSR) achieves superior results by referring to similar high quality images, compared to single image super-resolution (SISR). Intuitively, the more references, the better performance. However, previous RefSR methods have all focused on single-reference image training, while multiple reference images are often available in testing or practical applications. The root cause of such training-testing mismatch is the absence of publicly available multi-reference SR training datasets, which greatly hinders research efforts on multi-reference super-resolution. To this end, we construct a large-scale, multi-reference super-resolution dataset, named LMR. It contains 112,142 groups of 300x300 training images, which is 10x of the existing largest RefSR dataset. The image size is also much larger. More importantly, each group is equipped with 5 reference images with different similarity levels. Furthermore, we propose a new baseline method for multi-reference super-resolution: MRefSR, including a Multi-Reference Attention Module (MAM) for feature fusion of an arbitrary number of reference images, and a Spatial Aware Filtering Module (SAFM) for the fused feature selection. The proposed MRefSR achieves significant improvements over state-of-the-art approaches on both quantitative and qualitative evaluations. Our code and data would be made available soon.

['Zhaoxiang Zhang', 'Errui Ding', 'Dongliang He', 'Xin Li', 'Lin Zhang']

2023-03-09

['reference-based-super-resolution']

['computer-vision']

[ 3.07044238e-01 -4.99296844e-01 -1.86834008e-01 -4.22071785e-01 -1.40883172e+00 -1.63181558e-01 4.29865718e-01 -5.27498960e-01 -1.51089624e-01 8.79764318e-01 4.25006568e-01 3.62029880e-01 -2.32264951e-01 -6.83457434e-01 -5.10404170e-01 -6.25493646e-01 3.40404242e-01 4.84916531e-02 4.00676221e-01 -4.38669413e-01 3.28803778e-01 4.07044381e-01 -1.77846754e+00 4.74921584e-01 1.15366459e+00 8.56652558e-01 7.66613305e-01 3.43604535e-01 3.00499290e-01 5.34019828e-01 -4.57299203e-01 -2.48536259e-01 2.23541901e-01 -4.27672207e-01 -6.65887952e-01 -7.33392388e-02 8.12856197e-01 -4.82166320e-01 -4.79189634e-01 1.26402569e+00 6.81558311e-01 4.33683664e-01 2.72886902e-01 -6.24454558e-01 -1.33600402e+00 5.71110189e-01 -9.33627903e-01 8.69043410e-01 4.03466761e-01 -5.48352748e-02 8.75477672e-01 -1.24980366e+00 7.26655126e-01 1.29716790e+00 3.28096449e-01 4.82156485e-01 -1.18361032e+00 -8.58978391e-01 -8.48759040e-02 3.19627970e-01 -1.79610372e+00 -6.11725748e-01 5.89726508e-01 -1.03932835e-01 6.00264549e-01 4.14641410e-01 3.31650712e-02 1.04438925e+00 1.46381512e-01 3.03830802e-01 1.36574340e+00 -2.48891022e-02 -1.02676436e-01 4.17215526e-02 2.37333607e-02 2.91971173e-02 1.52918994e-01 2.88209558e-01 -5.98006904e-01 2.52712488e-01 1.32859492e+00 1.63806587e-01 -6.50631428e-01 1.64183825e-01 -1.31509840e+00 5.26595771e-01 5.57958126e-01 5.58061421e-01 -4.32349652e-01 -4.87209082e-01 4.50405851e-02 2.38745824e-01 5.16706944e-01 3.61905396e-01 -1.36312634e-01 2.40871012e-01 -1.14430583e+00 1.86981097e-01 -1.98311463e-01 8.00326586e-01 5.92592835e-01 7.41005912e-02 -3.34910482e-01 1.22848403e+00 -4.45064008e-02 5.87753534e-01 6.62440419e-01 -1.26493633e+00 4.66019779e-01 3.40132117e-01 4.33992952e-01 -1.15973163e+00 -2.01668739e-01 -6.42067790e-01 -1.23163652e+00 9.85415056e-02 2.81173903e-02 3.29305470e-01 -5.67520678e-01 1.65030873e+00 1.48474380e-01 3.55178088e-01 1.80654600e-01 1.60420656e+00 1.25973535e+00 4.73574907e-01 -2.68718272e-01 -3.87409985e-01 1.37936687e+00 -8.74786913e-01 -8.11243713e-01 -1.21856313e-02 -1.57630309e-01 -8.91296029e-01 1.13503456e+00 3.27090323e-01 -1.32550752e+00 -1.10309458e+00 -1.08468366e+00 -3.67054611e-01 -1.22930102e-01 2.60416985e-01 2.62184411e-01 1.80407584e-01 -1.08484912e+00 6.10428810e-01 -4.05761331e-01 -9.73473117e-02 3.91745687e-01 6.82381690e-02 -6.35756612e-01 -5.66020370e-01 -1.36602259e+00 9.95439708e-01 2.52708614e-01 8.28219056e-02 -6.14943922e-01 -8.51668239e-01 -8.28423440e-01 -1.09743856e-01 3.69488567e-01 -5.08424640e-01 8.46488357e-01 -5.43770015e-01 -1.17014098e+00 8.13116372e-01 -3.20957273e-01 -2.79567897e-01 2.66960919e-01 -3.15518796e-01 -1.00654876e+00 2.77502596e-01 2.95420140e-01 4.80241567e-01 8.52001190e-01 -1.54949594e+00 -6.49524331e-01 -4.12426025e-01 7.71580040e-02 3.48370939e-01 8.42986852e-02 3.39408815e-01 -4.97197002e-01 -8.09089839e-01 2.04678714e-01 -3.35152715e-01 -2.42008477e-01 -4.75364566e-01 -2.25191072e-01 1.43664077e-01 7.24084854e-01 -8.10709417e-01 1.27986562e+00 -2.24505591e+00 2.07079276e-01 -3.49241436e-01 5.11723936e-01 2.93443233e-01 -4.99808878e-01 -1.26533613e-01 -3.77549231e-01 1.56982824e-01 8.10041651e-02 -1.11776046e-01 -4.85644013e-01 -1.25347048e-01 -4.00841653e-01 4.54587966e-01 2.36375332e-01 8.52276921e-01 -8.58596146e-01 -6.75893009e-01 3.52966815e-01 8.94148111e-01 -1.74933508e-01 8.71653557e-02 3.00077200e-01 9.66237664e-01 -5.07638812e-01 6.19908154e-01 1.10424757e+00 -5.60891509e-01 -2.41666615e-01 -7.79111326e-01 -3.61159056e-01 -2.09632702e-02 -1.30927873e+00 1.84039915e+00 -4.96590495e-01 3.62209827e-01 -2.36747991e-02 -4.32431906e-01 1.09620595e+00 1.72694087e-01 4.93461221e-01 -1.14341819e+00 -7.58426636e-02 1.38004586e-01 -1.47438139e-01 -1.59695387e-01 9.38137591e-01 6.23739101e-02 2.49951199e-01 1.21272631e-01 -5.39131686e-02 2.96356559e-01 2.95828879e-01 1.72509775e-01 9.44427252e-01 1.30150199e-01 2.41874143e-01 -1.33451782e-02 7.21515715e-01 -2.72820085e-01 7.13998735e-01 6.16305113e-01 -8.90072361e-02 1.25735962e+00 -2.73598135e-01 -2.57931620e-01 -1.16940677e+00 -1.20084238e+00 -4.70097214e-01 8.08914483e-01 5.69815934e-01 -2.88354874e-01 -4.69804019e-01 -9.24007893e-02 -2.94082135e-01 4.88173127e-01 -6.69101417e-01 1.16084576e-01 -7.25205541e-01 -7.80098677e-01 -2.83755194e-02 5.07161975e-01 8.89020383e-01 -1.00778174e+00 -4.48230386e-01 2.72727739e-02 -5.95744312e-01 -1.34761477e+00 -7.44733512e-01 -5.00055730e-01 -7.93212533e-01 -9.89808202e-01 -9.18525636e-01 -3.58769685e-01 4.48279172e-01 9.35961366e-01 1.33377230e+00 5.07344007e-02 -2.30458975e-01 -2.50618346e-02 -4.88444328e-01 3.19662511e-01 -1.71900883e-01 -7.71954507e-02 7.73259997e-02 7.55671784e-02 1.68451458e-01 -5.88365734e-01 -8.40039015e-01 5.54116189e-01 -8.19364607e-01 2.14322597e-01 9.29974675e-01 9.13453996e-01 1.19786429e+00 1.74527451e-01 8.02130699e-01 -5.47361851e-01 4.33057725e-01 -3.46518576e-01 -5.17076552e-01 2.90899634e-01 -5.22149861e-01 -1.77759573e-01 6.34090066e-01 -4.00255203e-01 -1.29758799e+00 -3.60976547e-01 -5.01252851e-03 -7.90263593e-01 -2.49894008e-01 2.76378214e-01 -2.74935067e-01 -1.31872103e-01 5.61843336e-01 3.64293933e-01 -2.11236477e-01 -7.73763776e-01 4.33347940e-01 7.45056033e-01 1.05253232e+00 -2.83895284e-01 9.84051585e-01 3.48887593e-01 -1.68018669e-01 -6.86494648e-01 -1.11792433e+00 -4.25002337e-01 -6.62678480e-01 1.01553977e-01 8.51013303e-01 -1.33606374e+00 -3.44573140e-01 1.48832247e-01 -8.79628360e-01 -5.21399124e-05 -1.42713979e-01 5.39135158e-01 -4.32905704e-01 2.60201216e-01 -5.87958932e-01 -6.12181425e-01 -4.62449998e-01 -1.18374038e+00 1.16915846e+00 6.16192937e-01 1.36153266e-01 -5.18728197e-01 -1.66426420e-01 6.72172129e-01 9.03734267e-01 1.07221819e-01 2.68885463e-01 -3.41510445e-01 -7.37137079e-01 1.57168910e-01 -8.28882992e-01 2.27345735e-01 2.97610223e-01 -3.18325073e-01 -7.85534084e-01 -4.65002924e-01 7.14578107e-02 -1.46324158e-01 8.30449045e-01 4.65737313e-01 1.16471481e+00 1.61493085e-02 -1.08833820e-01 6.55482113e-01 1.67665505e+00 4.31669876e-02 9.73729193e-01 4.03651506e-01 7.22684801e-01 2.91461289e-01 1.18950760e+00 3.61640036e-01 5.58375418e-01 1.09295094e+00 2.43822992e-01 -2.91725367e-01 -5.45269132e-01 1.30751744e-01 1.92680702e-01 7.64404714e-01 -4.75578815e-01 1.21209607e-01 -3.92835379e-01 5.24443746e-01 -1.54724622e+00 -1.36977220e+00 -8.37553963e-02 2.29426956e+00 8.68571222e-01 -2.61100262e-01 -1.79061256e-02 -3.40983495e-02 1.00364852e+00 2.44859815e-01 -7.33203471e-01 3.76657724e-01 -4.91933465e-01 1.92661867e-01 2.57494271e-01 3.21618348e-01 -1.15935099e+00 7.21810997e-01 5.80023623e+00 1.11675894e+00 -1.09158528e+00 4.26412523e-01 6.64999068e-01 -2.65836358e-01 -2.03042507e-01 -4.13761854e-01 -9.67472017e-01 5.89137316e-01 8.78419161e-01 -3.43084693e-01 7.19926476e-01 4.48350221e-01 4.23342794e-01 -9.54240859e-02 -6.97595298e-01 1.40899110e+00 2.07183838e-01 -1.28579021e+00 1.07448690e-01 -2.03466535e-01 7.81508386e-01 1.56503499e-01 3.10679108e-01 2.12274179e-01 1.96859166e-01 -1.29251730e+00 4.33511525e-01 7.67824888e-01 1.35654926e+00 -7.85061955e-01 7.90873587e-01 2.71368958e-02 -1.51174617e+00 -6.24063872e-02 -5.93554080e-01 4.42654729e-01 1.33781955e-01 5.31567335e-01 1.94120750e-01 1.20666945e+00 9.91652906e-01 8.53168309e-01 -9.04235601e-01 8.86626661e-01 1.37079388e-01 1.04239769e-01 8.14959481e-02 8.84853601e-01 -2.98448354e-01 -2.51335442e-01 5.60349047e-01 8.29407632e-01 5.44216275e-01 5.22099257e-01 1.05649836e-01 1.08552253e+00 -7.66600147e-02 3.56659256e-02 -2.36918300e-01 2.63229817e-01 7.73181975e-01 1.60109174e+00 -4.39846128e-01 -3.03295940e-01 -7.36857355e-01 8.85014296e-01 7.65938759e-02 5.21322131e-01 -8.41329455e-01 -3.17098439e-01 6.54697180e-01 1.69323981e-01 3.53652179e-01 1.02110170e-01 -2.08092973e-01 -1.45615602e+00 6.28161132e-02 -1.14337361e+00 3.63072246e-01 -1.16651750e+00 -1.51010263e+00 1.00856841e+00 -5.56913763e-02 -1.45725715e+00 -9.75325629e-02 -2.73909792e-02 -2.60311216e-01 1.32862890e+00 -1.77713478e+00 -1.08572185e+00 -7.88018823e-01 6.42552674e-01 7.89585829e-01 -1.58002362e-01 5.96995711e-01 6.11851990e-01 -6.77419782e-01 6.00104570e-01 6.17260970e-02 1.04222082e-01 1.02338123e+00 -9.45048153e-01 2.61223465e-01 1.08419776e+00 4.26208712e-02 7.71355391e-01 6.81060433e-01 -5.09630740e-01 -1.28995955e+00 -1.28295493e+00 4.06186700e-01 -5.76340497e-01 3.76602530e-01 9.27521512e-02 -1.44507837e+00 3.89345616e-01 3.20120119e-02 5.07270157e-01 3.39071065e-01 -1.84848011e-01 -3.86327118e-01 -2.68597990e-01 -1.33878744e+00 5.15827119e-01 1.17215753e+00 -4.63785857e-01 -5.58946013e-01 -7.06262961e-02 9.71062183e-01 -6.57544136e-01 -1.51110423e+00 6.69972181e-01 3.24287683e-01 -1.14451456e+00 1.39021170e+00 3.42442431e-02 6.92526817e-01 -6.70553207e-01 -5.50198317e-01 -1.14804208e+00 -6.68058991e-01 -1.12513900e-01 -1.71819761e-01 1.34304845e+00 -6.49512485e-02 -5.79864979e-01 2.18665168e-01 3.79552960e-01 -1.70356229e-01 -7.88977444e-01 -7.77169585e-01 -7.52495646e-01 -5.25491983e-02 2.25946218e-01 8.98906767e-01 9.65531051e-01 -7.15502799e-01 4.30070877e-01 -4.83537853e-01 4.93150115e-01 9.58426476e-01 5.15883982e-01 5.06661773e-01 -1.21530652e+00 -1.23530410e-01 -3.56082618e-01 -2.06256926e-01 -8.22825611e-01 -6.83556944e-02 -6.45865023e-01 -1.93072423e-01 -1.61142313e+00 6.93963766e-01 -3.02934706e-01 -5.95130861e-01 2.16851935e-01 -3.85338545e-01 7.25402534e-01 6.37421906e-02 4.91830617e-01 -8.87642741e-01 5.31552553e-01 1.70592034e+00 1.87127814e-01 -1.25752077e-01 -3.38296026e-01 -1.02311635e+00 4.01894391e-01 5.19249499e-01 -3.40496227e-02 -2.87875712e-01 -2.59213626e-01 -2.87964672e-01 3.90516162e-01 4.92098480e-01 -1.02928376e+00 1.67587716e-02 -2.13037863e-01 7.59268403e-01 -9.57649827e-01 3.74756128e-01 -4.33360636e-01 5.03097951e-01 -2.57457376e-01 -3.65045011e-01 4.82885204e-02 -1.48663782e-02 3.84677082e-01 -4.92229968e-01 9.73995253e-02 1.09917307e+00 -2.15124577e-01 -9.14852262e-01 5.49036205e-01 4.49029922e-01 1.08797103e-01 6.57607317e-01 -2.21379042e-01 -6.97909057e-01 -8.60232338e-02 -5.64203799e-01 5.97191416e-02 8.01866233e-01 6.36328876e-01 9.59247172e-01 -1.48106766e+00 -1.14457357e+00 8.12317524e-03 6.02314286e-02 9.65447873e-02 8.67503524e-01 8.97217870e-01 1.93887502e-01 2.50188768e-01 -3.99933994e-01 -4.20237273e-01 -1.38490772e+00 8.21304679e-01 1.52693972e-01 -2.28660032e-01 -9.47194159e-01 3.74062419e-01 3.39023560e-01 2.30730679e-02 -2.87015945e-01 5.66179901e-02 -6.50578797e-01 -1.41291201e-01 1.28474772e+00 4.81776834e-01 5.62196830e-03 -1.18334174e+00 -2.30223566e-01 9.08388436e-01 -2.95588881e-01 4.41843197e-02 1.41088474e+00 -6.51738465e-01 -9.02946368e-02 3.18598270e-01 8.45316410e-01 3.89168933e-02 -1.15965128e+00 -5.51183462e-01 -3.32094133e-01 -9.60893810e-01 2.26904333e-01 -8.07923436e-01 -1.34354663e+00 5.95253885e-01 8.90158415e-01 -2.14657947e-01 1.57640553e+00 2.00766586e-02 6.90961897e-01 -1.32813111e-01 8.20744693e-01 -6.89790547e-01 9.05351788e-02 2.29487762e-01 1.23229873e+00 -1.55420542e+00 2.07424775e-01 -4.89395261e-01 -6.66705906e-01 7.31373906e-01 7.72135317e-01 -1.86993197e-01 1.75370038e-01 1.92426577e-01 -1.58813018e-02 2.99142245e-02 -8.19884837e-01 -3.28677118e-01 4.42444116e-01 6.66008711e-01 5.04623413e-01 -2.13570982e-01 -2.02312902e-01 9.95539486e-01 -1.12802878e-01 2.02909112e-01 4.96044397e-01 1.64341122e-01 -3.08545440e-01 -8.70904267e-01 -5.93544126e-01 4.69646096e-01 -6.43350005e-01 -2.27210939e-01 1.98382854e-01 5.75721443e-01 6.96713850e-02 1.08415496e+00 4.42494340e-02 -5.52750587e-01 3.49025309e-01 -5.53760707e-01 5.18766582e-01 -3.76092225e-01 -2.26302043e-01 5.58659993e-02 -2.13744536e-01 -9.41242814e-01 -7.36183405e-01 -6.25127792e-01 -1.10006917e+00 -2.74422914e-01 -3.49527359e-01 -5.05483262e-02 1.41790599e-01 6.43987536e-01 5.56446970e-01 7.33009517e-01 6.70680106e-01 -1.12453282e+00 -4.12229031e-01 -1.10162485e+00 -7.53409863e-01 5.61233759e-01 2.76898772e-01 -9.86266077e-01 -2.25063309e-01 2.22772937e-02]

[10.9736328125, -2.0942904949188232]

739dcca1-2980-4798-ac14-a67d68d91fee

on-the-liveliness-of-artificial-life

2302.10196

https://arxiv.org/abs/2302.10196v1

https://arxiv.org/pdf/2302.10196v1.pdf

On the Liveliness of Artificial Life

There has been on-going philosophical debate on whether artificial life models, also known as digital organisms, are truly alive. The main difficulty appears to be finding an encompassing and definite definition of life. By examining similarities and differences in recent definitions of life, we define life as "any system with a boundary to confine the system within a definite volume and protect the system from external effects, consisting of a program that is capable of improvisation, able to react and adapt to the environment, able to regenerate parts of it-self or its entirety, with energy system comprises of non-interference sets of secluded reactions for self-sustenance, is considered alive or a living system. Any incomplete system containing a program and can be re-assembled into a living system; thereby, converting the reassembled system for the purpose of the incomplete system, are also considered alive." Using this definition, we argue that digital organisms may not be the boundary case of life even though some digital organisms are not considered alive; thereby, taking the view that some form of digital organisms can be considered alive. In addition, we present an experimental framework based on continuity of the overall system and potential discontinuity of elements within the system for testing future definitions of life.

['Maurice HT Ling', 'Yong Zher Koh']

2023-02-19

['artificial-life']

['miscellaneous']

[ 3.23232889e-01 4.70979512e-01 2.87490994e-01 2.94395655e-01 5.77805579e-01 -9.91700590e-01 8.44045699e-01 8.21718350e-02 -9.06370431e-02 8.20228815e-01 -1.13819288e-02 -4.20082122e-01 -5.72431833e-02 -1.12019241e+00 -5.35755515e-01 -1.02832770e+00 -8.53975117e-02 2.31492504e-01 2.39216745e-01 -4.31259930e-01 -1.76387951e-02 5.12408197e-01 -1.92470467e+00 -5.19876480e-01 7.47928858e-01 3.48268449e-01 1.54081091e-01 7.51379132e-01 5.08955866e-02 6.77375674e-01 -8.11600089e-01 -8.85516927e-02 6.17872812e-02 -1.07860124e+00 -8.62224936e-01 1.49263486e-01 -3.63694370e-01 1.39543293e-02 -1.22008584e-01 7.42770970e-01 1.21857993e-01 -1.55108839e-01 5.79958558e-01 -1.13876677e+00 -7.99867153e-01 3.52874547e-01 4.17241484e-01 -3.66868913e-01 5.91460466e-01 4.41628039e-01 4.22015697e-01 -2.87088871e-01 6.43218577e-01 9.53463376e-01 4.66735005e-01 8.58904243e-01 -1.34756577e+00 -2.67797261e-02 -2.76884317e-01 -4.99766111e-01 -1.29815626e+00 -6.74112439e-01 2.25774080e-01 -4.63548303e-01 1.03643072e+00 9.28527296e-01 1.25275350e+00 7.67664373e-01 1.05110753e+00 -1.44091877e-03 8.29829514e-01 -5.52864730e-01 8.11767876e-01 1.12371176e-01 2.69331098e-01 4.46812481e-01 1.04853857e+00 1.81942493e-01 -1.17364839e-01 -1.31137550e-01 7.06610322e-01 -4.50094938e-01 -3.92502517e-01 -2.72496670e-01 -1.23720372e+00 -1.06518179e-01 -2.98296630e-01 8.23154747e-01 -2.79654622e-01 1.99721351e-01 1.74399137e-01 4.41012174e-01 2.32862458e-02 7.20880926e-01 -2.57522196e-01 -2.58583367e-01 -2.77278513e-01 8.74718204e-02 1.43738866e+00 4.63244945e-01 5.88274240e-01 5.64535595e-02 5.62702000e-01 4.99101095e-02 3.16541106e-01 3.01296115e-01 6.00037456e-01 -9.77861702e-01 -5.88824987e-01 1.21881008e+00 1.68369412e-01 -4.09770042e-01 -2.69757599e-01 -3.71485919e-01 -7.44593143e-01 5.80803037e-01 2.50452071e-01 -1.87631622e-02 -4.35519248e-01 1.78622353e+00 4.41906422e-01 -4.22815271e-02 6.35073066e-01 4.98847127e-01 7.29238749e-01 8.90473664e-01 -1.11235067e-01 -8.02507997e-01 1.09086168e+00 -2.77013659e-01 -7.90755510e-01 4.76967134e-02 4.54196513e-01 -2.54014313e-01 7.34278560e-01 4.23848987e-01 -1.45168710e+00 -3.42882007e-01 -1.70562351e+00 3.02878559e-01 -5.46374202e-01 -7.98478484e-01 6.11442268e-01 8.74247611e-01 -1.27089083e+00 5.83359361e-01 -6.82306886e-01 -8.32024395e-01 -3.56388837e-01 2.44069502e-01 -4.43821341e-01 5.35351992e-01 -1.04310811e+00 1.04160297e+00 4.73630220e-01 -1.06176540e-01 -1.12040651e+00 -1.21535659e-01 -6.11410975e-01 6.17543384e-02 4.18437552e-03 -1.09540784e+00 7.05374300e-01 -1.17598748e+00 -1.29207993e+00 1.10101402e+00 9.57692787e-02 -2.95189261e-01 3.68747264e-01 4.66993749e-01 -5.88823199e-01 -1.86835393e-01 -1.14331834e-01 1.58692583e-01 1.82380050e-01 -1.61453962e+00 -2.78336078e-01 -5.06053746e-01 4.02003229e-01 1.94993705e-01 -2.44002372e-01 -9.65871960e-02 -6.10232167e-02 -1.14957482e-01 4.39339697e-01 -9.36882675e-01 -9.98438150e-03 -1.17411189e-01 -1.42911628e-01 -2.38954708e-01 7.67335951e-01 -2.35512212e-01 1.26866400e+00 -2.15714574e+00 3.59535664e-01 -2.02208385e-01 2.40634650e-01 1.70540616e-01 1.23210035e-01 8.73638630e-01 -4.39275689e-02 5.75767815e-01 -4.93501633e-01 6.83960244e-02 -5.95345125e-02 5.35917878e-01 3.81639116e-02 5.29164732e-01 4.87814657e-03 5.67582846e-01 -8.62223923e-01 -3.87974292e-01 5.79998977e-02 5.51208138e-01 9.18368101e-02 1.98101625e-02 -1.78534850e-01 4.49214011e-01 -2.60593683e-01 7.09873080e-01 2.68808305e-01 1.17986307e-01 3.09385717e-01 4.10940975e-01 -5.88767231e-01 -9.51687172e-02 -1.16346192e+00 1.05641663e+00 -1.44359529e-01 3.81015927e-01 2.78622597e-01 -8.53386104e-01 8.97809267e-01 7.51183510e-01 4.34981942e-01 -4.86847848e-01 3.33782285e-01 2.82972634e-01 2.69397199e-01 -5.32770216e-01 3.37303847e-01 -2.96479017e-01 -1.43171415e-01 6.55587733e-01 -1.90484583e-01 -4.68504220e-01 3.05271357e-01 5.30260205e-02 1.27866626e+00 4.08593833e-01 5.91381013e-01 -6.11634135e-01 7.57223070e-01 -2.47939855e-01 7.62018144e-01 3.73529971e-01 -4.80434686e-01 1.06269658e-01 4.53928500e-01 -3.42678010e-01 -1.14038575e+00 -1.37622941e+00 -3.04815263e-01 4.97453570e-01 7.25752592e-01 -2.12395996e-01 -9.86515343e-01 -4.99394275e-02 -2.78812766e-01 9.37130272e-01 -6.75744534e-01 -3.10893059e-01 -3.13507259e-01 -7.90307581e-01 4.38274354e-01 -2.69234449e-01 4.78758514e-01 -1.17245007e+00 -1.20189261e+00 4.18536067e-01 2.07487017e-01 -2.41229162e-01 3.04269612e-01 3.61483574e-01 -7.66721010e-01 -9.83941913e-01 -1.91011187e-02 -6.53780699e-01 6.68633521e-01 2.24040613e-01 9.78618145e-01 8.27611864e-01 -1.13278806e-01 5.16645908e-01 -2.89074272e-01 -1.86819389e-01 -1.25028908e+00 -5.96760869e-01 3.78989220e-01 -2.54582942e-01 -2.24736974e-01 -8.80450964e-01 -4.17163640e-01 1.75598890e-01 -1.36422968e+00 2.17487276e-01 -1.75082013e-02 3.65008622e-01 3.52736741e-01 3.46697301e-01 8.52336824e-01 -2.28115425e-01 3.62987787e-01 -6.28299236e-01 -1.26633584e-01 5.32383502e-01 -6.14687145e-01 -7.05247000e-02 6.80775225e-01 -3.59021038e-01 -1.01812840e+00 -6.48928359e-02 2.51327790e-02 6.73875511e-01 -4.83460166e-02 1.90074563e-01 -9.07484472e-01 1.00705467e-01 5.34388959e-01 7.14632392e-01 2.08011121e-01 -7.98726305e-02 2.88654894e-01 7.36324131e-01 6.61462188e-01 -5.89598835e-01 4.27046776e-01 6.38520241e-01 2.02946424e-01 -9.63384390e-01 2.16105327e-01 1.94171369e-01 -6.58555269e-01 -7.03051805e-01 7.71955192e-01 -3.56590092e-01 -3.85029286e-01 5.62890828e-01 -8.58694375e-01 -1.09720230e-01 -7.83321738e-01 -4.92706373e-02 -7.55130231e-01 4.43637758e-01 -3.81736845e-01 -1.10737443e+00 -4.61722732e-01 -6.06896460e-01 3.47698271e-01 4.28454190e-01 -5.72268009e-01 -1.02349114e+00 5.37021995e-01 1.69327147e-02 2.63330400e-01 6.64705932e-01 8.62854242e-01 -6.16159439e-01 -3.05953801e-01 -4.62614834e-01 7.10740328e-01 3.26517403e-01 4.54179704e-01 6.68700099e-01 -5.50113440e-01 -4.20812219e-01 5.66328287e-01 9.71185341e-02 1.10828176e-01 -7.30102584e-02 1.21323898e-01 -4.76556301e-01 -5.30489683e-01 2.43292958e-01 1.64015710e+00 9.19893503e-01 1.05446446e+00 4.13100183e-01 1.64377447e-02 9.59561586e-01 2.93294340e-01 3.45709324e-01 2.62254417e-01 1.06653862e-01 5.14515460e-01 1.07501164e-01 -6.41159266e-02 9.17599723e-02 5.01042068e-01 9.31809962e-01 -1.14108659e-01 -5.54061353e-01 -8.13965261e-01 5.59905529e-01 -1.59715366e+00 -1.12939143e+00 -3.69346648e-01 2.29782557e+00 7.91953146e-01 1.62735879e-01 -2.91637448e-03 4.96371955e-01 7.70125806e-01 -3.89559299e-01 -6.97739005e-01 -6.45266533e-01 -3.79682094e-01 -6.59590140e-02 -2.79206168e-02 2.72350073e-01 -4.70344543e-01 5.43346763e-01 7.48569441e+00 1.91641882e-01 -8.75938118e-01 6.55938536e-02 3.77244771e-01 1.15600452e-01 -6.05405688e-01 6.36568666e-01 -2.10955456e-01 5.87807834e-01 1.10458541e+00 -8.21363688e-01 3.14470857e-01 3.27055991e-01 4.07083213e-01 -5.97498119e-01 -1.21156693e+00 1.85099810e-01 -8.17391425e-02 -9.61502910e-01 -1.34207774e-02 1.66928396e-01 5.41627765e-01 -6.44163013e-01 -6.24757409e-01 -9.22500268e-02 3.97554547e-01 -7.15016186e-01 1.13859475e+00 7.92513013e-01 5.07275581e-01 -4.73231018e-01 4.47196543e-01 7.89765239e-01 -9.92747545e-01 8.53290558e-02 4.82626669e-02 -6.63810253e-01 1.16394132e-01 3.97489220e-01 -5.05997278e-02 5.81241012e-01 4.21517223e-01 -2.67968252e-02 -4.92228359e-01 7.87737608e-01 1.40901297e-01 3.74199480e-01 -1.42551467e-01 -1.68643773e-01 -3.50121737e-01 -5.24895847e-01 8.94844532e-01 6.68383360e-01 4.01005596e-01 5.16890526e-01 -3.02317530e-01 1.08864760e+00 6.53763890e-01 -1.19885452e-01 -6.70416832e-01 -1.87285304e-01 5.67766666e-01 1.06172311e+00 -9.36546445e-01 -6.28050447e-01 -1.22256666e-01 7.07676888e-01 -5.48394144e-01 4.95682023e-02 -7.56194949e-01 -3.61475438e-01 6.06916666e-01 3.53333205e-01 -5.99736512e-01 -3.45866114e-01 -4.80797797e-01 -9.49344933e-01 -2.29531527e-01 -5.90391159e-01 -1.81307152e-01 -8.68240833e-01 -8.97745609e-01 3.67701828e-01 -2.54065096e-01 -9.10909593e-01 3.13393609e-03 -9.71312672e-02 -7.34900653e-01 5.99619031e-01 -1.86200634e-01 -1.05368257e+00 -2.79670000e-01 2.51103550e-01 1.04904324e-01 6.03139624e-02 9.46725428e-01 -3.75269622e-01 -7.26668119e-01 -5.57916239e-02 4.22379166e-01 -5.45326591e-01 7.34943748e-02 -8.66412878e-01 2.54079461e-01 1.19497180e+00 -4.79507804e-01 8.49120200e-01 1.12460518e+00 -9.44150925e-01 -1.74096489e+00 -7.74708569e-01 1.02157784e+00 -3.66778523e-01 3.37504178e-01 -4.67154145e-01 -8.34865093e-01 4.98168707e-01 3.11470687e-01 -7.35059381e-01 6.89592481e-01 -5.03432572e-01 1.57519996e-01 -4.01184745e-02 -1.56200957e+00 9.47388172e-01 1.21540117e+00 -5.95939718e-02 -8.82932305e-01 2.72731222e-02 9.47914302e-01 3.05627972e-01 -8.20095897e-01 3.34263146e-01 9.25390840e-01 -1.28527784e+00 6.86016798e-01 3.18707228e-02 7.84945954e-03 -6.85608089e-01 -4.55198400e-02 -6.92975163e-01 -5.07757008e-01 -6.30859554e-01 1.21749796e-01 1.58249068e+00 -6.40247092e-02 -1.34758139e+00 1.64918974e-01 9.50359643e-01 -2.48922363e-01 -4.28966671e-01 -1.21573794e+00 -1.10937333e+00 3.54044288e-01 7.81347156e-02 7.76762605e-01 1.13314044e+00 8.23027730e-01 2.02806100e-01 2.77073652e-01 -1.60251394e-01 5.55990517e-01 -1.07990518e-01 6.03253782e-01 -1.34123743e+00 -2.43355766e-01 -3.71308476e-01 -3.84651512e-01 -3.10880452e-01 -1.50215060e-01 -6.17727578e-01 9.44053456e-02 -1.85869384e+00 3.23762774e-01 -2.77485341e-01 9.91706699e-02 2.88103461e-01 4.59115118e-01 4.55827117e-02 -1.01559656e-02 6.20572746e-01 -2.73002535e-01 1.74563795e-01 9.59024131e-01 2.83878863e-01 -1.93656594e-01 -5.09050369e-01 -8.29316676e-01 5.65842569e-01 7.05480814e-01 -2.75816917e-01 -4.39063966e-01 4.32878762e-01 4.76449668e-01 4.76553917e-01 3.15957665e-01 -1.30849230e+00 -6.65136427e-02 -4.66419786e-01 1.43860672e-02 -2.23414347e-01 4.65890020e-02 -9.03568029e-01 1.45304132e+00 1.27293515e+00 1.10380128e-01 -2.15727016e-01 -4.38593663e-02 1.93693295e-01 2.78037518e-01 -5.21424770e-01 7.64949739e-01 -3.33737105e-01 -2.79937506e-01 -3.27511191e-01 -1.14906704e+00 -6.05529010e-01 1.80802858e+00 -1.18776941e+00 -5.54729700e-01 -3.00446246e-02 -6.98557079e-01 6.74536079e-02 1.56155205e+00 1.86566591e-01 1.78215533e-01 -9.08923686e-01 -3.86193752e-01 1.11194611e-01 -1.65923208e-01 -3.17651898e-01 -1.11526981e-01 2.08754122e-01 -9.69507754e-01 2.01036066e-01 -5.14868319e-01 -1.74930945e-01 -1.13285708e+00 7.06189752e-01 6.77583218e-01 3.12823921e-01 -3.66358817e-01 3.69197577e-01 4.34363246e-01 -1.24386832e-01 -2.10455030e-01 -1.22207582e-01 -1.37355939e-01 -1.66386366e-01 3.09222907e-01 4.27846342e-01 -2.29394361e-01 -7.36242414e-01 -4.83026266e-01 3.79947454e-01 9.13968027e-01 -2.65321493e-01 1.11223817e+00 -5.74873209e-01 -9.41564620e-01 9.32049990e-01 5.33388138e-01 -8.64933655e-02 -9.22790051e-01 6.12946272e-01 -3.37667495e-01 -3.17064598e-02 -5.01837313e-01 -8.69845510e-01 -2.96171308e-01 1.49626762e-01 3.34520370e-01 9.56317246e-01 1.29432499e+00 -6.52230233e-02 3.93905491e-01 -1.00435345e-02 7.34344959e-01 -8.75191629e-01 -8.57248232e-02 1.79464832e-01 9.56448078e-01 -2.79668123e-01 -1.01395674e-01 -2.46235147e-01 -1.04454644e-01 1.01033950e+00 4.94772732e-01 -2.25468785e-01 4.77118313e-01 6.19999588e-01 -4.97640997e-01 -5.54896295e-02 -1.13230968e+00 9.79838222e-02 -4.26691353e-01 7.33972371e-01 4.63028908e-01 4.09218594e-02 -1.13250268e+00 6.46697640e-01 -2.42110342e-01 8.34372826e-03 1.04635501e+00 1.15759861e+00 -9.63264883e-01 -9.59904075e-01 -6.24688745e-01 -8.27115253e-02 -2.72911787e-01 4.20241773e-01 -8.48077416e-01 8.01805675e-01 8.52227032e-01 1.16238225e+00 3.32118124e-01 -2.90690541e-01 1.63659453e-01 1.22206353e-01 6.61897480e-01 -4.05282408e-01 -7.85427392e-01 -3.08729053e-01 2.63843507e-01 -4.17087525e-02 -4.15772408e-01 -8.25315595e-01 -1.67645216e+00 -7.94856906e-01 -3.52421671e-01 1.50632516e-01 6.99797750e-01 8.06488037e-01 1.48090109e-01 4.57794666e-01 4.57912147e-01 -4.17498112e-01 -1.98013008e-01 -5.75707972e-01 -7.36946046e-01 1.65567592e-01 1.92597568e-01 -2.42985949e-01 -7.99212337e-01 5.49826622e-01]

[5.592956066131592, 4.153954982757568]

fd3f2294-7bc4-499d-97bf-b51e4f526779

learning-how-to-be-robust-deep-polynomial

1804.06504

http://arxiv.org/abs/1804.06504v2

http://arxiv.org/pdf/1804.06504v2.pdf

Learning how to be robust: Deep polynomial regression

Polynomial regression is a recurrent problem with a large number of applications. In computer vision it often appears in motion analysis. Whatever the application, standard methods for regression of polynomial models tend to deliver biased results when the input data is heavily contaminated by outliers. Moreover, the problem is even harder when outliers have strong structure. Departing from problem-tailored heuristics for robust estimation of parametric models, we explore deep convolutional neural networks. Our work aims to find a generic approach for training deep regression models without the explicit need of supervised annotation. We bypass the need for a tailored loss function on the regression parameters by attaching to our model a differentiable hard-wired decoder corresponding to the polynomial operation at hand. We demonstrate the value of our findings by comparing with standard robust regression methods. Furthermore, we demonstrate how to use such models for a real computer vision problem, i.e., video stabilization. The qualitative and quantitative experiments show that neural networks are able to learn robustness for general polynomial regression, with results that well overpass scores of traditional robust estimation methods.

['Patrick Perez', 'Patrick Bouthemy', 'Juan-Manuel Perez-Rua', 'Tomas Crivelli']

2018-04-17

['video-stabilization']

['computer-vision']

[ 1.38659984e-01 9.73260626e-02 -9.67966691e-02 -1.63558602e-01 -1.12106204e+00 -4.36775148e-01 4.37308639e-01 -5.83905727e-02 -3.42741400e-01 5.30667067e-01 3.67783941e-02 -1.65020928e-01 -2.88801156e-02 -7.06382543e-02 -1.26447284e+00 -8.49093676e-01 1.40382927e-02 2.74728924e-01 1.15003802e-01 -3.80750686e-01 3.55027020e-01 6.25713408e-01 -1.15860987e+00 -1.40125453e-01 6.04144812e-01 9.04867530e-01 -1.41467988e-01 7.65857816e-01 6.08577371e-01 7.52622604e-01 -3.41660053e-01 -3.26775283e-01 3.35908085e-01 -1.66452900e-01 -5.67082167e-01 1.71070114e-01 7.40567029e-01 -1.71206817e-01 -4.64153022e-01 1.04316521e+00 4.89759237e-01 2.91691154e-01 7.92866886e-01 -1.20198464e+00 -4.93579745e-01 4.01258379e-01 -6.37522280e-01 3.92395295e-02 1.88906953e-01 3.69955122e-01 9.15259123e-01 -9.07043397e-01 6.63478315e-01 1.13608229e+00 1.07277060e+00 5.23115158e-01 -1.52154756e+00 -1.19344220e-01 6.55252934e-02 -8.61113425e-04 -1.13651133e+00 -7.55273163e-01 8.15949738e-01 -7.08771348e-01 5.87238371e-01 5.85538372e-02 1.18722126e-01 1.39998472e+00 1.43219903e-01 6.98033750e-01 6.71872735e-01 -2.35909939e-01 1.68596193e-01 -3.78849544e-02 6.37706295e-02 7.05640137e-01 2.35308364e-01 1.65124699e-01 -2.01328188e-01 4.08107936e-02 8.75547111e-01 -1.91265598e-01 -4.69495714e-01 -8.04774821e-01 -1.35370421e+00 7.38024771e-01 2.65312314e-01 3.87200937e-02 -2.49759525e-01 8.38595092e-01 6.84202552e-01 4.12059098e-01 4.54107583e-01 6.76375151e-01 -5.73609054e-01 -9.49893426e-03 -1.28988540e+00 4.11343962e-01 6.29300356e-01 1.05564117e+00 5.31320632e-01 2.00794950e-01 -2.50758141e-01 6.52619183e-01 1.90004289e-01 2.05203861e-01 2.83901751e-01 -1.33763921e+00 3.96810114e-01 1.71979427e-01 3.41286868e-01 -1.17714965e+00 -4.02707428e-01 -4.67530608e-01 -8.92263114e-01 4.74595875e-01 9.06292140e-01 -1.25830024e-01 -7.04396486e-01 1.80545330e+00 1.35578230e-01 3.20946276e-01 -6.19018115e-02 9.17575657e-01 2.49972463e-01 3.90069008e-01 -3.62901896e-01 -2.36602932e-01 8.21726322e-01 -1.10666168e+00 -5.49963713e-01 -4.29651998e-02 6.20306134e-01 -6.61642432e-01 1.25958455e+00 6.46529913e-01 -1.04326987e+00 -4.05512571e-01 -1.02484512e+00 -3.52770656e-01 5.65047413e-02 2.64660537e-01 3.06513071e-01 2.70968854e-01 -1.01162672e+00 9.83501494e-01 -1.03838193e+00 -3.09472114e-01 3.36682469e-01 5.32996595e-01 -4.99674857e-01 2.17152089e-01 -6.57934844e-01 1.13316917e+00 -3.37518603e-02 5.27642250e-01 -1.03446209e+00 -6.20077491e-01 -9.07502294e-01 -1.35396570e-01 4.97556478e-01 -7.36734748e-01 1.34562373e+00 -1.29579234e+00 -1.63380420e+00 7.43821621e-01 -2.06463218e-01 -5.11947572e-01 1.09110224e+00 -5.73222697e-01 8.23490620e-02 -2.91213710e-02 -1.95059791e-01 4.09783125e-01 1.46932030e+00 -1.22815204e+00 -1.83872730e-01 -5.08582927e-02 -1.17294518e-02 -1.86039001e-01 -6.05016574e-02 1.84526607e-01 -4.68507469e-01 -8.07288229e-01 9.00286064e-03 -1.00455499e+00 -6.72934949e-01 1.61885217e-01 -4.27575767e-01 -5.57885952e-02 6.34867370e-01 -7.34801233e-01 1.14197290e+00 -2.09095383e+00 7.33397007e-01 2.74317265e-01 2.04165027e-01 -5.63551113e-02 -1.11054391e-01 2.51435459e-01 -3.21890682e-01 1.90457255e-02 -4.48323667e-01 -3.80419821e-01 2.62889862e-01 9.81487110e-02 -4.08083171e-01 9.92199242e-01 5.23088098e-01 8.92089546e-01 -7.90913224e-01 -2.12883517e-01 6.99206516e-02 4.66895252e-01 -7.23802447e-01 1.39852196e-01 -3.30583185e-01 6.65701151e-01 -2.02465743e-01 4.74373758e-01 2.73135453e-01 -1.65730342e-01 -2.62078106e-01 -2.53692269e-01 -1.13739595e-02 -1.13195017e-01 -1.19767272e+00 1.84449124e+00 -3.26540262e-01 1.00150251e+00 2.40252614e-01 -1.33860040e+00 6.52702868e-01 2.01320246e-01 4.59164202e-01 -5.43384738e-02 1.83291286e-01 4.69762087e-01 -1.47333309e-01 -7.68547773e-01 4.90408301e-01 1.81668818e-01 -4.42814082e-02 -1.20967746e-01 4.44521308e-02 -3.17615330e-01 3.73229645e-02 -1.85904369e-01 1.16046917e+00 6.62319243e-01 1.77675486e-01 -3.27899903e-01 6.18157685e-01 -4.35097562e-03 4.34808344e-01 5.34280062e-01 -1.79574162e-01 9.63293195e-01 9.84964132e-01 -4.87635881e-01 -1.29327774e+00 -4.94426340e-01 -4.27948460e-02 9.35204804e-01 -2.43312106e-01 -2.45798185e-01 -8.50052774e-01 -4.98843968e-01 -6.29660413e-02 3.05643708e-01 -6.17099524e-01 -2.31354743e-01 -8.69704306e-01 -5.08361876e-01 5.88209093e-01 4.11087304e-01 -4.01842222e-02 -7.21591473e-01 -4.33966458e-01 2.01610968e-01 1.05230035e-02 -1.29528320e+00 -6.29434705e-01 2.27731615e-01 -1.00940645e+00 -1.02425706e+00 -7.17901170e-01 -6.94526076e-01 7.62154996e-01 -4.99095693e-02 1.18271446e+00 1.29631773e-01 -1.03549249e-01 6.33413792e-01 7.84324482e-02 -8.85535851e-02 -4.57222939e-01 1.01844594e-01 2.26763591e-01 -6.45485818e-02 1.78543683e-02 -4.74054307e-01 -5.27113199e-01 3.12336445e-01 -8.65871668e-01 -3.24382663e-01 4.27007943e-01 9.65908229e-01 5.22504985e-01 -2.31604874e-01 3.74437183e-01 -6.92007065e-01 5.69632411e-01 -4.94556308e-01 -9.10610497e-01 -4.80498485e-02 -3.99638087e-01 4.04502541e-01 6.35727644e-01 -6.90783083e-01 -2.84624338e-01 5.13191462e-01 -3.55863087e-02 -8.66677582e-01 -3.69264632e-02 4.97167885e-01 4.57394645e-02 -1.17869921e-01 9.44881320e-01 -1.75018206e-01 1.83810368e-01 -2.10636795e-01 3.52173239e-01 6.03478551e-02 7.85297155e-01 -7.54315615e-01 1.06066680e+00 4.23217297e-01 3.88058603e-01 -1.01799333e+00 -5.80295801e-01 -3.09535086e-01 -7.57385671e-01 -1.21028401e-01 8.09893787e-01 -9.11560357e-01 -6.96090102e-01 3.29100966e-01 -1.32222843e+00 -6.36370838e-01 -3.32308859e-01 3.94813359e-01 -1.06032205e+00 6.64139807e-01 -5.30346990e-01 -6.94292784e-01 1.16492897e-01 -1.52133620e+00 1.18247521e+00 -1.79243907e-01 -2.99653709e-01 -1.11360359e+00 1.39735773e-01 5.39370477e-02 3.30564648e-01 5.46136141e-01 6.38169408e-01 -6.09397590e-01 -6.94324613e-01 -3.61371964e-01 2.01138910e-02 5.88355720e-01 -2.51926839e-01 5.37157118e-01 -1.04447794e+00 -2.51171649e-01 1.41967088e-02 -2.30977356e-01 9.66672778e-01 6.05376184e-01 1.28390944e+00 -2.89125502e-01 2.79815309e-02 8.74892473e-01 1.29632211e+00 -6.09544218e-01 6.25581861e-01 4.85100985e-01 1.01082838e+00 7.52126634e-01 3.99935246e-01 7.58474022e-02 1.07581839e-01 9.04213309e-01 7.06266403e-01 -3.20757516e-02 1.11091644e-01 5.22852456e-03 7.67868936e-01 5.02675831e-01 -4.26531762e-01 1.55574977e-01 -8.66170883e-01 4.56490517e-01 -2.27617383e+00 -8.98815930e-01 -4.98941988e-01 2.53641438e+00 8.07134271e-01 1.74087301e-01 1.50395200e-01 4.70628589e-02 5.34030676e-01 4.28489186e-02 -4.74055976e-01 -3.66717845e-01 4.45743352e-02 -2.95313522e-02 6.68453515e-01 6.18422449e-01 -1.29584908e+00 9.66663241e-01 6.50653028e+00 6.35087729e-01 -1.14743292e+00 -1.49497271e-01 4.35305029e-01 -7.53933936e-02 -1.69226192e-02 3.64762396e-02 -6.18125916e-01 3.83479804e-01 1.03030479e+00 3.30965966e-01 4.81920958e-01 7.47252107e-01 6.95534050e-01 1.30876020e-01 -1.67531478e+00 9.06702161e-01 -9.11546648e-02 -1.10971725e+00 -3.28546286e-01 -3.26455105e-03 6.80873454e-01 -9.53027084e-02 2.25087404e-01 2.98199832e-01 2.19410300e-01 -1.44635880e+00 7.26881027e-01 8.86490166e-01 4.55024064e-01 -5.23144603e-01 5.45290411e-01 2.69347340e-01 -5.72953463e-01 2.23585609e-02 -4.40251946e-01 7.70245120e-02 -9.35800821e-02 4.26452786e-01 -4.47606355e-01 3.26401711e-01 4.99455929e-01 8.78508091e-01 -5.27411342e-01 1.20883560e+00 -2.66980916e-01 6.66799366e-01 -3.61653179e-01 3.52187157e-01 1.89068228e-01 -3.83247823e-01 7.48616338e-01 1.30423439e+00 4.19247180e-01 -6.04341328e-01 -9.98320803e-02 8.30654860e-01 -9.45626944e-02 8.47838372e-02 -8.71134758e-01 1.43376157e-01 -1.80225343e-01 1.17909205e+00 -4.03623968e-01 -4.75666039e-02 -3.81731093e-01 1.06060064e+00 4.89931077e-01 6.29818559e-01 -1.04926789e+00 -7.46345371e-02 5.33640802e-01 1.48279086e-01 4.49947327e-01 -6.09362304e-01 -3.52717578e-01 -1.40179443e+00 2.69089013e-01 -1.05382431e+00 -5.60458302e-02 -7.44172573e-01 -1.10998642e+00 2.15572059e-01 -6.48954734e-02 -1.35740113e+00 -5.43906868e-01 -8.63085449e-01 -6.72721326e-01 6.40817404e-01 -1.48468423e+00 -1.08936381e+00 -1.64620697e-01 5.71437359e-01 5.54806352e-01 -7.84037113e-02 5.19096375e-01 2.86735147e-01 -8.85812402e-01 4.74249840e-01 2.70784765e-01 6.52857646e-02 1.06714964e+00 -1.40535223e+00 2.72359043e-01 1.25866616e+00 -7.63566792e-03 6.44205153e-01 1.39984858e+00 -1.32219359e-01 -1.67022526e+00 -9.52201307e-01 5.54385424e-01 -9.58718240e-01 1.06236696e+00 -2.79933363e-01 -1.03961384e+00 9.54478264e-01 2.74656061e-02 4.12398100e-01 1.02959059e-01 -1.24076769e-01 -4.07257974e-01 -2.05604602e-02 -8.22856665e-01 7.67086089e-01 8.06980073e-01 -4.35923398e-01 -3.77215147e-01 3.73599172e-01 6.80746436e-01 -6.72514677e-01 -7.08263516e-01 3.46612781e-01 4.51260209e-01 -7.76997328e-01 1.03693116e+00 -9.23671424e-01 9.23988760e-01 -4.22645837e-01 -1.89791545e-01 -1.31914115e+00 -1.08387433e-01 -1.05687869e+00 -5.39190650e-01 8.94421816e-01 4.85133022e-01 -2.39412084e-01 6.67614460e-01 7.54704475e-01 -2.88688540e-01 -6.34855151e-01 -8.94404292e-01 -6.36114419e-01 3.20731223e-01 -6.10382438e-01 -1.03072919e-01 9.08012748e-01 -3.31203073e-01 2.40725011e-01 -5.84304214e-01 4.41397965e-01 6.61959708e-01 -4.75856811e-01 1.17412746e+00 -1.13161778e+00 -3.89419734e-01 -8.07526350e-01 -5.34213424e-01 -1.06911373e+00 3.63346726e-01 -6.13681555e-01 3.23596090e-01 -1.08325338e+00 -2.96021521e-01 -1.03723720e-01 -1.04068294e-01 1.32199213e-01 -1.79199621e-01 1.70200825e-01 -9.67859924e-02 1.80828646e-01 -5.05065799e-01 4.12247568e-01 1.04729748e+00 -1.63631335e-01 -2.13985473e-01 3.30153048e-01 -6.13557816e-01 9.41388905e-01 7.33946860e-01 -3.56238097e-01 -2.68205851e-01 -4.23292100e-01 5.87207556e-01 -5.62973358e-02 7.05941200e-01 -8.88075233e-01 3.20737630e-01 -1.16892286e-01 2.04343304e-01 -5.71881793e-02 1.21402569e-01 -1.18787277e+00 -7.31127113e-02 2.22517431e-01 -4.83398587e-01 2.29118928e-01 7.82969221e-02 7.47934401e-01 -1.18618995e-01 -4.27283227e-01 8.63172650e-01 1.13487482e-01 -4.80396599e-01 2.87626296e-01 -3.08300644e-01 1.10258604e-03 7.78400242e-01 -2.31295660e-01 2.13365145e-02 -4.87017900e-01 -8.86998653e-01 1.63159475e-01 6.89204037e-01 3.29824686e-01 5.44933915e-01 -1.24418485e+00 -8.15121114e-01 1.13275005e-02 4.23296243e-02 2.80150026e-01 -2.58332908e-01 1.43733966e+00 -6.91599727e-01 1.61802005e-02 1.22254327e-01 -8.05544317e-01 -9.97011185e-01 7.20746696e-01 6.42440677e-01 -1.99192524e-01 -7.15778828e-01 4.83694971e-01 -4.97604385e-02 -2.64930785e-01 5.97904444e-01 -8.93124580e-01 4.26391251e-02 -3.60329188e-02 1.22403614e-01 4.17313486e-01 1.10946454e-01 -5.74625790e-01 -6.58569187e-02 7.57356763e-01 1.76147237e-01 -3.25711593e-02 1.44990611e+00 -3.88221554e-02 -2.13308349e-01 6.07819617e-01 1.14486158e+00 3.01939040e-01 -1.82531798e+00 -1.02343358e-01 5.20046413e-01 -1.89273968e-01 -7.41024986e-02 -1.90593138e-01 -1.05156565e+00 8.51526141e-01 2.22624585e-01 1.47823384e-03 8.15607488e-01 -3.36589664e-01 3.25817883e-01 6.58181190e-01 -7.24137411e-04 -1.09300637e+00 1.12106539e-01 6.66535318e-01 1.17227304e+00 -1.58715618e+00 8.49596262e-02 -3.92373540e-02 -5.05874872e-01 1.45566964e+00 2.52999216e-01 -7.43333995e-01 3.55962694e-01 2.15961710e-01 5.06601706e-02 -9.34368372e-02 -6.26850307e-01 4.99652699e-02 5.73383093e-01 5.13336480e-01 4.82320398e-01 -3.70939285e-01 -5.18773049e-02 3.69686335e-01 -5.91791421e-02 -9.71786082e-02 7.33442307e-01 5.00084639e-01 -3.05273861e-01 -9.14279878e-01 -4.28468823e-01 5.95165901e-02 -6.97564483e-01 -2.07559783e-02 -1.62842363e-01 8.73719692e-01 -4.31856930e-01 5.94944835e-01 -3.24903607e-01 -8.63547251e-02 4.11406577e-01 1.18810147e-01 5.93683064e-01 -2.98909873e-01 -5.27023256e-01 1.43718943e-01 -1.01706386e-01 -7.73158193e-01 -4.67460454e-01 -7.75333524e-01 -9.64984834e-01 -1.63134515e-01 -1.28354073e-01 -2.20815748e-01 7.34063268e-01 8.61513853e-01 6.36284351e-02 5.84321737e-01 4.83218670e-01 -1.19618845e+00 -9.58392322e-01 -8.00558031e-01 -1.97383225e-01 4.91340309e-01 1.09602880e+00 -5.28379381e-01 -5.07860720e-01 2.61265695e-01]

[10.42286205291748, -1.2867143154144287]

b846df75-988c-403a-842d-363b0bfb51c3

a-cramer-distance-perspective-on-non-crossing

2110.00535

https://arxiv.org/abs/2110.00535v2

https://arxiv.org/pdf/2110.00535v2.pdf

A Cramér Distance perspective on Quantile Regression based Distributional Reinforcement Learning

Distributional reinforcement learning (DRL) extends the value-based approach by approximating the full distribution over future returns instead of the mean only, providing a richer signal that leads to improved performances. Quantile Regression (QR) based methods like QR-DQN project arbitrary distributions into a parametric subset of staircase distributions by minimizing the 1-Wasserstein distance. However, due to biases in the gradients, the quantile regression loss is used instead for training, guaranteeing the same minimizer and enjoying unbiased gradients. Non-crossing constraints on the quantiles have been shown to improve the performance of QR-DQN for uncertainty-based exploration strategies. The contribution of this work is in the setting of fixed quantile levels and is twofold. First, we prove that the Cram\'er distance yields a projection that coincides with the 1-Wasserstein one and that, under non-crossing constraints, the squared Cram\'er and the quantile regression losses yield collinear gradients, shedding light on the connection between these important elements of DRL. Second, we propose a low complexity algorithm to compute the Cram\'er distance.

['Nicolas Bondoux', 'Alix Lhéritier']

2021-10-01

https://openreview.net/forum?id=weBSeGTv0i

https://openreview.net/pdf?id=weBSeGTv0i

neurips-2021-12

['distributional-reinforcement-learning']

['methodology']

[-2.36468568e-01 3.01830888e-01 -3.73145729e-01 -4.58424747e-01 -9.56269503e-01 -7.11382031e-01 3.72441411e-01 3.94758463e-01 -6.22447371e-01 1.04440010e+00 -1.67213865e-02 -4.59169030e-01 -6.73394680e-01 -9.38251555e-01 -9.15255547e-01 -8.34299207e-01 -5.39815128e-01 4.56083328e-01 -7.99612328e-02 -2.47845903e-01 3.80134761e-01 5.12305379e-01 -1.34118843e+00 -2.34665349e-01 1.06908298e+00 1.18493915e+00 -1.33299679e-01 5.10080159e-01 -3.69310118e-02 2.80894309e-01 -4.03358519e-01 -5.47064781e-01 4.55017686e-01 -4.01951492e-01 -2.89818794e-01 -5.13251305e-01 2.24141434e-01 -2.93399483e-01 3.17125358e-02 1.29879808e+00 3.68583202e-01 3.51742119e-01 9.52372372e-01 -1.35259461e+00 -4.63591337e-01 7.93694496e-01 -8.88241351e-01 -4.61884812e-02 1.65246069e-01 -2.19506934e-01 1.25321233e+00 -5.84221005e-01 4.96998459e-01 1.29915953e+00 7.43366659e-01 3.10090154e-01 -1.48906457e+00 -4.64151680e-01 9.42625105e-02 -2.97727466e-01 -1.21297860e+00 8.53590071e-02 4.07481521e-01 -3.83321583e-01 3.15192044e-01 1.08123325e-01 3.87533784e-01 7.02684104e-01 2.19856516e-01 7.49152005e-01 1.17982328e+00 -4.14906412e-01 5.57097971e-01 4.08764958e-01 -1.88370287e-01 4.03226227e-01 4.52683240e-01 4.40970451e-01 -3.55937183e-01 -2.10372970e-01 6.72868133e-01 -8.60363990e-02 -1.58480868e-01 -9.40792143e-01 -7.34573007e-01 1.18609583e+00 3.79584104e-01 2.62210071e-01 -2.47306287e-01 3.85025680e-01 3.06446075e-01 3.95665079e-01 5.01444042e-01 4.07876134e-01 -2.51377821e-01 -2.54136294e-01 -9.96651709e-01 5.60694456e-01 8.29954267e-01 8.10471654e-01 6.58594608e-01 1.10340118e-02 -4.13599551e-01 4.63218182e-01 2.91694164e-01 6.94514990e-01 1.44374952e-01 -9.05794203e-01 7.29977667e-01 1.81159973e-01 5.97041130e-01 -8.13350320e-01 -3.03784966e-01 -6.07271254e-01 -4.47555542e-01 4.71698761e-01 9.36259627e-01 -4.55489457e-01 -3.72045904e-01 2.16744685e+00 3.33197802e-01 -2.58292079e-01 1.56281330e-02 8.81110072e-01 -2.63868570e-01 4.30772007e-01 6.27056286e-02 -1.94017917e-01 8.07260692e-01 -2.46273473e-01 -3.33819211e-01 3.09798956e-01 5.21694899e-01 -2.20044121e-01 1.25545418e+00 5.58949113e-01 -1.00871027e+00 -2.49123856e-01 -1.05520296e+00 3.38367641e-01 -2.36107036e-01 -1.64728403e-01 3.22149396e-01 9.09189939e-01 -9.05280292e-01 1.07749927e+00 -7.43794203e-01 2.29207173e-01 4.83895987e-01 8.55916515e-02 1.51004456e-02 7.67104700e-02 -1.14647126e+00 8.52810383e-01 3.27111304e-01 4.32394482e-02 -9.64501202e-01 -8.51685464e-01 -7.43320584e-01 2.56270468e-01 4.43464756e-01 9.67204347e-02 1.01479113e+00 -7.18435585e-01 -1.42788696e+00 2.28263423e-01 3.23561192e-01 -7.83664525e-01 1.27998900e+00 -3.87654632e-01 6.66123703e-02 1.02892043e-02 1.93666086e-01 4.69863355e-01 7.08958685e-01 -1.05754077e+00 -6.66704893e-01 -4.25487012e-01 -1.72265712e-02 1.14988446e-01 -2.96440452e-01 -5.00712097e-01 3.33307803e-01 -5.41650593e-01 -3.38595003e-01 -5.50495267e-01 -3.60083669e-01 -1.51917160e-01 -3.67178798e-01 -4.01651144e-01 1.99646652e-01 -3.14164609e-01 1.18718672e+00 -2.13956594e+00 2.87405867e-03 6.65634751e-01 -1.68007523e-01 -3.25024366e-01 -7.67273456e-02 6.85009718e-01 1.13340743e-01 1.10075861e-01 -6.16388917e-01 -2.10145682e-01 5.07896841e-01 1.95383996e-01 -7.14223385e-01 7.55906224e-01 1.71509489e-01 6.73774600e-01 -1.14634442e+00 -1.33213446e-01 -5.23176324e-03 2.59683907e-01 -4.24991995e-01 -5.59977219e-02 -3.68243366e-01 2.07051337e-01 -4.28828269e-01 1.76342458e-01 6.34567022e-01 4.17755306e-01 1.12955309e-01 4.15905654e-01 -2.90286571e-01 -1.11220576e-01 -1.34307683e+00 1.45541537e+00 -4.75465447e-01 4.09457564e-01 -1.63537893e-03 -1.22652185e+00 1.23421144e+00 -7.47884959e-02 4.85236138e-01 -6.58171594e-01 -8.81047696e-02 4.69467640e-01 -3.56825441e-01 2.15145270e-03 5.14231682e-01 -5.77920139e-01 -2.32248545e-01 4.89888370e-01 -7.79412091e-02 -1.41311184e-01 2.70098865e-01 -1.71070658e-02 4.34039861e-01 6.27556086e-01 -6.37155259e-03 -8.29101682e-01 2.53515542e-01 -3.27856034e-01 5.01002610e-01 9.26006794e-01 5.17567433e-03 3.58869761e-01 1.34044218e+00 -1.28637075e-01 -1.14499271e+00 -1.58755648e+00 -4.89131838e-01 9.50568199e-01 1.18657112e-01 4.56260219e-02 -6.76836193e-01 -7.19984293e-01 6.45841122e-01 1.16870868e+00 -8.83740067e-01 -1.51023701e-01 -3.18972528e-01 -6.85572505e-01 5.76579690e-01 6.00254416e-01 1.65809095e-01 -5.70886195e-01 -9.48104322e-01 2.02558249e-01 1.69505239e-01 -4.07002866e-01 -4.55007255e-01 4.30931151e-01 -1.01665998e+00 -1.02453423e+00 -9.89313900e-01 -1.12784065e-01 4.46919560e-01 -4.25540179e-01 1.07800138e+00 -6.06007636e-01 -3.64014804e-02 3.32035989e-01 -5.63922934e-02 -6.30536556e-01 7.16833547e-02 -5.30956946e-02 -3.44943292e-02 1.11621000e-01 1.09703213e-01 -6.41497254e-01 -6.65172637e-01 2.00697199e-01 -8.17568779e-01 -6.93188429e-01 3.04919183e-01 8.67125809e-01 7.79787660e-01 -1.06379110e-02 1.08142316e+00 -7.19885349e-01 9.81010437e-01 -6.05892956e-01 -1.16958380e+00 1.97822034e-01 -1.00850046e+00 5.38432181e-01 7.90707350e-01 -3.99657011e-01 -9.50759172e-01 -3.53841275e-01 1.61605611e-01 -2.70459592e-01 2.14487568e-01 2.91101962e-01 1.15432486e-01 3.42849344e-01 6.62237287e-01 -2.74808686e-02 -7.09591713e-03 -4.91859108e-01 6.28811240e-01 4.18164968e-01 5.59010029e-01 -9.94424224e-01 5.85992634e-01 4.58110482e-01 3.72261107e-01 -5.37569821e-01 -9.86584961e-01 -1.09066457e-01 -3.54086369e-01 -1.25809923e-01 6.57126904e-01 -4.13295388e-01 -9.32073772e-01 -2.36540303e-01 -4.89091039e-01 -4.09770221e-01 -1.04850757e+00 7.34527826e-01 -1.02638042e+00 2.78907061e-01 -2.21328840e-01 -1.51861727e+00 -1.13589674e-01 -8.46003711e-01 8.04948747e-01 2.79334277e-01 2.74344459e-02 -1.04634631e+00 3.67666602e-01 -3.13882887e-01 3.67458522e-01 8.46871197e-01 9.82112527e-01 -6.67120457e-01 -1.01480506e-01 -2.44016737e-01 -1.38405472e-01 4.40961808e-01 -3.92784595e-01 -1.56297445e-01 -6.35073841e-01 -4.25422877e-01 -5.78241013e-02 -2.80377895e-01 1.01613462e+00 6.57386720e-01 9.76444721e-01 -2.36181766e-01 1.15225203e-01 4.22322661e-01 1.67506778e+00 1.88214332e-01 3.94575477e-01 3.88721704e-01 9.44837108e-02 8.26564848e-01 6.73497379e-01 7.38386273e-01 1.20562367e-01 3.26594830e-01 5.92602968e-01 3.59770119e-01 5.62070549e-01 -5.26464581e-01 5.27260602e-01 -7.31699616e-02 1.63297623e-01 4.18939628e-02 -6.11412287e-01 7.08773136e-01 -1.89235580e+00 -8.94481838e-01 6.54349625e-02 2.82117510e+00 9.29877818e-01 3.58803421e-01 4.86598581e-01 -3.13457549e-02 6.17837012e-01 -4.73705083e-02 -7.75315762e-01 -8.84618700e-01 -4.15059254e-02 4.08209890e-01 8.35334003e-01 7.65631080e-01 -7.77993858e-01 3.44357282e-01 5.95450830e+00 1.00260639e+00 -8.44450653e-01 -2.11203963e-01 6.64340079e-01 -1.95701882e-01 -6.58128798e-01 -9.68065187e-02 -7.20664740e-01 6.38524234e-01 9.39197779e-01 -2.79262185e-01 3.57080221e-01 1.21234155e+00 6.65138215e-02 -3.86425108e-01 -1.16723156e+00 4.93082702e-01 -4.65815216e-01 -8.91801953e-01 -3.35022777e-01 2.35389978e-01 7.65436590e-01 -1.21136948e-01 3.44253242e-01 4.16109949e-01 5.75919986e-01 -1.16581988e+00 9.87119257e-01 7.54225910e-01 8.62314165e-01 -1.51063204e+00 8.33312869e-01 3.91948044e-01 -8.50596488e-01 -2.97454327e-01 -6.01351142e-01 7.22283646e-02 1.03142016e-01 9.10310566e-01 -6.75575793e-01 5.85323274e-01 4.84480202e-01 3.07048917e-01 -1.33135036e-01 9.45944250e-01 -1.38504982e-01 4.52611834e-01 -5.21508992e-01 -3.15898657e-01 5.52717805e-01 -7.52068281e-01 5.51334739e-01 1.30244339e+00 4.83935565e-01 -5.36299706e-01 4.17612679e-02 1.21542978e+00 -8.71986225e-02 2.11007580e-01 -5.15037954e-01 6.78520475e-04 3.84861648e-01 8.94776046e-01 -4.77194548e-01 -6.04610238e-03 -4.33512814e-02 3.37934256e-01 2.40786806e-01 3.18158656e-01 -7.24408746e-01 -8.70859206e-01 6.90592587e-01 4.52164449e-02 5.15357733e-01 -1.42416507e-01 -4.61529493e-01 -6.65248990e-01 1.81776986e-01 -3.48553568e-01 5.09670019e-01 -1.80837438e-01 -1.31248927e+00 2.08241060e-01 3.32370162e-01 -9.15113628e-01 -4.67893571e-01 -5.61089158e-01 -5.56134164e-01 9.75490570e-01 -1.68324494e+00 -3.50586295e-01 3.16966206e-01 3.98168206e-01 -6.89078402e-03 -1.35841966e-02 4.79467958e-01 -8.99646133e-02 -3.02497596e-01 6.50691748e-01 7.65153408e-01 -1.22985825e-01 4.53719795e-01 -1.89081287e+00 -2.23433301e-02 4.94653165e-01 6.14861362e-02 3.02964449e-01 9.75561202e-01 -4.86009836e-01 -1.09718597e+00 -8.48318279e-01 4.63746786e-01 -3.62366527e-01 8.25319171e-01 -3.03976238e-01 -7.14221478e-01 3.23839158e-01 -1.09339900e-01 -1.44922718e-01 4.05042827e-01 8.00234824e-02 -3.89946878e-01 -3.48145813e-01 -1.41613078e+00 5.63650310e-01 4.91958290e-01 -2.23265082e-01 -3.44062924e-01 2.79785953e-02 5.30448556e-01 -6.96732923e-02 -1.06722772e+00 2.57420361e-01 6.85755074e-01 -1.18358946e+00 7.20705688e-01 -7.75351286e-01 3.93710345e-01 -1.35048687e-01 -3.56712222e-01 -1.28403878e+00 3.08013976e-01 -7.70289838e-01 5.30372411e-02 9.63530779e-01 3.75126719e-01 -8.49798918e-01 6.83755279e-01 3.71528238e-01 1.89721093e-01 -1.25334466e+00 -1.30571008e+00 -1.09715688e+00 7.28261232e-01 -5.60798049e-01 5.58250785e-01 3.94627810e-01 1.05718978e-01 -2.91266382e-01 -2.54953265e-01 -2.24001601e-01 1.08989072e+00 2.98218012e-01 4.18872505e-01 -1.13801897e+00 -5.20266891e-01 -7.04099894e-01 -6.55512810e-02 -9.60126519e-01 -6.04778491e-02 -7.53232718e-01 1.23953350e-01 -1.05030906e+00 -1.73163697e-01 -6.50157273e-01 -4.24540162e-01 2.07636524e-02 1.83659419e-01 -2.29385808e-01 1.29410699e-01 -9.78704244e-02 -3.70088935e-01 7.67926395e-01 1.10924745e+00 1.74453676e-01 -3.64818692e-01 2.44459391e-01 -7.75455117e-01 6.57792449e-01 8.65939617e-01 -5.19430339e-01 -5.09541452e-01 5.35912104e-02 5.85251629e-01 2.84447253e-01 1.13471799e-01 -5.74925482e-01 -8.26498196e-02 -2.16756180e-01 5.07376015e-01 -4.69248921e-01 -3.81767005e-02 -4.11464036e-01 -3.21889579e-01 4.65286851e-01 -7.76625335e-01 4.66075577e-02 -8.77548233e-02 8.81735802e-01 -2.94655953e-02 -5.78837335e-01 8.86756778e-01 5.73863387e-02 1.74482875e-02 2.22783666e-02 -2.09435165e-01 4.49631006e-01 9.52789068e-01 -9.98697057e-02 1.56244919e-01 -5.87555349e-01 -6.04515493e-01 4.32538509e-01 2.77653217e-01 -6.74440991e-03 3.31161588e-01 -1.30684626e+00 -8.08068931e-01 8.83593634e-02 -2.19848961e-01 -2.68791169e-02 -5.68924621e-02 8.48758101e-01 -2.08619609e-01 3.41281116e-01 -4.52690460e-02 -4.40099359e-01 -3.50900888e-01 4.32217658e-01 5.30109882e-01 -5.01622379e-01 -4.67496783e-01 7.44255126e-01 1.08516209e-01 -4.24214721e-01 5.01628458e-01 -3.45384777e-01 7.34767318e-02 2.90199608e-01 4.03473794e-01 7.24408984e-01 -1.48313746e-01 -1.65489467e-03 -8.46557319e-02 4.01818365e-01 1.33977950e-01 -5.80335021e-01 1.39815271e+00 1.49232475e-02 1.86842471e-01 5.33988297e-01 1.09494877e+00 1.11228108e-01 -1.69965208e+00 2.03244146e-02 5.83793402e-01 -4.80582058e-01 -2.63204247e-01 -7.74314761e-01 -7.88935721e-01 1.05476713e+00 5.39902627e-01 3.86610746e-01 8.68098795e-01 -2.37878487e-01 3.91862124e-01 3.23709041e-01 4.04066831e-01 -1.37175953e+00 -1.10740595e-01 1.44734934e-01 8.00854862e-01 -8.21083724e-01 -1.09805398e-01 3.26580137e-01 -7.82419682e-01 1.34432149e+00 1.12203076e-01 -5.22356451e-01 5.20191729e-01 2.03478590e-01 -3.31347287e-01 3.19924206e-01 -4.14530933e-01 -7.53946975e-02 7.73396492e-02 5.92791736e-01 2.43352085e-01 2.82146841e-01 -5.44359982e-01 5.28241694e-01 -3.62768084e-01 -2.43136659e-01 5.28481901e-01 5.71494520e-01 -6.24024689e-01 -1.08982444e+00 -2.72830039e-01 3.00402224e-01 -4.99895126e-01 7.80619234e-02 7.74613991e-02 9.73697603e-01 -1.79441154e-01 7.11339056e-01 1.08298972e-01 1.30806416e-01 4.15258586e-01 7.66254663e-02 5.00707209e-01 -1.13011308e-01 -1.80245444e-01 8.01151544e-02 -1.92736581e-01 -4.73264337e-01 1.76496610e-01 -8.11061680e-01 -1.15832496e+00 -1.93785489e-01 -2.01060787e-01 6.47657871e-01 6.97943032e-01 7.05956459e-01 2.44341558e-03 5.01262024e-02 8.27991664e-01 -3.93938452e-01 -1.61382115e+00 -7.10891187e-01 -1.03743398e+00 1.82245895e-01 4.85749215e-01 -6.69847310e-01 -5.66504240e-01 -6.47888899e-01]

[4.137958526611328, 2.6024088859558105]

4c582116-97c3-423d-9630-d10f50511b90

a-self-attention-guided-multi-scale-gradient

2210.06334

https://arxiv.org/abs/2210.06334v2

https://arxiv.org/pdf/2210.06334v2.pdf

A Self-attention Guided Multi-scale Gradient GAN for Diversified X-ray Image Synthesis

Imbalanced image datasets are commonly available in the domain of biomedical image analysis. Biomedical images contain diversified features that are significant in predicting targeted diseases. Generative Adversarial Networks (GANs) are utilized to address the data limitation problem via the generation of synthetic images. Training challenges such as mode collapse, non-convergence, and instability degrade a GAN's performance in synthesizing diversified and high-quality images. In this work, MSG-SAGAN, an attention-guided multi-scale gradient GAN architecture is proposed to model the relationship between long-range dependencies of biomedical image features and improves the training performance using a flow of multi-scale gradients at multiple resolutions in the layers of generator and discriminator models. The intent is to reduce the impact of mode collapse and stabilize the training of GAN using an attention mechanism with multi-scale gradient learning for diversified X-ray image synthesis. Multi-scale Structural Similarity Index Measure (MS-SSIM) and Frechet Inception Distance (FID) are used to identify the occurrence of mode collapse and evaluate the diversity of synthetic images generated. The proposed architecture is compared with the multi-scale gradient GAN (MSG-GAN) to assess the diversity of generated synthetic images. Results indicate that the MSG-SAGAN outperforms MSG-GAN in synthesizing diversified images as evidenced by the MS-SSIM and FID scores.

["Ruairi O'Reilly", 'Mubashir Husain Rehmani', 'Muhammad Muneeb Saad']

2022-10-09

['ms-ssim']

['computer-vision']

[ 3.59123766e-01 5.97026683e-02 2.18946636e-01 -4.01783474e-02 -9.31345701e-01 -1.95503563e-01 4.21306401e-01 -3.54402393e-01 2.60481844e-03 9.81413901e-01 1.65579110e-01 2.77002039e-03 -4.82982621e-02 -7.98465073e-01 -7.73439348e-01 -1.02431071e+00 2.12276861e-01 4.86225076e-02 -1.04363903e-01 -4.19499159e-01 2.18212202e-01 4.88968760e-01 -1.23495674e+00 5.82132638e-01 1.22078764e+00 8.57145488e-01 2.03421623e-01 8.05545509e-01 1.52265549e-01 6.51778221e-01 -1.04123342e+00 -2.96291500e-01 2.20119253e-01 -1.13568461e+00 -4.98143345e-01 -5.81597611e-02 2.21243799e-01 -7.82753825e-02 -2.19286427e-01 1.00645876e+00 1.07271385e+00 -2.37024382e-01 7.75731981e-01 -1.23995018e+00 -8.02554965e-01 2.31623873e-01 -5.96214414e-01 6.69889271e-01 1.77038386e-01 4.71153051e-01 4.06779319e-01 -6.77562296e-01 7.14220524e-01 1.20207751e+00 7.24695623e-01 7.64873087e-01 -1.23977602e+00 -6.47950947e-01 -4.12997484e-01 4.55502719e-02 -1.05258417e+00 9.47535858e-02 1.00456905e+00 -5.19854188e-01 5.99709511e-01 4.82071042e-01 5.53238511e-01 1.24194288e+00 1.01916492e+00 3.81760746e-01 1.32449508e+00 -3.01261634e-01 1.91556215e-01 -8.08604658e-02 -4.35947150e-01 6.55721545e-01 2.50740737e-01 1.66123986e-01 -5.00146389e-01 5.10242954e-02 1.08940017e+00 -1.32184520e-01 -2.74289638e-01 1.96085855e-01 -1.05085886e+00 8.41199100e-01 7.74719119e-01 5.36914349e-01 -5.85390210e-01 -1.25082955e-01 4.37430471e-01 4.23869908e-01 4.18185115e-01 1.06347215e+00 -5.83177432e-02 1.22948974e-01 -8.60595047e-01 8.53826106e-02 -1.41170537e-02 5.62799752e-01 3.71800274e-01 6.75802529e-01 -7.16910362e-01 7.63668954e-01 -1.98752850e-01 4.46357012e-01 1.26423192e+00 -6.31437242e-01 3.41208518e-01 8.44873726e-01 -2.66460389e-01 -1.02216601e+00 -3.88746738e-01 -8.47464442e-01 -1.17414415e+00 3.18643361e-01 -6.81088939e-02 -1.22883417e-01 -1.21762764e+00 1.67310941e+00 2.18284830e-01 1.94491759e-01 1.57033622e-01 7.72906601e-01 1.04725385e+00 6.99493647e-01 8.47665519e-02 -2.56682545e-01 1.04552412e+00 -8.46454084e-01 -6.97944880e-01 -8.85341093e-02 3.52619559e-01 -6.84910595e-01 1.30253625e+00 1.33779244e-02 -1.31510139e+00 -8.18075597e-01 -1.23849571e+00 3.13155144e-01 -2.23498195e-01 1.47191146e-02 1.23521291e-01 6.47645772e-01 -9.18983221e-01 5.75822532e-01 -8.10859382e-01 8.61118808e-02 7.65930474e-01 2.57119805e-01 -1.16536222e-01 -6.87874407e-02 -1.22416508e+00 7.19752848e-01 1.47118807e-01 -1.27331242e-01 -8.10630798e-01 -9.67555881e-01 -6.90333426e-01 -1.59593686e-01 -4.47824001e-01 -8.41496110e-01 4.74033505e-01 -1.24459970e+00 -1.38097966e+00 7.76123822e-01 3.76984119e-01 -2.19545364e-01 6.54842734e-01 3.63677710e-01 -4.96967256e-01 3.80780905e-01 2.17220038e-01 7.44486868e-01 9.41157341e-01 -1.24436474e+00 -7.06692934e-02 -3.82849932e-01 -4.12203223e-01 2.25973889e-01 -6.63150623e-02 -2.46921375e-01 2.91723311e-01 -1.22207928e+00 -7.48340320e-03 -8.91059279e-01 -1.32332444e-01 -4.36764032e-01 -6.34001851e-01 4.23891276e-01 8.07491243e-01 -7.52030730e-01 1.06587660e+00 -1.95646405e+00 1.07701898e-01 8.01514313e-02 7.64363557e-02 3.82746279e-01 -3.81855398e-01 8.34295154e-02 -2.95193195e-01 2.19907612e-01 -4.00345623e-01 1.10418506e-01 -6.23600781e-01 -4.78414446e-02 4.61881347e-02 3.76153558e-01 4.27757710e-01 1.33458745e+00 -8.02754223e-01 -4.13237423e-01 1.20734297e-01 5.00784218e-01 -5.52179039e-01 3.74765307e-01 -3.12122647e-02 1.08684516e+00 -3.31242979e-01 7.67328918e-01 6.82505786e-01 -3.05835307e-01 -1.68443650e-01 -4.80336428e-01 2.77502865e-01 -1.98435575e-01 -5.36495447e-01 1.35270786e+00 -3.25352430e-01 3.58192921e-01 -4.43685323e-01 -9.62180436e-01 9.15163398e-01 3.04013610e-01 5.91239274e-01 -1.13949668e+00 1.35394827e-01 1.99566364e-01 2.94312924e-01 -4.63194340e-01 1.76381722e-01 -5.23182571e-01 -1.10546954e-01 2.31311634e-01 6.04285579e-03 -3.53374958e-01 1.59042496e-02 -7.12773427e-02 9.80846286e-01 -2.25510061e-01 -3.70928831e-02 -3.35352689e-01 5.75412393e-01 -2.86621630e-01 4.24089938e-01 5.16753554e-01 1.38108417e-01 1.10905659e+00 5.59431255e-01 -3.56024593e-01 -1.44513571e+00 -1.06536877e+00 -9.21250209e-02 5.38700223e-01 8.27669427e-02 1.89694017e-01 -9.52466011e-01 -5.25073290e-01 -2.52700657e-01 6.10787928e-01 -8.16302776e-01 -7.93810368e-01 -5.47080338e-01 -1.29894876e+00 5.65559328e-01 4.81798053e-01 5.10804474e-01 -1.52842951e+00 -7.63339698e-01 1.53459981e-01 -1.80104688e-01 -7.93472409e-01 -5.92702448e-01 -1.67446122e-01 -8.90848458e-01 -7.92337418e-01 -1.08491540e+00 -7.00567961e-01 1.00638723e+00 -3.88751298e-01 1.08884096e+00 -5.73566705e-02 -8.55118394e-01 -6.78916201e-02 -2.90074259e-01 -4.07011926e-01 -1.02028394e+00 -2.28216499e-02 -2.43967876e-01 6.76086694e-02 -3.54949653e-01 -6.38924956e-01 -1.04321992e+00 4.46434736e-01 -1.22783875e+00 1.30727217e-01 6.90418720e-01 1.36871028e+00 8.09090614e-01 6.72854781e-02 1.18656838e+00 -9.22301352e-01 7.66126573e-01 -4.18020278e-01 -1.85198382e-01 1.51179895e-01 -8.11475575e-01 -1.34738293e-02 9.28196073e-01 -5.66989243e-01 -9.22713041e-01 -5.65845788e-01 -2.34702960e-01 -5.25097847e-01 2.55226791e-01 2.87321746e-01 -1.39388636e-01 -2.05511034e-01 9.43419099e-01 4.44761544e-01 1.84117183e-01 6.34406805e-02 3.52518074e-02 3.93341005e-01 4.84478533e-01 -1.78176641e-01 4.53398287e-01 2.39156529e-01 1.67012617e-01 -5.25870204e-01 -6.16722822e-01 1.53996632e-01 -6.98127374e-02 -2.58408457e-01 1.00043631e+00 -8.97623360e-01 -2.26969272e-01 9.68484700e-01 -7.10990548e-01 -3.07097018e-01 -4.98433918e-01 3.45432490e-01 -6.01629436e-01 -4.93034720e-02 -7.75609672e-01 -2.08260283e-01 -7.66169906e-01 -1.44179165e+00 1.00737083e+00 5.12807667e-01 -1.28924906e-01 -8.82572949e-01 -1.04963437e-01 5.91778696e-01 6.91293895e-01 9.13580418e-01 1.28803241e+00 -2.65739620e-01 -3.60743284e-01 -1.51355535e-01 2.04021372e-02 5.80074728e-01 3.36543113e-01 -3.12764019e-01 -6.38789177e-01 -5.88997543e-01 1.80118948e-01 -2.95482904e-01 5.41117609e-01 7.63864815e-01 1.19351149e+00 -4.43072349e-01 -5.40186614e-02 8.88196647e-01 1.51461196e+00 6.48491859e-01 1.10197794e+00 3.13841313e-01 8.29970658e-01 4.61646914e-02 3.31506610e-01 3.27548921e-01 -1.04628652e-01 4.51162070e-01 3.30842525e-01 -5.69230556e-01 -6.00848198e-01 -2.80929923e-01 1.26204908e-01 7.59316444e-01 6.57873303e-02 -3.10418218e-01 -5.91889322e-01 3.94147843e-01 -1.11506391e+00 -1.01109600e+00 2.40051761e-01 1.90432870e+00 9.21536922e-01 7.42523372e-02 -4.93075959e-02 1.60929188e-01 1.02258170e+00 -6.19317256e-02 -9.40060854e-01 -4.41090226e-01 -4.58946556e-01 4.51671124e-01 4.39258695e-01 2.68435985e-01 -6.04086995e-01 4.98565555e-01 6.01150274e+00 8.04885328e-01 -1.75558484e+00 1.31059811e-01 1.37141788e+00 -1.35075167e-01 -5.25069475e-01 -6.97000444e-01 -3.28777313e-01 1.00117159e+00 7.44935036e-01 -9.31146070e-02 2.84400821e-01 4.65784878e-01 9.56569985e-02 1.05101161e-01 -6.42614722e-01 9.89733875e-01 2.64559388e-01 -1.47167790e+00 3.84551942e-01 -1.00487478e-01 1.30807984e+00 -2.21769273e-01 5.86260736e-01 -7.16773495e-02 -1.28269702e-01 -1.37159765e+00 4.41048920e-01 6.31244719e-01 1.43633151e+00 -9.22407508e-01 8.20225596e-01 -1.14469893e-01 -6.39340460e-01 -2.22228646e-01 -1.09687887e-01 6.12896204e-01 5.39168827e-02 6.58025444e-01 -9.13257241e-01 5.59669018e-01 4.40231979e-01 2.36356676e-01 -6.94060624e-01 6.56851947e-01 3.04122791e-02 5.31956911e-01 1.36718184e-01 2.74053484e-01 1.82774782e-01 -2.98516184e-01 6.44648194e-01 8.35732698e-01 6.41785979e-01 -1.36767179e-01 -3.32601011e-01 9.82264400e-01 -1.08142965e-01 -4.89896759e-02 -3.71906340e-01 -7.14863390e-02 2.98425823e-01 9.38957214e-01 -7.33361483e-01 -2.22708106e-01 6.08080588e-02 1.05715287e+00 -1.62415609e-01 1.45523250e-01 -9.35731709e-01 -2.55542815e-01 1.61180958e-01 4.32338297e-01 2.44961709e-01 4.25857395e-01 -5.39785087e-01 -7.72783697e-01 -4.48392034e-02 -1.25066435e+00 3.73352170e-01 -1.00747776e+00 -1.19551599e+00 1.09191668e+00 -3.22972238e-01 -1.22100163e+00 -2.98007637e-01 -1.53833494e-01 -7.57446647e-01 9.94202077e-01 -1.36180079e+00 -1.31806231e+00 -5.96273482e-01 5.29313564e-01 6.07475162e-01 -5.69917142e-01 6.77075505e-01 3.55591625e-01 -5.46592951e-01 8.84231091e-01 1.50241792e-01 -1.56968340e-01 4.69587386e-01 -1.00551450e+00 2.24365890e-01 7.60857284e-01 -3.95605385e-01 2.24740073e-01 7.57616639e-01 -7.70953000e-01 -9.87167895e-01 -1.36467063e+00 1.70459285e-01 -8.11965242e-02 -2.73022074e-02 5.15323207e-02 -7.70295620e-01 1.74226999e-01 9.32441279e-03 1.45879656e-01 5.51487923e-01 -8.23776424e-01 7.48784617e-02 -2.57412106e-01 -1.69456065e+00 4.57230389e-01 6.86391234e-01 -2.95118868e-01 -1.37228236e-01 1.90850616e-01 5.91522872e-01 -5.59900582e-01 -1.05691314e+00 6.58868968e-01 2.63892382e-01 -1.05031121e+00 1.03783536e+00 -3.90052259e-01 9.93336022e-01 -1.59279689e-01 6.91991076e-02 -1.66122758e+00 -3.69448423e-01 -5.54730594e-01 3.32487077e-01 9.28059399e-01 3.51858109e-01 -6.35360718e-01 7.36946702e-01 1.38680533e-01 -3.62622231e-01 -1.11626410e+00 -9.94259179e-01 -5.96436501e-01 2.24435538e-01 5.50749540e-01 7.11529672e-01 8.57107997e-01 -4.98805463e-01 2.67857462e-01 -2.92332560e-01 -1.14234678e-01 4.26333576e-01 3.33498195e-02 3.44699264e-01 -6.75127745e-01 -1.28377840e-01 -4.19309258e-01 -6.49522603e-01 -1.91550016e-01 -1.21966146e-01 -8.10585022e-01 -2.80446142e-01 -1.34416723e+00 2.00243175e-01 -5.18814385e-01 -4.39569950e-01 5.45151345e-02 -4.80421335e-01 6.87149048e-01 3.93792577e-02 2.36901805e-01 -5.63116483e-02 6.05525732e-01 1.93643582e+00 -2.10923687e-01 -1.92747220e-01 -3.53016071e-02 -7.75122523e-01 2.26785719e-01 8.43999028e-01 -4.53247100e-01 -6.41505897e-01 -5.83743453e-02 1.00084402e-01 3.35775822e-01 2.62685716e-01 -1.33452880e+00 -2.00918436e-01 2.23583877e-02 7.97015011e-01 -3.21525604e-01 1.27812788e-01 -3.70828867e-01 5.99041224e-01 8.41320276e-01 -3.92783761e-01 3.49286556e-01 9.07566920e-02 2.05432072e-01 -3.48640442e-01 -3.19793858e-02 1.35602617e+00 -3.08379978e-01 -1.12111330e-01 3.23017836e-01 -1.85630023e-02 1.55798510e-01 1.25732946e+00 -4.28457230e-01 -2.11885124e-01 -2.56515652e-01 -4.75718915e-01 -2.66006470e-01 5.51286995e-01 2.82397211e-01 8.30599189e-01 -1.48601115e+00 -8.28174055e-01 5.77328801e-01 -3.62459235e-02 1.10809602e-01 7.76008666e-01 6.73212945e-01 -8.23995650e-01 9.58292335e-02 -7.81909406e-01 -5.79478264e-01 -9.89008427e-01 3.70033592e-01 6.33217335e-01 -6.86404526e-01 -4.88876790e-01 9.62854028e-01 3.94597352e-01 -2.16709837e-01 -3.85437042e-01 -6.49094358e-02 -1.09695077e-01 -1.04117721e-01 2.92558521e-01 3.49554360e-01 3.10317993e-01 -5.39736092e-01 -1.55299202e-01 4.61277455e-01 -3.50833163e-02 2.72782475e-01 1.42400122e+00 5.84267639e-03 -2.64623258e-02 1.73777983e-01 1.23908651e+00 -1.95843592e-01 -1.33813190e+00 3.71550411e-01 -6.89848781e-01 -2.49340981e-01 -1.65219586e-02 -1.17440605e+00 -1.43148243e+00 5.67630887e-01 1.10484838e+00 -7.34645799e-02 1.47175407e+00 -1.90362737e-01 1.03266609e+00 -5.12296617e-01 -5.55814207e-02 -8.83599341e-01 7.89387703e-01 -3.96842137e-02 1.11763871e+00 -1.04041255e+00 -8.22517276e-02 -1.13147520e-01 -9.14575219e-01 7.90258110e-01 8.35687459e-01 -1.78696409e-01 2.30726168e-01 3.86307895e-01 2.97727704e-01 -3.30139905e-01 -3.51470143e-01 5.19988716e-01 2.58735865e-01 7.06081152e-01 2.67071366e-01 -1.44196339e-02 -2.86341667e-01 3.74835461e-01 -3.79890203e-01 -3.02409641e-02 5.21124244e-01 7.79130995e-01 1.16150364e-01 -7.70751655e-01 -3.55916768e-01 7.53789783e-01 -7.59917021e-01 -1.33378342e-01 9.62051675e-02 3.48250061e-01 3.94545972e-01 5.33306420e-01 1.52084520e-02 -3.57911021e-01 1.66435823e-01 -1.83961734e-01 6.27211571e-01 -3.61936450e-01 -7.09703326e-01 -1.21873431e-01 -4.46314007e-01 -2.85762817e-01 -3.79439533e-01 -2.82761753e-01 -9.91309047e-01 -2.70772893e-02 -2.95503169e-01 -1.15915947e-02 7.02709079e-01 6.03939414e-01 6.12638712e-01 1.23496687e+00 1.03890264e+00 -4.91627455e-01 -4.62771505e-01 -1.00213993e+00 -4.72109616e-01 8.65210891e-01 3.44111741e-01 -4.96340603e-01 -3.67672235e-01 1.90740034e-01]

[14.016707420349121, -2.001892328262329]

9a2d9163-1e30-49fd-9a17-5e13470793d6

faithful-to-the-original-fact-aware-neural

1711.04434

http://arxiv.org/abs/1711.04434v1

http://arxiv.org/pdf/1711.04434v1.pdf

Faithful to the Original: Fact Aware Neural Abstractive Summarization

Unlike extractive summarization, abstractive summarization has to fuse different parts of the source text, which inclines to create fake facts. Our preliminary study reveals nearly 30% of the outputs from a state-of-the-art neural summarization system suffer from this problem. While previous abstractive summarization approaches usually focus on the improvement of informativeness, we argue that faithfulness is also a vital prerequisite for a practical abstractive summarization system. To avoid generating fake facts in a summary, we leverage open information extraction and dependency parse technologies to extract actual fact descriptions from the source text. The dual-attention sequence-to-sequence framework is then proposed to force the generation conditioned on both the source text and the extracted fact descriptions. Experiments on the Gigaword benchmark dataset demonstrate that our model can greatly reduce fake summaries by 80%. Notably, the fact descriptions also bring significant improvement on informativeness since they often condense the meaning of the source text.

['Furu Wei', 'Ziqiang Cao', 'Wenjie Li', 'Sujian Li']

2017-11-13

['summarization']

['natural-language-processing']

[ 4.25779164e-01 9.71221149e-01 -2.95890987e-01 -1.28879339e-01 -1.22886956e+00 -5.92607439e-01 6.38849854e-01 4.56548125e-01 -1.99113563e-01 1.27358496e+00 1.10278642e+00 -1.68932751e-01 4.78721142e-01 -7.04084694e-01 -1.06558585e+00 -1.83097288e-01 4.74413097e-01 4.56448942e-01 -1.66013241e-01 -5.25495172e-01 6.41540825e-01 -2.42086295e-02 -9.71855402e-01 7.72615850e-01 1.51140308e+00 2.60143638e-01 -2.44153187e-01 6.91645443e-01 -3.26919675e-01 1.04239202e+00 -1.54332447e+00 -9.42416191e-01 -4.99519296e-02 -8.23582172e-01 -1.00116718e+00 -2.92649027e-02 7.81557500e-01 -6.70420766e-01 -5.42639673e-01 1.18386972e+00 5.06575942e-01 -1.55561790e-01 7.90360332e-01 -1.04026413e+00 -9.00406182e-01 1.44689715e+00 -5.01443684e-01 2.79310048e-01 3.35386485e-01 2.12506607e-01 1.36779106e+00 -4.77537215e-01 8.03565800e-01 1.05906820e+00 3.28890949e-01 6.63448393e-01 -8.11781287e-01 -3.31756830e-01 7.93626457e-02 6.89678565e-02 -6.02882147e-01 -6.48983419e-01 9.13776457e-01 2.65220199e-02 1.24964285e+00 5.98304391e-01 6.87462211e-01 1.48319864e+00 4.25026417e-01 1.21739697e+00 4.29374576e-01 -1.99336037e-01 -4.14534807e-02 8.78051221e-02 4.08532232e-01 6.78025544e-01 1.11406159e+00 -6.19249761e-01 -8.13873231e-01 -2.01056585e-01 8.70527327e-02 -5.23731351e-01 -4.85787272e-01 4.75683421e-01 -1.09521711e+00 1.01108849e+00 2.62679547e-01 1.89772353e-01 -6.02625668e-01 1.83116198e-01 7.69351065e-01 9.62871388e-02 7.54987538e-01 1.20174587e+00 -1.85754031e-01 -2.04681009e-01 -1.35072958e+00 5.03129303e-01 1.11873126e+00 1.07110322e+00 2.60384828e-01 2.74994820e-01 -5.29525280e-01 4.06627774e-01 -3.58442217e-01 5.23351133e-01 6.12174392e-01 -9.60394144e-01 1.05614984e+00 6.40100420e-01 9.58671644e-02 -1.18423760e+00 -4.96383160e-02 -6.19070053e-01 -8.64094913e-01 -5.19864082e-01 1.04296789e-01 -4.16571736e-01 -6.82690561e-01 1.39214504e+00 -1.33846864e-01 -2.23656088e-01 4.16786492e-01 5.92710376e-01 1.24949896e+00 8.02855670e-01 -3.39929223e-01 -3.40345621e-01 1.14378941e+00 -1.07148528e+00 -1.05480838e+00 -5.95873415e-01 8.74854982e-01 -5.83082080e-01 8.91847849e-01 2.36236319e-01 -1.45302725e+00 4.04116921e-02 -1.47781420e+00 -6.35033131e-01 -6.51843920e-02 2.09706679e-01 2.70722717e-01 3.22383285e-01 -6.44379020e-01 8.26671243e-01 -6.06442928e-01 -3.59363556e-02 7.45266378e-01 -6.75724074e-02 -2.54928023e-01 1.13821752e-01 -1.24384856e+00 1.05265558e+00 6.72792971e-01 -1.37550309e-01 -4.65530664e-01 -7.74700463e-01 -8.70162785e-01 4.44970012e-01 6.33292496e-01 -1.18359792e+00 1.30427039e+00 -7.19438434e-01 -1.35139930e+00 4.79136199e-01 -3.21334928e-01 -8.39490592e-01 6.81086063e-01 -8.76707613e-01 7.79525936e-02 4.42226589e-01 4.45673585e-01 4.05100465e-01 7.37671912e-01 -1.25286794e+00 -5.72647989e-01 -7.85175115e-02 -1.70636475e-02 2.50208616e-01 -3.30086917e-01 -1.70854047e-01 -9.45369899e-02 -7.97811329e-01 -2.13679135e-01 -5.69204688e-01 -1.77940931e-02 -7.75931835e-01 -1.42746067e+00 -1.64257437e-01 6.96521819e-01 -1.16383290e+00 1.41667020e+00 -1.64370978e+00 2.41430610e-01 -2.92175919e-01 4.47691798e-01 4.09736395e-01 -9.43872631e-02 6.69822156e-01 1.67871267e-01 5.65935791e-01 -5.90468466e-01 -4.61422712e-01 3.85397896e-02 1.61410257e-01 -1.06339622e+00 2.66943991e-01 6.01878822e-01 1.16862893e+00 -1.06318128e+00 -5.48931062e-01 -2.53661931e-01 1.00742854e-01 -6.95274055e-01 1.38714507e-01 -5.17724931e-01 1.76513225e-01 -4.64766443e-01 2.14827985e-01 4.19592261e-01 -1.41226292e-01 -4.89450023e-02 -5.74478395e-02 2.11467505e-01 1.11328495e+00 -3.84817570e-01 1.67427790e+00 -1.25757858e-01 8.51771057e-01 -3.41894597e-01 -8.93399954e-01 5.47019303e-01 2.69579738e-01 1.57216713e-01 -4.49453861e-01 1.41937494e-01 2.32704997e-01 -9.10229608e-02 -5.11180937e-01 1.39349294e+00 -5.50944693e-02 -4.49236721e-01 7.34582841e-01 1.36189967e-01 -5.53633034e-01 4.20133919e-01 1.09165192e+00 1.12464023e+00 1.22544598e-02 4.20425951e-01 -5.60474582e-02 2.23710835e-01 4.75510806e-01 6.02131784e-01 1.03422737e+00 1.03081398e-01 7.45179713e-01 1.23604858e+00 -2.44061112e-01 -1.31097317e+00 -6.78852975e-01 3.93042058e-01 3.54474008e-01 -6.53174594e-02 -7.98774004e-01 -1.12755668e+00 -1.13470185e+00 -1.13206789e-01 1.69491386e+00 -5.16235650e-01 -5.53289950e-01 -8.92248929e-01 -7.04511940e-01 1.10354769e+00 2.69844353e-01 5.00418603e-01 -9.35656130e-01 -6.82337582e-01 3.08902234e-01 -9.47287202e-01 -1.21074319e+00 -4.41496909e-01 -1.07423626e-01 -7.36006916e-01 -8.59889150e-01 -3.90060723e-01 -1.61344349e-01 4.85972941e-01 1.75320596e-01 1.27132344e+00 2.98901260e-01 1.91131994e-01 -2.32648909e-01 -3.08899522e-01 -5.48185229e-01 -9.69222546e-01 5.14556527e-01 -2.62435496e-01 -4.08282995e-01 1.42148793e-01 -4.80673283e-01 -2.37244591e-01 -5.96110225e-01 -9.60043848e-01 3.25534642e-01 6.26064181e-01 8.76271129e-01 9.85899642e-02 -1.32690176e-01 7.93662310e-01 -1.18997741e+00 1.05562663e+00 -5.04065096e-01 1.52892824e-02 2.16674343e-01 -3.05169076e-01 4.43606019e-01 9.21266496e-01 -1.10856354e-01 -1.21660852e+00 -6.35868967e-01 -1.58934772e-01 -2.96397600e-02 1.88300207e-01 6.57765865e-01 -1.47190511e-01 8.62213135e-01 8.83715630e-01 4.05784756e-01 -1.81405306e-01 -9.92314145e-02 7.87565768e-01 7.08301425e-01 8.42883050e-01 -5.19251168e-01 7.23356545e-01 4.67659891e-01 -2.60895133e-01 -9.15847301e-01 -1.32053196e+00 -3.71238776e-02 -3.97179484e-01 2.09188640e-01 5.92907310e-01 -8.77318263e-01 -3.00838053e-01 6.41075522e-02 -1.88179719e+00 5.64338043e-02 -5.90900421e-01 3.66338086e-03 -4.32501346e-01 7.25798011e-01 -8.09735477e-01 -4.95309472e-01 -1.01922071e+00 -8.51752877e-01 1.16855979e+00 7.62608945e-02 -6.78135753e-01 -5.99259019e-01 -1.30023718e-01 5.90039551e-01 1.35878697e-02 4.47441578e-01 1.06275249e+00 -1.10032165e+00 -4.56390142e-01 -3.08991313e-01 -2.16323316e-01 3.80942762e-01 1.05386034e-01 1.85920864e-01 -7.36241043e-01 1.04714692e-01 1.88202098e-01 -3.59068662e-01 1.32453191e+00 2.63563007e-01 6.83200955e-01 -1.30318594e+00 4.52043535e-03 2.28813365e-01 9.98515487e-01 -4.00626570e-01 7.88804770e-01 6.54751062e-02 9.61511791e-01 7.34899819e-01 3.02263290e-01 4.12605256e-01 3.79857302e-01 2.46937200e-02 2.55099028e-01 3.32365930e-01 -3.60744655e-01 -6.97009027e-01 5.03374040e-01 1.08500743e+00 2.35473722e-01 -7.56817698e-01 -5.08899748e-01 6.46587670e-01 -1.87622368e+00 -1.35788023e+00 -4.79502678e-01 1.80010653e+00 1.16958082e+00 2.83988804e-01 -1.51975915e-01 7.42368251e-02 6.20850325e-01 4.84267682e-01 -3.98475766e-01 -6.97910368e-01 -3.99414748e-01 -3.05263009e-02 4.29401815e-01 6.81705832e-01 -7.91098714e-01 1.35771561e+00 5.74970675e+00 8.36928248e-01 -1.00204706e+00 -3.88263538e-02 6.00762427e-01 -4.83174175e-01 -6.63138807e-01 2.64661969e-03 -7.22228825e-01 5.78049421e-01 9.00621712e-01 -6.07339859e-01 -6.82724128e-03 6.47997320e-01 2.52544224e-01 -2.39872381e-01 -8.50561142e-01 4.16783452e-01 4.67450619e-01 -1.72333825e+00 7.50503838e-01 4.94607762e-02 8.38207603e-01 -1.76118255e-01 -2.37820059e-01 2.66486019e-01 4.38337535e-01 -9.42189872e-01 9.48483109e-01 3.32863480e-01 5.37343562e-01 -8.89862418e-01 8.93294275e-01 5.80285311e-01 -1.15151107e-01 1.67955160e-01 -5.15214026e-01 -1.66378871e-01 5.72850227e-01 1.00986409e+00 -1.06791520e+00 9.40065145e-01 5.88709302e-02 7.26114929e-01 -6.01588488e-01 5.40120423e-01 -9.32362437e-01 8.10162246e-01 -1.35209396e-01 -2.16380030e-01 3.48182589e-01 -1.09104581e-01 1.10688126e+00 1.36110389e+00 2.50734121e-01 7.03801811e-02 -2.24917442e-01 1.08768737e+00 -7.83917367e-01 -5.00504561e-02 -6.99517787e-01 -5.71394861e-01 2.77093321e-01 9.32377994e-01 -3.96965504e-01 -9.96583700e-01 -1.94930825e-02 1.26419461e+00 4.24798250e-01 2.54391104e-01 -7.86913812e-01 -6.11733556e-01 2.99780816e-01 -2.26950899e-01 2.31815457e-01 -2.32860707e-02 -9.82907236e-01 -1.61591351e+00 3.13181579e-01 -1.08834243e+00 1.77547529e-01 -7.55428255e-01 -9.66909170e-01 5.16736507e-01 -2.01106220e-01 -6.69504166e-01 -4.02499884e-01 1.17469132e-01 -8.77703726e-01 6.48602545e-01 -1.38526380e+00 -9.84082699e-01 1.50243128e-02 -7.32025206e-02 8.39716613e-01 9.86102298e-02 5.49946427e-01 -2.14739755e-01 -7.71309972e-01 4.43517178e-01 -1.62659928e-01 2.91398883e-01 7.33173966e-01 -1.34682965e+00 7.86495268e-01 1.40677714e+00 1.50996462e-01 7.82788217e-01 1.23802447e+00 -1.30759621e+00 -1.24313402e+00 -1.14006710e+00 1.37409711e+00 -6.53551280e-01 6.55604064e-01 -1.30247340e-01 -1.02975655e+00 7.83295274e-01 8.15975130e-01 -7.81911194e-01 5.40333092e-01 -1.57398075e-01 -5.36690712e-01 3.74660760e-01 -8.77919376e-01 9.41403031e-01 9.49338019e-01 -4.16573375e-01 -1.35091579e+00 4.97599632e-01 1.08195651e+00 -5.01847625e-01 -3.11715275e-01 8.93837512e-02 3.42475593e-01 -8.82875860e-01 4.98499751e-01 -1.03401911e+00 1.62853730e+00 8.66318643e-02 2.04833984e-01 -1.71556783e+00 -3.22285369e-02 -8.99811089e-01 -5.28291941e-01 1.29461944e+00 6.08048737e-01 -4.22748089e-01 7.24354744e-01 3.78045678e-01 -6.30272508e-01 -5.02352536e-01 -8.11739862e-01 -5.34459531e-01 4.82341766e-01 -4.52968888e-02 4.92419541e-01 8.57264996e-01 4.99936879e-01 1.03153932e+00 -5.06985724e-01 -8.61384198e-02 6.55064762e-01 1.07798666e-01 9.21261311e-01 -8.52418244e-01 -1.90855309e-01 -6.42731428e-01 1.35373816e-01 -1.06223679e+00 4.88061607e-01 -8.61102998e-01 9.34689343e-02 -2.02367878e+00 3.71126503e-01 3.62952411e-01 4.96789038e-01 3.74548823e-01 -7.10994363e-01 -1.00122571e-01 1.92202210e-01 2.40679592e-01 -5.12624800e-01 8.29116344e-01 1.25800765e+00 -3.75054300e-01 -9.33229625e-02 -2.71082103e-01 -1.40303993e+00 5.97159207e-01 7.98804641e-01 -7.04243183e-01 -2.55076557e-01 -6.42365992e-01 2.96467006e-01 2.03988239e-01 1.55909657e-01 -6.39538407e-01 1.92532554e-01 8.59860703e-02 5.60425669e-02 -6.60007656e-01 2.43114889e-01 -7.53825977e-02 -4.41857994e-01 6.09023929e-01 -6.55363977e-01 -1.47908449e-01 4.79831807e-02 5.56322038e-01 -1.58648670e-01 -3.27280462e-01 4.94176686e-01 -3.67497563e-01 9.73974690e-02 -2.51568109e-01 -4.80757833e-01 7.14779377e-01 3.63165796e-01 6.13164306e-02 -1.09513783e+00 -7.18334854e-01 -8.03271607e-02 -8.73126369e-03 4.07899410e-01 1.30323142e-01 5.15621960e-01 -8.31014752e-01 -1.22960794e+00 -2.35789388e-01 -2.03028440e-01 1.44038126e-01 5.26246428e-02 7.78364539e-01 -5.05418062e-01 6.52039528e-01 -1.07230946e-01 2.02840511e-02 -1.02307403e+00 4.16291744e-01 -1.58095703e-01 -5.98577082e-01 -9.08227026e-01 7.02252567e-01 -5.68863750e-02 5.88455983e-02 -2.99170315e-01 -4.44386989e-01 -2.74070874e-02 2.43827775e-01 5.84222257e-01 4.78121489e-01 1.75867617e-01 -4.54677403e-01 8.55974928e-02 -1.86162397e-01 -4.64786917e-01 -2.88353741e-01 1.40967798e+00 5.33150416e-03 -2.57748842e-01 1.04187429e-01 1.08463657e+00 5.79239011e-01 -9.19117391e-01 2.66074724e-02 5.74207455e-02 -1.14518486e-01 -8.58994275e-02 -8.18936586e-01 -6.87062919e-01 8.18895459e-01 -1.00496030e+00 3.72591496e-01 6.34690285e-01 -6.83556572e-02 1.44537938e+00 7.32283771e-01 -5.86133488e-02 -1.04311013e+00 1.34718701e-01 6.94469869e-01 1.27896762e+00 -9.68299389e-01 3.71960878e-01 -6.08892620e-01 -1.01026022e+00 1.06715512e+00 3.47198188e-01 -3.26858789e-01 -3.35676014e-01 8.28412548e-02 -2.55527377e-01 -2.60996133e-01 -7.85733283e-01 2.25145027e-01 1.02378227e-01 1.87276170e-01 3.25963646e-01 -8.72028321e-02 -5.93316853e-01 9.13259387e-01 -8.92344713e-01 -3.80501330e-01 1.50641954e+00 7.89900362e-01 -5.62030971e-01 -6.02231801e-01 -1.23435251e-01 6.67126894e-01 -8.86501729e-01 -5.18213212e-01 -1.02530313e+00 6.26345694e-01 -5.09079337e-01 1.10154164e+00 -1.05616145e-01 -2.23282561e-01 1.95363015e-01 2.69334108e-01 3.96500558e-01 -7.89665759e-01 -9.62503433e-01 -1.84901744e-01 8.48508120e-01 -4.24301237e-01 2.36304309e-02 -4.71551299e-01 -1.37568891e+00 -5.85137904e-01 -3.03212702e-01 3.69079322e-01 5.85556865e-01 1.09348071e+00 6.67443037e-01 6.77429199e-01 1.56929076e-01 -5.02355814e-01 -7.49883652e-01 -1.20373929e+00 -1.34017304e-01 4.26997066e-01 4.89914507e-01 -6.76040053e-02 -6.02597415e-01 1.57509401e-01]

[12.308525085449219, 9.33747386932373]

13f4345d-3924-4490-bfae-f98a49b8e50d

investigating-robustness-of-dialog-models-to

2110.00687

https://arxiv.org/abs/2110.00687v1

https://arxiv.org/pdf/2110.00687v1.pdf

Investigating Robustness of Dialog Models to Popular Figurative Language Constructs

Humans often employ figurative language use in communication, including during interactions with dialog systems. Thus, it is important for real-world dialog systems to be able to handle popular figurative language constructs like metaphor and simile. In this work, we analyze the performance of existing dialog models in situations where the input dialog context exhibits use of figurative language. We observe large gaps in handling of figurative language when evaluating the models on two open domain dialog datasets. When faced with dialog contexts consisting of figurative language, some models show very large drops in performance compared to contexts without figurative language. We encourage future research in dialog modeling to separately analyze and report results on figurative language in order to better test model capabilities relevant to real-world use. Finally, we propose lightweight solutions to help existing models become more robust to figurative language by simply using an external resource to translate figurative language to literal (non-figurative) forms while preserving the meaning to the best extent possible.

['Taylor Berg-Kirkpatrick', 'Eduard Hovy', 'Varun Gangal', 'Harsh Jhamtani']

2021-10-01

https://aclanthology.org/2021.emnlp-main.592

https://aclanthology.org/2021.emnlp-main.592.pdf

emnlp-2021-11

['open-domain-dialog']

['natural-language-processing']

[-0.24671592 0.39816138 0.23645288 -0.5178073 -0.07096735 -1.1066654 1.1250987 0.18496531 -0.18666522 0.6638656 0.6963275 -0.6960911 0.16649085 -0.7827582 0.01609207 0.24439907 0.23170331 0.8245817 0.22819589 -1.0371041 0.2213707 0.3886814 -0.9502544 0.7027994 0.45672315 0.15672596 0.29781374 0.7042901 -0.56911606 1.1454588 -1.0706757 -0.33773533 0.15487012 -0.4297445 -1.2186258 0.10800107 0.318608 -0.47108954 0.17964758 0.5675936 0.03453454 0.30038926 0.59447193 -1.174779 -0.6215838 0.9099168 0.08999245 -0.04904411 0.8234243 0.30851364 1.0255188 -0.40935236 0.96372306 2.194678 0.51412296 0.7246905 -1.539885 -0.16301467 0.1759576 -0.34103283 -0.8050918 -0.46856642 0.65270543 -0.30981958 1.4503217 0.69207215 0.4289512 0.85494584 0.3371962 0.34328726 1.331313 -0.6502154 0.03594111 0.7403343 0.33420202 0.5766863 -0.36509234 -0.3062678 -0.23524842 -0.6730187 0.8112449 -0.08939282 0.01297361 -0.03347617 -0.963947 0.97506917 0.5734114 0.8393154 -0.04685257 0.1335914 0.39336917 0.7204665 0.4399826 1.1226405 -0.4423195 -0.43917722 -0.14418218 0.33543736 1.2663867 0.9411264 0.4252959 -0.19144848 0.27202976 1.18457 0.308149 0.22786948 0.51276326 -1.3370554 0.16407968 0.94896275 0.44750425 -0.7342678 -0.58855873 0.2450564 0.08560316 0.04784276 0.7442119 -0.2569992 -0.092683 1.7547704 0.13184877 -0.5968665 0.48142022 0.8147621 0.9295575 0.78382427 0.27933872 -0.10359389 1.6337721 -0.45714334 -0.68858594 -0.40104055 1.1670078 -1.0941993 1.8230705 -0.15135956 -1.1086806 -0.30561417 -0.89390767 -0.36670768 -0.7473315 -0.19111522 0.58124703 0.6092859 -1.1264697 0.12224663 -0.5005621 -1.0475725 -0.5712812 0.18356602 -0.16265701 0.42317426 -1.399591 1.5879906 0.4295084 -0.41993624 -0.06327292 -0.38123 -0.88266593 0.17132033 0.55457056 -0.5795619 1.7561022 -0.8977837 -1.4534348 0.84696895 0.12986344 -0.6091825 0.33193848 -0.08239135 -0.2601979 -0.02375018 -0.13436574 1.0404208 0.29083204 -1.5435618 -0.22950238 -0.16948693 1.131705 0.29156125 -0.4034211 0.29725102 0.12762737 -0.49443766 -0.19175003 -1.1661999 0.18841395 -0.04792446 -0.2578212 -0.4265982 1.1923563 -0.3610063 1.2086648 -1.8878236 0.04394121 -0.28271613 0.31947792 -0.00945848 0.2222912 1.160764 0.25203145 0.24251464 -0.12593195 -0.27229246 0.28125322 0.45294416 -0.49059376 -0.2192705 -0.04549476 1.1231041 -0.78989804 -0.71201104 0.51199 -0.09904138 -0.7110446 0.3400533 -0.7274752 0.10753573 -0.29279172 0.19780643 0.28034127 -0.23607706 0.8367446 0.2028272 -0.25734308 0.7711842 -0.6053531 1.3836 -1.2567112 0.70390254 0.3555953 0.01356234 0.86332446 0.27347565 0.07649147 -0.40733802 0.23688856 0.01830523 0.47727334 -0.3637932 0.77626693 -0.69776917 -0.43649068 0.95531714 -0.3488236 -0.84306616 -0.0538099 0.7283255 0.6907966 -0.16278026 0.55850685 -0.6435214 0.49894083 0.33732113 -0.20381337 0.44008383 -0.12653762 -0.03922472 0.5973787 -0.5219344 -0.6160443 -1.1630806 -0.13982895 1.3993211 0.2524776 -0.8151203 -0.7436909 -0.5681368 -0.11616573 1.5900247 -0.30700225 -0.18470477 -0.80215687 -0.22520386 0.6184934 0.26777405 0.77725387 -1.2837161 -1.1511192 0.27044782 -0.16106541 -0.97366565 -0.20153204 -0.11242294 -0.763763 -0.6538074 -0.07909957 -0.5622413 0.22453955 0.3304045 1.3608793 0.37928852 0.15798147 0.7772159 -0.48898557 -0.38970375 -1.1945032 -0.03967493 -0.34475857 -0.766746 0.41888136 -0.33594692 -0.05671503 0.52426994 -1.0349416 0.35356188 -0.18373309 0.8483346 -0.6924886 -0.3784606 0.59026563 -1.1312776 1.4433188 -0.4356275 -0.14298183 0.3510393 -0.24387652 0.17304085 1.0068861 -0.378946 -1.4639904 -0.5192973 0.15658505 0.2043376 -0.22618252 0.29335216 0.09306268 0.37275043 1.0350279 -0.27150968 0.4407258 -0.3718839 0.7554684 0.98034066 0.05294011 -0.9824266 0.43583143 0.37760973 -0.34493575 -1.0946279 -0.15275066 -0.264592 -0.50694156 -0.25834775 0.65121555 -0.43503678 -0.7788905 -0.02717567 -1.2129245 -0.7233668 -0.08923516 -0.11947028 -0.51113117 0.5041942 -0.7956035 -1.0085008 -0.08952357 -1.0830002 1.0787934 -0.1070261 -1.2966765 -1.5267322 -0.02425771 0.34884724 0.7485778 -0.15592326 1.4632276 -0.9594793 -0.12651144 0.11427808 -0.06699644 -0.23870805 0.7103795 -0.14719598 -0.80735016 -0.11267301 0.13844615 -0.6956252 0.09365149 -0.18791103 0.1444903 -0.6949167 -0.11418668 -0.51381534 0.8064608 0.5630086 0.262878 0.27097937 0.06044245 1.1400073 0.8117263 0.3597793 0.49998778 1.0558194 -0.15258895 -0.0290795 -0.04297645 -0.2211624 0.3477624 0.33358383 0.6144305 -0.41464472 -1.0230825 0.18008274 -1.7750843 -0.8103151 0.03482258 1.6841528 0.95649624 0.3659597 0.34544107 -0.498479 0.38420904 0.29296708 -0.2092451 -1.0206208 0.23663343 -0.13721439 -0.27773175 1.0467311 -0.6386691 1.4416118 6.261052 0.12400929 -1.3817521 0.0479424 0.54313344 -0.0336134 -0.313089 0.3685908 -0.11393429 0.23138061 0.9331941 -0.2031228 0.5063065 0.6369889 0.46036208 -0.32031712 -1.2242625 0.4955393 -0.16128983 -1.1872102 0.049999 -0.19254185 -0.06943157 -0.60444957 -0.15235975 0.57855135 0.5126849 -0.79929644 0.31936014 0.0602783 0.309982 -0.1811212 0.46256697 0.58769804 -0.8295837 0.04104523 -0.20934905 -0.6846361 0.3071292 -0.48271003 -1.5755041 0.10755524 0.2384749 0.3252688 -0.538077 -0.02643802 0.27221793 0.31054547 -0.47452223 -0.5594582 0.32342008 -0.2977298 0.7501075 1.2078631 -0.26698503 0.34237832 0.41956288 0.9184704 0.03792668 0.24114794 -1.1043918 -0.05576245 0.733897 0.95747906 -0.97365165 -0.6464697 -0.43861037 0.7276893 0.1646468 0.05778496 -0.6732476 0.03594697 0.4556947 0.34707898 -0.61639816 -0.5209481 -0.37595865 -0.9501203 -0.29478005 -0.99703956 0.38454264 -1.1991503 -1.2945316 0.7104493 0.81840885 -0.88268304 -0.9530679 -0.6342434 -0.8267006 0.62016743 -0.38271123 -1.3644103 0.063278 0.17069523 0.96931064 0.08615736 1.2004495 -0.38919148 -0.02012804 0.32447603 -0.36515406 -0.2352585 1.0108725 -1.304666 0.588119 0.27781337 0.0975235 1.5067884 1.118416 -0.6993574 -1.2570566 -0.40161932 0.62420803 -0.83831507 1.0095527 -0.6739235 -0.93108565 0.910846 0.86834526 -0.88288313 0.64824796 0.38819218 -0.41545275 0.48034716 -1.4684783 1.1921005 0.99921364 -0.5655785 -1.3451355 0.48213243 0.95135367 -0.23596445 -0.6723371 -0.16486207 0.524791 -0.87030447 0.88512254 -0.84547293 0.2940767 -0.04483173 -0.32007468 -1.2993952 0.20930253 -0.9303316 0.27734798 1.080129 0.5163103 -1.0837047 0.22161946 1.3730928 -0.1092835 -0.36770704 -0.8037284 -0.6785183 0.4814183 -0.06614723 0.4791797 1.0355226 1.1025611 0.98893696 -0.2429324 -0.4562694 -0.42588714 0.47218406 1.0970126 -1.0545166 -0.46152037 -0.28843123 -0.15308978 -1.4267691 0.3496836 -0.737607 -0.11933232 -1.4376996 -0.09960474 -0.50407404 0.6127173 0.46055263 0.01926008 -0.09785337 0.8243607 0.36119878 -0.30959123 0.5419586 1.1748013 -0.07748537 -0.7434823 -0.2658092 -0.85948694 0.82658064 0.78134006 -0.1339605 -0.8353243 -0.13589513 -0.17205623 0.29262784 0.18216228 -0.5500616 -0.01427639 -0.6071999 -0.26718485 -0.01784247 0.9259726 -1.0251813 0.02001452 0.5978911 -0.6781983 0.52422273 0.42715555 -0.02131286 0.04277348 -0.30335304 0.70898306 -0.34486705 -0.55347145 -0.9054439 -1.0785956 -0.01199493 1.0748957 -0.11613762 -0.7196805 -1.0107996 -0.5751104 0.20356472 1.0391433 0.66806483 0.31943834 -0.7516368 -0.1725973 -0.19325583 0.25927132 -0.60720664 -0.19667141 0.18471864 -0.7845703 0.6859084 -0.37493497 -0.29841134 -1.4765537 0.53928417 0.49828652 -0.05247118 -0.6219472 0.3819292 0.48205808 -0.80927336 -0.20588706 -0.68520707 -0.08322877 -0.10816229 0.234425 0.04636771 -0.21412104 -0.7182554 -0.31254214 -0.12593874 -0.28761145 -0.7410657 0.76610065 -0.20587125 -0.22230555 0.82261246 1.0073147 0.30133596 -0.68715274 0.03627537 0.01020414 -0.33351806 -0.5217794 -1.1451997 -0.06965404 0.5316708 0.23319493 0.954938 0.77322376 0.23338142 0.75733703 0.9192689 0.6403383 -0.7674703 0.18468292 0.7205152 1.3383441 -0.9645407 0.14536121 -0.6138261 -0.96906847 1.1382073 1.0174153 0.00732314 0.55754566 0.21613286 0.6813642 -0.67614114 -1.1586385 0.24046622 -0.22357303 0.24082242 0.90055877 0.33954746 -0.5528152 0.21475862 -0.8108109 -0.5947986 0.8217986 1.3812289 -0.2902953 -1.2917085 -0.5248399 0.35874537 -0.07945769 -0.16093367 -1.6571655 1.3467526 -0.4867636 1.4815662 0.30875558 -0.24390128 0.34055474 0.4873559 0.3619272 -0.95850766 -1.3355705 -0.01609826 0.7878331 -0.11797579 -0.20491345 -0.38120064 -1.6387556 -0.81277245 -0.2852648 0.2642611 0.20633054 1.0048168 0.37176815 0.06644657 0.18087827 -0.44382063 -0.45965102 -1.4664639 -0.11315843 0.8060372 -0.14808843 -0.69625974 -0.08116532 -0.03952071]

[12.875751495361328, 8.018468856811523]

79c68806-7bb5-48f3-9553-64d9b6ded42d

matsci-nlp-evaluating-scientific-language

2305.08264

https://arxiv.org/abs/2305.08264v1

https://arxiv.org/pdf/2305.08264v1.pdf

MatSci-NLP: Evaluating Scientific Language Models on Materials Science Language Tasks Using Text-to-Schema Modeling

We present MatSci-NLP, a natural language benchmark for evaluating the performance of natural language processing (NLP) models on materials science text. We construct the benchmark from publicly available materials science text data to encompass seven different NLP tasks, including conventional NLP tasks like named entity recognition and relation classification, as well as NLP tasks specific to materials science, such as synthesis action retrieval which relates to creating synthesis procedures for materials. We study various BERT-based models pretrained on different scientific text corpora on MatSci-NLP to understand the impact of pretraining strategies on understanding materials science text. Given the scarcity of high-quality annotated data in the materials science domain, we perform our fine-tuning experiments with limited training data to encourage the generalize across MatSci-NLP tasks. Our experiments in this low-resource training setting show that language models pretrained on scientific text outperform BERT trained on general text. MatBERT, a model pretrained specifically on materials science journals, generally performs best for most tasks. Moreover, we propose a unified text-to-schema for multitask learning on \benchmark and compare its performance with traditional fine-tuning methods. In our analysis of different training methods, we find that our proposed text-to-schema methods inspired by question-answering consistently outperform single and multitask NLP fine-tuning methods. The code and datasets are publicly available at \url{https://github.com/BangLab-UdeM-Mila/NLP4MatSci-ACL23}.

['Bang Liu', 'Santiago Miret', 'Yu Song']

2023-05-14

['relation-classification']

['natural-language-processing']

[ 2.55888373e-01 2.08992928e-01 -3.06281477e-01 -2.84647077e-01 -1.42526698e+00 -9.88678634e-01 7.57053435e-01 4.78696585e-01 -5.45140088e-01 7.93556392e-01 4.68080461e-01 -5.85907102e-01 -2.19035119e-01 -8.75021935e-01 -1.28241098e+00 -3.99291277e-01 5.61594784e-01 8.48006308e-01 -4.80168536e-02 4.81295325e-02 4.29657042e-01 3.31526279e-01 -9.24026787e-01 7.73597419e-01 9.87358987e-01 8.86173069e-01 4.27047640e-01 5.81923366e-01 -7.97697723e-01 5.74960709e-01 -6.54720128e-01 -6.35678589e-01 2.58843511e-01 2.26972967e-01 -1.12657535e+00 -5.57124496e-01 7.96929181e-01 3.87092561e-01 -1.92615956e-01 7.43026614e-01 5.55702806e-01 2.06567377e-01 7.79627621e-01 -9.24133122e-01 -9.51317787e-01 1.09168684e+00 -1.55865029e-01 3.17120671e-01 6.07009113e-01 1.31678104e-01 1.27384257e+00 -9.72152114e-01 9.79332745e-01 1.31736863e+00 5.06721795e-01 5.70010424e-01 -1.10708225e+00 -8.23811233e-01 -2.62761116e-02 8.75956118e-02 -1.17700660e+00 -5.17032743e-01 3.78486395e-01 -4.88889605e-01 1.21024346e+00 1.14066832e-01 -1.01418681e-01 1.53041530e+00 4.45305198e-01 1.04218698e+00 1.14015722e+00 -4.98238921e-01 3.75696318e-03 1.51487216e-01 3.84943485e-01 6.65169537e-01 4.34606940e-01 -3.43999803e-01 -8.74341726e-01 -1.32486716e-01 6.11463964e-01 -3.05332273e-01 -2.79206812e-01 2.01600239e-01 -1.82218480e+00 6.28365636e-01 1.13083154e-01 5.36780715e-01 -1.85314491e-01 3.77738625e-02 6.41389132e-01 3.86849195e-01 7.02412307e-01 1.19257545e+00 -1.27036476e+00 -7.26955533e-02 -6.78133249e-01 3.23905289e-01 1.33779168e+00 1.52516913e+00 6.87946022e-01 -3.02400589e-01 -7.65837431e-01 9.68267798e-01 -7.63200298e-02 5.04721820e-01 2.66352028e-01 -8.92572343e-01 1.05300045e+00 6.67759240e-01 -1.33765638e-01 -4.18188244e-01 -4.55090046e-01 -2.53277570e-01 -7.51701176e-01 -6.34800196e-01 5.98191082e-01 -2.22323194e-01 -1.00262368e+00 1.40835118e+00 2.34386250e-02 5.77720627e-02 3.27719569e-01 3.37215185e-01 1.76617944e+00 1.09985471e+00 6.52348518e-01 -1.14230253e-01 1.67873132e+00 -1.13880217e+00 -7.29405761e-01 -3.06127340e-01 8.96322012e-01 -1.09334195e+00 1.44710720e+00 3.73778999e-01 -9.99450862e-01 -5.36178827e-01 -5.66851854e-01 -6.31491959e-01 -1.11487317e+00 3.76007706e-01 7.01108396e-01 2.68450052e-01 -7.72730947e-01 5.65710068e-01 -2.67958969e-01 -5.80534935e-01 4.45894837e-01 1.27040809e-02 -3.92926514e-01 -3.52228224e-01 -1.21092880e+00 8.17073464e-01 6.18679404e-01 -3.11021894e-01 -9.42080081e-01 -1.42778194e+00 -8.61908913e-01 1.63314477e-01 7.36600459e-01 -9.94005859e-01 1.25242293e+00 -2.03668624e-01 -1.39790154e+00 1.17793369e+00 -3.26706260e-01 -3.19409102e-01 2.55914778e-01 -2.79177994e-01 -2.02100039e-01 -1.48025369e-02 3.32675070e-01 6.93238974e-01 3.77960801e-01 -1.17970216e+00 -3.67218494e-01 -2.01325808e-02 -1.01768337e-02 -4.28410619e-02 -3.31119597e-01 1.97798669e-01 -5.33249080e-01 -7.21133232e-01 -4.34711307e-01 -7.18143046e-01 -4.22339402e-02 -3.21307659e-01 -8.19503725e-01 -8.30768406e-01 2.21053779e-01 -4.09938872e-01 8.61176014e-01 -1.73380947e+00 -3.14421169e-02 -3.09716374e-01 5.01325391e-02 1.51418045e-01 -5.91867387e-01 5.10743856e-01 -1.53741300e-01 5.95446169e-01 -8.50736424e-02 -3.67741793e-01 8.68375227e-02 1.49120703e-01 -4.66173321e-01 -1.52392998e-01 3.50893408e-01 1.19373560e+00 -7.98173487e-01 -6.61244869e-01 -7.08955973e-02 5.38428640e-06 -1.37845919e-01 2.48045281e-01 -8.05899441e-01 3.98720711e-01 -8.53961289e-01 8.68926823e-01 3.89986396e-01 -6.09522104e-01 -3.35687578e-01 -2.99470842e-01 -2.32300192e-01 7.11481452e-01 -6.60753071e-01 2.24859810e+00 -6.87972367e-01 4.06075865e-01 -1.33036584e-01 -1.05520535e+00 7.21619844e-01 5.85932732e-01 4.56172347e-01 -6.95042789e-01 -7.86798820e-02 1.50416031e-01 -2.29058728e-01 -7.08326936e-01 3.62100393e-01 7.35462606e-02 -2.29049876e-01 2.88121402e-01 4.99262393e-01 -5.24192393e-01 6.95267797e-01 3.02968174e-01 1.38809788e+00 2.77312338e-01 1.59051478e-01 -6.19304359e-01 6.09160781e-01 3.25772613e-01 3.34806651e-01 1.02949870e+00 1.22081429e-01 3.05751860e-01 2.58881778e-01 -3.50052625e-01 -8.80898416e-01 -1.11690688e+00 -5.34304500e-01 1.67965543e+00 -3.72260958e-01 -6.53401077e-01 -2.73406357e-01 -9.02638376e-01 1.60863906e-01 1.09434152e+00 -6.21916354e-01 3.51630896e-01 -5.15572011e-01 -8.34254742e-01 5.84055424e-01 2.74733871e-01 3.39368612e-01 -1.39030206e+00 2.14231446e-01 9.97504592e-02 -2.39122391e-01 -1.60780990e+00 -4.79238898e-01 4.20333773e-01 -7.19982266e-01 -9.64062929e-01 -6.44791007e-01 -1.00202107e+00 4.01332676e-01 3.28074046e-03 1.65107608e+00 -4.73363489e-01 -3.02986741e-01 5.33015549e-01 -2.61295915e-01 -8.92599642e-01 -5.59151888e-01 7.31961429e-01 -3.30889881e-01 -6.16788208e-01 4.64190066e-01 -1.64294139e-01 -1.42989084e-01 4.33305353e-02 -9.54898536e-01 2.84497321e-01 9.33502018e-01 5.93586624e-01 7.19672561e-01 -2.48917609e-01 7.82300293e-01 -1.33519328e+00 5.52108943e-01 -5.34819365e-01 -4.92727220e-01 8.44780385e-01 -5.03489196e-01 3.12427253e-01 6.18041873e-01 -5.40845811e-01 -1.43006742e+00 -2.51701146e-01 -2.18702853e-01 -5.39867468e-02 -4.63955015e-01 7.53889680e-01 -2.74196565e-01 -1.55571371e-01 7.89733648e-01 3.10933948e-01 -7.95084000e-01 -6.22658670e-01 5.09675324e-01 3.63337100e-01 2.66083777e-01 -1.35802138e+00 7.37394869e-01 2.64874082e-02 5.44938445e-02 -9.03445423e-01 -1.60009253e+00 -6.52275741e-01 -4.44871664e-01 1.69950977e-01 1.03587735e+00 -9.94245291e-01 -8.23303401e-01 3.08926832e-02 -1.45001936e+00 -5.31918228e-01 -3.92338365e-01 2.43377537e-01 -4.96645987e-01 5.68902604e-02 -1.00247824e+00 -2.13774338e-01 -9.21914935e-01 -9.54065263e-01 1.41891682e+00 1.02929629e-01 -6.84305355e-02 -1.34366381e+00 1.98870935e-02 8.23219717e-01 2.94216394e-01 9.94614232e-03 1.27848041e+00 -1.14035821e+00 -5.25817871e-01 4.37220931e-02 -4.78525937e-01 5.14200293e-02 1.88575059e-01 -1.50346518e-01 -9.88407314e-01 1.06739424e-01 -1.10027693e-01 -6.80243254e-01 1.10397422e+00 2.76998103e-01 1.64179802e+00 -4.42635655e-01 -4.63722557e-01 5.21250725e-01 1.25125253e+00 -1.01070635e-01 4.05145228e-01 5.88508070e-01 8.62104177e-01 7.94084430e-01 4.71305639e-01 8.18050206e-02 3.89099807e-01 1.03217348e-01 -2.22973704e-01 6.83160350e-02 -2.30439648e-01 -3.46901387e-01 3.10912460e-01 1.01462638e+00 2.05535322e-01 -6.12829268e-01 -1.25707829e+00 5.72110474e-01 -1.61774015e+00 -5.09599328e-01 -1.31944492e-01 1.70516777e+00 1.55217409e+00 6.69376776e-02 -4.56511080e-01 -5.65204620e-01 2.86972672e-01 2.68998090e-02 -5.41768134e-01 -1.35208011e-01 -3.35833758e-01 5.49875677e-01 6.96873903e-01 2.15214074e-01 -1.20342898e+00 1.41379988e+00 5.90711546e+00 1.36405432e+00 -8.45626354e-01 2.61143506e-01 6.82156265e-01 -2.76590120e-02 -5.08582830e-01 -1.33738503e-01 -1.20447099e+00 1.56045124e-01 1.23952353e+00 -4.81612265e-01 1.01339780e-01 7.05596507e-01 5.73402904e-02 1.82189897e-01 -1.52361691e+00 7.79875755e-01 -7.37670511e-02 -1.84479308e+00 6.29890978e-01 -3.08105767e-01 7.98817337e-01 3.01557124e-01 -2.64362693e-01 7.50728130e-01 6.75948799e-01 -1.06459236e+00 4.66771394e-01 4.38447475e-01 7.90441632e-01 -4.09107655e-02 5.74966848e-01 3.04742485e-01 -9.98321831e-01 3.27665806e-01 -4.40080315e-01 3.82755190e-01 -5.11673465e-02 8.90461445e-01 -9.29974079e-01 8.25830698e-01 8.16607952e-01 7.91609049e-01 -6.18399978e-01 7.95801580e-01 -3.54209870e-01 8.47583115e-01 -2.46729612e-01 -1.07606396e-01 1.06904589e-01 -6.73596710e-02 4.98005003e-01 1.73159277e+00 2.90302753e-01 1.55549273e-01 1.78739920e-01 1.19156373e+00 -7.62972772e-01 4.43428040e-01 -6.59475446e-01 -6.19185627e-01 2.44992182e-01 1.49980259e+00 -4.77516234e-01 -6.45029008e-01 -5.61662138e-01 4.93679941e-01 3.24039161e-01 5.00838399e-01 -4.75938559e-01 -2.48466149e-01 2.56139547e-01 -4.41454090e-02 4.11300585e-02 -2.80503154e-01 -6.58109248e-01 -1.38742769e+00 -1.33602902e-01 -7.78602660e-01 6.04816437e-01 -8.07344258e-01 -1.87085664e+00 4.13041830e-01 1.07579380e-01 -6.42590880e-01 2.17255861e-01 -1.03369570e+00 -5.21487772e-01 7.33516097e-01 -1.74446571e+00 -1.44699597e+00 -1.04450949e-01 5.65923274e-01 9.38889146e-01 -2.55079299e-01 9.20082569e-01 1.96324393e-01 -6.85677648e-01 5.23889124e-01 7.98410363e-03 2.07795933e-01 1.44421577e+00 -1.38639855e+00 5.60640991e-01 4.29453105e-01 1.10704742e-01 8.30483258e-01 5.35332382e-01 -8.95846665e-01 -1.83239233e+00 -1.32209361e+00 1.18535149e+00 -8.61038804e-01 1.17211294e+00 -6.69091165e-01 -9.47015882e-01 8.26619864e-01 4.52251643e-01 -2.75702268e-01 9.51703668e-01 3.89565945e-01 -5.12421966e-01 -1.96284223e-02 -9.78644729e-01 3.94282907e-01 1.22249186e+00 -5.20291328e-01 -8.34879756e-01 1.18040645e+00 1.27631843e+00 -4.78042006e-01 -1.56462014e+00 4.53468323e-01 1.30163684e-01 1.46681860e-01 1.11154139e+00 -8.01108778e-01 9.03228283e-01 1.97977185e-01 1.79740097e-02 -1.09851289e+00 -2.58253545e-01 -5.93225181e-01 1.31695062e-01 1.49228859e+00 9.51181352e-01 -6.20782912e-01 7.00073242e-01 6.31170273e-01 -4.31047767e-01 -5.65632522e-01 -7.10017622e-01 -8.46558511e-01 7.79583693e-01 -5.06434917e-01 2.07847834e-01 1.13962364e+00 -1.76745489e-01 8.46516728e-01 2.13205308e-01 -7.70432130e-02 5.71908891e-01 3.21161002e-01 5.96817732e-01 -1.23355007e+00 -1.62968054e-01 -4.31718886e-01 5.95615983e-01 -9.55990016e-01 6.35967135e-01 -1.31960464e+00 9.37450305e-02 -1.88511050e+00 3.76531363e-01 -5.08790076e-01 -1.47271320e-01 9.93385315e-01 -3.04484963e-01 -3.19263607e-01 1.14435367e-01 3.28567713e-01 -7.59725630e-01 4.68292862e-01 1.58250916e+00 -5.51430404e-01 2.38450971e-02 -1.51440695e-01 -8.03843498e-01 4.29590553e-01 5.94502985e-01 -5.74679911e-01 -7.08120763e-02 -7.70506918e-01 4.77467567e-01 4.23852773e-03 -1.39612421e-01 -6.73715949e-01 4.67750549e-01 -4.35717762e-01 3.41611236e-01 -3.52564037e-01 2.95348596e-02 -4.39908773e-01 -3.01191241e-01 5.47004528e-02 -7.86917388e-01 -1.91698387e-01 5.86636245e-01 2.75080353e-01 2.06815123e-04 -3.00096571e-01 2.83783317e-01 -5.84899783e-01 -4.34577733e-01 5.09450078e-01 -2.57163852e-01 4.54110414e-01 5.48662841e-01 4.96154010e-01 -9.35054898e-01 1.46961406e-01 -4.87277985e-01 5.34656048e-01 5.48202693e-02 5.81492960e-01 3.26520711e-01 -9.23400283e-01 -1.06598854e+00 -3.38851720e-01 4.46836472e-01 4.10895705e-01 1.05133265e-01 3.97510022e-01 -3.26490194e-01 1.10577977e+00 1.07627988e-01 -3.59127373e-01 -8.35487187e-01 6.66234612e-01 -1.90976307e-01 -8.13513398e-01 -2.21896127e-01 9.46983933e-01 5.52717447e-01 -9.38296318e-01 2.64913172e-01 -7.58694887e-01 -1.24962315e-01 7.56258937e-03 2.87969172e-01 6.31055012e-02 3.58389765e-01 7.18123391e-02 -2.18571395e-01 4.87869471e-01 -2.95045853e-01 1.51847184e-01 1.52380514e+00 3.88592303e-01 -3.02643031e-01 6.24645531e-01 1.10219467e+00 -4.84375358e-02 -5.25407076e-01 -6.47771180e-01 4.35249567e-01 1.17391296e-01 -4.93729860e-02 -1.31791985e+00 -7.96298563e-01 7.40903437e-01 -2.53483541e-02 -1.01613835e-01 8.22165072e-01 2.37805396e-01 7.63646841e-01 1.12985837e+00 2.19520181e-01 -8.88409317e-01 1.46150053e-01 8.79748881e-01 1.26745522e+00 -1.57316279e+00 2.68563330e-02 -7.34775424e-01 -4.62414771e-01 1.19911075e+00 6.97371125e-01 3.22235554e-01 7.25404441e-01 3.25911552e-01 -7.11591095e-02 -3.72876137e-01 -9.50983882e-01 -3.84578966e-02 5.59297681e-01 3.37943286e-01 8.15814555e-01 -4.23024222e-02 -1.76744416e-01 7.29525983e-01 -3.67924511e-01 1.25199690e-01 2.77621627e-01 9.21401024e-01 -1.33322671e-01 -1.26381981e+00 -1.64073795e-01 6.70168757e-01 -7.06229389e-01 -5.54293275e-01 -5.85334837e-01 5.84279597e-01 7.07819611e-02 8.02579165e-01 -1.72455415e-01 1.86720967e-01 4.16826695e-01 3.55814368e-01 4.58629996e-01 -1.16483068e+00 -9.73569989e-01 -1.20871544e-01 4.93473738e-01 -3.60276639e-01 -4.37509865e-01 -4.77897644e-01 -1.21588171e+00 -2.74414897e-01 -1.28640383e-02 3.33269328e-01 7.83877671e-01 1.04743850e+00 5.30814171e-01 6.83776259e-01 -1.24679379e-01 -5.33980787e-01 -3.00334603e-01 -1.24475455e+00 -1.01770289e-01 4.21630859e-01 -6.01464957e-02 -3.48736703e-01 -1.99402928e-01 2.20301747e-01]

[9.909339904785156, 8.488085746765137]

d2009e98-c2fc-4637-8d0f-9ea465f420e2

short-term-electricity-load-forecasting-using

2305.10559

https://arxiv.org/abs/2305.10559v1

https://arxiv.org/pdf/2305.10559v1.pdf

Short-Term Electricity Load Forecasting Using the Temporal Fusion Transformer: Effect of Grid Hierarchies and Data Sources

Recent developments related to the energy transition pose particular challenges for distribution grids. Hence, precise load forecasts become more and more important for effective grid management. Novel modeling approaches such as the Transformer architecture, in particular the Temporal Fusion Transformer (TFT), have emerged as promising methods for time series forecasting. To date, just a handful of studies apply TFTs to electricity load forecasting problems, mostly considering only single datasets and a few covariates. Therefore, we examine the potential of the TFT architecture for hourly short-term load forecasting across different time horizons (day-ahead and week-ahead) and network levels (grid and substation level). We find that the TFT architecture does not offer higher predictive performance than a state-of-the-art LSTM model for day-ahead forecasting on the entire grid. However, the results display significant improvements for the TFT when applied at the substation level with a subsequent aggregation to the upper grid-level, resulting in a prediction error of 2.43% (MAPE) for the best-performing scenario. In addition, the TFT appears to offer remarkable improvements over the LSTM approach for week-ahead forecasting (yielding a predictive error of 2.52% (MAPE) at the lowest). We outline avenues for future research using the TFT approach for load forecasting, including the exploration of various grid levels (e.g., grid, substation, and household level).

['Konstantin Hopf', 'Felix Haag', 'Elena Giacomazzi']

2023-05-17

['load-forecasting']

['miscellaneous']

[-2.96648353e-01 -2.56011426e-01 1.12462915e-01 -2.19362646e-01 -6.38062358e-01 -4.69951659e-01 8.37860286e-01 1.60436809e-01 3.93928707e-01 9.07145858e-01 3.12465250e-01 -6.27872646e-01 -4.15381491e-01 -1.15839493e+00 -1.72799584e-02 -1.05771971e+00 -5.91031432e-01 3.94164741e-01 -3.51592332e-01 -4.76439148e-01 -1.69571489e-01 5.17060876e-01 -1.27895999e+00 2.76851326e-01 1.12994707e+00 1.29681432e+00 3.18557054e-01 -3.70738879e-02 1.49396017e-01 6.94893539e-01 -7.37822592e-01 -1.68369338e-02 3.48832160e-02 -3.46129447e-01 -5.39617121e-01 -1.46884844e-01 -4.26353902e-01 -4.10981506e-01 -1.85674772e-01 8.44444275e-01 8.39556277e-01 1.98959887e-01 2.30922639e-01 -1.21438849e+00 -2.64019430e-01 1.05449820e+00 -4.44441259e-01 4.94091064e-01 3.22983086e-01 -8.87187868e-02 7.51648903e-01 -7.14429200e-01 -3.35251018e-02 8.29760611e-01 7.88383961e-01 -3.61897320e-01 -1.32661343e+00 -6.71600342e-01 1.68862492e-01 6.99178219e-01 -1.28024232e+00 -2.21437067e-01 8.23005915e-01 -4.05276150e-01 1.75949788e+00 2.29622334e-01 9.15705919e-01 5.90538085e-01 7.08910227e-01 5.78125954e-01 1.50243556e+00 -3.36724758e-01 2.72017598e-01 -1.11158103e-01 -1.07028745e-01 -2.72945762e-01 -1.99832961e-01 3.27801406e-01 -4.48316872e-01 -1.79345962e-02 3.10428977e-01 -2.76224375e-01 -3.83194387e-01 3.28335226e-01 -9.91120338e-01 8.11937749e-01 3.51464152e-01 9.92364168e-01 -9.58284855e-01 -3.90328676e-01 6.17857873e-01 4.68146592e-01 1.26224184e+00 8.14221948e-02 -8.21354210e-01 -2.73059458e-01 -1.50775266e+00 1.25453800e-01 9.09666598e-01 6.30990863e-01 4.11863983e-01 1.07872140e+00 -3.63398880e-01 6.42189443e-01 -8.39047208e-02 5.01821756e-01 5.13892233e-01 -2.91720271e-01 3.46887529e-01 9.66712534e-02 9.62855592e-02 -8.09867322e-01 -7.88574994e-01 -1.11437988e+00 -1.21941435e+00 1.88526019e-01 1.01146892e-01 -5.35065770e-01 -5.21011293e-01 1.51805496e+00 -7.13701025e-02 1.97552755e-01 1.49415478e-01 2.90201575e-01 3.78664017e-01 1.17583966e+00 1.07755736e-01 -7.80931294e-01 1.25020492e+00 -6.28980517e-01 -1.11598897e+00 9.73799638e-03 8.55964184e-01 -7.69024134e-01 2.31349885e-01 5.33707738e-01 -1.13262749e+00 -4.61682171e-01 -7.90011466e-01 5.65881908e-01 -6.79757714e-01 7.87214264e-02 2.57972240e-01 4.04331118e-01 -1.33754492e+00 9.37560856e-01 -7.89193988e-01 -5.12616098e-01 -1.54307351e-01 1.24808066e-01 1.94198564e-01 3.30898881e-01 -1.71499097e+00 1.59134245e+00 7.02680349e-01 5.35317302e-01 -5.46340287e-01 -1.03863478e+00 -8.21334839e-01 5.59970379e-01 -2.21119359e-01 -3.56409132e-01 1.16142535e+00 -3.79399925e-01 -1.54433846e+00 -2.14982972e-01 -3.63012522e-01 -1.06269157e+00 2.65223235e-01 2.22251654e-01 -1.05901527e+00 -3.88238579e-01 -1.02243550e-01 -1.02776036e-01 3.27142388e-01 -9.20608580e-01 -1.04245889e+00 -1.69611871e-01 -4.79339898e-01 1.94721758e-01 -2.73725957e-01 8.66166502e-03 8.12923133e-01 -8.78305018e-01 -1.34393692e-01 -4.13413167e-01 -1.17485717e-01 -1.04218817e+00 -1.20319739e-01 -7.30812848e-01 1.23101377e+00 -1.48779547e+00 1.61734664e+00 -1.86020958e+00 -2.85611361e-01 1.96828008e-01 -2.87600338e-01 1.55656517e-01 3.38699669e-01 1.09678125e+00 -5.42297781e-01 -1.77321687e-01 8.57226402e-02 -3.46591562e-01 3.88893545e-01 3.69555056e-01 -7.30217993e-01 3.35217237e-01 -1.68771878e-01 1.17023063e+00 -7.32202649e-01 2.95646071e-01 9.22359943e-01 6.12844706e-01 3.19753885e-01 -1.75165832e-02 6.77745615e-04 4.51617807e-01 -1.18761919e-01 3.74181628e-01 7.82948852e-01 -1.47333503e-01 1.91908881e-01 -2.97855437e-01 -6.50044739e-01 5.65297067e-01 -8.13197434e-01 1.20415485e+00 -6.43463731e-01 7.66991854e-01 -3.23932767e-01 -1.14885569e+00 9.88656342e-01 9.91659105e-01 7.75603414e-01 -1.13641143e+00 -2.14666933e-01 4.02498305e-01 1.45255312e-01 -5.50936311e-02 2.25533694e-01 -3.18249315e-01 9.38451067e-02 5.37864745e-01 1.38759151e-01 9.79446620e-02 5.09116054e-01 -3.23420882e-01 5.38832009e-01 -4.39441726e-02 1.93576574e-01 -8.17287922e-01 2.57670909e-01 -2.94042617e-01 6.07013047e-01 1.88092574e-01 2.44945407e-01 8.52106698e-03 1.76500887e-01 -6.80639148e-01 -1.01153064e+00 -6.25106454e-01 -6.03270233e-01 7.54496813e-01 -7.25805223e-01 -3.15331161e-01 -4.46726710e-01 -1.10168532e-01 1.19301125e-01 1.61082184e+00 -4.03093785e-01 1.50984734e-01 -4.91947711e-01 -1.33153868e+00 -2.85182688e-02 5.86739004e-01 7.00559258e-01 -9.61115837e-01 -4.55450028e-01 8.95496011e-01 -2.51337290e-01 -1.04655051e+00 1.08047903e-01 5.37341893e-01 -7.67257929e-01 -2.60369390e-01 -9.32245672e-01 -5.02323985e-01 7.13518411e-02 -3.30699742e-01 1.35696769e+00 -5.85279763e-01 5.27794898e-01 -2.81267911e-02 -3.14862490e-01 -3.87499124e-01 -1.49759382e-01 2.26673558e-01 1.01645589e-01 -2.48889551e-01 2.51822561e-01 -1.06324708e+00 -4.36903059e-01 1.88149080e-01 -3.76685858e-01 1.81297868e-01 8.88033863e-03 8.26312363e-01 1.58772632e-01 1.03418291e+00 1.25718379e+00 -3.02618504e-01 6.76592112e-01 -6.98246896e-01 -8.85498643e-01 2.23662034e-01 -1.15590024e+00 -5.26195347e-01 9.85397279e-01 1.04299560e-03 -1.19271612e+00 -5.63282549e-01 -2.54170299e-01 -1.10669553e-01 -1.18986918e-02 1.13679922e+00 1.76076144e-01 -2.03889441e-02 -1.17572837e-01 6.49556637e-01 -5.12897372e-01 -7.84757853e-01 -1.93027496e-01 4.19579059e-01 1.14311144e-01 -1.35934815e-01 7.90285289e-01 -1.96731258e-02 1.17797591e-01 -5.09586453e-01 -5.90292275e-01 6.97039813e-02 -5.97082376e-01 -4.22687531e-01 5.03588319e-01 -1.16089821e+00 -6.64138973e-01 7.99977958e-01 -8.71383429e-01 -5.23633122e-01 -4.75548744e-01 5.96638501e-01 -2.75099486e-01 -5.79977706e-02 -7.39381135e-01 -9.58098352e-01 -5.33199430e-01 -9.75755513e-01 8.06549609e-01 8.62498805e-02 -2.48549685e-01 -1.67104304e+00 -3.02654266e-01 -2.91909218e-01 1.06598175e+00 5.63350379e-01 1.18193662e+00 -5.97534657e-01 -1.29296675e-01 -7.79178366e-02 1.89336706e-02 2.95661390e-01 4.35047925e-01 -3.31663311e-01 -1.12643325e+00 -8.95314157e-01 3.82046372e-01 3.09488267e-01 1.94957420e-01 6.14270866e-01 8.83677363e-01 -3.83589536e-01 -2.26174310e-01 4.07231003e-01 1.49488461e+00 7.99518764e-01 5.07681906e-01 4.72749472e-01 2.16329515e-01 3.61540437e-01 2.26602539e-01 8.16739559e-01 8.13166499e-01 5.58079183e-01 1.38462707e-01 -3.25700700e-01 6.84089512e-02 -8.14036205e-02 2.36411452e-01 1.30432713e+00 1.28551144e-02 -4.60415244e-01 -9.64908838e-01 7.00733960e-01 -1.65384293e+00 -1.02622879e+00 -8.99192542e-02 2.01504159e+00 4.04433727e-01 1.53257832e-01 -5.51912524e-02 7.41116583e-01 4.74362224e-01 3.38361591e-01 -4.04128581e-01 -5.36539435e-01 -4.76873159e-01 2.03565270e-01 3.19497794e-01 3.41175258e-01 -8.06789100e-01 3.34640563e-01 6.39036989e+00 9.87108648e-01 -1.25967765e+00 9.21016634e-02 9.02141392e-01 2.92012215e-01 -2.16307208e-01 2.68494207e-02 -7.50718772e-01 8.78165781e-01 1.49637270e+00 -9.37639475e-01 6.18543804e-01 3.02375138e-01 8.85491550e-01 -2.22389087e-01 -8.85368228e-01 6.20682240e-01 -3.50294888e-01 -1.01486456e+00 -6.06883802e-02 2.98740357e-01 1.08453763e+00 3.11810970e-01 -1.13540523e-01 5.26647925e-01 9.52995941e-02 -9.13588107e-01 5.84050834e-01 5.93112707e-01 3.58790725e-01 -1.04764271e+00 1.08150005e+00 5.03676116e-01 -1.62174010e+00 -2.96991497e-01 -8.69935974e-02 -3.12836975e-01 9.08375442e-01 1.14577639e+00 -5.47303259e-01 1.33953202e+00 1.12903702e+00 1.21448350e+00 -1.55781806e-01 9.56783593e-01 9.26516578e-02 9.09449399e-01 -5.80686986e-01 5.76963723e-01 3.32707137e-01 -3.15481424e-01 3.45991552e-01 8.82770300e-01 1.03183413e+00 1.00894339e-01 3.18974316e-01 4.49289083e-01 4.03492361e-01 -2.13212669e-01 -4.10270005e-01 2.59835154e-01 5.31637073e-01 1.13056028e+00 -4.72276479e-01 -5.83814383e-01 -4.16787207e-01 4.33059782e-01 -2.96976030e-01 8.24193776e-01 -5.55190921e-01 -2.50432938e-01 4.42162246e-01 -9.40328538e-02 3.20096701e-01 -1.09755382e-01 -5.28478682e-01 -9.48015034e-01 -1.82322096e-02 -6.01164877e-01 4.66918528e-01 -9.97880340e-01 -1.47840285e+00 8.04410458e-01 3.26467335e-01 -1.34329975e+00 -1.07160795e+00 -2.27486461e-01 -9.45191681e-01 1.57863855e+00 -1.95286429e+00 -1.00531983e+00 5.71770929e-02 4.34836239e-01 5.65521061e-01 -1.40409723e-01 1.20026100e+00 4.88134444e-01 -5.61489999e-01 1.56261623e-01 7.36884296e-01 -1.58160716e-01 -1.99256241e-01 -1.37470829e+00 6.08654916e-01 9.39948440e-01 -2.61335254e-01 -9.42718536e-02 8.73067379e-01 -4.95688111e-01 -1.06799436e+00 -1.04040504e+00 1.34332693e+00 1.53998375e-01 7.72707224e-01 -1.91657662e-01 -1.00806129e+00 9.76995826e-01 1.09616017e+00 -4.69344378e-01 4.91284430e-01 4.42758575e-02 3.79531652e-01 -4.35185432e-01 -1.20862448e+00 1.71103880e-01 1.51126221e-01 -5.38551331e-01 -3.18636715e-01 3.46331000e-01 1.58675075e-01 -1.99360967e-01 -1.88744450e+00 8.59192073e-01 3.17348391e-01 -8.90577435e-01 5.69517255e-01 3.10882151e-01 -3.45831960e-01 -3.42856795e-01 -2.33912423e-01 -2.15207100e+00 -8.38369727e-01 -7.91233122e-01 -3.69367808e-01 1.40919197e+00 3.52202117e-01 -1.42941415e+00 3.02959383e-01 3.89784932e-01 -2.88754731e-01 -8.89297009e-01 -1.37513578e+00 -8.18348110e-01 3.59970540e-01 -4.03735310e-01 1.40353858e+00 1.23612523e+00 4.39714760e-01 6.91551566e-02 -3.76965284e-01 2.75706857e-01 6.69266522e-01 4.99423057e-01 -1.36627853e-01 -1.21703982e+00 2.65300363e-01 -6.19894862e-01 -7.23191723e-02 -6.48678720e-01 1.65329844e-01 -8.22541296e-01 -4.73879635e-01 -1.98118258e+00 -3.57695699e-01 -1.24400616e-01 -5.69954038e-01 7.08543837e-01 2.72957087e-01 -6.31026328e-02 3.67219567e-01 -1.02190161e-03 4.38205421e-01 9.27111089e-01 8.72685611e-01 -6.74232617e-02 1.64487511e-01 2.06640258e-01 -7.18619898e-02 3.82277697e-01 1.24183285e+00 9.20861587e-02 -3.46754074e-01 -5.99698246e-01 4.73646522e-02 4.85552728e-01 -2.69452333e-02 -1.16621637e+00 2.77036667e-01 2.53056467e-01 8.60145271e-01 -1.25662899e+00 2.60266781e-01 -9.13928509e-01 9.12217796e-01 5.31216979e-01 2.56669998e-01 7.94854105e-01 5.94883204e-01 -9.62631032e-02 -4.13165122e-01 2.67011046e-01 4.28724915e-01 1.99424490e-01 -6.13896132e-01 1.08444825e-01 -8.58805001e-01 -6.44215226e-01 9.07548070e-01 -1.20636448e-01 -3.65051746e-01 -5.60264945e-01 -1.02408278e+00 6.69531882e-01 -7.60904327e-02 3.60644728e-01 3.16271819e-02 -1.46100116e+00 -9.66633022e-01 3.99474263e-01 -4.74823624e-01 -4.96393740e-01 3.35608751e-01 8.96072328e-01 2.89897323e-01 1.00754082e+00 2.13712975e-02 -6.21580660e-01 -5.10839045e-01 2.34820783e-01 7.11183488e-01 -7.34994233e-01 -6.75800323e-01 1.00441799e-01 -1.28072008e-01 -1.97441965e-01 -1.28014177e-01 -5.62816381e-01 -4.63756084e-01 6.90996468e-01 4.20825958e-01 8.25646937e-01 7.46735692e-01 -7.84221292e-01 -1.91509761e-02 4.16653663e-01 3.19918215e-01 1.95881844e-01 1.59332311e+00 -4.46468234e-01 -4.34558153e-01 8.08220685e-01 8.18159640e-01 -7.47838616e-01 -9.01649714e-01 -1.78607985e-01 1.88131064e-01 -1.28092458e-02 4.91777986e-01 -1.36337793e+00 -1.34504044e+00 6.88472211e-01 7.25015819e-01 1.10149646e+00 1.58707845e+00 -5.81980944e-01 8.89159858e-01 -1.37174606e-01 8.90220702e-01 -8.50337744e-01 -1.19299829e+00 7.57205844e-01 7.91534603e-01 -6.42770529e-01 -1.83344096e-01 2.26191223e-01 -7.66517669e-02 1.02808201e+00 -5.83213754e-02 2.77976304e-01 1.09811521e+00 5.72348058e-01 -1.28081262e-01 1.41701192e-01 -1.23482978e+00 -3.87786590e-02 2.07299784e-01 3.69538635e-01 5.81752121e-01 4.27307844e-01 -2.43403599e-01 4.03503895e-01 -6.95414424e-01 9.85360667e-02 2.41571173e-01 7.64152646e-01 -6.77434653e-02 -9.19817626e-01 -2.40322664e-01 9.01763082e-01 -6.41101956e-01 -3.89898121e-01 6.83845341e-01 6.04849160e-01 -1.65601447e-02 1.25435448e+00 3.11184943e-01 -8.93753916e-02 4.90283817e-01 3.31894815e-01 1.98229011e-02 -1.41003251e-01 -9.57072854e-01 5.05669832e-01 2.35637985e-02 -2.88349271e-01 -3.23830366e-01 -9.30725932e-01 -7.78141141e-01 -6.70805454e-01 -5.16686618e-01 6.87148750e-01 5.42486548e-01 1.15300381e+00 1.93841651e-01 7.30383813e-01 1.05043972e+00 -1.35803950e+00 -4.21766490e-01 -1.45125520e+00 -1.12278819e+00 -1.73432082e-01 2.45277584e-01 -3.93614620e-01 -3.70942116e-01 -2.59307235e-01]

[6.149807929992676, 2.812535285949707]

4567a63c-6f9d-4a58-8058-586a05e67eac

unsupervised-domain-adaptation-of-a

2011.11499

https://arxiv.org/abs/2011.11499v1

https://arxiv.org/pdf/2011.11499v1.pdf

Unsupervised Domain Adaptation of a Pretrained Cross-Lingual Language Model

Recent research indicates that pretraining cross-lingual language models on large-scale unlabeled texts yields significant performance improvements over various cross-lingual and low-resource tasks. Through training on one hundred languages and terabytes of texts, cross-lingual language models have proven to be effective in leveraging high-resource languages to enhance low-resource language processing and outperform monolingual models. In this paper, we further investigate the cross-lingual and cross-domain (CLCD) setting when a pretrained cross-lingual language model needs to adapt to new domains. Specifically, we propose a novel unsupervised feature decomposition method that can automatically extract domain-specific features and domain-invariant features from the entangled pretrained cross-lingual representations, given unlabeled raw texts in the source language. Our proposed model leverages mutual information estimation to decompose the representations computed by a cross-lingual model into domain-invariant and domain-specific parts. Experimental results show that our proposed method achieves significant performance improvements over the state-of-the-art pretrained cross-lingual language model in the CLCD setting. The source code of this paper is publicly available at https://github.com/lijuntaopku/UFD.

['Rui Yan', 'Lidong Bing', 'Hwee Tou Ng', 'Hai Ye', 'Ruidan He', 'Juntao Li']

2020-11-23

['mutual-information-estimation']

['methodology']

[-4.76857156e-01 -5.34447670e-01 -5.72328150e-01 -6.06347620e-01 -1.35778284e+00 -8.50036860e-01 6.52472734e-01 -1.32717937e-01 -5.81970930e-01 7.24526286e-01 2.68730730e-01 -3.07079047e-01 4.09635186e-01 -4.29700851e-01 -6.96658075e-01 -2.24628732e-01 1.41763791e-01 6.07966423e-01 -2.19781935e-01 -1.72742143e-01 -3.44798416e-01 1.58436224e-01 -9.69825625e-01 4.81391937e-01 9.43388820e-01 4.64731008e-01 4.03212011e-01 1.84087932e-01 -5.30390143e-01 2.35485330e-01 4.01868410e-02 -4.85667378e-01 4.08539891e-01 -2.14224115e-01 -9.19054151e-01 5.60005046e-02 1.77442461e-01 5.32546081e-02 -3.10765147e-01 1.01623535e+00 1.63855985e-01 1.75065145e-01 6.85733616e-01 -7.49297261e-01 -9.85116780e-01 9.52579618e-01 -7.49468148e-01 3.73621315e-01 1.53971866e-01 -4.91236150e-02 1.16576791e+00 -1.49402165e+00 6.63610876e-01 1.41234183e+00 4.07511890e-01 4.44117516e-01 -1.28131723e+00 -9.73748982e-01 4.81176227e-01 1.45703014e-02 -1.53203177e+00 -6.51208460e-01 8.37862968e-01 -4.04769659e-01 1.38617826e+00 -4.50575739e-01 2.38238871e-02 1.17192781e+00 4.63897213e-02 1.20275354e+00 1.23824024e+00 -7.85643101e-01 -1.39967605e-01 4.84096915e-01 1.20950565e-01 6.37575388e-01 3.48623902e-01 -2.24958211e-02 -7.05835104e-01 -1.03820972e-01 4.37890381e-01 -1.90912768e-01 1.15776043e-02 -3.28781694e-01 -1.30794644e+00 9.89605308e-01 4.73098084e-02 8.62654388e-01 -3.20318699e-01 -3.55868220e-01 7.86615074e-01 5.32705545e-01 9.31483269e-01 2.29437947e-01 -1.17606080e+00 -1.14047244e-01 -8.32394958e-01 -2.02323750e-01 5.58557451e-01 1.12829924e+00 1.14142132e+00 1.36029840e-01 5.29870510e-01 1.14067888e+00 4.09645885e-01 6.49929225e-01 1.04599512e+00 -2.81533062e-01 9.25813675e-01 5.31482875e-01 -3.37309778e-01 -4.51252133e-01 -3.22577327e-01 -4.65632826e-01 -6.86963022e-01 -6.35600507e-01 1.11752652e-01 -3.40955496e-01 -6.05627775e-01 1.80104601e+00 3.96560431e-01 1.15762748e-01 6.79665446e-01 4.16733146e-01 3.77576202e-01 8.15122545e-01 2.33100638e-01 -1.69681802e-01 1.40169656e+00 -1.01256359e+00 -5.82971871e-01 -6.67132199e-01 1.18035483e+00 -1.13082540e+00 1.18856990e+00 1.68859139e-01 -8.35705936e-01 -6.73797190e-01 -8.15682948e-01 -2.55393893e-01 -6.29357338e-01 4.82893288e-01 4.70533937e-01 4.04313743e-01 -6.83659434e-01 -7.51717538e-02 -9.16041493e-01 -5.62605798e-01 -1.49689019e-02 8.62261429e-02 -6.25482798e-01 -6.73159599e-01 -1.57195103e+00 8.77000153e-01 7.02517331e-01 -2.45503142e-01 -5.10629773e-01 -6.84971452e-01 -1.21100426e+00 -3.09605628e-01 2.83557445e-01 -3.51935655e-01 1.22804213e+00 -1.06063068e+00 -1.35949302e+00 9.91458595e-01 -4.98536944e-01 -3.65712196e-01 1.34616867e-01 -3.85098845e-01 -7.68702209e-01 -1.51057854e-01 5.47471225e-01 1.68349355e-01 5.72101593e-01 -9.83309329e-01 -7.03003645e-01 -4.33840960e-01 -4.36503530e-01 4.62717175e-01 -5.71641088e-01 1.58450499e-01 -6.25374019e-01 -7.94022679e-01 -2.36001369e-02 -8.81145954e-01 1.83908623e-02 -6.93720460e-01 -1.08105540e-01 -5.93062222e-01 7.01381743e-01 -7.22131610e-01 1.16384399e+00 -2.15751719e+00 3.97991166e-02 -9.48512331e-02 -3.66261870e-01 3.55489463e-01 -3.65504861e-01 6.08953714e-01 1.19774409e-01 5.53077459e-02 3.64232901e-03 -5.81451952e-01 9.44913253e-02 3.25255603e-01 -3.88442636e-01 4.48536277e-01 2.93566346e-01 7.36446261e-01 -9.93597567e-01 -3.78406107e-01 1.34330213e-01 6.09191000e-01 -5.02848685e-01 1.24837950e-01 2.00200640e-02 4.19708908e-01 -6.62807882e-01 5.36964476e-01 7.26436377e-01 -1.27051100e-01 6.68851137e-01 -1.89940676e-01 -1.87264457e-01 7.54682660e-01 -7.43267596e-01 2.23699999e+00 -1.07905233e+00 2.77976304e-01 -2.02796161e-01 -1.05731416e+00 1.03651130e+00 4.50345933e-01 3.85670215e-01 -5.44207513e-01 5.80259524e-02 5.70067942e-01 -2.62820691e-01 -1.82308808e-01 5.52609801e-01 -4.12204385e-01 -7.33047783e-01 7.93630302e-01 6.29822493e-01 4.28581610e-03 2.49457777e-01 1.52465567e-01 6.37639523e-01 2.71976739e-01 7.61458933e-01 -4.06051725e-01 7.98317730e-01 -8.30369443e-03 6.49615467e-01 2.98367023e-01 -2.58268476e-01 2.63761263e-02 -2.76990235e-01 -3.09158504e-01 -8.14836323e-01 -9.74025905e-01 -2.11684138e-01 1.57535279e+00 -2.89968938e-01 -3.86785418e-01 -4.21357244e-01 -8.59825790e-01 1.41429216e-01 7.62077451e-01 -2.66789824e-01 -2.47455090e-02 -5.71491420e-01 -7.17811465e-01 5.81805766e-01 4.25639778e-01 5.23941338e-01 -7.05145240e-01 3.35634172e-01 3.77417743e-01 -3.24305058e-01 -1.46189511e+00 -7.70485580e-01 3.36530238e-01 -7.29157686e-01 -6.25169456e-01 -5.19546449e-01 -1.20802414e+00 5.38830936e-01 5.26260197e-01 1.15638924e+00 -5.64690292e-01 9.49001610e-02 5.02806425e-01 -4.80053663e-01 5.94118703e-03 -4.64900106e-01 6.18613899e-01 7.38061011e-01 1.41632259e-01 1.14909720e+00 -4.58854198e-01 4.52406295e-02 6.79779947e-02 -8.14875543e-01 -1.45665422e-01 7.00005829e-01 8.56077611e-01 9.06400204e-01 -7.57609382e-02 7.47323692e-01 -1.06799626e+00 5.55174589e-01 -9.26847219e-01 -5.94475389e-01 3.45342159e-01 -5.75985551e-01 5.15571833e-01 1.07483220e+00 -4.69545841e-01 -1.44937623e+00 1.83882996e-01 1.83826640e-01 -3.51266444e-01 -1.22080617e-01 9.78463233e-01 -2.24951982e-01 3.11651021e-01 4.86807078e-01 6.27116740e-01 -6.18284822e-01 -7.80775368e-01 6.23829961e-01 8.65329504e-01 3.16286504e-01 -1.06791389e+00 9.83025908e-01 2.75139332e-01 -7.44588614e-01 -6.79950833e-01 -1.16502047e+00 -8.74999106e-01 -1.37656581e+00 3.50439459e-01 5.94279826e-01 -1.66183817e+00 2.27283299e-01 2.21812442e-01 -1.05423903e+00 -2.95188755e-01 -6.63429573e-02 7.85545230e-01 -2.61044860e-01 3.71829152e-01 -4.31283206e-01 -4.34840411e-01 -2.94637471e-01 -8.68364155e-01 9.72513556e-01 3.78364474e-02 -2.76900567e-02 -1.76194715e+00 4.74742204e-01 2.64078707e-01 2.47226104e-01 -4.55754727e-01 7.75484562e-01 -1.02101243e+00 -1.37448445e-01 -1.19915128e-01 -6.24611937e-02 5.87044418e-01 4.99305844e-01 -4.13529813e-01 -8.89664948e-01 -6.22369647e-01 -1.43728659e-01 -7.13789523e-01 6.05723798e-01 2.40289867e-02 5.20847142e-01 -1.41441524e-01 -2.56959140e-01 7.19480455e-01 1.53116643e+00 -2.47130036e-01 -2.69667566e-01 3.29538077e-01 6.89427376e-01 4.00628656e-01 6.64704919e-01 3.88813585e-01 1.07835674e+00 5.55150390e-01 -5.96590936e-01 3.42700705e-02 -8.86408240e-02 -5.37884712e-01 9.27617192e-01 1.85408592e+00 4.08272177e-01 1.30014159e-02 -1.14172649e+00 9.97951984e-01 -1.56424558e+00 -6.70751393e-01 2.43105695e-01 1.93986309e+00 1.24373078e+00 -1.58560336e-01 -1.12908460e-01 -5.53138196e-01 4.97707933e-01 2.05848441e-02 -7.27465212e-01 -4.66795325e-01 -2.88700134e-01 2.92814493e-01 4.26587582e-01 6.61328614e-01 -1.20635462e+00 1.62562692e+00 5.18790388e+00 8.41736078e-01 -1.25165761e+00 5.72151244e-01 2.53748864e-01 -1.08056441e-02 -3.01862419e-01 -3.22526544e-02 -1.22147250e+00 2.18819022e-01 1.14692235e+00 -8.90488565e-01 3.84577364e-01 9.02491391e-01 6.81136474e-02 3.26122969e-01 -1.09444714e+00 8.74889910e-01 2.12219507e-01 -8.75400186e-01 2.36848563e-01 3.47786024e-02 9.49934244e-01 8.99972022e-01 2.29598403e-01 8.49174798e-01 8.08647692e-01 -4.91376519e-01 4.33406323e-01 -2.62769368e-02 1.02820873e+00 -7.16937423e-01 5.74003577e-01 4.91843253e-01 -1.63398552e+00 4.22727376e-01 -5.30125141e-01 1.77688599e-01 7.07765147e-02 5.47502220e-01 -9.95241284e-01 7.10611522e-01 5.33671319e-01 1.23076272e+00 -4.35578704e-01 2.83301864e-02 -2.27422953e-01 7.93368220e-01 -3.06939751e-01 4.30697918e-01 4.78180856e-01 -1.22620687e-01 3.05650294e-01 1.64184523e+00 4.25389439e-01 -2.79751182e-01 5.51468611e-01 5.36024988e-01 -3.85467291e-01 6.16159022e-01 -8.13477159e-01 -2.78768331e-01 5.20385444e-01 1.17095852e+00 1.38633894e-02 -5.14861107e-01 -9.19997990e-01 1.03783715e+00 7.34185815e-01 4.37401861e-01 -4.31613445e-01 -6.31067306e-02 8.70803177e-01 -3.24906290e-01 5.01106918e-01 -5.54677904e-01 1.52668152e-02 -1.92056060e+00 2.53128633e-02 -8.97059083e-01 6.01707816e-01 -1.31584793e-01 -2.06288910e+00 9.06378627e-01 -1.70805588e-01 -1.27672684e+00 -5.80117643e-01 -6.72956526e-01 7.36514619e-03 1.19062328e+00 -2.06016397e+00 -1.54814899e+00 4.61273521e-01 1.07456744e+00 9.61701751e-01 -7.34792769e-01 1.01540828e+00 3.64777446e-01 -4.97748345e-01 9.33968008e-01 7.29455471e-01 4.52222019e-01 1.08720207e+00 -8.71580005e-01 5.34002244e-01 1.06075168e+00 3.57128203e-01 9.55623150e-01 1.68093249e-01 -5.73167861e-01 -1.67497325e+00 -1.60215628e+00 1.43940163e+00 -4.72857654e-01 1.16084754e+00 -5.54144382e-01 -1.13098073e+00 1.20884418e+00 2.24508911e-01 3.57128412e-01 1.14385974e+00 4.18004066e-01 -8.49487782e-01 4.81032860e-03 -8.48911583e-01 2.56548226e-01 8.45069051e-01 -1.14100039e+00 -7.66521156e-01 4.26152915e-01 5.90367317e-01 -3.70477065e-02 -9.97262895e-01 4.42913771e-02 3.75595272e-01 -3.84192020e-01 7.64155865e-01 -7.00420022e-01 6.86629489e-02 -1.02035202e-01 -5.41523576e-01 -1.48701251e+00 -4.11251783e-01 -4.53602552e-01 1.81144625e-01 1.27000940e+00 6.26931131e-01 -9.07556832e-01 1.33555338e-01 1.59433559e-01 -9.83581319e-02 -2.97728598e-01 -1.11545289e+00 -1.18725014e+00 8.25655341e-01 -6.42534435e-01 3.63406450e-01 1.44154537e+00 3.31533879e-01 6.39037848e-01 -2.79453665e-01 2.88753897e-01 5.73384762e-01 2.47445077e-01 7.30055571e-01 -9.66313481e-01 -3.57759714e-01 -6.64094090e-02 6.76788157e-03 -1.27257264e+00 1.00083363e+00 -1.68335319e+00 -3.13430876e-02 -1.00588405e+00 2.33672172e-01 -6.65688634e-01 -5.51082253e-01 6.78830981e-01 -1.92908004e-01 -5.01543190e-03 1.39381260e-01 3.90871406e-01 -6.42459989e-01 6.61251903e-01 8.57716203e-01 -1.31188989e-01 -2.70005941e-01 -2.49485984e-01 -8.77376974e-01 6.88499629e-01 8.34378600e-01 -7.12628722e-01 -2.73785472e-01 -8.88997972e-01 -1.79106351e-02 -3.89517397e-01 -4.16885108e-01 -6.17281556e-01 5.88654988e-02 -2.12286860e-01 1.51526093e-01 -2.95272589e-01 1.73789188e-01 -7.76683331e-01 -3.72159392e-01 1.41676396e-01 -3.36358219e-01 1.00550078e-01 5.32699764e-01 4.69555378e-01 -5.10602534e-01 -2.65396982e-02 6.38123214e-01 -1.94695532e-01 -8.73793900e-01 3.51898193e-01 -4.87034380e-01 3.30286652e-01 7.54892886e-01 4.29337829e-01 -1.02135494e-01 -9.05667171e-02 -4.45707768e-01 1.78931966e-01 4.78517175e-01 8.21549952e-01 2.28789389e-01 -1.58269334e+00 -9.81946409e-01 4.73493785e-01 4.28061634e-01 -2.41687477e-01 1.22930810e-01 5.88837147e-01 1.88412786e-01 7.35039115e-01 1.43878683e-01 -5.97398520e-01 -8.79069388e-01 5.12817085e-01 2.02120051e-01 -8.22674930e-01 -2.73792535e-01 8.40209723e-01 5.22278786e-01 -1.23854375e+00 -4.05147523e-01 -2.95854926e-01 1.25078678e-01 2.31169388e-02 3.40054899e-01 -2.55197912e-01 8.71164277e-02 -1.19989967e+00 -5.49310029e-01 6.07163131e-01 -5.58069646e-01 -2.60736287e-01 1.35749912e+00 -5.55868924e-01 6.48381561e-02 8.78488779e-01 1.58761275e+00 2.18565866e-01 -9.08433259e-01 -1.27426040e+00 1.51634514e-01 -2.61768043e-01 2.26474628e-01 -7.49606609e-01 -7.95962214e-01 1.02829158e+00 2.70046890e-01 -3.86386961e-01 9.89142120e-01 2.42963448e-01 9.64437664e-01 4.85032976e-01 6.56484902e-01 -1.29005730e+00 -1.33661553e-01 1.06614006e+00 5.95517457e-01 -1.56234205e+00 -7.16979057e-02 -2.20618099e-02 -8.57289135e-01 1.09733498e+00 5.74411452e-01 -1.05930336e-01 9.57983851e-01 2.54017115e-01 3.56768489e-01 9.28801894e-02 -9.65341151e-01 -2.02224031e-01 5.50149322e-01 4.46020067e-01 9.34009016e-01 5.40202022e-01 4.27156053e-02 8.73619676e-01 -2.65458047e-01 -1.23005189e-01 1.20446168e-01 8.13016534e-01 -2.08787873e-01 -1.68822455e+00 -2.35178798e-01 -1.58663262e-02 -2.97506005e-01 -4.93912339e-01 -2.04126686e-01 7.76588500e-01 1.28880560e-01 7.78392196e-01 -1.93191674e-02 -1.91125050e-01 7.24532977e-02 6.28318191e-01 4.26477820e-01 -1.02444375e+00 -3.90931696e-01 2.80016631e-01 -7.90094584e-02 -2.73129970e-01 -5.41539371e-01 -1.07184410e+00 -1.26938224e+00 -3.14659700e-02 -8.68743956e-02 1.98686793e-01 6.24248564e-01 1.20260799e+00 5.72386026e-01 9.20270979e-02 7.76492238e-01 -6.09968305e-01 -4.90166843e-01 -1.15140796e+00 -4.26028669e-01 3.18447560e-01 2.95295835e-01 -6.51345372e-01 -3.59661579e-01 3.40715289e-01]

[10.959637641906738, 9.88442611694336]

2cb897df-02c6-4f32-922a-65971ad5c2ac

spelling-correction-with-denoising

2105.05977

https://arxiv.org/abs/2105.05977v1

https://arxiv.org/pdf/2105.05977v1.pdf

Spelling Correction with Denoising Transformer

We present a novel method of performing spelling correction on short input strings, such as search queries or individual words. At its core lies a procedure for generating artificial typos which closely follow the error patterns manifested by humans. This procedure is used to train the production spelling correction model based on a transformer architecture. This model is currently served in the HubSpot product search. We show that our approach to typo generation is superior to the widespread practice of adding noise, which ignores human patterns. We also demonstrate how our approach may be extended to resource-scarce settings and train spelling correction models for Arabic, Greek, Russian, and Setswana languages, without using any labeled data.

['Hector Urdiales', 'Alex Kuznetsov']

2021-05-12

['spelling-correction', 'bangla-spelling-error-correction']

['natural-language-processing', 'natural-language-processing']

[ 5.42043209e-01 -4.40110080e-02 3.10462080e-02 -1.55237122e-02 -7.17714369e-01 -9.51697528e-01 8.10052216e-01 6.88077137e-02 -4.81329262e-01 8.15808475e-01 1.08224958e-01 -7.13788152e-01 1.24426343e-01 -8.46340716e-01 -8.43347192e-01 -2.05439314e-01 7.62005746e-01 1.03766096e+00 1.12975143e-01 -8.80376935e-01 6.37120306e-01 4.83275294e-01 -1.43798959e+00 4.33460802e-01 1.17853975e+00 2.43497431e-01 2.79690593e-01 6.98601961e-01 -2.66959280e-01 5.82202792e-01 -1.02164388e+00 -7.58366168e-01 1.54173851e-01 -6.06297612e-01 -8.67789865e-01 -2.25287452e-01 5.03816664e-01 -6.21134089e-03 -1.14737488e-01 1.11204708e+00 4.27867234e-01 1.00147530e-01 7.24503934e-01 -8.63677740e-01 -1.16089833e+00 9.03221309e-01 8.10984802e-03 1.63755268e-01 6.90035820e-01 2.62464702e-01 1.02849686e+00 -9.99740303e-01 8.80221128e-01 1.08167148e+00 7.75188088e-01 6.76004052e-01 -1.19319999e+00 -4.69798148e-01 -2.90723026e-01 1.31473288e-01 -1.33575547e+00 -2.15970397e-01 3.96919996e-01 -2.39814311e-01 1.28133082e+00 3.21466953e-01 4.22926664e-01 1.46850479e+00 1.39084071e-01 6.93044782e-01 1.02645445e+00 -1.21309090e+00 5.24807815e-03 3.14653099e-01 -3.05542573e-02 5.93477190e-01 4.15986240e-01 3.68007630e-01 -6.98384047e-01 -1.79344147e-01 5.69757640e-01 -3.98384303e-01 -4.28689681e-02 1.10321283e-01 -1.25594473e+00 7.72641778e-01 -4.02261950e-02 4.12849247e-01 -1.95475966e-01 1.70900628e-01 1.84092134e-01 5.71373045e-01 3.49356204e-01 1.13141191e+00 -6.10521197e-01 -3.19598317e-01 -9.72974420e-01 6.25696361e-01 1.01193535e+00 1.45619380e+00 5.38703024e-01 1.59337610e-01 -2.99824685e-01 9.72392499e-01 1.48437675e-02 8.71860743e-01 8.52411807e-01 -6.70381129e-01 4.25028503e-01 4.76419955e-01 4.80420411e-01 -7.63309002e-01 -7.67671689e-02 -2.74669975e-01 -2.28598967e-01 3.05433362e-03 5.50389647e-01 4.75609954e-03 -1.24050426e+00 1.40552294e+00 -3.26000780e-01 -2.99790293e-01 -9.50001851e-02 5.56701422e-01 3.69467050e-01 5.20395637e-01 4.23290022e-02 -5.18356487e-02 1.03920889e+00 -1.06502545e+00 -7.52482474e-01 -1.80952534e-01 1.04395175e+00 -1.19933653e+00 1.68464708e+00 6.46416485e-01 -1.31672800e+00 -3.95130336e-01 -8.32534552e-01 -3.10065988e-02 -8.77860427e-01 8.26398879e-02 4.10595566e-01 9.01461720e-01 -9.76488888e-01 8.75417113e-01 -2.74687976e-01 -5.79600334e-01 -5.08491695e-02 2.08858117e-01 1.56959146e-02 9.52721015e-02 -1.41480076e+00 1.33751404e+00 5.18282294e-01 -7.91026205e-02 -5.09187520e-01 -7.15645015e-01 -6.19773746e-01 -3.78124863e-02 1.43057331e-01 -6.40163302e-01 1.72287071e+00 -1.18774831e+00 -1.60324693e+00 9.25525069e-01 -1.82166174e-01 -3.07398766e-01 4.52046156e-01 -1.11563504e-01 -7.47157454e-01 -3.28050226e-01 9.68069509e-02 3.91494155e-01 6.69391990e-01 -1.06795526e+00 -6.18268549e-01 7.76001588e-02 -4.34365362e-01 2.12760448e-01 -1.88259497e-01 4.25366253e-01 -2.35027507e-01 -1.18068194e+00 -1.70060992e-01 -1.05537939e+00 -2.65517563e-01 -7.71122634e-01 -4.56628323e-01 -2.89081275e-01 9.99788195e-02 -8.10808122e-01 1.57206488e+00 -1.66875565e+00 1.35539740e-01 7.53329635e-01 -3.20175558e-01 6.02136374e-01 -4.74412352e-01 7.50500023e-01 -5.98481968e-02 4.16641712e-01 -1.21825397e-01 3.66855748e-02 2.76918113e-01 7.75326863e-02 -5.93039989e-01 -2.58333772e-01 8.55514035e-02 1.25964653e+00 -1.15379333e+00 -1.66063011e-01 -1.49334028e-01 -1.62867114e-01 -7.35797882e-01 5.68477772e-02 -6.76017225e-01 -8.73579830e-02 6.12286776e-02 7.84129262e-01 3.39911401e-01 -1.33822545e-01 2.34454498e-01 4.20243979e-01 -1.12377457e-01 7.28706241e-01 -9.91327763e-01 1.51734889e+00 -5.73804200e-01 3.80514055e-01 -6.68897688e-01 -3.57062608e-01 8.61626565e-01 3.89837511e-02 -2.44878247e-01 -9.85027850e-01 -1.11967608e-01 9.48368967e-01 -9.73632634e-02 -4.60598648e-01 8.81742477e-01 -1.44763924e-02 -3.63866448e-01 7.40370035e-01 1.64199069e-01 -3.92876893e-01 6.03642404e-01 1.68670893e-01 1.02292073e+00 2.48772621e-01 2.06907153e-01 -2.35808924e-01 4.50309426e-01 4.27642971e-01 2.05588147e-01 1.22332680e+00 3.65307838e-01 5.44695556e-01 2.63702750e-01 -3.96419495e-01 -1.41847801e+00 -9.17191923e-01 3.53705399e-02 1.34124541e+00 -2.03427702e-01 -7.49764264e-01 -9.38687623e-01 -7.81569779e-01 1.04524709e-01 1.23967028e+00 -4.38957870e-01 -2.26884514e-01 -9.41779733e-01 -5.26078522e-01 9.61395025e-01 4.50248778e-01 8.97755548e-02 -1.79188359e+00 -1.68197230e-01 3.60397190e-01 -3.20385426e-01 -5.77488601e-01 -8.44374359e-01 2.18504235e-01 -5.72791159e-01 -7.59080589e-01 -7.51896262e-01 -1.06508398e+00 5.44460595e-01 -6.74479827e-02 1.34920335e+00 3.98729920e-01 -7.43357390e-02 3.13104875e-02 -5.19998193e-01 -6.76775455e-01 -1.10345721e+00 5.65773070e-01 9.39408839e-02 -6.69345200e-01 8.67288172e-01 -2.41466150e-01 -1.50410712e-01 2.03868330e-01 -9.70994532e-01 -1.05606504e-01 5.51339447e-01 1.03260231e+00 2.89577365e-01 -4.48958725e-01 4.24054831e-01 -1.42846441e+00 1.00752187e+00 -3.68908972e-01 -7.28485823e-01 5.66584527e-01 -1.01534057e+00 1.79754868e-01 7.21924603e-01 -4.52046275e-01 -1.00881910e+00 -8.48489925e-02 -2.65156835e-01 -1.41954403e-02 -1.73736408e-01 4.76164639e-01 -3.56476195e-02 -2.36258954e-01 1.14528954e+00 3.17894459e-01 -2.22951397e-01 -7.47033179e-01 4.68627393e-01 9.08821166e-01 5.02336502e-01 -4.22860891e-01 9.02298748e-01 -2.60205895e-01 -2.75640160e-01 -6.12244487e-01 -5.13394415e-01 -1.82468459e-01 -4.02598441e-01 1.69841498e-01 2.05329701e-01 -3.38668942e-01 -4.32390004e-01 6.90933943e-01 -1.34801388e+00 -5.19119918e-01 -5.05306780e-01 9.49580669e-02 -7.27397919e-01 4.26359624e-01 -7.10892498e-01 -3.69353384e-01 -3.28604370e-01 -6.71148658e-01 9.40290749e-01 1.43605262e-01 -6.55852556e-01 -9.10645187e-01 3.31850648e-01 2.82045007e-02 4.54831898e-01 -4.35740411e-01 1.38630915e+00 -1.09827745e+00 -4.30956811e-01 -3.83595414e-02 1.15952753e-01 2.69861400e-01 3.16886418e-02 8.84969905e-02 -5.40257037e-01 9.85344946e-02 -3.81823391e-01 -2.04869330e-01 6.23340428e-01 -1.60920635e-01 9.69876587e-01 -5.16523540e-01 -2.14389026e-01 6.89084053e-01 1.24097300e+00 3.02323043e-01 7.47159421e-01 6.28606141e-01 6.37825787e-01 6.22719109e-01 5.99787652e-01 8.80775526e-02 9.91053283e-02 7.43657112e-01 -1.24844328e-01 1.63208380e-01 -2.90178359e-01 -9.01813745e-01 3.11520010e-01 9.24018621e-01 1.65872443e-02 -6.50421560e-01 -1.14706182e+00 6.52347147e-01 -1.69793320e+00 -1.03854060e+00 -1.87217698e-01 2.04193902e+00 1.13214064e+00 6.97131753e-02 7.06742853e-02 1.82040736e-01 5.34548700e-01 -3.30057144e-01 -8.23863223e-02 -1.03030360e+00 -4.00186926e-01 7.44605720e-01 7.88898528e-01 6.73219204e-01 -7.07738698e-01 1.68246543e+00 7.92863989e+00 9.19255495e-01 -7.38026321e-01 -5.50745018e-02 1.54856846e-01 1.18468985e-01 -8.36656034e-01 -3.06377951e-02 -7.92816639e-01 7.59039581e-01 1.13406622e+00 -3.22653294e-01 9.97977972e-01 6.64035916e-01 4.69403714e-02 6.25716373e-02 -1.11284351e+00 7.47950554e-01 3.46844018e-01 -1.46021509e+00 5.03376722e-01 -2.29796022e-01 9.60715652e-01 -2.55461544e-01 7.09664971e-02 4.06724155e-01 7.82693744e-01 -1.17369699e+00 8.51371944e-01 5.25343657e-01 8.44828546e-01 -7.72515893e-01 4.59383249e-01 3.69137377e-01 -3.10293823e-01 1.73487924e-02 -2.68086076e-01 1.53766051e-01 -1.43253552e-02 4.33318578e-02 -1.08033550e+00 1.81117579e-01 3.83002341e-01 2.27915898e-01 -8.01274657e-01 9.63216722e-01 -5.89267194e-01 7.35369086e-01 -2.30059370e-01 -5.27927876e-01 1.41254574e-01 -1.66635334e-01 5.49884737e-01 1.64249980e+00 6.26032531e-01 -1.75855756e-01 -3.35097462e-01 7.97754347e-01 -2.17971638e-01 4.31415319e-01 -8.40799391e-01 -2.44986638e-01 6.69408619e-01 6.38541281e-01 -5.15653253e-01 -4.31669831e-01 -1.18540704e-01 1.36395192e+00 1.64452150e-01 5.38170099e-01 -5.57327032e-01 -8.81507218e-01 2.95384288e-01 1.87973350e-01 3.91743302e-01 1.94200650e-02 -5.79243362e-01 -1.11960554e+00 4.33028601e-02 -1.59409523e+00 2.58024424e-01 -9.32530761e-01 -1.22471690e+00 5.84548414e-01 -1.27923876e-01 -1.02254736e+00 -7.29815543e-01 -8.78612399e-01 -5.97334445e-01 1.26940620e+00 -1.32858527e+00 -1.08695054e+00 2.29442213e-03 3.54789615e-01 6.44710839e-01 -5.45082152e-01 1.02009165e+00 2.27767974e-01 -2.84494370e-01 1.04633462e+00 2.76180863e-01 -1.04351237e-01 9.28104877e-01 -1.41196954e+00 9.40385282e-01 9.07568753e-01 4.64936703e-01 9.75856960e-01 9.85032737e-01 -9.48778152e-01 -9.23688769e-01 -9.75376368e-01 1.89402723e+00 -7.57914722e-01 7.40056872e-01 -4.93526489e-01 -7.72423387e-01 7.87774801e-01 4.33316410e-01 -6.70113921e-01 5.67110479e-01 1.57118037e-01 -3.35524380e-01 2.25692481e-01 -9.26512480e-01 9.63740230e-01 1.28814137e+00 -4.71194595e-01 -8.94984543e-01 7.74460912e-01 6.73629045e-01 -5.26205659e-01 -2.72811472e-01 -3.46294045e-02 3.72836024e-01 -3.80159765e-01 5.93930244e-01 -1.17335451e+00 5.47162294e-01 -2.51052320e-01 7.21530095e-02 -1.86964774e+00 -5.06592572e-01 -1.20025241e+00 6.13131411e-02 1.14195907e+00 9.23971832e-01 -6.03437603e-01 6.29413426e-01 2.93323010e-01 -1.12752035e-01 -1.50413007e-01 -5.09771347e-01 -9.67356622e-01 5.80553710e-01 -1.74394786e-01 9.58567917e-01 7.88198590e-01 2.06379876e-01 1.84292555e-01 -3.43995899e-01 -2.65520155e-01 2.60091782e-01 -9.54061225e-02 5.39354622e-01 -9.22676325e-01 -3.70621324e-01 -5.39234936e-01 1.09504335e-01 -8.74553740e-01 1.97420970e-01 -1.13216054e+00 2.54694998e-01 -9.94437695e-01 -2.39098847e-01 -5.07857442e-01 -1.86212212e-01 2.40039155e-01 -2.65518069e-01 4.91377831e-01 2.80432642e-01 2.31330574e-01 -2.99533218e-01 2.30426013e-01 9.50151861e-01 2.55702212e-02 -2.11663604e-01 1.28917526e-02 -7.70487785e-01 6.50441408e-01 9.09309447e-01 -7.29830623e-01 -1.26374811e-01 -6.20456815e-01 8.68013501e-01 -3.71234715e-01 1.19534582e-01 -7.46273041e-01 4.01470095e-01 -3.06804955e-01 1.24079630e-01 -3.15138608e-01 -2.97716647e-01 -4.57222939e-01 1.39802828e-01 3.99407178e-01 -6.50819063e-01 7.39484251e-01 1.13359824e-01 1.85265675e-01 6.89375252e-02 -8.01928222e-01 4.40490067e-01 -2.83869594e-01 -5.70056319e-01 -2.02608630e-01 -7.58188128e-01 2.35059217e-01 6.50482118e-01 -2.74143100e-01 -3.52575660e-01 -1.00216679e-01 -5.18454194e-01 -2.03061216e-02 7.46806324e-01 4.71717149e-01 3.64590675e-01 -1.40516293e+00 -6.79379523e-01 5.10314703e-01 3.00723821e-01 -6.38470232e-01 -4.43489701e-01 2.43749022e-01 -1.11771286e+00 7.62127101e-01 -2.53130496e-01 5.91790266e-02 -9.56757665e-01 7.19801962e-01 3.16288590e-01 -3.38708937e-01 -2.38488153e-01 7.95485854e-01 -5.00645816e-01 -5.96804798e-01 1.76724046e-01 -2.49870881e-01 -2.70259790e-02 -1.18226498e-01 4.52629477e-01 3.19898874e-01 5.36912203e-01 -3.67533177e-01 -1.56268194e-01 2.82546610e-01 -2.83692032e-01 -4.61356431e-01 9.59663033e-01 5.25399633e-02 -4.49976288e-02 3.13243955e-01 6.62025809e-01 3.99484843e-01 -3.32102597e-01 -1.54431537e-01 4.45753932e-01 -4.40761030e-01 -5.36544979e-01 -1.31710255e+00 -3.61778647e-01 5.80671370e-01 2.33614355e-01 9.08892825e-02 8.66393387e-01 -3.44562024e-01 9.75413918e-01 8.77132773e-01 4.74751979e-01 -1.68808067e+00 -5.33312857e-02 9.77597296e-01 8.83280277e-01 -8.52463841e-01 -5.66494942e-01 -3.03975284e-01 -6.44577444e-01 1.08455575e+00 4.26099122e-01 -1.81329563e-01 1.42885476e-01 2.80540407e-01 2.63500392e-01 3.78253609e-02 -7.99340069e-01 -2.00434402e-01 7.14559779e-02 8.08017433e-01 4.94697630e-01 -5.81717379e-02 -7.60954976e-01 5.49880683e-01 -6.32477462e-01 9.36826393e-02 7.44179010e-01 8.88581693e-01 -2.11505696e-01 -1.65566313e+00 -2.93896019e-01 5.00176847e-01 -5.09010911e-01 -8.18374753e-01 -7.79541612e-01 7.37937212e-01 2.05264911e-01 7.23464429e-01 -7.75925964e-02 -3.72526586e-01 6.87105238e-01 4.70628530e-01 5.26421249e-01 -8.38733912e-01 -1.23972559e+00 -3.51066411e-01 3.02886128e-01 -4.74043638e-01 3.18581462e-01 -8.24168146e-01 -7.47381866e-01 -6.10994577e-01 -2.16743991e-01 3.08879703e-01 5.08905172e-01 6.48867667e-01 2.37823382e-01 1.09913848e-01 4.61510599e-01 -3.68901610e-01 -9.53223825e-01 -1.05754268e+00 -3.54286879e-01 7.91190624e-01 -6.80503319e-04 -1.76219076e-01 -3.23386192e-01 1.83508605e-01]

[11.17873477935791, 10.112317085266113]

1473d47a-b6ce-40c1-be48-3ae6418ba4cd

blind-extraction-of-target-speech-source

2111.03482

https://arxiv.org/abs/2111.03482v2

https://arxiv.org/pdf/2111.03482v2.pdf

Target Speech Extraction: Independent Vector Extraction Guided by Supervised Speaker Identification

This manuscript proposes a novel robust procedure for the extraction of a speaker of interest (SOI) from a mixture of audio sources. The estimation of the SOI is performed via independent vector extraction (IVE). Since the blind IVE cannot distinguish the target source by itself, it is guided towards the SOI via frame-wise speaker identification based on deep learning. Still, an incorrect speaker can be extracted due to guidance failings, especially when processing challenging data. To identify such cases, we propose a criterion for non-intrusively assessing the estimated speaker. It utilizes the same model as the speaker identification, so no additional training is required. When incorrect extraction is detected, we propose a ``deflation'' step in which the incorrect source is subtracted from the mixture and, subsequently, another attempt to extract the SOI is performed. The process is repeated until successful extraction is achieved. The proposed procedure is experimentally tested on artificial and real-world datasets containing challenging phenomena: source movements, reverberation, transient noise, or microphone failures. The method is compared with state-of-the-art blind algorithms as well as with current fully supervised deep learning-based methods.

['Jindrich Zdansky', 'Jaroslav Cmejla', 'Tomas Kounovsky', 'Zbynek Koldovsky', 'Jakub Jansky', 'Jiri Malek']

2021-11-05

['speech-extraction', 'speaker-identification']

['speech', 'speech']

[ 3.71167153e-01 -2.54637539e-01 5.53395748e-01 -6.31290302e-02 -1.19132853e+00 -7.12063313e-01 4.13744837e-01 1.72124624e-01 -4.64871705e-01 5.96909642e-01 1.60693318e-01 -2.73242921e-01 -9.02453810e-02 -1.05765492e-01 -5.70527256e-01 -1.10229266e+00 -6.54030815e-02 3.58700097e-01 1.21892663e-02 2.32510060e-01 3.93788546e-01 6.04633391e-01 -1.69218564e+00 -7.49424249e-02 5.79196632e-01 9.82670426e-01 3.95701341e-02 8.70434999e-01 -1.22566581e-01 5.64077914e-01 -1.18263400e+00 8.01037252e-02 3.68047915e-02 -4.55247521e-01 -5.60487747e-01 2.64821291e-01 2.02712685e-01 -3.77023906e-01 2.99453020e-01 9.41263914e-01 9.33236778e-01 1.64665669e-01 8.06665599e-01 -1.02840555e+00 2.21776947e-01 4.71264571e-01 -4.08177435e-01 2.85695165e-01 4.90028292e-01 -1.88383445e-01 6.92300737e-01 -1.36658335e+00 -3.83753367e-02 9.54793274e-01 4.93379980e-01 5.03047884e-01 -1.11967671e+00 -7.75128484e-01 -7.04207346e-02 2.17373237e-01 -1.54273999e+00 -1.19211960e+00 1.11698925e+00 -5.13396502e-01 5.69165170e-01 3.14935386e-01 2.69529134e-01 1.12729323e+00 -2.81418025e-01 5.87973654e-01 7.14681268e-01 -6.12291038e-01 6.13317430e-01 2.87587523e-01 1.12870336e-01 3.80510479e-01 7.94630125e-03 1.91485435e-01 -7.76359499e-01 -3.28709692e-01 2.50352532e-01 -5.53339124e-01 -6.43215656e-01 -9.27845910e-02 -1.10327947e+00 5.11069655e-01 2.83586364e-02 6.54074967e-01 -6.98303044e-01 -3.28402609e-01 2.48691276e-01 1.70444235e-01 4.80206281e-01 1.89225048e-01 -1.02855794e-01 -1.66249573e-02 -1.51224017e+00 -5.00956587e-02 8.91583383e-01 3.72273177e-01 4.01664227e-01 4.76729870e-01 1.25082940e-01 9.71083581e-01 5.75962782e-01 3.89247835e-01 6.04465544e-01 -5.22817016e-01 3.75358254e-01 2.70044565e-01 6.46538854e-01 -8.17403316e-01 -3.44672799e-01 -6.97192252e-01 -8.09650362e-01 4.76072967e-01 4.19718236e-01 -3.66864622e-01 -9.30968225e-01 1.44929361e+00 4.96089697e-01 3.85293335e-01 2.13463828e-01 1.07905102e+00 9.01630938e-01 6.20630324e-01 -2.17241094e-01 -5.55277050e-01 1.12027633e+00 -5.08486986e-01 -9.53769088e-01 -3.63018990e-01 -8.39300528e-02 -1.08861995e+00 3.31639379e-01 7.06191659e-01 -9.87645507e-01 -7.31461346e-01 -1.23042166e+00 6.38729334e-01 -9.47003737e-02 3.73382866e-01 -7.30161369e-02 7.87600219e-01 -9.45132852e-01 1.88565627e-01 -6.71972334e-01 -8.23956951e-02 2.17720447e-03 4.02731627e-01 -4.19815719e-01 3.59935760e-01 -9.77868199e-01 5.93519807e-01 1.22825608e-01 5.73975801e-01 -1.20308208e+00 -3.39467794e-01 -8.55993330e-01 2.93075666e-02 1.15087077e-01 -3.81357312e-01 1.13614380e+00 -1.20173073e+00 -1.71664369e+00 6.28284633e-01 -6.71887875e-01 -3.42648506e-01 5.88565886e-01 -3.01699340e-01 -6.61264122e-01 2.81251341e-01 6.25407994e-02 -6.61213845e-02 1.85448146e+00 -1.54515696e+00 -5.54851174e-01 -4.38473165e-01 -3.81947547e-01 1.19049013e-01 -3.31726134e-01 3.86286318e-01 -2.26804614e-01 -6.07962489e-01 5.96540451e-01 -7.42767692e-01 3.87209088e-01 -3.43705237e-01 -7.46027052e-01 -1.38121232e-01 7.79139936e-01 -1.01171637e+00 1.12897086e+00 -2.51050520e+00 2.08317950e-01 3.52516294e-01 2.29594499e-01 4.42037046e-01 8.79237726e-02 2.62892306e-01 -3.07156593e-01 -1.52484193e-01 -4.96101379e-01 -8.56575906e-01 -2.19027385e-01 -4.54736084e-01 -2.80459315e-01 7.44941831e-01 9.07085910e-02 2.07792539e-02 -4.84422386e-01 -4.18003559e-01 2.85721421e-01 6.41467035e-01 -1.04538679e-01 6.41339004e-01 3.89960349e-01 8.12621057e-01 1.49159264e-02 5.87920070e-01 7.52257288e-01 1.53341308e-01 -1.42032921e-01 -2.18851909e-01 -1.74405828e-01 4.51373518e-01 -1.72514403e+00 1.22873378e+00 -5.57244420e-01 7.53704548e-01 7.92677164e-01 -1.26647031e+00 9.63894486e-01 9.18292046e-01 1.67449474e-01 9.56896767e-02 1.99627295e-01 5.91822028e-01 6.19554818e-02 -5.03458321e-01 1.27633706e-01 -1.72810301e-01 2.93404341e-01 4.31779921e-01 2.74204671e-01 1.65642127e-01 -8.08386728e-02 -3.93791758e-02 9.18144405e-01 -1.48660168e-01 2.61349767e-01 -8.78527015e-02 1.02055621e+00 -6.51106656e-01 4.15747583e-01 7.32267857e-01 -4.09247249e-01 5.64471960e-01 -7.27724805e-02 1.06813110e-01 -4.41026956e-01 -1.19327497e+00 -2.18958780e-02 8.56144965e-01 -8.00767764e-02 3.54698040e-02 -8.72573495e-01 -5.37480593e-01 -2.07432047e-01 6.61472619e-01 -3.89497817e-01 8.52705985e-02 -5.95712602e-01 -5.89923263e-01 4.82400835e-01 1.53330609e-01 4.43561912e-01 -1.10409164e+00 -6.14428997e-01 5.30202389e-01 -5.40322661e-01 -9.05240357e-01 -2.74364382e-01 3.37648094e-01 -5.07164896e-01 -9.18226659e-01 -9.48567092e-01 -7.89800167e-01 6.29075110e-01 2.88810015e-01 6.64601803e-01 -1.26917586e-01 1.10452697e-02 3.42078507e-01 -1.11551598e-01 -4.30862546e-01 -8.81030321e-01 -3.70089114e-01 4.09035683e-01 7.15394497e-01 2.92316079e-01 -5.40146112e-01 -5.70539773e-01 4.04201776e-01 -5.46948075e-01 -5.19163668e-01 2.43855163e-01 7.34495163e-01 1.32678494e-01 6.14422977e-01 9.47519243e-01 -7.99162537e-02 6.94479346e-01 -2.99320698e-01 -6.79677606e-01 -1.43653005e-01 -4.02077436e-02 -9.53045189e-02 6.97737992e-01 -5.98300695e-01 -1.22826326e+00 3.72340903e-02 -1.48594722e-01 -4.00352746e-01 -7.69393027e-01 4.12295908e-01 -3.85625601e-01 4.88029234e-02 6.09383464e-01 5.22265494e-01 -2.49544814e-01 -6.99059725e-01 -1.66052938e-01 1.23953533e+00 6.19405925e-01 7.22641498e-02 9.09988165e-01 2.82010168e-01 -4.85177308e-01 -1.15954518e+00 -4.13374245e-01 -7.16224730e-01 -4.88239706e-01 -4.94615197e-01 6.28506243e-01 -7.96397686e-01 -5.63833296e-01 9.13166106e-01 -1.44411361e+00 3.58469784e-01 2.06784025e-01 8.55167687e-01 -2.45748505e-01 3.33836377e-01 -2.66826183e-01 -1.54064977e+00 -4.98899579e-01 -1.21757925e+00 9.21555042e-01 7.65209049e-02 -4.58615512e-01 -6.69085801e-01 -7.93548226e-02 5.02828538e-01 4.43024039e-01 -1.63033113e-01 5.71085274e-01 -1.05468738e+00 -1.49702281e-01 -4.73218709e-01 2.10985333e-01 5.78877687e-01 5.48653543e-01 -1.10841446e-01 -1.48998356e+00 -4.10779148e-01 5.88187695e-01 7.91655481e-02 5.88650286e-01 4.27452207e-01 5.78230262e-01 -3.83788168e-01 -2.12264881e-01 2.88931698e-01 1.13437045e+00 4.44372237e-01 1.15091935e-01 1.37961298e-01 4.99351770e-01 7.04459131e-01 1.97695717e-01 4.54311758e-01 -3.35255787e-02 5.94540536e-01 4.78605062e-01 4.97819893e-02 -1.05880432e-01 1.17603339e-01 5.16929388e-01 8.48468304e-01 1.46550521e-01 -4.35025960e-01 -8.33231270e-01 7.88131833e-01 -1.24796617e+00 -7.94716179e-01 5.80628179e-02 2.63360739e+00 6.02291167e-01 2.99909443e-01 1.37300834e-01 1.12628794e+00 1.09591138e+00 1.24948651e-01 -5.00199258e-01 -9.40937102e-02 7.79495835e-02 1.14717178e-01 1.21300474e-01 8.05069327e-01 -1.18784773e+00 4.84329790e-01 5.38124323e+00 6.22263551e-01 -1.42196250e+00 1.00861147e-01 1.91648006e-01 -6.25718161e-02 1.72631830e-01 -4.06372219e-01 -7.54917026e-01 4.67350841e-01 9.50689733e-01 2.44747683e-01 4.05637056e-01 3.60081851e-01 4.51177835e-01 -1.88075796e-01 -1.14328969e+00 1.23669636e+00 3.77832055e-01 -7.44878411e-01 -3.13639730e-01 -1.99449539e-01 6.25325367e-02 -3.07260573e-01 -1.10277034e-01 -1.02009572e-01 -1.82437837e-01 -6.47988021e-01 9.88827229e-01 4.13464636e-01 3.98329914e-01 -8.54408979e-01 7.44554818e-01 6.05446041e-01 -1.08110988e+00 -2.92716563e-01 2.00291201e-01 1.89573884e-01 2.13630065e-01 8.07882786e-01 -9.78029132e-01 4.72968787e-01 6.49105191e-01 2.23967910e-01 -2.74408370e-01 1.26179719e+00 -3.47067505e-01 1.06922352e+00 -4.93514270e-01 2.03774318e-01 -1.20932043e-01 2.09140912e-01 1.31617844e+00 1.18030834e+00 5.58171928e-01 -2.88924843e-01 -2.03579724e-01 7.22395003e-01 9.17794853e-02 1.35012329e-01 -5.34411848e-01 1.67432144e-01 7.20662594e-01 1.09271729e+00 -7.11852908e-01 -2.94594109e-01 2.47920621e-02 1.24650359e+00 -4.68935929e-02 6.51313186e-01 -6.05289638e-01 -4.99701500e-01 2.40325227e-01 -1.81121826e-01 5.26567817e-01 -5.40809631e-02 -3.56183350e-02 -9.61537838e-01 2.88515121e-01 -9.58921254e-01 7.93112293e-02 -6.28634632e-01 -1.03971171e+00 8.79706800e-01 -3.05097908e-01 -1.55371559e+00 -5.86804986e-01 -1.26284167e-01 -8.18868041e-01 1.01669347e+00 -1.39980471e+00 -6.57338381e-01 -1.08423889e-01 6.42454147e-01 6.36053383e-01 -4.70459670e-01 1.00789034e+00 3.22754592e-01 -6.26269042e-01 6.44865155e-01 4.62368168e-02 1.41651765e-01 7.03479290e-01 -1.07804632e+00 7.48561621e-02 1.19828248e+00 1.81198314e-01 5.87963223e-01 1.04356122e+00 -3.27733725e-01 -1.09124398e+00 -7.93156803e-01 1.08858120e+00 -5.32257333e-02 4.99259174e-01 -6.01501107e-01 -9.04732823e-01 9.22357738e-02 4.79375198e-02 -1.65551424e-01 8.42172563e-01 -3.24430972e-01 -1.98233962e-01 -1.78137839e-01 -1.18994486e+00 1.96884766e-01 3.49058479e-01 -6.16380692e-01 -7.02787459e-01 1.03051290e-01 3.22532326e-01 -3.59938736e-03 -4.46305484e-01 2.54008144e-01 3.07591707e-01 -1.02680433e+00 9.50018942e-01 -3.43795009e-02 -1.65263161e-01 -5.26101470e-01 -6.67305216e-02 -1.41834569e+00 -7.00025931e-02 -8.31957042e-01 -3.21639836e-01 1.60260236e+00 3.07953387e-01 -5.89720786e-01 3.66716176e-01 2.24324629e-01 1.50447518e-01 -5.37160784e-03 -1.48178923e+00 -7.98110723e-01 -5.86415410e-01 -5.10127306e-01 4.98287559e-01 7.18633056e-01 -2.04253569e-01 3.56745690e-01 -4.90998983e-01 7.28577316e-01 1.05847228e+00 -4.28616256e-02 6.58335447e-01 -1.32540894e+00 -2.77440578e-01 -2.55122811e-01 -3.37148547e-01 -7.43167102e-01 1.78826571e-01 -2.93986708e-01 4.47087705e-01 -1.30418229e+00 -5.23119509e-01 2.41299230e-03 -4.16991770e-01 1.72776923e-01 -2.18710601e-01 1.19982541e-01 3.27335447e-02 3.05734009e-01 -7.81763792e-02 3.75476986e-01 3.57374310e-01 -2.87829667e-01 -4.90460366e-01 6.29750311e-01 -3.61032873e-01 9.43877637e-01 7.89220333e-01 -5.53606451e-01 -1.37734547e-01 -8.93347189e-02 -2.22201481e-01 4.49660122e-01 5.84972680e-01 -1.31437302e+00 3.24053288e-01 5.73230624e-01 3.02546322e-01 -6.72786951e-01 5.74458897e-01 -1.00559628e+00 -2.73965765e-02 3.15639347e-01 -2.52650619e-01 -4.12901729e-01 2.71140695e-01 5.48613012e-01 -3.38799179e-01 -5.41550636e-01 8.21613073e-01 1.84545189e-01 -2.17788890e-01 -2.98666894e-01 -8.60300422e-01 -4.36372817e-01 5.65227091e-01 -1.80794597e-01 3.16756070e-01 -6.77019000e-01 -1.05105388e+00 -3.02281469e-01 -2.11004019e-01 2.92803764e-01 8.23234856e-01 -1.00399864e+00 -9.39089119e-01 4.52798784e-01 7.28250539e-04 -3.04150820e-01 1.96602955e-01 7.51085281e-01 2.58849841e-03 2.52070516e-01 4.01049882e-01 -6.22520208e-01 -1.57879376e+00 5.95152915e-01 4.94009346e-01 1.68813109e-01 -3.65990698e-01 9.41018343e-01 5.03849089e-02 6.52229879e-03 7.39914775e-01 7.63890846e-03 -6.07105672e-01 3.48530680e-01 8.55240703e-01 3.75570536e-01 4.48363423e-01 -9.51685905e-01 -6.07224584e-01 3.55155826e-01 8.75263214e-02 -5.49425960e-01 1.06561625e+00 -4.60244626e-01 -8.18434954e-02 6.86181486e-01 1.22717488e+00 3.63177776e-01 -8.32418263e-01 -2.38714889e-01 -2.32060760e-01 -3.22770715e-01 2.98514068e-01 -7.40743577e-01 -8.63090575e-01 9.19897020e-01 1.07642448e+00 4.19759095e-01 1.19481015e+00 -2.25261360e-01 4.12731975e-01 1.44681975e-01 2.88690299e-01 -8.25165331e-01 -1.70586184e-01 8.19609314e-02 1.07012153e+00 -1.23421168e+00 -4.22645628e-01 3.65334637e-02 -1.36603519e-01 9.87561762e-01 1.63492799e-01 1.27968788e-01 8.89037251e-01 1.35553598e-01 3.59345883e-01 1.07104175e-01 -1.49641484e-01 2.36680191e-02 3.77409041e-01 5.31196177e-01 2.44122937e-01 -1.35599822e-01 6.77205771e-02 5.53824663e-01 -2.95781106e-01 -8.03371370e-02 4.03474122e-01 7.59332359e-01 -5.35057724e-01 -5.95183849e-01 -1.14107144e+00 1.03904605e-01 -7.11751819e-01 -1.68284178e-01 -3.69439721e-01 1.20067336e-01 -6.26946539e-02 1.73001444e+00 -9.38819572e-02 -1.97891161e-01 1.24261692e-01 4.28597331e-01 1.01339839e-01 -2.94081807e-01 -6.18381381e-01 6.73053265e-01 -1.80441886e-02 -2.21404225e-01 -5.48183918e-01 -5.30494452e-01 -1.02540553e+00 2.17914507e-01 -7.02945769e-01 4.76649076e-01 1.10052955e+00 1.11413980e+00 9.60941240e-02 5.40897012e-01 1.00040817e+00 -1.13539803e+00 -3.55226576e-01 -1.18570840e+00 -5.39033592e-01 2.38938868e-01 1.07222676e+00 -6.98139668e-01 -1.20434523e+00 8.11365098e-02]

[14.83825969696045, 5.781285285949707]

2cea51e1-7814-492c-ba3d-da99ed06fb9d

minimum-levels-of-interpretability-for

2307.00660

https://arxiv.org/abs/2307.00660v1

https://arxiv.org/pdf/2307.00660v1.pdf

Minimum Levels of Interpretability for Artificial Moral Agents

As artificial intelligence (AI) models continue to scale up, they are becoming more capable and integrated into various forms of decision-making systems. For models involved in moral decision-making, also known as artificial moral agents (AMA), interpretability provides a way to trust and understand the agent's internal reasoning mechanisms for effective use and error correction. In this paper, we provide an overview of this rapidly-evolving sub-field of AI interpretability, introduce the concept of the Minimum Level of Interpretability (MLI) and recommend an MLI for various types of agents, to aid their safe deployment in real-world settings.

['Cosmin Badea', 'Avish Vijayaraghavan']

2023-07-02

['decision-making']

['reasoning']

[-9.10672843e-02 7.76838303e-01 -2.77946144e-01 -5.26778162e-01 3.75006825e-01 -5.57668149e-01 8.43075812e-01 4.52015132e-01 -4.54401702e-01 6.46797299e-01 6.85941428e-02 -6.30374849e-01 -2.19334111e-01 -5.38835645e-01 -1.98555261e-01 -2.88475573e-01 1.65077895e-02 7.32262492e-01 -4.27729189e-01 -3.24981093e-01 2.70653069e-01 3.63125801e-01 -1.34468150e+00 -1.21525399e-01 1.17021155e+00 5.84372401e-01 -6.30316377e-01 7.06674039e-01 6.23613238e-01 1.40412796e+00 -6.72060668e-01 -8.18495333e-01 1.81913868e-01 -4.25805837e-01 -1.14180768e+00 -2.92679250e-01 -3.15505087e-01 -7.15437710e-01 -9.42223072e-02 1.09838164e+00 -9.37318355e-02 -2.18161240e-01 1.00639164e+00 -2.06527066e+00 -9.25557435e-01 6.80914521e-01 1.02954820e-01 -2.59373993e-01 4.30369675e-01 5.93594730e-01 1.04207551e+00 -5.58091747e-03 7.31496394e-01 1.60537243e+00 5.49605489e-01 1.08333457e+00 -1.00181711e+00 -2.68008232e-01 9.33221206e-02 4.73070592e-01 -9.31850493e-01 -4.20384407e-01 4.26045418e-01 -6.13858640e-01 1.01645982e+00 4.51122165e-01 1.08536375e+00 9.49671209e-01 7.19481349e-01 8.03822339e-01 1.30239868e+00 -4.54372585e-01 8.86901200e-01 1.83341682e-01 4.20589626e-01 4.79284227e-01 1.03098667e+00 4.88383263e-01 -3.06599557e-01 -5.41349411e-01 7.73513317e-01 -1.41810134e-01 2.24872917e-01 -3.77370924e-01 -9.82845426e-01 1.01939893e+00 2.65950024e-01 1.50913566e-01 -7.83196807e-01 6.13193095e-01 2.80998081e-01 5.31216204e-01 1.56658307e-01 9.38684106e-01 -8.19651857e-02 -3.15953135e-01 -8.21397528e-02 5.73163807e-01 1.10770059e+00 5.55562198e-01 3.48145783e-01 1.05477735e-01 3.91615368e-02 9.44830179e-02 8.85644197e-01 5.22766948e-01 9.26194116e-02 -1.76547527e+00 -3.60206068e-01 9.26070869e-01 5.82999289e-01 -8.02474976e-01 -4.46546495e-01 -1.53371140e-01 -4.35588360e-01 9.83380020e-01 1.21551849e-01 -3.03115826e-02 -6.13970876e-01 1.72254121e+00 2.70492464e-01 -2.69787312e-01 5.62675714e-01 9.12974656e-01 2.94812053e-01 4.09913301e-01 3.21164638e-01 -7.23508000e-02 1.31451130e+00 -6.83463454e-01 -8.89028609e-01 -6.97773337e-01 7.60762155e-01 -1.15289904e-01 7.41231978e-01 4.14747536e-01 -1.13027787e+00 3.35780382e-01 -1.39245212e+00 -5.54954521e-02 -2.41786748e-01 -6.17618024e-01 1.26350272e+00 7.48703778e-01 -1.03684235e+00 2.97202110e-01 -7.67239273e-01 -5.34084141e-01 6.93747520e-01 3.81753474e-01 -2.30202779e-01 9.81505290e-02 -1.15450060e+00 1.62077570e+00 2.26674780e-01 1.03029147e-01 -1.13276100e+00 -1.44166827e-01 -7.88335204e-01 -1.49472132e-01 2.00199038e-01 -1.16267550e+00 1.33195817e+00 -1.53039801e+00 -1.46400666e+00 1.09548283e+00 2.29634508e-01 -7.50486314e-01 5.31171262e-01 -3.28836679e-01 -1.99062869e-01 -2.81320768e-03 -5.35463542e-02 5.87847292e-01 5.03407598e-01 -1.45051110e+00 -5.87531328e-01 -4.12305146e-01 7.63918519e-01 4.61060315e-01 -5.95014822e-03 4.44541633e-01 5.18633664e-01 -3.09662968e-01 -3.49372804e-01 -1.02796555e+00 -2.97916770e-01 3.48187119e-01 4.70420122e-02 -4.41242039e-01 5.18822253e-01 -4.36817527e-01 1.16846037e+00 -1.81438649e+00 2.51046330e-01 5.38770407e-02 6.88853204e-01 3.38432670e-01 7.78986439e-02 2.98220634e-01 3.32574815e-01 4.32189405e-01 -2.69787967e-01 -1.93660229e-01 4.53564525e-01 7.23854780e-01 1.91277400e-01 4.74989265e-01 2.66802788e-01 7.95514286e-01 -1.19241345e+00 -3.25937301e-01 3.39843154e-01 1.04026236e-01 -5.96742630e-01 4.22997475e-01 -2.49846026e-01 4.13123012e-01 -7.33629823e-01 5.05295277e-01 4.56777513e-02 -4.00932916e-02 2.13779032e-01 6.37876213e-01 1.88471168e-01 1.68329537e-01 -6.44948721e-01 9.55953240e-01 -2.54109591e-01 8.31882119e-01 3.03877831e-01 -5.08673906e-01 5.72268784e-01 5.64179718e-01 -1.05686791e-01 -4.79262710e-01 3.95441264e-01 2.41987973e-01 5.94812632e-01 -4.98650193e-01 1.39153257e-01 -4.13987726e-01 -2.00196311e-01 1.08522964e+00 -4.13470924e-01 -4.54191774e-01 4.81499843e-02 2.56191701e-01 1.09998214e+00 -4.00513634e-02 9.11663949e-01 -5.98843277e-01 5.29467583e-01 2.26737872e-01 7.95753241e-01 7.63848186e-01 -7.64506996e-01 -9.39629227e-02 4.38868523e-01 -1.10238004e+00 -1.09867930e+00 -6.48498356e-01 -1.53924869e-02 8.18427563e-01 2.51819044e-01 -2.33660102e-01 -1.24431181e+00 -5.87008417e-01 -5.01553668e-03 1.54398477e+00 -8.55221212e-01 -5.21692812e-01 -3.12652707e-01 -4.86090213e-01 6.52697384e-01 4.33433592e-01 3.48669380e-01 -1.32521498e+00 -1.60944307e+00 -5.47316484e-02 -1.30827129e-01 -7.13970602e-01 3.17342132e-01 -9.27511156e-02 -6.43010914e-01 -1.00266767e+00 9.89086032e-02 -6.71178699e-02 8.73788178e-01 1.27185490e-02 1.33011830e+00 8.55602205e-01 1.15423620e-01 7.17382610e-01 -5.16189754e-01 -9.81799304e-01 -1.28114522e+00 -5.25425255e-01 4.25529420e-01 -4.83447105e-01 5.48840821e-01 -2.91287690e-01 -5.04255712e-01 2.16169283e-01 -9.55813944e-01 3.54270905e-01 2.81897664e-01 5.74395418e-01 -3.20326239e-01 -4.40734066e-02 3.59213144e-01 -6.63024783e-01 1.10251427e+00 -4.53316450e-01 -2.77922988e-01 5.64558923e-01 -1.10477340e+00 -2.09805109e-02 6.25951767e-01 -3.31684351e-02 -9.95494902e-01 -7.96974242e-01 4.21721548e-01 2.62696892e-01 -2.38988727e-01 4.07811165e-01 -1.64050773e-01 3.55038308e-02 7.89695680e-01 -5.06431103e-01 1.76500723e-01 1.18569270e-01 3.98416489e-01 9.90235090e-01 2.66003877e-01 -6.89325571e-01 5.93190730e-01 3.59719783e-01 -1.59409061e-01 -3.45989853e-01 -6.36789382e-01 4.97174025e-01 -2.82250077e-01 -6.84932351e-01 8.31366658e-01 -4.32459444e-01 -1.08961928e+00 4.73110974e-01 -1.22891104e+00 -4.56509918e-01 -2.19512075e-01 4.09249038e-01 -8.85726571e-01 8.40861574e-02 -3.34246069e-01 -1.33404016e+00 -3.70232701e-01 -1.33810174e+00 4.34687793e-01 5.51571906e-01 -9.83215928e-01 -1.17666554e+00 5.67510687e-02 7.72520900e-01 3.86820108e-01 2.50119030e-01 1.15061259e+00 -9.28860784e-01 -2.87540227e-01 -4.89604264e-01 1.39253944e-01 3.42507750e-01 -6.79319203e-02 3.25275153e-01 -8.78521204e-01 -6.04485497e-02 3.03778648e-01 -5.41570246e-01 -1.99761346e-01 2.52778500e-01 4.37035471e-01 -9.26480770e-01 -3.98473293e-02 1.25238717e-01 8.02897871e-01 7.22148061e-01 5.76036870e-01 7.94900775e-01 1.91739425e-01 7.39505112e-01 9.36982393e-01 6.19604766e-01 8.69384110e-01 4.13684130e-01 6.62537217e-01 9.56865028e-02 4.27419245e-01 -1.02446191e-01 4.90827680e-01 4.11502361e-01 -6.83452129e-01 -3.05895239e-01 -1.31717336e+00 1.58920344e-02 -2.38163352e+00 -9.29851890e-01 2.83806175e-01 2.08829021e+00 6.04555905e-01 1.43646970e-01 6.88922554e-02 4.18541208e-02 5.86456537e-01 -2.45338380e-01 -8.33599627e-01 -1.22015369e+00 -6.16130084e-02 -4.77796525e-01 1.81614220e-01 7.61679769e-01 -5.83761930e-01 9.66163933e-01 8.15320778e+00 1.33842111e-01 -5.07379174e-01 9.38989818e-02 1.11672783e+00 4.57778871e-01 -7.69583941e-01 4.32993412e-01 9.88748968e-02 2.08720833e-01 1.06848264e+00 -5.57708979e-01 7.41953909e-01 8.36087525e-01 3.96782786e-01 -4.32250023e-01 -1.63022602e+00 4.53947157e-01 -9.82485712e-02 -9.73065674e-01 1.31377652e-01 4.65125032e-02 3.60471040e-01 -4.15832072e-01 -9.24219713e-02 3.15697752e-02 1.08327389e+00 -1.43947911e+00 1.18460548e+00 3.47136110e-01 1.19350672e-01 -7.38412976e-01 1.26145637e+00 4.94885594e-01 -1.49445117e-01 -3.98823559e-01 -1.95886195e-01 -7.45309591e-01 1.26274303e-01 -7.76393786e-02 -6.07181489e-01 2.09554926e-01 5.86775184e-01 4.74433482e-01 -2.53728598e-01 5.62149882e-01 -4.79881644e-01 3.19420546e-01 -2.75150985e-01 -3.50547969e-01 2.16003999e-01 -4.87629294e-01 6.75833046e-01 6.06434226e-01 -2.07675084e-01 5.65256178e-01 -3.21958691e-01 9.98426557e-01 3.59935135e-01 -4.76859003e-01 -5.62492311e-01 -4.48304951e-01 6.89318240e-01 8.67290020e-01 -6.04274809e-01 -4.46733207e-01 1.63650930e-01 6.87183678e-01 2.20068023e-01 1.77254200e-01 -8.36569071e-01 2.66842604e-01 1.02724659e+00 -3.32287610e-01 -8.44817579e-01 -1.75771207e-01 -6.81547105e-01 -9.86599028e-01 -3.34172010e-01 -1.40977716e+00 3.29945773e-01 -9.44353104e-01 -1.01030564e+00 4.40121949e-01 2.44441226e-01 -7.93246567e-01 -4.69615161e-01 -6.53036058e-01 -6.84076190e-01 2.25918174e-01 -1.09995139e+00 -1.34619153e+00 -1.12152100e-01 -9.71832499e-03 1.94828827e-02 -1.76392809e-01 1.10954237e+00 -4.53087866e-01 -5.68754196e-01 4.51283395e-01 -1.48034647e-01 -1.35810301e-01 3.07055324e-01 -9.73262310e-01 3.22115451e-01 7.16720998e-01 -3.68468344e-01 9.99157012e-01 1.28233385e+00 -5.49700558e-01 -1.54319859e+00 -4.01790589e-01 6.19946778e-01 -8.26044738e-01 6.54997945e-01 -7.84999952e-02 -3.76136363e-01 1.12664771e+00 3.16292465e-01 -5.69905937e-01 8.50887775e-01 8.42298120e-02 -3.00848454e-01 -8.87635574e-02 -1.73337436e+00 1.04849708e+00 9.83557463e-01 -1.94256127e-01 -1.04402769e+00 2.25159138e-01 6.43083572e-01 9.23232585e-02 -8.69130254e-01 2.13272363e-01 5.92570722e-01 -1.31291127e+00 6.27597034e-01 -7.94494390e-01 3.37755889e-01 -5.38645089e-01 5.94965219e-02 -1.20940971e+00 -5.25694251e-01 -1.02504992e+00 -9.98353809e-02 8.46198916e-01 2.04733804e-01 -1.22690904e+00 4.91766989e-01 1.74171090e+00 9.46220756e-02 -3.83719862e-01 -1.00554514e+00 -3.79332423e-01 4.04145978e-02 -3.29539359e-01 6.70233309e-01 1.06086802e+00 7.16088176e-01 1.02899007e-01 -3.37160110e-01 -5.25144264e-02 9.56094265e-01 -3.01224709e-01 5.81176341e-01 -1.47762704e+00 -5.95375104e-03 -7.12611973e-01 -8.38698030e-01 -2.85173148e-01 1.06402107e-01 -4.98110384e-01 3.99764366e-02 -1.70044303e+00 3.41643065e-01 -5.33331394e-01 -9.35133100e-02 7.82711744e-01 -2.24159285e-01 -1.87677905e-01 3.94796371e-01 3.38672400e-01 -7.44056821e-01 4.95835811e-01 8.32589507e-01 1.12846084e-01 2.84175158e-01 -2.03455672e-01 -9.95010018e-01 1.29301143e+00 8.82885814e-01 -5.45471966e-01 -4.45254743e-01 -4.21269000e-01 5.58639884e-01 -6.48796838e-03 4.44627196e-01 -8.89545262e-01 1.10141188e-02 -9.83016491e-01 -1.08686477e-01 3.56406510e-01 2.01233327e-01 -1.10140586e+00 5.15606225e-01 1.01244998e+00 -5.78640640e-01 2.33840764e-01 -1.46800473e-01 1.32096812e-01 1.82262152e-01 -5.21299720e-01 8.15573692e-01 1.06105106e-02 -4.34673846e-01 -2.34427169e-01 -1.09950185e+00 -2.07480028e-01 1.69445682e+00 -4.41252381e-01 -6.42956078e-01 -7.35465229e-01 -3.08437705e-01 4.69965279e-01 1.31890106e+00 3.61354887e-01 5.54210246e-01 -1.02538955e+00 -6.97858751e-01 -2.02026233e-01 2.86612570e-01 -2.12920770e-01 -6.68905824e-02 3.48950207e-01 -1.01489496e+00 3.66267979e-01 -6.34455085e-01 9.12629534e-03 -1.03703666e+00 3.58817339e-01 5.40503144e-01 1.66060954e-01 -4.89232183e-01 5.91268599e-01 4.54679757e-01 -3.27097088e-01 7.48615637e-02 -1.70599401e-01 -3.59734535e-01 -5.78229904e-01 6.60896957e-01 3.95965278e-01 -6.89185381e-01 -7.13945687e-01 -6.11775577e-01 9.46220085e-02 5.99740120e-03 -3.23373348e-01 1.24612236e+00 -2.64209777e-01 -5.63721359e-01 5.01594365e-01 1.31961077e-01 -6.30878389e-01 -1.05792797e+00 3.50609481e-01 1.43330887e-01 -3.71031940e-01 6.47241296e-03 -1.12662029e+00 -5.61789453e-01 5.58877409e-01 3.34245592e-01 3.57161134e-01 7.65324414e-01 -2.35691369e-01 1.74268901e-01 6.35017753e-01 7.33948946e-01 -1.36836708e+00 -2.55389005e-01 3.01673532e-01 8.69981647e-01 -1.14388430e+00 3.65171701e-01 -2.45068837e-02 -1.19755864e+00 9.78450835e-01 6.85097098e-01 4.08624746e-02 2.89299458e-01 1.78481013e-01 3.87693197e-01 -2.16787651e-01 -1.03198516e+00 3.32865834e-01 -1.85916826e-01 9.90184844e-01 3.20525765e-01 5.28674304e-01 -4.91514117e-01 5.00525653e-01 -1.71864584e-01 1.15256295e-01 1.00530326e+00 1.18119991e+00 -5.47912598e-01 -1.01476371e+00 -3.73001665e-01 1.17543906e-01 -2.69116104e-01 2.08756715e-01 -9.54673886e-01 5.94419897e-01 -3.93510982e-03 1.41678321e+00 -6.85286224e-02 -3.35379571e-01 -5.81742004e-02 -1.47049263e-01 2.41495088e-01 -2.18918607e-01 -8.06529343e-01 -7.19491720e-01 5.03308356e-01 -5.71179092e-01 -6.69796288e-01 -5.93541622e-01 -1.42225623e+00 -9.92429078e-01 -6.32549152e-02 3.23192894e-01 5.26719928e-01 1.09749591e+00 3.30925405e-01 7.25660399e-02 4.03639644e-01 -4.93459374e-01 -5.42151928e-01 -6.25501931e-01 -2.13799551e-01 5.24446666e-01 1.25958905e-01 -6.44368708e-01 -5.36723197e-01 -1.63780347e-01]

[8.958528518676758, 6.204000949859619]

87822b0e-9451-420a-9930-ef538f82118e

multi-template-temporal-siamese-network-for

2211.13812

https://arxiv.org/abs/2211.13812v1

https://arxiv.org/pdf/2211.13812v1.pdf

Multi-Template Temporal Siamese Network for Long-Term Object Tracking

Siamese Networks are one of most popular visual object tracking methods for their high speed and high accuracy tracking ability as long as the target is well identified. However, most Siamese Network based trackers use the first frame as the ground truth of an object and fail when target appearance changes significantly in next frames. They also have dif iculty distinguishing the target from similar other objects in the frame. We propose two ideas to solve both problems. The first idea is using a bag of dynamic templates, containing diverse, similar, and recent target features and continuously updating it with diverse target appearances. The other idea is to let a network learn the path history and project a potential future target location in a next frame. This tracker achieves state-of-the-art performance on the long-term tracking dataset UAV20L by improving the success rate by a large margin of 15% (65.4 vs 56.6) compared to the state-of-the-art method, HiFT. The of icial python code of this paper is publicly available.

['Won-Sook Lee', 'Ali Sekhavati']

2022-11-24

['visual-object-tracking']

['computer-vision']

[-3.57331663e-01 -5.01794338e-01 -3.28858078e-01 1.45302013e-01 -1.34464994e-01 -8.88366938e-01 5.67181289e-01 -2.19920442e-01 -7.14903235e-01 7.47513294e-01 -2.74679899e-01 2.82266110e-01 4.88967523e-02 -4.37582344e-01 -6.44390047e-01 -8.49702358e-01 -3.63734096e-01 4.36707616e-01 1.10865974e+00 -3.41053642e-02 -1.26755998e-01 8.79010916e-01 -1.39023674e+00 -4.39451247e-01 3.80885094e-01 1.09539890e+00 2.06739098e-01 6.63427055e-01 2.73271292e-01 6.81122661e-01 -7.53357232e-01 -3.79225343e-01 5.66387057e-01 -1.54617041e-01 -2.62881756e-01 -1.47082865e-01 1.00595951e+00 -2.90684491e-01 -6.90644741e-01 1.33515036e+00 3.10103655e-01 1.03050143e-01 3.67153347e-01 -1.68991542e+00 -1.97917610e-01 2.67637998e-01 -7.63775289e-01 5.09823442e-01 -1.01999929e-02 4.07774031e-01 6.46567822e-01 -6.23223901e-01 8.16613436e-01 1.05819917e+00 9.73479211e-01 7.24508107e-01 -9.30230618e-01 -1.17703950e+00 5.41185021e-01 1.91352099e-01 -1.28084731e+00 -3.98899406e-01 3.28057826e-01 -6.87148929e-01 5.18070281e-01 -1.19419150e-01 1.06058931e+00 9.90440130e-01 3.55457157e-01 6.08134747e-01 4.50526118e-01 1.31242037e-01 -1.56009510e-01 -2.65071839e-01 3.99370380e-02 9.74856794e-01 6.92482948e-01 8.10622156e-01 -5.35652399e-01 -2.40507394e-01 6.61183655e-01 2.10849211e-01 -3.11718017e-01 -7.07549572e-01 -1.57985818e+00 5.96106291e-01 8.98648918e-01 1.72691867e-01 -5.76604962e-01 4.58187073e-01 2.15148509e-01 -2.81527452e-03 2.23036289e-01 2.16409132e-01 -4.03145820e-01 -6.77202716e-02 -1.06278610e+00 3.03925961e-01 6.67773902e-01 8.52051973e-01 4.64792341e-01 4.41959351e-01 -4.20639336e-01 1.83116451e-01 6.08185649e-01 9.81803894e-01 2.08699286e-01 -9.97869730e-01 1.04273036e-01 3.32653791e-01 5.45433879e-01 -1.09691918e+00 -2.97947317e-01 -7.60498047e-01 -6.02474868e-01 6.89354241e-01 6.92550540e-01 -5.83594263e-01 -1.04062784e+00 2.02632737e+00 6.69065475e-01 5.95033824e-01 -2.50037946e-02 7.53165424e-01 7.43903756e-01 6.17365777e-01 1.73374936e-01 -1.32971823e-01 9.71180737e-01 -1.13183749e+00 -5.84115624e-01 -5.21843612e-01 1.63517609e-01 -7.45714188e-01 -1.26585588e-01 1.11395251e-02 -6.38640583e-01 -7.36716866e-01 -9.48528111e-01 8.18466246e-01 -3.40293050e-01 3.81196648e-01 5.78540206e-01 4.43682641e-01 -1.11078262e+00 5.80031991e-01 -1.10696948e+00 -8.47832859e-01 5.82029879e-01 4.42745566e-01 -3.66195679e-01 1.74435228e-01 -7.87272453e-01 9.67857301e-01 5.74738383e-01 3.73577476e-02 -1.45475614e+00 -7.49593794e-01 -6.36437118e-01 -1.34703130e-01 4.92413461e-01 -5.37695527e-01 1.22723234e+00 -9.45568204e-01 -1.18289208e+00 5.59760630e-01 -1.00743413e-01 -7.85465419e-01 6.40410066e-01 -3.85943711e-01 -5.16426921e-01 -1.23372450e-01 2.07050249e-01 1.04409814e+00 1.17268097e+00 -1.09016991e+00 -1.29112995e+00 -3.54137927e-01 -1.07970566e-01 9.71917212e-02 -2.06980124e-01 4.25498374e-02 -5.76023519e-01 -6.79365218e-01 -1.14740156e-01 -1.21027005e+00 -1.13730781e-01 1.00257051e+00 7.47267306e-02 -2.39240199e-01 1.42541504e+00 -2.97821254e-01 8.89603376e-01 -2.05069661e+00 1.62528548e-03 -1.27878726e-01 5.04052877e-01 8.39920461e-01 -3.07252437e-01 5.54132424e-02 1.54248014e-01 -4.79822040e-01 3.79495680e-01 -1.53196514e-01 -3.08773369e-01 -7.38732442e-02 -2.38397896e-01 8.23225200e-01 -3.89823169e-02 9.07831550e-01 -1.21637774e+00 -3.64999741e-01 1.90328747e-01 4.85090464e-01 8.85790959e-02 1.47642285e-01 -2.53692329e-01 4.17972445e-01 -3.43253881e-01 5.46180487e-01 5.06735146e-01 -2.33825654e-01 -1.12168007e-01 -2.74218082e-01 -4.62587625e-01 -4.31921810e-01 -1.09279394e+00 1.44219041e+00 3.86644661e-01 9.91364717e-01 5.69556132e-02 -3.67405474e-01 8.72597516e-01 2.44860604e-01 7.83402264e-01 -5.59782572e-02 1.86717436e-01 -1.23343393e-01 3.08347762e-01 -2.60198792e-03 2.28529349e-01 2.99865901e-01 4.19942766e-01 2.13644803e-02 1.17062166e-01 4.31626797e-01 4.93067205e-01 1.54166207e-01 1.21976388e+00 3.84941578e-01 1.56351075e-01 -5.80730513e-02 4.68928546e-01 4.32557017e-01 9.40693319e-01 8.96076500e-01 -7.55686045e-01 1.38267711e-01 4.69986312e-02 -7.18977690e-01 -7.28511751e-01 -1.03334248e+00 1.78643465e-01 7.89126396e-01 4.22713727e-01 -2.84518033e-01 -3.62185746e-01 -8.10560405e-01 2.39347756e-01 2.05850795e-01 -7.83340335e-01 -1.66912735e-01 -4.74341214e-01 -4.17757839e-01 5.09643972e-01 6.58668876e-01 7.55178869e-01 -8.60595584e-01 -9.48855221e-01 2.91756511e-01 4.45324033e-02 -1.14297783e+00 -6.18873715e-01 -8.05429444e-02 -7.23535180e-01 -1.32049096e+00 -8.71057749e-01 -5.77313423e-01 5.05848587e-01 6.95708632e-01 8.90416682e-01 8.84711649e-03 -2.18514159e-01 2.80766517e-01 -2.18317449e-01 -5.30062437e-01 -1.04586340e-01 -2.13480622e-01 4.43081498e-01 -2.20754705e-02 3.02776396e-01 2.78251320e-02 -4.50378209e-01 5.51392198e-01 -1.63075209e-01 -3.58842582e-01 4.68144208e-01 6.38190269e-01 5.66105843e-01 9.04257819e-02 -1.64927952e-02 -1.24012940e-01 -1.39365315e-01 -2.96857417e-01 -1.35014081e+00 3.16162854e-01 -4.83152717e-01 -2.65523314e-01 4.73838657e-01 -8.91583979e-01 -5.99654853e-01 4.97853070e-01 2.72414505e-01 -1.01495278e+00 -2.72546839e-02 -1.07213326e-01 2.66976178e-01 -6.96871161e-01 4.86058384e-01 9.67640728e-02 1.65712878e-01 -8.61849412e-02 8.35167617e-02 -2.02329352e-01 7.34593928e-01 -6.94948956e-02 1.42995393e+00 6.01672590e-01 1.70316458e-01 -5.25374413e-01 -1.19041443e+00 -7.02327669e-01 -6.75053775e-01 -8.53978038e-01 9.37574387e-01 -1.01132774e+00 -8.83660018e-01 6.93376482e-01 -1.17078316e+00 -2.84360737e-01 -3.03857237e-01 6.54669762e-01 -1.20696194e-01 1.86505899e-01 -1.23911187e-01 -6.34547949e-01 -2.94486672e-01 -8.38037372e-01 8.34685445e-01 6.85998917e-01 1.44132361e-01 -9.36864138e-01 2.83051699e-01 -4.12767261e-01 5.91667414e-01 6.30332589e-01 -1.79753408e-01 -4.17758524e-01 -7.64936626e-01 -4.89634365e-01 -3.63745332e-01 -3.44192721e-02 1.43336058e-01 3.57980043e-01 -6.50002718e-01 -8.73705924e-01 -3.45030665e-01 7.18545914e-02 9.04952288e-01 6.52298093e-01 4.55354273e-01 -1.70606524e-02 -9.98336792e-01 6.56405985e-01 1.44698870e+00 3.91834825e-01 1.00267999e-01 2.49575451e-01 6.36820853e-01 6.31316602e-02 1.06032157e+00 2.83825219e-01 8.76412019e-02 8.86720419e-01 9.03156042e-01 -2.53825635e-02 -3.56865913e-01 -2.10248247e-01 6.28310442e-01 1.60201937e-01 -4.00361717e-02 -3.87996137e-01 -8.18263113e-01 6.35287404e-01 -2.13032794e+00 -1.28009152e+00 -9.58842859e-02 2.38511109e+00 2.20651478e-01 3.17085832e-01 4.44017500e-01 -5.58728456e-01 9.79910135e-01 2.92627156e-01 -8.13710868e-01 7.48336196e-01 -7.19733611e-02 -4.71382469e-01 9.74346578e-01 2.44756430e-01 -1.47523737e+00 1.17220521e+00 6.05703115e+00 6.34920478e-01 -1.14758968e+00 1.15348600e-01 -9.91858020e-02 -1.02609277e-01 9.77709770e-01 -5.61826751e-02 -1.43688226e+00 4.27829593e-01 7.49027431e-01 -2.91749477e-01 1.41417995e-01 8.75340521e-01 -7.26384744e-02 -2.17250943e-01 -7.71598697e-01 9.51088965e-01 2.37111107e-01 -1.30495906e+00 -4.24391419e-01 -4.40040370e-03 6.49744213e-01 5.68862021e-01 3.04052625e-02 1.65348724e-01 7.30665982e-01 -5.21512806e-01 8.41775894e-01 7.63996422e-01 3.70625705e-01 -5.16730309e-01 6.57321990e-01 2.34777004e-01 -1.86290824e+00 -1.31603509e-01 -5.18140554e-01 1.76597744e-01 1.94823548e-01 9.64053422e-02 -4.25217330e-01 3.91432643e-01 1.13255954e+00 1.25766766e+00 -8.76447856e-01 2.00008011e+00 -2.18486071e-01 4.61576939e-01 -5.98087370e-01 -4.12588567e-02 5.77775896e-01 1.95065036e-01 1.07174563e+00 8.35926294e-01 3.63390625e-01 -2.48201534e-01 7.19477832e-01 5.46159923e-01 1.71734065e-01 -5.14247715e-01 -8.12198877e-01 1.12559833e-01 4.99112964e-01 1.45116508e+00 -7.72150993e-01 -3.90421838e-01 -2.55820811e-01 6.83113337e-01 1.30084470e-01 1.96770281e-01 -1.21677148e+00 -2.33335957e-01 9.83723462e-01 -2.40094721e-01 7.40862906e-01 -3.66511077e-01 5.28592527e-01 -9.79300141e-01 -1.63627952e-01 -4.53816891e-01 4.32424396e-01 -7.08979547e-01 -1.16068864e+00 8.15820694e-01 -1.16085954e-01 -1.75279164e+00 4.15799906e-03 -6.82439506e-01 -6.98735237e-01 5.67639530e-01 -1.43380594e+00 -1.32603526e+00 -8.17584455e-01 5.34727573e-01 4.80373949e-01 -4.58923697e-01 4.44045097e-01 1.89014807e-01 -6.17430091e-01 2.90422171e-01 1.16824694e-01 4.83111203e-01 8.61252248e-01 -9.33411479e-01 4.88083124e-01 1.36683238e+00 2.11309373e-01 2.69005567e-01 7.11300850e-01 -1.06955528e+00 -1.30816150e+00 -1.49739289e+00 5.96257627e-01 -6.17223680e-01 1.04853559e+00 -1.33225664e-01 -6.14213109e-01 6.99553788e-01 1.64083958e-01 4.72031504e-01 2.58718058e-02 -3.61607641e-01 -2.03695625e-01 -3.04682732e-01 -9.10415769e-01 4.58196551e-01 1.05153346e+00 8.64281058e-02 -3.34603608e-01 4.77063298e-01 5.52779913e-01 -5.99007249e-01 -8.00969541e-01 4.58665550e-01 7.46627331e-01 -8.12117398e-01 9.94601965e-01 -5.03068328e-01 -6.05372548e-01 -9.40251291e-01 1.34005859e-01 -1.13945508e+00 -6.82788432e-01 -7.12681353e-01 -5.08247435e-01 1.08289289e+00 1.43872291e-01 -5.75136721e-01 1.01015639e+00 -1.60216987e-02 2.30300263e-01 -3.78750950e-01 -8.07839394e-01 -1.41380775e+00 -4.31742847e-01 3.35639976e-02 3.52044076e-01 5.42028785e-01 -7.34385252e-01 1.25123516e-01 -4.97378588e-01 5.75643897e-01 1.18855846e+00 4.43333760e-02 1.03593004e+00 -1.71794438e+00 1.96680009e-01 -3.31713319e-01 -7.39472032e-01 -1.13935840e+00 2.65057832e-01 -5.33077717e-01 2.63768345e-01 -1.23115635e+00 7.66714737e-02 -3.59309971e-01 -2.18789577e-01 6.42481863e-01 -5.38520813e-02 3.21501821e-01 5.89632988e-01 3.07979822e-01 -9.79937494e-01 5.52394331e-01 1.02605891e+00 -2.78685838e-01 1.16795413e-02 2.87790269e-01 1.09756496e-02 6.30163610e-01 6.36306822e-01 -9.46932554e-01 -5.34371287e-02 -3.91996831e-01 -4.96854991e-01 -1.14249825e-01 8.30645740e-01 -1.75959313e+00 7.66498744e-01 -4.64944690e-02 7.76024818e-01 -1.02605736e+00 5.92350125e-01 -1.20434690e+00 4.64984506e-01 1.00858951e+00 1.84271291e-01 5.06786704e-01 5.48170149e-01 9.63697374e-01 -1.36785328e-01 -7.67073706e-02 1.12385583e+00 -1.17294528e-02 -1.18298519e+00 9.59460557e-01 -2.76359886e-01 1.06576636e-01 1.47614551e+00 -3.18544120e-01 -5.33240199e-01 -2.58924186e-01 -5.28889298e-01 4.39150155e-01 6.71076238e-01 7.85713673e-01 3.81520301e-01 -1.33555865e+00 -6.54928923e-01 7.86463320e-02 -1.19338054e-02 -4.69074070e-01 -2.06939027e-01 9.10788536e-01 -3.36378008e-01 4.49831039e-01 -5.04371226e-01 -9.77612555e-01 -1.56233776e+00 5.95432341e-01 7.25904286e-01 -7.01801479e-02 -7.00534284e-01 9.25864220e-01 1.00271493e-01 1.69615418e-01 6.21684074e-01 1.05733797e-01 -2.71115959e-01 -3.61253023e-02 7.06956685e-01 3.49918008e-01 -4.70852643e-01 -1.12715852e+00 -5.84097207e-01 9.93198156e-01 -2.57475674e-02 2.55156785e-01 1.12022531e+00 1.23364635e-01 -2.61469260e-02 2.00463831e-01 5.77690363e-01 -3.75856906e-01 -1.93687177e+00 -3.15612942e-01 1.45155229e-02 -7.19918370e-01 6.46664500e-02 -6.75832868e-01 -1.56282890e+00 4.20130998e-01 1.14955425e+00 8.80423561e-02 7.36352921e-01 -8.72172490e-02 5.57182729e-01 4.71279472e-01 5.21664441e-01 -6.10420525e-01 1.25494540e-01 8.38436067e-01 6.17824018e-01 -1.26789558e+00 1.56088680e-01 -2.03706890e-01 -4.14717495e-01 1.11848938e+00 1.05641973e+00 -2.66968876e-01 6.94923818e-01 2.61531234e-01 2.11003914e-01 -3.70894611e-01 -6.71254873e-01 -6.00380599e-01 4.03544873e-01 9.95753467e-01 -1.15883537e-01 -3.17991108e-01 3.84540915e-01 -3.95391405e-01 1.77369833e-01 -1.90125778e-01 1.18834078e-01 9.04576480e-01 -5.55402160e-01 -7.88010240e-01 -5.20253181e-01 2.97331065e-01 -3.37510645e-01 2.97113270e-01 -3.22955549e-01 1.04171216e+00 9.38122794e-02 8.51137936e-01 -4.14845087e-02 -2.32305363e-01 2.97655135e-01 -3.03094000e-01 4.72234577e-01 -3.67729247e-01 -6.53201103e-01 2.89493185e-02 -1.72395125e-01 -8.09439659e-01 -8.22895348e-01 -9.40630555e-01 -8.42437446e-01 -2.49941230e-01 -6.21929288e-01 -3.40151675e-02 5.47408998e-01 7.66512454e-01 2.28744626e-01 4.63007927e-01 2.85010725e-01 -1.27363575e+00 -2.93889225e-01 -6.53973758e-01 -2.55493283e-01 3.18698846e-02 5.87336838e-01 -1.20325994e+00 -2.36948803e-01 -3.84734608e-02]

[6.438023090362549, -2.0490269660949707]

b9d2896c-9ff9-4448-8099-9a20564ef32e

retinexflow-for-ct-metal-artifact-reduction

2306.10520

https://arxiv.org/abs/2306.10520v1

https://arxiv.org/pdf/2306.10520v1.pdf

RetinexFlow for CT metal artifact reduction

Metal artifacts is a major challenge in computed tomography (CT) imaging, significantly degrading image quality and making accurate diagnosis difficult. However, previous methods either require prior knowledge of the location of metal implants, or have modeling deviations with the mechanism of artifact formation, which limits the ability to obtain high-quality CT images. In this work, we formulate metal artifacts reduction problem as a combination of decomposition and completion tasks. And we propose RetinexFlow, which is a novel end-to-end image domain model based on Retinex theory and conditional normalizing flow, to solve it. Specifically, we first design a feature decomposition encoder for decomposing the metal implant component and inherent component, and extracting the inherent feature. Then, it uses a feature-to-image flow module to complete the metal artifact-free CT image step by step through a series of invertible transformations. These designs are incorporated in our model with a coarse-to-fine strategy, enabling it to achieve superior performance. The experimental results on on simulation and clinical datasets show our method achieves better quantitative and qualitative results, exhibiting better visual performance in artifact removal and image fidelity

['Dong Liang', 'Kun Shang', 'Yinsheng Li', 'Ce Wang', 'Jiandong Su']

2023-06-18

['metal-artifact-reduction', 'computed-tomography-ct']

['medical', 'methodology']

[ 3.32708389e-01 -2.17006356e-01 1.81330562e-01 -8.85369256e-02 -7.30689168e-01 -6.29333928e-02 1.09255448e-01 -3.84886146e-01 -9.07403901e-02 4.16454285e-01 4.91331249e-01 -3.20932940e-02 -3.37124199e-01 -5.34086168e-01 -4.40481305e-01 -8.15398157e-01 1.09600082e-01 -2.55880002e-02 1.36666834e-01 1.25870392e-01 7.10315034e-02 2.25761443e-01 -8.60761046e-01 3.73308390e-01 1.03685558e+00 1.07286644e+00 4.71474886e-01 2.54979640e-01 1.96993463e-02 1.07427347e+00 -1.92909956e-01 1.25177447e-02 3.86459559e-01 -6.12214983e-01 -5.76112449e-01 4.33340997e-01 -1.01602145e-01 -5.59286177e-01 -6.20577276e-01 1.17087090e+00 7.01894403e-01 -2.34711036e-01 6.91527903e-01 -7.69688725e-01 -8.11674535e-01 3.94157112e-01 -8.19646239e-01 1.97032973e-01 1.57358259e-01 2.46088967e-01 1.86248988e-01 -1.12010109e+00 5.21273136e-01 9.08817291e-01 6.37643635e-01 4.52930599e-01 -1.14753640e+00 -6.86615109e-01 -2.26930469e-01 -3.16878892e-02 -1.33783317e+00 -3.54319096e-01 8.75856638e-01 -6.02015793e-01 4.10323381e-01 2.43703783e-01 7.27781594e-01 6.70808375e-01 6.53411746e-01 5.41693509e-01 1.20873046e+00 -2.73421407e-01 9.93801355e-02 -1.02593742e-01 -1.83198541e-01 7.61800826e-01 2.71377027e-01 2.57319748e-01 -3.46065938e-01 -2.26766348e-01 1.38589358e+00 5.58114231e-01 -8.71756971e-01 -4.17004406e-01 -1.12981844e+00 4.74175692e-01 6.91784382e-01 2.94911742e-01 -7.62869835e-01 2.11769000e-01 1.46427989e-01 9.07973051e-02 1.78977549e-01 -5.54738380e-02 1.49015352e-01 2.88178623e-01 -8.23594689e-01 -3.67645994e-02 2.43812427e-01 9.09218490e-01 4.43671197e-01 1.31956249e-01 -3.98492694e-01 8.84558380e-01 6.10020220e-01 5.64947844e-01 9.05331314e-01 -7.06340849e-01 2.67805368e-01 4.21461999e-01 6.69232151e-03 -8.83296490e-01 -5.11750042e-01 -7.75026023e-01 -1.14416528e+00 3.98427904e-01 -9.19635519e-02 3.65058556e-02 -1.21295333e+00 1.47473335e+00 3.06445420e-01 4.83645976e-01 -2.57922590e-01 1.30623722e+00 6.42064273e-01 3.89395237e-01 1.78811979e-02 -6.71937585e-01 1.70462978e+00 -8.07359695e-01 -1.12903523e+00 -9.49512124e-02 2.04251736e-01 -1.07048452e+00 9.77702320e-01 5.63758314e-01 -1.28650260e+00 -4.24630255e-01 -1.27106261e+00 -7.98673928e-02 3.58352751e-01 1.84914976e-01 4.01253253e-01 5.75749159e-01 -5.04514813e-01 5.43028474e-01 -1.13314688e+00 2.00578853e-01 5.08729100e-01 2.45527089e-01 -1.02759846e-01 -2.65342206e-01 -8.24731708e-01 8.23940575e-01 -1.56153575e-01 2.94536740e-01 -9.82620597e-01 -9.94509578e-01 -6.43700957e-01 -7.06700310e-02 3.73182952e-01 -1.00448871e+00 1.18560731e+00 -8.33831906e-01 -1.54124343e+00 3.69005680e-01 -8.74357969e-02 -6.30294904e-02 7.56385922e-01 -2.40068868e-01 -4.95895594e-01 1.59543306e-01 1.88548759e-01 1.21602630e-02 1.13628566e+00 -1.34917700e+00 -3.24663162e-01 -4.13612515e-01 -4.51821506e-01 1.66203782e-01 -2.31716946e-01 -7.09541067e-02 -5.47776461e-01 -1.07264638e+00 6.89198434e-01 -7.02904403e-01 -4.56804514e-01 4.18345600e-01 -1.76502928e-01 5.00020146e-01 5.06589711e-01 -9.96339440e-01 1.48176336e+00 -2.45293427e+00 -3.73954363e-02 4.40648824e-01 5.93699336e-01 -1.17031179e-01 1.48923069e-01 9.96947959e-02 -3.46458912e-01 -1.26027256e-01 -6.02584779e-01 -1.82351023e-01 -4.48585540e-01 -2.34274305e-02 5.98313734e-02 6.53155088e-01 1.50763080e-01 6.10397518e-01 -7.53744423e-01 -5.84820271e-01 2.57001549e-01 6.49703324e-01 -6.55485690e-01 1.94475636e-01 3.52939457e-01 6.66968882e-01 -7.86907971e-01 5.50260842e-01 1.12714767e+00 -5.23336947e-01 1.61014721e-02 -5.39865911e-01 -1.20196491e-01 7.37918867e-03 -1.30137074e+00 1.93379569e+00 -6.93951666e-01 7.62858894e-04 1.70176119e-01 -3.48823309e-01 5.81624329e-01 4.91459221e-01 1.00834811e+00 -7.69868910e-01 3.52377534e-01 3.68344605e-01 5.70792034e-02 -8.62060249e-01 -1.20778782e-02 -5.31477213e-01 3.75508696e-01 3.64813030e-01 -3.25943291e-01 -7.15348199e-02 -3.16038817e-01 1.62767127e-01 1.15364981e+00 -7.94887617e-02 1.77562907e-01 -2.25783899e-01 5.71021199e-01 -7.85090104e-02 7.17023373e-01 2.09513918e-01 -1.18465662e-01 1.14567280e+00 3.43515575e-02 -2.40968153e-01 -9.01334107e-01 -1.38988757e+00 -3.14543843e-01 8.47806111e-02 3.74931335e-01 -1.82763845e-01 -7.85317779e-01 -3.01987857e-01 -2.62274027e-01 1.80861101e-01 -5.29242635e-01 -3.99928838e-01 -6.58223391e-01 -9.13234890e-01 -7.49917924e-02 5.38699627e-01 7.16843545e-01 -7.06700563e-01 -5.46411574e-01 5.38191199e-01 -5.67685962e-01 -8.87130380e-01 -8.94561648e-01 -5.02166413e-02 -1.07730305e+00 -1.05640912e+00 -9.78297770e-01 -9.14204776e-01 9.93813097e-01 4.12926495e-01 5.95732272e-01 3.00371915e-01 -7.51254201e-01 -3.20368670e-02 -2.03791946e-01 -1.53183892e-01 -2.27541402e-01 -6.31657302e-01 -1.50071695e-01 3.32491130e-01 -2.38513321e-01 -6.40109599e-01 -1.14040422e+00 3.31491441e-01 -1.21384847e+00 1.71306610e-01 9.38668728e-01 1.01284981e+00 7.15986609e-01 2.11639911e-01 2.09449664e-01 -7.06317842e-01 6.46844506e-01 -2.93534398e-01 -3.88353258e-01 1.33896932e-01 -5.56107223e-01 -7.40281213e-03 5.60896575e-01 -2.63673365e-01 -1.23121560e+00 3.33120197e-01 -1.09281398e-01 -7.29712605e-01 4.36690390e-01 4.51714873e-01 -1.25406668e-01 -3.03430855e-01 7.28682458e-01 5.30243516e-01 2.03631938e-01 -5.10121346e-01 -3.39669883e-02 6.40086830e-01 5.60138047e-01 -3.17337930e-01 7.74063170e-01 7.76518703e-01 4.63168183e-03 -5.60472429e-01 -4.27126497e-01 -3.53415132e-01 -2.67274618e-01 -3.12287778e-01 8.04623485e-01 -1.00572097e+00 -6.65997326e-01 3.75998735e-01 -9.38559055e-01 1.63339153e-01 -2.01739043e-01 8.06170940e-01 -5.54590166e-01 6.13042653e-01 -7.29388177e-01 -5.01131237e-01 -6.16904616e-01 -1.46462619e+00 8.21090698e-01 1.19478948e-01 1.08347304e-01 -4.32677835e-01 -9.30952933e-03 1.25674978e-01 5.74450970e-01 1.28083497e-01 8.43781710e-01 3.19106460e-01 -8.13894272e-01 1.50132358e-01 -2.75884718e-01 4.14453357e-01 4.97632980e-01 -4.25201148e-01 -7.55482316e-01 -3.87805849e-01 7.13368714e-01 2.46411383e-01 6.36425674e-01 6.37379587e-01 1.31141448e+00 -6.97193667e-02 -3.14367920e-01 8.57184529e-01 1.76452911e+00 2.22466156e-01 9.64551687e-01 1.34413674e-01 5.33341050e-01 1.78526148e-01 4.55063075e-01 6.63227379e-01 1.82134300e-01 5.51341712e-01 3.11722875e-01 -2.66177177e-01 -6.48395479e-01 -1.69967458e-01 2.17964351e-02 1.05541337e+00 -1.29321590e-01 2.82470644e-01 -7.58081138e-01 5.85264742e-01 -1.66704106e+00 -5.22859514e-01 -2.02880412e-01 2.13418722e+00 6.53913021e-01 -1.40109450e-01 -3.26705396e-01 1.32969052e-01 6.53005421e-01 -2.74644226e-01 -4.79386330e-01 3.97381037e-01 3.52116883e-01 2.94375896e-01 5.25895476e-01 3.70042026e-01 -8.21903288e-01 3.42885941e-01 6.26180029e+00 8.86535227e-01 -1.23967052e+00 3.89093965e-01 2.19325081e-01 2.08575055e-02 -4.88477796e-01 -1.23489514e-01 2.11509150e-02 6.20215416e-01 1.02322564e-01 -4.08450738e-02 4.36551273e-01 4.32464659e-01 5.08084774e-01 -1.46386446e-02 -7.39241421e-01 1.28570664e+00 -9.59895998e-02 -1.11578357e+00 5.29753454e-02 -3.72317173e-02 5.39716959e-01 -3.53577226e-01 6.43298104e-02 -2.73563176e-01 1.02505848e-01 -7.89479375e-01 6.68892980e-01 6.44206285e-01 6.85540617e-01 -6.01888001e-01 4.84383076e-01 1.37691200e-01 -1.10690856e+00 -7.94601962e-02 -3.61237466e-01 3.67200196e-01 2.64779001e-01 9.32111144e-01 -5.42445600e-01 9.20776486e-01 6.94325089e-01 6.21811271e-01 -3.73063125e-02 1.49040747e+00 -1.52070358e-01 2.36323923e-01 -1.74885795e-01 3.46352577e-01 -2.17691854e-01 -1.93094790e-01 4.55951065e-01 7.77218163e-01 5.56091964e-01 5.54820240e-01 2.33334109e-01 9.44293499e-01 6.48091733e-02 1.87750340e-01 -2.04246238e-01 5.95259011e-01 3.47004801e-01 1.25725400e+00 -6.85927689e-01 -2.03785583e-01 -5.56800485e-01 1.09515536e+00 -2.27301806e-01 3.50746930e-01 -1.06374657e+00 -3.93287241e-01 4.16703582e-01 3.80962491e-01 7.58208036e-02 -8.32773373e-02 -6.10769749e-01 -1.22420740e+00 2.83004999e-01 -8.60543072e-01 1.78386807e-01 -8.99088025e-01 -1.18401325e+00 6.06356204e-01 -4.33243543e-01 -1.87956178e+00 2.09587708e-01 -4.03132856e-01 -4.51140612e-01 9.70088840e-01 -1.58212388e+00 -1.12819898e+00 -6.72113717e-01 9.49279785e-01 4.07503515e-01 2.82780558e-01 3.43081981e-01 7.67239690e-01 -2.77274996e-01 4.66362834e-01 1.64325669e-01 8.09519142e-02 7.44420171e-01 -7.10955918e-01 -1.66903242e-01 9.18862581e-01 -3.87317479e-01 5.59751034e-01 4.17498291e-01 -8.11561406e-01 -1.49953508e+00 -1.05942976e+00 1.94198981e-01 -2.00998262e-02 3.28978509e-01 -2.22668499e-02 -8.01186919e-01 4.87225920e-01 1.11925051e-01 2.40026385e-01 4.89805400e-01 -5.92456639e-01 2.84400750e-02 -1.82670251e-01 -1.37551284e+00 6.24384224e-01 1.16110301e+00 -2.90801078e-01 -5.97310722e-01 2.09058493e-01 6.24980330e-01 -6.68387473e-01 -1.06083155e+00 4.75003392e-01 5.69373965e-01 -7.14207351e-01 1.04393566e+00 -8.48061964e-02 7.25492895e-01 -6.60050035e-01 9.43284556e-02 -1.32966149e+00 -7.97464490e-01 -4.51811850e-01 2.73536325e-01 9.19795036e-01 2.29236454e-01 -7.15082109e-01 6.57190681e-01 4.62582409e-01 -5.09818077e-01 -5.42872310e-01 -9.46848392e-01 -6.93634987e-01 -1.35178223e-01 -2.75812775e-01 5.72180450e-01 8.59674275e-01 3.84137630e-02 -1.56531036e-01 -4.71880347e-01 5.48983037e-01 8.97304833e-01 7.04381913e-02 3.20692778e-01 -7.37868965e-01 -5.22197664e-01 -2.24806219e-01 -2.56924480e-01 -1.04196811e+00 -7.72195518e-01 -8.44235241e-01 2.01411739e-01 -1.48987997e+00 4.37287331e-01 -4.50882554e-01 -4.04715508e-01 2.81063855e-01 -1.63134694e-01 2.52878964e-01 1.11764409e-01 4.01540518e-01 1.76715199e-02 6.27814531e-01 1.94164681e+00 -2.41317138e-01 -2.52056457e-02 -1.73947081e-01 -7.09448993e-01 6.83000624e-01 3.60186428e-01 -6.55396581e-01 -3.83454174e-01 -7.75041640e-01 -1.59801468e-01 4.06474441e-01 4.22348917e-01 -1.27073228e+00 2.74510354e-01 -1.00776114e-01 6.76816106e-01 -3.87531251e-01 2.37504691e-01 -1.29731643e+00 4.39982802e-01 9.05119240e-01 3.76009718e-02 -1.42742112e-01 4.21369746e-02 4.04813766e-01 -3.48628789e-01 -2.60335177e-01 9.84164774e-01 -1.70210332e-01 -3.20179999e-01 5.43945611e-01 -2.97152936e-01 -2.08847210e-01 9.37539279e-01 -1.80503264e-01 1.12631554e-02 -1.52505904e-01 -9.21471477e-01 1.25608966e-01 2.41940051e-01 2.77632624e-01 1.00067627e+00 -1.57175136e+00 -7.77874351e-01 5.56093097e-01 -2.64528636e-02 9.31463316e-02 8.31002355e-01 1.06897354e+00 -7.68756092e-01 -7.68765062e-02 -2.69010544e-01 -6.89085603e-01 -1.03986847e+00 5.02229393e-01 4.95867848e-01 -3.77623230e-01 -9.00178432e-01 5.39488554e-01 5.85693002e-01 9.09621362e-03 -6.53810948e-02 -4.42770869e-01 9.17886198e-02 -5.00974238e-01 6.76741481e-01 3.08189452e-01 2.64192969e-01 -3.12812805e-01 -3.65243196e-01 1.03907919e+00 -2.16365620e-01 -1.37051478e-01 1.26368678e+00 -3.37373048e-01 -2.65420750e-02 -3.08640063e-01 1.14358127e+00 2.94171274e-02 -1.08037055e+00 -3.80554736e-01 -4.34082627e-01 -8.88819993e-01 4.18324977e-01 -8.63039970e-01 -1.46503079e+00 8.23022723e-01 1.14401579e+00 -1.94959179e-01 1.58185661e+00 -2.61652470e-01 1.00807428e+00 -3.99522930e-01 4.38213676e-01 -8.07777584e-01 2.83509701e-01 -1.77288149e-02 1.13604176e+00 -9.49267924e-01 2.30836242e-01 -1.00239885e+00 -3.98860395e-01 9.46514189e-01 4.69815820e-01 -2.43805632e-01 9.80488598e-01 5.17503202e-01 2.27636009e-01 -2.44700253e-01 -2.71893442e-01 7.58827925e-02 1.56825677e-01 4.39563334e-01 3.10512096e-01 8.65687057e-03 -6.33385897e-01 8.93329740e-01 7.71684349e-02 5.56904495e-01 4.76979315e-01 1.17337584e+00 -3.43490869e-01 -9.67933059e-01 -6.55302465e-01 4.22990322e-01 -5.95368445e-01 -1.99856311e-01 3.25366348e-01 5.01912892e-01 9.82903391e-02 8.97601187e-01 -2.48426154e-01 -2.88551629e-01 6.82197511e-01 -4.06098217e-01 5.52340686e-01 -5.03693819e-01 -5.83223462e-01 6.34062171e-01 -4.07842189e-01 -5.44222713e-01 -1.43684521e-01 -4.80358750e-01 -1.52886486e+00 -1.75686684e-02 -5.45348287e-01 -1.58126473e-01 6.14189208e-01 6.61132932e-01 3.24190527e-01 1.14981282e+00 9.25806642e-01 -4.86187905e-01 -5.24046421e-01 -8.50358725e-01 -7.19855368e-01 5.65465569e-01 4.54828292e-01 -6.83426917e-01 -1.53307483e-01 1.15555614e-01]

[13.508604049682617, -2.558328628540039]

308a95ba-d1f5-4e2a-a205-384a2270659c

low-rank-time-frequency-synthesis

http://papers.nips.cc/paper/5522-low-rank-time-frequency-synthesis

http://papers.nips.cc/paper/5522-low-rank-time-frequency-synthesis.pdf

Low-Rank Time-Frequency Synthesis

Many single-channel signal decomposition techniques rely on a low-rank factorization of a time-frequency transform. In particular, nonnegative matrix factorization (NMF) of the spectrogram -- the (power) magnitude of the short-time Fourier transform (STFT) -- has been considered in many audio applications. In this setting, NMF with the Itakura-Saito divergence was shown to underly a generative Gaussian composite model (GCM) of the STFT, a step forward from more empirical approaches based on ad-hoc transform and divergence specifications. Still, the GCM is not yet a generative model of the raw signal itself, but only of its STFT. The work presented in this paper fills in this ultimate gap by proposing a novel signal synthesis model with low-rank time-frequency structure. In particular, our new approach opens doors to multi-resolution representations, that were not possible in the traditional NMF setting. We describe two expectation-maximization algorithms for estimation in the new model and report audio signal processing results with music decomposition and speech enhancement.

['Cédric Févotte', 'Matthieu Kowalski']

2014-12-01

neurips-2014-12

['audio-signal-processing']

['audio']

[ 5.47559917e-01 -7.33040720e-02 3.80475491e-01 1.98040791e-02 -8.16963732e-01 -6.81707501e-01 4.26352054e-01 -3.08272809e-01 -2.98575103e-01 5.98013222e-01 6.10713065e-01 -2.19337016e-01 -5.95616996e-01 -3.63962561e-01 -4.47108120e-01 -1.05876148e+00 -7.25349486e-02 -4.63844985e-02 -2.77960747e-01 -4.25604224e-01 -6.22003786e-02 2.27492526e-01 -1.74517155e+00 3.37903887e-01 7.56093800e-01 7.04727888e-01 3.64887238e-01 1.05541027e+00 1.66336864e-01 4.27701294e-01 -4.25715715e-01 -2.37744555e-01 3.00416857e-01 -5.83163321e-01 -4.37794864e-01 2.23780066e-01 2.33570486e-01 5.83989322e-02 -1.81789532e-01 1.04768908e+00 3.11100006e-01 2.50632346e-01 6.61707222e-01 -1.16284704e+00 -4.57853019e-01 5.57790160e-01 -3.66730720e-01 5.12413345e-02 4.45311308e-01 -5.65610409e-01 1.17652607e+00 -1.22387815e+00 4.62920904e-01 1.06640482e+00 7.67676592e-01 7.91124031e-02 -1.51359892e+00 -2.83462733e-01 -1.35352656e-01 8.55195969e-02 -1.32753515e+00 -4.52405304e-01 8.74264479e-01 -5.46262920e-01 9.00904715e-01 5.57170570e-01 6.54487789e-01 1.01115680e+00 5.12699783e-02 7.07543373e-01 1.18881536e+00 -7.60192037e-01 6.84445575e-02 -2.59268314e-01 -1.24399684e-01 1.24581404e-01 1.23201162e-01 2.63998918e-02 -7.52366364e-01 -3.73764098e-01 8.59681010e-01 -2.19920784e-01 -5.41582227e-01 -4.13726598e-01 -1.34746110e+00 7.07530916e-01 -2.95809627e-01 7.42913008e-01 -6.48564577e-01 1.91165507e-01 3.35867763e-01 6.27131701e-01 6.62120402e-01 3.88883293e-01 -3.72370332e-01 -3.00236881e-01 -1.34273303e+00 2.31238768e-01 8.53342950e-01 3.90318096e-01 4.69972014e-01 4.63460624e-01 5.04533947e-02 7.41734028e-01 3.22887570e-01 5.33596158e-01 4.15492594e-01 -1.09665763e+00 3.66826773e-01 -1.69211134e-01 2.18311235e-01 -9.96267676e-01 -3.66031051e-01 -8.56616378e-01 -1.02481759e+00 -5.16026793e-03 5.81391871e-01 -1.08934246e-01 -3.39186221e-01 1.94324887e+00 3.25755596e-01 6.03056431e-01 1.27200797e-01 7.77959645e-01 6.40213769e-03 7.12272584e-01 -5.32711685e-01 -8.01575422e-01 1.16641378e+00 -4.69296843e-01 -1.08935571e+00 2.77467072e-01 2.73143589e-01 -1.20286167e+00 7.06827879e-01 1.14165461e+00 -1.27007866e+00 -6.46502852e-01 -1.12536502e+00 1.27219900e-01 1.74240872e-01 3.19734335e-01 5.79823792e-01 8.95697713e-01 -1.23620844e+00 7.76796818e-01 -5.96601546e-01 6.51262328e-03 -2.78263509e-01 7.46400049e-03 -5.37582338e-01 5.13381846e-02 -1.18026292e+00 5.45351624e-01 -1.12615908e-02 3.55591655e-01 -4.97454166e-01 -6.65475905e-01 -5.41922212e-01 -1.48922773e-02 3.41599494e-01 -7.36534834e-01 1.21815562e+00 -9.45994854e-01 -1.47621822e+00 4.01872814e-01 -4.47536528e-01 -4.41010416e-01 3.93435806e-01 -4.59491104e-01 -6.20034456e-01 2.57132053e-01 -9.89366546e-02 -2.06279203e-01 1.81653225e+00 -1.03301311e+00 -2.09534362e-01 -2.25044772e-01 -2.75438249e-01 -5.03223687e-02 -3.04260969e-01 -3.72350253e-02 2.31651276e-01 -1.29926229e+00 5.29173493e-01 -7.71693707e-01 -3.05875093e-01 -4.91277725e-01 -1.98526412e-01 1.96973071e-01 3.02932322e-01 -7.85180688e-01 1.51243103e+00 -2.53540158e+00 4.64904755e-01 2.77782172e-01 1.00757547e-01 2.69934963e-02 -1.36906072e-01 8.36675346e-01 -6.72703505e-01 -2.73649454e-01 -1.21129125e-01 -5.45687854e-01 8.49154368e-02 8.64464492e-02 -8.36086750e-01 7.04720259e-01 1.69539526e-01 4.62202787e-01 -8.76553357e-01 -3.26195955e-02 9.04764533e-02 1.00705779e+00 -6.98530257e-01 -2.33606771e-01 1.68184310e-01 4.58745360e-01 -6.32613450e-02 5.57683595e-02 5.49276054e-01 4.64832895e-02 8.78824443e-02 -5.41282535e-01 -4.89772290e-01 1.90264806e-01 -1.66105819e+00 1.91447449e+00 -4.04115230e-01 5.60373306e-01 5.57821512e-01 -1.05708849e+00 7.90059984e-01 7.32031941e-01 9.21563864e-01 -1.72485113e-01 -5.24660461e-02 5.57077527e-01 5.36187775e-02 -2.89492458e-01 7.33357906e-01 -5.23503125e-01 1.75097585e-01 3.63683909e-01 2.22618863e-01 -9.53868870e-03 3.84045482e-01 4.14288342e-02 8.81072819e-01 1.78185388e-01 4.63078648e-01 -2.99175739e-01 5.93061626e-01 -5.49500644e-01 2.34037653e-01 6.97014689e-01 1.93946809e-01 8.86713743e-01 2.83761084e-01 1.50522992e-01 -1.10913742e+00 -1.19089746e+00 -2.21940652e-01 1.02042186e+00 -7.21450984e-01 -8.22770000e-01 -8.12331319e-01 8.33996013e-02 -1.71258956e-01 5.30854702e-01 -4.29602921e-01 7.10896403e-02 -4.91214544e-01 -6.00380242e-01 4.13091451e-01 1.85189426e-01 -1.34959131e-01 -3.64673883e-01 -4.11078811e-01 5.71842968e-01 -3.87178123e-01 -1.06731534e+00 -4.04013962e-01 4.02041823e-01 -1.02331650e+00 -6.30455017e-01 -9.38802004e-01 -3.46181065e-01 3.36260468e-01 3.36169302e-01 8.07422638e-01 -4.81460273e-01 -2.46121913e-01 7.23554552e-01 -5.57734489e-01 -2.97753066e-01 -3.33508372e-01 -4.28178549e-01 4.50908720e-01 7.13708580e-01 -1.36105254e-01 -1.33053589e+00 -3.64927679e-01 2.03399107e-01 -1.08598471e+00 1.18961990e-01 5.52442014e-01 8.99955750e-01 5.15024543e-01 4.72429305e-01 5.87740541e-01 -4.34347123e-01 7.93219388e-01 -2.41911918e-01 -3.72197062e-01 -4.39141765e-02 -3.41481090e-01 -6.00452758e-02 7.35543013e-01 -6.93690240e-01 -9.55136359e-01 4.54133675e-02 -2.00570703e-01 -6.27970338e-01 2.73756593e-01 6.94530129e-01 -1.97832122e-01 2.47752860e-01 6.71072662e-01 5.54294288e-01 -6.35403618e-02 -8.59559000e-01 6.39832914e-01 4.46228623e-01 8.41974020e-01 -5.97973824e-01 1.04484737e+00 6.45312726e-01 1.94741249e-01 -1.27115142e+00 -6.01916671e-01 -6.77957177e-01 -6.19367659e-01 -1.73973680e-01 6.17392302e-01 -8.79427969e-01 -4.45893109e-01 3.19291979e-01 -1.14157355e+00 1.69188932e-01 -6.51741982e-01 9.80827332e-01 -7.48958647e-01 6.68318868e-01 -5.98554969e-01 -1.31859493e+00 -7.59758204e-02 -7.80184090e-01 9.97566819e-01 -4.15495366e-01 -4.38650638e-01 -6.94255292e-01 2.13449478e-01 6.89582899e-02 5.08189142e-01 9.41459164e-02 6.14064813e-01 -2.45877936e-01 -1.35131881e-01 -1.66144133e-01 2.47000590e-01 7.24826038e-01 1.01683408e-01 -9.58091393e-02 -1.10918033e+00 -4.21235234e-01 7.66514778e-01 2.96024799e-01 7.87350833e-01 7.47895241e-01 6.09846354e-01 -2.21244082e-01 2.54888773e-01 5.65997124e-01 1.33119905e+00 -3.21283489e-02 4.88299519e-01 -8.96073580e-02 4.12337661e-01 4.75956947e-01 5.49982607e-01 8.39473784e-01 -1.28566548e-01 7.96700001e-01 7.49045983e-02 1.75018385e-01 -1.03231976e-02 -2.56222606e-01 6.93565071e-01 1.31641459e+00 -3.87643665e-01 2.02311650e-02 -5.10624051e-01 4.61053252e-01 -1.77520061e+00 -1.19045746e+00 -5.26933789e-01 2.40541267e+00 5.50975442e-01 -5.99464914e-03 3.36728185e-01 1.04499793e+00 4.86493200e-01 1.35778308e-01 -1.98409960e-01 -2.03828603e-01 -2.76119918e-01 5.27248144e-01 2.50145018e-01 6.56977892e-01 -1.04708469e+00 1.52923793e-01 6.01632738e+00 9.71049070e-01 -9.57217753e-01 3.10504526e-01 -1.66204274e-01 -3.53628658e-02 -4.05637264e-01 3.18212323e-02 -4.27766711e-01 2.36548275e-01 1.18870652e+00 -2.89099455e-01 6.97782159e-01 3.92363191e-01 5.30238032e-01 1.05620451e-01 -9.50826526e-01 1.24776661e+00 1.85502600e-02 -9.07579601e-01 -1.32985711e-01 3.15493345e-01 4.56333011e-01 -2.85960823e-01 3.58664393e-01 5.69040515e-02 -4.12917912e-01 -8.96246195e-01 9.61730540e-01 6.21100128e-01 6.59421563e-01 -6.34697795e-01 2.39981771e-01 3.66936922e-01 -1.39063692e+00 -1.49800390e-01 -1.44885033e-01 -1.61156639e-01 4.80111480e-01 1.21174800e+00 -5.51520705e-01 9.28880751e-01 2.77088344e-01 6.74121916e-01 1.97352152e-02 9.17769492e-01 7.78441271e-03 8.13724339e-01 -4.76421118e-01 5.62273562e-01 -8.54340289e-03 -4.95281070e-01 1.14272690e+00 1.26086628e+00 8.68599474e-01 1.36301341e-02 -6.91509340e-03 6.01776958e-01 3.85091722e-01 1.85957998e-01 -3.56161892e-01 -4.13866907e-01 7.14751482e-02 1.16746247e+00 -6.14710152e-01 4.69958745e-02 -3.70345592e-01 8.43082368e-01 -4.45876062e-01 5.08523345e-01 -5.81303895e-01 -1.57445878e-01 6.83090091e-01 2.94964612e-01 4.59797323e-01 -5.58330536e-01 5.54569289e-02 -1.29968321e+00 2.11671129e-01 -1.02037418e+00 8.12642127e-02 -7.82284081e-01 -1.15751278e+00 5.55735648e-01 -6.35009483e-02 -1.60802007e+00 -6.93634212e-01 -5.41861773e-01 -2.21298918e-01 1.09962571e+00 -1.21479690e+00 -9.46295261e-01 5.15496075e-01 8.76809359e-01 3.30048174e-01 -6.87846020e-02 1.06512380e+00 5.96575856e-01 -6.35842904e-02 1.54402584e-01 5.34117699e-01 -3.18978965e-01 5.17905951e-01 -1.35264349e+00 1.17886528e-01 9.98212516e-01 5.80658317e-01 1.00137782e+00 1.28242803e+00 -4.70189631e-01 -1.61114907e+00 -6.33685470e-01 9.26253557e-01 -3.00538629e-01 9.38459754e-01 -4.29972410e-01 -8.23094010e-01 4.29172158e-01 1.35074064e-01 -1.83665738e-01 9.63997900e-01 1.85816541e-01 -3.09999824e-01 -6.65897876e-02 -6.15071714e-01 3.00035298e-01 7.66052127e-01 -9.13139224e-01 -6.80051744e-01 1.09515764e-01 5.24647474e-01 -1.45675793e-01 -8.47535074e-01 1.47388250e-01 7.03842163e-01 -1.08912551e+00 1.27016532e+00 -4.56700891e-01 2.99043059e-01 -6.76734865e-01 -6.92862928e-01 -1.34671295e+00 -4.44601566e-01 -1.33477378e+00 -2.67113656e-01 1.16399348e+00 2.18294114e-01 -4.64150786e-01 3.04940939e-01 -6.60618544e-02 -2.23880112e-01 -4.54088211e-01 -1.15295172e+00 -9.23138678e-01 -1.95405245e-01 -1.11816204e+00 2.12949440e-01 9.25291359e-01 5.64952344e-02 4.39731807e-01 -7.38100648e-01 2.54913449e-01 6.58844411e-01 1.05070099e-01 6.24423563e-01 -1.29602027e+00 -9.89992023e-01 -3.48539054e-01 -3.44911128e-01 -9.97502267e-01 -6.89437762e-02 -8.03789854e-01 -1.57290280e-01 -1.16503477e+00 -2.16623276e-01 2.02483028e-01 -2.94447452e-01 -1.41361699e-01 1.94401354e-01 1.57625720e-01 4.36592251e-01 1.11355588e-01 -1.00650407e-01 4.19175595e-01 1.10506964e+00 1.43231347e-01 -1.55233666e-01 1.61884859e-01 -5.31826377e-01 6.75682902e-01 3.65744561e-01 -3.78643125e-01 -7.26427376e-01 2.47418359e-02 7.65259624e-01 3.85706455e-01 2.32791662e-01 -1.06434858e+00 2.07107574e-01 2.22538635e-01 1.17194124e-01 -5.46690524e-01 8.09507489e-01 -8.39544117e-01 7.78339326e-01 1.04444116e-01 -1.33960545e-01 4.24985103e-02 -6.87381253e-02 8.44012320e-01 -4.78260130e-01 -1.75166219e-01 4.86825287e-01 1.96052827e-02 -2.18875781e-01 -2.22789317e-01 -5.98809958e-01 -1.86476842e-01 3.89091641e-01 -3.10664415e-01 2.93000609e-01 -6.80050850e-01 -1.14090955e+00 -6.24770284e-01 7.68584162e-02 1.77641422e-01 6.15582883e-01 -1.33666909e+00 -1.00180328e+00 4.09206748e-01 -3.36900204e-01 -5.21476507e-01 5.64037144e-01 1.34807348e+00 1.68942332e-01 4.17919099e-01 3.24522369e-02 -4.98542517e-01 -1.30960166e+00 4.39737350e-01 1.53043000e-02 -3.99073511e-01 -5.28003573e-01 7.23676383e-01 2.56710142e-01 3.77951004e-02 -1.29313618e-01 -3.74322712e-01 -1.64081544e-01 5.15048027e-01 7.68453121e-01 3.30389202e-01 2.32414708e-01 -1.00524461e+00 -9.71196443e-02 5.84469736e-01 4.12272513e-01 -7.16721535e-01 1.46555758e+00 -3.32605869e-01 -3.35998952e-01 8.92218411e-01 1.33637238e+00 5.72380483e-01 -8.27789903e-01 -1.49556994e-01 2.28266930e-03 -4.54433024e-01 2.26924300e-01 -3.91764313e-01 -6.81568444e-01 8.07264626e-01 3.37730616e-01 5.24686933e-01 1.45524657e+00 -5.67042410e-01 6.30252063e-01 2.16721579e-01 5.43764114e-01 -9.50520396e-01 -2.01776568e-02 4.85332549e-01 1.14738536e+00 -4.52257931e-01 -8.90461132e-02 -4.08396155e-01 -1.13429494e-01 1.36623776e+00 -5.22128046e-01 -1.99771911e-01 7.44718075e-01 3.27468574e-01 -1.82767555e-01 1.92558154e-01 -6.06488705e-01 -2.69159079e-01 4.65356141e-01 5.35634875e-01 6.61547184e-01 1.02300107e-01 -4.48765934e-01 8.75611782e-01 -5.79016924e-01 -7.74031803e-02 5.16131759e-01 6.95252836e-01 -3.53258103e-01 -1.44979489e+00 -9.38842237e-01 3.04652721e-01 -5.82190335e-01 -2.85983324e-01 -1.56599477e-01 4.12817866e-01 -5.46803102e-02 1.21649945e+00 -3.90608013e-01 -4.67078775e-01 3.75572979e-01 4.03695971e-01 6.20614529e-01 -3.63764048e-01 -4.50134009e-01 9.11889374e-01 -2.14027874e-02 -4.89295602e-01 -5.20400524e-01 -9.79243457e-01 -7.68031180e-01 6.81876484e-03 -3.87946457e-01 4.31604028e-01 8.22091818e-01 6.38383210e-01 -7.17735440e-02 5.60370743e-01 6.64346457e-01 -1.10426939e+00 -7.53901184e-01 -8.77358854e-01 -1.08516574e+00 2.94482559e-01 7.38663256e-01 -4.77496117e-01 -6.89936399e-01 1.32326484e-01]

[15.450004577636719, 5.5705885887146]

9c257ee8-7a1c-4cd8-9245-b2a12cccc619

transition-propagation-graph-neural-networks

2304.07501

https://arxiv.org/abs/2304.07501v1

https://arxiv.org/pdf/2304.07501v1.pdf

Transition Propagation Graph Neural Networks for Temporal Networks

Researchers of temporal networks (e.g., social networks and transaction networks) have been interested in mining dynamic patterns of nodes from their diverse interactions. Inspired by recently powerful graph mining methods like skip-gram models and Graph Neural Networks (GNNs), existing approaches focus on generating temporal node embeddings sequentially with nodes' sequential interactions. However, the sequential modeling of previous approaches cannot handle the transition structure between nodes' neighbors with limited memorization capacity. Detailedly, an effective method for the transition structures is required to both model nodes' personalized patterns adaptively and capture node dynamics accordingly. In this paper, we propose a method, namely Transition Propagation Graph Neural Networks (TIP-GNN), to tackle the challenges of encoding nodes' transition structures. The proposed TIP-GNN focuses on the bilevel graph structure in temporal networks: besides the explicit interaction graph, a node's sequential interactions can also be constructed as a transition graph. Based on the bilevel graph, TIP-GNN further encodes transition structures by multi-step transition propagation and distills information from neighborhoods by a bilevel graph convolution. Experimental results over various temporal networks reveal the efficiency of our TIP-GNN, with at most 7.2\% improvements of accuracy on temporal link prediction. Extensive ablation studies further verify the effectiveness and limitations of the transition propagation module. Our code is available at \url{https://github.com/doujiang-zheng/TIP-GNN}.

['Chun Chen', 'Ji Zhao', 'Xinyu Wang', 'Xingen Wang', 'Mingli Song', 'Yunzhi Hao', 'Tianli Zhang', 'Zunlei Feng', 'Tongya Zheng']

2023-04-15

['graph-mining', 'memorization']

['graphs', 'natural-language-processing']

[-2.50037938e-01 1.72571614e-01 -5.70030630e-01 -1.51940733e-01 3.05453658e-01 -3.13192338e-01 5.86187422e-01 3.07358384e-01 -6.77273124e-02 4.34598297e-01 3.42296623e-02 -6.67525232e-01 -4.20041800e-01 -1.22204244e+00 -5.38435400e-01 -4.16599065e-01 -9.00995255e-01 4.57835197e-01 4.58307236e-01 -3.40351194e-01 -2.30232596e-01 2.81432301e-01 -8.74759197e-01 -1.17699839e-01 6.90362394e-01 7.10841954e-01 9.91807599e-03 5.96337676e-01 -3.77141863e-01 8.43609214e-01 -2.96848863e-02 -4.75976706e-01 4.37936969e-02 -3.67082238e-01 -7.43510664e-01 -2.95936018e-01 -1.83528095e-01 -1.41598970e-01 -1.12334144e+00 8.34460735e-01 1.42412111e-01 2.58018434e-01 3.77249956e-01 -1.54479635e+00 -9.12302256e-01 1.39602757e+00 -5.59773386e-01 5.39254367e-01 2.35490769e-01 9.82664600e-02 1.34478307e+00 -7.65087366e-01 6.21135950e-01 1.07187963e+00 9.17579532e-01 3.80485982e-01 -1.17362416e+00 -8.06934178e-01 6.01419985e-01 3.59568954e-01 -1.40823054e+00 -2.88186558e-02 8.73776972e-01 -2.45019346e-01 1.01772237e+00 1.83275208e-01 1.00834560e+00 1.11591101e+00 3.68440636e-02 6.13634467e-01 3.51625860e-01 1.74938608e-02 -2.69317269e-01 -2.71136463e-01 4.95462596e-01 9.71305728e-01 3.06660891e-01 -3.71371955e-02 -4.11326200e-01 -1.05845772e-01 9.06325459e-01 4.77925420e-01 -1.37885258e-01 -9.52934027e-02 -1.02461290e+00 7.02079117e-01 9.19392407e-01 5.88986993e-01 -3.16045165e-01 5.87796688e-01 4.26048428e-01 4.66003329e-01 5.80646932e-01 -8.12525600e-02 -3.37197632e-01 -1.93019465e-01 -5.10821462e-01 -2.85252929e-01 1.03315008e+00 1.06207991e+00 7.69907117e-01 -6.93962257e-03 -2.06554979e-01 7.07877815e-01 4.91037279e-01 1.81963071e-01 3.71925354e-01 -3.84547681e-01 3.60456496e-01 8.94981503e-01 -4.94565129e-01 -1.48132455e+00 -4.66103911e-01 -5.56394219e-01 -1.32227111e+00 -8.19542468e-01 1.79353267e-01 -2.24880263e-01 -7.56906152e-01 1.76718104e+00 4.50824678e-01 7.43012607e-01 -3.32781613e-01 4.46245968e-01 9.82598484e-01 8.37492287e-01 -6.11210801e-03 -1.77949384e-01 1.02369392e+00 -1.10776830e+00 -6.27894759e-01 2.53201015e-02 7.73597360e-01 -1.64042741e-01 7.28613496e-01 -2.13235214e-01 -8.90735149e-01 -2.47912779e-01 -6.63558841e-01 1.63053006e-01 -3.87933850e-01 -3.66695255e-01 1.07052183e+00 2.59291291e-01 -1.32943642e+00 8.48131120e-01 -1.09502101e+00 -6.48254216e-01 4.35847163e-01 5.26324332e-01 -2.60910779e-01 1.14804633e-01 -1.57162702e+00 2.91328639e-01 4.58691299e-01 3.87827873e-01 -6.67546988e-01 -7.06816196e-01 -1.02685022e+00 3.96534950e-01 4.20559883e-01 -5.23757696e-01 9.96172965e-01 -5.78452051e-01 -1.26370311e+00 2.09235787e-01 -3.65731567e-01 -5.95250368e-01 1.88781455e-01 2.17374161e-01 -7.92963564e-01 5.66213876e-02 -1.44982308e-01 2.31750593e-01 4.26824540e-01 -8.85551989e-01 -3.19848716e-01 3.13321277e-02 1.85166523e-01 -3.83037105e-02 -1.00619555e+00 -3.03558946e-01 -9.47723389e-01 -6.61478996e-01 1.33053005e-01 -1.00589669e+00 -4.98388678e-01 -1.05270125e-01 -5.77009201e-01 -5.41556776e-01 8.30230117e-01 -3.77931684e-01 2.10876060e+00 -1.92272294e+00 -1.38982475e-01 7.13648856e-01 7.48593330e-01 2.20527902e-01 -3.15502137e-01 8.95783603e-01 -1.74421053e-02 3.90844792e-01 -5.42262159e-02 -3.15307260e-01 4.38534692e-02 3.58169496e-01 -4.73162532e-02 2.94457316e-01 -5.46529964e-02 1.40062821e+00 -1.20206356e+00 -4.30503130e-01 2.71432716e-02 5.46398222e-01 -4.61377501e-01 -3.11060976e-02 -2.03681812e-01 8.54346752e-02 -5.88276327e-01 4.62516069e-01 3.89389157e-01 -9.24636841e-01 7.44242847e-01 -1.51964173e-01 2.34319553e-01 3.98155153e-01 -8.41810524e-01 1.27975726e+00 -2.46178359e-01 5.22801936e-01 -2.02977344e-01 -9.95247006e-01 8.39223027e-01 2.41059929e-01 7.46474326e-01 -6.07698381e-01 4.77285273e-02 1.32135041e-02 2.74778336e-01 -3.83241713e-01 3.37596178e-01 3.43654126e-01 3.58413230e-03 7.39915371e-01 -1.63584650e-02 6.97426558e-01 4.80272472e-01 7.05833018e-01 1.46298194e+00 -2.97492027e-01 -6.15281239e-02 -3.69693451e-02 2.86917984e-01 -2.84671783e-01 6.47763371e-01 6.23022437e-01 -1.64579093e-01 8.10198188e-02 6.33599043e-01 -6.31828725e-01 -6.74155891e-01 -9.78455186e-01 3.14759672e-01 1.02310991e+00 1.61600396e-01 -9.95969415e-01 -2.29683429e-01 -7.33126223e-01 8.95488635e-02 1.52458921e-01 -7.80278802e-01 -3.09166431e-01 -8.53247702e-01 -7.68542826e-01 5.03782749e-01 4.93578494e-01 3.43136907e-01 -1.07826495e+00 2.01366812e-01 4.00331885e-01 -1.59622490e-01 -1.06692886e+00 -9.16893542e-01 -1.44185126e-01 -9.60118592e-01 -1.14948738e+00 -2.19350234e-01 -9.46080744e-01 7.57733166e-01 4.40135241e-01 1.11736727e+00 5.55814922e-01 2.73607881e-03 3.32095951e-01 -5.09316623e-01 2.39688262e-01 -1.25105441e-01 4.46872771e-01 1.51706383e-01 1.29196882e-01 4.36209053e-01 -1.40986550e+00 -7.68530548e-01 3.65932018e-01 -8.00854206e-01 1.23478219e-01 5.17005086e-01 6.43133461e-01 3.05067837e-01 3.59736651e-01 4.58428144e-01 -1.29039156e+00 7.90160596e-01 -1.07986057e+00 -3.54810119e-01 2.68481106e-01 -9.52106953e-01 -2.47131307e-02 7.29985535e-01 -8.14316213e-01 -5.55517673e-01 -4.12468612e-01 4.47854027e-02 -5.96772790e-01 4.00224000e-01 1.10226417e+00 2.14391232e-01 -3.26116197e-02 3.55301499e-01 3.73325586e-01 5.70915602e-02 -1.93358809e-01 3.38411510e-01 1.38609841e-01 5.60980923e-02 -5.48477352e-01 1.09846807e+00 3.60487461e-01 2.57252827e-02 -6.76423073e-01 -3.74394268e-01 -3.63912076e-01 -3.62411201e-01 -2.64757097e-01 3.93382937e-01 -6.14678621e-01 -1.01538813e+00 4.19881523e-01 -9.67252374e-01 -6.80554628e-01 -1.90092728e-03 4.05148357e-01 1.68704212e-01 6.44068599e-01 -1.22351229e+00 -5.94736576e-01 -4.98589277e-01 -5.16576827e-01 2.39881799e-01 3.42888474e-01 -2.44902089e-01 -1.66583288e+00 9.22896862e-02 -6.79940358e-02 5.97702920e-01 2.14010671e-01 8.99860740e-01 -6.64542317e-01 -8.11397076e-01 -1.74937710e-01 -2.85194308e-01 -2.48617560e-01 3.85267705e-01 2.20547870e-01 -3.35023791e-01 -3.50421906e-01 -7.60210752e-01 2.04643562e-01 7.86500037e-01 4.02220547e-01 1.14332402e+00 -6.26175106e-01 -7.62857020e-01 6.99099839e-01 1.26556766e+00 1.98076487e-01 4.15143728e-01 -2.79949624e-02 1.17534256e+00 3.99638921e-01 1.01159096e-01 5.02406180e-01 9.21501338e-01 3.63456875e-01 5.47429502e-01 1.31470084e-01 -5.39390706e-02 -6.95422590e-01 4.31592822e-01 1.32752013e+00 -1.84755623e-01 -5.20088911e-01 -9.79455888e-01 7.52424300e-01 -2.11193037e+00 -1.03012073e+00 -4.33884382e-01 1.74154925e+00 7.25220561e-01 3.38502914e-01 1.90538496e-01 -1.63878977e-01 8.29243958e-01 4.67036635e-01 -5.89226961e-01 -2.85401165e-01 1.76326677e-01 -3.45208403e-03 4.88414913e-01 5.56017339e-01 -8.33391190e-01 1.03625941e+00 5.04453135e+00 8.09524894e-01 -8.59025359e-01 6.99025691e-02 4.73140568e-01 9.12144631e-02 -5.98766029e-01 1.95015162e-01 -7.76522517e-01 6.07872307e-01 1.23285592e+00 -4.37834978e-01 6.00569487e-01 4.53899443e-01 3.07054162e-01 4.79804128e-01 -9.50382590e-01 8.14749479e-01 -3.06282699e-01 -1.54941523e+00 1.28598914e-01 1.05763502e-01 6.55178487e-01 1.75057620e-01 3.24103422e-02 5.32468438e-01 7.27097332e-01 -1.05421138e+00 1.02055177e-01 5.01052439e-01 6.85446799e-01 -5.83581567e-01 4.29843038e-01 2.49690071e-01 -2.05524850e+00 -4.16148901e-02 -3.51806842e-02 -1.90903172e-01 2.51011312e-01 8.38796973e-01 -8.95234168e-01 9.00220513e-01 7.10774839e-01 1.49648321e+00 -4.64778453e-01 7.90448964e-01 -1.80055201e-01 9.75011706e-01 -4.89606261e-01 -3.75570029e-01 4.29076433e-01 -4.08553839e-01 5.00582457e-01 1.28700185e+00 2.86243051e-01 2.13111475e-01 2.42024362e-01 8.19940567e-01 -3.60129267e-01 -7.61576816e-02 -6.95132673e-01 -5.98328233e-01 9.27357495e-01 1.34635234e+00 -7.82993257e-01 -1.29999042e-01 -4.53663528e-01 7.52117395e-01 6.94941998e-01 5.92189968e-01 -9.43968892e-01 -4.64471221e-01 7.50349939e-01 4.70868826e-01 3.79403144e-01 -4.93378252e-01 2.15216011e-01 -1.04987681e+00 1.07523955e-01 -4.14226443e-01 6.82487249e-01 -3.59819531e-01 -1.46672964e+00 8.37250590e-01 -3.38853756e-03 -1.02220452e+00 3.55213247e-02 -2.20581055e-01 -1.00026441e+00 5.95094085e-01 -1.16827571e+00 -1.32499719e+00 -1.97538674e-01 7.90624917e-01 1.84515998e-01 1.19043939e-01 6.25081241e-01 4.23611432e-01 -8.23048830e-01 8.67971718e-01 -8.66345093e-02 6.56825662e-01 1.22525878e-01 -1.05341303e+00 6.94864333e-01 8.52170646e-01 3.07946622e-01 9.83727038e-01 3.67723227e-01 -8.41580391e-01 -1.50874376e+00 -1.47535455e+00 1.08414686e+00 -9.96460691e-02 1.29226148e+00 -4.24372315e-01 -1.01266170e+00 1.09478891e+00 1.37539014e-01 3.50463867e-01 6.67477190e-01 4.51551974e-01 -3.99040222e-01 -1.87892437e-01 -6.61360383e-01 8.71578455e-01 1.61772847e+00 -5.84892154e-01 3.14445943e-02 3.03193480e-01 1.05143380e+00 -2.61361569e-01 -1.08572388e+00 3.60031366e-01 4.98636425e-01 -6.06220961e-01 9.04275894e-01 -5.95924258e-01 2.43644714e-01 -2.39524662e-01 3.67581785e-01 -1.21583664e+00 -6.22355700e-01 -9.35803235e-01 -8.94013584e-01 1.22535813e+00 5.72088301e-01 -9.81066763e-01 9.74722087e-01 3.39034081e-01 -1.47177652e-02 -1.16111803e+00 -6.70384049e-01 -5.91328442e-01 -3.30015630e-01 -5.17197073e-01 6.57575965e-01 1.19525576e+00 1.67734325e-01 4.51336116e-01 -5.35229921e-01 2.36091405e-01 4.53263968e-01 2.21934006e-01 5.88880181e-01 -1.25717568e+00 -4.74632859e-01 -6.23042345e-01 -3.98794711e-01 -1.38853073e+00 1.91885799e-01 -1.20441961e+00 -3.25132489e-01 -1.77213621e+00 1.58620164e-01 -7.97992468e-01 -5.64669669e-01 6.07631207e-01 -4.73203026e-02 2.97768973e-02 -5.52987605e-02 1.56152427e-01 -9.07823384e-01 6.33792162e-01 1.31702864e+00 -2.68449396e-01 -4.34358507e-01 -4.18974981e-02 -5.84317267e-01 4.17288691e-01 8.68651807e-01 -3.80708128e-01 -8.73439312e-01 -4.26737458e-01 5.32290041e-01 9.25774574e-02 2.52212018e-01 -5.85258961e-01 7.34146535e-01 -8.07006434e-02 -1.22341305e-01 -4.94937241e-01 2.11134151e-01 -7.13266373e-01 5.47916532e-01 6.29379809e-01 -1.21405870e-01 2.50816375e-01 -6.11199066e-02 1.05350101e+00 -1.22193225e-01 3.81849825e-01 1.49005055e-01 1.69882253e-02 -7.27147341e-01 1.23830938e+00 -2.50697792e-01 -6.62870631e-02 9.41020310e-01 -2.15662003e-01 -3.67315114e-01 -5.95530808e-01 -1.09693801e+00 8.57919514e-01 5.57241067e-02 5.46262681e-01 8.04904044e-01 -1.57683325e+00 -4.15994108e-01 7.30973333e-02 -1.68651007e-02 5.72134256e-02 3.73292327e-01 1.22566736e+00 -3.36278468e-01 2.78614819e-01 2.23583117e-01 -3.35877150e-01 -1.19541502e+00 4.82870579e-01 3.41172636e-01 -7.03242779e-01 -7.73540318e-01 9.70119834e-01 -8.81340057e-02 -4.62100685e-01 1.06347963e-01 -3.90581816e-01 -4.06486213e-01 -6.05443493e-02 1.49056911e-01 2.57767171e-01 -3.61449838e-01 -4.47451472e-01 -3.01524043e-01 2.46808276e-01 -3.32180023e-01 3.16390634e-01 1.55605352e+00 -7.39635378e-02 -5.01683831e-01 5.46868563e-01 1.19680667e+00 -2.37147018e-01 -9.26170349e-01 -7.08756924e-01 6.20151386e-02 -2.83166260e-01 -3.39400083e-01 -7.98564330e-02 -1.36211109e+00 5.33476472e-01 -2.46498194e-02 6.80137873e-01 9.65285838e-01 1.26926228e-01 1.21127355e+00 3.81761461e-01 3.84139061e-01 -4.97230828e-01 1.96059898e-01 8.15038562e-01 3.57986450e-01 -9.84643638e-01 -3.15577507e-01 -5.73035002e-01 -1.44460604e-01 1.02850544e+00 7.97689080e-01 -1.01522483e-01 1.41654706e+00 -4.68470575e-03 -5.31872213e-01 -4.56569940e-01 -1.06783199e+00 -9.96219516e-02 1.29769817e-01 3.16437572e-01 3.30782503e-01 1.78909078e-01 -2.76937306e-01 6.58955514e-01 5.77440448e-02 -2.51445621e-01 4.07034189e-01 7.19515383e-01 -1.06824063e-01 -1.33946824e+00 4.46060330e-01 8.74072552e-01 -1.45124182e-01 -4.22663867e-01 -2.49575555e-01 7.11720228e-01 -2.37345710e-01 7.91311026e-01 1.23903744e-01 -9.84609783e-01 5.89830205e-02 -2.74457663e-01 1.10580586e-01 -7.33266830e-01 -5.56828082e-01 -3.56758028e-01 1.60825506e-01 -5.54769397e-01 -3.67362410e-01 -4.71265674e-01 -1.38299823e+00 -9.91949439e-01 -1.58838019e-01 4.56713408e-01 -6.17848448e-02 5.44378877e-01 5.15433133e-01 6.89157903e-01 6.72698438e-01 -4.42413688e-01 4.23854142e-02 -9.37022269e-01 -5.52546382e-01 3.51599753e-01 1.95356980e-01 -4.32249308e-01 -3.49538058e-01 -3.26905280e-01]

[7.233029842376709, 5.972745418548584]

6818b0ed-44ba-41cc-b883-09d89f906f4f

physics-informed-neural-networks-for-9

2207.14230

https://arxiv.org/abs/2207.14230v1

https://arxiv.org/pdf/2207.14230v1.pdf

Physics-informed neural networks for diffraction tomography

We propose a physics-informed neural network as the forward model for tomographic reconstructions of biological samples. We demonstrate that by training this network with the Helmholtz equation as a physical loss, we can predict the scattered field accurately. It will be shown that a pretrained network can be fine-tuned for different samples and used for solving the scattering problem much faster than other numerical solutions. We evaluate our methodology with numerical and experimental results. Our physics-informed neural networks can be generalized for any forward and inverse scattering problem.

['Demetri Psaltis', 'Ahmed B. Ayoub', 'Carlo Gigli', 'Amirhossein Saba']

2022-07-28

['tomographic-reconstructions']

['medical']

[ 4.80478257e-01 -6.94841426e-03 4.82501000e-01 -7.36350358e-01 -6.03459179e-01 2.37831876e-01 -1.85510062e-03 -4.81759429e-01 -5.82169652e-01 1.14836991e+00 -1.02715425e-01 -3.03714365e-01 -2.81802982e-01 -8.85299146e-01 -1.07687187e+00 -1.08884943e+00 -2.06013739e-01 8.16417933e-01 1.85631126e-01 -3.17915171e-01 -5.09620756e-02 9.34773266e-01 -7.78034627e-01 3.34532171e-01 3.49415213e-01 1.00662100e+00 2.95531005e-01 7.96534717e-01 7.24320486e-02 1.02948248e+00 -3.24446440e-01 1.98939443e-01 8.48103240e-02 -4.33652937e-01 -9.70602691e-01 -6.46684110e-01 2.42915750e-01 -7.51338422e-01 -6.98392391e-01 8.42427492e-01 7.49964595e-01 2.36529574e-01 7.66953945e-01 -6.01133168e-01 -8.72527540e-01 7.25401223e-01 -2.63866037e-01 4.05149847e-01 3.57303806e-02 4.02081273e-02 3.98476809e-01 -1.05819058e+00 8.76006365e-01 1.10350668e+00 1.38594306e+00 1.13821661e+00 -1.32524967e+00 -5.60427010e-01 -2.98856944e-01 2.55664140e-01 -1.08359110e+00 -3.52534920e-01 8.84446204e-01 -6.46794168e-03 8.62080336e-01 1.73361316e-01 7.12364197e-01 1.03009164e+00 7.82733858e-01 5.05685925e-01 1.06943202e+00 -5.74344873e-01 2.73508310e-01 -1.43034816e-01 3.88040096e-01 7.59729385e-01 2.53470927e-01 3.97690356e-01 -4.38305020e-01 -3.29175711e-01 9.84730899e-01 -1.66538492e-01 -7.94665217e-01 -1.40282243e-01 -1.02269232e+00 6.22769356e-01 7.93574035e-01 2.21609369e-01 -4.50045109e-01 7.03002036e-01 1.23407573e-01 6.70013279e-02 5.15680194e-01 7.06303060e-01 -3.48216176e-01 4.93364364e-01 -8.93783271e-01 4.07325208e-01 1.15125227e+00 5.95746160e-01 6.11167610e-01 3.05868596e-01 2.26692453e-01 5.74829161e-01 2.85534263e-01 1.00229335e+00 7.92534836e-03 -1.56019664e+00 -5.12559712e-01 -4.72154886e-01 3.87357116e-01 -7.55226791e-01 -5.96372545e-01 -3.96534652e-01 -1.24406111e+00 4.01149541e-01 4.40204561e-01 -4.26823050e-01 -1.12720847e+00 1.62482440e+00 2.51799434e-01 5.97374976e-01 1.76881894e-01 1.16487467e+00 1.30224288e+00 8.89505863e-01 -1.73359931e-01 -4.57851350e-01 6.14244580e-01 -5.27016580e-01 -9.46260273e-01 -4.15515713e-02 3.06162089e-01 -5.06909192e-01 3.88693154e-01 4.53961700e-01 -1.52158988e+00 -1.36929154e-01 -1.04815030e+00 -1.25313967e-01 1.28794491e-01 -4.88295346e-01 7.48190165e-01 1.44159392e-01 -1.11608577e+00 1.41141450e+00 -1.24185121e+00 -2.18026966e-01 6.66798234e-01 4.73943084e-01 1.05276369e-01 -6.50988743e-02 -1.23452818e+00 8.43316913e-01 -1.42233595e-01 3.66005450e-01 -9.48122859e-01 -1.11830378e+00 -3.84948105e-01 -1.57651424e-01 -4.59514499e-01 -1.02037704e+00 1.21630466e+00 -4.64149028e-01 -1.77632964e+00 6.73659682e-01 -1.07922181e-01 -4.29584950e-01 1.11715861e-01 8.58919472e-02 -3.06618899e-01 4.74886805e-01 -1.96271151e-01 8.32931846e-02 4.85177606e-01 -1.77698457e+00 1.53739929e-01 -1.10826511e-02 -2.72652268e-01 -1.80590272e-01 6.36392534e-02 -2.21481144e-01 2.49471352e-01 -2.75481313e-01 5.24400949e-01 -7.42249548e-01 -6.40523374e-01 6.35016859e-01 -4.41657960e-01 6.01041436e-01 7.34418809e-01 -5.75345993e-01 2.63162494e-01 -1.43746138e+00 1.95848480e-01 2.45682344e-01 5.56711495e-01 2.88565040e-01 -1.51544407e-01 1.95597947e-01 -2.17319988e-02 -3.78629774e-01 -4.03893977e-01 -1.23687454e-01 -4.38430935e-01 3.55714768e-01 -3.45446944e-01 7.68054128e-01 -2.62706429e-01 1.05603313e+00 -8.39510500e-01 -2.21697912e-01 6.42866492e-02 9.60864961e-01 -6.53203130e-01 1.93891943e-01 -8.70843455e-02 7.88605154e-01 -6.37843907e-01 2.91320324e-01 1.03878164e+00 -8.29605758e-01 8.18517208e-02 -5.36384463e-01 2.46189043e-01 1.13140726e-02 -5.94554126e-01 1.44605136e+00 -6.42270744e-01 7.34125614e-01 6.52686119e-01 -1.28048158e+00 5.67091286e-01 2.20475763e-01 6.86502218e-01 -5.49348295e-01 4.51706439e-01 3.34412217e-01 -1.31443873e-01 -4.89627451e-01 -3.16508859e-02 -1.05560362e+00 5.59833407e-01 7.20333278e-01 2.09598452e-01 -2.80246377e-01 -5.53495347e-01 6.97589293e-02 1.22533274e+00 -7.24254549e-03 -1.55318558e-01 -6.32502735e-01 4.12789345e-01 2.74166256e-01 2.42005661e-01 9.44074214e-01 -3.58617343e-02 6.82074487e-01 -3.35677594e-01 -1.03905404e+00 -9.84938979e-01 -1.24354160e+00 -5.43960929e-01 6.34805799e-01 2.95858473e-01 4.38378185e-01 -7.56702960e-01 -4.38222811e-02 -6.25682399e-02 6.95264280e-01 -6.36816263e-01 -1.69800162e-01 -9.81299996e-01 -1.22804773e+00 5.85953057e-01 4.44491953e-01 2.42058992e-01 -1.19137836e+00 -4.74183947e-01 3.15218091e-01 -2.73312241e-01 -1.24514353e+00 1.19617410e-01 6.53236508e-01 -8.15879166e-01 -1.00538051e+00 -9.62863803e-01 -6.97596967e-01 5.61778724e-01 5.08993566e-02 1.19402564e+00 3.62935543e-01 -4.27134335e-01 4.66244876e-01 9.59587619e-02 -3.90342414e-01 -7.73205698e-01 -4.82386351e-01 5.64835258e-02 -4.02220458e-01 1.27330393e-01 -1.03565931e+00 -8.04266870e-01 -1.07743874e-01 -6.58865511e-01 -1.24456771e-01 1.34204611e-01 1.06976366e+00 7.72669196e-01 -5.69398291e-02 1.41184673e-01 -1.02010429e+00 4.01021153e-01 -1.96316749e-01 -3.82459193e-01 -2.01645717e-02 -2.82733381e-01 3.02286178e-01 8.49735498e-01 -5.27953804e-01 -1.26060474e+00 -1.11640856e-01 -6.42624319e-01 -4.23154414e-01 -1.03854127e-01 3.16961259e-01 4.48534906e-01 -1.19446707e+00 9.23207164e-01 4.20842022e-01 3.42752300e-02 -2.99124181e-01 -1.37312129e-01 1.14206970e-01 5.53617775e-01 -6.65338159e-01 9.51206982e-01 1.04921043e+00 6.96356058e-01 -1.11763906e+00 -1.14559805e+00 -1.11285336e-01 -2.96296537e-01 -3.24563742e-01 7.81941175e-01 -3.85850102e-01 -1.03483796e+00 6.27911329e-01 -1.39205492e+00 -6.57256722e-01 -5.10218322e-01 7.10369885e-01 -7.46599674e-01 2.28711069e-01 -1.21960270e+00 -6.62555814e-01 -4.87515777e-01 -9.03920472e-01 7.76233912e-01 -5.06941192e-02 2.66347062e-02 -1.57230556e+00 3.24283272e-01 -3.50003429e-02 6.69158936e-01 3.89999986e-01 7.73985386e-01 -1.47840440e-01 -6.82845950e-01 1.22400165e-01 -2.92623788e-01 1.63547516e-01 -2.06788018e-01 -2.13583827e-01 -1.17451286e+00 -2.69057602e-01 7.75372446e-01 -6.41419470e-01 1.16793275e+00 1.24662101e+00 1.30090809e+00 -2.66075611e-01 -7.48362005e-01 1.42933929e+00 1.66194165e+00 -8.77371728e-02 5.70320308e-01 -5.19316494e-02 6.60764337e-01 8.19045082e-02 -2.79495776e-01 1.68136299e-01 -7.28119016e-02 4.66688611e-02 2.95149237e-01 -2.77528286e-01 -3.44937503e-01 2.79436588e-01 -2.54026204e-01 9.53168452e-01 -2.64233887e-01 -4.39250231e-01 -8.68039131e-01 9.70771089e-02 -1.36243081e+00 -1.06417906e+00 -4.18412209e-01 1.49939477e+00 7.53614128e-01 -2.20843658e-01 -6.91633940e-01 3.51317674e-02 3.40766609e-01 -1.01002827e-01 -7.99073040e-01 -8.60514566e-02 -8.10129754e-03 7.53832698e-01 6.52848482e-01 9.08444881e-01 -7.30061054e-01 4.61224914e-01 8.75694752e+00 4.40156698e-01 -1.43629992e+00 3.80067855e-01 5.24983704e-01 -1.65171772e-01 -4.63375777e-01 -5.20318270e-01 -4.98896927e-01 3.36270110e-04 8.86622667e-01 7.21516684e-02 6.38450563e-01 2.74251699e-01 1.94455087e-01 4.53953929e-02 -1.13485873e+00 7.17860758e-01 -2.86604941e-01 -1.88104737e+00 1.84924737e-01 -1.45682961e-01 7.23681927e-01 4.73186672e-01 -5.78439683e-02 -2.21335068e-01 5.57128668e-01 -1.08520448e+00 2.97668129e-01 8.91272724e-01 7.63880849e-01 -4.65492040e-01 6.65866792e-01 4.21099007e-01 -5.85696936e-01 4.06582266e-01 -7.20901430e-01 -5.73454350e-02 4.75920290e-01 9.67108130e-01 -4.99881089e-01 1.66173428e-01 8.71134877e-01 5.96466839e-01 4.91119832e-01 8.40251684e-01 4.65198010e-01 6.30473971e-01 -5.84945738e-01 -2.74779797e-01 1.78465277e-01 1.01508185e-01 7.43455410e-01 1.09192050e+00 3.98340285e-01 7.40857065e-01 -1.19491123e-01 1.41096830e+00 -4.01385799e-02 -3.37409973e-01 -3.79155219e-01 3.34327221e-01 -7.52397776e-02 9.56574380e-01 -7.76927114e-01 -5.68087548e-02 1.29167810e-01 6.58474028e-01 2.87652910e-01 1.00603926e+00 -7.26976216e-01 -1.54055625e-01 4.46420521e-01 4.29113448e-01 2.28520676e-01 5.06809354e-02 -1.89862713e-01 -1.06467652e+00 -4.52925026e-01 -1.41065940e-01 -3.35043997e-01 -1.21019876e+00 -1.61272609e+00 7.78784156e-01 2.54865326e-02 -5.89437783e-01 1.56115294e-01 -1.19090319e+00 -5.06413460e-01 9.75268543e-01 -1.64594841e+00 -8.28785717e-01 -3.59646559e-01 5.73476493e-01 -2.83124298e-01 1.18066773e-01 1.08212829e+00 1.60170540e-01 -7.65564442e-02 1.39724910e-01 4.88923728e-01 2.36530334e-01 2.39576936e-01 -1.01404858e+00 3.02587628e-01 4.23075020e-01 -5.05178928e-01 6.87120795e-01 1.14258635e+00 -5.11586487e-01 -1.54469514e+00 -1.08670890e+00 4.91833061e-01 3.81354205e-02 7.48126388e-01 -2.31335964e-02 -1.22206819e+00 8.31263900e-01 1.49941429e-01 7.36389756e-01 4.72483456e-01 -1.61566526e-01 3.24811786e-02 4.76176552e-02 -1.60909367e+00 1.25810519e-01 1.09658611e+00 -2.32876763e-01 -2.59881169e-01 8.15964341e-01 5.50121844e-01 -7.74740458e-01 -8.87331486e-01 4.08153981e-01 5.27819157e-01 -9.70036089e-01 1.42865396e+00 -4.90623444e-01 6.59687757e-01 3.16511095e-01 -1.42825902e-01 -1.43132699e+00 -6.58032596e-01 -3.77600461e-01 -1.66648373e-01 6.45661876e-02 2.34552786e-01 -6.52543545e-01 1.06676710e+00 3.82341623e-01 -2.64416654e-02 -8.11264753e-01 -1.05285048e+00 -6.96889758e-01 6.80863023e-01 -4.97356921e-01 2.05562338e-01 6.93728447e-01 -3.84997725e-01 2.78128330e-02 -3.24070126e-01 4.53058273e-01 1.37966251e+00 1.09739140e-01 -3.02661620e-02 -1.31426871e+00 -4.41876292e-01 -3.23436745e-02 -4.51305769e-02 -1.28013337e+00 3.81149650e-01 -9.63129461e-01 6.87599301e-01 -1.56015599e+00 3.40540737e-01 -5.71335077e-01 -2.77491719e-01 2.98224622e-03 3.23552966e-01 5.84987581e-01 -3.38254005e-01 1.47448331e-01 -2.62070715e-01 5.03482342e-01 1.85279334e+00 -2.35503584e-01 1.93735063e-01 -7.04246610e-02 -1.40556335e-01 1.00122261e+00 6.00017965e-01 -7.88439095e-01 -8.92118737e-02 -6.29710436e-01 1.89474493e-01 3.18268597e-01 9.64290917e-01 -1.53928745e+00 3.06322396e-01 -1.52651072e-01 8.12904894e-01 -4.88574803e-01 8.96075726e-01 -8.46531391e-01 5.40947691e-02 8.70106459e-01 -5.07459939e-01 -8.83075237e-01 3.46258372e-01 5.49578786e-01 3.41439903e-01 -2.98965901e-01 1.20629311e+00 -5.46222627e-01 -5.21242879e-02 4.83773291e-01 -5.49220562e-01 2.02831566e-01 3.73206973e-01 4.92468253e-02 1.22876670e-02 -6.18497372e-01 -1.12577009e+00 -2.97681808e-01 2.82770842e-01 -7.66776681e-01 1.00316000e+00 -1.27249157e+00 -6.67289972e-01 3.66202623e-01 -5.28027356e-01 -2.22670704e-01 2.74370879e-01 5.84542572e-01 -1.23018062e+00 1.41419604e-01 -3.01349849e-01 -8.21154356e-01 -8.22107732e-01 2.88846552e-01 1.17391062e+00 -2.05702156e-01 -7.93345511e-01 1.17923605e+00 1.28903836e-01 -8.42913091e-01 2.21319757e-02 -5.43378055e-01 1.50310412e-01 -1.07802296e+00 7.42552698e-01 2.59963691e-01 -2.91919976e-01 -4.64929283e-01 -3.84231508e-01 8.37326705e-01 3.37544173e-01 5.64968437e-02 1.97250712e+00 -6.28409488e-03 -3.14642221e-01 3.81536573e-01 1.38476181e+00 -4.69172388e-01 -1.20711207e+00 -3.68789673e-01 -9.15172756e-01 2.09752589e-01 6.77060544e-01 -7.68208921e-01 -1.50885820e+00 9.76842701e-01 4.74581480e-01 1.43671349e-01 1.01039636e+00 -1.11521654e-01 1.34739935e+00 1.02967310e+00 5.04029393e-01 -5.58611095e-01 -1.93252698e-01 6.22667730e-01 7.82156289e-01 -1.10921168e+00 3.01127672e-01 -5.04081607e-01 2.44663671e-01 1.33241189e+00 2.53198028e-01 -6.64367259e-01 1.54860580e+00 1.16986942e+00 1.37440935e-01 -3.96716535e-01 -5.30167162e-01 5.79742968e-01 1.06922448e-01 8.94799590e-01 3.52196008e-01 -2.68304795e-01 3.42389017e-01 1.93671465e-01 1.22249566e-01 6.66069984e-01 6.41768694e-01 9.01935160e-01 -6.23582125e-01 -6.49899960e-01 -3.64886761e-01 4.57368612e-01 -4.52349901e-01 -1.54874787e-01 -6.70849904e-02 6.27481163e-01 -8.19512382e-02 4.21130598e-01 -8.33981931e-02 -9.93743539e-02 -3.57986987e-02 -1.80075765e-01 1.14170873e+00 -3.34191084e-01 -2.35234663e-01 -1.03635281e-01 -3.21828902e-01 -6.93991065e-01 -8.65145743e-01 -2.21440867e-01 -1.79187155e+00 -8.05003643e-01 -1.68659508e-01 6.85710385e-02 2.12758571e-01 1.04562485e+00 -6.58496544e-02 8.43359292e-01 1.35391116e-01 -1.39109015e+00 -5.04810929e-01 -5.95048964e-01 -7.80053735e-01 7.12975338e-02 9.61151421e-01 -5.32310903e-01 -6.12741113e-01 5.50473481e-02]

[12.58035659790039, -2.624521255493164]

be55c91e-67d0-4e3b-b929-c1600de78c1b

the-pose-knows-video-forecasting-by

1705.00053

http://arxiv.org/abs/1705.00053v1

http://arxiv.org/pdf/1705.00053v1.pdf

The Pose Knows: Video Forecasting by Generating Pose Futures

Current approaches in video forecasting attempt to generate videos directly in pixel space using Generative Adversarial Networks (GANs) or Variational Autoencoders (VAEs). However, since these approaches try to model all the structure and scene dynamics at once, in unconstrained settings they often generate uninterpretable results. Our insight is to model the forecasting problem at a higher level of abstraction. Specifically, we exploit human pose detectors as a free source of supervision and break the video forecasting problem into two discrete steps. First we explicitly model the high level structure of active objects in the scene---humans---and use a VAE to model the possible future movements of humans in the pose space. We then use the future poses generated as conditional information to a GAN to predict the future frames of the video in pixel space. By using the structured space of pose as an intermediate representation, we sidestep the problems that GANs have in generating video pixels directly. We show through quantitative and qualitative evaluation that our method outperforms state-of-the-art methods for video prediction.

['Martial Hebert', 'Abhinav Gupta', 'Kenneth Marino', 'Jacob Walker']

2017-04-28

the-pose-knows-video-forecasting-by-1

http://openaccess.thecvf.com/content_iccv_2017/html/Walker_The_Pose_Knows_ICCV_2017_paper.html

http://openaccess.thecvf.com/content_ICCV_2017/papers/Walker_The_Pose_Knows_ICCV_2017_paper.pdf

iccv-2017-10

['human-pose-forecasting']

['computer-vision']

[ 4.22259808e-01 6.26185954e-01 9.93024260e-02 -3.38868886e-01 -5.29087663e-01 -5.88148415e-01 8.92913401e-01 -8.45006466e-01 3.58757526e-02 8.79234195e-01 5.12845218e-01 2.38437485e-03 6.68034554e-01 -9.85528111e-01 -1.37970674e+00 -6.57754242e-01 2.57581860e-01 5.82691491e-01 1.74135953e-01 -2.64080524e-01 -1.38279274e-01 4.23713148e-01 -1.65116632e+00 6.26442015e-01 4.72051650e-01 8.12992096e-01 -5.57694258e-03 1.00120819e+00 2.16684550e-01 1.39316320e+00 -5.56627035e-01 -5.19817770e-01 4.12976652e-01 -8.56644154e-01 -6.81977153e-01 5.78825176e-01 5.33297181e-01 -6.38532281e-01 -7.59194851e-01 7.17742145e-01 -1.91337522e-02 1.93533704e-01 7.96511769e-01 -1.39251304e+00 -3.97748947e-01 3.47555637e-01 -2.98465073e-01 -2.96547294e-01 7.36060798e-01 5.27856648e-01 6.56077147e-01 -6.81756198e-01 1.17706001e+00 1.33150041e+00 6.62808776e-01 1.01706588e+00 -1.45718932e+00 -3.28534245e-01 3.19830745e-01 1.30977035e-01 -1.13894963e+00 -4.14590716e-01 1.07026231e+00 -7.75914073e-01 7.51136184e-01 2.61538506e-01 1.13263535e+00 1.68778062e+00 3.26586485e-01 9.91879463e-01 7.83718884e-01 -3.40841025e-01 2.50038862e-01 -1.63515985e-01 -6.63741291e-01 8.16093326e-01 -3.79540443e-01 4.86997098e-01 -5.46321869e-01 8.12317654e-02 1.11233294e+00 1.27907649e-01 -2.38062069e-01 -5.82737923e-01 -1.22473466e+00 9.15802360e-01 5.11654139e-01 -1.16708398e-01 -7.72158384e-01 6.27218127e-01 3.97804528e-02 -7.95352273e-03 4.41940457e-01 4.77462173e-01 -1.89928725e-01 6.18267246e-02 -1.05812407e+00 7.06292629e-01 6.77128971e-01 8.55532050e-01 6.84706092e-01 4.12439644e-01 -3.93741369e-01 1.40878707e-01 2.33671665e-01 2.13021338e-01 2.49924019e-01 -1.36942649e+00 2.35444874e-01 1.11350179e-01 3.81442666e-01 -9.16005731e-01 1.23767547e-01 -2.95880437e-02 -5.96626818e-01 5.10357082e-01 3.88373494e-01 -4.38629568e-01 -1.40759385e+00 1.80824053e+00 3.84444058e-01 5.47789931e-01 -1.85636923e-01 8.85777652e-01 3.24445248e-01 1.18524086e+00 1.33501574e-01 -1.10196635e-01 8.89867663e-01 -1.07544827e+00 -5.92292309e-01 -2.58541435e-01 1.35246187e-01 -3.51023436e-01 7.07210124e-01 3.56603712e-01 -1.44200003e+00 -9.26578224e-01 -8.50720108e-01 -2.95679215e-02 -1.47746354e-01 -1.15551531e-01 5.30838192e-01 3.09078753e-01 -1.34041631e+00 6.63292944e-01 -1.30519581e+00 -8.33646208e-02 3.68112236e-01 2.30619222e-01 -2.04648897e-01 3.56063284e-02 -1.04956353e+00 7.90653646e-01 1.93877876e-01 7.61457533e-02 -1.54672229e+00 -6.09333813e-01 -1.03293371e+00 -4.22560610e-02 4.21315342e-01 -1.22882533e+00 1.27599847e+00 -1.52190936e+00 -1.51394868e+00 5.81425607e-01 -2.91504234e-01 -6.41271889e-01 8.57323825e-01 -1.86840281e-01 -4.42369878e-02 1.35285914e-01 -2.12622851e-01 1.32836127e+00 1.07510448e+00 -1.69155419e+00 -5.15799165e-01 -8.01435560e-02 3.34918261e-01 2.51360834e-01 2.21226364e-01 -3.37126493e-01 -5.10371923e-01 -9.39575613e-01 -1.99228659e-01 -1.22103882e+00 -4.79502797e-01 8.10008720e-02 -5.11171103e-01 1.13393776e-01 9.00592446e-01 -9.95348930e-01 7.97397614e-01 -1.66342485e+00 7.15298116e-01 -6.80197543e-03 2.04386845e-01 5.23780435e-02 -4.03456055e-02 3.17876577e-01 -1.31637856e-01 -5.14739007e-02 -3.96743380e-02 -6.93320870e-01 -1.36266768e-01 4.58041012e-01 -4.72101092e-01 1.37562960e-01 3.68849665e-01 1.26061678e+00 -9.20452356e-01 -4.05483663e-01 5.13116300e-01 8.12156796e-01 -1.02380002e+00 5.73387742e-01 -9.22375023e-01 9.59614515e-01 -4.77693409e-01 3.35383743e-01 2.41168484e-01 -1.26386851e-01 2.39543334e-01 -2.31048331e-01 1.59027413e-01 -1.12915346e-02 -7.80838311e-01 1.78570807e+00 -2.98367113e-01 7.08338797e-01 -2.70716071e-01 -1.03675616e+00 4.25863147e-01 3.93558770e-01 5.28922737e-01 -8.14553201e-02 -6.97780848e-02 -3.73787761e-01 -2.65616685e-01 -3.76850247e-01 4.94190961e-01 -3.51334721e-01 -1.21872383e-03 8.55758265e-02 1.71400040e-01 -2.24686846e-01 -7.31462464e-02 1.67848676e-01 1.02950573e+00 1.02162707e+00 -1.51853159e-01 2.73630977e-01 1.92993835e-01 1.81300342e-01 5.56526363e-01 5.85819840e-01 1.84407160e-01 9.94059265e-01 5.90957403e-01 -7.33470440e-01 -1.40015566e+00 -1.10540915e+00 5.38698316e-01 7.40155876e-01 -1.50838584e-01 -3.40279251e-01 -1.14598501e+00 -7.11137533e-01 -3.68046135e-01 9.36630368e-01 -1.09757662e+00 -1.29823402e-01 -7.96530366e-01 -2.57861227e-01 1.09285384e-01 7.86038816e-01 1.95287362e-01 -1.28212106e+00 -8.31575394e-01 2.32466593e-01 -3.44863296e-01 -1.07015145e+00 -3.12462032e-01 -1.78738996e-01 -6.29323423e-01 -8.11308742e-01 -8.40771258e-01 -5.87183774e-01 7.71482527e-01 -3.77823025e-01 1.43508780e+00 -5.66373952e-02 -3.53570551e-01 5.00206113e-01 -2.71176994e-01 -2.00987682e-01 -8.20715547e-01 -4.28925484e-01 -1.25992388e-01 2.33704984e-01 -6.95278719e-02 -5.63024044e-01 -7.89400041e-01 4.45802063e-02 -9.13322628e-01 6.52784348e-01 2.14236215e-01 8.95651519e-01 7.51184583e-01 -4.57595773e-02 -1.65116880e-02 -1.14240897e+00 -1.11569520e-02 -3.57759863e-01 -5.81855059e-01 1.95181608e-01 -6.93344995e-02 1.20829359e-01 6.40637100e-01 -5.37821531e-01 -1.32304752e+00 6.62254870e-01 -1.85099065e-01 -1.02855027e+00 -4.23480302e-01 2.08829731e-01 -2.38748819e-01 3.25721025e-01 6.56566560e-01 3.89338225e-01 -3.15516330e-02 -1.48441613e-01 5.15263319e-01 -8.84990692e-02 7.39808619e-01 -5.61613023e-01 9.07770693e-01 5.49492359e-01 1.33839428e-01 -5.82587779e-01 -8.85318935e-01 2.86660790e-01 -7.08750904e-01 -6.28093123e-01 1.37442422e+00 -1.11280918e+00 -4.49516654e-01 3.86983067e-01 -1.31592786e+00 -6.97684169e-01 -4.66467172e-01 5.97281829e-02 -1.09120345e+00 1.01586878e-02 -6.41006827e-01 -7.91583955e-01 2.46927574e-01 -1.23019230e+00 1.37382913e+00 6.94632307e-02 -3.03510755e-01 -1.04938960e+00 4.39690575e-02 3.53910327e-01 4.06373106e-02 9.96042490e-01 5.80434382e-01 -3.12602073e-02 -1.00068402e+00 -2.09406212e-01 4.60817337e-01 2.66663164e-01 -1.86393842e-01 7.67135471e-02 -8.06491971e-01 -6.10269457e-02 -2.33999379e-02 -2.15317905e-01 6.73587620e-01 6.83339894e-01 1.46915042e+00 -5.90057909e-01 -4.73422915e-01 6.51970088e-01 1.29902387e+00 1.91155568e-01 1.05044329e+00 -8.03875849e-02 9.99123275e-01 5.71339726e-01 5.16146481e-01 4.90190208e-01 2.62681752e-01 8.33346546e-01 5.83982110e-01 -1.16567358e-01 -2.74983793e-01 -1.04282808e+00 5.11166930e-01 2.23492473e-01 -4.00931120e-01 -6.56679928e-01 -7.48182297e-01 3.28978002e-01 -1.97489095e+00 -1.30756545e+00 1.71560153e-01 1.89186656e+00 6.08335853e-01 3.26478258e-02 3.79146248e-01 -1.57979205e-01 6.62450552e-01 1.90425262e-01 -5.79448760e-01 -2.44997088e-02 1.72532886e-01 6.72954246e-02 2.67332733e-01 5.25685251e-01 -1.01024270e+00 1.03971577e+00 6.83617544e+00 3.56501430e-01 -9.25083339e-01 -1.08725086e-01 1.00911450e+00 -2.40366757e-01 -3.71603549e-01 2.07000554e-01 -5.20431161e-01 5.41270256e-01 9.69129741e-01 2.43930638e-01 6.81085944e-01 7.89698184e-01 2.02460110e-01 1.27758741e-01 -1.58263278e+00 8.01217973e-01 1.36558548e-01 -1.74283755e+00 4.56548870e-01 2.31387675e-01 1.04065144e+00 -2.13258713e-01 1.11934170e-01 2.57748306e-01 5.50411224e-01 -1.15580559e+00 1.21697998e+00 1.08216083e+00 7.48511970e-01 -6.02670550e-01 3.84116173e-01 6.29168093e-01 -7.92811930e-01 7.89950639e-02 -5.94175681e-02 -1.77521020e-01 7.04139650e-01 5.31928428e-02 -6.80616021e-01 1.72392979e-01 3.27249348e-01 6.51210070e-01 -3.51186842e-01 4.50974852e-01 -4.10082340e-01 5.46405256e-01 -1.54753193e-01 3.55279386e-01 1.82650268e-01 -1.44840777e-01 4.16421711e-01 7.81786382e-01 4.70119596e-01 3.46483618e-01 2.38723397e-01 1.07651627e+00 -7.29862154e-02 -6.79486394e-01 -7.11550176e-01 -8.25842321e-02 5.83586097e-02 8.44687045e-01 -4.58778828e-01 -5.80152810e-01 -2.14358807e-01 1.30988228e+00 1.02758847e-01 5.86015701e-01 -1.08730197e+00 2.88531303e-01 6.06841624e-01 4.78931069e-01 4.89178896e-01 -1.92685455e-01 4.56171967e-02 -1.41081727e+00 -8.02427381e-02 -1.05387509e+00 3.29736695e-02 -1.33907986e+00 -8.59668791e-01 6.36366367e-01 4.92159948e-02 -1.22995842e+00 -1.16120100e+00 -6.00178540e-01 -6.03702962e-01 7.27479100e-01 -7.88301408e-01 -1.49982095e+00 -1.90668240e-01 6.40927672e-01 9.31366324e-01 -2.16204654e-02 7.10936368e-01 -1.91979468e-01 -2.21949816e-01 2.43781224e-01 -2.23879829e-01 2.46389270e-01 5.05386665e-02 -1.20354414e+00 7.21705675e-01 9.84248161e-01 5.07028103e-01 2.57105291e-01 1.09212410e+00 -7.91267037e-01 -1.39941657e+00 -1.16318631e+00 5.62141240e-01 -9.44486678e-01 2.95433640e-01 -5.70764422e-01 -4.69928652e-01 1.16725063e+00 1.86278448e-01 1.54487580e-01 1.49025127e-01 -3.30779761e-01 -5.64030372e-02 3.76447290e-01 -1.02214706e+00 8.09111893e-01 1.12855828e+00 -5.15396655e-01 -3.88677418e-01 4.07458246e-01 8.36394131e-01 -6.37746096e-01 -5.77235222e-01 2.46968955e-01 5.59331834e-01 -1.13456607e+00 1.18240535e+00 -8.76471996e-01 1.12092400e+00 -2.00347409e-01 -1.29483804e-01 -1.41809154e+00 -2.33633295e-01 -8.01183343e-01 -4.28102285e-01 9.17384863e-01 2.48555973e-01 -8.79982393e-03 1.30893254e+00 9.06420350e-01 -1.09935164e-01 -6.89470708e-01 -5.49383998e-01 -3.32079023e-01 -1.79412197e-02 -4.04424429e-01 5.51665604e-01 6.06483757e-01 -4.23431128e-01 2.32660115e-01 -9.66301322e-01 -5.86299831e-03 5.44115722e-01 -3.65507640e-02 1.00882602e+00 -7.66548991e-01 -7.10502744e-01 1.24951579e-01 -5.23340762e-01 -1.25926387e+00 3.69057357e-01 -3.39290798e-01 3.34070832e-01 -1.45389652e+00 5.70467636e-02 1.40647953e-02 1.26584351e-01 3.33467305e-01 -5.20559959e-02 3.37681085e-01 4.37106073e-01 8.39143470e-02 -4.56902951e-01 6.76328719e-01 1.41526830e+00 -6.80393428e-02 1.27649968e-02 -8.04190338e-02 -2.27827355e-01 9.40391183e-01 2.97630966e-01 -2.58401513e-01 -5.88532507e-01 -5.48002124e-01 7.77471364e-02 7.33438313e-01 7.14706719e-01 -1.09617579e+00 -7.72322714e-02 -4.18964088e-01 9.37335372e-01 -4.55815971e-01 9.19720471e-01 -7.46944308e-01 7.77188897e-01 3.69609922e-01 -4.75584328e-01 1.99456662e-02 -1.54166371e-01 8.01712275e-01 -1.46482527e-01 1.11304425e-01 5.39037824e-01 -4.59591389e-01 -6.72518909e-01 5.41471004e-01 -5.04228354e-01 -2.22586542e-01 1.11853659e+00 -3.17570657e-01 2.15307951e-01 -9.97680128e-01 -1.19602370e+00 -2.67453734e-02 8.71644616e-01 5.31471133e-01 5.08462846e-01 -1.33598220e+00 -4.82859254e-01 2.69528151e-01 -3.02972198e-01 1.97811544e-01 3.42596024e-01 1.78995162e-01 -7.08341241e-01 2.38961115e-01 -3.67037654e-01 -5.54017663e-01 -9.04155195e-01 7.79534101e-01 4.08900201e-01 -1.94009066e-01 -5.64317226e-01 9.06286716e-01 6.62779570e-01 -2.04999279e-02 1.00228541e-01 2.59599909e-02 1.35627091e-01 -2.58545846e-01 2.66714513e-01 6.98979537e-04 -4.81369942e-01 -9.34107065e-01 1.01837613e-01 1.85697764e-01 2.17608556e-01 -4.28481787e-01 1.41824210e+00 -2.87952330e-02 2.08146304e-01 3.89335632e-01 1.01543927e+00 -2.42063284e-01 -2.10240078e+00 2.41953507e-01 -4.65833962e-01 -5.27596653e-01 -8.92481655e-02 -6.84717417e-01 -1.09289706e+00 8.39642704e-01 3.11113596e-01 1.37233183e-01 9.72481012e-01 5.40053993e-02 7.31244266e-01 4.02067304e-02 5.06805778e-01 -8.48837793e-01 1.38208613e-01 1.35088235e-01 9.41190124e-01 -9.24074233e-01 -2.75507957e-01 -3.30992222e-01 -8.91224802e-01 9.56174493e-01 7.64261782e-01 -3.69701952e-01 3.87630343e-01 2.51584619e-01 -2.81110629e-02 -1.43824262e-03 -9.45364118e-01 1.94719061e-01 3.52530092e-01 7.90700257e-01 4.27428246e-01 2.39431318e-02 3.43446374e-01 1.72404855e-01 -3.29774529e-01 2.22913653e-01 5.18745661e-01 7.73534775e-01 -1.93574131e-02 -1.15529191e+00 -3.73231351e-01 3.91241789e-01 -2.69737810e-01 2.48318687e-01 -3.58535498e-01 6.06863916e-01 3.03732038e-01 5.96317112e-01 1.95594132e-01 -5.75909376e-01 1.39842497e-03 1.74478278e-01 7.64020443e-01 -5.77611923e-01 -1.84125498e-01 1.21835321e-01 -5.27769281e-03 -9.08786595e-01 -4.93989736e-01 -9.59678113e-01 -8.88801217e-01 -1.56130046e-01 1.11290939e-01 -1.64797246e-01 4.64479327e-01 9.05458808e-01 2.81472117e-01 7.48633862e-01 4.64156121e-01 -1.60337269e+00 -3.31229001e-01 -6.78869903e-01 -2.10131139e-01 6.92847610e-01 4.45605040e-01 -6.12589896e-01 -1.43050715e-01 8.25307965e-01]

[10.731184959411621, -0.649075984954834]

9f846c9e-1e59-498c-b11f-f0773eec2b17

designing-game-of-theorems

1906.08549

https://arxiv.org/abs/1906.08549v1

https://arxiv.org/pdf/1906.08549v1.pdf

Designing Game of Theorems

"Theorem proving is similar to the game of Go. So, we can probably improve our provers using deep learning, like DeepMind built the super-human computer Go program, AlphaGo." Such optimism has been observed among participants of AITP2017. But is theorem proving really similar to Go? In this paper, we first identify the similarities and differences between them and then propose a system in which various provers keep competing against each other and changing themselves until they prove conjectures provided by users.

['Yutaka Nagashima']

2019-06-20

['game-of-go']

['playing-games']

[-5.93817532e-01 4.14844632e-01 -1.05457031e-03 -1.16008617e-01 -3.65115464e-01 -7.25849509e-01 3.13569099e-01 1.69398859e-02 -4.22686636e-02 1.14335573e+00 -1.42572418e-01 -1.09619677e+00 -1.71911776e-01 -1.16125500e+00 -1.17302918e+00 -3.65684450e-01 -2.61552185e-01 8.77103209e-01 4.74066883e-01 -7.73293018e-01 3.66466194e-01 -1.73761159e-01 -1.25397217e+00 6.80785120e-01 7.91439831e-01 4.59855497e-01 -1.06986135e-01 1.05257547e+00 7.04538003e-02 1.53730404e+00 -4.56331819e-01 -8.95950615e-01 4.37206089e-01 -5.05248785e-01 -1.79350054e+00 -5.70435405e-01 4.30367082e-01 -5.16850114e-01 -4.21153486e-01 1.57646203e+00 8.53845403e-02 -1.13624975e-01 -1.47024527e-01 -1.92882705e+00 -5.60540974e-01 1.47330320e+00 -2.97807813e-01 2.71403193e-01 6.86593831e-01 4.33451027e-01 1.42504680e+00 1.25448048e-01 7.45314538e-01 1.43892741e+00 7.26477563e-01 8.80140901e-01 -1.20024741e+00 -9.24679756e-01 -1.66883141e-01 1.00867736e+00 -1.22070646e+00 -8.51357952e-02 8.17007959e-01 -2.91501611e-01 9.51083660e-01 4.32323039e-01 6.24678433e-01 1.09408617e+00 5.72742462e-01 9.83856857e-01 1.00320077e+00 -3.71193141e-01 1.80358380e-01 2.24250868e-01 1.10617224e-02 1.03566456e+00 4.11899894e-01 7.98173547e-02 -3.86406511e-01 -2.03937784e-01 3.10414165e-01 -6.63437009e-01 -1.14297360e-01 -4.42667335e-01 -1.34881222e+00 1.01124978e+00 3.69530916e-01 5.70754349e-01 -2.44307682e-01 4.78586316e-01 5.69410622e-01 8.71047974e-01 2.05845028e-01 1.07152057e+00 -7.63729155e-01 -1.12996913e-01 -9.91901398e-01 8.28414321e-01 1.34226859e+00 7.85159945e-01 4.21985477e-01 -4.23111349e-01 1.15472794e-01 -2.39670232e-01 2.63600588e-01 1.61154911e-01 9.33343619e-02 -1.50522316e+00 3.76549661e-01 4.57327873e-01 1.65285096e-01 -1.14734566e+00 -4.89051580e-01 -3.03827852e-01 -7.48503625e-01 3.42553407e-01 6.58376276e-01 -5.53937793e-01 -1.81916356e-01 1.57475507e+00 1.97948515e-01 3.87509763e-01 4.70090546e-02 8.64371657e-01 7.49208689e-01 5.35788596e-01 -1.85007989e-01 1.27750799e-01 9.46913719e-01 -8.77171516e-01 -5.23212075e-01 1.68479338e-01 8.70194316e-01 -2.32589573e-01 6.10361218e-01 9.93331611e-01 -1.34341097e+00 -4.28186268e-01 -1.29651797e+00 6.96976529e-03 -2.68326163e-01 -6.81303203e-01 1.18100893e+00 3.52749020e-01 -1.34101200e+00 1.23473716e+00 -7.58080542e-01 3.18424664e-02 4.13055032e-01 6.81471705e-01 -2.40913570e-01 5.96278980e-02 -1.67301130e+00 1.14342523e+00 9.61302578e-01 9.08065811e-02 -1.08657670e+00 -4.24538374e-01 -5.28073430e-01 2.16721267e-01 8.92985523e-01 -1.08778536e+00 1.72800076e+00 -8.94160807e-01 -1.40513945e+00 7.74215043e-01 -1.05354739e-02 -9.72713470e-01 7.24017978e-01 -1.64986596e-01 -9.59645137e-02 -2.73109823e-01 1.97449133e-01 4.84517664e-01 1.42407432e-01 -1.15964735e+00 -9.70177114e-01 -1.53108314e-01 1.18208396e+00 -2.88987279e-01 5.89537323e-01 1.59184650e-01 3.37356150e-01 3.20135415e-01 1.50873497e-01 -6.97918534e-01 -3.15728307e-01 -1.82873920e-01 -4.62050140e-01 -9.91874397e-01 4.85538721e-01 -4.48997855e-01 1.13533413e+00 -1.69180048e+00 1.62494227e-01 2.67004132e-01 8.99944603e-01 1.21109262e-01 1.25304565e-01 1.25124395e-01 -1.14645310e-01 5.54162323e-01 5.76965988e-01 4.24521685e-01 4.39394653e-01 2.18555510e-01 3.28325108e-03 3.15207362e-01 5.55981584e-02 1.05956459e+00 -1.60181475e+00 -7.12778330e-01 1.11405477e-01 -4.22527879e-01 -8.56939793e-01 -6.18027598e-02 -9.22392309e-01 1.75511345e-01 -3.69815320e-01 5.86159229e-01 8.75703871e-01 -5.38492441e-01 5.77618897e-01 2.32437626e-01 -2.59285606e-02 6.81465149e-01 -1.12206459e+00 1.53663862e+00 1.23155579e-01 8.12896609e-01 1.75429076e-01 -1.28925276e+00 4.49081987e-01 6.45740747e-01 5.49893863e-02 -6.68646753e-01 5.07735431e-01 2.14988127e-01 7.79691279e-01 -6.75615907e-01 2.13927686e-01 -2.92115688e-01 4.83375117e-02 4.50675666e-01 1.27829388e-01 -3.55800599e-01 5.87525904e-01 3.61233056e-01 1.23676896e+00 2.00727925e-01 2.16061905e-01 -4.12718803e-01 6.45034552e-01 4.14841861e-01 4.92414325e-01 1.35374570e+00 -3.39342535e-01 -2.31561154e-01 1.08649731e+00 -9.87957835e-01 -1.17876768e+00 -7.96760201e-01 3.37471664e-01 1.09412336e+00 3.33882213e-01 -6.76277161e-01 -7.18340397e-01 -9.29975986e-01 -5.56590371e-02 8.48393202e-01 -4.55106258e-01 -1.59068659e-01 -5.14404893e-01 6.29681796e-02 7.81206667e-01 1.91981152e-01 8.37682247e-01 -1.29158390e+00 -4.70499933e-01 3.08285862e-01 -6.01655126e-01 -5.79553902e-01 4.37707067e-01 3.89941752e-01 -4.69007969e-01 -1.50871325e+00 4.46250848e-02 -6.01756632e-01 -1.47204459e-01 1.01194039e-01 1.37909687e+00 5.51143169e-01 3.68946970e-01 -4.71003979e-01 -3.36382300e-01 -6.74330354e-01 -8.61128032e-01 2.83841431e-01 1.77116245e-01 -8.55337918e-01 5.83200216e-01 -6.66930437e-01 -1.52558744e-01 -1.21978864e-01 -3.82690847e-01 3.45507830e-01 3.35225224e-01 7.18174756e-01 -3.35516810e-01 6.16531491e-01 1.85766533e-01 -9.29937422e-01 7.29907572e-01 -4.93473083e-01 -8.40089738e-01 4.41734970e-01 -5.29093623e-01 3.78438294e-01 9.20527041e-01 -2.00112090e-01 -4.52592105e-01 -5.29727519e-01 -2.08382621e-01 -1.84451118e-01 -5.66898845e-02 5.20040214e-01 -1.69355974e-01 1.70247629e-02 8.60173702e-01 -1.43886982e-02 -3.28971446e-01 8.96304026e-02 1.60664573e-01 1.93757415e-01 6.04561627e-01 -1.12955868e+00 1.15823352e+00 -8.01282562e-03 -2.48744935e-02 1.57009382e-02 -1.01251054e+00 -1.00311579e-03 -2.85114229e-01 -3.80148962e-02 5.00356793e-01 -4.58444595e-01 -1.92902994e+00 1.67900771e-01 -1.75808883e+00 -5.09726882e-01 -6.02028854e-02 2.50241250e-01 -5.45332074e-01 2.65199184e-01 -5.97636402e-01 -7.92167783e-01 -3.18730414e-01 -9.15308416e-01 4.04348761e-01 3.20369393e-01 -5.33810139e-01 -9.08905029e-01 3.64862770e-01 4.64202315e-01 4.38734740e-01 1.82983756e-01 9.28892016e-01 -9.06389475e-01 -5.16350746e-01 7.14876205e-02 -4.12509620e-01 3.71678919e-01 -1.34950191e-01 -7.37901181e-02 -7.70372093e-01 -1.61910936e-01 -1.44147426e-01 -2.81960338e-01 2.75074422e-01 2.88706303e-01 1.30036294e+00 -7.57974565e-01 -1.57150447e-01 3.49130660e-01 1.23624289e+00 2.33704686e-01 7.48904943e-01 9.20284688e-01 3.04820389e-01 1.93497211e-01 2.38081306e-01 5.60973547e-02 5.72659016e-01 2.14786790e-02 6.41739845e-01 2.31279358e-01 4.77202743e-01 -3.41419071e-01 4.77670394e-02 1.49675220e-01 -4.19023544e-01 -1.64706349e-01 -1.18017101e+00 4.00249481e-01 -2.09362006e+00 -1.51902616e+00 -4.42030340e-01 1.38285911e+00 1.19988072e+00 7.58185565e-01 1.16093755e-01 3.99152637e-01 5.76159418e-01 -1.95314124e-01 -1.96635008e-01 -6.86951041e-01 3.55521031e-02 6.20370448e-01 3.52268785e-01 6.59674346e-01 -9.56207931e-01 1.17418611e+00 7.04275131e+00 5.09363055e-01 -1.00337362e+00 2.11558379e-02 3.39180708e-01 2.22733319e-01 -1.47884682e-01 5.00238717e-01 -2.04938054e-01 3.17469537e-01 9.51053143e-01 -6.86295331e-01 7.53951430e-01 1.03323734e+00 1.19319215e-01 -1.33315846e-01 -1.61050141e+00 4.32001442e-01 -1.93652064e-01 -1.77763486e+00 -1.33742571e-01 -1.66536123e-01 6.52212441e-01 5.24058975e-02 -4.67279494e-01 8.54337156e-01 1.02517152e+00 -1.25833356e+00 7.84367204e-01 3.21878672e-01 1.80784278e-02 -9.17829812e-01 1.31803632e+00 6.55840099e-01 -4.14726615e-01 -1.77292097e-02 -2.53750622e-01 -8.51053476e-01 -3.99947405e-01 1.98073704e-02 -1.25997794e+00 7.81583846e-01 7.78234243e-01 1.21110164e-01 -5.14929473e-01 1.16975164e+00 -5.63637495e-01 6.87600613e-01 -2.11656064e-01 -4.89477634e-01 4.69429940e-01 1.93442225e-01 6.06083989e-01 1.04452705e+00 -3.03736657e-01 1.96361601e-01 1.96638748e-01 1.44111192e+00 -2.37393215e-01 -5.82365453e-01 -4.89153892e-01 -5.08637428e-02 4.36677225e-02 1.06599545e+00 -4.53353882e-01 -7.08276510e-01 -5.23788705e-02 5.70334673e-01 4.80719879e-02 -1.20596439e-01 -1.14451957e+00 -4.68007296e-01 2.78016895e-01 1.79186419e-01 1.32923489e-02 4.77778725e-02 -3.06038916e-01 -1.23979390e+00 6.85252100e-02 -1.36365783e+00 1.86639950e-01 -9.39600825e-01 -1.15892851e+00 2.28224352e-01 -3.37591954e-02 -6.49776161e-01 -1.55341655e-01 -8.03009331e-01 -9.58676457e-01 5.92927337e-01 -1.28527021e+00 -9.74534810e-01 6.30587637e-02 6.42098486e-01 2.47193307e-01 1.43345326e-01 7.81714141e-01 9.36400518e-02 -1.66034773e-01 5.69338083e-01 -2.50913858e-01 4.81666982e-01 1.96307063e-01 -1.52600098e+00 4.87494171e-01 7.00765073e-01 1.30106390e-01 6.70964122e-01 1.42339754e+00 -3.30371499e-01 -1.56864285e+00 -3.29425693e-01 1.26523089e+00 -5.83081782e-01 1.36113906e+00 -2.29831904e-01 -8.54114771e-01 7.23203242e-01 3.91272843e-01 -4.31366891e-01 6.19920455e-02 8.26250732e-01 -4.14394766e-01 -4.94157262e-02 -1.06955290e+00 7.00458765e-01 8.69824886e-01 -6.37565315e-01 -1.07576036e+00 8.03089559e-01 8.57232809e-01 -5.59063911e-01 -2.22224861e-01 2.33298928e-01 6.18311167e-01 -1.11887240e+00 5.13643026e-01 -1.34200382e+00 8.51190031e-01 -5.45029044e-01 8.89896378e-02 -1.13092828e+00 -5.53837478e-01 -9.12662148e-01 -3.49214524e-01 6.48164928e-01 5.56015491e-01 -2.26633906e-01 1.03187656e+00 5.85429013e-01 1.02718249e-01 -4.57016408e-01 -9.68887091e-01 -7.71615505e-01 5.13440907e-01 -7.29894400e-01 6.40669048e-01 1.40657401e+00 9.13981736e-01 4.55303818e-01 -2.24630550e-01 3.95657808e-01 4.45971221e-01 1.13678589e-01 9.38555002e-01 -1.39277565e+00 -6.69859409e-01 -7.83243299e-01 -3.72617185e-01 -5.69544971e-01 4.27345127e-01 -1.18951726e+00 1.64666250e-01 -1.40912545e+00 4.98380721e-01 -5.59281893e-02 -5.30950129e-01 7.91042089e-01 9.82756689e-02 -2.60379732e-01 9.02631134e-02 -3.85631859e-01 -1.39604509e+00 -1.56857952e-01 1.19827771e+00 -7.12487400e-01 -3.20633613e-02 2.49195114e-01 -1.15834188e+00 9.28810954e-01 8.53316963e-01 -4.02865797e-01 1.41554251e-01 -4.30812001e-01 1.02030444e+00 2.14548975e-01 6.97370827e-01 -1.14061058e+00 6.56201780e-01 -6.08423829e-01 -6.42886478e-03 -3.78938556e-01 -5.31992137e-01 -6.32686675e-01 3.69472466e-02 8.19154859e-01 -4.16644573e-01 8.48384276e-02 3.29237729e-01 -2.25814208e-01 -5.00466973e-02 -3.28090876e-01 4.77251202e-01 -4.16691482e-01 -6.38012350e-01 -8.48236084e-02 -3.71376187e-01 -4.01666649e-02 8.33447456e-01 1.19364507e-01 -3.68235111e-01 -6.24684334e-01 -5.83141148e-01 7.86596894e-01 2.28228606e-02 1.05402991e-01 3.65905553e-01 -9.51708615e-01 -1.05550897e+00 -5.08665264e-01 -2.89540887e-01 1.71047658e-01 -3.33660357e-02 8.71733248e-01 -9.82954085e-01 5.10305226e-01 -3.46966714e-01 -2.77478874e-01 -1.26114941e+00 8.36229086e-01 7.35794067e-01 -6.18450105e-01 -5.16366899e-01 1.11479306e+00 -2.08855510e-01 -4.41049099e-01 2.16594666e-01 -6.18607521e-01 1.17672794e-01 -6.60724401e-01 7.01763272e-01 1.91556141e-01 -9.36887935e-02 4.37142789e-01 -5.47046363e-01 -3.01257998e-01 -1.59769475e-01 1.30958468e-01 1.51864135e+00 3.36868465e-01 -5.55070639e-01 3.22773963e-01 6.35104179e-01 -2.51864433e-01 -3.99991274e-01 -1.91310987e-01 3.76615554e-01 -5.23263179e-02 -1.89299747e-01 -9.25604045e-01 -5.30982614e-01 6.47229970e-01 -3.80870923e-02 6.31739914e-01 6.81528509e-01 1.92067698e-01 6.42013431e-01 1.18738306e+00 8.21907341e-01 -1.10996103e+00 -2.12150320e-01 1.09152889e+00 3.66247416e-01 -1.47257853e+00 2.80688375e-01 2.56623924e-01 -2.94633865e-01 1.14797068e+00 7.68153906e-01 -6.61365271e-01 8.48916620e-02 2.70942062e-01 -4.56396580e-01 -4.49019104e-01 -1.23590791e+00 1.31719023e-01 -4.37191613e-02 4.16580409e-01 4.97437447e-01 1.52781934e-01 -4.48643833e-01 6.34438574e-01 -8.90528202e-01 5.01188099e-01 4.94392544e-01 7.72779763e-01 -5.90502024e-01 -1.04095542e+00 -4.13090795e-01 -3.87337059e-02 -4.07327622e-01 -1.48746073e-01 -5.77996075e-01 1.13882923e+00 9.29176629e-01 1.12414706e+00 -3.37168634e-01 -7.57173359e-01 2.86537874e-02 7.50278607e-02 8.08771551e-01 -3.93223822e-01 -6.83850586e-01 -7.87971079e-01 3.49401474e-01 -5.98791540e-01 -1.91557392e-01 -2.50847191e-01 -1.08551764e+00 -1.51348269e+00 -5.60401142e-01 7.57176518e-01 3.62850696e-01 1.38930702e+00 -3.70077461e-01 5.44254899e-01 5.06164253e-01 -3.24343085e-01 -5.97811222e-01 -8.82270932e-01 -4.01305765e-01 -6.01628423e-02 3.85664403e-01 -2.35023901e-01 -4.71471161e-01 -3.70541126e-01]

[8.918739318847656, 7.043548107147217]

94ec8ba7-54f7-425e-9e55-b439f2d4ba22

time-series-classification-using-the-hidden

1506.05085

http://arxiv.org/abs/1506.05085v2

http://arxiv.org/pdf/1506.05085v2.pdf

Time Series Classification using the Hidden-Unit Logistic Model

We present a new model for time series classification, called the hidden-unit logistic model, that uses binary stochastic hidden units to model latent structure in the data. The hidden units are connected in a chain structure that models temporal dependencies in the data. Compared to the prior models for time series classification such as the hidden conditional random field, our model can model very complex decision boundaries because the number of latent states grows exponentially with the number of hidden units. We demonstrate the strong performance of our model in experiments on a variety of (computer vision) tasks, including handwritten character recognition, speech recognition, facial expression, and action recognition. We also present a state-of-the-art system for facial action unit detection based on the hidden-unit logistic model.

['Laurens van der Maaten', 'Hamdi Dibeklioğlu', 'Wenjie Pei', 'David M. J. Tax']

2015-06-16

['action-unit-detection', 'facial-action-unit-detection']

['computer-vision', 'computer-vision']

[ 5.37223458e-01 9.56336483e-02 -7.97998130e-01 -6.52528524e-01 -7.66867697e-02 -1.78491641e-02 8.33693087e-01 -5.11182487e-01 -4.01806496e-02 4.03636396e-01 5.78716993e-02 -2.84209281e-01 3.73082340e-01 -4.74581748e-01 -2.97634691e-01 -1.02239537e+00 -4.69752342e-01 2.47237742e-01 3.44550729e-01 1.95432901e-01 -1.40802162e-02 4.53997493e-01 -1.45128262e+00 5.86773455e-01 1.15272924e-01 1.12007046e+00 -2.64434874e-01 8.29220474e-01 1.20830588e-01 1.36609840e+00 -5.07114589e-01 1.89524721e-02 -3.88874590e-01 -5.32829106e-01 -5.78296840e-01 4.52036232e-01 -2.45432988e-01 -2.62180865e-01 -3.76769632e-01 7.76832223e-01 -1.20340310e-01 -4.64248769e-02 1.15462899e+00 -1.53509915e+00 -6.30114377e-01 4.94057536e-01 -2.75663316e-01 1.68568254e-01 1.79829121e-01 -1.30608857e-01 8.70052934e-01 -5.60713410e-01 5.40117800e-01 1.65182066e+00 7.15267479e-01 6.72305465e-01 -1.46704960e+00 -5.88874996e-01 3.95359039e-01 3.78199756e-01 -1.08959424e+00 -5.32837808e-01 6.56387031e-01 -7.23231018e-01 1.45487273e+00 -4.27677259e-02 4.51086909e-01 1.74852383e+00 7.39108443e-01 1.08005404e+00 1.27169025e+00 -6.77669585e-01 4.13484901e-01 -4.23272282e-01 5.95201850e-01 9.34892714e-01 -4.24589455e-01 1.13321938e-01 -6.18384778e-01 -3.56740415e-01 1.04017007e+00 2.91143239e-01 3.93523037e-01 7.90482610e-02 -8.55264127e-01 9.58818913e-01 -3.06916744e-01 2.38810614e-01 -3.19067776e-01 5.41915536e-01 7.67467394e-02 3.88452977e-01 6.73883975e-01 -5.98720789e-01 -2.56705761e-01 -1.32340193e-01 -9.98986363e-01 -4.84662831e-01 8.90892684e-01 8.07907939e-01 4.78984863e-01 2.81664729e-01 -2.34641686e-01 4.88359839e-01 6.54165447e-01 4.15132076e-01 8.03436756e-01 -1.04696536e+00 -5.47780730e-02 3.52157891e-01 -1.26705021e-01 -5.81224859e-01 -1.65669709e-01 7.64146373e-02 -9.39543486e-01 5.10134041e-01 1.38425335e-01 -1.68677196e-02 -1.50606740e+00 1.68985069e+00 -1.09144226e-01 3.93412769e-01 5.01948968e-02 1.85185820e-01 2.97118634e-01 9.27877963e-01 2.79154450e-01 -7.29162395e-01 1.14965022e+00 -7.73569345e-01 -1.24218655e+00 1.85045581e-02 2.82531828e-01 -4.72532183e-01 2.30246946e-01 3.67569715e-01 -8.29508185e-01 -6.14642084e-01 -8.60059738e-01 1.63530931e-02 -2.54664302e-01 1.40784889e-01 9.57954168e-01 6.24851584e-01 -7.97759533e-01 7.58069754e-01 -1.86656237e+00 -3.90388876e-01 1.28392011e-01 4.79739219e-01 -2.77051270e-01 2.88721919e-01 -9.78365302e-01 7.82206595e-01 1.25165107e-02 2.85142660e-01 -1.34390497e+00 3.82923335e-01 -9.26043272e-01 -2.56858058e-02 1.11099128e-02 -1.67815238e-01 1.41530180e+00 -1.08998120e+00 -1.98660183e+00 6.68191850e-01 -7.41724670e-01 -3.07996869e-01 2.54951835e-01 1.46163166e-01 -5.86056650e-01 -2.88597532e-02 -2.54584163e-01 5.34827948e-01 1.43230569e+00 -5.77162325e-01 -3.01818728e-01 -3.15330356e-01 -5.33233523e-01 -4.55899209e-01 -1.54790163e-01 5.49529374e-01 -1.59822553e-01 -7.30210721e-01 3.26844513e-01 -1.12474263e+00 -4.14003104e-01 -2.73754448e-02 -2.45163560e-01 -6.67645097e-01 1.14468408e+00 -4.79190588e-01 1.21476424e+00 -2.25671387e+00 2.91412026e-01 -6.11022161e-03 5.14527038e-02 -3.62064540e-01 6.25282750e-02 5.22520006e-01 -4.29538697e-01 -1.67317633e-02 -1.04569077e-01 -8.53407323e-01 1.10863678e-01 8.47361624e-01 -6.10802472e-01 6.49360478e-01 2.93497860e-01 8.87291491e-01 -5.23794234e-01 -3.90451968e-01 -6.40447140e-02 5.33894181e-01 4.58552153e-04 4.49056961e-02 -2.27600053e-01 1.80376440e-01 -3.51904362e-01 8.40540648e-01 -1.24794021e-01 -5.32192528e-01 1.45653114e-01 4.31321889e-01 -2.44356111e-01 3.08753163e-01 -8.65945160e-01 1.16269863e+00 1.16412170e-01 8.28396857e-01 -4.27827030e-01 -9.12483275e-01 8.02083910e-01 9.72985685e-01 4.46664751e-01 -1.20571561e-01 1.78776607e-01 -1.66178554e-01 -1.98738217e-01 -3.57895166e-01 -5.48533462e-02 -3.73454630e-01 -2.49823257e-01 6.98848367e-01 4.19071168e-01 4.42600340e-01 2.83613428e-02 -2.71966398e-01 1.15434849e+00 2.64083356e-01 2.91032732e-01 2.34076202e-01 9.60756913e-02 -4.08216208e-01 7.33988166e-01 5.72989941e-01 -3.49257469e-01 3.23620677e-01 9.83171940e-01 -5.49955130e-01 -7.02031076e-01 -8.80239248e-01 -9.86138508e-02 1.19506907e+00 -6.47767365e-01 -4.22983795e-01 -3.93826514e-01 -4.09619093e-01 -2.21824527e-01 4.11388606e-01 -9.80910361e-01 -1.77005693e-01 -2.36348242e-01 -8.34462106e-01 6.65267050e-01 9.33578372e-01 2.22218126e-01 -1.37436152e+00 -5.60857117e-01 2.54534721e-01 1.47109888e-02 -1.19861186e+00 -2.73729414e-01 5.98224282e-01 -1.10999608e+00 -6.24196351e-01 -6.12742305e-01 -7.55642951e-01 6.55418158e-01 -4.33324724e-01 5.61524212e-01 -5.39708436e-01 -4.13086474e-01 4.60540265e-01 -2.64881760e-01 -4.16296005e-01 -6.23437107e-01 -4.29440230e-01 2.46220708e-01 5.80055058e-01 5.43002188e-01 -5.24884820e-01 -1.02716774e-01 5.07640541e-01 -8.88564169e-01 -1.46448659e-02 5.16461372e-01 7.82896340e-01 4.71712202e-01 2.20458642e-01 -3.32587324e-02 -5.32788754e-01 3.91140193e-01 -3.64859074e-01 -6.26313329e-01 3.34913105e-01 -4.81471479e-01 2.71586806e-01 -5.57435094e-04 -1.10551703e+00 -8.90173733e-01 4.28932935e-01 1.51153103e-01 -7.24121571e-01 -2.67302930e-01 6.81474864e-01 2.91253865e-01 2.72041619e-01 7.01972395e-02 3.59285206e-01 2.62918055e-01 -5.59841156e-01 6.91182241e-02 6.03150547e-01 3.80399853e-01 -5.15110902e-02 8.44018459e-02 7.59485781e-01 2.20323175e-01 -1.03050852e+00 -5.64071119e-01 -3.46310943e-01 -1.08250058e+00 -1.80870831e-01 1.34851193e+00 -7.54140615e-01 -7.08946168e-01 9.72269118e-01 -1.27092922e+00 -6.16094828e-01 -2.66587645e-01 7.91055322e-01 -8.47186923e-01 4.14775670e-01 -1.25550890e+00 -1.38272798e+00 1.71889693e-01 -8.75426054e-01 1.17528677e+00 -1.35552451e-01 -5.10668039e-01 -1.32967138e+00 3.62823486e-01 -2.44575426e-01 1.86701342e-02 4.04677898e-01 1.08935452e+00 -2.16511592e-01 -3.98421615e-01 -4.34675992e-01 4.82586235e-01 5.62712967e-01 2.18773335e-01 5.53027093e-01 -8.85437429e-01 -6.10151188e-03 3.03816915e-01 -2.78530478e-01 1.17860293e+00 5.41803122e-01 9.35223043e-01 -3.17368865e-01 -5.19835353e-01 3.94263387e-01 7.05786824e-01 3.14410746e-01 7.03854024e-01 -2.99225092e-01 3.92081559e-01 5.77449501e-01 1.02525935e-01 5.63795507e-01 9.48812217e-02 4.90946561e-01 1.55244440e-01 -1.75462008e-01 1.81378290e-01 -2.04502150e-01 1.13638210e+00 8.82810533e-01 -5.22332609e-01 -2.26044655e-01 -9.44081008e-01 2.85480082e-01 -2.07969904e+00 -1.32439637e+00 -1.84826016e-01 1.81151903e+00 7.06314921e-01 3.47131759e-01 7.56212547e-02 2.23164871e-01 6.45220995e-01 1.82260528e-01 -4.05147970e-01 -4.35239404e-01 -1.11705981e-01 2.72928089e-01 2.64653176e-01 4.90139484e-01 -1.27224374e+00 1.00000882e+00 8.69248009e+00 4.18050498e-01 -1.29834127e+00 1.19698592e-01 5.67907453e-01 2.33719423e-01 3.62617731e-01 1.13617443e-01 -1.12850356e+00 3.68295461e-01 1.34786201e+00 1.92120627e-01 -5.62709346e-02 6.41339481e-01 9.12472084e-02 -2.95367204e-02 -1.36034882e+00 8.02456677e-01 1.11466028e-01 -9.04883862e-01 -2.87614521e-02 4.10840511e-01 5.57244718e-01 3.40740196e-02 3.58851254e-01 2.69613534e-01 5.95706761e-01 -1.13389301e+00 6.87108338e-01 6.41275942e-01 9.66775417e-01 -2.62040943e-01 3.17558616e-01 6.01094246e-01 -1.02679229e+00 -2.15693399e-01 -2.26826042e-01 -4.54400957e-01 2.54518658e-01 2.19496265e-01 -6.52380347e-01 -2.50969887e-01 5.63096046e-01 9.19092238e-01 -3.08635205e-01 3.77359241e-01 -6.31779969e-01 1.12649870e+00 -3.25211555e-01 -6.27436675e-04 2.36972585e-01 2.29243524e-02 2.99482286e-01 1.10598719e+00 1.44885793e-01 3.22251767e-01 3.02112341e-01 6.03540003e-01 5.15315592e-01 -4.97240841e-01 -5.30728579e-01 -6.11616373e-01 -4.41018701e-01 8.53635311e-01 -1.02970636e+00 -4.37773675e-01 -5.35026610e-01 1.19725752e+00 3.44065167e-02 6.47743523e-01 -7.65259922e-01 5.52871358e-03 4.12274390e-01 -2.37721410e-02 6.43418610e-01 -8.25791121e-01 2.51863807e-01 -1.29573953e+00 -2.33137324e-01 -3.32020432e-01 6.20865941e-01 -7.46059537e-01 -1.49988484e+00 8.57560277e-01 3.66811186e-01 -1.15793002e+00 -8.18926573e-01 -9.05563056e-01 -5.86625636e-01 6.60438120e-01 -1.17571139e+00 -1.24077976e+00 2.73049653e-01 8.43140244e-01 4.81290370e-01 -1.39611945e-01 1.27248335e+00 -3.39003861e-01 -6.65199220e-01 2.57515192e-01 3.80284414e-02 4.76059645e-01 3.96505088e-01 -1.03853643e+00 4.57308263e-01 6.19450510e-01 1.82459950e-01 6.35116279e-01 5.38399816e-01 -8.52794170e-01 -9.67382312e-01 -8.11230600e-01 1.37026179e+00 -5.77089489e-01 9.62879241e-01 -7.28566468e-01 -6.87081337e-01 1.29778719e+00 9.01310891e-02 1.14980072e-01 1.15954137e+00 2.71720827e-01 -4.10805047e-01 3.30132514e-01 -5.46079457e-01 4.22685862e-01 7.09067285e-01 -8.61920536e-01 -6.98069930e-01 3.65011334e-01 4.42752302e-01 1.90851614e-01 -7.49656439e-01 3.65904659e-01 9.31339920e-01 -4.80883688e-01 5.52225649e-01 -1.00534773e+00 2.91781962e-01 1.20407350e-01 -1.87586099e-01 -7.45788693e-01 -8.19896877e-01 -9.85638559e-01 -6.39946461e-01 7.34400630e-01 2.73079842e-01 -5.90830505e-01 8.29463303e-01 7.24648774e-01 2.82307297e-01 -4.61014986e-01 -1.27110255e+00 -9.69553053e-01 -2.58004963e-01 -3.95380765e-01 -2.87341643e-02 9.09831405e-01 2.07134664e-01 5.58847666e-01 -5.46005368e-01 1.22037351e-01 6.34744406e-01 1.46764264e-01 1.36553422e-01 -1.32645261e+00 -4.52640116e-01 -3.17859262e-01 -8.42179418e-01 -1.15269208e+00 4.90039498e-01 -5.46599030e-01 -4.61767949e-02 -1.13943446e+00 1.68674976e-01 4.90652360e-02 -6.07460260e-01 1.04206967e+00 3.00920516e-01 1.16811231e-01 -1.72549039e-01 6.43524885e-01 -7.12442160e-01 6.83652222e-01 6.35795355e-01 -1.03607021e-01 -2.28017300e-01 5.00001490e-01 2.11579517e-01 9.73123610e-01 4.77629632e-01 -7.88664341e-01 -6.67335764e-02 -1.54453292e-01 -1.28334135e-01 5.63910306e-01 2.91295171e-01 -8.02146852e-01 3.38604242e-01 -3.95574242e-01 5.96859396e-01 -6.73616767e-01 7.04279780e-01 -7.35711098e-01 6.63874224e-02 5.36741972e-01 -4.71327633e-01 6.52825758e-02 -2.29731590e-01 6.81130826e-01 -4.36799198e-01 6.62559224e-03 5.23990929e-01 -6.41740561e-02 -7.11320877e-01 4.13917631e-01 -1.19808602e+00 -9.56486821e-01 1.14162767e+00 -2.27036864e-01 -4.64996770e-02 -6.73527420e-01 -1.54467618e+00 -8.53591338e-02 -1.00246102e-01 5.11992097e-01 6.28674805e-01 -1.31457782e+00 -5.15159786e-01 5.87821782e-01 -2.75686737e-02 -7.31025994e-01 -2.79746115e-01 8.52600813e-01 8.01556930e-02 5.61706603e-01 -3.67064834e-01 -9.67555404e-01 -1.61515141e+00 6.26226366e-01 1.55161038e-01 -4.17685241e-01 -3.64656180e-01 6.90389395e-01 9.38892365e-02 2.02232629e-01 6.09243870e-01 -7.68090844e-01 -3.93226087e-01 2.19658092e-01 5.48335850e-01 3.20872396e-01 -4.29459125e-01 -7.34191120e-01 -2.57136911e-01 2.96804041e-01 -3.58476378e-02 -6.16473615e-01 1.33156216e+00 1.01334900e-01 -3.56350511e-01 1.41032434e+00 1.07425249e+00 -6.33481383e-01 -1.50681722e+00 -2.08128482e-01 1.35238692e-01 1.64565235e-01 7.96074122e-02 -3.89784366e-01 -4.43637013e-01 1.09367180e+00 6.12629294e-01 3.37251246e-01 9.40993369e-01 1.60552412e-01 2.29905620e-01 4.55866903e-01 4.15787160e-01 -1.06937075e+00 1.59008309e-01 7.93387413e-01 7.19063282e-01 -1.03862011e+00 6.39454834e-03 -3.47922742e-01 -5.44612408e-01 1.32048643e+00 3.89203504e-02 -5.42217009e-02 1.26663554e+00 4.99562323e-01 -9.05677117e-03 -1.13169849e-01 -1.49822235e+00 -2.05783561e-01 2.09486619e-01 3.53105038e-01 4.47983086e-01 2.00950503e-01 1.23745687e-01 6.14047647e-01 2.51271844e-01 2.93661922e-01 3.69544536e-01 1.43593526e+00 -4.06855255e-01 -1.24082983e+00 -3.31561804e-01 -6.22084253e-02 -6.22968733e-01 2.00190619e-01 -4.07488436e-01 2.64893830e-01 -2.14824453e-01 9.03517604e-01 3.81098479e-01 -3.15968454e-01 -6.77228197e-02 6.92647994e-01 6.07985139e-01 -7.96760261e-01 -3.52229029e-02 6.04109406e-01 -1.67742535e-01 -5.77968836e-01 -6.64868653e-01 -9.69087481e-01 -1.13683617e+00 8.69196132e-02 -2.94363081e-01 -2.28415340e-01 7.57950485e-01 9.84543562e-01 -1.59572661e-02 3.39470625e-01 6.32432520e-01 -7.27871716e-01 -5.93635082e-01 -1.20606124e+00 -6.52533710e-01 8.67364779e-02 5.39906144e-01 -6.59950078e-01 -3.78103465e-01 7.79157579e-01]

[8.491016387939453, 1.0129456520080566]

0639852c-2629-4522-880e-12e659e2bacb

towards-understanding-generalization-of-macro

2305.05248

https://arxiv.org/abs/2305.05248v2

https://arxiv.org/pdf/2305.05248v2.pdf

Towards Understanding Generalization of Macro-AUC in Multi-label Learning

Macro-AUC is the arithmetic mean of the class-wise AUCs in multi-label learning and is commonly used in practice. However, its theoretical understanding is far lacking. Toward solving it, we characterize the generalization properties of various learning algorithms based on the corresponding surrogate losses w.r.t. Macro-AUC. We theoretically identify a critical factor of the dataset affecting the generalization bounds: \emph{the label-wise class imbalance}. Our results on the imbalance-aware error bounds show that the widely-used univariate loss-based algorithm is more sensitive to the label-wise class imbalance than the proposed pairwise and reweighted loss-based ones, which probably implies its worse performance. Moreover, empirical results on various datasets corroborate our theory findings. To establish it, technically, we propose a new (and more general) McDiarmid-type concentration inequality, which may be of independent interest.

['Yilong Yin', 'Chongxuan Li', 'Guoqiang Wu']

2023-05-09

['multi-label-learning']

['methodology']

[ 1.27185658e-01 8.33928883e-02 -4.73759770e-01 -5.66392839e-01 -9.47336435e-01 -4.39199686e-01 -6.98385462e-02 7.95180917e-01 -4.38792855e-01 1.13864362e+00 -5.02605975e-01 -1.28641158e-01 -7.61026442e-01 -6.07731044e-01 -7.21807063e-01 -1.11558843e+00 -9.68478471e-02 4.41428930e-01 1.93682779e-02 3.62688005e-02 2.94321448e-01 2.81857461e-01 -1.61090159e+00 4.24256250e-02 1.25657976e+00 1.23687506e+00 -4.81588870e-01 1.44511551e-01 2.74738103e-01 6.76375508e-01 -5.53097367e-01 -6.67991996e-01 3.24428260e-01 -4.59375292e-01 -7.31716514e-01 -1.34803012e-01 4.26311672e-01 6.43210858e-02 5.27765095e-01 1.28817928e+00 4.33124453e-01 -2.60017533e-02 1.07282257e+00 -1.64227688e+00 -3.62869352e-01 6.12505913e-01 -1.10585880e+00 2.24627703e-01 4.93608341e-02 -6.00950420e-01 1.15019596e+00 -6.01922095e-01 1.57534823e-01 9.72368896e-01 8.90994787e-01 1.73770204e-01 -1.14040732e+00 -8.37009430e-01 -1.83391813e-02 2.94750422e-01 -1.56791127e+00 6.87870234e-02 6.54003024e-01 -4.55269277e-01 5.06543033e-02 4.18791771e-01 3.98172848e-02 5.85105419e-01 3.13650161e-01 5.35142124e-01 1.23576999e+00 -6.50721312e-01 1.32962018e-01 5.37204266e-01 2.78497040e-01 6.04559720e-01 9.76646006e-01 -1.16102986e-01 -2.87390262e-01 -4.69126821e-01 2.45530590e-01 -2.17348769e-01 -2.68039346e-01 -7.37902284e-01 -6.42126977e-01 9.00621653e-01 2.03761403e-02 2.06783712e-01 -1.29273668e-01 -6.61416650e-02 4.88176346e-01 4.61828619e-01 8.18786681e-01 4.93969657e-02 -4.56007421e-01 2.35811651e-01 -9.36387241e-01 2.99177945e-01 6.07177377e-01 6.44349158e-01 6.66746855e-01 -4.97255534e-01 -1.00717701e-01 8.58904958e-01 3.69852841e-01 1.75845429e-01 1.46688029e-01 -6.36860609e-01 6.82088435e-01 5.26884139e-01 2.63954401e-01 -1.10707438e+00 -5.93674242e-01 -8.20852458e-01 -7.59305179e-01 7.87288398e-02 9.75943863e-01 -8.39198977e-02 6.98797405e-02 1.90800655e+00 5.35081446e-01 1.38829663e-01 -8.11436251e-02 7.27099240e-01 1.34661183e-01 5.56838512e-02 3.73728782e-01 -8.58259141e-01 9.92348552e-01 -5.52992284e-01 -6.40061438e-01 4.17789459e-01 9.30518448e-01 -5.88630855e-01 8.95281732e-01 6.40372097e-01 -1.01371074e+00 -2.36507311e-01 -1.13553703e+00 3.72455359e-01 -1.34811729e-01 1.77447736e-01 4.22811717e-01 1.12705576e+00 -5.56861162e-01 6.73480988e-01 -2.63432115e-01 -2.18990609e-01 4.08993691e-01 1.91453665e-01 -2.73418128e-01 -2.58768685e-02 -1.18033707e+00 8.18745792e-01 4.09913778e-01 -1.20845653e-01 -3.52764159e-01 -9.85555172e-01 -4.86823112e-01 -1.08981565e-01 2.25169271e-01 -5.53323269e-01 9.55068171e-01 -8.93455327e-01 -1.01893306e+00 1.04225421e+00 2.08991140e-01 -3.50702167e-01 1.01709437e+00 2.50379927e-02 -2.56459206e-01 1.97663426e-01 4.95145172e-02 -5.03854975e-02 6.15255296e-01 -1.33966267e+00 -6.94708765e-01 -7.30999053e-01 1.42067418e-01 1.22049600e-01 -5.20251453e-01 -7.28320852e-02 5.55604219e-01 -3.94781858e-01 7.92817306e-03 -5.46765387e-01 -1.32012278e-01 -5.48828626e-03 -4.07336652e-01 -4.51752096e-01 2.60070354e-01 -1.72311664e-01 1.44178426e+00 -2.00504160e+00 -2.44006485e-01 4.78272259e-01 9.31968912e-02 9.40565020e-02 1.38694748e-01 3.83496374e-01 -2.03809351e-01 2.51975298e-01 -2.14921698e-01 -1.02937557e-01 3.18034180e-02 -2.25290924e-01 -7.80501068e-02 1.01452351e+00 -1.07330777e-01 3.53941202e-01 -8.19543600e-01 -7.42431283e-01 -5.53395487e-02 1.85093015e-01 -4.21841204e-01 -8.34564269e-02 -1.86971892e-02 2.82576174e-01 -4.79184330e-01 5.09400189e-01 1.11382282e+00 -2.78247297e-01 3.91952813e-01 -3.01545560e-01 1.81360487e-02 -4.00338948e-01 -1.15147555e+00 1.11376405e+00 -3.44217330e-01 -3.55770737e-02 -1.53517500e-01 -1.47901452e+00 9.52188849e-01 1.53232038e-01 6.79299772e-01 -4.65923309e-01 2.77957737e-01 5.54007351e-01 -1.10155441e-01 -3.23018134e-01 1.60510823e-01 -7.79436171e-01 -6.10398315e-02 4.05056328e-01 -2.46805832e-01 3.89968872e-01 2.72060961e-01 -1.76092669e-01 6.19430125e-01 -2.39558429e-01 7.55431712e-01 -7.45554745e-01 8.20467889e-01 -3.67674172e-01 5.42998493e-01 6.46337986e-01 -5.01066864e-01 3.96136254e-01 8.52213383e-01 -1.09109648e-01 -8.17059577e-01 -1.12098026e+00 -7.61246085e-01 1.00765514e+00 4.16733295e-01 -5.53378602e-03 -7.08381653e-01 -7.97307372e-01 3.70677382e-01 6.12604201e-01 -6.56240284e-01 -1.85714141e-01 -1.25662044e-01 -1.41638947e+00 7.40981460e-01 2.49095753e-01 2.54484683e-01 -2.19099954e-01 -4.78881508e-01 -3.88900712e-02 -2.10449010e-01 -6.53009772e-01 -1.87646881e-01 1.49920478e-01 -8.87524128e-01 -1.60733199e+00 -8.32617044e-01 -3.65288615e-01 6.88409209e-01 6.39391458e-03 1.06942105e+00 -3.28387581e-02 -2.51942202e-02 2.19484702e-01 -3.68048131e-01 -4.93101865e-01 -3.23401481e-01 2.12518554e-02 1.48908287e-01 1.85640529e-01 4.13672209e-01 -6.48738086e-01 -5.71449280e-01 5.72360098e-01 -8.96449566e-01 -5.53366840e-01 4.09610122e-01 6.35860622e-01 8.19367230e-01 3.78425777e-01 1.19236672e+00 -1.06867707e+00 5.85946202e-01 -7.71055639e-01 -5.44903398e-01 7.67278075e-01 -1.07989240e+00 -2.49057692e-02 6.63231790e-01 -2.75966972e-01 -8.65744233e-01 -2.86687762e-01 1.54701322e-01 -1.76835316e-03 2.91644614e-02 4.00717616e-01 -3.62459958e-01 -1.94590405e-01 4.61522877e-01 -1.10907473e-01 -1.71939477e-01 -3.85450691e-01 2.13266656e-01 7.06681609e-01 6.34972975e-02 -8.69089544e-01 4.84095544e-01 4.86626655e-01 4.20045853e-01 -5.52897394e-01 -1.38213634e+00 -4.72921044e-01 -3.92958105e-01 -3.33899885e-01 5.34306705e-01 -6.64869487e-01 -1.05079126e+00 4.39432889e-01 -9.75119710e-01 1.17172487e-01 -1.35472342e-01 4.86263901e-01 -7.07132578e-01 5.67902744e-01 -3.93756777e-01 -1.10375524e+00 -2.13186890e-01 -7.24291444e-01 7.14214146e-01 1.23441540e-01 -5.87866991e-04 -1.18726575e+00 2.34064102e-01 4.89132911e-01 2.83724125e-02 5.65627813e-01 1.27949786e+00 -8.84195387e-01 3.76166217e-02 -1.85495540e-01 -4.70126331e-01 4.75445449e-01 -1.04243122e-01 -1.38220161e-01 -8.45972896e-01 -3.86784881e-01 1.61723390e-01 -3.55499178e-01 7.19258368e-01 5.85495472e-01 1.55500424e+00 -2.20864192e-01 -2.24558339e-01 2.15591490e-01 1.90136385e+00 -2.52788980e-02 3.81197959e-01 3.48091543e-01 2.67546028e-01 7.52217174e-01 9.77630377e-01 9.51118529e-01 3.49038899e-01 6.74966872e-01 5.40017426e-01 1.42917246e-01 5.66202581e-01 -1.08862139e-01 1.00227274e-01 7.85021961e-01 5.22948131e-02 -4.58901942e-01 -5.82480073e-01 2.79970348e-01 -1.75634301e+00 -7.67392516e-01 -4.89862531e-01 2.55018425e+00 1.07084835e+00 3.91505994e-02 2.99425006e-01 6.38351560e-01 1.00490940e+00 -1.87082477e-02 -5.52637875e-01 -4.69886810e-01 -3.56123626e-01 4.60592099e-02 6.06273293e-01 4.25099194e-01 -1.05351162e+00 5.76701798e-02 6.15286255e+00 1.24645627e+00 -5.85565090e-01 3.00559014e-01 9.45560515e-01 -1.08489573e-01 -3.60426694e-01 -2.61557668e-01 -8.46214473e-01 4.96125996e-01 8.88414979e-01 -5.44066250e-01 -1.92964271e-01 7.30947316e-01 2.35842660e-01 -2.95424074e-01 -1.20045173e+00 8.21607172e-01 1.70655653e-01 -7.15146780e-01 -1.11064911e-01 8.36982131e-02 7.90177703e-01 -6.64940000e-01 1.59648597e-01 5.05298935e-02 -1.77415207e-01 -8.13763142e-01 7.28826761e-01 5.60968757e-01 9.44086850e-01 -1.11738956e+00 1.03464127e+00 4.40001100e-01 -9.80758250e-01 -2.10690722e-01 -3.98314446e-01 -1.63811460e-01 -9.26809534e-02 1.35607493e+00 -1.81833759e-01 8.87998164e-01 3.44434291e-01 5.26926339e-01 -5.46619177e-01 1.18535841e+00 1.19767062e-01 4.87104833e-01 -2.03537986e-01 1.52382433e-01 -1.68986008e-01 -3.66950154e-01 2.11406171e-01 8.81621957e-01 3.39956403e-01 -1.77711278e-01 -5.14345877e-02 5.98718524e-01 -1.43189445e-01 6.27646863e-01 -2.80238122e-01 3.85271519e-01 5.19148886e-01 1.13704216e+00 -6.98561311e-01 -3.66095058e-03 -4.13936853e-01 5.12782037e-01 3.44602287e-01 1.46262109e-01 -1.06964993e+00 -3.36681813e-01 6.05340779e-01 6.03328049e-02 8.78900439e-02 4.72174674e-01 -6.16988063e-01 -9.59341288e-01 3.73242170e-01 -5.75371861e-01 7.20534265e-01 -1.12119041e-01 -1.77662098e+00 1.91917747e-01 1.10909522e-01 -1.35547256e+00 1.66125953e-01 -5.21156788e-01 -2.89272755e-01 5.22655368e-01 -1.66832149e+00 -7.00390756e-01 -4.12058868e-02 4.32466507e-01 -1.90090701e-01 1.17189161e-01 6.33859992e-01 7.81159878e-01 -7.30464756e-01 1.33706176e+00 5.89682460e-01 -2.43872494e-01 8.11720908e-01 -1.10445619e+00 -7.03228831e-01 5.25732160e-01 -1.92583457e-01 2.19649002e-01 8.94768536e-01 -2.80581892e-01 -5.94180405e-01 -1.02722001e+00 8.06324244e-01 -2.94970125e-01 6.72879338e-01 1.54984385e-01 -6.68486774e-01 3.56170535e-01 -3.38887841e-01 5.11556752e-02 1.08181751e+00 8.35691914e-02 -4.63939488e-01 -5.62830448e-01 -1.66195309e+00 1.99011364e-03 9.22044814e-01 -3.77768934e-01 -2.35151082e-01 3.95852715e-01 4.14686739e-01 -1.12809436e-02 -1.29586720e+00 7.59322107e-01 8.07724833e-01 -1.33499563e+00 9.62702692e-01 -7.33827353e-01 5.37037730e-01 -2.75447257e-02 -3.76520365e-01 -1.08447158e+00 1.96566451e-02 -1.89372972e-02 7.95274333e-04 1.33099866e+00 4.06273395e-01 -8.80935669e-01 7.04926014e-01 3.86684507e-01 2.92285830e-01 -1.26312172e+00 -1.23170590e+00 -1.17799222e+00 7.17479765e-01 -3.14668596e-01 5.43414235e-01 1.06139076e+00 1.65451363e-01 -2.04230309e-01 -3.82551372e-01 1.31036267e-01 1.15310585e+00 1.86450839e-01 1.57811925e-01 -1.58767068e+00 -1.65247157e-01 -4.10292327e-01 -4.64044005e-01 -5.07136703e-01 2.85234213e-01 -8.54261220e-01 -3.07181448e-01 -1.04020917e+00 4.97944444e-01 -9.41358984e-01 -8.76730740e-01 5.81790470e-02 -1.07602641e-01 2.75382668e-01 4.20890050e-03 1.02233090e-01 -8.56215119e-01 4.86912072e-01 1.12498832e+00 1.36763915e-01 3.76950279e-02 2.83876657e-01 -9.38739419e-01 7.44655967e-01 9.39693630e-01 -7.93960512e-01 -4.41383451e-01 1.83302656e-01 6.00899339e-01 6.62232637e-02 1.04258269e-01 -9.92018163e-01 -4.63084504e-02 -3.42680871e-01 3.87977585e-02 -4.52848136e-01 -1.56796634e-01 -8.32095742e-01 9.33258459e-02 4.16640103e-01 -6.79270566e-01 -2.29735494e-01 -1.50403634e-01 7.82882631e-01 -1.44285679e-01 -4.97828662e-01 1.07188892e+00 2.13792756e-01 6.24828823e-02 2.07099915e-01 7.39866942e-02 4.49518770e-01 1.30912530e+00 -1.00750923e-01 -3.87869090e-01 -1.46071538e-01 -3.78813714e-01 1.17681980e-01 2.75723100e-01 -7.28499442e-02 9.98083428e-02 -1.41106701e+00 -8.52322280e-01 -1.23742662e-01 3.87341708e-01 -4.31833446e-01 2.84365147e-01 1.42959809e+00 -2.19976425e-01 3.88536990e-01 -8.13199356e-02 -3.59399706e-01 -1.19845796e+00 3.96118164e-01 4.91149187e-01 -4.98295903e-01 4.57309894e-02 7.15464473e-01 1.56912789e-01 -1.99133754e-01 2.82139242e-01 -1.50410756e-01 -8.28666165e-02 4.32549685e-01 4.70436603e-01 9.57463384e-01 1.92660540e-01 -4.89803225e-01 -5.29762328e-01 6.38615131e-01 9.47115645e-02 2.44242415e-01 1.00760496e+00 -3.49290103e-01 -3.19762617e-01 6.32833660e-01 1.38952208e+00 -7.01246634e-02 -9.25705016e-01 -1.25934750e-01 2.93419957e-01 -4.66412246e-01 -2.52372414e-01 -7.77259350e-01 -9.57037032e-01 8.17436695e-01 6.21814787e-01 3.53251457e-01 1.30459058e+00 -7.79856592e-02 5.55976927e-01 4.37122397e-03 5.86017549e-01 -1.20145619e+00 -1.85047105e-01 3.35645191e-02 6.48508012e-01 -1.18350375e+00 2.39777103e-01 -6.50235951e-01 -3.46698314e-01 9.33574080e-01 4.28187281e-01 1.17406577e-01 9.19708908e-01 1.33283203e-02 -8.02724659e-02 1.26004532e-01 -4.94131744e-01 1.08197525e-01 1.21249281e-01 4.66902494e-01 5.54146647e-01 2.61087358e-01 -1.14828777e+00 1.01191401e+00 1.00584393e-02 -4.41079552e-04 4.08473134e-01 2.58435667e-01 -3.36543351e-01 -1.15053380e+00 -4.10324216e-01 5.60852230e-01 -6.95636690e-01 1.79103285e-01 -2.92014927e-02 7.12995291e-01 3.50587100e-01 1.07954204e+00 -2.75246710e-01 -1.94859669e-01 2.47272223e-01 3.57886404e-02 7.33868062e-01 -9.92741808e-02 -3.12292397e-01 -4.88824964e-01 -4.73884828e-02 -4.03917640e-01 -8.31468105e-01 -6.76371634e-01 -9.86712098e-01 -5.07928371e-01 -6.83029771e-01 3.75352979e-01 3.59862119e-01 1.04523242e+00 -1.23873904e-01 -9.17816907e-03 8.91713262e-01 -1.80323929e-01 -1.06255877e+00 -7.48769701e-01 -1.13944936e+00 4.63459343e-01 1.96989730e-01 -9.63574290e-01 -8.39336395e-01 -3.55161488e-01]

[9.00698184967041, 4.195162296295166]

579c3a6c-adc6-4e26-bba0-0bad25ba4322

a-boosting-algorithm-for-positive-unlabeled

2205.09485

https://arxiv.org/abs/2205.09485v4

https://arxiv.org/pdf/2205.09485v4.pdf

A Boosting Algorithm for Positive-Unlabeled Learning

Positive-unlabeled (PU) learning deals with binary classification problems when only positive (P) and unlabeled (U) data are available. Many recent PU methods are based on neural networks, but little has been done to develop boosting algorithms for PU learning, despite boosting algorithms' strong performance on many fully supervised classification problems. In this paper, we propose a novel boosting algorithm, AdaPU, for PU learning. Similarly to AdaBoost, AdaPU aims to optimize an empirical exponential loss, but the loss is based on the PU data, rather than on positive-negative (PN) data. As in AdaBoost, we learn a weighted combination of weak classifiers by learning one weak classifier and its weight at a time. However, AdaPU requires a very different algorithm for learning the weak classifiers and determining their weights. This is because AdaPU learns a weak classifier and its weight using a weighted positive-negative (PN) dataset with some negative data weights $-$ the dataset is derived from the original PU data, and the data weights are determined by the current weighted classifier combination, but some data weights are negative. Our experiments showed that AdaPU outperforms neural networks on several benchmark PU datasets, including a large-scale challenging cyber security dataset.

['Weitong Chen', 'Miao Xu', 'Nan Ye', 'Chenhao Zhang', 'Mingzhe Zhang', 'Yawen Zhao']

2022-05-19

['activity-detection']

['computer-vision']

[ 2.47030072e-02 -2.74518609e-01 -8.73656631e-01 -4.93083000e-01 -7.17151701e-01 -4.08765763e-01 1.98199809e-01 5.87101936e-01 -5.49679458e-01 1.08907223e+00 -1.05332263e-01 -6.55730605e-01 2.98260659e-01 -1.26766694e+00 -7.01041281e-01 -9.17849004e-01 -9.48405713e-02 3.60930175e-01 3.27619940e-01 -2.48484418e-01 1.15785688e-01 5.14803603e-02 -1.60509980e+00 5.38938463e-01 6.85441136e-01 1.58173954e+00 -2.35740319e-01 7.22007453e-01 -1.92463681e-01 1.02929688e+00 -5.74103177e-01 -2.80175984e-01 5.87812781e-01 -3.68985265e-01 -6.77939832e-01 -5.22909284e-01 2.03716531e-01 -5.80158755e-02 2.59570628e-01 9.65329707e-01 4.54919428e-01 -3.35142836e-02 7.64786482e-01 -1.56837034e+00 -7.37931967e-01 4.73890543e-01 -7.94972062e-01 1.92799315e-01 4.26758766e-01 -2.18080997e-01 1.30429983e+00 -7.75546074e-01 3.80117506e-01 1.07549834e+00 1.06022251e+00 6.86434448e-01 -1.10894895e+00 -9.51845467e-01 2.08473772e-01 5.44867277e-01 -9.32435513e-01 2.99350083e-01 9.40530717e-01 -3.08626354e-01 8.93981397e-01 2.43250787e-01 8.41243625e-01 1.04384339e+00 3.85576725e-01 1.06042778e+00 1.51840305e+00 -8.57718945e-01 6.75799429e-01 4.20039892e-01 1.01605630e+00 6.36600316e-01 1.38418414e-02 4.24211562e-01 -3.27330500e-01 -5.72359324e-01 -8.68684575e-02 1.32470191e-01 -9.33915302e-02 -4.91911083e-01 -6.02030516e-01 9.14509535e-01 6.04607284e-01 2.24952549e-01 -1.63593546e-01 -4.19624984e-01 5.42336822e-01 7.06043303e-01 6.99908137e-01 4.06493545e-02 -1.10344505e+00 -3.65836658e-02 -1.41936213e-01 1.32908329e-01 9.48979378e-01 4.54699397e-01 1.20735824e+00 -1.72867000e-01 1.67986602e-02 1.04354620e+00 9.49398801e-02 3.13053697e-01 7.91503191e-01 -2.67034590e-01 3.88471514e-01 6.04984164e-01 9.80928615e-02 -6.62082016e-01 -3.10280234e-01 -1.64269313e-01 -7.41993606e-01 6.32205307e-01 5.25293589e-01 -4.78996515e-01 -8.70790064e-01 1.65121579e+00 4.42845911e-01 2.04077303e-01 1.45038933e-01 4.84505385e-01 6.58307076e-01 7.84488499e-01 2.53816754e-01 -5.09695590e-01 8.89775991e-01 -9.94945586e-01 -4.50079292e-01 -4.25410479e-01 7.34759152e-01 -5.26851714e-01 1.05747974e+00 7.63368607e-01 -5.33158064e-01 -5.59583366e-01 -1.47669482e+00 4.08078492e-01 -6.36366427e-01 -2.34176487e-01 5.54579139e-01 9.39720213e-01 -9.22680259e-01 6.05636299e-01 -1.99646786e-01 -1.42892122e-01 3.59970272e-01 3.74594122e-01 -2.72403866e-01 -2.30614051e-01 -1.42863357e+00 1.12357926e+00 7.54566610e-01 -2.01097474e-01 -3.81034315e-01 -4.75639462e-01 -8.58261704e-01 2.75852066e-02 1.97111800e-01 -3.51839244e-01 1.27466798e+00 -1.25487614e+00 -1.25844240e+00 4.95060295e-01 1.26225233e-01 -5.75125813e-01 4.59224254e-01 1.38444044e-02 -4.25185978e-01 -3.33649039e-01 -2.04108253e-01 5.67626595e-01 9.88636732e-01 -1.46416199e+00 -1.29418230e+00 -5.90099871e-01 -2.52374429e-02 -1.60207730e-02 -7.44613290e-01 -5.44081628e-02 2.04066500e-01 -5.35938382e-01 1.98111609e-01 -6.17199779e-01 -1.05556473e-01 -6.83211684e-02 1.40973762e-01 -5.74262917e-01 1.34311914e+00 -5.99447250e-01 1.20162463e+00 -1.94966125e+00 -4.98149693e-01 5.29839396e-01 -4.11998965e-02 5.45938849e-01 -1.13860220e-01 -7.97551721e-02 -4.45634961e-01 -2.42994055e-01 -4.35910583e-01 8.81135240e-02 -2.59922802e-01 4.47485834e-01 -3.12009960e-01 2.30440795e-01 1.42426908e-01 5.01358390e-01 -1.26453125e+00 -3.80798459e-01 1.41082585e-01 -4.34660017e-02 -4.49134409e-01 1.61405906e-01 -1.02067262e-01 -6.29972965e-02 -2.94222176e-01 9.16756332e-01 7.69607425e-01 1.66826725e-01 2.18671143e-01 1.26812816e-01 2.60698766e-01 -1.22328661e-02 -1.09069443e+00 1.01269925e+00 -5.60882807e-01 2.66392916e-01 9.12001804e-02 -1.65474319e+00 1.37079930e+00 2.73139417e-01 6.22927487e-01 -6.76040828e-01 3.97191584e-01 2.66349912e-01 5.32081611e-02 -1.65444404e-01 9.74301342e-03 -5.39808571e-01 1.91219170e-02 6.19280100e-01 7.90901482e-02 -2.70599667e-02 2.25444764e-01 -1.90497100e-01 1.12083495e+00 7.73860291e-02 7.27884293e-01 -1.27895400e-01 8.03189099e-01 8.84287432e-02 1.12124264e+00 8.21204066e-01 -7.72126079e-01 2.62144774e-01 5.19478977e-01 -9.00964260e-01 -6.94596171e-01 -1.15719223e+00 -3.65948707e-01 1.32209396e+00 6.10025339e-02 -3.71313840e-01 -5.03924906e-01 -1.55378270e+00 2.05038995e-01 6.35461807e-01 -6.58442259e-01 -2.82615900e-01 -4.49237436e-01 -1.33006394e+00 -5.29071800e-02 7.66812921e-01 3.30163628e-01 -1.16365206e+00 -1.21633656e-01 4.28015262e-01 9.10408348e-02 -3.58577400e-01 -7.47762099e-02 7.27667332e-01 -8.55729043e-01 -1.42061770e+00 -2.52608687e-01 -1.08837533e+00 7.48322010e-01 2.39315316e-01 1.20501757e+00 -1.33550301e-01 -1.40773565e-01 3.39209259e-01 -6.06156111e-01 -1.03325081e+00 -3.56141925e-01 -3.12722772e-01 3.82384241e-01 -5.89866303e-02 7.89689541e-01 -4.92183596e-01 -1.43337622e-01 3.84679884e-01 -6.22902811e-01 -3.31443310e-01 4.51932192e-01 1.51862204e+00 4.51975405e-01 1.94927424e-01 1.00255489e+00 -9.59780395e-01 7.82359123e-01 -5.66710353e-01 -4.22569752e-01 4.73684996e-01 -9.32637930e-01 -2.17026323e-01 8.45153272e-01 -8.05223644e-01 -1.10141575e+00 5.16641662e-02 -5.85283101e-01 -1.60885379e-01 8.61746892e-02 2.66108602e-01 -1.87081054e-01 -2.54857332e-01 1.21798360e+00 -8.61082897e-02 2.68058963e-02 -2.13743806e-01 2.11671591e-01 9.69855607e-01 1.46393567e-01 -8.70294809e-01 4.51489061e-01 2.42505193e-01 -4.10164684e-01 -4.09153104e-01 -1.01956153e+00 -6.09761834e-01 -5.16588151e-01 -3.22216243e-01 4.34264630e-01 -5.32258689e-01 -6.33263588e-01 6.32894158e-01 -7.46432841e-01 -2.84557700e-01 -6.13516629e-01 4.93770927e-01 -3.14784855e-01 3.60785186e-01 -6.87979996e-01 -9.95361924e-01 -5.96326530e-01 -6.92514062e-01 1.45331532e-01 9.59248319e-02 -2.49908209e-01 -8.79354179e-01 3.03418398e-01 1.92010924e-01 2.84385867e-02 -9.62171257e-02 9.35050547e-01 -8.17529082e-01 3.69854152e-01 -8.71121049e-01 -2.69648075e-01 1.26534045e+00 1.12977982e-01 -1.96592137e-01 -1.18933785e+00 -4.19644713e-01 3.51881832e-01 -9.01066065e-01 7.93385267e-01 2.22112060e-01 1.48772740e+00 -2.95973569e-01 -1.35123104e-01 6.35034367e-02 1.24960899e+00 6.17819607e-01 1.79057032e-01 2.40968406e-01 5.02828896e-01 3.95467877e-01 8.93404663e-01 3.58143300e-01 1.71280056e-01 2.43663285e-02 4.84243155e-01 -3.58547159e-02 3.13646674e-01 -1.02652803e-01 5.38157761e-01 7.96798468e-01 9.61259604e-02 3.30744743e-01 -1.03957617e+00 1.65180549e-01 -2.12246561e+00 -9.28768933e-01 6.07205816e-02 2.42418265e+00 1.07553029e+00 7.25975096e-01 1.59411252e-01 7.18582571e-01 7.76603878e-01 1.10419840e-02 -6.22354031e-01 -5.84498167e-01 -1.17343636e-02 3.96173447e-01 4.45252389e-01 3.81966829e-01 -1.41574681e+00 3.41648251e-01 6.96230078e+00 9.43738937e-01 -1.27974212e+00 2.98701167e-01 1.14198124e+00 6.77416846e-02 -8.84154812e-02 -1.12236999e-01 -7.82187104e-01 5.60126722e-01 6.51024759e-01 4.47237678e-02 2.76977271e-02 1.49626744e+00 -4.54144776e-01 -3.12514424e-01 -1.12399948e+00 9.81291711e-01 1.43222511e-01 -1.12844193e+00 -2.29685053e-01 -2.95770019e-01 7.37826228e-01 -1.22660361e-02 -1.53124362e-01 1.01754904e+00 7.89902687e-01 -3.09088022e-01 5.66165328e-01 -1.12889454e-01 1.92884862e-01 -8.85002375e-01 1.14155853e+00 7.67575443e-01 -9.86248136e-01 -7.78597653e-01 -5.61210811e-01 -6.66395366e-01 -4.76643980e-01 7.16859758e-01 -7.03679621e-01 2.85733044e-01 1.04098463e+00 8.81274581e-01 -2.86605418e-01 8.74694943e-01 -4.02308106e-01 7.10344732e-01 -1.96150005e-01 -3.45744312e-01 1.29021242e-01 -3.98702085e-01 2.24281047e-02 1.00152922e+00 1.43377021e-01 1.59352079e-01 6.51698053e-01 1.20603599e-01 -7.42775872e-02 4.83139098e-01 -5.67169070e-01 6.67445540e-01 2.74072230e-01 1.28966689e+00 -4.06482697e-01 -5.99497974e-01 -7.07393050e-01 2.67278433e-01 6.27238095e-01 1.85858309e-02 -5.08706808e-01 -1.16590418e-01 4.24597591e-01 -2.37817034e-01 -1.23354837e-01 2.78050810e-01 -6.27965868e-01 -1.08586597e+00 7.48433098e-02 -6.58096910e-01 1.21174395e+00 -4.08542186e-01 -2.02041364e+00 2.33744770e-01 -2.21050322e-01 -1.61828268e+00 -1.27851203e-01 -9.55048144e-01 -9.15764213e-01 7.46715844e-01 -1.45319021e+00 -8.63339484e-01 1.44638985e-01 5.81721067e-01 2.04304859e-01 -2.00646698e-01 9.49128926e-01 8.42158645e-02 -5.32668173e-01 3.76528680e-01 2.87100792e-01 2.58558601e-01 9.43774343e-01 -1.37443662e+00 -8.93138051e-02 6.05434597e-01 -1.57371491e-01 2.01538011e-01 4.72216606e-01 -5.08235097e-01 -1.00845170e+00 -8.22239757e-01 8.71822894e-01 -3.71541679e-01 8.17113101e-01 -1.26186222e-01 -8.18855464e-01 4.53627408e-01 3.00891958e-02 6.67069077e-01 1.16991162e+00 5.26658535e-01 -6.77444816e-01 -6.59372151e-01 -1.71214759e+00 1.06021255e-01 5.92271090e-01 -1.37106821e-01 -7.49041855e-01 3.73712659e-01 2.57461905e-01 -1.19739041e-01 -6.48686409e-01 7.64377117e-01 6.92574084e-01 -1.02144003e+00 9.60605681e-01 -8.05278957e-01 2.13092044e-01 -1.53818116e-01 2.61410680e-02 -1.59408450e+00 -3.41481596e-01 3.82454723e-01 -4.67389375e-01 8.07884395e-01 4.12149072e-01 -9.52273250e-01 1.11305118e+00 4.79005843e-01 5.32387234e-02 -1.02847993e+00 -9.69042063e-01 -7.65513301e-01 1.67537391e-01 -7.18701363e-01 2.60505646e-01 1.24685836e+00 3.97333503e-01 4.55829769e-01 -4.96672690e-01 -2.94544071e-01 6.23143911e-01 1.15656830e-01 4.12073791e-01 -1.47498047e+00 -4.25625622e-01 -3.62850517e-01 -4.15616930e-01 -2.94470012e-01 3.30618143e-01 -7.73705184e-01 1.64389044e-01 -1.15615153e+00 1.60166129e-01 -8.38673472e-01 -9.23569202e-01 9.32643056e-01 -5.57203710e-01 3.28735381e-01 2.42801141e-02 1.70099493e-02 -2.96067268e-01 3.90889466e-01 8.87584746e-01 -7.65518904e-01 -3.31114531e-01 3.81909072e-01 -4.95108455e-01 1.03057075e+00 1.00233400e+00 -4.20073420e-01 -5.11549830e-01 1.61260501e-01 5.61939087e-03 5.65143395e-03 -2.34092444e-01 -1.07877457e+00 -4.27852869e-02 -1.41205236e-01 8.44467580e-01 -6.35604501e-01 2.16764480e-01 -9.85276878e-01 -5.68686783e-01 8.99732888e-01 -2.81091575e-02 -1.83262408e-01 -1.10254757e-01 6.37573004e-01 -3.89742374e-01 -5.96805990e-01 9.66973484e-01 -2.12041378e-01 -9.35553551e-01 3.52006644e-01 -2.73756236e-01 -1.66674227e-01 1.38117826e+00 -1.72835112e-01 -2.77085990e-01 -1.12823695e-01 -9.04741466e-01 3.29314202e-01 2.09010646e-01 3.95466864e-01 5.68357170e-01 -1.44122076e+00 -5.44229984e-01 3.44188422e-01 3.50500673e-01 -3.25321406e-02 8.54100846e-03 2.21376494e-01 -2.30193719e-01 -9.02277380e-02 -6.56098127e-01 -4.67312008e-01 -1.34570944e+00 9.40105617e-01 4.55665030e-02 -4.82268780e-01 -1.47343814e-01 9.95436132e-01 -1.56363279e-01 -9.24740732e-01 4.40004528e-01 4.79200073e-02 -4.43649590e-01 4.52259928e-01 8.51408064e-01 4.38574404e-01 4.12955970e-01 -1.44835100e-01 -7.37758800e-02 1.04422867e-01 -2.51820087e-01 2.66796827e-01 1.28418136e+00 5.02139628e-01 -2.30029896e-01 5.37926733e-01 1.30041277e+00 -3.64828497e-01 -1.03153706e+00 -5.57702720e-01 1.66493028e-01 -3.93146574e-01 -1.69065803e-01 -7.35372484e-01 -8.98622334e-01 6.99544191e-01 7.68609703e-01 2.29167268e-01 1.22376049e+00 -4.23076957e-01 7.32086420e-01 6.32888556e-01 7.97342539e-01 -1.60476267e+00 1.56694483e-02 7.06918538e-01 3.96833986e-01 -1.65285993e+00 -1.09916115e-02 -3.43587816e-01 -4.45813626e-01 1.10686529e+00 1.00006282e+00 -2.12857977e-01 1.16558158e+00 3.43813658e-01 2.84561753e-01 5.46785116e-01 -7.23818183e-01 2.85725929e-02 -5.46700992e-02 9.26189184e-01 1.98345900e-01 1.74902216e-01 -7.88791418e-01 6.36412024e-01 4.67389710e-02 1.65013820e-01 1.02906518e-01 1.53923166e+00 -5.86150646e-01 -1.59378564e+00 -7.72281587e-01 9.60025251e-01 -2.72355050e-01 2.43886665e-01 -2.33467802e-01 3.46727729e-01 3.89798820e-01 8.75202298e-01 -4.48854407e-03 -7.31936932e-01 1.28438815e-01 3.87871444e-01 2.40360469e-01 -5.23483574e-01 -6.61298633e-01 -5.28064847e-01 6.24015070e-02 -1.92672372e-01 -2.77686089e-01 -5.52911162e-01 -9.78548527e-01 -1.70901150e-01 -1.79797128e-01 4.39909905e-01 5.89091837e-01 8.19720387e-01 -3.89874011e-01 1.62246540e-01 1.17380321e+00 -7.70924985e-01 -9.15363550e-01 -9.53432620e-01 -6.72407866e-01 4.63004053e-01 1.42431825e-01 -7.58173585e-01 -3.24798822e-01 -2.07196593e-01]

[9.619516372680664, 3.526097297668457]

65bcea58-999f-46f4-ab8f-0e2936e20db2

meta-temporal-point-processes

2301.12023

https://arxiv.org/abs/2301.12023v1

https://arxiv.org/pdf/2301.12023v1.pdf

Meta Temporal Point Processes

A temporal point process (TPP) is a stochastic process where its realization is a sequence of discrete events in time. Recent work in TPPs model the process using a neural network in a supervised learning framework, where a training set is a collection of all the sequences. In this work, we propose to train TPPs in a meta learning framework, where each sequence is treated as a different task, via a novel framing of TPPs as neural processes (NPs). We introduce context sets to model TPPs as an instantiation of NPs. Motivated by attentive NP, we also introduce local history matching to help learn more informative features. We demonstrate the potential of the proposed method on popular public benchmark datasets and tasks, and compare with state-of-the-art TPP methods.

['Gabriel L. Oliveira', 'Frederick Tung', 'Mohamed Osama Ahmed', 'Wonho Bae']

2023-01-27

['point-processes']

['methodology']

[ 4.85078156e-01 -9.54609830e-03 -1.54773235e-01 -2.43512854e-01 -6.59391940e-01 -5.15794933e-01 1.33537757e+00 2.75169760e-01 -5.26101999e-02 5.50024629e-01 4.74838823e-01 5.70544489e-02 -1.75958961e-01 -1.07314813e+00 -9.32464242e-01 -9.04709697e-01 -3.25693816e-01 8.80801797e-01 5.39665341e-01 1.07765220e-01 3.14694405e-01 1.96459949e-01 -1.41283703e+00 7.02987790e-01 3.97143900e-01 1.09811652e+00 9.71263647e-02 5.50309837e-01 -3.01576674e-01 1.36114252e+00 -6.94930136e-01 -5.62094271e-01 1.55974954e-01 -3.34815055e-01 -7.88800836e-01 -9.93112624e-02 -2.68372118e-01 1.80139989e-01 -3.27718198e-01 5.83122313e-01 2.02151045e-01 4.02364939e-01 9.47513223e-01 -1.48944092e+00 -5.36777318e-01 8.69268477e-01 -3.48060310e-01 4.06208485e-01 1.95475176e-01 4.23599064e-01 1.09626186e+00 -7.48690307e-01 3.66044134e-01 1.49275935e+00 7.44967163e-01 4.30967182e-01 -1.27516460e+00 -1.64322019e-01 4.16161269e-01 3.20993245e-01 -7.57452130e-01 1.30022485e-02 8.51328552e-01 -2.92976201e-01 1.10894394e+00 -1.24464944e-01 7.93817937e-01 1.81947410e+00 7.47071803e-01 1.28837669e+00 9.11909461e-01 -2.54387051e-01 7.88325548e-01 -6.94809556e-01 2.37127677e-01 4.27806914e-01 -3.17700773e-01 1.49326563e-01 -1.04476655e+00 -5.03593564e-01 5.70450008e-01 3.74686390e-01 1.66218936e-01 -1.71045125e-01 -1.20800662e+00 5.07702827e-01 -8.56321454e-02 1.27508804e-01 -9.47291553e-01 7.16234088e-01 5.14999568e-01 1.74267620e-01 6.22974217e-01 5.43432832e-02 -5.89860976e-01 -5.19192517e-01 -5.42468667e-01 3.28731030e-01 1.17131400e+00 7.23481655e-01 5.32969236e-01 -2.72541434e-01 -9.07793820e-01 6.41757429e-01 2.67721832e-01 2.26170301e-01 6.22920871e-01 -1.05454493e+00 6.00810170e-01 3.72541308e-01 3.60987306e-01 -4.29043174e-01 4.93987054e-02 -2.44663030e-01 -6.35837972e-01 -1.92893848e-01 1.60468847e-01 -8.31823871e-02 -9.94016469e-01 1.71077240e+00 1.37087941e-01 1.09600019e+00 1.76301897e-01 5.14643081e-02 2.62418061e-01 1.38943028e+00 3.92790794e-01 -3.65373194e-01 1.17117393e+00 -1.34740984e+00 -6.05135500e-01 -1.65356502e-01 -6.70946091e-02 -1.25453636e-01 9.35342252e-01 6.01711035e-01 -1.26459801e+00 -5.10292411e-01 -6.32789493e-01 4.56862241e-01 -2.19465986e-01 -2.26014867e-01 3.01781893e-01 3.44800234e-01 -1.12258840e+00 1.12709117e+00 -1.35395861e+00 -3.29631597e-01 5.73844135e-01 4.47654165e-02 2.18042821e-01 1.86292470e-01 -1.12964034e+00 7.44840741e-01 7.65840173e-01 -5.52337021e-02 -1.75461137e+00 -7.32226014e-01 -5.63255906e-01 3.72399569e-01 5.00143707e-01 -8.00653219e-01 1.72098291e+00 -6.92179620e-01 -1.76280236e+00 5.41762471e-01 -5.42574465e-01 -1.00031745e+00 6.77570343e-01 -3.81087720e-01 -4.29644078e-01 1.81543440e-01 -5.08469343e-02 3.19833219e-01 1.31803942e+00 -1.38142073e+00 -1.06143379e+00 -6.34255633e-03 -2.93558180e-01 2.45378509e-01 3.60264592e-02 2.20380917e-01 -5.57302296e-01 -4.46433067e-01 1.61636006e-02 -7.73158014e-01 -2.72381812e-01 -4.58485067e-01 -3.56975049e-01 -7.40104616e-01 8.24499190e-01 -4.60181385e-01 1.04122293e+00 -1.92207503e+00 2.39178225e-01 2.32817624e-02 2.40175039e-01 -1.15862958e-01 -8.93723592e-02 1.12061059e+00 1.44761965e-01 3.54622990e-01 -5.22266507e-01 -1.06441438e+00 1.96488366e-01 6.59818470e-01 -6.94463193e-01 1.43762052e-01 3.63508195e-01 9.48840618e-01 -1.33621895e+00 -3.82280111e-01 5.50384112e-02 1.75950274e-01 2.38708541e-01 2.90950030e-01 -8.55468929e-01 5.54268003e-01 -5.99985778e-01 6.90364599e-01 1.71030954e-01 -3.47847819e-01 -3.58664356e-02 4.11942929e-01 -4.45682667e-02 1.86059371e-01 -7.81126976e-01 1.56686282e+00 -4.29990083e-01 2.53887713e-01 -8.01972508e-01 -9.16708767e-01 8.56830716e-01 7.35872567e-01 7.31538832e-01 -1.85871080e-01 -9.97348055e-02 -5.84797151e-02 -2.43411183e-01 -4.19940025e-01 3.64006549e-01 -2.89449304e-01 -1.07153477e-02 7.56785333e-01 3.24844897e-01 1.00421302e-01 1.86830923e-01 -2.20778435e-01 1.61624146e+00 6.45255208e-01 4.51296240e-01 5.08914068e-02 4.42723393e-01 -2.96028376e-01 9.19545889e-01 1.27799070e+00 -2.91800916e-01 4.20683324e-01 8.42051268e-01 -7.65473545e-01 -1.12122440e+00 -1.46576715e+00 2.99550563e-01 1.18703711e+00 -2.30233103e-01 -4.60704535e-01 -3.78670067e-01 -7.90274024e-01 -1.21250905e-01 8.63099098e-01 -8.93669367e-01 -6.65446669e-02 -8.43202591e-01 -9.61346328e-01 4.75683302e-01 7.81750381e-01 6.76374853e-01 -1.70146668e+00 -6.98339164e-01 7.31126964e-01 -1.01830415e-01 -8.16494226e-01 -2.06592634e-01 3.04492533e-01 -1.09348643e+00 -7.13843107e-01 -3.23894620e-01 -7.22473502e-01 1.22445665e-01 -2.25390688e-01 1.41492331e+00 -4.73668605e-01 3.83078486e-01 5.76626718e-01 -4.18825537e-01 -9.33045506e-01 -8.22734535e-01 1.86825860e-02 7.74517469e-03 4.15995568e-01 2.07957879e-01 -9.30840075e-01 -4.95456100e-01 8.48932713e-02 -9.68438327e-01 -2.48951335e-02 3.52797121e-01 8.42394352e-01 7.94237673e-01 4.63955373e-01 4.56508517e-01 -7.73946643e-01 8.94225717e-01 -8.54452252e-01 -2.36632600e-01 4.50042307e-01 -2.54862517e-01 2.38786325e-01 6.51699126e-01 -6.86100602e-01 -1.66109061e+00 -7.32806921e-02 2.51848221e-01 -6.64838552e-01 -1.90555394e-01 6.46148980e-01 9.36294161e-03 6.27384841e-01 4.66753066e-01 7.08917022e-01 -2.43523598e-01 -2.94838518e-01 3.46223325e-01 4.72570285e-02 3.84861529e-01 -1.33134472e+00 6.54795289e-01 7.12176323e-01 2.44370461e-01 -3.07903975e-01 -9.56755996e-01 -4.54157382e-01 -6.25025213e-01 -3.31100553e-01 6.24946833e-01 -6.16349101e-01 -8.07300389e-01 6.87027574e-01 -1.24328661e+00 -7.72590995e-01 -6.33792639e-01 1.25126407e-01 -1.16064358e+00 9.94857848e-02 -1.04118156e+00 -1.09043181e+00 -4.54836398e-01 -7.30643094e-01 1.32434464e+00 8.71478021e-02 -1.97417155e-01 -1.48573983e+00 4.27960843e-01 -1.38382539e-01 1.60509720e-01 4.62925553e-01 9.09719706e-01 -1.10963523e+00 -6.04784489e-01 -1.25953645e-01 5.09960413e-01 4.35358495e-01 2.53088951e-01 2.49524489e-02 -9.93897259e-01 -1.94725636e-02 6.03731513e-01 -8.34767893e-02 9.01784837e-01 3.44805926e-01 1.40451896e+00 -4.56361890e-01 -5.90333164e-01 1.84327751e-01 1.41259336e+00 6.84103370e-01 6.86635852e-01 4.54635531e-01 4.62744415e-01 6.44041359e-01 4.49678332e-01 5.51757514e-01 2.10932121e-01 1.08873233e-01 3.48334938e-01 5.31785488e-01 3.61964017e-01 -7.04533041e-01 5.13718367e-01 7.37280071e-01 -4.00264591e-01 -6.43839121e-01 -1.44581091e+00 6.11269474e-01 -2.52598596e+00 -1.46151173e+00 1.45498857e-01 2.00420976e+00 7.56855905e-01 3.40230942e-01 1.95771009e-02 3.18326168e-02 8.11336219e-01 4.95677829e-01 -5.48090816e-01 -7.25498199e-02 -7.11701214e-02 3.48540485e-01 8.16901326e-02 1.46375254e-01 -1.24855566e+00 9.45375860e-01 6.30968475e+00 7.80987322e-01 -7.72976875e-01 3.25336039e-01 7.52600253e-01 -9.33806971e-02 6.72718734e-02 3.01054139e-02 -7.75125563e-01 7.29723096e-01 1.22884905e+00 -4.04948890e-01 2.60641724e-01 6.79677904e-01 3.39394122e-01 2.47924849e-01 -1.72260821e+00 8.01527500e-01 -5.73513098e-02 -1.47146535e+00 1.23777129e-01 -1.32621139e-01 9.19830680e-01 1.31560966e-01 2.62657493e-01 7.08718419e-01 9.96814489e-01 -1.00899005e+00 9.66501474e-01 9.75921810e-01 -8.97123590e-02 -4.79596019e-01 5.46530962e-01 5.76106727e-01 -1.33869696e+00 -4.58867669e-01 -1.88205436e-01 -3.50287557e-02 4.67110217e-01 3.85957003e-01 -9.63979661e-01 6.77560091e-01 7.52383173e-01 1.12283385e+00 -2.35573202e-01 1.11149418e+00 -5.29165983e-01 1.21157193e+00 -1.74820900e-01 -1.26415566e-01 3.98874938e-01 -1.57774866e-01 7.34401226e-01 1.07352769e+00 4.35404688e-01 -1.85769051e-01 3.28185618e-01 8.59526992e-01 -1.04636222e-01 -2.65182883e-01 -5.73794782e-01 1.48668319e-01 7.12810636e-01 9.46391582e-01 -7.35127509e-01 -5.34464478e-01 -3.00863028e-01 1.07184780e+00 3.17388564e-01 4.68147844e-01 -8.94714952e-01 4.69107091e-01 3.78986061e-01 -3.27430218e-01 3.84011179e-01 9.43209082e-02 -1.12676978e-01 -1.10381913e+00 8.29415321e-02 -5.63490093e-01 6.42050445e-01 -9.46510732e-01 -2.06249356e+00 3.68950784e-01 2.72155285e-01 -1.30968082e+00 -3.11660975e-01 -3.20439279e-01 -1.22558463e+00 8.73406470e-01 -1.22697759e+00 -1.27916849e+00 5.24343289e-02 4.97133940e-01 9.20672834e-01 -1.42032027e-01 5.95760643e-01 -5.84942460e-01 -3.57420713e-01 -2.98813045e-01 2.17532709e-01 -1.17556155e-01 2.59686291e-01 -1.53076303e+00 8.15091193e-01 8.05388451e-01 8.13258067e-02 6.27338886e-01 7.96638370e-01 -9.86270189e-01 -1.11373711e+00 -1.47363877e+00 6.71556890e-01 -8.72034252e-01 1.06404757e+00 -2.71096110e-01 -1.07360947e+00 1.15369129e+00 3.69684875e-01 -1.13684289e-01 5.67795873e-01 7.72634447e-02 2.66744029e-02 -2.47422308e-02 -1.05664015e+00 8.13401103e-01 1.32537115e+00 -3.77445877e-01 -1.14748299e+00 4.34925914e-01 1.07938969e+00 -3.16686004e-01 -7.99236774e-01 2.21241638e-01 1.22336440e-01 -4.91869748e-01 9.31186199e-01 -6.80115283e-01 7.98492312e-01 -1.47087216e-01 -4.45744544e-02 -1.60648072e+00 -4.30656344e-01 -7.35155761e-01 -8.57735991e-01 1.39208126e+00 1.01276293e-01 -6.51648164e-01 8.84499490e-01 2.72441953e-01 -2.51036197e-01 -8.11591983e-01 -9.92095470e-01 -1.04985797e+00 6.12738691e-02 -6.07235432e-01 9.77787137e-01 6.16675735e-01 -2.79510200e-01 3.09500843e-01 -4.70625490e-01 3.55691582e-01 7.08276451e-01 -7.07456321e-02 1.76995814e-01 -1.50122929e+00 -6.98062241e-01 -3.67806286e-01 -2.94679180e-02 -7.32156813e-01 3.13951612e-01 -6.02743089e-01 3.09704751e-01 -1.65215993e+00 2.01210693e-01 -3.51387024e-01 -5.95636785e-01 4.23557490e-01 -1.65594742e-01 -5.68851709e-01 2.13337407e-01 6.71625018e-01 -6.70274079e-01 9.42047596e-01 8.24585140e-01 -1.94959521e-01 -2.37601757e-01 4.52673197e-01 -8.95372406e-02 7.57100821e-01 1.10694742e+00 -7.18814969e-01 -7.34172106e-01 -3.08263570e-01 2.05655724e-01 5.06420970e-01 3.25808257e-01 -1.10056472e+00 4.40686405e-01 -3.68044525e-01 2.61369854e-01 -7.73019791e-01 6.15079761e-01 -6.89010918e-01 4.68724579e-01 5.26721895e-01 -5.63028574e-01 2.59195328e-01 -1.94700584e-01 1.28263235e+00 -3.23914587e-01 -3.50798547e-01 1.64867252e-01 -5.51245034e-01 -7.28429258e-01 7.23411441e-01 -4.36407179e-01 -2.77595520e-01 1.15677488e+00 1.19772978e-01 -2.75458097e-01 -3.52842450e-01 -8.02661836e-01 3.18611503e-01 5.06198704e-02 2.82526374e-01 4.05688018e-01 -1.12619042e+00 -4.30496991e-01 -4.31465685e-01 1.09258592e-01 1.87078699e-01 1.30519956e-01 4.72042054e-01 -1.85465783e-01 7.19471276e-02 -1.09503187e-01 -6.99097812e-01 -8.18004489e-01 7.24256098e-01 4.62493360e-01 -9.17175889e-01 -7.59214938e-01 4.90000367e-01 1.05807073e-01 -2.54814565e-01 2.72391170e-01 -3.52045655e-01 -3.35356653e-01 -1.68992504e-01 4.99605685e-01 3.45320106e-01 -2.16864958e-01 5.41178621e-02 1.81168675e-01 9.96264280e-04 2.58129109e-02 -7.03659356e-01 1.57136297e+00 1.21764861e-01 -2.50198454e-01 1.28814673e+00 6.99467838e-01 -5.03525794e-01 -1.84583521e+00 -7.25209653e-01 5.74501753e-01 -7.21157864e-02 -6.36729300e-01 -7.86083639e-01 -4.67671126e-01 7.10868120e-01 2.97434360e-01 3.73580724e-01 9.95968401e-01 2.63099492e-01 5.30294895e-01 4.40470487e-01 6.41991138e-01 -1.03030467e+00 5.77813864e-01 8.83732736e-01 8.23882878e-01 -8.82532835e-01 -5.74056208e-01 -2.54809618e-01 -9.30934787e-01 1.05769122e+00 2.80941755e-01 -2.33239949e-01 6.61842287e-01 5.41879097e-03 -6.15461886e-01 -1.22640222e-01 -1.48968852e+00 1.19004615e-01 -1.72227070e-01 6.58893824e-01 -4.74937409e-02 3.52945551e-02 -4.77391221e-02 7.73434937e-01 -3.10817151e-03 3.34264845e-01 3.16928595e-01 1.44707668e+00 -3.05441380e-01 -1.26313245e+00 -2.44785950e-01 4.67779577e-01 -4.66196418e-01 -1.57632411e-01 1.66782022e-01 4.25421745e-01 7.73025528e-02 7.08026648e-01 3.45620692e-01 -2.02974781e-01 1.33338764e-01 6.60417259e-01 3.49650592e-01 -9.00776982e-01 -8.20861161e-01 -2.65238792e-01 1.39690444e-01 -3.65676314e-01 -5.65032244e-01 -1.10941410e+00 -1.00166559e+00 -2.33059358e-02 6.43707454e-01 -4.21972983e-02 3.56483549e-01 1.08073986e+00 4.44097333e-02 6.09884560e-01 6.41366422e-01 -6.93948030e-01 -7.66415477e-01 -1.08592272e+00 -4.30133015e-01 4.73452210e-01 6.69911057e-02 -5.93426883e-01 -1.12126425e-01 4.81931686e-01]

[8.474377632141113, 5.676767826080322]

c12bd267-019d-42a7-85d6-b0624c1694ee

mindstorms-in-natural-language-based

2305.17066

https://arxiv.org/abs/2305.17066v1

https://arxiv.org/pdf/2305.17066v1.pdf

Mindstorms in Natural Language-Based Societies of Mind

Both Minsky's "society of mind" and Schmidhuber's "learning to think" inspire diverse societies of large multimodal neural networks (NNs) that solve problems by interviewing each other in a "mindstorm." Recent implementations of NN-based societies of minds consist of large language models (LLMs) and other NN-based experts communicating through a natural language interface. In doing so, they overcome the limitations of single LLMs, improving multimodal zero-shot reasoning. In these natural language-based societies of mind (NLSOMs), new agents -- all communicating through the same universal symbolic language -- are easily added in a modular fashion. To demonstrate the power of NLSOMs, we assemble and experiment with several of them (having up to 129 members), leveraging mindstorms in them to solve some practical AI tasks: visual question answering, image captioning, text-to-image synthesis, 3D generation, egocentric retrieval, embodied AI, and general language-based task solving. We view this as a starting point towards much larger NLSOMs with billions of agents-some of which may be humans. And with this emergence of great societies of heterogeneous minds, many new research questions have suddenly become paramount to the future of artificial intelligence. What should be the social structure of an NLSOM? What would be the (dis)advantages of having a monarchical rather than a democratic structure? How can principles of NN economies be used to maximize the total reward of a reinforcement learning NLSOM? In this work, we identify, discuss, and try to answer some of these questions.

['Jürgen Schmidhuber', 'Bernard Ghanem', 'Deng-Ping Fan', 'Mengmeng Xu', 'Yuhui Wang', 'Wenyi Wang', 'Aleksandar Stanić', 'Weimin Shi', 'Imanol Schlag', 'Aditya Ramesh', 'Piotr Piękos', 'Jinjie Mai', 'Shuming Liu', 'Guohao Li', 'Bing Li', 'Louis Kirsch', 'Kazuki Irie', 'Vincent Herrmann', 'Hasan Abed Al Kader Hammoud', 'Abdullah Hamdi', 'Anand Gopalakrishnan', 'Róbert Csordás', 'Dylan R. Ashley', 'Francesco Faccio', 'Haozhe Liu', 'Mingchen Zhuge']

2023-05-26

['image-captioning']

['computer-vision']

[ 1.16959317e-02 8.36141229e-01 3.15811992e-01 -4.57963757e-02 6.40233904e-02 -5.00307262e-01 1.06336153e+00 -2.54717410e-01 -5.31402051e-01 7.60495365e-01 3.36656660e-01 -3.51560652e-01 4.62739579e-02 -7.17208922e-01 -5.91934443e-01 -6.48959279e-01 2.29838528e-02 8.82600069e-01 -1.34616747e-01 -8.77344489e-01 1.94474235e-01 1.46559462e-01 -1.43009043e+00 4.01809029e-02 1.08273661e+00 4.40798730e-01 4.27442491e-01 7.55012691e-01 -1.50549293e-01 1.36834407e+00 -5.57663441e-01 -5.19381523e-01 5.00530228e-02 -6.79407835e-01 -1.01599824e+00 2.21275911e-01 -1.49054855e-01 -2.87086517e-02 -2.13390350e-01 9.09374177e-01 2.65163869e-01 1.98598504e-01 6.98459744e-01 -1.37632084e+00 -1.22938991e+00 6.85242176e-01 -3.89913589e-01 -3.50743830e-01 5.38275480e-01 6.35939300e-01 7.98850894e-01 -3.71787161e-01 7.57469833e-01 1.90586472e+00 1.94082126e-01 8.96428704e-01 -1.26433837e+00 -3.03695351e-01 -9.86060314e-03 -8.66533592e-02 -1.14537179e+00 -6.52841210e-01 4.38317776e-01 -3.39870095e-01 1.14126456e+00 1.46720305e-01 1.05148983e+00 1.03615093e+00 3.76407474e-01 6.52552962e-01 1.03571987e+00 -5.86696386e-01 4.03535992e-01 3.01709563e-01 -4.08458352e-01 1.01813650e+00 4.53793526e-01 -9.58892405e-02 -8.33544731e-01 -1.83391660e-01 9.70140517e-01 -5.17182350e-01 -5.52404895e-02 -5.39865494e-01 -1.71457219e+00 1.00667334e+00 3.59739900e-01 3.76886696e-01 -3.13225210e-01 5.63104451e-01 1.21244823e-03 7.11420119e-01 1.97959498e-01 1.02585804e+00 -2.52224896e-02 1.53430447e-01 -2.39976853e-01 1.06445983e-01 1.18663383e+00 6.22601926e-01 8.01428914e-01 2.15792537e-01 5.23446679e-01 5.02734363e-01 6.87533557e-01 8.21077168e-01 5.70643663e-01 -1.61383188e+00 -3.61134484e-02 8.66022289e-01 2.60971904e-01 -9.82683957e-01 -5.39454043e-01 -9.67797861e-02 -8.27008426e-01 4.73722070e-01 3.53289336e-01 -6.22945070e-01 -4.88659769e-01 1.99551499e+00 3.01631689e-01 -3.60424787e-01 7.01695144e-01 1.02681446e+00 5.27271032e-01 8.80145729e-01 1.36056226e-02 3.95942591e-02 1.38491476e+00 -7.87656724e-01 -3.05798441e-01 -4.82587129e-01 7.60509372e-01 -3.64837229e-01 9.97008562e-01 1.29762724e-01 -1.39312720e+00 -1.21227711e-01 -9.70478892e-01 -2.05123704e-02 -6.10976636e-01 -4.21399951e-01 9.65079248e-01 4.31662887e-01 -1.58646452e+00 1.49379507e-01 -4.58190382e-01 -9.73341703e-01 -2.51914021e-02 6.03165865e-01 -4.49965179e-01 4.70477343e-01 -1.27323186e+00 1.40472949e+00 3.84322137e-01 -1.18794683e-02 -8.88755441e-01 7.51498993e-03 -7.36044407e-01 -1.79035127e-01 4.58095789e-01 -1.64276469e+00 1.17442179e+00 -1.68343401e+00 -1.61651790e+00 1.22268891e+00 7.66276941e-02 -5.11057436e-01 1.90223873e-01 2.10570931e-01 -2.72679865e-01 2.71553487e-01 -4.56489064e-02 1.21669400e+00 6.07643008e-01 -1.69566584e+00 -3.19039345e-01 -3.98072094e-01 4.51414704e-01 5.82224131e-01 -2.47531489e-01 -1.67375840e-02 1.94212094e-01 -1.31470576e-01 -1.70079708e-01 -9.65534866e-01 -3.60379666e-01 2.62356503e-03 -2.65909303e-02 -3.91697735e-01 3.85530442e-01 -1.25874039e-02 2.46950507e-01 -1.94227099e+00 7.19566822e-01 1.39587419e-03 6.90136611e-01 8.48994532e-05 -4.85332251e-01 6.26328886e-01 4.97029841e-01 6.37829751e-02 2.95736119e-02 -2.09917635e-01 4.79079753e-01 5.38904727e-01 -1.45459190e-01 2.81869441e-01 7.29895905e-02 1.25796211e+00 -1.12195969e+00 -6.07812762e-01 -2.17237826e-02 4.10932332e-01 -3.71210873e-01 9.43374634e-02 -5.19021332e-01 4.13091630e-01 -4.95427579e-01 3.99375468e-01 -1.09849915e-01 -7.12451398e-01 3.79660219e-01 3.45005661e-01 -6.58580065e-02 -4.67915893e-01 -7.46294796e-01 1.70958006e+00 -4.11110908e-01 7.32856095e-01 6.21461868e-01 -6.74018145e-01 7.96714485e-01 3.33994627e-01 1.86475828e-01 -8.19265366e-01 4.76884007e-01 2.67099768e-01 3.93997341e-01 -7.07810819e-01 4.26421583e-01 -4.25700426e-01 -1.89606741e-01 8.91957641e-01 2.12864190e-01 -6.93102062e-01 1.89832166e-01 5.72806478e-01 8.56919587e-01 -2.84014735e-03 4.32432704e-02 -4.43910390e-01 4.53924954e-01 1.48818985e-01 1.17220454e-01 8.43986511e-01 -2.73214191e-01 1.30048975e-01 4.08120006e-01 -6.54375076e-01 -1.20160520e+00 -1.12884867e+00 6.04171634e-01 1.27416074e+00 3.54473978e-01 1.63111966e-02 -9.48984504e-01 2.42294833e-01 -1.21176846e-01 9.05694485e-01 -5.59390903e-01 -1.86715886e-01 -3.33654344e-01 -6.80501938e-01 7.16310501e-01 -1.80947185e-01 5.06351233e-01 -1.46505463e+00 -1.37317705e+00 -1.12387486e-01 -6.23254776e-02 -9.57512975e-01 1.12821860e-02 -5.24439290e-02 -5.00859261e-01 -6.03794217e-01 -1.10727131e+00 -8.82072926e-01 7.16008186e-01 2.00254887e-01 1.23571289e+00 2.01936334e-01 -1.13834888e-01 9.16346669e-01 -1.63138404e-01 -7.20419288e-01 -9.14350450e-01 -6.11638390e-02 4.60427046e-01 -7.14556947e-02 1.54422224e-01 -8.46038461e-01 -6.06163740e-01 1.76582988e-02 -1.02936995e+00 5.44746876e-01 8.45069587e-01 5.12436211e-01 -4.45204973e-01 -4.70730424e-01 8.06575835e-01 -3.92625868e-01 9.48176384e-01 -5.53411007e-01 -3.17944199e-01 5.70124447e-01 -2.79704630e-01 3.27353299e-01 4.29083794e-01 -5.10740399e-01 -1.23257852e+00 -3.48080188e-01 4.19058502e-01 9.00610760e-02 -1.15730464e-01 3.21515203e-01 1.39364675e-01 -2.39996612e-01 7.28078306e-01 3.11081111e-01 6.29718482e-01 -3.05829719e-02 8.34574997e-01 5.08363307e-01 4.92118388e-01 -8.42052341e-01 8.00976932e-01 4.41229343e-01 -4.14917953e-02 -1.05024743e+00 -2.39771903e-01 2.21925333e-01 -2.28381887e-01 -4.92961824e-01 1.14805079e+00 -1.00252914e+00 -1.31850159e+00 4.73128647e-01 -1.31698716e+00 -6.44729793e-01 -5.00054300e-01 3.38216156e-01 -8.78679454e-01 2.68710315e-01 -5.77018559e-01 -9.90514576e-01 -2.11937800e-01 -9.23184216e-01 6.08170748e-01 7.56502926e-01 -5.10848820e-01 -1.10985720e+00 2.45923832e-01 5.99212170e-01 8.44957709e-01 -1.75157581e-02 1.18580461e+00 -4.63760108e-01 -5.31144202e-01 3.96284312e-01 -1.68689072e-01 -8.81305113e-02 -1.97241992e-01 -2.39668369e-01 -6.28861845e-01 -1.40197277e-01 -6.58279732e-02 -9.46738362e-01 5.10961592e-01 -3.81319821e-02 2.18188222e-02 -4.02577490e-01 -1.46836475e-01 2.31272906e-01 1.23629558e+00 3.90954733e-01 4.20857519e-01 3.97335589e-01 4.30410802e-01 1.08385754e+00 -2.37867147e-01 3.19543332e-01 1.06505418e+00 -6.23422675e-02 4.37035590e-01 -2.75846541e-01 1.23783760e-01 -2.10080538e-02 6.73954785e-01 1.10060751e+00 -4.69112307e-01 -3.93767357e-01 -1.19675779e+00 4.62990254e-01 -2.02946234e+00 -8.62987399e-01 2.61186421e-01 1.67368734e+00 8.13955843e-01 -2.17193872e-01 7.94720312e-04 -7.31162310e-01 5.04989505e-01 1.30044013e-01 -6.96900129e-01 -7.18367696e-01 -5.90898395e-01 -4.04454321e-01 1.36867955e-01 4.32811409e-01 -4.71350998e-01 1.09339273e+00 6.47440386e+00 2.79476583e-01 -8.55545521e-01 -3.06387004e-02 5.75997412e-01 -1.57517761e-01 -5.77341259e-01 1.87344760e-01 -1.99477211e-01 1.69951394e-02 1.14279056e+00 -2.85167426e-01 1.02315032e+00 2.16583297e-01 1.78785622e-01 -5.89557886e-01 -1.00099516e+00 9.56096888e-01 2.86292017e-01 -1.38366950e+00 3.18936646e-01 1.76177263e-01 8.19321394e-01 1.94380343e-01 -2.04887185e-02 1.11960940e-01 1.20730472e+00 -1.30247283e+00 9.50821042e-01 8.71101081e-01 2.89982170e-01 -4.19263601e-01 4.60268617e-01 7.32725203e-01 -6.29194081e-01 -3.35294276e-01 -2.68337011e-01 -3.61811966e-01 3.33508372e-01 -6.83391392e-02 -2.76930988e-01 2.87753284e-01 2.73554265e-01 -2.32336111e-02 -5.01353025e-01 4.15230334e-01 2.64766663e-01 -1.97533503e-01 -2.06565097e-01 -5.94541132e-01 4.46697176e-01 -3.70768607e-01 6.37452960e-01 7.59003937e-01 1.75758570e-01 4.07169789e-01 -1.81039914e-01 1.26167858e+00 2.68133171e-02 -9.17524025e-02 -8.40643048e-01 -5.03317416e-01 6.78459480e-02 1.24297297e+00 -8.21768522e-01 -3.89550418e-01 -3.72004151e-01 9.03536260e-01 1.28643036e-01 4.55065697e-01 -5.34194052e-01 -3.05875272e-01 5.14888465e-01 -1.60786718e-01 -3.16986620e-01 -3.68954629e-01 -5.15467972e-02 -1.40773213e+00 -7.19515800e-01 -1.27313101e+00 -2.13244587e-01 -1.30005181e+00 -1.22217262e+00 6.33121789e-01 -2.42325533e-02 -3.59043270e-01 -4.43354607e-01 -5.16414881e-01 -4.60904568e-01 4.79470998e-01 -9.48737562e-01 -1.45509136e+00 -5.62391281e-02 6.07260823e-01 2.76018798e-01 -5.14549911e-01 8.57567310e-01 -3.42541426e-01 -3.57622802e-01 8.21424723e-02 7.87890181e-02 7.68045783e-02 4.52156752e-01 -9.81107831e-01 3.66916806e-01 2.79188842e-01 9.75370556e-02 6.88199639e-01 9.12136734e-01 -3.51318449e-01 -1.99702036e+00 -2.12223858e-01 8.40023994e-01 -5.92795074e-01 9.81845260e-01 -4.34175819e-01 -4.40878928e-01 6.08425915e-01 9.48460996e-01 -8.19946647e-01 3.37300718e-01 -1.28454044e-01 -2.24759355e-01 1.17196754e-01 -9.99727666e-01 1.09461057e+00 9.13339674e-01 -4.88803923e-01 -8.57614160e-01 3.45880389e-01 8.75710964e-01 3.66750032e-01 -4.98274952e-01 -1.33038774e-01 7.25674152e-01 -9.30886209e-01 9.11168694e-01 -5.23097694e-01 4.85087037e-01 -2.90793329e-01 -1.26899675e-01 -1.41493249e+00 -9.55246612e-02 -9.33274508e-01 1.97431520e-01 8.48485947e-01 4.26611960e-01 -1.12022614e+00 4.33594733e-01 8.88895452e-01 1.01396114e-01 -2.53679037e-01 -8.09680521e-01 -4.41222876e-01 3.14406902e-01 1.55807093e-01 4.94217098e-01 1.01313460e+00 6.13777876e-01 7.61375427e-01 -2.70438343e-01 -3.86110157e-01 7.13920593e-01 -1.88458204e-01 7.89014995e-01 -1.45812893e+00 -3.76357108e-01 -6.29340172e-01 -8.30171108e-02 -6.27681613e-01 2.99582541e-01 -7.46185124e-01 1.29364014e-01 -1.70316017e+00 3.94317478e-01 1.09190717e-02 9.31689739e-02 3.97505552e-01 4.31400359e-01 -4.59548738e-03 6.84042275e-01 4.24525797e-01 -1.04206085e+00 4.51747835e-01 1.49738443e+00 -8.15431848e-02 -8.66825581e-02 -8.41939569e-01 -1.01265907e+00 1.07230282e+00 7.38006234e-01 -3.53436614e-03 -5.46935141e-01 -6.78855717e-01 9.58805740e-01 2.43331879e-01 6.19878232e-01 -9.78336096e-01 6.74004972e-01 -4.58282411e-01 2.36610413e-01 2.14901179e-01 5.71824729e-01 -5.55942833e-01 3.00431788e-01 6.66259646e-01 -3.44086736e-01 2.70910591e-01 -1.97679132e-01 3.42114002e-01 1.64655328e-01 -1.08310878e-01 5.74236274e-01 -9.82350707e-01 -6.73611343e-01 -4.82609600e-01 -1.08249211e+00 2.72479206e-02 1.10450470e+00 -1.16119608e-01 -7.08801091e-01 -8.21910203e-01 -4.69985723e-01 5.05143464e-01 9.22563374e-01 2.56678790e-01 4.35023993e-01 -7.84054279e-01 -7.47720599e-01 -6.52371943e-02 -4.55151796e-02 -2.09292382e-01 3.21346074e-02 6.83193207e-01 -7.35159457e-01 5.35248995e-01 -4.93503839e-01 -1.29928276e-01 -6.53900802e-01 4.20492828e-01 5.88975906e-01 1.39848813e-01 -2.76767880e-01 9.84507680e-01 6.31390274e-01 -5.43019652e-01 -3.81495990e-02 1.06088199e-01 -5.86627163e-02 1.08583774e-02 2.59840906e-01 1.85257003e-01 -9.24745083e-01 -8.19665194e-01 -2.36033008e-01 4.61872965e-01 2.72257090e-01 -6.72167599e-01 1.25412369e+00 -4.49417382e-01 -6.86870933e-01 6.36429965e-01 5.08275986e-01 -3.53547394e-01 -8.35831046e-01 1.01392411e-01 -1.63234130e-01 2.68834382e-01 -3.70822132e-01 -8.59874785e-01 -6.34245932e-01 8.10965657e-01 1.38908222e-01 5.20524442e-01 8.65923107e-01 4.23568994e-01 6.97861016e-01 8.53596210e-01 5.89092076e-01 -1.22430694e+00 4.24244523e-01 5.88010371e-01 8.40580285e-01 -1.24709737e+00 -2.05788285e-01 3.84782046e-01 -1.02030933e+00 9.75094080e-01 5.53036630e-01 -8.19902420e-02 1.26229867e-01 6.60083294e-02 1.60351187e-01 -4.69384402e-01 -1.31979394e+00 -2.78564036e-01 -2.06763014e-01 7.15785861e-01 1.27226532e-01 1.43854529e-01 -2.16288902e-02 1.40480742e-01 -2.99771696e-01 -3.82806733e-03 6.66305661e-01 8.64107728e-01 -9.22926307e-01 -8.76144886e-01 -5.99929929e-01 -6.64154142e-02 1.23349726e-01 1.31254032e-01 -9.18253481e-01 7.36111879e-01 2.99518943e-01 1.19823098e+00 8.95950794e-02 -2.71597654e-01 -1.47165373e-01 7.82648996e-02 4.77692813e-01 -2.37445220e-01 -5.84799528e-01 -2.62279570e-01 1.30041882e-01 -2.23518506e-01 -6.43545270e-01 -3.84574741e-01 -1.40741491e+00 -8.68836522e-01 3.64517830e-02 2.15478972e-01 5.71903586e-01 1.01693904e+00 3.66611898e-01 7.33632669e-02 1.84519351e-01 -9.49348450e-01 -4.41098154e-01 -5.80922663e-01 -4.25784796e-01 1.81149766e-01 3.35815489e-01 -2.17689559e-01 -3.44299406e-01 1.56020686e-01]

[4.369327545166016, 1.5192407369613647]

76d606f2-2331-4dfc-bc2a-99e44ffb306f

unsupervised-translation-of-german-lower-1

https://aclanthology.org/2021.wmt-1.104

https://aclanthology.org/2021.wmt-1.104.pdf

Unsupervised Translation of German–Lower Sorbian: Exploring Training and Novel Transfer Methods on a Low-Resource Language

This paper describes the methods behind the systems submitted by the University of Groningen for the WMT 2021 Unsupervised Machine Translation task for German–Lower Sorbian (DE–DSB): a high-resource language to a low-resource one. Our system uses a transformer encoder-decoder architecture in which we make three changes to the standard training procedure. First, our training focuses on two languages at a time, contrasting with a wealth of research on multilingual systems. Second, we introduce a novel method for initializing the vocabulary of an unseen language, achieving improvements of 3.2 BLEU for DE->DSB and 4.0 BLEU for DSB->DE.Lastly, we experiment with the order in which offline and online back-translation are used to train an unsupervised system, finding that using online back-translation first works better for DE->DSB by 2.76 BLEU. Our submissions ranked first (tied with another team) for DSB->DE and third for DE->DSB.

['Gertjan van Noord', 'Antonio Toral', 'Ahmet Üstün', 'Lukas Edman']

wmt-emnlp-2021-11

['unsupervised-machine-translation']

['natural-language-processing']

[-1.16837539e-01 1.65955946e-01 -3.02252531e-01 -4.09496248e-01 -1.39422524e+00 -8.98631155e-01 1.02828074e+00 -9.87471640e-02 -7.30475068e-01 1.32988584e+00 3.65369767e-01 -8.32030833e-01 2.71601081e-01 -4.30945486e-01 -7.29692757e-01 -1.99545667e-01 3.70945454e-01 1.22927296e+00 -2.01430231e-01 -8.68986368e-01 -1.60734430e-01 2.45298594e-01 -7.62318194e-01 4.58501965e-01 6.93298638e-01 1.19376168e-01 2.85164177e-01 6.47507727e-01 7.39043439e-03 3.24749857e-01 -4.62409377e-01 -9.53363240e-01 4.45054650e-01 -6.32719219e-01 -1.29643595e+00 -3.89562041e-01 5.32152474e-01 -2.18184218e-01 -2.63101101e-01 7.50442505e-01 6.44079566e-01 -5.78172132e-02 5.43091297e-01 -7.31824636e-01 -6.53183818e-01 1.02016914e+00 -2.55068392e-01 2.32779130e-01 5.83833516e-01 -1.34215161e-01 1.10705090e+00 -1.03794026e+00 1.10765707e+00 1.25366592e+00 6.39277041e-01 7.46263266e-01 -1.13401067e+00 -5.81938446e-01 -2.05617771e-01 -1.23936899e-01 -1.33993542e+00 -8.84804904e-01 1.00750774e-01 -5.18041492e-01 1.65566635e+00 3.17316204e-01 4.29474741e-01 1.44793272e+00 2.12217763e-01 5.91932356e-01 1.33609486e+00 -8.30281556e-01 -3.82489890e-01 5.47540069e-01 -2.78276026e-01 4.13076729e-01 2.22508684e-01 8.44960362e-02 -6.26323998e-01 1.49477482e-01 5.04541934e-01 -6.41347408e-01 7.34149292e-02 1.21196754e-01 -1.66356313e+00 7.44732499e-01 -8.77321362e-02 5.01356483e-01 -6.82569668e-02 -2.86523014e-01 5.18433750e-01 8.72521341e-01 5.75247347e-01 7.65380502e-01 -8.97862673e-01 -4.64673907e-01 -1.17297006e+00 9.86001119e-02 9.52356398e-01 1.25181162e+00 6.01087987e-01 -1.81032255e-01 -3.73756103e-02 1.05467653e+00 1.71156332e-01 6.15604937e-01 5.57153702e-01 -5.54622710e-01 9.40427184e-01 -5.52310310e-02 3.25066030e-01 -4.52712595e-01 -2.03966916e-01 -4.61811662e-01 -4.70320433e-01 -4.82132643e-01 5.15247762e-01 -3.06696206e-01 -8.81546438e-01 1.60949016e+00 -8.18597302e-02 -7.82665551e-01 3.71438533e-01 8.23731840e-01 6.85362101e-01 8.66123557e-01 -2.23547742e-01 -3.16947192e-01 1.25893462e+00 -1.37693632e+00 -7.37112999e-01 -4.01545197e-01 1.02774608e+00 -1.43701291e+00 1.05198872e+00 3.38751823e-01 -1.38782477e+00 -4.61022079e-01 -8.92480671e-01 -3.77760977e-01 -6.72884524e-01 3.81988108e-01 3.41044694e-01 6.26389921e-01 -1.49173796e+00 5.16003132e-01 -5.89611351e-01 -1.00205588e+00 -4.21039701e-01 3.09025645e-01 -5.59087634e-01 -7.53401294e-02 -1.64149725e+00 1.44770348e+00 5.03179550e-01 -1.22222200e-01 -8.14187646e-01 -3.57397318e-01 -7.19989121e-01 -4.64385241e-01 -1.53031364e-01 -4.94031876e-01 1.53087866e+00 -9.04478431e-01 -1.80651224e+00 1.39977908e+00 -2.83251733e-01 -4.65195745e-01 8.10805857e-01 -3.13258350e-01 -8.32772613e-01 -2.66565382e-01 4.97879952e-01 6.50912285e-01 2.80246496e-01 -7.80003488e-01 -7.34649003e-01 -6.71964735e-02 4.81469417e-03 5.09662449e-01 -2.41801336e-01 5.92813849e-01 -4.28394765e-01 -6.95982158e-01 -1.10781146e-02 -1.00720549e+00 1.91444699e-02 -8.04192483e-01 -2.86686033e-01 -1.02938734e-01 2.52135515e-01 -1.23661339e+00 1.18807602e+00 -1.81578076e+00 3.32600564e-01 -2.85546649e-02 -4.23849672e-02 2.40033507e-01 -1.78212285e-01 1.02686584e+00 -2.12298512e-01 2.27850258e-01 3.31148393e-02 -4.30724204e-01 7.10458262e-03 2.06517264e-01 -3.12568247e-01 2.19537452e-01 2.32615367e-01 8.44673455e-01 -1.19421911e+00 -3.39670807e-01 -1.93097845e-01 1.97123528e-01 -2.30993405e-01 5.17724641e-02 1.43069193e-01 5.06711245e-01 9.95396078e-03 6.03040457e-01 5.24062514e-01 1.37017474e-01 4.78495419e-01 1.20931864e-01 -4.87029910e-01 9.82239127e-01 -5.88567317e-01 1.80264497e+00 -8.07010412e-01 9.19121027e-01 5.50027601e-02 -6.87912464e-01 9.89527166e-01 5.91736913e-01 2.33397931e-01 -8.66040587e-01 -5.18187508e-02 1.10997796e+00 7.01151565e-02 -1.88769579e-01 8.41895223e-01 -4.10037488e-01 -2.20625296e-01 4.13237512e-01 3.45511734e-01 -3.12573433e-01 5.54324329e-01 2.10547790e-01 7.53785014e-01 4.30127859e-01 1.91739634e-01 -6.76198542e-01 4.16891813e-01 1.82975248e-01 5.05914211e-01 5.13953686e-01 -1.38452992e-01 5.71323574e-01 2.34368786e-01 -3.41903210e-01 -1.47408783e+00 -8.61676276e-01 -1.48632303e-01 1.09928668e+00 -2.34858468e-01 -6.40787840e-01 -7.75043905e-01 -6.04093909e-01 -2.76092947e-01 8.22218537e-01 -2.35741839e-01 5.96276298e-02 -7.75224149e-01 -5.53968072e-01 8.48146021e-01 1.32428482e-01 4.69615340e-01 -7.76232064e-01 7.40160793e-02 3.00122678e-01 -6.81767285e-01 -1.27229476e+00 -7.30350494e-01 2.83622384e-01 -7.68236041e-01 -3.36712509e-01 -1.04309845e+00 -1.09028709e+00 3.86719346e-01 -1.18907839e-01 1.52003241e+00 -3.02574158e-01 1.92660496e-01 1.07991330e-01 -3.98660213e-01 -3.06314021e-01 -8.10462177e-01 7.05805242e-01 3.80707353e-01 -5.20616651e-01 6.49312496e-01 -6.58409595e-02 -1.44833654e-01 4.19689983e-01 -4.44316268e-01 2.56607085e-01 6.68778121e-01 9.11749899e-01 2.63524383e-01 -6.24783635e-01 2.59672970e-01 -9.12419915e-01 7.19739735e-01 -3.19512457e-01 -5.03148139e-01 3.28157246e-01 -8.02487969e-01 2.74635293e-02 6.15892410e-01 -1.91725150e-01 -7.12168515e-01 -9.12267640e-02 -2.63464838e-01 1.28542587e-01 1.71394199e-01 4.40073609e-01 -1.41137213e-01 1.60953283e-01 6.49779141e-01 8.09060559e-02 -2.64102638e-01 -7.24678218e-01 3.29031616e-01 1.14718795e+00 2.02382818e-01 -7.18714476e-01 7.34877408e-01 -5.12218215e-02 -7.19972253e-01 -5.82254231e-01 -5.03734589e-01 -5.54664731e-01 -8.28008175e-01 8.23789090e-02 7.50884712e-01 -1.16855586e+00 -7.46892486e-03 4.48349923e-01 -1.27444160e+00 -5.06491244e-01 -2.05325603e-01 8.65869164e-01 -5.26617050e-01 4.54466119e-02 -7.94894814e-01 -3.59826952e-01 -4.71577317e-01 -1.38599086e+00 1.26080179e+00 -2.28508249e-01 -5.44465721e-01 -1.09905767e+00 4.64904040e-01 5.41590512e-01 3.34260583e-01 -3.44003737e-01 6.66279674e-01 -7.78865933e-01 3.75222974e-02 1.78885609e-02 -1.70846656e-01 3.57177585e-01 2.14006051e-01 -1.84912816e-01 -5.65610945e-01 -7.00845480e-01 -3.10874254e-01 -3.50810945e-01 4.04889017e-01 -1.48376778e-01 6.67480975e-02 -3.90261650e-01 -3.13890904e-01 4.67537612e-01 1.21954775e+00 5.30047677e-02 5.97744167e-01 7.55778790e-01 4.21452194e-01 3.99020642e-01 6.11479700e-01 -7.88137913e-02 5.40097356e-01 9.98301744e-01 -4.59392995e-01 -2.90006191e-01 -3.08890194e-01 -3.85952085e-01 9.47135627e-01 1.61280084e+00 -8.14309791e-02 -2.36616358e-01 -1.17359376e+00 1.02403140e+00 -1.37079144e+00 -5.35205185e-01 -7.89463222e-02 2.39383411e+00 1.45954001e+00 1.13033906e-01 2.91092575e-01 -3.91841620e-01 7.11609006e-01 -9.58710760e-02 2.34799072e-01 -1.16022575e+00 -2.61682272e-01 4.49431807e-01 8.07208776e-01 9.04513121e-01 -8.26013684e-01 1.56309998e+00 6.96643972e+00 6.31248355e-01 -1.13887322e+00 2.65903801e-01 3.67802143e-01 -1.21116064e-01 -3.53515297e-01 3.68011713e-01 -1.07506955e+00 3.68419915e-01 1.51528084e+00 -2.64512956e-01 7.14630961e-01 3.65358382e-01 1.81625202e-01 2.32150346e-01 -1.28300965e+00 8.46901596e-01 3.63207683e-02 -9.75179732e-01 2.03275099e-01 1.47760332e-01 1.01438105e+00 5.49502730e-01 -1.42501846e-01 6.90234244e-01 5.87040246e-01 -9.82267976e-01 8.46702158e-01 2.57127315e-01 1.13881731e+00 -5.13165236e-01 8.19209039e-01 2.77869970e-01 -8.54762852e-01 5.35997927e-01 -4.02605653e-01 -2.63029337e-02 1.02444664e-01 5.24994791e-01 -1.20033276e+00 8.25965047e-01 6.75705314e-01 6.57885313e-01 -4.22387123e-01 5.04123628e-01 -3.34607333e-01 6.69823170e-01 -3.59863400e-01 -2.53487285e-02 4.86906528e-01 -3.17271054e-01 6.69522405e-01 1.70401657e+00 4.43738788e-01 -3.51800978e-01 1.28966108e-01 2.43040591e-01 -4.32392448e-01 4.09826756e-01 -7.66554832e-01 -2.94770688e-01 4.14067447e-01 9.94042456e-01 -3.16065252e-01 -5.60021758e-01 -1.74650669e-01 1.31331670e+00 3.63669097e-01 3.17458898e-01 -7.39675045e-01 -6.22354269e-01 5.05557835e-01 1.45873472e-01 1.11740686e-01 -4.64589506e-01 -2.92417035e-02 -1.40562522e+00 8.45170543e-02 -1.51735890e+00 1.91015169e-01 -7.47957528e-01 -1.04307139e+00 9.32599783e-01 -8.66486281e-02 -1.17334080e+00 -5.45501053e-01 -6.10538840e-01 1.57580674e-01 1.35613036e+00 -1.41357696e+00 -1.09702373e+00 3.80007893e-01 2.45828196e-01 7.67068624e-01 -4.86031055e-01 1.25857818e+00 7.26877570e-01 -3.95700008e-01 8.97659659e-01 5.92297614e-01 3.06109637e-01 1.45368493e+00 -1.19677234e+00 9.20682549e-01 8.67588520e-01 3.46140176e-01 8.85170937e-01 6.88277423e-01 -7.53690600e-01 -1.14724159e+00 -7.98555017e-01 2.06960154e+00 -7.52601027e-01 9.24108863e-01 -6.57838762e-01 -3.38261425e-01 9.34830070e-01 7.27165818e-01 -6.18085682e-01 5.36976159e-01 3.86188656e-01 -3.95995736e-01 -3.10870837e-02 -9.80065584e-01 7.56430030e-01 1.00758255e+00 -9.06705499e-01 -7.58770585e-01 7.34314382e-01 6.67902708e-01 -6.83160365e-01 -1.08628178e+00 4.52107303e-02 5.87050080e-01 -3.84648740e-01 6.63003564e-01 -8.84101331e-01 6.02562428e-01 -6.82728216e-02 -2.37780839e-01 -1.60671926e+00 -2.70173013e-01 -1.19046199e+00 3.93309474e-01 1.05645454e+00 1.00781858e+00 -8.54172051e-01 2.67282128e-01 5.41061610e-02 -8.24212953e-02 -3.52110386e-01 -9.98012006e-01 -9.68324423e-01 5.26647866e-01 -5.43762110e-02 5.26185453e-01 1.13479877e+00 1.03369482e-01 6.70092881e-01 -4.06259388e-01 -2.85519838e-01 1.85978830e-01 -1.74800754e-01 7.26029634e-01 -7.97478914e-01 -2.83333868e-01 -3.90791208e-01 -1.66196510e-01 -1.15816271e+00 -1.34686381e-01 -1.25821984e+00 7.25407377e-02 -1.57992399e+00 8.85701403e-02 -2.34318227e-01 -1.12823203e-01 5.48769534e-01 4.24202019e-03 4.59336966e-01 9.92106572e-02 4.47350651e-01 -2.39243165e-01 9.37914848e-02 1.10636771e+00 -2.14918390e-01 -7.15335459e-02 -2.00183570e-01 -6.82204843e-01 2.67081439e-01 7.02146053e-01 -5.13831317e-01 -9.66284573e-02 -1.12312627e+00 4.71010625e-01 -1.97449476e-01 -2.27518320e-01 -6.03704989e-01 -4.33214679e-02 1.28881052e-01 6.92198500e-02 -4.27936524e-01 8.52737576e-02 -4.99091774e-01 4.20795456e-02 2.32306212e-01 -3.38954240e-01 5.84826112e-01 4.47287470e-01 -1.67321071e-01 -2.78774232e-01 -2.14262068e-01 6.44181430e-01 -1.79738760e-01 -4.19806033e-01 -9.52181742e-02 -6.11367643e-01 3.01967174e-01 5.15772402e-01 -6.99689612e-02 -1.60648152e-01 -4.27372843e-01 -7.59709775e-01 1.24202870e-01 4.55823898e-01 6.76141798e-01 1.41922161e-01 -1.31055176e+00 -1.30914104e+00 1.95723057e-01 1.18394248e-01 -5.29827833e-01 -5.13672709e-01 9.76590753e-01 -8.76174867e-01 7.37802684e-01 -2.16466159e-01 -3.09976637e-01 -9.82680500e-01 3.02466042e-02 3.98593783e-01 -7.17171013e-01 -3.60917240e-01 9.92274106e-01 -3.57600123e-01 -9.97520030e-01 -1.62187532e-01 6.38689995e-02 2.30504826e-01 1.76497772e-01 1.24830902e-01 1.70973301e-01 6.11119926e-01 -8.39720905e-01 -5.25939167e-01 3.60198975e-01 -5.22148907e-01 -7.84317076e-01 1.06964302e+00 -2.54834622e-01 -4.05791372e-01 6.13904238e-01 1.36730957e+00 3.77117097e-01 -3.78559798e-01 -2.63196379e-01 2.57043183e-01 -1.62521973e-01 -4.08797786e-02 -1.18911791e+00 -5.09046376e-01 7.27501094e-01 3.13362479e-01 -1.69621900e-01 8.19112122e-01 -2.79930055e-01 1.03709364e+00 6.06122971e-01 5.80790997e-01 -1.51091754e+00 -5.98122835e-01 1.07231307e+00 8.42295349e-01 -1.12724054e+00 2.91480385e-02 -1.81051325e-02 -5.70205808e-01 1.11025155e+00 1.43557802e-01 2.39423007e-01 1.80182934e-01 1.07008763e-01 4.33499724e-01 1.71302289e-01 -5.37940562e-01 -4.41081636e-02 2.55933017e-01 3.19230199e-01 9.86283302e-01 2.96123385e-01 -7.86121547e-01 2.37146437e-01 -7.89407909e-01 -8.54810253e-02 3.29574168e-01 7.90910184e-01 8.17875564e-02 -1.72098505e+00 -2.44422555e-01 3.89129221e-02 -4.94848073e-01 -6.35923803e-01 -8.39201152e-01 1.07927799e+00 -9.29045901e-02 9.13401484e-01 -1.25170708e-01 -4.17086214e-01 2.67887354e-01 3.37446034e-01 5.99191070e-01 -8.36570084e-01 -9.12324488e-01 2.62156844e-01 6.81052387e-01 -2.91979015e-01 -2.81539679e-01 -7.89240420e-01 -7.73803115e-01 -4.46110904e-01 -2.47874573e-01 6.97202861e-01 8.78220737e-01 9.65157092e-01 1.74289361e-01 1.69016197e-01 6.99601412e-01 -3.36083591e-01 -6.14291608e-01 -1.14872122e+00 -2.31151104e-01 1.31452441e-01 1.43117875e-01 -1.48011133e-01 -2.33094484e-01 1.51493087e-01]

[11.494739532470703, 10.336833000183105]

5f43468e-05e1-4fcf-b5e7-4e1d4b850fab

beyond-gauss-image-set-matching-on-the

1507.08711

http://arxiv.org/abs/1507.08711v1

http://arxiv.org/pdf/1507.08711v1.pdf

Beyond Gauss: Image-Set Matching on the Riemannian Manifold of PDFs

State-of-the-art image-set matching techniques typically implicitly model each image-set with a Gaussian distribution. Here, we propose to go beyond these representations and model image-sets as probability distribution functions (PDFs) using kernel density estimators. To compare and match image-sets, we exploit Csiszar f-divergences, which bear strong connections to the geodesic distance defined on the space of PDFs, i.e., the statistical manifold. Furthermore, we introduce valid positive definite kernels on the statistical manifolds, which let us make use of more powerful classification schemes to match image-sets. Finally, we introduce a supervised dimensionality reduction technique that learns a latent space where f-divergences reflect the class labels of the data. Our experiments on diverse problems, such as video-based face recognition and dynamic texture classification, evidence the benefits of our approach over the state-of-the-art image-set matching methods.

['Mathieu Salzmann', 'Mehrtash Harandi', 'Mahsa Baktashmotlagh']

2015-07-31

beyond-gauss-image-set-matching-on-the-1

http://openaccess.thecvf.com/content_iccv_2015/html/Harandi_Beyond_Gauss_Image-Set_ICCV_2015_paper.html

http://openaccess.thecvf.com/content_iccv_2015/papers/Harandi_Beyond_Gauss_Image-Set_ICCV_2015_paper.pdf

iccv-2015-12

['set-matching', 'supervised-dimensionality-reduction']

['computer-vision', 'computer-vision']

[-1.12027273e-01 -3.49594146e-01 -1.78237081e-01 -5.55973649e-01 -3.96522433e-01 -4.68146265e-01 6.42713368e-01 -1.28052324e-01 -2.55848318e-01 1.42536879e-01 -2.57559001e-01 8.17333758e-02 -5.09869576e-01 -8.34891677e-01 -6.22735560e-01 -9.52667415e-01 -1.18126482e-01 3.85206848e-01 -9.95643996e-03 -1.26300126e-01 5.45872509e-01 6.15201831e-01 -1.71206486e+00 2.31065247e-02 6.49774134e-01 9.89285588e-01 -1.02345094e-01 3.96179438e-01 -1.27002910e-01 3.27825904e-01 -6.14689365e-02 -5.92110097e-01 2.10772172e-01 -3.63254488e-01 -6.82870507e-01 3.45065087e-01 9.00172532e-01 -2.12420840e-02 -5.25764823e-01 1.41138339e+00 -5.60456477e-02 2.13018551e-01 1.20008254e+00 -1.55626345e+00 -1.01514113e+00 -2.89945960e-01 -5.62376022e-01 2.56438348e-02 2.18107909e-01 -2.78496414e-01 7.52865791e-01 -9.54161704e-01 6.39882088e-01 1.43198907e+00 8.21578801e-01 4.93021131e-01 -1.50095224e+00 -3.74935567e-01 -2.78863341e-01 2.74718374e-01 -1.51901352e+00 -5.27366757e-01 7.59393573e-01 -7.71540165e-01 2.14631066e-01 2.00630069e-01 3.64664048e-01 7.65387833e-01 3.06897074e-01 4.60449219e-01 1.09030235e+00 -5.51353872e-01 2.78093040e-01 1.84419945e-01 2.26457641e-01 1.04670405e+00 1.05426669e-01 1.22698862e-02 -3.82105798e-01 -3.78333807e-01 1.09705138e+00 1.55523151e-01 -2.73685038e-01 -1.04111648e+00 -1.08522952e+00 1.01635540e+00 -1.65960938e-02 4.79984224e-01 -2.10255474e-01 1.13295093e-01 1.32616699e-01 4.20324594e-01 6.86115205e-01 -1.57252923e-01 1.39098689e-01 7.94346258e-02 -7.64057279e-01 1.31408468e-01 8.30379069e-01 8.24161768e-01 1.10847998e+00 -4.38071579e-01 -1.34034157e-02 8.40019584e-01 5.86656749e-01 6.27110243e-01 1.66537583e-01 -1.01386940e+00 1.29421949e-01 3.32350522e-01 -6.74574822e-02 -1.25201881e+00 9.00924429e-02 1.71646297e-01 -7.80516148e-01 1.74114689e-01 7.37915039e-01 5.51265895e-01 -5.54096103e-01 1.75454235e+00 3.15513104e-01 5.74986815e-01 -2.76679322e-02 9.27315593e-01 1.66818351e-01 2.82096773e-01 -2.66071588e-01 5.32499589e-02 1.22833669e+00 -4.34584707e-01 -4.15583432e-01 5.13042986e-01 5.63216269e-01 -7.71348774e-01 1.08504033e+00 -1.40713574e-02 -9.35065448e-01 -3.67556036e-01 -8.50871623e-01 2.83382028e-01 -2.24010110e-01 1.02635846e-01 5.37940204e-01 8.88487637e-01 -1.12990808e+00 1.03815484e+00 -1.10141504e+00 -5.19097388e-01 3.87071013e-01 2.35381678e-01 -6.67714059e-01 -1.43686786e-01 -6.49737000e-01 5.97396314e-01 -6.21283911e-02 -1.02837890e-01 -5.80325365e-01 -8.79090965e-01 -9.16783571e-01 -1.69927269e-01 1.46278413e-02 -6.40196562e-01 6.98300004e-01 -9.57471013e-01 -1.45109904e+00 1.28524375e+00 -1.32930830e-01 5.20578772e-02 2.88039237e-01 1.09010108e-01 -4.17764395e-01 5.77034652e-01 -4.50946763e-02 4.64236200e-01 1.29492891e+00 -1.11591887e+00 -1.58992067e-01 -5.56142330e-01 -2.45753288e-01 -1.36941046e-01 -7.03522980e-01 2.54557151e-02 -2.14731708e-01 -5.40503561e-01 2.12587014e-01 -9.91910100e-01 8.72378796e-02 5.68861723e-01 -1.75584123e-01 -3.92437607e-01 8.51400316e-01 -4.20143157e-01 1.03096831e+00 -2.46906257e+00 4.56907898e-01 5.01837254e-01 2.27742121e-01 -8.00284892e-02 -3.55921268e-01 3.03233773e-01 -2.94930220e-01 -5.41339442e-02 -3.21409345e-01 -5.23380756e-01 5.98893091e-02 3.42105359e-01 -1.90664336e-01 1.14331615e+00 3.47388834e-01 6.93717360e-01 -8.12004685e-01 -3.07624310e-01 2.92756826e-01 6.17023826e-01 -4.17252749e-01 9.58666652e-02 1.11370847e-01 4.30380613e-01 -5.02051473e-01 2.36450762e-01 8.98621321e-01 -2.12811574e-01 -7.03938454e-02 -3.68534237e-01 9.97521728e-02 -1.58132181e-01 -1.16663361e+00 1.88367307e+00 -2.45262310e-01 5.49874604e-01 -6.34547323e-02 -1.29882216e+00 7.93523312e-01 -5.82210980e-02 6.52549982e-01 -4.41900313e-01 -4.44595143e-02 2.70635307e-01 -2.91716099e-01 -3.48653346e-01 2.54788667e-01 -1.17971890e-01 1.64040491e-01 4.95328188e-01 2.34342307e-01 1.05045356e-01 8.77258703e-02 1.43516377e-01 8.13340604e-01 1.22863360e-01 -2.71716028e-01 -9.22921062e-01 7.78143585e-01 -4.93086159e-01 9.50346664e-02 4.94526714e-01 1.07112646e-01 6.30141973e-01 4.69456702e-01 -2.09586143e-01 -1.14285731e+00 -1.44265246e+00 -7.22616613e-01 5.97521245e-01 3.21512222e-01 -2.97588646e-01 -1.09386992e+00 -6.09314263e-01 2.90181339e-01 2.81308085e-01 -7.12496758e-01 -2.64117748e-01 -3.48800868e-01 -4.83354568e-01 3.93188328e-01 3.25292617e-01 2.42840052e-01 -5.27329147e-01 -1.42491519e-01 -1.84821367e-01 1.59957603e-01 -9.32713389e-01 -7.41455078e-01 -5.29160678e-01 -9.08832014e-01 -1.36299014e+00 -8.85357141e-01 -8.87080789e-01 9.20906603e-01 3.92236829e-01 1.01657426e+00 -7.10964501e-02 -5.45351863e-01 1.18161607e+00 3.70708220e-02 1.27690643e-01 -5.18553257e-01 -4.11221117e-01 5.42027235e-01 5.51745594e-01 4.22849596e-01 -6.57465279e-01 -5.62185645e-01 7.11704254e-01 -1.05513024e+00 -2.53472358e-01 -8.35370738e-03 6.61819220e-01 7.42291510e-01 -7.90734543e-04 1.42987058e-01 -4.19794410e-01 5.05510151e-01 -4.32170779e-01 -8.90227497e-01 6.12653434e-01 -4.20321971e-01 4.40376550e-01 3.84022444e-01 -6.16913974e-01 -7.42839694e-01 -3.05907935e-01 4.56235290e-01 -8.94116461e-01 -1.29061088e-01 1.17187887e-01 9.11202095e-03 -6.02914691e-01 4.06773359e-01 1.74776882e-01 5.23496389e-01 -5.98462045e-01 6.05044961e-01 6.95610464e-01 6.21541917e-01 -6.84274077e-01 8.39820802e-01 8.72234702e-01 3.33356917e-01 -1.06526399e+00 -6.64988041e-01 -8.06676149e-01 -8.22224855e-01 -1.80597648e-01 9.36096728e-01 -5.25912046e-01 -9.60025191e-01 7.99331486e-01 -8.72710705e-01 -1.29616380e-01 -2.49904290e-01 6.12031937e-01 -9.47419941e-01 6.85842454e-01 -7.25863814e-01 -7.48959363e-01 -2.73950677e-02 -8.49183083e-01 1.29414737e+00 1.29431933e-01 1.31118432e-01 -1.54881656e+00 2.33170018e-01 5.77930622e-02 3.41044068e-01 1.95456054e-02 1.10990655e+00 -3.45138520e-01 -5.36238372e-01 -6.37298748e-02 -2.84837097e-01 5.14104247e-01 3.27896327e-01 6.25778586e-02 -7.78360903e-01 -4.57493335e-01 1.48335174e-01 2.38191541e-02 7.62862444e-01 2.99764335e-01 1.31305397e+00 -2.27286890e-01 -4.20169264e-01 7.41819203e-01 1.37754524e+00 -2.58066535e-01 7.13042438e-01 -5.23879714e-02 7.78657436e-01 8.36476445e-01 3.58200997e-01 3.32438231e-01 2.69954830e-01 8.09774041e-01 2.26255115e-02 1.68265015e-01 2.22768844e-03 -1.51915789e-01 3.74146879e-01 9.42030668e-01 4.18392383e-02 6.73282966e-02 -6.22188091e-01 1.84408695e-01 -1.89268553e+00 -1.03421021e+00 -1.99369401e-01 2.66702151e+00 6.02261007e-01 -3.24429810e-01 1.63232207e-01 -2.45807711e-02 8.84183407e-01 -1.02199338e-01 -2.65141338e-01 8.97311196e-02 -7.81850442e-02 3.17229003e-01 4.00273174e-01 4.22353327e-01 -1.20807970e+00 6.05601251e-01 6.26619339e+00 1.01118362e+00 -1.17857158e+00 4.83143590e-02 4.88394856e-01 2.74620295e-01 -4.04947579e-01 6.77103475e-02 -5.77588022e-01 6.15430892e-01 8.10024858e-01 -8.81774798e-02 5.87874174e-01 6.12022102e-01 -1.98753819e-01 -2.28731949e-02 -1.47333527e+00 1.36984706e+00 3.16797107e-01 -1.38267732e+00 -3.12121175e-02 5.39827108e-01 5.84011197e-01 -1.65309444e-01 3.98304403e-01 -1.72055542e-01 -2.12572459e-02 -9.45089281e-01 6.47137523e-01 9.94146705e-01 9.55489099e-01 -5.17959893e-01 3.19179982e-01 4.25247513e-02 -1.17066002e+00 3.18758696e-01 -6.13917112e-01 2.77473837e-01 -1.26088873e-01 5.76002538e-01 -2.37632558e-01 4.71920937e-01 6.03268981e-01 1.05997121e+00 -4.97791260e-01 1.00569642e+00 2.74909258e-01 3.81466895e-01 -4.26917553e-01 1.63718238e-01 -4.49530818e-02 -1.06304133e+00 5.90215564e-01 9.09811318e-01 3.94107968e-01 -3.83322239e-02 1.13693833e-01 1.10927474e+00 4.84902831e-03 1.33340344e-01 -6.73034668e-01 -2.57716149e-01 1.77922919e-01 1.27968693e+00 -6.90086782e-01 -1.60270974e-01 -4.57485348e-01 1.15470803e+00 2.90269852e-01 3.47110480e-01 -4.52293366e-01 -2.78192520e-01 1.08934081e+00 1.61373496e-01 2.13546112e-01 -5.73025346e-01 2.49883875e-01 -1.39244628e+00 2.21701071e-01 -3.92456561e-01 1.85984343e-01 -5.80012441e-01 -1.93599141e+00 9.02723745e-02 2.42679253e-01 -1.22455812e+00 -1.42421141e-01 -9.76558447e-01 -5.98132730e-01 7.78225720e-01 -1.41831934e+00 -9.09791172e-01 -1.71381027e-01 9.05686796e-01 -1.61847427e-01 -1.80200487e-01 9.77356613e-01 3.83903742e-01 -3.23665082e-01 6.16988838e-01 6.55986786e-01 3.23444992e-01 5.46694756e-01 -1.15334404e+00 3.17112625e-01 3.94816935e-01 4.14535880e-01 7.55535960e-01 2.08403587e-01 -3.15773576e-01 -1.74528754e+00 -7.25768328e-01 4.97625858e-01 -5.64457774e-01 1.02163994e+00 -4.76733088e-01 -1.09414756e+00 3.60677242e-01 -1.97879612e-01 2.50275701e-01 8.50106537e-01 -4.07922752e-02 -5.60895622e-01 -9.13590714e-02 -1.15252864e+00 4.35236514e-01 1.13038671e+00 -9.28836584e-01 -1.80864781e-01 4.92471308e-01 1.67157918e-01 4.00835089e-02 -1.09985411e+00 2.37710088e-01 6.33459926e-01 -9.44486022e-01 1.01690221e+00 -6.96818113e-01 1.86298303e-02 -1.40419275e-01 -3.81778985e-01 -1.06311417e+00 -3.80749047e-01 -5.54730177e-01 5.54913580e-02 1.16941726e+00 -2.32003495e-01 -6.17121220e-01 7.65443146e-01 4.92055118e-01 2.14560181e-01 -5.14527678e-01 -1.13291764e+00 -1.06099021e+00 1.19643852e-01 -3.20074975e-01 3.83142531e-01 1.03277731e+00 -5.45510799e-02 -1.51646331e-01 -6.27204552e-02 2.13104021e-02 9.78850782e-01 6.88794404e-02 5.53308189e-01 -1.48945141e+00 -3.46796125e-01 -6.00289464e-01 -1.10262036e+00 -7.71031380e-01 6.57545924e-01 -1.06571984e+00 -2.22027570e-01 -7.75535703e-01 3.44291955e-01 -7.23469913e-01 -1.85412899e-01 2.16222540e-01 2.82233924e-01 3.00328255e-01 6.35128021e-02 4.97647673e-01 -5.08625388e-01 7.10588038e-01 1.11372232e+00 3.33425403e-02 3.42766866e-02 -6.53319508e-02 -9.58568975e-02 7.18981385e-01 4.38093305e-01 -3.57194036e-01 -3.04814339e-01 -2.06723958e-01 -6.54352531e-02 2.96594258e-02 7.49113381e-01 -8.83734822e-01 8.40281919e-02 -1.50257334e-01 5.02850600e-02 -1.85438227e-02 3.67391586e-01 -7.26289749e-01 2.18628094e-01 1.34276718e-01 -1.42639682e-01 -1.85466409e-01 -3.26675400e-02 7.18347788e-01 -2.06310377e-01 -3.47798049e-01 8.96711111e-01 2.63041586e-01 -4.49046195e-01 6.91227615e-01 -2.01670945e-01 4.63075787e-02 1.02549636e+00 -3.32408875e-01 -1.65170841e-02 -2.03704029e-01 -5.55978358e-01 -1.51283637e-01 7.99950063e-01 4.99645442e-01 7.47285306e-01 -1.72578061e+00 -5.54422081e-01 6.36235297e-01 3.93726081e-01 -5.84692657e-01 1.36094913e-01 1.02633476e+00 -3.47098678e-01 1.06442578e-01 -1.61376074e-01 -9.73986030e-01 -1.32360864e+00 3.83374006e-01 5.73280990e-01 2.73315609e-01 -5.25436699e-01 6.31021619e-01 4.23371077e-01 -4.63823408e-01 -4.77711931e-02 -7.08937645e-02 8.38619843e-02 -2.22425740e-02 4.23300534e-01 3.76997024e-01 -1.69440396e-02 -8.95011604e-01 -4.62254018e-01 1.07606101e+00 1.16724983e-01 -2.21051380e-01 1.04334092e+00 -7.17296004e-02 -3.86081159e-01 4.36316758e-01 1.82426894e+00 -2.93676674e-01 -1.22220576e+00 -3.54563475e-01 1.84262455e-01 -8.19098830e-01 -6.88959882e-02 -3.27661410e-02 -8.95222783e-01 9.42427099e-01 9.13147867e-01 9.66271311e-02 8.61214817e-01 2.32678086e-01 4.21689868e-01 2.38438934e-01 5.39928079e-01 -8.20995450e-01 2.55772501e-01 2.78167188e-01 6.17450953e-01 -1.14282036e+00 -3.60952646e-01 -6.14848197e-01 -2.08888903e-01 1.40575337e+00 -7.39783840e-03 -4.67910051e-01 1.04358053e+00 -2.83334441e-02 -1.19575731e-01 -2.66174376e-01 -1.69705376e-01 -6.02855980e-02 6.38898730e-01 6.62188172e-01 3.45536649e-01 1.36070654e-01 -1.20133482e-01 -6.62626419e-03 1.23350660e-03 -3.58025581e-02 2.62004226e-01 7.13165045e-01 -2.58377075e-01 -1.42610943e+00 -4.17022943e-01 4.40938652e-01 -2.58137286e-01 3.66025530e-02 -2.11859625e-02 5.55418432e-01 -2.94388443e-01 6.88676178e-01 5.28263927e-01 -2.57812947e-01 3.52335423e-02 9.70389470e-02 1.12773764e+00 -4.09406930e-01 1.48696974e-01 -1.94297954e-01 -5.15701473e-01 -7.00294554e-01 -6.71117842e-01 -8.03455114e-01 -1.04896200e+00 -4.75440472e-01 -3.39600652e-01 1.74930751e-01 8.14327598e-01 9.36881661e-01 3.14054102e-01 -2.73730129e-01 8.35609257e-01 -7.69177914e-01 -9.41703916e-01 -5.88959932e-01 -1.08065271e+00 8.32129359e-01 1.15663320e-01 -9.02358532e-01 -3.78748953e-01 1.65063396e-01]

[8.015192985534668, 3.7190709114074707]

0e09a01a-977b-4e16-b9f4-c33c729320a6

hymer-a-hybrid-machine-learning-framework-for

1909.08074

https://arxiv.org/abs/1909.08074v3

https://arxiv.org/pdf/1909.08074v3.pdf

MER-SDN: Machine Learning Framework for Traffic Aware Energy Efficient Routing in SDN

Software Defined Networking (SDN) achieves programmability of a network through separation of the control and data planes. It enables flexibility in network management and control. Energy efficiency is one of the challenging global problems which has both economic and environmental impact. A massive amount of information is generated in the controller of an SDN based network. Machine learning gives the ability to computers to progressively learn from data without having to write specific instructions. In this work, we propose MER-SDN: a machine learning framework for traffic-aware energy efficient routing in SDN. Feature extraction, training, and testing are the three main stages of the learning machine. Experiments are conducted on Mininet and POX controller using real-world network topology and dynamic traffic traces from SNDlib. Results show that our approach achieves more than 65\% feature size reduction, more than 70% accuracy in parameter prediction of an energy efficient heuristics algorithm, also our prediction refine heuristics converges the predicted value to the optimal parameters values with up to 25X speedup as compared to the brute force method.

['Oznur Ozkasap', 'Beakal Gizachew Assefa']

2019-08-27

['traffic-classification', 'parameter-prediction']

['miscellaneous', 'miscellaneous']

[ 3.46536525e-02 -1.60762891e-01 -6.91875637e-01 -4.02986795e-01 2.32466772e-01 -4.76343215e-01 8.88931751e-02 -2.72643715e-01 -9.57992300e-03 1.16761267e+00 -4.77085441e-01 -6.11357272e-01 -5.82591474e-01 -7.82183290e-01 -1.05395876e-01 -5.45996666e-01 -4.53635931e-01 9.18807626e-01 4.69383389e-01 1.57291576e-01 3.90271395e-01 1.00436711e+00 -1.43370223e+00 -1.58354312e-01 3.94604594e-01 1.42418730e+00 1.56462699e-01 8.50644410e-01 -3.46608073e-01 8.80895913e-01 -3.50488424e-01 2.89566427e-01 6.97468460e-01 -2.24697679e-01 -9.92189527e-01 -6.73284580e-04 -1.77204937e-01 -3.23254094e-02 -2.73243666e-01 7.09488690e-01 2.23073468e-01 4.78230231e-02 1.42162472e-01 -1.79337621e+00 3.79993558e-01 6.89559162e-01 -4.08687562e-01 6.07682824e-01 -2.93389231e-01 3.51887524e-01 1.07391381e+00 -4.94400896e-02 7.51040518e-01 9.78178263e-01 1.87394723e-01 4.00744408e-01 -1.37451005e+00 -1.05427098e+00 -2.63909519e-01 5.43251872e-01 -1.08074319e+00 -8.97696793e-01 7.30237365e-01 6.03480078e-03 1.36142194e+00 2.33756870e-01 8.09590638e-01 3.49576265e-01 3.74388754e-01 -4.01795022e-02 6.27086937e-01 -3.58343810e-01 5.89907467e-01 1.20518468e-01 -9.15316716e-02 1.00100493e+00 3.88495833e-01 4.23972249e-01 -1.70823678e-01 -1.35183007e-01 6.58329725e-01 -5.14178462e-02 1.26975998e-01 -4.61422473e-01 -9.48211372e-01 6.14885807e-01 4.00926232e-01 6.25996664e-02 -5.39625585e-01 5.59624851e-01 6.62307382e-01 5.10741115e-01 -1.99718952e-01 3.96060437e-01 -9.97408867e-01 -6.08416080e-01 -1.09945655e+00 -2.22702011e-01 1.16498959e+00 1.12623441e+00 7.74421513e-01 3.29645306e-01 4.75753665e-01 4.65275794e-01 4.81629342e-01 5.63138664e-01 1.62647367e-01 -1.35497308e+00 5.57363629e-02 6.82999313e-01 -2.93972433e-01 -5.60662508e-01 -6.39301002e-01 -2.76530236e-01 -9.51414704e-01 4.83595669e-01 5.19610196e-02 -5.99368334e-01 -6.99476779e-01 1.58991730e+00 2.93264866e-01 5.29516339e-01 -7.27250129e-02 6.42434835e-01 -1.82270762e-02 1.16782236e+00 1.74526423e-01 -6.66010201e-01 1.18590105e+00 -1.12963569e+00 -5.32221377e-01 -6.69828476e-03 1.60777539e-01 -6.95740879e-01 3.69194716e-01 3.31946790e-01 -9.73563910e-01 -3.71641695e-01 -1.22559822e+00 4.40285772e-01 -4.53599095e-01 -2.95945466e-01 7.89357960e-01 7.70869792e-01 -9.48053360e-01 8.00953627e-01 -8.88671279e-01 -8.11902106e-01 6.33864403e-01 1.00224817e+00 5.11172116e-02 2.45753288e-01 -5.74442923e-01 6.93650365e-01 6.60843015e-01 -6.18667603e-01 -7.78206527e-01 -1.07099664e+00 -1.94524094e-01 3.27856213e-01 6.14428401e-01 -5.38229287e-01 1.06270885e+00 -7.32292473e-01 -1.62689984e+00 4.09135580e-01 1.33846179e-01 -5.30332386e-01 -2.68198475e-02 5.01048267e-01 -1.06221330e+00 1.83569621e-02 -2.96193749e-01 5.65895796e-01 6.62856221e-01 -7.93662369e-01 -9.49374795e-01 -6.85930401e-02 -2.36468539e-01 -5.07956684e-01 8.97891000e-02 -1.32574523e-02 1.67360798e-01 -9.60907340e-02 -7.85774812e-02 -9.41913068e-01 -4.87652987e-01 7.26916566e-02 -3.87230426e-01 -1.79014996e-01 1.54450738e+00 6.86651766e-02 1.33122313e+00 -1.64655769e+00 -1.53845742e-01 8.53817821e-01 4.91912991e-01 3.08732927e-01 1.77411720e-01 2.57236183e-01 -2.70509953e-03 3.01263213e-01 3.53795499e-01 5.25563300e-01 8.35691839e-02 5.31280279e-01 5.41224051e-03 9.55276564e-02 -6.06701784e-02 2.65010834e-01 -5.81180036e-01 -7.51685083e-01 4.99435008e-01 1.23366937e-01 -5.63652515e-01 -2.48513557e-02 -4.42734659e-01 2.52244949e-01 -5.21530807e-01 6.86514318e-01 5.20974934e-01 -5.94007671e-01 7.55261362e-01 -3.79101813e-01 -2.25068673e-01 3.15279245e-01 -1.21857584e+00 1.23799539e+00 -4.98295814e-01 6.78229630e-01 -1.52053684e-02 -1.12537348e+00 8.21441770e-01 2.94933707e-01 1.04273927e+00 -7.82760859e-01 4.21014518e-01 2.75770068e-01 2.69051880e-01 -3.43261302e-01 -2.40617588e-01 -1.96076646e-01 1.54011041e-01 8.31865907e-01 2.12009981e-01 5.10157108e-01 6.10411584e-01 -2.58814305e-01 1.62751424e+00 -4.67589945e-01 4.59405452e-01 -4.77859676e-01 7.17566669e-01 6.06740452e-02 1.00260806e+00 1.96648866e-01 -6.18228078e-01 -6.52238429e-01 6.07293904e-01 -8.81755710e-01 -1.30653155e+00 -1.15522254e+00 1.94104314e-02 9.04645741e-01 -1.20000325e-01 -1.48316503e-01 -5.11577606e-01 -5.97623527e-01 -1.21152885e-01 6.11225963e-01 6.45426959e-02 -9.39372256e-02 -7.76869953e-01 -4.95411754e-01 2.96540976e-01 1.08446293e-01 6.33175611e-01 -1.01816380e+00 -9.22853410e-01 5.59768438e-01 3.64505678e-01 -1.50523257e+00 -2.36855164e-01 6.02299988e-01 -1.28210878e+00 -1.24269354e+00 8.22099447e-01 -7.95785546e-01 4.15068567e-01 -2.74741173e-01 1.36694705e+00 8.19700286e-02 -8.66602421e-01 -2.67385423e-01 -4.96573858e-02 -1.36113510e-01 -4.27899271e-01 3.67023796e-01 2.87985861e-01 -2.19625443e-01 6.05107129e-01 -1.39007092e+00 -6.53887510e-01 4.27384853e-01 -3.05116028e-01 -4.49176617e-02 7.40302622e-01 1.71109378e-01 7.25242376e-01 6.05327547e-01 6.69792831e-01 -8.66369963e-01 -2.62147337e-02 -7.20849037e-01 -1.16580927e+00 9.06243473e-02 -1.31485748e+00 6.09206378e-01 1.06031322e+00 3.95899191e-02 -5.04077196e-01 3.03293586e-01 3.48832041e-01 -3.47992271e-01 -3.33017893e-02 -3.02907586e-01 -5.50710320e-01 -2.93870479e-01 2.55822688e-01 -1.37014449e-01 3.46687883e-02 -3.23470414e-01 -6.70747533e-02 4.34020877e-01 -4.91929837e-02 -5.08079231e-01 1.01722693e+00 2.32421413e-01 7.87260234e-01 -7.17145026e-01 -1.63941219e-01 -4.26665880e-02 -3.59895170e-01 -4.57935661e-01 5.21462202e-01 -1.07685432e-01 -1.13117468e+00 -5.50040126e-01 -6.93254709e-01 -1.50865182e-01 -3.03292036e-01 4.41206455e-01 -5.88175178e-01 -3.06989104e-01 -4.20832962e-01 -3.93528283e-01 -5.86525619e-01 -1.03130662e+00 3.00616492e-03 4.76387739e-01 -9.74426568e-02 -8.58897150e-01 1.61442786e-01 2.52669811e-01 8.07108760e-01 2.18379900e-01 1.55158961e+00 -9.40689564e-01 -1.20485389e+00 2.32971981e-01 -4.42964733e-01 1.05866663e-01 2.55676240e-01 5.59819639e-01 -4.45793927e-01 -1.39120787e-01 -3.03282589e-01 -2.31566504e-02 1.53469101e-01 4.27862406e-01 1.32868576e+00 -4.56780910e-01 -5.08723855e-01 8.74717832e-01 2.29319835e+00 7.21049249e-01 5.56229413e-01 5.82371280e-02 2.94350535e-01 -6.44651279e-02 2.34241188e-01 8.46973836e-01 -1.64206579e-01 4.62910444e-01 7.20654428e-01 4.52288926e-01 -6.85446039e-02 6.92227930e-02 1.76123038e-01 7.70153880e-01 -1.27619207e-01 -4.36466277e-01 -8.16067100e-01 -1.39955031e-02 -1.38017714e+00 -1.24347353e+00 1.26434341e-01 1.90783393e+00 4.07774121e-01 5.76004505e-01 2.32678339e-01 4.65124995e-01 6.29174769e-01 -7.50562642e-03 -9.30925846e-01 -9.69188690e-01 5.50490975e-01 6.94314659e-01 9.22067463e-01 2.14889348e-01 -6.99638963e-01 8.38813603e-01 5.62925577e+00 8.47176254e-01 -1.20974779e+00 -6.83663264e-02 4.40641761e-01 -2.60549903e-01 3.12562823e-01 6.18316494e-02 -6.46504998e-01 7.64782846e-01 1.86456084e+00 -5.42549491e-01 1.15065646e+00 8.87870610e-01 7.78019965e-01 1.16715714e-01 -1.26367927e+00 1.21727300e+00 -8.09548736e-01 -1.64134097e+00 1.22110853e-02 2.53056049e-01 5.59543371e-01 1.81140512e-01 -4.57723111e-01 3.25179189e-01 6.23056591e-01 -1.01319432e+00 5.74079603e-02 3.23358238e-01 6.90910459e-01 -1.28021097e+00 4.87743050e-01 4.35166694e-02 -1.44144511e+00 -5.04174173e-01 -3.06112528e-01 4.19088900e-02 1.60195887e-01 7.63643086e-01 -9.32037652e-01 1.45883307e-01 4.42019552e-01 4.50816780e-01 -4.33457233e-02 1.04660869e+00 6.06636465e-01 5.86205900e-01 -6.22745991e-01 -1.17991939e-01 1.10123403e-01 -2.95750707e-01 4.85022098e-01 9.51547444e-01 2.11708784e-01 1.27748728e-01 2.71775573e-01 7.87891150e-01 -5.75411856e-01 6.58603609e-02 -3.33746105e-01 -2.65997469e-01 9.48269010e-01 1.44169772e+00 -9.14113343e-01 -1.66314691e-01 -2.61042893e-01 4.11418825e-01 -1.32125750e-01 3.85444239e-02 -1.28075254e+00 -7.87626386e-01 1.19839668e+00 3.40025634e-01 3.52092892e-01 -2.55390197e-01 -4.59332108e-01 -4.13338542e-01 -3.63231242e-01 -5.45193195e-01 2.50385225e-01 -3.41287673e-01 -8.08142245e-01 6.04439437e-01 -3.70917141e-01 -1.16067028e+00 -1.24949895e-01 -5.41615069e-01 -5.87249458e-01 2.64423281e-01 -1.56531775e+00 -4.40132827e-01 -5.15983284e-01 6.45950675e-01 6.43560946e-01 -6.28317654e-01 8.11699748e-01 7.55421937e-01 -9.65118468e-01 2.14207962e-01 1.97619393e-01 2.87270872e-04 2.82431126e-01 -8.67651045e-01 -1.10493988e-01 5.55388033e-01 -3.05093110e-01 -6.86452538e-02 7.47309148e-01 -2.82481223e-01 -1.76774657e+00 -1.22312367e+00 6.78361893e-01 3.01611006e-01 7.27174699e-01 8.57347995e-02 1.09901413e-01 5.82898378e-01 1.85019329e-01 3.18331361e-01 8.68985653e-01 -3.34179014e-01 1.39750123e-01 -1.04211485e+00 -1.70513928e+00 3.22408050e-01 1.02015626e+00 1.34049505e-01 1.40656099e-01 3.39368135e-01 4.62394118e-01 2.39114568e-01 -9.50339854e-01 1.81686327e-01 4.74403888e-01 -1.02361941e+00 6.77319884e-01 -9.81090546e-01 -5.31338528e-02 -3.74817312e-01 -3.51005316e-01 -8.11524749e-01 -2.56146669e-01 -8.61172676e-01 -6.13515496e-01 1.19265616e+00 5.31862736e-01 -3.27298224e-01 1.11159396e+00 3.28665018e-01 9.61033106e-02 -7.04395950e-01 -9.44160402e-01 -9.45654094e-01 -6.91899955e-01 -3.09655428e-01 9.42756593e-01 7.75059998e-01 -8.13660324e-02 5.84631979e-01 1.70782372e-01 2.87481487e-01 1.10476720e+00 1.54134765e-01 5.13835669e-01 -1.84073162e+00 2.71242168e-02 -6.00618541e-01 -7.60998726e-01 -6.67503834e-01 2.36652374e-01 -8.80353749e-01 -5.87184191e-01 -1.26151276e+00 -7.15988949e-02 -8.60137820e-01 -6.06291234e-01 3.33559066e-01 1.05695939e+00 -2.79460609e-01 -9.22337174e-02 -5.06689325e-02 -8.50821316e-01 -3.17854285e-02 8.49496663e-01 1.42158955e-01 -2.76833326e-01 8.70807543e-02 -2.23340526e-01 4.85632181e-01 1.48599267e+00 -6.40934348e-01 -9.16434228e-01 2.36383080e-02 -1.07136838e-01 2.91060358e-01 7.73833022e-02 -1.47344768e+00 2.74962425e-01 -6.84366286e-01 3.58819157e-01 -2.23250985e-01 -1.03220776e-01 -1.69884515e+00 6.76123381e-01 8.83766413e-01 -3.66659686e-02 1.60142288e-01 -4.41186540e-02 5.74558735e-01 2.22480446e-01 2.12403595e-01 1.40659761e+00 8.18952024e-02 -1.01897407e+00 6.86083317e-01 -5.56716084e-01 1.84693381e-01 1.55434513e+00 -3.43492538e-01 -2.48546153e-01 1.93216711e-01 -7.44342148e-01 3.71047497e-01 1.87491655e-01 2.01565415e-01 4.38782871e-01 -1.21027207e+00 -2.05997810e-01 3.90921712e-01 -4.38504666e-01 -6.50339305e-01 -1.04222109e-03 6.41596556e-01 -1.22004294e+00 8.93744648e-01 -8.39140654e-01 -4.18333948e-01 -1.19009233e+00 5.58310688e-01 7.38217711e-01 -5.33306003e-01 -2.95897990e-01 3.71349096e-01 -1.02700269e+00 3.01274341e-02 3.32288802e-01 -2.71199703e-01 1.84455842e-01 -5.18946111e-01 2.18343794e-01 1.20364594e+00 -1.68906860e-02 -4.33017313e-02 -7.87352800e-01 3.73200893e-01 1.03348263e-01 3.41425002e-01 1.57895815e+00 -1.91929311e-01 -5.29276371e-01 -1.79913878e-01 1.26144278e+00 -4.19851124e-01 -8.59615445e-01 -7.08556734e-03 4.58853006e-01 -3.21139336e-01 5.20004690e-01 -1.00612068e+00 -1.73156500e+00 4.53301013e-01 1.08839798e+00 2.45356560e-01 1.45278430e+00 -4.10711527e-01 1.18341506e+00 7.12582648e-01 5.37842870e-01 -1.37046909e+00 -1.82021976e-01 3.26553762e-01 -2.41596669e-01 -1.11258101e+00 -3.15613672e-02 -3.56069922e-01 3.51102813e-03 1.57272112e+00 8.65277946e-01 2.01137047e-02 1.24906445e+00 5.64159751e-01 -3.85596544e-01 -2.97956288e-01 -1.38398361e+00 9.99778062e-02 -5.89918196e-01 4.33074921e-01 -1.60342842e-01 1.94383383e-01 8.79973620e-02 1.67735681e-01 -1.48927346e-01 2.76925027e-01 3.89328122e-01 9.02868211e-01 -1.06097281e+00 -1.39206755e+00 1.63186803e-01 7.86799014e-01 -3.69323432e-01 1.71655506e-01 2.82783210e-01 7.01208770e-01 2.80556083e-01 1.06928611e+00 2.32353866e-01 -6.64032578e-01 7.35797808e-02 2.26080015e-01 2.68091798e-01 -3.00096184e-01 -2.50629038e-01 -3.79813075e-01 2.41441593e-01 -1.03897405e+00 -2.97778808e-02 -2.79510975e-01 -1.88087714e+00 -1.18205833e+00 1.86192170e-01 3.46796751e-01 1.09901154e+00 8.13536167e-01 6.45844936e-01 9.16136026e-01 1.25855982e+00 -4.27753508e-01 -2.33074069e-01 -3.47282887e-01 -7.34490037e-01 -4.30191040e-01 4.08460259e-01 -5.71098566e-01 -3.83498132e-01 1.65823862e-01]

[5.83573579788208, 1.757981538772583]

da664aea-46dd-429a-9012-a87c3d61825b

online-visual-multi-object-tracking-via

1611.06011

http://arxiv.org/abs/1611.06011v2

http://arxiv.org/pdf/1611.06011v2.pdf

Online Visual Multi-Object Tracking via Labeled Random Finite Set Filtering

This paper proposes an online visual multi-object tracking algorithm using a top-down Bayesian formulation that seamlessly integrates state estimation, track management, clutter rejection, occlusion and mis-detection handling into a single recursion. This is achieved by modeling the multi-object state as labeled random finite set and using the Bayes recursion to propagate the multi-object filtering density forward in time. The proposed filter updates tracks with detections but switches to image data when mis-detection occurs, thereby exploiting the efficiency of detection data and the accuracy of image data. Furthermore the labeled random finite set framework enables the incorporation of prior knowledge that mis-detections of long tracks which occur in the middle of the scene are likely to be due to occlusions. Such prior knowledge can be exploited to improve occlusion handling, especially long occlusions that can lead to premature track termination in on-line multi-object tracking. Tracking performance are compared to state-of-the-art algorithms on well-known benchmark video datasets.

['Ba-Tuong Vo', 'Ba-Ngu Vo', 'Du Yong Kim']

2016-11-18

['occlusion-handling']

['computer-vision']

[ 5.34387343e-02 -4.16613787e-01 -8.45916495e-02 4.50254045e-02 -4.41255122e-01 -6.72269225e-01 5.16782820e-01 2.85216391e-01 -5.32521605e-01 7.35725880e-01 -2.39317253e-01 -1.18444659e-01 -1.93166077e-01 -4.56163973e-01 -8.02503169e-01 -5.80433249e-01 -1.00037515e-01 7.54577041e-01 9.20866191e-01 4.48396325e-01 -2.18014289e-02 9.18084800e-01 -1.80480063e+00 -5.66315129e-02 4.31536674e-01 8.90622854e-01 3.63684058e-01 9.02255833e-01 -1.56483620e-01 5.55453300e-01 -5.82852066e-01 -1.64461151e-01 3.34453136e-01 2.83768952e-01 -6.23700721e-03 4.16634738e-01 7.77829349e-01 -4.40730900e-01 -5.62245809e-02 1.17716444e+00 2.67181635e-01 3.56948636e-02 5.05900145e-01 -1.35987115e+00 2.29828507e-01 4.59568910e-02 -6.78934932e-01 2.83302128e-01 -1.80492774e-02 1.89223126e-01 5.03120363e-01 -8.50892842e-01 6.67816699e-01 1.35620701e+00 8.15861881e-01 1.17394276e-01 -1.24587333e+00 -6.49532974e-01 4.63491678e-01 1.20868005e-01 -1.33560383e+00 -4.10699069e-01 2.13157728e-01 -9.06330347e-01 4.56701756e-01 2.96259969e-01 6.64895296e-01 5.42914510e-01 2.84700602e-01 7.41763532e-01 7.94594705e-01 -4.23742473e-01 7.10684508e-02 1.37770399e-01 3.45804036e-01 6.52467668e-01 8.43635380e-01 5.37694454e-01 -5.90065420e-01 -1.75118834e-01 6.80236995e-01 8.68276432e-02 2.91323680e-02 -8.20962727e-01 -1.02349842e+00 5.14549732e-01 9.95012149e-02 -3.30539537e-04 -5.66225111e-01 1.72022909e-01 4.42543536e-01 9.33734849e-02 2.05170467e-01 -3.39463830e-01 -2.73616880e-01 1.70379430e-01 -1.28097463e+00 3.54854107e-01 4.96919245e-01 1.32332599e+00 6.25454724e-01 1.59167394e-01 -6.17695391e-01 2.69580245e-01 7.69225717e-01 8.42245102e-01 -2.90870428e-01 -6.46153808e-01 8.39923322e-02 3.84955227e-01 7.33385444e-01 -5.39030075e-01 -3.50882202e-01 -9.02765274e-01 -4.63165432e-01 6.95405424e-01 6.04186416e-01 -7.73900524e-02 -1.01015401e+00 1.55226970e+00 8.52747083e-01 3.98157418e-01 -2.25373134e-01 8.26343954e-01 3.44676018e-01 3.07294011e-01 1.71353742e-01 -7.06372321e-01 1.50438523e+00 -7.17060089e-01 -1.02080345e+00 -1.37354910e-01 -7.13333394e-03 -1.07854259e+00 4.15279064e-03 4.44755316e-01 -9.45156515e-01 -9.79776740e-01 -1.04081249e+00 5.57291865e-01 -2.29418218e-01 3.05606604e-01 3.85827303e-01 7.61905253e-01 -5.91145873e-01 4.37457711e-01 -9.97704387e-01 -3.11727792e-01 4.83519435e-01 4.37213093e-01 1.63727924e-01 -6.66806996e-02 -5.53403914e-01 1.03313577e+00 5.54345250e-01 1.70081824e-01 -1.09679544e+00 -7.85779774e-01 -4.91503417e-01 -2.36895680e-01 8.16735744e-01 -8.06555510e-01 9.98186767e-01 -5.91434479e-01 -1.05772328e+00 4.85829920e-01 -3.48814517e-01 -5.47856152e-01 9.40486908e-01 -7.41084754e-01 -4.47124958e-01 -2.13501707e-01 1.17765494e-01 3.06329548e-01 1.27636027e+00 -1.30751252e+00 -1.11882269e+00 -3.45228255e-01 -3.44525307e-01 7.25988075e-02 2.36138046e-01 1.39878556e-01 -8.39693666e-01 -6.25034571e-01 1.31753102e-01 -9.93626773e-01 -2.43628174e-01 2.78216392e-01 -3.45994502e-01 6.35888800e-02 1.05357063e+00 -2.57550895e-01 1.21539557e+00 -1.91745448e+00 2.06146426e-02 2.99903572e-01 3.99640426e-02 3.75921607e-01 1.19947769e-01 3.11401159e-01 3.14813644e-01 -6.02403998e-01 6.35717690e-01 -5.27623415e-01 -1.37540121e-02 2.84616679e-01 -2.81615674e-01 9.24026310e-01 -2.43910849e-02 3.75844806e-01 -7.47691035e-01 -6.09830320e-01 7.97920704e-01 3.79090279e-01 -2.44884297e-01 1.85713097e-01 -4.62103665e-01 6.14178300e-01 -4.31677103e-01 5.84454119e-01 1.06752503e+00 -7.42758587e-02 -4.14539464e-02 -4.86289561e-01 -7.32162416e-01 -4.06179756e-01 -1.90421665e+00 1.32011330e+00 9.36510339e-02 2.84594625e-01 2.24548340e-01 -3.42848659e-01 4.87721175e-01 2.04595387e-01 6.30839586e-01 -1.97821617e-01 2.11568102e-01 -9.18251052e-02 -1.78371027e-01 -1.04657926e-01 7.94603705e-01 1.63257852e-01 3.25438559e-01 1.20229926e-02 3.66251846e-03 4.81106788e-01 4.76953954e-01 1.38253957e-01 8.79377425e-01 4.53536332e-01 9.15855393e-02 -3.55131745e-01 5.64286113e-01 1.77531987e-01 8.93797815e-01 1.26708138e+00 -2.95490623e-01 1.77645341e-01 -2.74795234e-01 -3.32704425e-01 -8.37953925e-01 -1.22713256e+00 -2.68609256e-01 1.02012444e+00 3.76876354e-01 -2.55945921e-01 -2.40567401e-01 -3.36863250e-01 2.05058917e-01 2.53397971e-01 -4.05595809e-01 2.01521620e-01 -4.36902881e-01 -6.22315824e-01 5.16258366e-02 4.64324653e-01 -1.38821779e-02 -6.84250712e-01 -1.26177227e+00 6.96764171e-01 2.97636062e-01 -1.16949713e+00 -3.46380115e-01 3.40082377e-01 -8.73197913e-01 -1.20258892e+00 -5.54177165e-01 -1.75467938e-01 5.47015429e-01 9.36814696e-02 9.48363125e-01 1.03488110e-01 -6.64943814e-01 4.91221070e-01 -1.26604557e-01 -6.70142829e-01 -4.66364205e-01 -5.40401936e-01 2.41178170e-01 2.10509405e-01 1.40073657e-01 6.48548231e-02 -4.63721961e-01 4.76739705e-01 -4.61055905e-01 1.02587327e-01 3.71539831e-01 7.06568003e-01 8.97094786e-01 1.77016586e-01 1.66921049e-01 -6.25510752e-01 -1.93372041e-01 -7.48286769e-02 -1.64691544e+00 4.24596846e-01 -3.14741641e-01 -1.82447899e-02 -1.17362207e-02 -8.17183137e-01 -1.00399983e+00 4.84643042e-01 3.01435113e-01 -1.00580311e+00 -1.48516253e-01 -3.20507623e-02 5.42207528e-03 -3.44240904e-01 1.86322317e-01 -4.04621884e-02 -1.51135237e-03 -5.71640491e-01 3.20970029e-01 8.94480869e-02 6.94028497e-01 -3.73502940e-01 8.85845602e-01 7.87440479e-01 4.20629084e-01 -7.80334175e-01 -7.94040024e-01 -1.05849636e+00 -7.12847710e-01 -8.99331033e-01 8.16870511e-01 -1.15702856e+00 -8.70998502e-01 5.09004593e-01 -1.13128805e+00 2.70189613e-01 -1.61357999e-01 6.61123812e-01 -3.99186313e-01 3.05075586e-01 -3.49505186e-01 -1.51175618e+00 -1.10207498e-01 -1.10958648e+00 1.05370641e+00 3.98977786e-01 6.13098815e-02 -7.63550401e-01 -9.76111963e-02 -2.28685196e-02 3.48138183e-01 1.62559792e-01 1.50977746e-01 -4.65340465e-01 -1.30743110e+00 -2.71990687e-01 -1.32252807e-02 -2.03506798e-01 -2.54003495e-01 1.68269128e-01 -8.49471033e-01 -7.59009898e-01 -2.80437142e-01 1.47614479e-01 6.95611954e-01 8.03201914e-01 3.74297827e-01 2.17820659e-01 -8.98136258e-01 2.28137121e-01 1.75014651e+00 1.66748360e-01 2.13304564e-01 3.30639511e-01 5.26270211e-01 2.94750512e-01 1.31172633e+00 8.90918195e-01 8.86805877e-02 1.04423583e+00 6.74901366e-01 1.49967219e-03 -2.67119646e-01 1.43348724e-01 1.30709767e-01 1.37294725e-01 1.93531111e-01 -2.73377359e-01 -5.88028014e-01 6.15631163e-01 -2.16388798e+00 -1.04012120e+00 -6.69720471e-01 2.62701106e+00 2.57085413e-01 4.85281050e-01 3.87373954e-01 -2.33367402e-02 1.00678170e+00 -1.97417006e-01 -4.81518745e-01 4.46454704e-01 8.65919366e-02 -2.38145903e-01 1.04665446e+00 5.21200180e-01 -1.41083634e+00 7.73401737e-01 6.06874466e+00 6.53560579e-01 -6.63748205e-01 2.66792595e-01 -2.63478518e-01 -1.35233387e-01 4.18287784e-01 1.10702828e-01 -1.68134725e+00 4.23802674e-01 6.21829152e-01 1.97020441e-01 -7.80748725e-02 5.49035251e-01 2.22510666e-01 -5.63480675e-01 -1.07340157e+00 9.94301975e-01 -2.47593135e-01 -1.25650585e+00 -2.20009252e-01 1.70621648e-01 4.53774363e-01 5.95842935e-02 -1.44785270e-01 7.20899180e-02 5.55960894e-01 -2.41213590e-01 1.19758058e+00 8.75453830e-01 2.03002110e-01 -5.59663951e-01 4.71594721e-01 4.46543008e-01 -1.69248378e+00 -1.37729183e-01 -1.94698900e-01 2.50321358e-01 7.10694969e-01 6.77792907e-01 -5.76973796e-01 8.16966236e-01 6.35722816e-01 4.20003504e-01 -5.02025187e-01 1.75013757e+00 2.62749732e-01 1.41819850e-01 -7.05437362e-01 3.38226050e-01 -9.10260230e-02 3.47857215e-02 1.08014715e+00 1.16193473e+00 8.93910602e-02 -3.06459755e-01 8.34544301e-01 6.03512704e-01 6.24393463e-01 -3.50984067e-01 -3.22335362e-01 3.53706986e-01 5.79535484e-01 1.10010922e+00 -9.07049358e-01 -5.33649981e-01 -5.99651337e-01 5.65479159e-01 -5.26715890e-02 2.13135943e-01 -1.04612482e+00 2.91663706e-01 5.79662740e-01 3.65561992e-02 9.12690043e-01 -8.89953449e-02 8.26641619e-02 -9.18389738e-01 -1.70144364e-01 -3.63666356e-01 6.12296879e-01 -5.07180870e-01 -1.07434642e+00 3.46264094e-01 2.08944827e-01 -1.52786303e+00 -1.35534778e-01 -4.86473978e-01 -2.03674242e-01 8.30370426e-01 -1.45141292e+00 -1.30045962e+00 -1.56023502e-01 4.35858011e-01 5.40462732e-01 -8.41304287e-02 3.87655437e-01 7.39627421e-01 -3.75568092e-01 2.54877239e-01 4.11240906e-02 -1.79010406e-01 6.12664998e-01 -8.72881174e-01 4.27406132e-02 1.25562012e+00 -8.17232009e-04 4.63547558e-01 1.09189963e+00 -1.12288511e+00 -1.34943593e+00 -1.17404389e+00 3.85138035e-01 -3.80459309e-01 6.08231664e-01 -2.27074102e-01 -8.05067480e-01 6.47664130e-01 -1.14349291e-01 2.86557615e-01 3.37652773e-01 8.11462477e-02 1.48787331e-02 4.74040164e-03 -9.77185726e-01 2.92174608e-01 7.50835121e-01 -2.24528294e-02 -3.73986185e-01 2.69732326e-01 8.41803998e-02 -5.93936205e-01 -7.56144822e-01 6.18417859e-01 7.45603561e-01 -6.70811713e-01 1.16363156e+00 -3.26497734e-01 -8.91472518e-01 -1.05446947e+00 1.46761276e-02 -7.79136360e-01 -4.85036194e-01 -5.55067241e-01 -5.27092934e-01 1.31170654e+00 -1.18530527e-01 -1.39859527e-01 6.73848569e-01 2.15690181e-01 -3.21540721e-02 -3.25067388e-03 -1.08173132e+00 -1.02548218e+00 -7.07626581e-01 -4.13264155e-01 4.43866476e-02 3.74262542e-01 -8.20415616e-01 5.27469534e-03 -7.89164722e-01 8.18513036e-01 1.54003942e+00 3.25775146e-01 9.44916546e-01 -1.69190347e+00 -3.32526416e-01 -9.05375183e-02 -4.78739321e-01 -1.02996659e+00 -1.88590616e-01 -2.42445335e-01 3.39310080e-01 -1.22936630e+00 2.88237602e-01 -5.70056379e-01 -2.96415567e-01 6.68691620e-02 -2.19867319e-01 1.79272264e-01 6.01475835e-01 2.03371808e-01 -1.14703023e+00 1.64842531e-01 9.02275264e-01 1.11423843e-01 -1.04306266e-02 4.35433954e-01 1.79532528e-01 7.74286091e-01 1.64016917e-01 -8.81144881e-01 1.75862070e-02 -2.52064884e-01 -1.52701661e-01 3.08135062e-01 4.92020816e-01 -1.37760532e+00 7.41258025e-01 -2.40832428e-03 5.48114836e-01 -1.55837905e+00 5.94777584e-01 -1.37092805e+00 8.82679105e-01 6.86690569e-01 1.29749030e-01 -1.20168272e-02 5.80028951e-01 1.02339041e+00 1.91683486e-01 -3.22647452e-01 1.00104034e+00 2.78005842e-02 -7.99979210e-01 3.06831002e-01 -5.14890790e-01 -2.63990909e-01 1.31589770e+00 -4.69103962e-01 8.24762136e-02 9.00942981e-02 -1.21240628e+00 5.70410252e-01 4.29975241e-01 5.19733548e-01 1.64397374e-01 -1.09591663e+00 -4.75475162e-01 2.29935542e-01 2.70513054e-02 -2.35122681e-01 3.79594415e-01 9.28847969e-01 -9.41492543e-02 2.54064560e-01 -1.65771917e-01 -1.21449065e+00 -1.92763460e+00 7.00813413e-01 1.88146219e-01 -4.07531768e-01 -6.62170589e-01 5.73088467e-01 3.37412283e-02 3.59024614e-01 7.19921112e-01 -1.13304488e-01 -1.59099847e-01 1.22288726e-01 8.33928168e-01 5.39343953e-01 -1.42779872e-01 -6.93704665e-01 -5.71703374e-01 8.40133369e-01 -2.47030407e-01 1.12892529e-02 8.63503873e-01 -3.93395543e-01 3.44260782e-01 5.59075594e-01 2.78642714e-01 4.12963070e-02 -1.68370962e+00 -3.98284376e-01 2.60026544e-01 -5.94574571e-01 1.77399918e-01 -6.81513429e-01 -8.03953350e-01 3.59031439e-01 1.20763695e+00 -1.13446601e-01 6.22504115e-01 -2.41534822e-02 1.63220614e-01 5.38996682e-02 5.95662057e-01 -1.01459336e+00 -1.73157126e-01 5.77664435e-01 3.53358984e-01 -1.08706367e+00 4.48771298e-01 -5.13943136e-01 -2.77971894e-01 8.87733340e-01 4.78081793e-01 4.41711769e-02 6.74516082e-01 8.08304548e-01 -9.57587063e-02 -1.00663126e-01 -6.70202732e-01 -7.11400688e-01 3.47625196e-01 5.36871195e-01 -1.73575968e-01 -9.78612676e-02 2.33165622e-02 1.18319420e-02 5.92724383e-01 3.43764395e-01 1.90144181e-01 1.22029638e+00 -7.04208434e-01 -9.05218065e-01 -9.50921535e-01 3.47537965e-01 -6.48357511e-01 2.40435213e-01 4.59932745e-01 6.71560049e-01 3.13728809e-01 8.27926397e-01 2.39766613e-01 2.63087988e-01 3.60265911e-01 -1.38125911e-01 7.22564518e-01 -5.27477026e-01 -5.90300441e-01 7.55890608e-01 9.37425718e-02 -4.87744242e-01 -6.73172712e-01 -9.91690516e-01 -1.02833200e+00 1.18216709e-03 -9.47430074e-01 -7.68779740e-02 6.86854184e-01 9.33525145e-01 3.07603180e-01 9.59545135e-01 -1.47814527e-01 -1.00108075e+00 -5.90929806e-01 -8.28294158e-01 -4.71140891e-01 2.58359522e-01 6.28171265e-01 -1.31020391e+00 -2.24828348e-03 1.93738893e-01]

[6.580289363861084, -1.959050178527832]

1b0531a5-151d-43f6-822a-76fffd1fa37f

racai-gec-a-hybrid-approach-to-grammatical

https://aclanthology.org/W14-1705

https://aclanthology.org/W14-1705.pdf

RACAI GEC -- A hybrid approach to Grammatical Error Correction

['Ionut Paul V{\\u{a}}duva', 'Tiberiu Boro{\\textcommabelow{s}}', 'Verginica Barbu Mititelu', 'Stefan Daniel Dumitrescu', 'Adrian Zafiu']

2014-06-01

ws-2014-6

['grammatical-error-detection']

['natural-language-processing']

[-8.63703638e-02 1.71006292e-01 -6.22772932e-01 -4.08054382e-01 -8.41685571e-03 -9.08429027e-01 6.55310392e-01 -6.53472245e-01 -2.85945535e-01 1.06888819e+00 -4.63127941e-02 -1.01159286e+00 -3.91567826e-01 -9.63214397e-01 -4.95059669e-01 -6.31337762e-01 -9.79754329e-01 7.25764990e-01 3.30370307e-01 -6.93831444e-01 7.03166842e-01 7.88774848e-01 -1.68942046e+00 7.18545914e-01 7.04417467e-01 8.52217197e-01 2.49141872e-01 1.14950800e+00 -1.95044339e-01 1.55633950e+00 -7.48382092e-01 -5.46825826e-01 3.13719302e-01 -1.23176083e-01 -7.22945035e-01 -1.01074085e-01 9.28529128e-02 -8.59008506e-02 -2.09758401e-01 9.22211111e-01 5.37373662e-01 4.49454933e-02 1.08379531e+00 -1.42548037e+00 -5.91619551e-01 6.10313773e-01 -4.01565880e-02 1.21627934e-01 1.03678203e+00 -5.39447069e-01 1.19919395e+00 -1.13026452e+00 7.20913768e-01 1.26888943e+00 8.66221786e-01 5.44149756e-01 -1.22286928e+00 -1.94712028e-01 -3.26822817e-01 -9.51717794e-02 -1.46558487e+00 -3.25250506e-01 4.25783843e-02 -2.08119690e-01 1.66093647e+00 1.26596653e+00 1.20609856e+00 1.01401424e+00 1.26658809e+00 8.34431887e-01 1.04267764e+00 -5.13792276e-01 3.35295945e-01 3.66983831e-01 1.54683650e-01 6.33519173e-01 8.40953708e-01 5.26628852e-01 -7.06372619e-01 -9.13127720e-01 9.33553874e-01 -2.94925272e-01 1.71355158e-01 -5.05680561e-01 -9.05919552e-01 6.91228509e-01 1.78732842e-01 3.83959889e-01 -1.39880210e-01 9.89067405e-02 1.26390755e-01 5.30987144e-01 -2.58292928e-02 6.47037446e-01 -9.11868811e-01 -1.33165747e-01 -8.71728659e-01 5.10332465e-01 1.25398111e+00 1.52653182e+00 1.24482810e-01 2.94908643e-01 -9.34252143e-02 3.17179203e-01 8.92314315e-01 1.01808000e+00 4.28362608e-01 -1.36146402e+00 -6.87414408e-02 1.72361732e-01 5.01781464e-01 -8.52631688e-01 -6.33224547e-01 -9.64177120e-03 -8.93263519e-01 4.49267089e-01 3.49161088e-01 4.57367361e-01 -8.02827001e-01 5.07305264e-01 4.33481112e-02 -2.34125629e-01 4.53833073e-01 5.55570945e-02 4.99930978e-01 3.76208365e-01 -1.34477139e-01 -5.73289394e-01 1.06082785e+00 -1.36716676e+00 -1.35299087e+00 2.33215362e-01 9.05734658e-01 -1.07320261e+00 4.35900748e-01 5.33875942e-01 -1.55548143e+00 -1.37560293e-01 -1.08699942e+00 1.78573877e-01 -7.27255583e-01 -3.14239264e-01 8.57801437e-01 1.43120694e+00 -1.60129595e+00 9.73287821e-01 -4.91727620e-01 6.59165755e-02 1.20568443e-02 8.24621081e-01 -2.64718989e-03 4.62812334e-01 -1.33193445e+00 1.08501506e+00 2.22979754e-01 -1.21242590e-01 -1.65216476e-01 -2.13068098e-01 -8.23704481e-01 -5.63443303e-01 -4.78693932e-01 -5.29636025e-01 1.44139910e+00 -2.59346128e-01 -1.65295815e+00 9.71794367e-01 -1.42069459e-01 -1.97814897e-01 6.14786744e-01 -1.28011424e-02 -8.31891418e-01 2.42498964e-01 -1.89849049e-01 5.76383233e-01 9.28263724e-01 -1.35132408e+00 -7.59897232e-01 -1.67359829e-01 -1.23336017e-01 2.66287565e-01 -1.25510961e-01 1.89734384e-01 2.11616129e-01 -1.12999000e-01 3.27147305e-01 -7.20919967e-01 -2.53068686e-01 -5.32041907e-01 -1.46512717e-01 -7.10518599e-01 7.70373225e-01 -4.51523662e-01 1.83705616e+00 -1.67618537e+00 -2.06720144e-01 3.98590982e-01 3.57815564e-01 -1.24705513e-03 2.40583986e-01 1.08380008e+00 -2.76906848e-01 7.32199550e-01 4.11965609e-01 -1.10722095e-01 2.24991128e-01 4.85861301e-01 -4.16602850e-01 3.05609167e-01 -1.29282743e-01 1.15307164e+00 -1.16605783e+00 -5.23096442e-01 4.11106765e-01 9.42391157e-02 -4.58732933e-01 4.79237735e-01 2.99364805e-01 1.47170946e-01 -3.56553018e-01 1.39399457e+00 1.15709066e+00 -1.31984919e-01 1.45911396e-01 5.30878425e-01 -3.79135728e-01 3.55090618e-01 -6.96863770e-01 1.05554795e+00 7.08333924e-02 5.00173986e-01 1.02364108e-01 -7.94621468e-01 3.33247900e-01 8.61540735e-01 4.37155962e-01 -9.67555881e-01 -2.26398129e-02 6.92409754e-01 1.12803578e-01 -6.07703328e-01 7.58228302e-01 3.94563079e-02 -3.64872098e-01 6.40070081e-01 -2.37588286e-01 -6.59476995e-01 9.64643434e-02 3.08100313e-01 6.22585893e-01 -5.18246442e-02 5.62923312e-01 -1.05613089e+00 7.86340594e-01 -1.82965681e-01 -1.81299388e-01 1.03415680e+00 -3.09923887e-01 3.19085121e-01 2.99841821e-01 -6.65102363e-01 -6.45341039e-01 -1.12307119e+00 -4.89381433e-01 1.30636716e+00 3.24267983e-01 -4.39044595e-01 -9.54439282e-01 -2.49762803e-01 1.77620783e-01 6.89606130e-01 -5.90509653e-01 3.84124845e-01 -5.03739953e-01 -8.32535863e-01 7.39044368e-01 3.45434904e-01 -5.07752821e-02 -1.33414865e+00 -6.58416986e-01 1.25490099e-01 -2.20292807e-01 -6.63697243e-01 -6.23428151e-02 4.48765576e-01 -1.35989368e+00 -5.18594682e-01 -6.66252747e-02 -8.20914626e-01 5.87345481e-01 2.46782884e-01 1.27047324e+00 5.39230824e-01 -2.31483161e-01 4.26904231e-01 -1.21292919e-01 -4.95818377e-01 -4.59671497e-01 -8.00336525e-02 5.28869390e-01 -5.87835789e-01 5.19427478e-01 -2.50617653e-01 -7.29350567e-01 5.37953973e-01 -6.88540697e-01 1.62748516e-01 1.79803044e-01 1.04410267e+00 1.35816500e-01 -9.34035778e-02 1.22507080e-01 -6.38007045e-01 8.72274399e-01 -1.69219792e-01 -3.78732830e-01 5.77745810e-02 -6.77108407e-01 -3.74140263e-01 3.21430594e-01 -3.25342178e-01 -1.01981449e+00 -4.87835288e-01 -9.82677937e-02 2.45538145e-01 1.11353043e-02 -1.46784872e-01 6.47139177e-02 -5.24923325e-01 8.02199244e-01 9.25758183e-02 1.99174434e-02 -6.80815242e-03 3.01039815e-01 7.09525108e-01 -6.82967342e-03 -6.68678164e-01 8.44880998e-01 4.91470337e-01 7.98524171e-02 -9.57177758e-01 -1.52186140e-01 -2.60129690e-01 -9.51962709e-01 -6.54426932e-01 6.56643391e-01 -6.78531289e-01 -9.10833478e-01 3.91110867e-01 -9.38691139e-01 -3.38627815e-01 -3.91645581e-01 4.25431967e-01 -1.01278400e+00 2.75717527e-02 -3.90154392e-01 -1.27895141e+00 -5.10977268e-01 -1.02017939e+00 9.43384409e-01 5.30070923e-02 -5.10597289e-01 -1.26927447e+00 5.87685481e-02 2.71537274e-01 1.81734428e-01 -1.73075795e-01 6.90226793e-01 -2.38256708e-01 -4.24233019e-01 -1.53791070e-01 2.34436691e-02 -1.39755070e-01 1.70832314e-02 4.95917559e-01 -9.81751978e-01 -5.31145096e-01 6.65065646e-02 -1.92070693e-01 -1.08835101e-01 6.52520418e-01 5.91872573e-01 -2.29931593e-01 -8.56000841e-01 5.40386558e-01 1.38545322e+00 3.85070026e-01 5.32770038e-01 7.28214979e-01 1.41836226e-01 5.53460240e-01 9.17806149e-01 4.63203549e-01 1.30579369e-02 3.28798652e-01 2.40537539e-01 1.49327129e-01 1.11720070e-01 -1.54819340e-01 3.77893507e-01 1.16112018e+00 -8.18235934e-01 -2.69281328e-01 -5.07867396e-01 4.42987174e-01 -1.72482407e+00 -1.40330648e+00 -4.32368398e-01 6.90478683e-01 6.25676990e-01 1.56016424e-01 -1.48347050e-01 3.35214496e-01 4.99015123e-01 -2.03574806e-01 -1.19133167e-01 -1.06291151e+00 -1.43546045e-01 3.15233678e-01 7.37729073e-01 1.00061214e+00 -7.20721722e-01 1.03317809e+00 1.29781246e+01 1.02230716e+00 2.21112028e-01 1.03134915e-01 5.16071796e-01 3.48020852e-01 -4.36954498e-01 -4.56139445e-02 -1.04416132e+00 2.72933897e-02 1.38140702e+00 -4.30666685e-01 6.85999811e-01 5.44219851e-01 3.44648361e-01 -4.23268199e-01 -1.26188684e+00 5.26221812e-01 9.73738134e-02 -1.40886843e+00 -2.83300440e-04 6.85225725e-01 7.73699820e-01 -5.08050561e-01 6.22419357e-01 3.24184299e-01 6.09259963e-01 -1.14389277e+00 8.60300779e-01 2.53660440e-01 1.03040910e+00 -6.05088234e-01 5.67372203e-01 1.68872893e-01 -1.14389896e+00 -2.20873043e-01 -8.77727985e-01 -1.00755692e+00 3.93533185e-02 -1.81779593e-01 -4.29956943e-01 3.48861217e-01 9.58353162e-01 2.99398601e-01 -3.93658698e-01 9.95779395e-01 -4.78694476e-02 1.04875881e-02 -2.71853864e-01 -4.48467314e-01 4.83122796e-01 -3.54241252e-01 4.66730654e-01 1.00164843e+00 2.48499006e-01 3.51035744e-01 -9.84472036e-02 4.01770771e-01 5.45058846e-01 3.29446048e-02 -1.19659424e+00 -1.78908288e-01 2.83276141e-01 9.16795909e-01 -4.83487815e-01 -4.22520459e-01 -2.00212970e-01 8.62069130e-01 -3.55488248e-02 5.01107454e-01 -6.11489356e-01 -4.35615242e-01 9.72222984e-01 -1.27327025e-01 -1.14700586e-01 -3.48497719e-01 -6.23769283e-01 -7.30352640e-01 -5.89872956e-01 -4.54965204e-01 5.93606755e-02 -5.53365827e-01 -1.39813089e+00 5.79277515e-01 -2.27688253e-02 -1.40553558e+00 -6.99901402e-01 -1.27676582e+00 -4.76714373e-01 4.92853165e-01 -1.11898029e+00 -1.10984349e+00 2.50124663e-01 4.52870727e-01 1.64141744e-01 -5.34416080e-01 1.39563632e+00 3.57715860e-02 1.00637585e-01 9.24474537e-01 6.69434488e-01 -7.37814724e-01 5.56605101e-01 -1.27867436e+00 5.68737745e-01 -1.37897313e-01 -4.31265175e-01 9.05828118e-01 6.28349900e-01 -5.39804697e-01 -1.41196322e+00 -3.66917729e-01 1.08350635e+00 -9.83769417e-01 6.55218959e-01 -3.86345625e-01 4.23767231e-02 7.88592756e-01 7.15902448e-01 -6.18741751e-01 8.21781039e-01 -1.83753878e-01 1.80774391e-01 5.75296998e-01 -1.39248300e+00 6.12354755e-01 1.66275799e+00 -4.63594139e-01 -6.25784039e-01 7.60327101e-01 8.13696027e-01 -6.94087505e-01 -1.30082703e+00 3.34633321e-01 8.65424156e-01 -8.75409484e-01 1.61978090e+00 -1.32660246e+00 -4.43697497e-02 2.81152606e-01 -2.61993498e-01 -9.32519078e-01 -5.96193194e-01 -1.23518765e+00 -5.33532679e-01 -5.83747849e-02 5.96577883e-01 -1.13057327e+00 3.42365682e-01 8.74560475e-01 -2.82833427e-01 -6.42737269e-01 -1.06996536e+00 -1.32016802e+00 -3.35779637e-02 -1.45572275e-01 4.90409225e-01 7.63798356e-01 6.83744550e-01 1.09839931e-01 -6.36873543e-02 -1.00294888e-01 5.33176839e-01 9.55312885e-03 4.41501856e-01 -1.34294486e+00 3.86843324e-01 -5.75816095e-01 -3.07655483e-01 -9.45992947e-01 -8.85957032e-02 -8.12076271e-01 -6.53862000e-01 -1.28511906e+00 -8.32044985e-03 -1.91056758e-01 -1.11109078e-01 -1.65725678e-01 3.67937148e-01 2.13746816e-01 1.20859891e-02 1.03788137e-01 -3.71160030e-01 6.18435517e-02 1.29639816e+00 6.91750320e-05 -1.62315920e-01 4.85058486e-01 -4.81304944e-01 7.84440815e-01 8.58408585e-02 -2.99253196e-01 -6.78878546e-01 6.11881316e-02 6.69384480e-01 4.61409837e-02 3.24159935e-02 -7.42885649e-01 5.37211418e-01 -3.75702560e-01 4.78586555e-01 -1.32223868e+00 1.30741090e-01 -9.61415648e-01 6.95283338e-02 9.40189242e-01 2.74610907e-01 1.20424610e-02 8.66204947e-02 5.39500564e-02 -1.44871444e-01 -5.70943117e-01 9.21121240e-01 -4.07591403e-01 -4.92852688e-01 -5.20386267e-03 -1.03226590e+00 8.97834301e-02 9.92593169e-01 -7.84614205e-01 -3.59281451e-01 -4.20183957e-01 -8.29068601e-01 -1.95836127e-02 6.50830388e-01 3.03609259e-02 7.30431557e-01 -1.51530886e+00 -2.30721906e-01 7.18729138e-01 -3.22939813e-01 -3.74400020e-01 -1.70157343e-01 6.58265352e-01 -1.32361674e+00 1.02442718e+00 -5.41665435e-01 -4.55340147e-01 -1.14228773e+00 4.78126436e-01 4.28307921e-01 -2.41845414e-01 -2.15481281e-01 1.11954463e+00 2.71224789e-02 -8.23025763e-01 1.85185194e-01 -9.21545625e-02 -7.54407048e-01 3.55081353e-03 6.88606799e-01 1.05194807e+00 -2.91290224e-01 -6.04341030e-01 -4.56784427e-01 6.65885091e-01 2.32151806e-01 -2.87484169e-01 9.17833567e-01 -2.19243199e-01 -9.89108324e-01 4.28274393e-01 8.38715494e-01 -1.36269778e-01 -3.69319022e-02 4.16855574e-01 1.25943512e-01 -8.02164078e-01 -4.32406247e-01 -3.55811834e-01 -1.85641110e-01 5.54822803e-01 5.34874737e-01 8.99602413e-01 8.57008278e-01 -3.02566767e-01 8.18335712e-01 9.66778398e-01 5.72402716e-01 -1.68019545e+00 -2.13140488e-01 6.89524531e-01 9.12339568e-01 -9.28350806e-01 5.44190466e-01 -7.27165341e-01 -4.14997995e-01 1.32979155e+00 4.68304873e-01 -1.55325383e-01 1.27306652e+00 5.74917436e-01 1.14069022e-02 -3.70670199e-01 -9.44949508e-01 1.12705544e-01 3.75366658e-01 1.11147714e+00 5.06513238e-01 5.07374525e-01 -9.66500878e-01 3.21953118e-01 -7.28706717e-01 -2.34555230e-01 4.90474731e-01 1.41972518e+00 -6.43810987e-01 -1.20391917e+00 -7.23931909e-01 4.61561680e-01 -5.49773455e-01 -1.16372630e-01 -5.28106689e-01 8.46754074e-01 -5.76629937e-02 1.49448860e+00 -1.97535474e-03 -5.03491640e-01 4.11356747e-01 1.41089618e-01 7.21762300e-01 -1.23501487e-01 -9.04846430e-01 4.15413082e-01 3.84890139e-01 -1.23056793e+00 -8.58632207e-01 -1.05834293e+00 -1.40667629e+00 -1.19437599e+00 -5.12782812e-01 1.89310342e-01 3.83317530e-01 3.90289724e-01 -2.06836104e-01 2.85260603e-02 9.80917513e-01 -1.07949340e+00 -3.90341938e-01 -9.39418614e-01 -1.00026262e+00 -7.84516707e-02 2.89751232e-01 -8.00943017e-01 -7.83523321e-01 2.83909619e-01]

[-7.23227071762085, 3.811842918395996]

50221572-b8a3-4629-b49b-54ae708328c0

on-the-effectiveness-of-knowledge-graph

2206.00983

https://arxiv.org/abs/2206.00983v2

https://arxiv.org/pdf/2206.00983v2.pdf

On the Effectiveness of Knowledge Graph Embeddings: a Rule Mining Approach

We study the effectiveness of Knowledge Graph Embeddings (KGE) for knowledge graph (KG) completion with rule mining. More specifically, we mine rules from KGs before and after they have been completed by a KGE to compare possible differences in the rules extracted. We apply this method to classical KGEs approaches, in particular, TransE, DistMult and ComplEx. Our experiments indicate that there can be huge differences between the extracted rules, depending on the KGE approach for KG completion. In particular, after the TransE completion, several spurious rules were extracted.

['Ana Ozaki', 'Ricardo Guimarães', 'Johanna Jøsang']

2022-06-02

['knowledge-graph-embeddings', 'knowledge-graph-embeddings']

['graphs', 'methodology']

[-2.38519982e-01 6.55162632e-01 -3.05422515e-01 -1.76697206e-02 3.23598444e-01 -5.64525068e-01 4.95101720e-01 6.88832104e-01 -2.42736533e-01 7.23835886e-01 9.26197246e-02 -4.50525641e-01 -6.94731712e-01 -1.34968472e+00 -7.22070873e-01 -7.23623186e-02 -4.13158178e-01 5.20077705e-01 5.97602844e-01 -3.09122115e-01 6.97070435e-02 4.43132430e-01 -1.48006189e+00 2.09886864e-01 8.27586591e-01 6.60251319e-01 -3.00698549e-01 3.05618316e-01 -4.61665988e-01 9.46273029e-01 -5.52345753e-01 -9.97740269e-01 1.07203625e-01 -5.88542856e-02 -9.56553280e-01 -6.88865483e-02 1.41945451e-01 9.64013934e-02 -6.37186706e-01 1.11198235e+00 -7.61022046e-02 3.31225507e-02 7.84735620e-01 -1.60027659e+00 -5.67353725e-01 1.22724473e+00 -4.02678519e-01 5.42235300e-02 2.78864115e-01 -3.12175244e-01 1.16078758e+00 -1.12115216e+00 1.14238334e+00 9.84609067e-01 8.37427557e-01 3.18802297e-01 -8.87718916e-01 -5.14668167e-01 9.41581726e-02 7.85483897e-01 -1.65562820e+00 -2.40632921e-01 5.85403204e-01 -3.74497622e-01 1.20124066e+00 1.03206471e-01 8.42775762e-01 5.77935994e-01 1.34585068e-01 4.90168631e-01 8.60542357e-01 -6.06434047e-01 3.10643137e-01 3.59881520e-01 7.28563249e-01 1.08350968e+00 1.25583708e+00 6.46093413e-02 -5.89791715e-01 -1.94879323e-01 5.08430421e-01 -4.66079980e-01 -3.80082428e-01 -4.44480032e-01 -7.90841281e-01 9.62074637e-01 2.53763217e-02 3.49908799e-01 -3.32043260e-01 -1.40492350e-01 5.50092101e-01 3.51428658e-01 -1.30211189e-02 5.18439054e-01 -7.70087183e-01 1.30426377e-01 -4.56205875e-01 1.54397249e-01 1.22371054e+00 1.04021204e+00 9.93780494e-01 1.10555552e-01 7.72643536e-02 3.43205273e-01 1.72697812e-01 -2.84502774e-01 1.52769968e-01 -3.48797321e-01 4.12658900e-01 1.00780749e+00 2.07181387e-02 -1.38966763e+00 -1.11965150e-01 -1.33088961e-01 -5.14556468e-01 5.57183512e-02 2.64737397e-01 -2.33947486e-01 -9.56635594e-01 1.18654728e+00 5.03671646e-01 3.70200992e-01 2.86481053e-01 2.80043364e-01 1.07419062e+00 1.85639501e-01 -1.85079058e-03 -2.81559229e-01 9.93431687e-01 -8.02916110e-01 -9.55942631e-01 -4.46713753e-02 9.34070885e-01 -3.54776919e-01 6.34906530e-01 5.70534587e-01 -4.88659203e-01 -4.00158018e-01 -1.19318914e+00 3.62182081e-01 -1.13180780e+00 -1.60345227e-01 8.13708484e-01 7.72460401e-01 -8.13738346e-01 9.25899863e-01 -4.74456817e-01 -4.24691051e-01 3.29772085e-01 2.69703478e-01 -5.85022390e-01 -2.69418061e-01 -1.49958956e+00 9.65719938e-01 1.29206896e+00 -1.02395877e-01 -4.91223544e-01 -6.98721886e-01 -1.05022204e+00 9.50166807e-02 1.11313236e+00 -5.23389935e-01 5.36481559e-01 -1.75552756e-01 -7.38849163e-01 5.61035812e-01 2.02749372e-01 -6.26699686e-01 1.67959392e-01 -1.58483118e-01 -1.10759747e+00 -8.89570639e-02 -8.66215453e-02 4.84137982e-02 5.25199354e-01 -1.28426647e+00 -5.51789582e-01 -4.42765951e-01 -4.03390229e-02 -3.48063670e-02 -3.28285128e-01 -1.13920756e-01 -6.22617245e-01 -4.47180241e-01 -2.77419925e-01 -5.17200828e-01 -1.93464011e-02 -5.92262864e-01 -6.56046569e-01 -6.57559097e-01 9.06284213e-01 -6.27243817e-01 1.84838271e+00 -2.11732817e+00 -2.13820115e-01 7.93187678e-01 6.64961457e-01 5.25668025e-01 5.80683537e-02 7.56560504e-01 -3.53656895e-02 7.51083374e-01 2.22709924e-01 4.01907772e-01 -7.97538012e-02 9.21122670e-01 -1.02160431e-01 -1.80009454e-02 6.36585578e-02 1.10399318e+00 -9.12955284e-01 -6.53501034e-01 1.26071736e-01 -7.89721757e-02 -2.81363308e-01 -9.93324742e-02 -2.91308105e-01 -5.34365177e-01 -3.91347975e-01 5.87828934e-01 6.16394818e-01 -7.93626234e-02 9.53104436e-01 -5.19863963e-01 -6.72785491e-02 2.84998953e-01 -1.57117951e+00 8.83084476e-01 6.23065718e-02 2.74111807e-01 -2.99057245e-01 -1.18336654e+00 8.64634395e-01 4.30103913e-02 2.24820495e-01 -2.11855277e-01 8.73596594e-03 7.23402053e-02 7.30036870e-02 -5.00889480e-01 7.62345135e-01 7.39976987e-02 8.49133134e-02 1.52433112e-01 2.81736702e-01 1.99144214e-01 7.41849601e-01 5.88013470e-01 1.33706951e+00 -1.10895239e-01 8.95865619e-01 -2.17187628e-01 1.62633806e-01 3.60546142e-01 9.46659327e-01 8.25301766e-01 -2.68439353e-02 -4.52578738e-02 7.42977858e-01 -5.64789534e-01 -5.88829398e-01 -8.50767434e-01 2.09330723e-01 4.18288052e-01 2.40855575e-01 -1.20941305e+00 -2.98689127e-01 -1.33691049e+00 4.53371584e-01 8.42836916e-01 -6.34644210e-01 -3.76557976e-01 -4.15242285e-01 -5.00797093e-01 7.00209498e-01 6.52469933e-01 3.89186889e-01 -1.07647157e+00 -3.13013226e-01 5.70744634e-01 2.74981678e-01 -1.24487197e+00 -1.96452767e-01 3.48307282e-01 -6.38949931e-01 -1.79954374e+00 5.68363905e-01 -6.50221825e-01 4.87160236e-01 2.83515826e-02 1.36930144e+00 2.81816900e-01 -2.45707780e-01 4.30550694e-01 -9.09661591e-01 -5.29985785e-01 -4.89668041e-01 -1.93210155e-01 2.35323250e-01 -2.17774764e-01 7.89437473e-01 -5.46336055e-01 -1.50220776e-02 2.13461384e-01 -9.24535334e-01 -3.58371228e-01 4.76159990e-01 4.96857554e-01 6.65429533e-01 9.61949646e-01 5.67031085e-01 -1.32265520e+00 9.83742237e-01 -5.82815349e-01 -3.24553788e-01 7.82373607e-01 -1.37238634e+00 2.87808537e-01 5.58740258e-01 -3.30882132e-01 -6.67468965e-01 -3.60030323e-01 1.19946763e-01 -6.74436748e-01 -3.34713794e-02 1.08526766e+00 -7.89484903e-02 -1.18615873e-01 5.70926309e-01 5.05982973e-02 -3.42741936e-01 -3.32589328e-01 5.04715800e-01 3.46244425e-01 2.95698643e-01 -4.18923080e-01 8.17904532e-01 -2.20283424e-03 -2.12828532e-01 -5.83238423e-01 -5.04287004e-01 -4.04504128e-02 -3.91851932e-01 -9.82520655e-02 7.37528324e-01 -3.50182116e-01 -5.57005882e-01 1.39650926e-02 -1.03807306e+00 5.31663187e-02 -6.84390247e-01 5.71303368e-01 7.62889832e-02 5.35429716e-01 -3.06777447e-01 -5.65269828e-01 -2.00338051e-01 -2.95151442e-01 3.89322937e-02 1.65827736e-01 -3.93740237e-01 -1.12601531e+00 7.85927624e-02 -1.97190419e-03 -1.83239654e-01 4.72265601e-01 1.36984742e+00 -1.07409763e+00 -2.38742530e-01 1.01730056e-01 -8.59516636e-02 3.10653627e-01 5.48849583e-01 3.49601358e-01 -3.00330818e-01 9.31875873e-03 -5.29689729e-01 5.25055220e-03 8.71001184e-01 -8.88632238e-02 9.13750827e-01 -6.84175193e-01 -7.35138357e-01 3.35846752e-01 1.64492559e+00 3.75897825e-01 8.46816599e-01 3.47641796e-01 7.30650425e-01 4.24563140e-01 7.50577569e-01 3.53421539e-01 3.59338015e-01 4.90015715e-01 1.66099548e-01 3.02523494e-01 -9.17354450e-02 -5.80793560e-01 3.55904549e-01 9.69305336e-01 -2.26309687e-01 -1.94915906e-01 -8.90889466e-01 8.89886498e-01 -1.76880705e+00 -8.07991624e-01 -4.27135289e-01 2.03437591e+00 8.28061998e-01 2.98057556e-01 -1.63409829e-01 6.37227058e-01 7.63993740e-01 1.27732188e-01 -3.02177131e-01 -9.06737566e-01 -2.59035397e-02 6.04903817e-01 6.96682453e-01 4.36039060e-01 -8.04199338e-01 1.23358309e+00 6.96201181e+00 9.32349145e-01 -3.65156651e-01 -4.78772670e-02 -2.44866088e-01 5.01706183e-01 -5.13776243e-01 3.30589861e-01 -9.47237372e-01 1.97451681e-01 5.74469328e-01 -5.41698456e-01 3.03146541e-01 8.09673786e-01 -3.85072500e-01 -5.77021465e-02 -9.62106168e-01 7.23514080e-01 -4.25556004e-02 -1.60856402e+00 3.24214697e-01 -3.82966995e-02 6.48047149e-01 -3.69547248e-01 -6.59379244e-01 6.46132350e-01 8.00414681e-01 -7.55710423e-01 2.54482955e-01 2.49196157e-01 5.42753279e-01 -1.00070548e+00 7.11263359e-01 7.72910491e-02 -1.46788490e+00 8.39138776e-02 -3.40616792e-01 9.30195376e-02 -3.19635645e-02 1.24594569e+00 -1.34222448e+00 1.51489341e+00 7.18504727e-01 5.29010832e-01 -6.35672688e-01 5.56702316e-01 -5.99928617e-01 7.11784840e-01 -1.58489093e-01 8.49670097e-02 -2.12943763e-01 -4.81909573e-01 4.93305981e-01 1.15939724e+00 2.29077205e-01 -5.47198765e-02 1.99747026e-01 7.33000636e-01 -1.41415045e-01 -1.83003936e-02 -1.06792903e+00 -4.63253200e-01 8.11290681e-01 9.91743445e-01 -6.34679794e-01 -5.82665682e-01 -6.99417531e-01 5.85577250e-01 5.13765693e-01 3.20216238e-01 -7.15465128e-01 -8.03474784e-01 5.81598043e-01 2.68848360e-01 9.30570245e-01 -2.71512508e-01 1.25108466e-01 -1.00579095e+00 2.44721889e-01 -7.67882109e-01 8.57658923e-01 -5.61204731e-01 -1.25988162e+00 4.08828110e-01 1.65029168e-01 -6.79213822e-01 2.99164113e-02 -4.18293774e-01 -6.95080936e-01 1.87198982e-01 -1.37500429e+00 -7.61362672e-01 -5.53214476e-02 8.89965713e-01 2.21070219e-02 -1.34262204e-01 7.41622448e-01 1.65127963e-01 -5.63269973e-01 7.04585314e-01 -2.12679029e-01 3.81935030e-01 1.98738664e-01 -1.24029827e+00 1.80636570e-01 8.75218272e-01 6.59423411e-01 6.87534094e-01 5.97555757e-01 -1.28128076e+00 -1.44379008e+00 -1.42671406e+00 1.08854949e+00 -2.72784263e-01 8.91253531e-01 -1.80941477e-01 -1.07624793e+00 1.00898933e+00 1.15108207e-01 1.30356103e-01 5.11207819e-01 4.90015984e-01 -6.51266754e-01 -8.04518759e-02 -1.16943252e+00 5.15154064e-01 1.40014696e+00 -2.80057997e-01 -1.03878391e+00 -1.21246234e-01 5.54552376e-01 1.13730289e-01 -1.23201299e+00 7.88143098e-01 4.03280765e-01 -7.92653739e-01 6.13257706e-01 -9.04600799e-01 1.39071003e-01 -5.46363235e-01 1.71366617e-01 -1.44958222e+00 -3.50852549e-01 -3.68238688e-01 -8.20042729e-01 1.31087708e+00 5.93994915e-01 -8.14977288e-01 7.44342744e-01 2.45729357e-01 9.78703871e-02 -9.14981604e-01 -7.24852800e-01 -1.27202737e+00 -2.46235251e-01 -3.47078145e-01 8.99709761e-01 1.46357751e+00 3.11103284e-01 2.38414600e-01 -2.19299451e-01 2.65471250e-01 4.15915906e-01 2.67750144e-01 7.44367957e-01 -1.67106855e+00 -2.09161267e-01 -6.73830435e-02 -6.38780236e-01 -4.35346365e-01 9.65884477e-02 -1.10000002e+00 -5.70503891e-01 -1.89750409e+00 4.01460677e-02 -2.86031693e-01 -4.77613837e-01 8.49610686e-01 -1.46902204e-01 -3.34270000e-01 -1.33564293e-01 -1.36261955e-01 -5.09805858e-01 3.30244452e-01 1.09003496e+00 -2.56265372e-01 -4.78882313e-01 -6.14616036e-01 -9.55789506e-01 6.40716910e-01 8.17723930e-01 -8.02179277e-01 -5.42387784e-01 4.14847657e-02 6.52616858e-01 -2.58076042e-01 5.33860875e-03 -6.98072016e-01 2.65172392e-01 -2.77732730e-01 -2.54418626e-02 -4.23272550e-01 -3.79850656e-01 -6.35436237e-01 4.42026228e-01 2.90075272e-01 2.70308346e-01 -1.02021649e-01 3.90383422e-01 8.20878744e-01 -5.09599686e-01 -3.89777422e-01 1.12688944e-01 4.08434821e-03 -1.19856191e+00 2.72011727e-01 -3.85896355e-01 1.04420029e-01 1.08830047e+00 -6.55396283e-01 -3.72525841e-01 1.99045539e-02 -1.16192460e+00 2.59652734e-01 1.83641553e-01 4.13953900e-01 1.00706041e+00 -1.49940395e+00 -4.77826446e-01 1.15529664e-01 4.14002508e-01 -2.29223371e-02 -2.70038903e-01 7.68107951e-01 -1.26625165e-01 7.90182278e-02 -6.69577345e-02 2.94841260e-01 -1.49461746e+00 8.71677339e-01 -2.81863883e-02 -7.31485724e-01 -6.90663695e-01 4.02857780e-01 -3.57773215e-01 -3.09097469e-01 -8.91911685e-02 -3.40718150e-01 -5.78205049e-01 3.28603595e-01 1.45688534e-01 6.64443195e-01 2.80366451e-01 3.47121693e-02 -5.91295421e-01 4.69967127e-01 -3.34241837e-01 2.43498296e-01 1.40887189e+00 4.96171743e-01 -2.26430565e-01 1.02951296e-01 6.45983875e-01 4.90553886e-01 -2.90441364e-01 -7.56359622e-02 4.66256857e-01 -5.61697781e-01 -1.27416253e-01 -5.75223327e-01 -1.14258921e+00 6.42851517e-02 7.85521939e-02 5.14219642e-01 8.35846484e-01 1.80274293e-01 4.72045243e-01 5.45197904e-01 6.57186747e-01 -1.47831953e+00 -3.84343356e-01 5.25032997e-01 6.83246791e-01 -7.87523687e-01 3.16744506e-01 -8.68153334e-01 -5.13583004e-01 9.60216701e-01 4.87265974e-01 -1.67399868e-02 9.63628769e-01 1.82369903e-01 -5.18239617e-01 -7.23683774e-01 -6.83022380e-01 -4.26292956e-01 9.90111232e-02 8.14329445e-01 -1.94848374e-01 2.93680340e-01 -7.95380652e-01 9.48834777e-01 -1.31995335e-01 1.96696505e-01 7.10100710e-01 9.21771765e-01 -3.80656004e-01 -1.43844163e+00 -2.86417012e-03 8.25926900e-01 -1.64509982e-01 -1.38149142e-01 -1.00048935e+00 1.42807674e+00 5.29109478e-01 1.02028799e+00 -4.75502223e-01 -9.73362386e-01 5.53560078e-01 4.21693325e-01 5.78571677e-01 -7.62563527e-01 -3.62247050e-01 -6.99465692e-01 8.33220661e-01 -4.73224670e-01 -2.30130658e-01 -2.90899098e-01 -1.45224535e+00 -3.92914742e-01 -7.70685732e-01 4.94092792e-01 1.04819953e-01 7.05853939e-01 6.70094788e-01 4.19673115e-01 4.32791412e-01 4.14139330e-01 -7.50978291e-02 -7.93577671e-01 -1.12074316e+00 6.96803212e-01 -5.40681362e-01 -9.82996821e-01 -4.82972413e-01 -1.06275640e-01]

[8.814497947692871, 7.858809947967529]

2c9949ff-6164-4829-9085-0ae236baeb90

learning-a-predictive-model-for-music-using

1709.08842

http://arxiv.org/abs/1709.08842v1

http://arxiv.org/pdf/1709.08842v1.pdf

Learning a Predictive Model for Music Using PULSE

Predictive models for music are studied by researchers of algorithmic composition, the cognitive sciences and machine learning. They serve as base models for composition, can simulate human prediction and provide a multidisciplinary application domain for learning algorithms. A particularly well established and constantly advanced subtask is the prediction of monophonic melodies. As melodies typically involve non-Markovian dependencies their prediction requires a capable learning algorithm. In this thesis, I apply the recent feature discovery and learning method PULSE to the realm of symbolic music modeling. PULSE is comprised of a feature generating operation and L1-regularized optimization. These are used to iteratively expand and cull the feature set, effectively exploring feature spaces that are too large for common feature selection approaches. I design a general Python framework for PULSE, propose task-optimized feature generating operations and various music-theoretically motivated features that are evaluated on a standard corpus of monophonic folk and chorale melodies. The proposed method significantly outperforms comparable state-of-the-art models. I further discuss the free parameters of the learning algorithm and analyze the feature composition of the learned models. The models learned by PULSE afford an easy inspection and are musicologically interpreted for the first time.

['Jonas Langhabel']

2017-09-26

['music-modeling']

['music']

[ 3.32280070e-01 -3.79824549e-01 -1.39949962e-01 -4.42414954e-02 -5.05448222e-01 -6.91251457e-01 6.31766915e-01 -2.88941637e-02 -1.82621300e-01 4.73314404e-01 2.23239124e-01 1.59084350e-02 -6.76925838e-01 -4.98770267e-01 -3.44290346e-01 -5.48764765e-01 -3.08822066e-01 7.85343528e-01 2.58728117e-02 -4.47217196e-01 6.51362479e-01 3.70600164e-01 -2.05787587e+00 7.70513356e-01 3.91851157e-01 8.62718523e-01 3.27341944e-01 7.84561992e-01 -1.08274437e-01 7.78730094e-01 -3.28665346e-01 -1.71914354e-01 2.20680654e-01 -8.00626993e-01 -9.53541994e-01 -8.96272808e-02 7.30980486e-02 6.27720714e-01 1.91220790e-02 6.57933891e-01 6.25479817e-01 5.05525887e-01 5.70266008e-01 -1.18310595e+00 -7.28101805e-02 1.21231639e+00 -3.04539176e-03 -2.79962216e-02 5.34106314e-01 -1.46939963e-01 1.38254285e+00 -9.53073144e-01 6.42103791e-01 8.93815577e-01 8.72913837e-01 4.89485174e-01 -1.39964235e+00 -6.48744166e-01 -3.35560769e-01 6.75369680e-01 -1.31394744e+00 -4.22971547e-01 1.07165504e+00 -6.69846714e-01 1.04189229e+00 8.48741710e-01 1.11388481e+00 7.42796481e-01 1.13936961e-02 9.32630956e-01 1.05026996e+00 -7.81007648e-01 1.23902515e-01 5.48861884e-02 -1.40084922e-01 4.65152860e-01 -7.08823919e-01 2.50588655e-01 -1.10664999e+00 -2.55897701e-01 6.73833549e-01 -2.65254378e-01 1.28569826e-02 -1.16481312e-01 -1.53581595e+00 6.34027362e-01 5.64398803e-03 4.82487261e-01 -4.83274639e-01 7.01632202e-02 4.74888653e-01 5.45805633e-01 -3.95295583e-02 9.50127959e-01 -6.80922210e-01 -5.63596308e-01 -1.37217391e+00 6.59824073e-01 1.11238122e+00 6.66322112e-01 5.34253657e-01 1.84789762e-01 -7.28849471e-02 9.88788843e-01 1.08689137e-01 -5.89519441e-02 7.12055862e-01 -8.37782621e-01 -9.32931453e-02 5.86957395e-01 -3.10347110e-01 -7.18006968e-01 -4.95581865e-01 -6.99323595e-01 -5.73311806e-01 2.18672439e-01 4.46248978e-01 4.14046437e-01 -3.35194767e-01 1.34787405e+00 1.95461616e-01 3.31045687e-01 -3.31241220e-01 7.03388751e-01 5.98455429e-01 4.87909526e-01 3.88322324e-02 -5.59902608e-01 1.15010142e+00 -9.98718739e-01 -5.24241269e-01 2.02859730e-01 2.75316328e-01 -1.37567008e+00 1.29738498e+00 1.05850649e+00 -1.14612567e+00 -7.04492509e-01 -8.71230006e-01 1.72329813e-01 -3.36584561e-02 2.24459931e-01 8.41654003e-01 3.54395747e-01 -3.78383070e-01 1.44179809e+00 -6.35328174e-01 4.79127094e-02 1.99361935e-01 5.86601794e-01 -1.41117200e-01 5.29051900e-01 -8.59807312e-01 8.12356770e-01 8.72189581e-01 -1.24351487e-01 -8.53499532e-01 -8.62056196e-01 -4.32982653e-01 -1.01465091e-01 3.11130404e-01 -4.96540844e-01 1.37340450e+00 -1.24542642e+00 -1.73037100e+00 9.25540924e-01 4.66818511e-02 -3.81092966e-01 2.81780541e-01 -2.95601606e-01 -4.97804165e-01 -2.08201900e-01 -2.17843443e-01 2.07464859e-01 1.01340604e+00 -8.36858988e-01 -4.27538037e-01 3.99840809e-02 -3.87669891e-01 2.69269705e-01 -1.12723947e-01 3.04623693e-01 -1.57292143e-01 -1.08364606e+00 2.90373802e-01 -1.18460286e+00 -2.18476191e-01 -6.47219539e-01 -4.58168626e-01 -9.91996899e-02 4.17563677e-01 -5.50116718e-01 1.63496065e+00 -2.07136464e+00 5.26259243e-01 4.31156456e-01 -2.91798234e-01 -2.03529730e-01 1.12479500e-01 6.71247125e-01 -2.86149949e-01 -2.46751815e-01 -1.30710155e-01 -7.20297024e-02 3.10927540e-01 1.15749121e-01 -4.08992410e-01 1.79221883e-01 -2.81485409e-01 8.49242032e-01 -6.52135015e-01 -4.32963282e-01 3.01137835e-01 3.67747039e-01 -7.28611887e-01 1.85330287e-01 -6.24701023e-01 7.19382048e-01 -1.73146099e-01 5.69616139e-01 -8.72618929e-02 1.86634064e-03 1.66779533e-01 -2.24660978e-01 -3.74869227e-01 6.82006359e-01 -1.44124961e+00 2.05386043e+00 -2.78473258e-01 5.55065215e-01 -4.44709808e-01 -7.57501781e-01 1.12881482e+00 3.71688336e-01 6.81184769e-01 -1.58556223e-01 1.51880994e-01 3.69820803e-01 3.64260793e-01 -4.80104387e-01 5.92802703e-01 -6.11664116e-01 -1.12406813e-01 5.91835797e-01 2.31699079e-01 -4.75433141e-01 3.86971354e-01 -3.40546310e-01 8.22147727e-01 7.42945254e-01 5.67772090e-01 -3.42527956e-01 7.60378718e-01 2.02094615e-01 6.04720592e-01 4.49657887e-01 3.73373717e-01 4.04290795e-01 1.72485292e-01 -7.40078390e-01 -1.09802473e+00 -7.17854500e-01 -1.05351411e-01 1.83700538e+00 -4.48749900e-01 -1.05526197e+00 -4.51916337e-01 -8.69419873e-02 -1.10832356e-01 7.74493635e-01 -3.67902458e-01 2.34965072e-03 -7.31846869e-01 -4.89921898e-01 4.52095389e-01 1.43119559e-01 -1.27825275e-01 -1.74300158e+00 -5.40608644e-01 5.20555317e-01 -1.59464106e-02 -4.97754246e-01 -1.77106231e-01 5.01922846e-01 -8.97451341e-01 -1.11812830e+00 -9.12407488e-02 -9.24452305e-01 -1.74322262e-01 -4.98978794e-01 1.37171876e+00 2.60577947e-02 -5.86079597e-01 3.28333706e-01 -4.58869457e-01 -5.79796374e-01 -6.57640338e-01 2.92974472e-01 2.18294486e-01 -7.88198560e-02 2.49450654e-01 -1.07278633e+00 -2.74020702e-01 2.22513124e-01 -6.24517262e-01 4.61543053e-01 4.50894594e-01 8.43454421e-01 9.64056075e-01 -2.49427743e-02 4.30789292e-01 -7.75456071e-01 5.09633541e-01 -1.53249070e-01 -2.61238873e-01 1.37870178e-01 -5.49178720e-01 1.72443718e-01 6.26225710e-01 -7.83663809e-01 -6.19422495e-01 5.69949508e-01 -1.63466781e-01 -3.51385802e-01 -1.28204808e-01 6.29670382e-01 -8.39138180e-02 -4.97745313e-02 8.82858634e-01 3.10855478e-01 -2.06210405e-01 -8.54339361e-01 2.75461048e-01 4.50607091e-01 6.52921200e-01 -8.49998534e-01 6.83137059e-01 1.16117271e-02 2.79450446e-01 -9.67294931e-01 -7.88975716e-01 -4.29743260e-01 -8.30964506e-01 -4.08869267e-01 4.24704999e-01 -4.95705754e-01 -9.51564670e-01 5.14230914e-02 -8.35057735e-01 -2.52990097e-01 -7.90620744e-01 6.79455519e-01 -1.16262484e+00 1.31733552e-01 -4.93278265e-01 -8.25176835e-01 -4.29554611e-01 -8.18431675e-01 6.99849546e-01 1.39778078e-01 -8.75409544e-01 -8.25960636e-01 4.84610587e-01 1.82134971e-01 1.84473604e-01 1.28654212e-01 1.18978536e+00 -1.02871358e+00 -3.30701649e-01 -3.13833430e-02 7.69006491e-01 1.40889660e-01 -8.47580656e-02 1.02892686e-02 -1.17705417e+00 6.02379665e-02 -1.21161610e-01 -3.26099634e-01 7.60419130e-01 1.86319649e-01 1.19961023e+00 -5.62151484e-02 5.60654961e-02 7.25808024e-01 1.05127013e+00 3.06535866e-02 3.25953573e-01 7.24082232e-01 5.28238237e-01 5.48888385e-01 5.86857736e-01 8.14826548e-01 -1.69995263e-01 9.43898082e-01 2.62455851e-01 6.30709291e-01 -7.98442736e-02 -4.06895101e-01 3.66362721e-01 1.31130922e+00 -6.21243834e-01 5.78833640e-01 -8.32609832e-01 2.90910006e-01 -2.03101325e+00 -1.26815128e+00 -2.27363020e-01 2.24082303e+00 1.11349547e+00 3.14979777e-02 4.75034475e-01 9.60539758e-01 4.55507755e-01 -1.14462771e-01 -1.14578873e-01 -4.16620076e-01 -2.58292496e-01 8.60689223e-01 -1.38026252e-01 3.55658293e-01 -1.32268620e+00 1.09010506e+00 6.61968899e+00 9.80942130e-01 -9.28461730e-01 -4.89552170e-02 -1.50212109e-01 -3.67669970e-01 -1.91440016e-01 2.88575679e-01 -5.02677143e-01 1.02495790e-01 8.75434995e-01 -2.57744938e-01 9.30442691e-01 1.07025194e+00 1.92119673e-01 2.48653382e-01 -1.34245133e+00 1.34331191e+00 -3.54326032e-02 -1.54250073e+00 8.04349631e-02 -1.35047495e-01 6.97120309e-01 -1.97953314e-01 -4.57899980e-02 3.70441079e-01 -3.46450388e-01 -1.26576221e+00 9.53294754e-01 9.11998689e-01 5.93198180e-01 -8.73317599e-01 2.60681599e-01 3.66349429e-01 -1.24001420e+00 -2.72506118e-01 -1.45144790e-01 -4.96905327e-01 2.22511292e-01 2.64657021e-01 -7.21218884e-01 3.61785352e-01 4.38232303e-01 9.38534915e-01 -5.60828507e-01 1.34777141e+00 -3.00382942e-01 8.81824791e-01 -3.44458193e-01 -6.33108541e-02 -6.07107840e-02 -2.09733233e-01 8.94438744e-01 1.25595021e+00 3.55446339e-01 -1.66930929e-01 3.13756227e-01 7.29712486e-01 3.46449554e-01 6.14586294e-01 -8.24303478e-02 -2.48184472e-01 2.51545489e-01 1.29520214e+00 -7.58218169e-01 -1.37595922e-01 1.73170537e-01 8.28297377e-01 -8.54722261e-02 -1.69065312e-01 -5.83850086e-01 -1.34237409e-01 4.57310975e-01 2.57813603e-01 2.06145182e-01 -2.00905800e-01 -3.91207665e-01 -1.07577097e+00 -3.95939559e-01 -1.25446212e+00 3.61314178e-01 -4.41171706e-01 -1.14205742e+00 5.83050728e-01 -1.24691926e-01 -1.38705468e+00 -7.39074290e-01 -5.96053481e-01 -6.51629388e-01 6.87525988e-01 -9.14838731e-01 -1.26225615e+00 -4.16977273e-04 8.21700156e-01 6.92849636e-01 -7.80443907e-01 1.45986116e+00 2.72923440e-01 -8.95588174e-02 2.43694097e-01 -6.04566373e-02 -1.62038431e-01 5.86891592e-01 -1.26662374e+00 2.91214045e-02 2.58866996e-01 1.15362573e+00 3.94750506e-01 9.23606813e-01 -5.40236354e-01 -1.45950615e+00 -5.20547092e-01 1.01786780e+00 -2.37571180e-01 8.82494152e-01 -2.07909286e-01 -6.59423411e-01 4.08668965e-01 -7.46245608e-02 -3.81246984e-01 1.25531912e+00 3.51713628e-01 -2.63224810e-01 -1.92345330e-03 -4.42525029e-01 3.94088179e-01 1.01965296e+00 -6.46498501e-01 -8.91745269e-01 3.64923030e-01 1.32424846e-01 -2.67220974e-01 -9.79706585e-01 2.20288396e-01 1.07645535e+00 -1.02411485e+00 9.39779699e-01 -8.61989617e-01 2.78437972e-01 -4.54119086e-01 -2.49421373e-01 -1.06417704e+00 -4.86278564e-01 -1.28614521e+00 -3.49009752e-01 1.04596317e+00 5.22067606e-01 1.55958951e-01 7.93072939e-01 1.05831400e-01 -2.46231154e-01 -5.56555808e-01 -8.90487671e-01 -7.40334690e-01 -2.68186748e-01 -1.07972467e+00 4.93921429e-01 1.01584828e+00 3.58858675e-01 4.05788898e-01 -6.03528917e-01 -5.17685890e-01 5.82255363e-01 6.14776433e-01 8.19231212e-01 -1.63949573e+00 -9.79263604e-01 -7.65699267e-01 -5.88963628e-01 -5.51074803e-01 1.60898134e-01 -1.37959766e+00 -1.19836971e-01 -9.41607594e-01 6.60424903e-02 -4.52451050e-01 -5.25647342e-01 5.71121812e-01 2.30963871e-01 5.34043550e-01 4.91374344e-01 4.82377112e-01 -4.31912750e-01 3.51027608e-01 1.05351412e+00 7.56985545e-02 -7.02564418e-01 3.48201782e-01 -2.51441866e-01 1.04255140e+00 9.52091336e-01 -6.50391757e-01 -1.55509844e-01 2.40152121e-01 8.15128326e-01 -1.97245046e-01 2.84281850e-01 -1.05785549e+00 2.35491931e-01 -3.57419670e-01 4.09578770e-01 -5.62467575e-01 5.48404992e-01 -6.93964660e-01 7.00217366e-01 4.65517282e-01 -4.51085359e-01 -8.07426870e-02 5.82506582e-02 8.39875713e-02 -3.48428011e-01 -5.20600021e-01 5.10861278e-01 -2.51620770e-01 -9.32979882e-01 5.28193377e-02 -2.04029351e-01 -1.50057301e-01 7.97246933e-01 -1.35619000e-01 5.05566597e-01 -1.85045362e-01 -1.38349068e+00 -4.92313325e-01 1.71608761e-01 4.97173160e-01 5.04142880e-01 -1.49564481e+00 -8.58939946e-01 1.95449010e-01 -3.64115462e-02 -5.09043157e-01 2.00108960e-02 9.29731071e-01 -5.68550408e-01 -1.01394281e-02 -4.39116627e-01 -4.86206889e-01 -1.60306525e+00 4.55537647e-01 5.36553003e-02 -2.39318430e-01 -6.66098952e-01 7.92338610e-01 -4.60045248e-01 -5.14113188e-01 1.84226274e-01 -3.21509689e-02 -3.66870344e-01 2.54389346e-01 5.90856373e-01 5.10128021e-01 -9.55298170e-02 -7.63006270e-01 -2.95746356e-01 6.12614930e-01 4.09992516e-01 -2.73572415e-01 1.81552804e+00 2.91587770e-01 -4.23153371e-01 1.13312054e+00 5.06634474e-01 1.84610084e-01 -7.68405974e-01 -9.21741500e-02 5.72787702e-01 -2.65551567e-01 -3.33278567e-01 -9.03328061e-01 -3.90192062e-01 6.99424326e-01 3.24963659e-01 3.39747310e-01 1.18245471e+00 2.81633288e-02 3.60032290e-01 4.30644006e-01 1.96978495e-01 -1.26104927e+00 -6.81300089e-03 8.87099266e-01 1.12989461e+00 -6.73921108e-01 2.87604302e-01 -9.68734473e-02 -7.90121615e-01 1.52702463e+00 3.63443457e-02 -4.49686170e-01 7.20954120e-01 3.54063272e-01 -1.95360839e-01 -1.07968580e-02 -8.54910851e-01 -1.96185797e-01 9.74562466e-01 2.27525860e-01 8.58809948e-01 2.13087633e-01 -4.72651303e-01 1.20525575e+00 -1.12447500e+00 -1.76377799e-02 8.14507231e-02 6.52065694e-01 -5.10502040e-01 -1.56850851e+00 -4.87320870e-01 3.38932514e-01 -5.75033307e-01 -2.63595402e-01 -4.87772077e-01 6.68968141e-01 5.11941612e-01 6.07431889e-01 -3.41535330e-01 -7.31807411e-01 2.19272554e-01 6.99337840e-01 9.04712617e-01 -5.90919256e-01 -1.34172690e+00 4.79757398e-01 1.87379688e-01 -4.21449840e-01 -5.92180252e-01 -8.43761921e-01 -1.14954066e+00 -1.60903394e-01 -2.89997935e-01 3.53395313e-01 9.41485763e-01 1.08008385e+00 -2.63471622e-02 5.07847130e-01 4.51383919e-01 -1.29086089e+00 -4.62353081e-01 -1.07860219e+00 -7.30141878e-01 4.59043711e-01 -2.55831867e-01 -4.57419872e-01 6.83551375e-03 3.95300001e-01]

[15.946097373962402, 5.3287506103515625]

acc2a2b4-3c32-4293-b30a-986691121abb

seed-driven-document-ranking-for-systematic

2112.04090

https://arxiv.org/abs/2112.04090v1

https://arxiv.org/pdf/2112.04090v1.pdf

Seed-driven Document Ranking for Systematic Reviews: A Reproducibility Study

Screening or assessing studies is critical to the quality and outcomes of a systematic review. Typically, a Boolean query retrieves the set of studies to screen. As the set of studies retrieved is unordered, screening all retrieved studies is usually required for high-quality systematic reviews. Screening prioritisation, or in other words, ranking the set of studies, enables downstream activities of a systematic review to begin in parallel. We investigate a method that exploits seed studies -- potentially relevant studies used to seed the query formulation process -- for screening prioritisation. Our investigation aims to reproduce this method to determine if it is generalisable on recently published datasets and determine the impact of using multiple seed studies on effectiveness.We show that while we could reproduce the original methods, we could not replicate their results exactly. However, we believe this is due to minor differences in document pre-processing, not deficiencies with the original methodology. Our results also indicate that our reproduced screening prioritisation method, (1) is generalisable across datasets of similar and different topicality compared to the original implementation, (2) that when using multiple seed studies, the effectiveness of the method increases using our techniques to enable this, (3) and that the use of multiple seed studies produces more stable rankings compared to single seed studies. Finally, we make our implementation and results publicly available at the following URL: https://github.com/ielab/sdr

['Guido Zuccon', 'Ahmed Mourad', 'Harrisen Scells', 'Shuai Wang']

2021-12-08

['document-ranking']

['natural-language-processing']

[ 3.41147840e-01 -5.11613190e-02 -5.72701335e-01 -1.59116685e-02 -1.00944734e+00 -9.33543324e-01 4.39734817e-01 5.46694994e-01 -5.00223219e-01 7.07063496e-01 5.18661261e-01 -9.05250967e-01 -5.71685672e-01 -6.54156029e-01 -7.85701275e-01 -1.85214095e-02 1.24106161e-01 2.19506249e-01 4.73565489e-01 3.38586599e-01 8.22851717e-01 5.51314414e-01 -1.49348783e+00 3.01144272e-01 6.80048466e-01 2.23805860e-01 3.99829805e-01 5.05235195e-01 9.98821706e-02 3.86417896e-01 -5.86782694e-01 -1.86300963e-01 1.86117887e-01 -4.09930855e-01 -9.20595944e-01 -2.95397073e-01 2.24492773e-01 -2.88150847e-01 3.04726928e-01 8.33366334e-01 4.22640920e-01 -2.82193273e-01 6.98816657e-01 -9.69454169e-01 -4.74702895e-01 4.82254505e-01 -4.13670003e-01 3.55773151e-01 7.39317477e-01 5.87543137e-02 9.50984597e-01 -9.39167976e-01 1.00206482e+00 9.86573279e-01 5.20625830e-01 5.82799353e-02 -1.00858617e+00 -9.81433213e-01 -7.61363143e-03 -3.41016829e-01 -1.28347635e+00 -6.80036247e-01 2.59430707e-01 -6.51672959e-01 9.01678205e-01 3.41420263e-01 8.06024849e-01 6.04570806e-01 3.88515025e-01 4.85913921e-03 1.27710772e+00 -6.80590451e-01 2.40472913e-01 2.48235703e-01 1.57572150e-01 2.52364784e-01 1.15496325e+00 6.64938837e-02 -2.76646793e-01 -5.24167955e-01 4.24632341e-01 -7.81743601e-02 -2.86785066e-01 -5.63014150e-02 -1.21036696e+00 6.90728307e-01 -5.28615527e-02 3.28525841e-01 -5.30128658e-01 -4.13815938e-02 3.78035784e-01 1.74803570e-01 3.85219038e-01 8.75991046e-01 -3.11299205e-01 -2.70142943e-01 -1.40733576e+00 2.69808292e-01 8.80453348e-01 8.31211805e-01 4.49293435e-01 -6.30644917e-01 -4.03005891e-02 7.44367123e-01 3.62709969e-01 3.61417711e-01 2.44506881e-01 -1.03674591e+00 4.28204536e-01 8.24886382e-01 2.83924550e-01 -8.43929529e-01 -4.29005653e-01 -5.47396298e-03 -2.76455879e-01 -5.61381280e-02 1.74033672e-01 -1.88413382e-01 -7.69689620e-01 1.50260246e+00 9.82370600e-03 -4.45136100e-01 -1.23392902e-01 6.04291499e-01 1.07072282e+00 5.65778077e-01 2.79097676e-01 -5.91931343e-01 1.46047175e+00 -3.01863819e-01 -7.86739528e-01 7.80728608e-02 7.09912896e-01 -1.26746762e+00 1.05642998e+00 4.50639546e-01 -1.46261930e+00 -3.36374715e-02 -1.15452528e+00 2.47311413e-01 -5.98472536e-01 -1.06454529e-01 3.03684175e-01 7.53865600e-01 -1.26240408e+00 3.39505345e-01 -5.44965506e-01 -6.24865830e-01 1.60579443e-01 3.46874565e-01 -2.83879399e-01 -8.25073272e-02 -1.18475461e+00 1.08202052e+00 8.75706226e-02 -2.63001442e-01 -3.64725977e-01 -7.47604609e-01 -5.90403557e-01 -2.54683979e-02 5.80813229e-01 -7.18945920e-01 1.19555962e+00 -4.81725395e-01 -8.03932905e-01 6.78568244e-01 -5.07140338e-01 6.46896437e-02 8.88746604e-02 1.38411120e-01 -2.69365937e-01 3.13215196e-01 5.88007927e-01 3.07807326e-01 -3.66573334e-02 -1.08266842e+00 -7.22613573e-01 -2.74390310e-01 -1.36845186e-02 2.74772137e-01 -2.56062120e-01 8.49783719e-01 -7.49952853e-01 -5.51747322e-01 -6.30791634e-02 -9.06619310e-01 -3.20344061e-01 -4.80592787e-01 -2.41104770e-03 -1.64771438e-01 3.10399711e-01 -5.21915138e-01 1.77204061e+00 -1.82607436e+00 -3.28846216e-01 2.45110884e-01 1.23716697e-01 6.80785924e-02 -7.55690336e-02 9.97698784e-01 -1.15622744e-01 9.27650630e-01 -8.85955989e-03 7.57591724e-02 -2.87070006e-01 -2.56419718e-01 -1.47675769e-02 4.99244779e-01 3.21678340e-01 7.53856063e-01 -8.34288716e-01 -5.73053062e-01 4.44913004e-03 1.82171464e-01 -4.47354198e-01 -2.21452400e-01 3.18169355e-01 -3.47016573e-01 -4.79129255e-01 6.45906985e-01 6.21746123e-01 -3.78762305e-01 3.70674431e-01 -8.57884958e-02 -7.15203941e-01 8.87517810e-01 -1.29909480e+00 9.77987051e-01 -4.16898161e-01 4.69785839e-01 6.74603581e-02 -5.92974484e-01 7.37301588e-01 4.05471623e-01 3.96745503e-01 -5.80209553e-01 -2.65965849e-01 5.43994367e-01 1.69534877e-01 -4.45737302e-01 5.28765500e-01 -6.35343511e-03 4.09865901e-02 7.62727499e-01 -2.87845612e-01 -3.68845999e-01 5.77675700e-01 4.28758085e-01 1.20915651e+00 -2.57511914e-01 5.79312086e-01 -4.72736567e-01 -1.78557839e-02 3.87676865e-01 2.89203584e-01 9.31287110e-01 1.18432313e-01 8.30827594e-01 7.19685555e-01 1.82840079e-01 -1.02287292e+00 -7.23324716e-01 -4.84920263e-01 5.00164688e-01 -2.49997869e-01 -7.28064954e-01 -4.24174398e-01 -3.92462760e-01 -8.54413658e-02 5.15940189e-01 -6.51456237e-01 -1.64442099e-04 -1.55372843e-01 -8.17552805e-01 4.63942438e-01 2.63191134e-01 -6.18100017e-02 -1.06101537e+00 -1.10166013e+00 1.06668003e-01 1.49822116e-01 -6.48054302e-01 -4.14047986e-01 1.92177013e-01 -9.14089799e-01 -1.40758801e+00 -1.00227427e+00 -5.61015725e-01 5.88399649e-01 3.79443973e-01 1.04158819e+00 3.44677180e-01 8.64867121e-02 5.20829678e-01 -4.87509608e-01 -6.00182235e-01 -5.18928647e-01 4.75973636e-02 -6.35897815e-02 -9.82133806e-01 5.16646564e-01 -4.28850986e-02 -6.55878544e-01 1.75181910e-01 -1.19812572e+00 -3.22996289e-01 7.62675643e-01 5.11241734e-01 5.16595185e-01 1.29597276e-01 8.10910046e-01 -1.07463288e+00 1.17980087e+00 -6.08847857e-01 -7.60782182e-01 3.42275828e-01 -1.26107168e+00 -3.03201973e-01 1.15774632e-01 -3.66893947e-01 -5.96163809e-01 -3.15630138e-01 1.02411784e-01 5.04661463e-02 -5.35735004e-02 1.26853704e+00 4.64784503e-02 8.67518634e-02 5.44248521e-01 -3.35240662e-01 2.06832111e-01 -3.16356212e-01 -1.33581057e-01 9.18182671e-01 -3.17562699e-01 -3.38957429e-01 2.80425608e-01 1.75244644e-01 -8.83687362e-02 -4.84197676e-01 -3.41457799e-02 -7.70591855e-01 -2.33151466e-01 -8.07040483e-02 4.69202250e-01 -8.87330234e-01 -4.48756039e-01 -1.45489141e-01 -1.01827323e+00 -3.26000541e-01 2.08382076e-03 7.73867786e-01 1.37667414e-02 3.30803096e-01 -3.33931863e-01 -7.54456818e-01 -2.80721366e-01 -1.57794642e+00 8.40461612e-01 1.77951351e-01 -8.27917337e-01 -8.66419971e-01 3.16831619e-01 7.82923121e-03 2.38953516e-01 -7.85152987e-02 8.10681760e-01 -8.13318849e-01 -2.27992862e-01 -3.40270370e-01 -3.02498043e-01 -7.84116983e-02 6.19105101e-01 6.22700274e-01 -3.70668679e-01 -2.02853143e-01 -1.21426031e-01 1.27017826e-01 5.11525035e-01 9.06484127e-01 6.49448514e-01 -3.67662281e-01 -5.04754126e-01 -5.99422790e-02 1.35898924e+00 6.73123300e-01 6.60878003e-01 8.39100838e-01 2.22605705e-01 1.03134203e+00 8.86899590e-01 2.39975125e-01 3.48529696e-01 6.28659308e-01 -9.07860547e-02 -2.29126394e-01 -4.70466763e-02 -1.09907286e-02 1.89748436e-01 7.25510776e-01 1.27128854e-01 -1.54096767e-01 -1.32174397e+00 9.72592950e-01 -1.48132420e+00 -7.40281761e-01 -3.57002914e-01 2.72722697e+00 8.60848904e-01 4.85713601e-01 4.55118179e-01 -1.43059209e-01 6.51534438e-01 -1.91182852e-01 -1.35884613e-01 -8.55762541e-01 6.19057268e-02 3.44550967e-01 8.05251062e-01 3.89854491e-01 -3.91825944e-01 4.18151736e-01 7.30387735e+00 4.06024724e-01 -9.77034032e-01 -2.79240223e-04 5.23444831e-01 -2.95652539e-01 -6.37094200e-01 3.97440404e-01 -7.53146827e-01 4.93883252e-01 1.20285845e+00 -6.80155993e-01 -4.19127941e-02 2.77432144e-01 7.66974390e-01 -5.15719950e-01 -8.84059966e-01 2.01985195e-01 -2.53793150e-01 -1.57878160e+00 8.78446698e-02 3.57586861e-01 5.74529886e-01 -9.52617824e-02 -1.77117791e-02 -1.06234081e-01 3.47640514e-01 -9.53916609e-01 6.52637362e-01 1.17094010e-01 9.35572267e-01 -5.74622452e-01 8.85633647e-01 -9.47759748e-02 -9.13933456e-01 1.22226231e-01 -1.87605247e-01 -1.40624672e-01 1.07994378e-01 7.60787785e-01 -1.02053511e+00 8.15912962e-01 9.07319546e-01 5.26889920e-01 -7.51505256e-01 1.42982066e+00 3.24011892e-02 1.00856411e+00 -1.81417704e-01 -3.31988871e-01 9.35885161e-02 -1.92273781e-02 4.54732955e-01 1.45636463e+00 4.04202104e-01 1.11684509e-01 -2.89878190e-01 6.88447058e-01 1.46808296e-01 3.74705940e-01 -8.62513781e-01 -3.17267627e-01 8.45086098e-01 8.71360779e-01 -9.94221449e-01 -2.65750349e-01 -5.55585086e-01 2.06228849e-02 -2.16853827e-01 4.47323740e-01 -2.86610186e-01 -5.39153814e-01 -5.87403998e-02 4.63283300e-01 2.18679905e-01 5.95371649e-02 -6.15267396e-01 -5.29663444e-01 1.93110436e-01 -1.23157442e+00 5.94335735e-01 -9.01090860e-01 -7.22255230e-01 3.07051837e-01 6.04447305e-01 -1.19972801e+00 -1.79720491e-01 -2.99974710e-01 -2.93742895e-01 1.37093043e+00 -1.03969550e+00 -5.74582160e-01 1.96502417e-01 -9.10138041e-02 4.07078981e-01 2.76895881e-01 4.29531902e-01 1.21437281e-01 -4.15966451e-01 3.31346393e-01 -9.38762203e-02 -3.63788754e-01 1.00055552e+00 -1.01706862e+00 4.26957309e-02 8.83413136e-01 -2.88582474e-01 1.44712877e+00 6.53736889e-01 -1.30414462e+00 -1.17887008e+00 -5.80662310e-01 1.09962940e+00 -4.74869341e-01 8.24365914e-01 6.74343631e-02 -9.02199209e-01 3.27373058e-01 4.34601396e-01 -9.49659407e-01 7.40882635e-01 3.73140424e-01 8.48294497e-02 1.19867928e-01 -9.23343360e-01 7.96376705e-01 6.01130068e-01 -2.75317550e-01 -7.73354769e-01 1.24102801e-01 5.63588619e-01 -2.55042493e-01 -1.08159649e+00 5.72749436e-01 8.35204661e-01 -5.24445534e-01 6.90764427e-01 -2.19319776e-01 8.10792804e-01 -2.77088374e-01 8.87595713e-02 -1.01006544e+00 -3.94155920e-01 -1.75360888e-01 6.19794309e-01 1.30656850e+00 9.76988196e-01 -7.78232396e-01 3.90956432e-01 7.76312232e-01 -2.68357471e-02 -8.84909332e-01 -7.49429286e-01 -6.26924396e-01 3.20170164e-01 -1.89283103e-01 6.19341850e-01 9.45699930e-01 1.48710176e-01 1.32970691e-01 2.18352869e-01 1.37312651e-01 3.02000910e-01 -2.67206226e-02 4.67894733e-01 -1.13525271e+00 1.08954579e-01 -7.45658278e-01 1.84956312e-01 -2.71589160e-01 -4.14679021e-01 -5.57641804e-01 -1.55250788e-01 -2.22833323e+00 5.96598029e-01 -6.10173583e-01 -2.99101472e-01 5.80584109e-01 -3.61842185e-01 3.23742218e-02 7.81206787e-02 5.70377409e-01 -3.36991131e-01 -3.01076800e-01 9.43183541e-01 1.80853814e-01 -5.30569732e-01 2.99414191e-02 -1.24933946e+00 1.94309175e-01 6.91069961e-01 -8.66956532e-01 -4.78236705e-01 -3.79919447e-02 7.15283751e-01 -9.58975703e-02 3.48159015e-01 -4.22976017e-01 2.98523933e-01 -3.32284272e-01 7.15402141e-02 -6.97220027e-01 -3.21335196e-01 -6.00907683e-01 6.67437971e-01 6.08846903e-01 -4.23245013e-01 6.43171847e-01 7.08358884e-01 -9.05030519e-02 -1.67206153e-01 -9.77161407e-01 3.63660604e-02 -2.70993650e-01 -2.43982390e-01 -2.99395114e-01 -8.51480842e-01 -3.39983851e-02 5.15647888e-01 -3.78728777e-01 -4.61159825e-01 -3.07924896e-01 -9.75675583e-02 2.61523455e-01 1.03190613e+00 2.69586354e-01 4.65611130e-01 -9.23078418e-01 -6.32474899e-01 -4.66921389e-01 1.25962034e-01 -9.83944237e-02 -1.77915558e-01 1.07565379e+00 -4.24372792e-01 6.54709458e-01 1.03743397e-01 -1.69137433e-01 -1.47675729e+00 5.53410113e-01 -4.12656628e-02 -3.19270641e-01 -2.54365832e-01 2.97066897e-01 1.49500102e-01 -1.79277927e-01 7.33939707e-02 -4.33282971e-01 -3.53868842e-01 3.88530016e-01 4.74174887e-01 3.01559091e-01 3.69025320e-01 -2.33254209e-01 -7.95468688e-01 4.20127600e-01 -2.81690359e-01 -5.94747961e-01 1.36825669e+00 -8.63956809e-02 -3.03842276e-01 6.14462972e-01 1.02994597e+00 6.02309644e-01 -3.70780557e-01 5.26607811e-01 1.11105196e-01 -2.35414237e-01 1.56643763e-01 -9.66946363e-01 -4.96132404e-01 1.85986310e-01 2.86972344e-01 4.95019585e-01 1.22406876e+00 2.53658611e-02 -1.29906222e-01 -3.07505708e-02 6.93562329e-02 -8.67383718e-01 -4.59255695e-01 -5.94037175e-02 8.58546972e-01 -9.01961207e-01 6.51206791e-01 -4.12014335e-01 -4.97158945e-01 8.38763773e-01 1.83642417e-01 3.12748045e-01 6.36291087e-01 2.50617564e-01 -2.40426138e-02 -6.56292915e-01 -8.82187366e-01 8.35998431e-02 4.06373531e-01 1.76790550e-01 7.34747231e-01 -1.78405002e-01 -1.31058812e+00 6.01339281e-01 4.82565202e-02 6.16718411e-01 9.00000751e-01 1.19127798e+00 -4.00364920e-02 -1.22399294e+00 -7.32553840e-01 9.07748163e-01 -7.81036556e-01 -4.83978182e-01 -7.33589768e-01 1.15428889e+00 -2.15223879e-01 1.30505085e+00 -6.29318133e-02 -1.34812444e-01 5.63062727e-01 -3.17896813e-01 1.04151428e-01 -9.11407351e-01 -8.07949543e-01 2.94623613e-01 5.09398460e-01 -2.58191884e-01 -7.36725390e-01 -9.87092197e-01 -8.02263677e-01 -5.77009022e-02 -6.17884755e-01 5.69746912e-01 7.32522964e-01 6.60758436e-01 6.47682250e-01 4.11997765e-01 3.81355882e-01 -2.16135040e-01 -2.14803681e-01 -7.74650812e-01 -2.46996745e-01 -8.94416682e-03 1.31761417e-01 -7.95129895e-01 -5.22250652e-01 -1.09371990e-01]

[8.77573013305664, 8.44133472442627]

54cf93b4-8295-49ef-ad73-071f297b5ad5

language-models-are-multilingual-chain-of

2210.03057

https://arxiv.org/abs/2210.03057v1

https://arxiv.org/pdf/2210.03057v1.pdf

Language Models are Multilingual Chain-of-Thought Reasoners

We evaluate the reasoning abilities of large language models in multilingual settings. We introduce the Multilingual Grade School Math (MGSM) benchmark, by manually translating 250 grade-school math problems from the GSM8K dataset (Cobbe et al., 2021) into ten typologically diverse languages. We find that the ability to solve MGSM problems via chain-of-thought prompting emerges with increasing model scale, and that models have strikingly strong multilingual reasoning abilities, even in underrepresented languages such as Bengali and Swahili. Finally, we show that the multilingual reasoning abilities of language models extend to other tasks such as commonsense reasoning and word-in-context semantic judgment. The MGSM benchmark is publicly available at https://github.com/google-research/url-nlp.

['Jason Wei', 'Dipanjan Das', 'Denny Zhou', 'Sebastian Ruder', 'Yi Tay', 'Hyung Won Chung', 'Soroush Vosoughi', 'Suraj Srivats', 'Xuezhi Wang', 'Markus Freitag', 'Mirac Suzgun', 'Freda Shi']

2022-10-06

['gsm8k']

['natural-language-processing']

[-5.19801497e-01 2.42231503e-01 -3.13604511e-02 -2.16080382e-01 -9.00991976e-01 -1.18018055e+00 6.81584239e-01 3.24345440e-01 -3.43554497e-01 8.92984033e-01 3.93681258e-01 -8.79975617e-01 -4.98984873e-01 -1.25992608e+00 -8.54134977e-01 -4.20685783e-02 6.02766037e-01 6.76600039e-01 -1.94755286e-01 -5.49080670e-01 3.52670610e-01 3.44149210e-02 -9.22551930e-01 3.49482596e-01 1.57859123e+00 -9.24194045e-03 2.82313079e-01 5.92361152e-01 -1.67786375e-01 1.61970019e+00 -3.61759514e-01 -1.10809422e+00 4.24401835e-02 -2.71973431e-01 -1.37570083e+00 -7.31528401e-01 8.81541371e-01 1.85859343e-03 -4.88248527e-01 1.29713786e+00 2.59705901e-01 2.60261923e-01 5.76847136e-01 -8.43729436e-01 -1.61246789e+00 1.41507411e+00 -1.37787297e-01 4.89552051e-01 9.44245815e-01 2.16638774e-01 1.29080617e+00 -8.19660127e-01 7.74209976e-01 1.54437113e+00 8.11893880e-01 4.56417769e-01 -1.25577641e+00 -8.43013465e-01 1.52435780e-01 5.75545669e-01 -1.30254066e+00 -9.53857675e-02 4.72212523e-01 -7.13821888e-01 1.00668418e+00 2.41656810e-01 7.72219062e-01 9.44535375e-01 1.99445277e-01 8.21336091e-01 1.88283730e+00 -4.23574477e-01 -1.91640973e-01 -2.88966268e-01 4.98956919e-01 8.72259319e-01 3.35533768e-01 -2.11408272e-01 -5.89122295e-01 5.45487218e-02 6.29043281e-01 -5.39998651e-01 1.53432041e-01 2.72093654e-01 -1.39587188e+00 7.04030693e-01 3.29636037e-01 3.24237913e-01 2.57255808e-02 2.06962228e-01 -5.75869204e-03 8.10573459e-01 2.09535256e-01 8.59621823e-01 -5.53479731e-01 -9.54921246e-02 -6.27362847e-01 8.36020231e-01 5.85868359e-01 9.47286546e-01 5.07535756e-01 -7.08149821e-02 7.44059309e-02 1.19609904e+00 2.13988960e-01 9.28110182e-01 5.09402275e-01 -1.48852193e+00 9.95556653e-01 5.09941399e-01 -2.34853819e-01 -9.30179834e-01 -3.83271694e-01 -4.94519889e-01 -6.00172102e-01 -3.22062045e-01 9.01518881e-01 6.93833828e-02 -2.94770479e-01 2.18565249e+00 1.72497213e-01 -1.10011540e-01 4.16781604e-01 5.66142559e-01 1.12579393e+00 5.51511765e-01 4.93040681e-01 1.89339980e-01 1.36329782e+00 -9.38027978e-01 -3.92492145e-01 -4.53935325e-01 8.44311357e-01 -9.46485639e-01 1.32434893e+00 3.25424314e-01 -1.74908483e+00 -6.01293564e-01 -5.96424341e-01 -7.45761931e-01 -6.39790118e-01 -1.85213447e-01 8.03999603e-01 4.30619091e-01 -1.34068060e+00 1.94372699e-01 -4.71662641e-01 -2.77554750e-01 2.76846468e-01 -1.69577762e-01 -2.73100764e-01 -5.76725900e-01 -1.77853429e+00 1.24365246e+00 4.79127735e-01 -3.15171927e-01 -7.30795026e-01 -1.06801260e+00 -1.01857710e+00 -9.00574997e-02 1.04701310e-01 -7.30562985e-01 1.50732887e+00 -6.29478574e-01 -1.21290207e+00 1.36036777e+00 6.30204082e-02 -1.07739300e-01 5.46087086e-01 -2.07037359e-01 -3.59019548e-01 -1.52157843e-02 6.82327509e-01 4.83229965e-01 -1.07122354e-01 -6.59409761e-01 -4.94336247e-01 -2.23527119e-01 5.97507775e-01 3.93441856e-01 -2.19280403e-02 3.76256704e-01 1.89718470e-01 -9.64119315e-01 3.88084501e-01 -9.30704117e-01 1.75172120e-01 -5.88771582e-01 -2.37512246e-01 -6.75325215e-01 -3.39458048e-01 -1.19898725e+00 1.17084742e+00 -1.46814811e+00 3.79650533e-01 1.38548106e-01 2.71363556e-01 -2.28780836e-01 -1.76760688e-01 4.72129375e-01 -2.69899130e-01 2.75946796e-01 4.53153700e-02 1.64820388e-01 7.28120327e-01 8.05925727e-02 -4.08312052e-01 6.66327178e-02 -5.61505705e-02 1.54439092e+00 -1.26006699e+00 -4.57354367e-01 -2.22185608e-02 -1.57096654e-01 -8.13993931e-01 -2.19980359e-01 -2.23016247e-01 3.63299876e-01 -1.57425597e-01 6.49150312e-01 4.36587751e-01 -3.58857453e-01 4.30924892e-01 5.56635320e-01 -8.44807923e-02 8.86790276e-01 -8.42938602e-01 1.73002803e+00 -5.33549428e-01 7.67090738e-01 -3.01938355e-01 -6.66371226e-01 3.42464209e-01 1.81424752e-01 -3.43626171e-01 -9.83361304e-01 -1.42818838e-01 3.88518214e-01 6.12675369e-01 -3.76263350e-01 3.54434013e-01 -9.71087813e-02 -4.96861547e-01 6.55457914e-01 7.93170109e-02 -5.37127614e-01 6.49057448e-01 4.56799150e-01 8.14271808e-01 5.75791299e-02 2.90976256e-01 -9.04715300e-01 7.51118660e-01 1.68899104e-01 4.86103773e-01 8.93324196e-01 8.12455267e-02 -1.58006772e-01 4.56502348e-01 -3.02743942e-01 -9.00906742e-01 -1.51620269e+00 -2.52419382e-01 1.47761154e+00 -3.43264818e-01 -4.06209469e-01 -7.55604386e-01 -2.81056017e-01 1.67553648e-01 1.26894939e+00 -2.91578382e-01 3.23844887e-02 -7.03261256e-01 -5.33296585e-01 1.11291647e+00 5.75819552e-01 5.70918441e-01 -9.22999680e-01 1.96594417e-01 1.23868942e-01 -4.84015107e-01 -1.24359334e+00 9.36441198e-02 -3.67726684e-01 -5.52983463e-01 -9.26491439e-01 -4.70541388e-01 -9.93518233e-01 4.27126616e-01 -3.18869948e-01 1.68324375e+00 3.13882709e-01 5.22874035e-02 6.01675332e-01 -7.46431500e-02 -3.20077300e-01 -6.66684210e-01 6.28894661e-03 2.35026687e-01 -1.19211209e+00 6.77666605e-01 -5.72212398e-01 3.21493447e-02 -1.93057224e-01 -4.61333632e-01 4.58943188e-01 1.21121079e-01 5.14414907e-01 2.44141385e-01 2.21764259e-02 5.55093229e-01 -7.41875947e-01 7.11132944e-01 -6.45077586e-01 -6.69644058e-01 5.20393431e-01 -3.83411348e-01 -7.43914098e-02 7.27828503e-01 -3.27363849e-01 -1.07217908e+00 -8.55664074e-01 -1.50589198e-01 3.46008241e-01 -1.72010258e-01 6.67861760e-01 -1.37530919e-02 -1.06568977e-01 5.93974054e-01 4.95744020e-01 -6.47208691e-01 -3.63668859e-01 4.78462398e-01 3.13758969e-01 5.92867076e-01 -1.75454438e+00 9.75027919e-01 -2.27007493e-01 -8.90488774e-02 -7.13218391e-01 -1.16832328e+00 1.83775052e-01 -3.76360953e-01 -1.71156853e-01 9.79446232e-01 -1.37571990e+00 -1.05805469e+00 6.82766140e-01 -1.11278987e+00 -9.04496789e-01 1.58712462e-01 4.94574249e-01 -6.33300662e-01 1.74276978e-01 -1.38997877e+00 -2.69196898e-01 -9.76212695e-02 -1.01949513e+00 3.82396162e-01 3.01619172e-01 -5.68696797e-01 -1.54932201e+00 1.18245319e-01 1.13022339e+00 3.48812565e-02 -1.88433439e-01 1.78860044e+00 -7.07765043e-01 -5.11912286e-01 4.68111813e-01 -2.17877731e-01 1.38473630e-01 -3.67381573e-01 -2.36180767e-01 -5.37082076e-01 8.08765292e-02 -2.51227528e-01 -9.13226306e-01 6.49546027e-01 7.03219995e-02 1.01485288e+00 -2.62216359e-01 3.80685240e-01 4.75207239e-01 1.40953076e+00 -6.59281984e-02 4.78572994e-01 4.38018173e-01 8.16277385e-01 4.32737112e-01 1.99797153e-01 -3.13963205e-01 1.48682272e+00 1.58817545e-01 -5.66643775e-01 5.11834979e-01 -2.64256239e-01 -5.78540504e-01 4.21494484e-01 1.50386715e+00 -2.77563244e-01 7.26741105e-02 -1.94530678e+00 7.52888322e-01 -1.70284033e+00 -1.08865070e+00 -3.08569282e-01 1.69868529e+00 1.58167040e+00 -2.32672304e-01 -2.46980786e-01 -1.96890756e-01 4.45992321e-01 -5.93300536e-02 -1.05538458e-01 -6.15976393e-01 -4.36548561e-01 5.63795805e-01 8.82471278e-02 1.13383222e+00 -6.33686900e-01 1.59259415e+00 5.99002457e+00 1.09855151e+00 -4.73514318e-01 2.64802277e-01 5.59686661e-01 7.07943067e-02 -8.14702749e-01 -1.49748921e-01 -6.22344971e-01 3.18488240e-01 1.02140319e+00 -3.22370499e-01 8.94843936e-01 1.55097008e-01 -9.96742919e-02 -1.36494726e-01 -1.09881830e+00 7.02960491e-01 9.57144517e-03 -1.13887310e+00 4.37742770e-02 -3.04268569e-01 1.32495105e+00 -1.65586197e-03 1.52154028e-01 6.62221134e-01 1.15119374e+00 -1.29620945e+00 1.04460859e+00 6.91801846e-01 5.49166441e-01 -7.43110180e-01 8.28193873e-02 4.05411929e-01 -9.27625239e-01 -6.34068772e-02 -3.35115731e-01 -7.64484525e-01 -3.29263270e-01 2.45757490e-01 -3.03479224e-01 2.45239511e-01 5.02155364e-01 5.43898344e-01 -1.01673329e+00 3.46495420e-01 -1.03335893e+00 8.02786410e-01 -8.76275301e-02 -5.99467605e-02 1.64794743e-01 -2.95660943e-01 3.53307664e-01 9.73682165e-01 1.21244878e-01 4.46725816e-01 1.88056469e-01 1.18218720e+00 -1.65222302e-01 2.15932369e-01 -4.02967215e-01 -3.27863932e-01 5.72748899e-01 8.96568000e-01 -5.65640330e-01 -5.36598921e-01 -5.28029919e-01 6.48891270e-01 9.38172102e-01 4.39769119e-01 -7.69809127e-01 -9.75235626e-02 6.52069271e-01 -1.07255355e-01 -4.26209748e-01 -4.56940234e-01 -5.75710177e-01 -1.50779927e+00 -2.09890619e-01 -1.26865447e+00 4.64770675e-01 -1.37477934e+00 -1.58091640e+00 -2.64997602e-01 -4.20931838e-02 -4.25147921e-01 -3.48124683e-01 -9.28495407e-01 -3.18549454e-01 1.07233822e+00 -1.14264238e+00 -1.25527322e+00 7.29457289e-02 7.54706681e-01 4.12810951e-01 -2.62094289e-01 8.35036576e-01 2.92397767e-01 -4.07074660e-01 6.01096153e-01 6.54198676e-02 7.31459558e-01 6.06912792e-01 -1.64375544e+00 7.16653883e-01 8.66998732e-01 2.60487795e-01 1.11821914e+00 4.29899693e-01 -8.84575009e-01 -1.20430434e+00 -8.60560358e-01 1.61760092e+00 -1.01784313e+00 1.40248668e+00 -3.61867398e-01 -6.81237280e-01 1.36379468e+00 5.59396267e-01 -4.83217180e-01 8.19098353e-01 4.57364798e-01 -7.07718015e-01 3.77311468e-01 -9.10699189e-01 1.02919364e+00 1.40707576e+00 -1.06671107e+00 -1.25111210e+00 8.47034037e-01 6.67499065e-01 -7.36958504e-01 -1.31610894e+00 6.44375980e-02 4.19457734e-01 -6.36477947e-01 1.10894978e+00 -1.15270615e+00 1.23410869e+00 -1.48779899e-02 -2.20590845e-01 -1.47488832e+00 -7.86072612e-01 -2.70136237e-01 1.13916710e-01 1.27460825e+00 5.69670737e-01 -8.10493588e-01 1.02641389e-01 5.87050915e-01 1.60953090e-01 -3.72663289e-01 -7.03805089e-01 -6.98561430e-01 1.28654146e+00 -7.85165846e-01 6.70928478e-01 1.71990299e+00 2.61981487e-01 2.89707124e-01 4.70479906e-01 2.99409956e-01 7.32626200e-01 3.19704175e-01 2.92978346e-01 -1.19295406e+00 -3.12100321e-01 -7.11228251e-01 -2.42599696e-01 -4.09963757e-01 1.04976964e+00 -1.74649429e+00 -4.99976665e-01 -1.73770106e+00 2.98373759e-01 -4.25134271e-01 -1.63042709e-01 5.97568452e-01 -5.18691301e-01 3.48742574e-01 4.48118985e-01 -2.92205334e-01 -8.75033021e-01 -1.22002155e-01 1.42406845e+00 -1.12372801e-01 4.52622414e-01 -6.48695171e-01 -8.37291062e-01 1.05566728e+00 1.07540715e+00 -3.16668183e-01 -3.44954938e-01 -1.00388730e+00 1.09311342e+00 8.29384997e-02 5.98842502e-01 -7.82748163e-01 2.53707379e-01 -6.60025120e-01 7.46559560e-01 -1.17183421e-02 -1.83816642e-01 -2.01873243e-01 -3.69382873e-02 5.04864514e-01 -5.77247381e-01 5.71685255e-01 3.15078795e-01 -3.09306920e-01 -2.42838915e-02 -1.83488488e-01 5.77546358e-01 -4.68148053e-01 -7.93929040e-01 -2.61331350e-01 -4.67154145e-01 1.07266402e+00 7.24547267e-01 3.10822517e-01 -8.87110472e-01 -1.69200048e-01 -7.49468803e-01 4.34110850e-01 4.01184082e-01 4.71235573e-01 8.81157666e-02 -1.67104101e+00 -1.09790993e+00 -3.94292511e-02 1.35805272e-03 -2.79469788e-01 3.30023527e-01 8.83780420e-01 -1.00594807e+00 8.02725196e-01 -8.93050060e-02 -4.02270071e-03 -7.34240830e-01 4.12947178e-01 4.01259601e-01 -1.97923437e-01 -1.22925676e-01 1.14299428e+00 5.33160828e-02 -1.26391113e+00 -4.87597525e-01 -4.00526702e-01 -8.19682255e-02 -9.03406739e-02 4.46665764e-01 6.53246105e-01 -3.39453697e-01 -5.63967109e-01 -2.27307290e-01 7.51953363e-01 1.77470952e-01 -1.96978495e-01 1.10798264e+00 3.09520662e-02 -8.78652930e-01 5.70344090e-01 6.37621403e-01 6.04948044e-01 -1.62371635e-01 -3.54060709e-01 6.22228272e-02 -7.54327923e-02 -2.74108797e-01 -1.17990017e+00 -6.93960547e-01 5.50849915e-01 -2.41841063e-01 -8.63804296e-02 6.68437660e-01 1.11054480e-01 4.33889717e-01 7.54901826e-01 5.04705727e-01 -1.33804572e+00 -5.31514406e-01 1.37525594e+00 7.64476717e-01 -9.72770572e-01 -1.27575755e-01 -2.53096730e-01 -4.21362668e-01 7.34612882e-01 7.81202972e-01 -2.28830591e-01 3.81437689e-01 2.10805088e-01 9.82328281e-02 -1.27331959e-02 -9.24472690e-01 -5.77918105e-02 3.12516838e-01 3.20308864e-01 8.96535158e-01 6.15841210e-01 -6.62787676e-01 1.02580059e+00 -1.35139716e+00 -2.22132400e-01 5.12943685e-01 5.78590572e-01 -1.31360620e-01 -8.00917685e-01 -6.80787981e-01 2.24586263e-01 -5.83686650e-01 -7.99603641e-01 -5.93410730e-01 7.26618171e-01 2.43185073e-01 8.31241906e-01 6.06583767e-02 3.16358507e-02 -5.08934148e-02 6.24558210e-01 9.98554349e-01 -5.60835838e-01 -6.73933089e-01 -6.78160250e-01 2.06823125e-01 -3.66587341e-01 -1.69472530e-01 -6.57806218e-01 -1.29457188e+00 -9.20924544e-01 4.68824238e-01 -1.38951512e-02 -4.85875010e-02 1.22846353e+00 -2.45861709e-01 5.61366975e-01 -2.99433470e-01 5.20014092e-02 -3.82587284e-01 -8.90953362e-01 -4.81811106e-01 2.81044781e-01 -1.58738438e-02 -2.95432985e-01 -3.10204268e-01 -5.09347618e-02]

[9.784581184387207, 7.395485877990723]

d204017d-1f7d-441c-ac26-36e4b3f69254

unsupervised-extraction-of-market-moving

2001.09466

https://arxiv.org/abs/2001.09466v3

https://arxiv.org/pdf/2001.09466v3.pdf

From Stock Prediction to Financial Relevance: Repurposing Attention Weights to Assess News Relevance Without Manual Annotations

We present a method to automatically identify financially relevant news using stock price movements and news headlines as input. The method repurposes the attention weights of a neural network initially trained to predict stock prices to assign a relevance score to each headline, eliminating the need for manually labeled training data. Our experiments on the four most relevant US stock indices and 1.5M news headlines show that the method ranks relevant news highly, positively correlated with the accuracy of the initial stock price prediction task.

['Johannes Hoffart', 'Luciano Del Corro']

2020-01-26

https://aclanthology.org/2021.econlp-1.6

https://aclanthology.org/2021.econlp-1.6.pdf

emnlp-econlp-2021-11

['stock-price-prediction', 'stock-prediction']

['time-series', 'time-series']

[-5.85151076e-01 3.17380250e-01 -7.41134465e-01 -3.06922793e-01 -7.43549645e-01 -7.58640468e-01 8.24696481e-01 3.53388280e-01 -6.04920566e-01 9.19728398e-01 1.14438426e+00 -4.31822121e-01 -2.18824148e-02 -1.07906294e+00 -6.66293204e-01 -8.31729472e-02 -1.53533667e-01 6.18371129e-01 1.55670047e-01 -4.57256496e-01 9.61243510e-01 2.80708134e-01 -1.18561351e+00 -6.40928149e-02 4.95889902e-01 1.55175817e+00 6.84518591e-02 3.92949373e-01 -2.43360445e-01 1.40374315e+00 -9.34648752e-01 -8.13381493e-01 5.30653775e-01 -1.72740787e-01 -7.54008353e-01 -3.56076658e-01 3.71124268e-01 -4.22749162e-01 -2.07350880e-01 1.24164438e+00 -2.87055299e-02 1.60607874e-01 7.30912030e-01 -6.46806955e-01 -1.27439749e+00 1.24074626e+00 -3.59473079e-01 1.26720715e+00 6.41227514e-02 -5.62137246e-01 1.86782598e+00 -9.10499215e-01 8.53825986e-01 5.83177328e-01 6.92181528e-01 -2.09769726e-01 -7.03218400e-01 -4.68013167e-01 3.12690735e-01 -7.47092487e-03 -6.20157957e-01 -2.39157453e-01 7.11869597e-01 -6.45924330e-01 1.06193388e+00 2.47435495e-01 9.88354743e-01 5.57686925e-01 6.32786691e-01 4.95343328e-01 6.81114674e-01 -4.38646935e-02 3.46241146e-01 2.47603133e-01 5.25315106e-01 2.09320672e-02 5.92519939e-01 1.04060337e-01 -5.47989666e-01 -3.63136142e-01 7.39011765e-01 -1.06365584e-01 -1.76889569e-01 5.41609704e-01 -1.00544512e+00 1.25994146e+00 7.47016132e-01 4.92553741e-01 -1.06292415e+00 -1.62694722e-01 2.48641700e-01 6.66393280e-01 1.07039583e+00 1.21842897e+00 -5.78580499e-01 2.94905286e-02 -1.02551866e+00 2.52822578e-01 8.83981526e-01 3.51222992e-01 1.71002686e-01 4.78465199e-01 -8.71821195e-02 4.33659971e-01 1.72673643e-01 3.87576371e-01 8.51810813e-01 -6.90333307e-01 4.19037640e-01 4.32846993e-01 4.40624088e-01 -1.22717404e+00 -5.52619636e-01 -1.06417191e+00 -5.10489382e-02 9.20118317e-02 -7.26810470e-02 -3.45626831e-01 -4.22419906e-01 1.09527206e+00 -3.14644843e-01 -8.80791247e-02 2.57108212e-01 7.63678670e-01 5.62714994e-01 1.04608428e+00 -9.89906117e-02 -4.27798092e-01 1.09645534e+00 -8.48289669e-01 -5.51968873e-01 -2.26345822e-01 -7.29272962e-02 -7.57022798e-01 4.84946370e-01 1.94740430e-01 -1.34172881e+00 -3.59204024e-01 -1.02462101e+00 3.83065194e-01 -4.41279680e-01 -2.04241157e-01 3.48054439e-01 -1.94890112e-01 -1.01278651e+00 8.25281739e-01 -2.50936210e-01 4.32290345e-01 1.60419002e-01 1.70567691e-01 4.53178734e-02 8.90654325e-01 -1.57606924e+00 1.22241652e+00 5.37834764e-01 -3.79254222e-01 -2.76326060e-01 -6.37297809e-01 -6.43828154e-01 4.27945852e-01 -1.80087790e-01 -2.50842452e-01 1.59900331e+00 -1.10712790e+00 -1.20363438e+00 4.92316753e-01 4.25400168e-01 -1.10161471e+00 4.07098323e-01 -4.44166094e-01 -6.36355281e-01 2.76525319e-01 5.11830211e-01 2.44524077e-01 5.78490376e-01 -5.55236876e-01 -1.11705565e+00 1.36931986e-01 -9.14736837e-02 2.00972676e-01 -3.08678210e-01 5.23472011e-01 6.21913671e-02 -1.12970090e+00 2.24715732e-02 -3.26849967e-01 -1.72203943e-01 -9.47294474e-01 -2.08774745e-01 -3.19918931e-01 4.65937816e-02 -1.03289771e+00 1.22655153e+00 -1.90874362e+00 -1.30172476e-01 4.29529339e-01 -3.43144089e-02 -4.10413206e-01 8.15843120e-02 5.03773466e-02 -3.58551443e-01 1.51215613e-01 4.26389515e-01 3.24806511e-01 2.05746993e-01 -4.40149695e-01 -9.63777363e-01 1.65217251e-01 2.63535887e-01 1.00637567e+00 -7.00714290e-01 9.51622203e-02 -3.26006114e-01 -7.58367702e-02 -3.90531689e-01 2.61807386e-02 -2.56975472e-01 -3.75880957e-01 -2.81574517e-01 2.70428270e-01 8.95206630e-02 -5.53125799e-01 -3.20066601e-01 1.46406770e-01 -4.86869276e-01 1.08220708e+00 -6.05134726e-01 2.51305729e-01 -1.81848404e-03 8.35091650e-01 -4.87576604e-01 -6.58216178e-01 1.21218872e+00 2.78246999e-01 3.62953603e-01 -7.70498574e-01 3.28609020e-01 3.32690746e-01 -1.17904089e-01 -1.76005065e-01 9.04932737e-01 -3.72878909e-01 -3.93090934e-01 7.85514057e-01 -1.94567263e-01 2.70027757e-01 1.85581461e-01 2.15482563e-01 6.09240770e-01 -3.70598376e-01 4.40669924e-01 -4.40176934e-01 2.13019416e-01 1.94572866e-01 5.15752196e-01 4.61818933e-01 2.71020606e-02 4.83668983e-01 6.94963515e-01 -8.49505007e-01 -1.04789340e+00 -5.47248483e-01 -8.15600380e-02 1.30949831e+00 -3.53995919e-01 4.78626937e-02 -3.87742579e-01 -4.67804819e-01 2.65578985e-01 1.11776686e+00 -9.33189571e-01 2.09139019e-01 -3.21126848e-01 -8.20405245e-01 -1.41386122e-01 6.40454233e-01 1.02292098e-01 -1.32930803e+00 -6.83831930e-01 4.88309473e-01 1.09142818e-01 -6.55311763e-01 -4.51288521e-01 3.49020839e-01 -7.06509709e-01 -6.98045790e-01 -1.16699314e+00 -8.73464406e-01 5.81523597e-01 -3.39760333e-01 1.23684740e+00 -2.77537227e-01 5.29133677e-01 -1.76834449e-01 -2.36752391e-01 -6.94319010e-01 -3.00473243e-01 3.96369100e-01 9.66450050e-02 -1.15895405e-01 4.59031135e-01 -1.53969437e-01 -3.72843057e-01 -1.04092024e-01 -3.94761622e-01 -2.60420412e-01 4.37234968e-01 7.20561504e-01 2.32161641e-01 8.69536400e-03 1.26822698e+00 -7.59486973e-01 1.04032886e+00 -8.06125462e-01 -1.17184508e+00 -2.28492240e-03 -8.53484273e-01 1.24266654e-01 3.98494273e-01 -1.39652982e-01 -8.40476096e-01 -3.07309210e-01 -1.44574508e-01 1.89246297e-01 4.35042560e-01 1.35891485e+00 8.32092762e-01 3.50661099e-01 6.50300145e-01 1.42398819e-01 -3.11280519e-01 -4.45004284e-01 -4.99247648e-02 3.52131814e-01 7.59452045e-01 1.85762674e-01 8.02857041e-01 -3.74029912e-02 -6.47796571e-01 -3.14610869e-01 -1.22138143e+00 -2.28722855e-01 -2.96802759e-01 -2.58991063e-01 6.68150008e-01 -9.40837383e-01 -3.10377568e-01 2.57536918e-01 -9.44653928e-01 5.07959090e-02 -5.72766721e-01 9.90111828e-01 -2.42823988e-01 -2.30525315e-01 -1.07097816e+00 -6.90573454e-01 -7.41844773e-01 -5.78931332e-01 1.83169067e-01 2.83654332e-01 -6.24978244e-01 -1.06176686e+00 3.37005734e-01 -1.03705097e-02 7.66758263e-01 2.19517350e-01 6.92188442e-01 -1.33217728e+00 -2.47823909e-01 -7.92018592e-01 -1.80720747e-01 8.03324208e-02 8.11811723e-03 -1.22039579e-02 -7.08194733e-01 2.10015178e-01 1.84814230e-01 -1.92263618e-01 1.04121029e+00 8.37961555e-01 6.90079257e-02 -7.81644106e-01 2.15585500e-01 1.02421381e-01 1.14491963e+00 3.35986823e-01 1.72734246e-01 1.15795815e+00 2.70217601e-02 5.50248444e-01 2.90633976e-01 5.12089133e-01 2.47337431e-01 6.43284917e-02 6.89754114e-02 3.43795232e-02 5.37733674e-01 -1.75271153e-01 6.50141537e-01 9.22172248e-01 -2.22389176e-01 2.08670914e-01 -7.50096440e-01 5.47350168e-01 -1.52323246e+00 -1.31036758e+00 2.91334599e-01 1.84102690e+00 8.11834872e-01 9.36426401e-01 4.59727138e-01 -6.36629015e-02 7.09153771e-01 2.82660395e-01 -4.87287611e-01 -2.26506531e-01 -3.31015706e-01 -7.76056349e-02 8.73519421e-01 7.03994751e-01 -1.19572341e+00 6.43741071e-01 8.11254406e+00 4.88869809e-02 -1.03108037e+00 -2.59711057e-01 9.83154595e-01 -2.97085255e-01 -5.80211580e-01 -4.78773713e-01 -8.38797510e-01 6.23658299e-01 1.20678318e+00 -9.63801742e-01 -1.12708576e-01 1.21990252e+00 1.95703581e-01 2.02733353e-02 -5.75270414e-01 2.33034089e-01 -1.27114225e-02 -2.09440041e+00 9.90157351e-02 1.94868967e-02 1.01253545e+00 4.76775885e-01 3.33743095e-01 1.46151900e-01 7.51741171e-01 -3.98211658e-01 1.23436701e+00 7.09032774e-01 1.97697822e-02 -9.45496380e-01 1.26201284e+00 1.96931642e-02 -7.56836474e-01 -2.91136622e-01 -5.07536769e-01 -4.99797374e-01 3.93920392e-01 5.32361746e-01 -8.05385113e-01 2.86054797e-02 7.00156391e-01 8.21228802e-01 -4.87512767e-01 1.09816790e+00 -2.36930847e-01 6.68307185e-01 -1.35693282e-01 -2.79330015e-01 6.10302806e-01 -4.28165533e-02 3.16599756e-01 8.64782274e-01 5.09622812e-01 2.47008055e-01 -2.49729559e-01 6.91659212e-01 -3.09198231e-01 3.58749151e-01 -4.32605445e-01 -2.09903300e-01 3.58042717e-01 6.18456900e-01 -8.36936712e-01 -7.51504540e-01 -5.93527257e-01 4.57751095e-01 1.72657087e-01 1.39686733e-01 -4.67226893e-01 -4.92076486e-01 3.28153163e-01 1.96705282e-01 5.65956473e-01 3.15982997e-01 -4.73532021e-01 -1.04271197e+00 -3.10810894e-01 -2.87534624e-01 6.15610838e-01 -1.01853430e+00 -1.58659434e+00 9.86944437e-01 -2.80470014e-01 -1.12717426e+00 -5.17950952e-01 -4.76984739e-01 -1.04615748e+00 9.70204592e-01 -1.71948564e+00 -7.08853304e-02 4.45268095e-01 -2.08644755e-02 4.58297819e-01 -7.23385632e-01 4.29315656e-01 -8.16318542e-02 -5.14703989e-01 1.49034426e-01 3.55709672e-01 5.54342806e-01 3.16368610e-01 -1.37892675e+00 1.20186245e+00 7.10075200e-01 7.15068653e-02 4.11215067e-01 7.64522851e-01 -9.76475954e-01 -2.55065292e-01 -7.82977581e-01 1.57653773e+00 -3.40480268e-01 1.40610981e+00 3.95614445e-01 -8.06417704e-01 8.04292321e-01 6.50502920e-01 -6.82446539e-01 7.80005217e-01 -8.47913884e-03 -2.97421694e-01 2.50619128e-02 -7.87331641e-01 2.26389214e-01 1.32234395e-01 -3.46040428e-01 -1.44879913e+00 5.11095107e-01 9.78087485e-01 -8.06826428e-02 -1.04607105e+00 -1.91062629e-01 4.04688567e-01 -6.08426213e-01 8.40528965e-01 -7.96088815e-01 7.84425974e-01 1.40051991e-01 9.07090306e-02 -1.35807419e+00 -9.85716581e-01 -3.39461237e-01 7.13564232e-02 8.18312824e-01 1.27260530e+00 -9.74352598e-01 6.57785714e-01 9.33943808e-01 -3.95177230e-02 -3.48212242e-01 -7.74926603e-01 -6.18135631e-01 5.36686815e-02 7.53327319e-03 8.25432897e-01 1.15366018e+00 5.10287523e-01 4.67404872e-01 -2.93591738e-01 5.14924899e-02 1.84245646e-01 5.53388417e-01 -3.46394069e-02 -1.69552159e+00 -2.83665687e-01 -1.12209594e+00 -2.68691421e-01 -6.50260687e-01 2.02111632e-01 -7.11066186e-01 -1.81427628e-01 -1.46907163e+00 -1.07123539e-01 -4.23112400e-02 -7.92309880e-01 3.10136110e-01 -1.28774300e-01 2.43382588e-01 1.61460757e-01 6.24131799e-01 -3.22310477e-01 1.69817716e-01 7.00594425e-01 -1.50334150e-01 -2.35786453e-01 4.00407791e-01 -1.15762997e+00 7.84159720e-01 1.09367657e+00 -3.32981408e-01 -7.19157383e-02 -1.06072024e-01 9.41429377e-01 2.43081339e-02 -1.75667375e-01 -8.37899327e-01 1.68229997e-01 -3.43147248e-01 7.41022944e-01 -8.31172585e-01 -2.00397447e-01 -4.85347509e-01 9.26996619e-02 5.68848908e-01 -8.81092131e-01 7.74451315e-01 9.37717035e-02 3.23158503e-01 -5.37011981e-01 -6.23628199e-01 5.55331409e-01 -1.76589683e-01 -6.65320277e-01 -8.51177797e-02 -5.84582984e-01 6.33762628e-02 7.97497094e-01 1.25853345e-01 -4.64601070e-01 -8.17192137e-01 -7.44506538e-01 3.37303691e-02 2.29633078e-01 4.44922686e-01 4.73079592e-01 -1.38832188e+00 -1.05561399e+00 1.68498725e-01 -2.38948151e-01 -6.79261863e-01 -4.36948746e-01 2.42120922e-01 -7.03599393e-01 7.12529004e-01 -3.94391567e-01 4.60791230e-01 -4.91479367e-01 6.52699888e-01 1.96285665e-01 -2.44740784e-01 -7.10811317e-01 1.03011370e+00 -1.48788825e-01 4.31868970e-01 1.49514869e-01 -6.45116568e-01 -9.57759559e-01 8.74929070e-01 1.03425920e+00 1.96106732e-01 4.11118902e-02 -8.23926091e-01 8.39417204e-02 3.03633809e-01 -2.99886137e-01 -4.43310976e-01 1.91554320e+00 9.43538472e-02 1.44134790e-01 5.72512746e-01 9.00769472e-01 1.80191752e-02 -1.09426332e+00 -3.45937818e-01 6.92401588e-01 -1.27250776e-01 4.17365879e-01 -7.53445745e-01 -1.22702885e+00 5.45962108e-03 -1.70479923e-01 9.17419136e-01 6.28621519e-01 1.72473937e-02 8.63019109e-01 4.30734336e-01 -1.84216015e-02 -1.46068537e+00 -1.66146979e-01 7.11825192e-01 1.16617298e+00 -9.46274698e-01 2.29870230e-01 3.68002623e-01 -9.20813680e-01 1.08068848e+00 1.01494081e-01 -5.18498778e-01 9.76068318e-01 1.94389194e-01 3.69967580e-01 -4.35233474e-01 -7.73716271e-01 1.26554295e-01 4.78262067e-01 -7.11211786e-02 3.17413837e-01 -6.59505278e-02 -2.11408809e-01 8.88166785e-01 -8.68772686e-01 -2.93520689e-01 8.31733167e-01 6.61090136e-01 -1.06804538e+00 -1.94301382e-01 -4.51981455e-01 9.13186550e-01 -1.03336453e+00 -3.69488299e-01 -5.40770113e-01 4.90405589e-01 -6.27049685e-01 3.72692198e-01 7.07031250e-01 -1.96519777e-01 4.03436065e-01 3.21093291e-01 -6.68217182e-01 -4.53028172e-01 -9.84714270e-01 3.28404516e-01 1.98153570e-01 -8.17878637e-03 -2.63601869e-01 -9.07603979e-01 -9.91433918e-01 -3.88245195e-01 -2.32362628e-01 6.41126275e-01 2.18756825e-01 6.46002948e-01 2.71351993e-01 3.70455623e-01 9.35716569e-01 -8.59416664e-01 -7.60355711e-01 -9.76632833e-01 -7.78677523e-01 3.48813772e-01 6.08396351e-01 -3.90720606e-01 -6.43300474e-01 1.02012847e-02]

[4.419174671173096, 4.291524887084961]

4334d5f1-0850-4435-a1b6-422579e681a7

on-the-convergence-of-decentralized-federated

2303.10695

https://arxiv.org/abs/2303.10695v1

https://arxiv.org/pdf/2303.10695v1.pdf

On the Convergence of Decentralized Federated Learning Under Imperfect Information Sharing

Decentralized learning and optimization is a central problem in control that encompasses several existing and emerging applications, such as federated learning. While there exists a vast literature on this topic and most methods centered around the celebrated average-consensus paradigm, less attention has been devoted to scenarios where the communication between the agents may be imperfect. To this end, this paper presents three different algorithms of Decentralized Federated Learning (DFL) in the presence of imperfect information sharing modeled as noisy communication channels. The first algorithm, Federated Noisy Decentralized Learning (FedNDL1), comes from the literature, where the noise is added to their parameters to simulate the scenario of the presence of noisy communication channels. This algorithm shares parameters to form a consensus with the clients based on a communication graph topology through a noisy communication channel. The proposed second algorithm (FedNDL2) is similar to the first algorithm but with added noise to the parameters, and it performs the gossip averaging before the gradient optimization. The proposed third algorithm (FedNDL3), on the other hand, shares the gradients through noisy communication channels instead of the parameters. Theoretical and experimental results demonstrate that under imperfect information sharing, the third scheme that mixes gradients is more robust in the presence of a noisy channel compared with the algorithms from the literature that mix the parameters.

['Stanislaw H /. Zak', 'Abolfazl Hashemi', 'Antesh Upadhyay', 'Vishnu Pandi Chellapandi']

2023-03-19

['distributed-optimization']

['methodology']

[-4.70415324e-01 -7.28362352e-02 6.14439622e-02 -8.44763443e-02 -4.33296323e-01 -5.20521343e-01 7.47949779e-01 1.64126530e-01 -4.51234996e-01 9.36711311e-01 1.54288381e-01 -5.66971265e-02 -2.75482357e-01 -6.71750426e-01 -7.16077387e-01 -1.32437348e+00 -5.72053134e-01 5.01503125e-02 -6.64481297e-02 -2.49583006e-01 8.88815597e-02 -1.10984199e-01 -1.04808521e+00 -3.64043862e-01 7.67605305e-01 9.76659715e-01 1.57055110e-01 6.14326477e-01 -2.18935892e-01 8.94207299e-01 -6.88806295e-01 -3.96092713e-01 8.28051686e-01 -5.06502569e-01 -3.62219512e-01 6.06197491e-02 -2.92107593e-02 -2.60201454e-01 -2.91209072e-01 1.35192549e+00 8.18772197e-01 -1.53849155e-01 1.42834812e-01 -1.44220293e+00 -3.66464406e-01 1.22186756e+00 -3.73816043e-01 -1.83006480e-01 2.00061113e-01 2.39440709e-01 7.11973727e-01 -4.51746017e-01 6.92245424e-01 1.28160059e+00 6.22829914e-01 3.94298732e-01 -8.45491946e-01 -7.37465918e-01 4.53876913e-01 3.91090691e-01 -1.27039015e+00 -1.81901604e-01 8.82700920e-01 -2.44522318e-01 1.03030413e-01 3.78569309e-03 6.82146907e-01 6.35794342e-01 3.66493553e-01 6.65091336e-01 1.50557256e+00 -3.89023185e-01 8.97673488e-01 1.49822712e-01 -1.33382872e-01 5.59687912e-01 6.89967752e-01 2.19455391e-01 -6.94083750e-01 -6.47380352e-01 2.36481145e-01 4.00334634e-02 -5.95050395e-01 -4.13914084e-01 -1.13743782e+00 8.01116467e-01 5.22074938e-01 5.34723699e-01 -7.57408261e-01 3.00774664e-01 4.37043995e-01 1.08221149e+00 4.76407260e-01 -1.77555799e-01 -1.60313383e-01 2.47636706e-01 -7.15661824e-01 3.90600502e-01 1.57111216e+00 7.61754692e-01 1.22719193e+00 2.66103148e-01 4.23900746e-02 3.11949581e-01 4.16666955e-01 6.80439949e-01 3.95106018e-01 -1.01301718e+00 4.50643718e-01 4.35974360e-01 3.34203750e-01 -9.63484883e-01 -2.38222972e-01 -6.70531392e-01 -1.10551226e+00 3.69678497e-01 4.06710118e-01 -1.12517881e+00 -1.57094494e-01 1.70756173e+00 7.26008594e-01 1.98539719e-01 4.10427868e-01 1.20656502e+00 3.23112905e-01 5.57964504e-01 -5.11683881e-01 -6.06394768e-01 4.07318920e-01 -9.83996212e-01 -1.03939092e+00 2.37606615e-01 6.00839615e-01 -7.30523229e-01 5.97803742e-02 4.92667824e-01 -6.72885180e-01 1.79447487e-01 -1.12159383e+00 8.66162896e-01 -3.67884755e-01 -4.55198586e-01 5.93286991e-01 8.30228150e-01 -1.45195699e+00 6.54510617e-01 -4.69350159e-01 -5.17531574e-01 3.54292430e-02 4.09846604e-01 -1.74158856e-01 -3.17938775e-01 -1.00189626e+00 5.97606719e-01 -5.86741092e-03 1.35895982e-01 -1.19486821e+00 -3.85297835e-01 -4.31042582e-01 -1.56002656e-01 6.02815926e-01 -9.04661715e-01 1.32370198e+00 -1.30188489e+00 -1.74513006e+00 -7.82415830e-03 1.32434055e-01 -6.05798304e-01 9.42675173e-01 -1.26120523e-02 1.99781787e-02 -1.07976459e-01 -1.37862667e-01 -2.88946956e-01 7.23280191e-01 -1.45232761e+00 -5.98044991e-01 -5.74744821e-01 3.73240747e-02 4.22208667e-01 -3.34500045e-01 -3.09062093e-01 -6.22600503e-03 -3.71046692e-01 -1.84726298e-01 -9.20147717e-01 -6.18973494e-01 9.08652097e-02 -3.33905041e-01 -2.29396090e-01 1.36447084e+00 -1.57544434e-01 8.50602865e-01 -1.86836958e+00 2.03678399e-01 4.97196078e-01 1.61810786e-01 -1.21487388e-02 -1.57730281e-01 1.19400656e+00 5.62331080e-01 4.52595856e-03 -3.76905762e-02 -5.33879280e-01 1.24393173e-01 4.15632844e-01 2.33242169e-01 1.00336039e+00 -8.07656825e-01 2.86060840e-01 -1.09465933e+00 -3.14047188e-01 -4.81195971e-02 3.22198391e-01 -3.48736167e-01 2.70465612e-01 -1.28756359e-01 6.32371008e-01 -9.89172339e-01 4.30005640e-01 7.38424182e-01 -9.73212942e-02 5.13258696e-01 1.49793610e-01 -5.44882596e-01 -3.61885726e-01 -1.89239132e+00 1.48948586e+00 -3.00801158e-01 6.39967546e-02 1.32832301e+00 -1.19086957e+00 7.98432171e-01 7.15257466e-01 8.40858877e-01 -2.84443647e-01 3.82086396e-01 5.74292779e-01 -2.34243050e-01 -2.90876538e-01 6.26518503e-02 1.31876022e-01 3.71709131e-02 8.34610641e-01 2.13926181e-01 -2.68001527e-01 2.29089841e-01 6.50057614e-01 1.36137736e+00 -3.04613560e-01 1.88494533e-01 -2.76118755e-01 6.71819508e-01 -1.44982696e-01 7.84902096e-01 1.21488798e+00 -3.58108759e-01 -6.84375018e-02 2.01859206e-01 -3.39587510e-01 -8.89330387e-01 -4.39187080e-01 6.70075834e-01 7.42452741e-01 4.54152763e-01 -3.70348901e-01 -7.59998739e-01 -5.07884324e-01 3.13584059e-01 -6.70418888e-02 -3.00857127e-01 1.78239033e-01 -2.88679656e-02 -5.58038890e-01 3.70906323e-01 -4.01921123e-01 1.18806219e+00 -4.99988258e-01 -1.49314538e-01 4.01489258e-01 3.49565670e-02 -6.36773407e-01 -6.01810455e-01 8.42038989e-02 -5.66660345e-01 -1.40860236e+00 -8.31019700e-01 -5.55853248e-01 5.48071623e-01 3.19714159e-01 5.30891538e-01 1.88659847e-01 -4.87533435e-02 1.19904280e+00 -5.44258237e-01 -7.20844209e-01 -3.46329689e-01 -1.53896540e-01 2.39554062e-01 6.91968560e-01 -1.60220116e-01 -6.15003109e-01 -8.10238421e-01 1.85644999e-01 -1.01982033e+00 -4.41828042e-01 6.27929270e-01 8.64345849e-01 2.37751156e-01 8.14657211e-02 8.07651341e-01 -8.77427936e-01 9.76041496e-01 -7.84197211e-01 -7.59156525e-01 1.23524934e-01 -7.37242401e-01 -2.19160803e-02 1.06951606e+00 -2.20707729e-01 -1.06582654e+00 3.10206562e-01 4.96143490e-01 -2.21176669e-01 1.65715769e-01 4.85440582e-01 1.41014621e-01 -7.57610977e-01 4.86969918e-01 2.68658906e-01 3.25885922e-01 -4.96243358e-01 6.08285904e-01 9.32872236e-01 -5.54749221e-02 -5.08852661e-01 7.98330665e-01 6.89516485e-01 -2.18203999e-02 -6.89829886e-01 -4.71745610e-01 -5.03340960e-01 1.54984295e-02 -5.41725755e-01 2.48381272e-01 -8.50414693e-01 -1.06809163e+00 9.50505316e-01 -1.02787602e+00 -2.30986327e-01 -1.48808375e-01 8.45599294e-01 -5.29324830e-01 4.43071514e-01 -5.39506912e-01 -1.06677246e+00 -6.74772203e-01 -9.30650771e-01 2.30285272e-01 3.30155134e-01 6.07111752e-01 -1.20985246e+00 3.73980671e-01 -3.28154713e-02 1.14773369e+00 2.56178766e-01 1.30472898e-01 -6.80999517e-01 -7.63665676e-01 -1.77135736e-01 4.34180200e-01 4.79564339e-01 1.57536834e-01 -3.77785444e-01 -5.19078791e-01 -7.55283475e-01 1.70925498e-01 -4.85148638e-01 4.30318177e-01 2.17594251e-01 2.99106359e-01 -7.66211390e-01 -1.49944695e-02 2.06414908e-01 1.83290482e+00 -3.06128636e-02 -1.76171690e-01 1.52582273e-01 1.85108066e-01 3.96796077e-01 1.44203320e-01 1.10270107e+00 5.98740280e-01 1.85028777e-01 9.62730885e-01 6.36213645e-02 1.45066366e-01 -1.14711979e-02 5.49920857e-01 1.19447410e+00 2.38073170e-02 -1.26392007e-01 -4.34861273e-01 5.83073080e-01 -2.17901731e+00 -8.52862775e-01 1.03837401e-01 2.18363714e+00 6.23748600e-01 -5.43848038e-01 -5.57935573e-02 3.45885567e-02 1.11129463e+00 3.48080546e-01 -6.82373941e-01 -3.58438313e-01 -3.93067241e-01 -2.74620593e-01 9.75895286e-01 3.79244715e-01 -5.94792426e-01 4.16047990e-01 5.91033745e+00 4.65925902e-01 -1.23096633e+00 3.16610128e-01 2.56179422e-01 8.71847495e-02 -8.96284282e-02 3.31767142e-01 -4.28152889e-01 4.64620322e-01 5.36339521e-01 -6.81404293e-01 9.01710927e-01 7.05331922e-01 6.64046466e-01 -3.50571781e-01 -7.49451280e-01 9.26793218e-01 -1.88095551e-02 -1.06581199e+00 -8.70604888e-02 -1.02831639e-01 1.49477470e+00 4.35932040e-01 -2.59800643e-01 2.42679007e-02 1.12841153e+00 -3.48236710e-01 5.36352396e-01 6.98769510e-01 -6.31310567e-02 -6.02747440e-01 1.04293239e+00 6.90418065e-01 -1.05176997e+00 -5.97828865e-01 -2.62488753e-01 -2.74989694e-01 6.77444860e-02 8.50677490e-01 -4.84274358e-01 1.03854477e+00 7.58167088e-01 6.81847751e-01 4.69811400e-03 1.64176643e+00 -6.88024089e-02 7.95162261e-01 -4.91115749e-01 -4.59655076e-01 4.55735803e-01 -3.79840493e-01 8.59466553e-01 7.10887253e-01 3.28401387e-01 -5.25585040e-02 6.96435452e-01 4.93945092e-01 -3.21808606e-01 4.49719757e-01 -6.71306133e-01 3.07032883e-01 6.80085421e-01 1.30425799e+00 1.14952028e-01 -4.91538614e-01 -5.98375678e-01 6.89627469e-01 1.44560844e-01 7.18568504e-01 1.86833437e-03 -3.63547266e-01 3.72682154e-01 -3.45337272e-01 1.67403281e-01 -2.44612038e-01 1.82587028e-01 -9.53748286e-01 1.10883601e-01 -8.44997346e-01 5.03835261e-01 -2.04960287e-01 -1.61644721e+00 2.10403651e-01 -4.24523354e-01 -1.03680050e+00 -4.37257290e-02 1.53917387e-01 -7.93887317e-01 5.86050093e-01 -1.72126329e+00 -8.11803877e-01 -3.77467304e-01 1.21970773e+00 2.42300421e-01 -1.18507691e-01 5.99575162e-01 2.37376094e-01 -3.73117357e-01 1.41916782e-01 7.80910552e-01 -9.84515622e-02 8.68278861e-01 -1.04177856e+00 -6.68759882e-01 5.70792854e-01 -4.37638611e-01 3.19904417e-01 8.28193963e-01 -5.60264111e-01 -2.08385181e+00 -1.19798625e+00 5.98884761e-01 6.72448158e-01 7.31644392e-01 -4.84029241e-02 -3.48228425e-01 5.61273515e-01 6.59904122e-01 2.33291164e-01 3.53067338e-01 -4.59327757e-01 -1.49162501e-01 -6.23735666e-01 -1.52886915e+00 2.37107560e-01 5.66287518e-01 -1.62012681e-01 3.51609639e-03 5.97928107e-01 3.97107005e-01 -1.21734731e-01 -8.13027501e-01 -1.10220753e-01 1.52116030e-01 -8.67328107e-01 -3.73573452e-02 -2.57772766e-03 -5.70848823e-01 -5.20678461e-01 -9.08678770e-02 -2.04690647e+00 1.57423213e-01 -1.34166563e+00 1.88006043e-01 1.00607979e+00 5.82720637e-02 -1.21071911e+00 6.43600702e-01 1.58679098e-01 2.11393148e-01 -4.26142365e-01 -1.19686294e+00 -8.17708373e-01 3.69547643e-02 7.03558996e-02 5.16599953e-01 9.04201627e-01 1.93556488e-01 -1.20250203e-01 -4.70807672e-01 2.23054618e-01 1.11581182e+00 -8.41342360e-02 1.00256741e+00 -1.00778806e+00 -1.77924931e-01 -5.91623336e-02 -2.61893690e-01 -9.29421365e-01 2.18882831e-03 -6.03021562e-01 -4.51010838e-02 -1.55650544e+00 5.77309951e-02 -7.16045141e-01 -8.92918110e-02 2.21068248e-01 1.91085353e-01 -2.19405219e-01 5.66646159e-01 3.61061931e-01 -1.04063356e+00 6.84038162e-01 1.09799707e+00 -3.14809203e-01 -1.92963421e-01 2.34500676e-01 -5.11567712e-01 4.39830899e-01 9.47682858e-01 -5.57797968e-01 -2.58626997e-01 -3.32161278e-01 6.05926942e-03 3.24181855e-01 1.94034278e-01 -9.34158385e-01 1.11452639e+00 -7.59456828e-02 -4.28716153e-01 -1.72593713e-01 -9.61766765e-02 -1.14323294e+00 3.42180699e-01 7.66906023e-01 -2.72491336e-01 6.06391113e-04 -5.86124003e-01 1.03060102e+00 -3.43815893e-01 -1.13976285e-01 5.37832677e-01 -6.48626983e-01 -2.91709512e-01 2.77335525e-01 -6.37969434e-01 -2.13368773e-01 1.14854062e+00 2.95523584e-01 -4.67767626e-01 -1.10311842e+00 -7.07688212e-01 8.20187867e-01 3.68235588e-01 1.11411884e-01 3.15729976e-01 -9.77226496e-01 -8.28884780e-01 -3.09564739e-01 -5.05713940e-01 -1.41803920e-01 2.32058708e-02 1.00285208e+00 -1.39237434e-01 2.33979166e-01 2.16102660e-01 -4.01180506e-01 -9.75847721e-01 3.05145711e-01 6.19973361e-01 -2.45609209e-01 -1.97221458e-01 1.97289497e-01 -5.17066419e-01 -6.12389565e-01 7.37991631e-01 3.89807016e-01 8.81782770e-02 -1.94605932e-01 4.61525589e-01 7.45229363e-01 1.16997942e-01 -2.53666937e-01 5.73486537e-02 5.30913532e-01 2.79388368e-01 -1.39362603e-01 1.32066286e+00 -5.46810627e-01 -4.60548609e-01 4.41937685e-01 9.03733611e-01 2.97394730e-02 -1.06289589e+00 -7.97487676e-01 -8.81911293e-02 -3.49819154e-01 1.63718432e-01 -6.22922599e-01 -1.52076733e+00 2.99872663e-02 6.31207347e-01 3.66115183e-01 9.44438875e-01 -3.65153044e-01 4.27601218e-01 7.40253568e-01 1.02432656e+00 -1.24733961e+00 -3.14026885e-02 5.00686944e-01 5.08095741e-01 -1.02169669e+00 -9.76994634e-02 -2.28875637e-01 -3.62896085e-01 9.33159590e-01 4.32103634e-01 -4.15247411e-01 1.01679981e+00 3.68165284e-01 2.39285320e-01 -1.48013895e-02 -9.76481915e-01 -2.29519993e-01 -7.60276496e-01 5.94236076e-01 1.35558648e-02 6.24165265e-03 -1.12184632e+00 2.60040849e-01 2.30934069e-01 1.32131070e-01 6.99761212e-01 1.27681243e+00 -6.44151092e-01 -1.12653077e+00 -7.51380086e-01 2.05370978e-01 -6.78856790e-01 3.08011860e-01 -5.72906256e-01 4.75127935e-01 1.23807572e-01 1.41025674e+00 -5.01981080e-01 -7.86718279e-02 2.87192434e-01 -2.90589184e-01 -8.80514085e-02 -1.28341153e-01 -1.12209630e+00 -1.57768935e-01 -1.20328330e-01 -6.01331055e-01 -8.48399639e-01 -2.39032939e-01 -1.11745882e+00 -5.11024892e-01 -5.52987516e-01 8.12410176e-01 1.25850594e+00 6.73362434e-01 4.32935417e-01 -6.99291304e-02 1.32278502e+00 -9.44905758e-01 -1.26425016e+00 -8.29217196e-01 -1.00078642e+00 2.28230938e-01 4.87620115e-01 -3.22726727e-01 -9.33210671e-01 -3.16225976e-01]

[6.1687235832214355, 4.985624313354492]

5fc6076e-d68e-4db7-a047-23ca9449da54

a-multi-modal-approach-to-single-modal-visual

2305.06179

https://arxiv.org/abs/2305.06179v2

https://arxiv.org/pdf/2305.06179v2.pdf

A Multi-modal Approach to Single-modal Visual Place Classification

Visual place classification from a first-person-view monocular RGB image is a fundamental problem in long-term robot navigation. A difficulty arises from the fact that RGB image classifiers are often vulnerable to spatial and appearance changes and degrade due to domain shifts, such as seasonal, weather, and lighting differences. To address this issue, multi-sensor fusion approaches combining RGB and depth (D) (e.g., LIDAR, radar, stereo) have gained popularity in recent years. Inspired by these efforts in multimodal RGB-D fusion, we explore the use of pseudo-depth measurements from recently-developed techniques of ``domain invariant" monocular depth estimation as an additional pseudo depth modality, by reformulating the single-modal RGB image classification task as a pseudo multi-modal RGB-D classification problem. Specifically, a practical, fully self-supervised framework for training, appropriately processing, fusing, and classifying these two modalities, RGB and pseudo-D, is described. Experiments on challenging cross-domain scenarios using public NCLT datasets validate effectiveness of the proposed framework.

['Kenta Tsukahara', 'Kanji Tanaka', 'Tomoya Iwasaki']

2023-05-10

['monocular-depth-estimation', 'robot-navigation']

['computer-vision', 'robots']

[ 3.05671006e-01 -3.89394134e-01 8.01972970e-02 -7.43911803e-01 -8.64378631e-01 -6.84675157e-01 7.80058265e-01 -2.73335706e-02 -6.24890268e-01 7.92663991e-01 1.12755619e-01 9.47760865e-02 7.88696781e-02 -6.87506914e-01 -7.16273010e-01 -7.82222450e-01 5.11215210e-01 5.48398755e-02 1.32337973e-01 -1.46279648e-01 3.53381187e-01 5.56889772e-01 -2.03197360e+00 6.18549436e-02 7.22543716e-01 1.30442643e+00 2.63799340e-01 4.53949362e-01 -2.00095311e-01 4.87920702e-01 -2.21289515e-01 -9.70646366e-02 4.47390527e-01 -1.32533208e-01 -5.39837360e-01 3.02910298e-01 5.84323406e-01 -3.04414362e-01 -3.69701207e-01 1.10706520e+00 5.78107297e-01 6.52986839e-02 5.60318708e-01 -1.50462973e+00 -4.91898865e-01 -2.47946873e-01 -7.01771140e-01 -3.08670681e-02 9.15047705e-01 1.76896155e-02 4.80962873e-01 -8.30199480e-01 5.71136355e-01 1.25042617e+00 6.30106449e-01 4.33308572e-01 -9.69986677e-01 -5.56614399e-01 1.14790723e-01 3.60579163e-01 -1.25146341e+00 -4.59313571e-01 1.10981691e+00 -4.18934941e-01 7.36984372e-01 7.73266330e-02 6.20217741e-01 1.15197909e+00 1.49762690e-01 8.06071341e-01 1.70439088e+00 -4.77877647e-01 4.32797968e-01 2.40399957e-01 -2.84414411e-01 5.79459429e-01 1.96966678e-01 2.95900613e-01 -8.75800967e-01 1.09110326e-01 4.33965534e-01 2.30671287e-01 -4.34542477e-01 -8.84780467e-01 -1.08817923e+00 4.78194058e-01 7.13598490e-01 2.51345150e-02 -3.42574924e-01 4.62477840e-03 1.09977067e-01 1.45124495e-01 5.41432559e-01 -6.89631104e-02 -5.35413325e-01 -2.72747457e-01 -6.84696198e-01 1.65843889e-01 2.81266928e-01 9.92020488e-01 1.10157323e+00 -2.57288873e-01 5.75170875e-01 7.40824521e-01 6.72866285e-01 9.38918293e-01 5.76951683e-01 -9.13158834e-01 5.96723080e-01 5.79789877e-01 7.09639490e-02 -8.46409738e-01 -5.24801314e-01 -1.17203658e-02 -7.82667816e-01 3.11556816e-01 3.49040747e-01 3.89934629e-02 -9.41546917e-01 1.51227748e+00 6.10526621e-01 -8.79350603e-02 1.99634492e-01 1.12752521e+00 8.67836058e-01 1.23378463e-01 -1.93886254e-02 4.34477329e-02 1.15334153e+00 -5.04876673e-01 -5.57748079e-01 -6.24102294e-01 2.02883452e-01 -5.56755424e-01 6.52449310e-01 3.86552811e-01 -5.12296915e-01 -6.46245599e-01 -1.19175088e+00 -1.68420985e-01 -9.46345091e-01 -1.07433408e-01 6.13094449e-01 7.34472930e-01 -9.38456237e-01 7.25120604e-02 -7.26424932e-01 -6.56861365e-01 1.19714804e-01 1.57140359e-01 -9.00080740e-01 -5.74840248e-01 -1.11119103e+00 8.94588828e-01 3.33931148e-01 2.67220914e-01 -4.44243938e-01 -2.69647121e-01 -1.26375294e+00 -8.57037783e-01 8.03448856e-02 -6.07223630e-01 8.92241180e-01 -7.97742009e-01 -1.34364092e+00 1.22185743e+00 -3.21696907e-01 -9.26939622e-02 4.59372371e-01 -3.06940794e-01 -3.07161391e-01 1.17382951e-01 2.21335873e-01 7.52386987e-01 8.33230674e-01 -1.66652024e+00 -9.29801166e-01 -1.21964252e+00 9.57599841e-03 7.03903615e-01 1.13062628e-01 -6.17731154e-01 -2.84039974e-01 -6.44107237e-02 8.86778653e-01 -1.03908038e+00 -5.68372160e-02 1.20729789e-01 -2.26715833e-01 2.33753070e-01 7.58166194e-01 -5.19421041e-01 3.10348064e-01 -2.08684087e+00 2.34170943e-01 8.15165639e-02 -1.33876622e-01 -3.16489190e-01 3.87912750e-01 5.22366092e-02 7.99225271e-02 -3.84358764e-01 -3.13980371e-01 -6.69465780e-01 -6.43129973e-03 4.76568848e-01 -4.05969769e-02 9.19978797e-01 -1.23080097e-01 5.93900919e-01 -9.82580662e-01 -4.24772203e-01 7.06645191e-01 5.37541807e-01 -2.32263744e-01 1.61976635e-01 1.17459565e-01 9.73852813e-01 -1.75568163e-01 1.17183661e+00 9.62725878e-01 2.22851560e-01 -1.67769745e-01 -3.04665327e-01 -2.47860253e-01 -6.95076138e-02 -1.18459320e+00 2.39876413e+00 -4.61594045e-01 6.46785200e-01 1.32856846e-01 -8.24952841e-01 8.05821180e-01 9.87469181e-02 4.64794576e-01 -1.16750014e+00 1.39562443e-01 4.10277128e-01 -7.86590040e-01 -4.64526772e-01 7.40194798e-01 -2.47115046e-01 -3.09375167e-01 6.45198673e-02 -2.98426244e-02 -4.91673023e-01 -3.15415412e-01 -3.31819952e-01 8.50904107e-01 5.84761620e-01 4.07002658e-01 3.65226001e-01 6.18284285e-01 8.51664171e-02 3.70083064e-01 4.34330016e-01 -6.75736487e-01 8.42134476e-01 -1.85019001e-01 -1.81959182e-01 -8.94154310e-01 -1.21284258e+00 -2.38903180e-01 6.96653366e-01 6.47873044e-01 2.30971262e-01 -1.61409169e-01 -4.14872766e-01 3.75296742e-01 2.01276585e-01 -5.40026903e-01 -2.30981149e-02 -2.62615025e-01 -5.37250876e-01 4.45151359e-01 3.96002591e-01 1.08463764e+00 -6.25939548e-01 -1.06339312e+00 7.42523521e-02 -4.65348750e-01 -1.48576570e+00 1.27916947e-01 5.38238049e-01 -8.70133519e-01 -1.06825101e+00 -9.10174072e-01 -5.29277861e-01 4.31330800e-01 6.96836233e-01 7.17538178e-01 -6.63493633e-01 -2.96444178e-01 9.99489665e-01 -4.35717702e-01 -1.77255630e-01 6.51213676e-02 -1.93266481e-01 1.74091056e-01 6.35627285e-02 5.44564426e-01 -5.50933003e-01 -7.03116834e-01 1.43476307e-01 -8.30397725e-01 1.20223518e-02 6.00843012e-01 5.72094500e-01 7.15086043e-01 -2.34229058e-01 7.29382262e-02 -4.20051217e-01 1.56188875e-01 -4.56044585e-01 -4.94740546e-01 1.56743318e-01 -4.78983998e-01 -6.59825839e-03 -2.22700667e-02 1.75338099e-03 -1.21000957e+00 5.00193715e-01 1.94077566e-02 -4.19190228e-01 -8.24854374e-01 4.36001450e-01 -4.63584572e-01 -3.54025632e-01 5.97924411e-01 4.55484182e-01 1.79892462e-02 -2.68995196e-01 6.77465379e-01 1.03589320e+00 7.91561246e-01 -3.82468075e-01 9.05515373e-01 9.71550107e-01 1.99974239e-01 -9.23044860e-01 -5.08892775e-01 -9.73506451e-01 -9.92623627e-01 -4.43965554e-01 8.72498572e-01 -1.45638406e+00 -4.46385264e-01 8.24913502e-01 -1.16229165e+00 3.37725133e-02 1.24153718e-01 4.58321840e-01 -6.44214928e-01 5.34696996e-01 -1.47633344e-01 -9.87269104e-01 1.40792191e-01 -9.84468818e-01 1.53487074e+00 4.52796429e-01 2.34588593e-01 -8.35200548e-01 2.47780293e-01 5.93592823e-01 1.01931609e-01 7.46211231e-01 4.70696986e-01 1.63645774e-01 -6.14439070e-01 -1.34660423e-01 -3.75770897e-01 2.35670686e-01 3.76549602e-01 -5.90984821e-01 -1.34659147e+00 -8.55834037e-02 -3.72064556e-03 -4.50867057e-01 6.53153300e-01 1.42211780e-01 5.64250112e-01 2.77863383e-01 -2.89408445e-01 8.71262729e-01 1.75153732e+00 2.22370714e-01 4.39967334e-01 7.77156711e-01 8.91936719e-01 5.35401165e-01 8.84447217e-01 7.02533603e-01 8.77079308e-01 7.13357985e-01 6.59664094e-01 -1.09699920e-01 5.26253916e-02 -1.18142255e-01 2.20866486e-01 3.18826288e-01 6.76723793e-02 1.29511937e-01 -1.03608000e+00 4.08145398e-01 -1.72235048e+00 -6.94862068e-01 4.13881987e-02 2.11367655e+00 5.77638507e-01 -9.58417058e-02 -1.15874223e-01 4.15324092e-01 3.80171984e-01 1.68316871e-01 -6.75235868e-01 1.19141787e-01 -5.22310376e-01 -1.40802696e-01 9.83019531e-01 2.14723870e-01 -1.31585693e+00 7.05828488e-01 5.04292011e+00 1.55958593e-01 -1.33145058e+00 7.31237605e-02 1.72895223e-01 1.37173086e-01 -2.54845798e-01 -2.18447402e-01 -4.90489125e-01 2.62648612e-01 4.70913082e-01 2.77347922e-01 3.72359306e-01 8.52478802e-01 -9.84791294e-02 -7.41171658e-01 -1.05935585e+00 1.75111985e+00 4.71764028e-01 -5.94827771e-01 -3.81683201e-01 1.12721726e-01 5.80397069e-01 2.16100439e-01 1.54556349e-01 1.51598141e-01 3.02665010e-02 -6.28164291e-01 9.92566288e-01 4.29351032e-01 7.83130646e-01 -5.35383105e-01 5.95600009e-01 2.92583674e-01 -1.32482767e+00 -1.99703291e-01 -1.86863780e-01 -1.28665611e-01 2.17233449e-01 3.93438786e-01 -3.79761726e-01 9.03622508e-01 1.10057771e+00 1.07883394e+00 -8.40744495e-01 9.98121619e-01 -2.41682336e-01 -4.75868970e-01 -3.69456530e-01 3.68337423e-01 3.58024091e-02 -1.34413123e-01 2.60093540e-01 7.52852917e-01 3.20684582e-01 1.34164378e-01 -1.04409553e-01 4.72830951e-01 4.10404563e-01 -3.06323946e-01 -1.01166713e+00 3.95312279e-01 4.06508058e-01 1.01683819e+00 -5.71719825e-01 5.59218489e-02 -5.19906342e-01 1.41277707e+00 4.23630662e-02 5.42418540e-01 -7.04947710e-01 -1.20828189e-01 8.45444202e-01 -2.57948339e-01 8.46361443e-02 -6.40067041e-01 -5.73212266e-01 -1.26634920e+00 1.63002297e-01 -5.62617064e-01 1.77565023e-01 -1.15447247e+00 -1.09940863e+00 4.44446206e-01 9.44440290e-02 -1.63138211e+00 -2.96546966e-01 -8.23882937e-01 2.60118753e-01 8.35221052e-01 -2.17769670e+00 -1.44767141e+00 -1.01697743e+00 1.03849483e+00 2.84830153e-01 4.33821827e-02 6.55542016e-01 4.18174922e-01 -8.32572132e-02 2.76279509e-01 8.96711200e-02 -1.88999951e-01 8.57384026e-01 -1.24173212e+00 2.72612832e-02 7.90675282e-01 -1.46392241e-01 2.69138902e-01 6.57569408e-01 -4.96674299e-01 -1.73774266e+00 -9.35955524e-01 6.46364927e-01 -4.92490172e-01 1.61198258e-01 -3.49527001e-01 -2.94275463e-01 4.55343634e-01 -1.34933829e-01 6.17099069e-02 6.23539984e-01 -2.18825027e-01 -4.22268480e-01 -3.03274333e-01 -1.49837101e+00 3.44690323e-01 1.20205891e+00 -1.01429498e+00 -5.70427358e-01 9.06117726e-03 3.74331415e-01 -8.20827365e-01 -6.64328098e-01 4.98363942e-01 6.76162362e-01 -1.33687007e+00 1.25342667e+00 2.64207035e-01 1.33806258e-01 -3.62406194e-01 -8.32434893e-01 -1.25508118e+00 3.02817345e-01 1.39496818e-01 1.34557441e-01 1.02903593e+00 -9.38205570e-02 -6.26335919e-01 8.95153344e-01 6.69681787e-01 -7.88190886e-02 -1.14066511e-01 -1.35286880e+00 -4.52063173e-01 -3.03708076e-01 -6.55132055e-01 4.64141101e-01 7.30638862e-01 -9.52572823e-02 -3.29775810e-02 -2.80946285e-01 4.33098197e-01 9.59421098e-01 9.63126346e-02 9.85770881e-01 -1.31743562e+00 5.05741164e-02 -6.09774776e-02 -8.55104387e-01 -1.24921119e+00 1.57558724e-01 -5.23683488e-01 4.18346584e-01 -1.60548544e+00 -2.14659616e-01 -5.35282135e-01 -1.65629461e-01 1.81008086e-01 1.32849202e-01 4.93399829e-01 2.94881966e-02 3.22033525e-01 -5.41317701e-01 8.78620028e-01 9.78439748e-01 -1.23175949e-01 -8.54850002e-03 -2.75938988e-01 -2.55878627e-01 6.06436610e-01 4.01075214e-01 -2.25908965e-01 -3.92452270e-01 -4.36920583e-01 2.44362414e-01 1.83777809e-01 5.03412604e-01 -1.32530153e+00 4.56858307e-01 -1.14741564e-01 5.85059822e-01 -9.90631521e-01 8.51220846e-01 -1.27028680e+00 2.14574183e-03 2.55077183e-01 3.32072586e-01 2.26147935e-01 1.45271033e-01 8.17637324e-01 -4.89826053e-01 4.21793967e-01 6.55873358e-01 -3.80007684e-01 -1.35692441e+00 9.01257172e-02 -6.96808845e-02 -2.63578326e-01 1.00466716e+00 -8.11851919e-01 -2.74999976e-01 -2.59217411e-01 -4.54462737e-01 2.61237472e-02 7.18856931e-01 6.38377309e-01 9.95959938e-01 -1.46972799e+00 -5.88468537e-02 5.20457923e-01 5.99677205e-01 2.72636682e-01 4.26471382e-01 7.96003282e-01 -6.10212207e-01 6.17474377e-01 -5.07092893e-01 -9.78716969e-01 -1.19047654e+00 3.42658073e-01 4.12032187e-01 2.76105970e-01 -3.35135669e-01 8.03538263e-01 -5.30968383e-02 -8.71812046e-01 3.66367280e-01 -4.10175055e-01 -5.94657175e-02 1.65072724e-01 7.24897087e-02 2.82445043e-01 2.52587587e-01 -1.04959583e+00 -8.05919170e-01 9.81877208e-01 5.45205295e-01 -5.55799007e-01 1.10313213e+00 -9.51905847e-01 -8.43780302e-03 8.72636855e-01 1.22599089e+00 -5.21167278e-01 -1.30870903e+00 -4.98463750e-01 -1.94751043e-02 -5.99015951e-01 2.06228625e-02 -6.72934294e-01 -7.68981695e-01 8.70441437e-01 1.14347029e+00 -1.61878929e-01 1.29716086e+00 -1.36727720e-01 6.24840915e-01 2.18724176e-01 1.04604065e+00 -1.16150045e+00 3.58750522e-02 5.07835567e-01 7.14702487e-01 -1.69385600e+00 9.87031832e-02 -1.38812080e-01 -5.13904452e-01 9.47992384e-01 5.92210770e-01 3.18048269e-01 7.81107485e-01 -1.41885862e-01 4.79864657e-01 -1.62502080e-02 -7.23366216e-02 -5.04577100e-01 9.01193768e-02 8.98488998e-01 1.63074985e-01 -1.72204878e-02 3.28797579e-01 1.06589325e-01 -2.13282034e-01 1.43625244e-01 2.82529891e-01 1.48925495e+00 -2.91266024e-01 -9.56037998e-01 -7.55711317e-01 -1.45759135e-01 1.59929186e-01 2.79034197e-01 -2.94277132e-01 7.65111685e-01 5.08884013e-01 1.11501217e+00 -2.80101430e-02 -6.07029080e-01 4.89128619e-01 1.47010982e-01 8.19754362e-01 -2.72542834e-01 -1.50944009e-01 -1.82763994e-01 -2.30055064e-01 -7.54913568e-01 -1.02151334e+00 -9.99202490e-01 -9.81168509e-01 -2.12335125e-01 3.13942321e-02 -4.19703305e-01 1.23066354e+00 1.18773985e+00 5.22854179e-02 2.05262005e-01 4.92597640e-01 -1.31122768e+00 -2.15016931e-01 -6.69085324e-01 -7.21251190e-01 5.43849468e-01 8.29617083e-01 -1.00476599e+00 -4.94865149e-01 -5.99098243e-02]

[8.307612419128418, -2.332390308380127]

85673192-a9bd-4be3-85bb-3898359e419d

learning-to-jointly-generate-and-separate

http://openaccess.thecvf.com/content_ICCV_2019/html/Ma_Learning_to_Jointly_Generate_and_Separate_Reflections_ICCV_2019_paper.html

http://openaccess.thecvf.com/content_ICCV_2019/papers/Ma_Learning_to_Jointly_Generate_and_Separate_Reflections_ICCV_2019_paper.pdf

Learning to Jointly Generate and Separate Reflections

Existing learning-based single image reflection removal methods using paired training data have fundamental limitations about the generalization capability on real-world reflections due to the limited variations in training pairs. In this work, we propose to jointly generate and separate reflections within a weakly-supervised learning framework, aiming to model the reflection image formation more comprehensively with abundant unpaired supervision. By imposing the adversarial losses and combinable mapping mechanism in a multi-task structure, the proposed framework elegantly integrates the two separate stages of reflection generation and separation into a unified model. The gradient constraint is incorporated into the concurrent training process of the multi-task learning as well. In particular, we built up an unpaired reflection dataset with 4,027 images, which is useful for facilitating the weakly-supervised learning of reflection removal model. Extensive experiments on a public benchmark dataset show that our framework performs favorably against state-of-the-art methods and consistently produces visually appealing results.

[' Ling-Yu Duan', ' Alex C. Kot', ' Boxin Shi', ' Renjie Wan', 'Daiqian Ma']

2019-10-01

iccv-2019-10

['reflection-removal']

['computer-vision']

[ 7.18344212e-01 -2.79758405e-02 2.94178277e-01 -1.93538308e-01 -1.26357937e+00 -2.71245629e-01 7.63886929e-01 -6.85360610e-01 -1.77018136e-01 6.55537546e-01 2.18687534e-01 -2.83592552e-01 3.42284143e-02 -6.83449924e-01 -1.03264225e+00 -1.12204611e+00 3.21939409e-01 -6.50275424e-02 -5.07544614e-02 -2.42138639e-01 9.45435092e-02 1.55872852e-01 -1.41685212e+00 5.73197186e-01 1.11754155e+00 8.49122524e-01 8.52880403e-02 2.96304762e-01 3.52072358e-01 8.62459600e-01 -3.82191360e-01 -4.35167134e-01 6.41738832e-01 -5.07257819e-01 -2.92396575e-01 1.77411899e-01 9.13117111e-01 -2.41927013e-01 -3.29571426e-01 6.40137613e-01 7.85743654e-01 5.18644825e-02 6.79008007e-01 -9.70068276e-01 -9.29435313e-01 3.29526186e-01 -8.21806252e-01 -3.34594190e-01 2.50439405e-01 2.33847089e-02 9.99548793e-01 -1.13588357e+00 2.73479462e-01 9.39010322e-01 7.47677267e-01 4.89579201e-01 -1.25937963e+00 -8.20684314e-01 2.45870531e-01 -1.08962126e-01 -1.24829972e+00 -5.83275259e-01 1.18000472e+00 -1.93921223e-01 5.77340066e-01 3.41461092e-01 2.56558537e-01 1.48464584e+00 8.91290456e-02 8.15136373e-01 1.40423346e+00 -4.35770810e-01 -2.65690178e-01 2.42674500e-01 -1.89865053e-01 8.79019380e-01 4.15736213e-02 2.78827816e-01 -7.07539260e-01 -6.14306293e-02 7.72242665e-01 1.28330633e-01 -3.90805215e-01 -2.57384062e-01 -1.01746106e+00 5.77652574e-01 4.15317893e-01 -7.72307366e-02 -3.41333896e-02 -6.84574395e-02 1.69662312e-01 5.34115613e-01 8.51917386e-01 7.54098371e-02 -1.19789489e-01 7.53015995e-01 -6.54892147e-01 8.11098367e-02 2.82731652e-01 8.96269798e-01 7.99799442e-01 2.24346086e-01 -4.36491728e-01 1.08588326e+00 2.04800159e-01 6.58001721e-01 8.67688060e-02 -6.15755081e-01 6.98724508e-01 3.84807140e-01 -9.10435058e-03 -5.97269237e-01 -2.26506457e-01 -7.61361539e-01 -1.05958486e+00 4.67100561e-01 1.68281555e-01 -1.55395269e-01 -9.27964807e-01 1.80571353e+00 2.44793132e-01 5.99305332e-01 3.23599249e-01 8.91775429e-01 1.16442025e+00 4.78373080e-01 -3.22432041e-01 -7.39137381e-02 1.11557925e+00 -1.51381576e+00 -3.69782388e-01 -3.15238297e-01 3.60868543e-01 -1.04907012e+00 1.30839944e+00 6.47921860e-01 -1.02600563e+00 -7.65233576e-01 -1.18676233e+00 -2.10964188e-01 1.19302891e-01 5.29429197e-01 7.03437388e-01 6.78854465e-01 -7.58837223e-01 1.33593544e-01 -2.73319125e-01 3.56826007e-01 6.06686532e-01 -9.71160904e-02 -4.03833359e-01 -4.91655141e-01 -8.90347183e-01 6.61112785e-01 -2.83840925e-01 5.10233998e-01 -1.09786904e+00 -9.02787566e-01 -8.14697921e-01 -3.57512504e-01 5.67419112e-01 -9.37750399e-01 8.22585166e-01 -8.00355136e-01 -1.66952896e+00 9.64891851e-01 6.85744733e-02 5.95040284e-02 9.98350739e-01 -5.67027926e-01 -2.78833270e-01 3.45167965e-02 -1.43929467e-01 1.32330567e-01 1.31304455e+00 -1.95288134e+00 -2.51209170e-01 -2.37330079e-01 -3.04873195e-02 4.53970343e-01 -3.56980979e-01 -1.31628841e-01 -5.26462436e-01 -7.57528186e-01 5.79264127e-02 -9.02126372e-01 -6.56769574e-02 1.44344829e-02 -5.88255882e-01 1.98184609e-01 8.60832691e-01 -3.83382022e-01 5.79170287e-01 -2.15328646e+00 2.38267273e-01 5.11465929e-02 2.50640661e-01 1.64950684e-01 -5.47025740e-01 2.66416103e-01 -2.84858376e-01 -2.83960760e-01 -2.36745730e-01 -9.35803354e-01 -1.43008828e-01 -2.28546992e-01 -6.12315357e-01 6.24050796e-01 1.61810026e-01 9.63066041e-01 -5.91072381e-01 -2.79920667e-01 1.58770666e-01 8.17118943e-01 -3.82385343e-01 4.57463920e-01 -6.00493466e-03 8.04620206e-01 -5.38365483e-01 6.49087906e-01 8.86213362e-01 -1.70783728e-01 3.08439862e-02 -4.92358685e-01 -8.46365765e-02 -3.32793593e-02 -9.06934619e-01 1.88744736e+00 -9.03241098e-01 2.95368105e-01 -7.30306581e-02 -9.79835510e-01 1.12669873e+00 2.48215586e-01 5.13462424e-01 -1.16224527e+00 -1.65615961e-01 1.21115766e-01 -3.17438960e-01 -3.48337024e-01 1.91465735e-01 -3.87422115e-01 -1.30676895e-01 7.78686225e-01 -1.21573776e-01 -9.37040225e-02 -3.71288091e-01 1.42544052e-02 9.96855974e-01 5.80462098e-01 -1.73924938e-01 1.87571958e-01 6.81328356e-01 -5.55427790e-01 5.35170078e-01 7.05339372e-01 3.19145232e-01 1.16698527e+00 -4.19774540e-02 -3.12309176e-01 -9.54884112e-01 -1.53276563e+00 -2.58324862e-01 1.14196289e+00 2.57644266e-01 -7.86730722e-02 -2.74447113e-01 -6.35650098e-01 -2.14410484e-01 3.09457570e-01 -5.35074115e-01 -3.17830890e-01 -7.66872585e-01 -1.10592687e+00 5.49612045e-01 3.00890923e-01 8.26735973e-01 -1.02952826e+00 -1.93495065e-01 -1.58352762e-01 -3.33081007e-01 -1.30637658e+00 -3.15698534e-01 8.93478990e-02 -5.45038462e-01 -1.04870892e+00 -9.89613831e-01 -8.38850260e-01 7.68133044e-01 7.52914131e-01 1.12811327e+00 2.21382126e-01 -6.79550827e-01 3.53668600e-01 -3.22108179e-01 -2.50726104e-01 -1.90520033e-01 -2.78914332e-01 -3.56417567e-01 2.12464809e-01 -2.93204129e-01 -8.27096343e-01 -7.49496222e-01 4.26616132e-01 -8.39086771e-01 4.85150635e-01 1.01146257e+00 1.07600260e+00 4.59302485e-01 -4.32613157e-02 5.62581062e-01 -1.15615094e+00 4.39620405e-01 -2.12180048e-01 -3.71543497e-01 5.71090281e-01 -5.18388271e-01 -6.76290616e-02 9.12397504e-01 -2.37515256e-01 -1.83195186e+00 1.95096359e-02 1.16749212e-01 -2.67366976e-01 3.06528136e-02 -5.83870038e-02 -2.70381480e-01 -4.56934154e-01 6.12296164e-01 4.40068781e-01 -1.30835578e-01 -4.68408257e-01 4.49380338e-01 3.03438902e-01 5.62140048e-01 -8.66880000e-01 1.28992450e+00 7.60801911e-01 1.41031623e-01 -7.59286344e-01 -1.19502676e+00 -3.34593326e-01 -4.83122736e-01 -2.06093878e-01 4.62811410e-01 -1.42566657e+00 -6.38645232e-01 9.49571609e-01 -9.03687298e-01 -5.93417525e-01 1.38226986e-01 2.12564006e-01 -6.70999646e-01 5.18421233e-01 -6.33028328e-01 -7.86928296e-01 -5.97680807e-01 -1.05224228e+00 1.28670192e+00 -8.39023441e-02 5.49981654e-01 -8.78125310e-01 2.52620429e-01 9.65188384e-01 2.98696727e-01 2.13950306e-01 7.88634300e-01 -2.95734107e-02 -9.17154789e-01 1.91441119e-01 -4.81560886e-01 6.67122543e-01 1.70666307e-01 -2.88418800e-01 -1.37827277e+00 -4.22556758e-01 8.83930176e-02 -7.79854357e-01 1.59369993e+00 1.57321077e-02 1.26145887e+00 2.56361663e-02 -9.82007086e-02 1.01237583e+00 1.51842785e+00 -3.21494520e-01 1.07788777e+00 2.49626487e-01 1.04813218e+00 6.82480514e-01 6.81558371e-01 4.58765775e-01 2.52384454e-01 6.12627864e-01 4.37741339e-01 -7.96861589e-01 -3.88391882e-01 -1.81495592e-01 5.63053191e-01 5.96959114e-01 -3.78131837e-01 -4.07840997e-01 -3.73875439e-01 -1.39975669e-02 -1.81740582e+00 -1.01539028e+00 -2.55193356e-02 2.23133326e+00 7.73611665e-01 1.38842473e-02 -1.06664628e-01 1.19790047e-01 3.48046154e-01 6.77750289e-01 -2.96062917e-01 1.60705581e-01 -3.51206094e-01 3.77213985e-01 3.73898000e-01 4.29919660e-01 -1.07991374e+00 7.19317377e-01 6.06732225e+00 6.94969594e-01 -1.06267273e+00 -2.91702822e-02 6.03124857e-01 -3.09580803e-01 -6.97374523e-01 -3.01762909e-01 -7.07768798e-01 1.46323115e-01 2.15416983e-01 2.64788330e-01 3.07010055e-01 1.72662973e-01 3.93704534e-01 1.66863739e-01 -9.57051218e-01 8.94190013e-01 5.34252465e-01 -1.07152987e+00 1.88796341e-01 -1.34203941e-01 9.57309425e-01 -8.62231180e-02 6.84515238e-01 1.37426436e-01 3.93453762e-02 -1.24527156e+00 6.06858373e-01 6.50490522e-01 9.57981467e-01 -6.56107187e-01 3.67857367e-01 1.05878688e-01 -1.06540179e+00 -5.91101311e-02 -1.94572896e-01 1.45275772e-01 1.28352880e-01 6.80972278e-01 -3.88471782e-01 1.09984720e+00 6.19026244e-01 9.67649758e-01 -4.56077456e-01 7.77597547e-01 -5.01279712e-01 2.73392260e-01 -3.24223600e-02 5.94746292e-01 6.34004772e-02 -6.90582693e-01 2.92601228e-01 9.50957716e-01 1.52762637e-01 -1.68404043e-01 2.31813535e-01 8.10750306e-01 -1.62854448e-01 -9.01793465e-02 -6.05659425e-01 4.25602973e-01 8.13632682e-02 1.47930908e+00 -2.45758697e-01 1.23170823e-01 -8.54822874e-01 1.25515878e+00 3.94232005e-01 6.24857545e-01 -1.04683602e+00 -1.10851536e-02 4.89181668e-01 -1.78123526e-02 3.02595764e-01 -5.83241731e-02 -3.29736978e-01 -1.24791324e+00 5.82975864e-01 -1.02020943e+00 1.94325700e-01 -7.16503739e-01 -1.49961102e+00 5.17797172e-01 -3.93442482e-01 -1.51419628e+00 2.57112116e-01 -5.75928688e-01 -1.15389431e+00 8.93165529e-01 -2.26300216e+00 -1.82160902e+00 -7.86124945e-01 8.13201368e-01 2.76220739e-01 -2.75141269e-01 7.47688055e-01 5.86028337e-01 -7.98452437e-01 8.88950944e-01 1.82603300e-01 -1.05861403e-01 1.21475554e+00 -9.82436836e-01 1.11495875e-01 8.84988427e-01 2.37122983e-01 4.14009899e-01 4.61680919e-01 -2.87023216e-01 -1.39203072e+00 -1.15737188e+00 2.14053869e-01 -2.14064583e-01 3.93570602e-01 -6.84899271e-01 -7.75588453e-01 7.89883971e-01 7.43401796e-02 1.82828605e-01 9.11786199e-01 1.53059393e-01 -8.77357721e-01 -3.50886017e-01 -8.65091205e-01 6.66982770e-01 1.37754810e+00 -6.65786684e-01 -3.28050882e-01 5.85805058e-01 5.95958114e-01 -4.12748158e-01 -6.20016813e-01 4.81216311e-01 5.96378207e-01 -1.28311503e+00 1.56596148e+00 -2.40921214e-01 8.05067658e-01 -1.88809440e-01 2.51993425e-02 -1.23636603e+00 -1.00945987e-01 -7.98693538e-01 2.11416304e-01 1.20143902e+00 5.36186218e-01 -7.35247374e-01 6.78991795e-01 2.78825089e-02 -5.34365892e-01 -7.87105381e-01 -6.07863009e-01 -6.16523206e-01 2.03443795e-01 -2.88463533e-01 1.35071754e-01 6.19338751e-01 -5.95098972e-01 3.98382038e-01 -1.06041837e+00 2.36577347e-01 1.16621137e+00 4.93611068e-01 1.07110095e+00 -9.84691262e-01 -5.71202576e-01 -1.22504808e-01 2.24065244e-01 -1.16493702e+00 4.70171571e-01 -7.55979359e-01 1.31500840e-01 -1.16795874e+00 5.69297552e-01 -6.44439638e-01 -4.74605411e-01 5.61652005e-01 -3.29082668e-01 7.01725423e-01 -8.18905234e-03 1.97521970e-01 -7.20359027e-01 1.12280142e+00 1.65756416e+00 -7.92685673e-02 -7.04125687e-02 1.72885373e-01 -1.07984364e+00 5.37641585e-01 5.41482389e-01 -2.05457985e-01 -6.70099854e-01 -6.84219062e-01 2.52595991e-01 -1.06743932e-01 7.09514558e-01 -7.54072785e-01 1.01007178e-01 1.48693338e-01 3.60291064e-01 -3.82304549e-01 6.69342935e-01 -6.89768970e-01 -5.61005324e-02 2.56106835e-02 -2.69178122e-01 -6.04878664e-01 9.68895555e-02 7.27858186e-01 -1.48909464e-01 2.50363708e-01 8.00818920e-01 1.15356129e-02 -5.00340581e-01 4.16376263e-01 2.98951775e-01 5.19110821e-02 9.75603342e-01 -1.10525474e-01 -7.05037713e-01 -1.41134828e-01 -3.78146112e-01 1.44214213e-01 3.99865776e-01 5.46819091e-01 8.51852238e-01 -1.25489926e+00 -1.00054204e+00 3.99934143e-01 4.49932426e-01 1.28800809e-01 6.12594903e-01 7.13706255e-01 -1.44862365e-02 -9.00949985e-02 -2.85690159e-01 -2.92907864e-01 -1.50497866e+00 2.67054975e-01 3.11034948e-01 -3.04937035e-01 -1.00702500e+00 7.63965905e-01 6.27928793e-01 -5.58525026e-01 1.91418439e-01 2.10266396e-01 2.81626433e-02 -2.63034940e-01 5.43264627e-01 3.34440827e-01 2.71490756e-02 -3.81353408e-01 -1.27762437e-01 9.02681053e-01 -2.57138878e-01 1.20572001e-01 1.67368007e+00 -9.97005031e-02 5.29578142e-02 2.28089437e-01 1.33975279e+00 2.34804332e-01 -1.56181002e+00 -5.72849214e-01 -6.31073952e-01 -5.26488364e-01 6.88933879e-02 -8.01442921e-01 -1.33473313e+00 7.24407256e-01 2.57525235e-01 -3.68937194e-01 1.28013551e+00 -1.05949320e-01 8.60187888e-01 4.32684720e-01 2.76968449e-01 -7.66315520e-01 7.72329092e-01 2.05697253e-01 9.68513727e-01 -1.56813145e+00 2.92393357e-01 -8.21373224e-01 -6.07713461e-01 9.60556030e-01 6.13465965e-01 -2.25148886e-01 3.08906704e-01 4.62642133e-01 3.44145536e-01 -1.92686021e-01 -6.70478046e-01 5.20508960e-02 5.47117710e-01 7.02468336e-01 5.39062142e-01 -3.46770078e-01 1.33592440e-02 3.83544654e-01 1.09021254e-01 -2.95889497e-01 3.15774590e-01 6.19877934e-01 -1.65049046e-01 -1.26815534e+00 -2.45925218e-01 7.66642168e-02 -2.25630909e-01 -2.95754284e-01 -2.68825978e-01 5.49154043e-01 -4.31690887e-02 8.79196942e-01 -2.25965455e-01 -2.87926108e-01 2.56312817e-01 -3.70947212e-01 8.24005902e-01 -5.63558280e-01 -5.76300144e-01 1.25927046e-01 6.31531775e-02 -5.84694505e-01 -6.25155807e-01 -4.26418513e-01 -8.88422668e-01 -7.15220207e-03 -2.38578171e-01 -3.14507663e-01 2.20831364e-01 9.87741709e-01 1.56244606e-01 7.52849221e-01 1.19100213e+00 -9.81063068e-01 -6.79460406e-01 -7.11756825e-01 -6.29218996e-01 6.20713055e-01 4.20288414e-01 -6.38029933e-01 -7.02775359e-01 -1.30973026e-01]

[10.589333534240723, -2.765544891357422]

5766955e-bb57-4b9f-b828-77496b539c7b

spatial-temporal-anomaly-detection-for-sensor

2212.07757

https://arxiv.org/abs/2212.07757v1

https://arxiv.org/pdf/2212.07757v1.pdf

Spatial-Temporal Anomaly Detection for Sensor Attacks in Autonomous Vehicles

Time-of-flight (ToF) distance measurement devices such as ultrasonics, LiDAR and radar are widely used in autonomous vehicles for environmental perception, navigation and assisted braking control. Despite their relative importance in making safer driving decisions, these devices are vulnerable to multiple attack types including spoofing, triggering and false data injection. When these attacks are successful they can compromise the security of autonomous vehicles leading to severe consequences for the driver, nearby vehicles and pedestrians. To handle these attacks and protect the measurement devices, we propose a spatial-temporal anomaly detection model \textit{STAnDS} which incorporates a residual error spatial detector, with a time-based expected change detection. This approach is evaluated using a simulated quantitative environment and the results show that \textit{STAnDS} is effective at detecting multiple attack types.

['David Wallom', 'Devki Jha', 'Martin Higgins']

2022-12-15

['change-detection']

['computer-vision']

[ 7.69434497e-02 -1.55150250e-01 1.27318934e-01 -3.13252419e-01 -1.51375130e-01 -6.85887098e-01 7.77034283e-01 3.27670604e-01 -8.07575107e-01 6.57326281e-01 -5.82623243e-01 -9.13613856e-01 -1.78049698e-01 -1.06013548e+00 -5.63729823e-01 -6.34755433e-01 -2.01467514e-01 -3.13502252e-02 1.13467968e+00 -9.72858816e-02 4.94249195e-01 1.08173931e+00 -1.85551012e+00 -3.94942641e-01 7.45845437e-01 1.28469825e+00 -3.79004836e-01 7.62380719e-01 -5.42388894e-02 1.36749968e-01 -7.44730055e-01 -2.36714289e-01 3.43226522e-01 4.58859235e-01 1.48862869e-01 -6.32765353e-01 1.33857861e-01 -6.56289101e-01 -2.43633449e-01 1.34676421e+00 1.62633091e-01 2.75300490e-03 6.43663168e-01 -1.93179548e+00 3.57149392e-01 -2.83062130e-01 -4.42841470e-01 5.65343976e-01 1.87797204e-01 4.10144448e-01 -1.62197445e-02 -7.19132662e-01 1.04153655e-01 1.13647950e+00 5.80355108e-01 2.37404943e-01 -6.93447292e-01 -1.20599484e+00 -4.08490986e-01 5.02983749e-01 -1.41254532e+00 -5.69590688e-01 4.49500114e-01 -3.60342830e-01 7.73273587e-01 1.71429753e-01 1.64158762e-01 7.03113019e-01 1.01676691e+00 -1.69333313e-02 9.42556143e-01 -7.86312595e-02 2.85341620e-01 2.80568480e-01 2.21800253e-01 6.05916321e-01 9.53901172e-01 9.30116057e-01 -1.97900534e-01 -2.31370300e-01 -6.34752810e-02 -1.24414094e-01 1.80339515e-01 -1.94877878e-01 -6.93268836e-01 8.44123304e-01 1.44076109e-01 -2.39141043e-02 -5.19551933e-01 1.34782583e-01 4.16109681e-01 3.35832745e-01 3.75046097e-02 -2.84497112e-01 -8.53168070e-02 -1.97514981e-01 -3.37256908e-01 2.04033822e-01 3.49913865e-01 8.49594116e-01 5.63711941e-01 2.85610735e-01 1.31486058e-01 -2.36947369e-02 6.81595802e-01 1.36945868e+00 8.89907330e-02 -5.91329575e-01 4.00542378e-01 3.71733069e-01 3.76927555e-01 -1.24894381e+00 -4.96982992e-01 -8.69068056e-02 -4.77051258e-01 1.01440287e+00 1.04847522e-02 -2.81381756e-01 -9.72724259e-01 1.24863291e+00 5.90608239e-01 4.75560039e-01 1.58646524e-01 3.65311921e-01 3.54534626e-01 5.00048995e-01 2.13487372e-01 -1.79783791e-01 1.08444023e+00 3.27185467e-02 -9.60504949e-01 -6.96410313e-02 4.72120345e-01 -6.63682044e-01 1.04827203e-01 4.48889524e-01 -4.91814464e-01 -6.11299276e-01 -1.54216099e+00 6.05318785e-01 -8.60160947e-01 -4.63998109e-01 -1.34845689e-01 1.38735390e+00 -5.32655180e-01 1.73370227e-01 -6.54281318e-01 -1.97030723e-01 4.27831084e-01 4.52367187e-01 -2.21755743e-01 -2.63336375e-02 -1.30448246e+00 1.20065641e+00 1.69120669e-01 1.41511947e-01 -7.88370907e-01 -2.50780433e-01 -8.96843493e-01 -5.76436937e-01 3.43388498e-01 -1.98919959e-02 9.27519619e-01 4.55229640e-01 -1.03138387e+00 3.73345256e-01 -2.75730014e-01 -1.14041471e+00 6.63262248e-01 -2.54206806e-01 -1.26469779e+00 3.26702446e-02 2.89046317e-01 2.16982424e-01 9.56106126e-01 -9.55463052e-01 -1.13660216e+00 -6.32565081e-01 -6.57525003e-01 -4.62354571e-01 1.53356731e-01 1.06703803e-01 4.67409462e-01 1.31234661e-01 1.17959939e-01 -9.00008321e-01 -1.10062309e-01 4.98114489e-02 -4.94560421e-01 -1.19091056e-01 1.76662052e+00 -1.04112670e-01 9.63148892e-01 -2.30576873e+00 -1.03480804e+00 6.39045000e-01 3.82230766e-02 9.32674885e-01 3.63348663e-01 3.00301582e-01 3.96749437e-01 -4.94203679e-02 -5.81374392e-02 -1.26978941e-02 -2.00826138e-01 3.72831941e-01 -3.60574722e-01 9.02319908e-01 1.01775922e-01 4.65143800e-01 -6.00741923e-01 -4.79801185e-02 9.76525366e-01 4.73584712e-01 -7.21039437e-03 -3.90622020e-02 3.25982779e-01 1.49854466e-01 -8.31645310e-01 5.50776958e-01 9.78614151e-01 9.53064978e-01 -8.56990695e-01 -5.92622790e-04 -7.77411759e-01 2.84601271e-01 -1.23332107e+00 2.74227411e-01 -5.00907674e-02 9.69689131e-01 1.49257004e-01 -7.62741387e-01 1.16021764e+00 2.69201487e-01 1.59238532e-01 -9.41777766e-01 5.41168928e-01 4.60413575e-01 -2.01466903e-02 -5.82093298e-01 5.57062745e-01 1.69919103e-01 -6.97279125e-02 9.70782191e-02 -7.13110924e-01 2.14236937e-02 -2.72955507e-01 1.39234409e-01 1.27449274e+00 -4.88287479e-01 1.05852619e-01 -4.59164456e-02 9.58446622e-01 3.92292924e-02 3.72646511e-01 7.19574869e-01 -7.74529517e-01 -3.38571310e-01 -1.52858913e-01 -2.19305739e-01 -9.25705314e-01 -1.32317507e+00 -5.52772403e-01 1.06987730e-01 5.34588337e-01 7.57764578e-02 -2.94897765e-01 -6.88853562e-01 5.55318892e-01 1.14381170e+00 -1.73772112e-01 -6.73740506e-01 -3.26860785e-01 -1.35880366e-01 1.09598458e+00 3.06123853e-01 8.10689211e-01 -4.87016261e-01 -1.10165143e+00 3.17263335e-01 3.88800561e-01 -1.35045600e+00 1.80787757e-01 -4.40346375e-02 -3.65239739e-01 -1.17701197e+00 4.36277807e-01 1.02165006e-02 3.54397863e-01 9.16426301e-01 2.01014176e-01 1.48782715e-01 -5.70954025e-01 4.36040312e-01 -6.41464368e-02 -1.34355736e+00 -7.34053731e-01 -8.06080341e-01 5.86408973e-01 2.09853664e-01 8.82878184e-01 -2.24626288e-01 -3.82322550e-01 5.67440689e-01 -6.40292585e-01 -8.09733152e-01 2.16629282e-01 1.40617013e-01 7.83073232e-02 6.70898438e-01 6.56495333e-01 -5.49798429e-01 5.83618939e-01 -5.40651739e-01 -1.13243687e+00 -4.88043159e-01 -8.94927979e-01 -2.46034488e-01 2.87008762e-01 -8.63072798e-02 -5.89388192e-01 -1.91839859e-01 -2.13153541e-01 -1.36022359e-01 -7.12021410e-01 -1.16861612e-01 -3.26701403e-01 -4.88474160e-01 5.46551764e-01 2.02416796e-02 3.60431314e-01 -9.60734561e-02 -2.80005634e-02 8.55835080e-01 6.87313795e-01 2.84133583e-01 1.48586762e+00 7.66002834e-01 5.73350966e-01 -1.42538548e+00 -3.86050530e-02 -6.17967725e-01 -4.35778916e-01 -5.92852712e-01 8.63887250e-01 -7.35219836e-01 -8.86690736e-01 6.64720356e-01 -1.14707410e+00 5.09977639e-01 1.65647343e-01 6.20406687e-01 4.63770293e-02 3.87783974e-01 1.42078117e-01 -1.46170330e+00 7.84202814e-02 -1.05060792e+00 8.28925133e-01 2.81205267e-01 6.18628301e-02 -6.73295617e-01 -2.11441323e-01 8.92461985e-02 6.36755049e-01 4.54334497e-01 5.78410864e-01 -7.89431155e-01 -7.27387190e-01 -8.07516813e-01 -2.24707201e-01 3.64039212e-01 1.54864132e-01 1.99186578e-01 -8.91500354e-01 -1.18727431e-01 2.58293092e-01 4.22024876e-01 5.42417407e-01 1.96672440e-01 5.10478020e-01 -1.39604822e-01 -7.58760750e-01 3.13838601e-01 1.19329858e+00 8.84331286e-01 8.73222709e-01 3.49493325e-01 3.34521651e-01 5.47036409e-01 1.09271598e+00 1.95326358e-01 1.96793586e-01 6.54922605e-01 9.55824494e-01 3.46687198e-01 2.70496190e-01 2.49773115e-02 5.21109998e-01 -3.20051461e-01 4.20529068e-01 -3.04946393e-01 -6.19161844e-01 3.81715775e-01 -1.40925825e+00 -1.20415556e+00 -8.07053149e-01 2.62077260e+00 -2.12035626e-01 8.03124428e-01 7.89728016e-02 7.46548712e-01 8.71512413e-01 -6.99997097e-02 -5.62485218e-01 -6.79377317e-01 3.29453588e-01 5.32963639e-03 1.31888783e+00 6.72988355e-01 -1.06217170e+00 5.75132370e-01 5.53654289e+00 4.41107363e-01 -1.17916501e+00 -3.46431136e-02 1.01311030e-02 2.61560112e-01 1.24538697e-01 -1.36424676e-01 -1.21418798e+00 6.51988268e-01 1.36341202e+00 -1.09615535e-01 -2.38979101e-01 5.86681068e-01 5.88648558e-01 -5.18988371e-01 -4.06788141e-01 5.65707803e-01 -1.92922354e-01 -8.26126456e-01 -3.23473394e-01 4.60901022e-01 -1.35824487e-01 1.50075480e-01 -4.18336503e-02 3.81146255e-03 3.13069642e-01 -8.18216443e-01 5.57924509e-01 1.52687997e-01 2.84930885e-01 -1.08909082e+00 8.08877409e-01 5.26068389e-01 -1.28698123e+00 -2.40073755e-01 -2.50755370e-01 -4.66726981e-02 7.17033863e-01 5.48419595e-01 -1.05097842e+00 5.03246486e-01 5.02399743e-01 8.01748261e-02 -5.10711133e-01 9.91085529e-01 -9.16787311e-02 6.45720482e-01 -8.02058458e-01 -2.90690869e-01 2.94618011e-01 -1.02326553e-02 1.31478691e+00 7.65688062e-01 6.09257817e-01 -2.52580009e-02 -7.19087897e-04 5.49461126e-01 8.15366626e-01 -4.50001091e-01 -1.35819733e+00 3.45533192e-01 1.00144053e+00 8.27947736e-01 -4.54412520e-01 6.38160259e-02 -2.55902141e-01 2.53538221e-01 -6.04399681e-01 2.00785145e-01 -1.02525353e+00 -7.54438698e-01 1.01249599e+00 5.35113275e-01 3.01649004e-01 -7.55069613e-01 -1.03101254e-01 -1.00698777e-01 -7.74395157e-05 -2.29461625e-01 4.22685295e-02 -3.93231452e-01 -7.13610470e-01 3.46910954e-01 2.20269457e-01 -1.41317284e+00 -2.13872254e-01 -8.61863315e-01 -5.93864918e-01 6.54052615e-01 -1.59016275e+00 -7.34952390e-01 -2.72633791e-01 8.97287607e-01 7.63830692e-02 -6.64740324e-01 3.40034783e-01 4.84645277e-01 -4.42709297e-01 3.67581427e-01 -1.66655689e-01 6.10955693e-02 3.38079542e-01 -4.94055033e-01 4.21286702e-01 1.37382603e+00 -2.56790638e-01 2.65160084e-01 1.10993266e+00 -1.03416812e+00 -1.45641899e+00 -1.18523598e+00 5.73322475e-01 -2.50128806e-01 5.97204328e-01 -1.40366405e-01 -6.60009623e-01 3.07384938e-01 -6.70831203e-02 2.79713199e-02 2.27736592e-01 -6.64602995e-01 -1.98506191e-01 -3.10368687e-01 -1.70242119e+00 3.75167161e-01 5.58388889e-01 -3.38017523e-01 -4.32354897e-01 -5.24820834e-02 5.25700271e-01 -1.26494458e-02 -1.75179824e-01 7.65819252e-01 5.74290693e-01 -1.20306063e+00 9.30544317e-01 -2.24709436e-01 -7.97978163e-01 -6.72011495e-01 -1.88092679e-01 -9.56041634e-01 5.49094155e-02 -5.74721634e-01 2.30110615e-01 1.13990533e+00 2.31565341e-01 -1.48917520e+00 5.29264748e-01 3.84631217e-01 -9.27724615e-02 1.19353734e-01 -1.64389372e+00 -9.89230275e-01 -4.59193110e-01 -9.58831072e-01 5.91550171e-01 3.91623139e-01 -4.48722243e-01 1.63597465e-01 -2.05869064e-01 1.12503147e+00 1.01808214e+00 -9.18762743e-01 1.01478016e+00 -1.59951484e+00 6.03606522e-01 -2.33290628e-01 -1.21002901e+00 -5.92852652e-01 -1.97020411e-01 -7.61916041e-02 3.83190364e-01 -1.01480722e+00 -9.37197685e-01 -6.29430473e-01 -1.41718522e-01 -5.15844934e-02 2.59543151e-01 3.56113493e-01 -3.67152482e-01 -1.99721947e-01 -6.43479154e-02 3.28913331e-01 3.39806348e-01 3.96391265e-02 1.17170878e-01 6.16686761e-01 3.95548567e-02 8.25044572e-01 8.19377720e-01 -8.43782127e-01 -4.61689651e-01 1.27127558e-01 6.21917062e-02 -3.51024382e-02 7.17347562e-01 -1.58270788e+00 4.66714263e-01 -8.42485279e-02 4.62669246e-02 -1.13618422e+00 4.77353126e-01 -1.42804468e+00 2.19486537e-03 9.98778582e-01 3.55403394e-01 4.30956870e-01 5.41062117e-01 9.45020974e-01 -4.90833782e-02 -2.85383224e-01 9.24652457e-01 4.79912907e-01 -7.22767591e-01 2.74689794e-01 -1.06840837e+00 -6.24189317e-01 1.76034641e+00 -6.18170440e-01 -6.78678215e-01 -4.36314374e-01 -9.05578434e-02 2.65395552e-01 7.60957450e-02 5.39293170e-01 8.58134389e-01 -1.14198244e+00 -4.87675130e-01 8.14357102e-01 2.00655386e-01 -2.74626285e-01 -8.50436389e-02 6.26341939e-01 -3.20795655e-01 7.42702127e-01 -1.86768830e-01 -7.56323397e-01 -1.56525052e+00 5.04111648e-01 2.76113510e-01 3.94248813e-01 -3.69532108e-01 4.65771735e-01 -4.22218323e-01 -3.63265304e-03 2.19907477e-01 -2.27271631e-01 -3.22894841e-01 -2.66765445e-01 7.97277868e-01 7.59240508e-01 2.12908491e-01 -1.09643793e+00 -5.87029994e-01 5.45603096e-01 -8.07270408e-02 -3.19785386e-01 4.83060569e-01 -4.55586314e-01 1.95454761e-01 2.76883423e-01 6.80089056e-01 3.80648762e-01 -9.06187117e-01 -7.75774661e-03 2.47846559e-01 -6.12522960e-01 2.97245711e-01 -3.04816246e-01 -6.69131458e-01 8.58381689e-01 1.14453876e+00 7.81224489e-01 5.61379731e-01 -5.46890378e-01 9.42720830e-01 4.60272253e-01 8.46047163e-01 -7.75945544e-01 -3.55857372e-01 5.48421562e-01 1.59892365e-01 -1.24873114e+00 -4.36080545e-02 -5.24966300e-01 -1.94488317e-01 7.57386982e-01 5.23679316e-01 -5.32463044e-02 1.03503180e+00 6.43360615e-01 1.82743728e-01 -1.00210935e-01 -3.99741948e-01 -2.50796169e-01 -2.91835796e-02 1.07389128e+00 -4.25528079e-01 4.58940640e-02 -2.24730372e-01 -1.11240946e-01 -1.49858266e-01 -2.52467841e-01 5.56271136e-01 1.17148983e+00 -1.19530892e+00 -1.02919328e+00 -6.79085135e-01 4.99253809e-01 -4.20020223e-01 3.48994315e-01 -6.58399286e-03 8.97178173e-01 2.30519965e-01 1.56198776e+00 1.71134770e-01 -6.75938189e-01 5.49529552e-01 1.64378896e-01 -1.43246308e-01 -1.46080405e-01 6.94977045e-02 -4.71912026e-01 3.59199405e-01 -7.27171779e-01 -7.70444497e-02 -7.32048869e-01 -1.45828569e+00 -3.72594684e-01 -2.90260822e-01 2.08312526e-01 1.14868057e+00 1.34076118e+00 4.19848382e-01 3.67087334e-01 7.87394524e-01 -2.91907847e-01 -5.39486051e-01 -6.62323952e-01 -5.34384727e-01 -8.64076316e-02 8.08302164e-01 -1.23668861e+00 -6.25249922e-01 -6.85086370e-01]

[5.328400611877441, 7.341818332672119]

38f05644-3d2b-49ad-9abd-e99139aefb3e

counterfactual-explanations-and-predictive

2306.03980

https://arxiv.org/abs/2306.03980v1

https://arxiv.org/pdf/2306.03980v1.pdf

Counterfactual Explanations and Predictive Models to Enhance Clinical Decision-Making in Schizophrenia using Digital Phenotyping

Clinical practice in psychiatry is burdened with the increased demand for healthcare services and the scarce resources available. New paradigms of health data powered with machine learning techniques could open the possibility to improve clinical workflow in critical stages of clinical assessment and treatment in psychiatry. In this work, we propose a machine learning system capable of predicting, detecting, and explaining individual changes in symptoms of patients with Schizophrenia by using behavioral digital phenotyping data. We forecast symptoms of patients with an error rate below 10%. The system detects decreases in symptoms using changepoint algorithms and uses counterfactual explanations as a recourse in a simulated continuous monitoring scenario in healthcare. Overall, this study offers valuable insights into the performance and potential of counterfactual explanations, predictive models, and change-point detection within a simulated clinical workflow. These findings lay the foundation for further research to explore additional facets of the workflow, aiming to enhance its effectiveness and applicability in real-world healthcare settings. By leveraging these components, the goal is to develop an actionable, interpretable, and trustworthy integrative decision support system that combines real-time clinical assessments with sensor-based inputs.

['Omar Costilla-Reyes', 'Francisco Gomez', 'Juan Sebastian Canas']

2023-06-06

['change-point-detection']

['time-series']

[ 4.78951931e-01 5.41680276e-01 -1.29700989e-01 -6.79173410e-01 -2.85013944e-01 -2.29427695e-01 3.35494488e-01 8.16274285e-01 -2.39966780e-01 9.31198478e-01 3.61124873e-01 -5.28378189e-01 -6.94471657e-01 -5.33909917e-01 -9.07474607e-02 -5.43032527e-01 -3.47356528e-01 7.22856760e-01 -5.88602722e-01 2.14545920e-01 1.11104548e-01 6.67287409e-01 -1.33459044e+00 9.08991456e-01 8.84431303e-01 8.20648313e-01 1.31722510e-01 4.59240943e-01 4.29004759e-01 7.47077346e-01 -3.75374079e-01 -2.54670501e-01 -2.44342368e-02 -4.22771007e-01 -3.83275986e-01 4.32375260e-02 -3.50541830e-01 -6.18921876e-01 3.15146953e-01 5.25396407e-01 7.08815932e-01 -5.01177132e-01 3.53233308e-01 -1.22407496e+00 -3.70909691e-01 6.64180458e-01 1.23522043e-01 4.96870615e-02 6.94004238e-01 7.44153440e-01 3.29702765e-01 -3.34162652e-01 4.54021841e-01 8.98287058e-01 8.79706562e-01 6.72751546e-01 -1.21843708e+00 -6.38236821e-01 6.41665468e-03 3.01062703e-01 -8.11273813e-01 -5.31491697e-01 1.29893586e-01 -6.61371052e-01 9.94181514e-01 5.96021235e-01 1.25239897e+00 1.27960336e+00 8.08256567e-01 1.23759083e-01 1.34605503e+00 -2.77040333e-01 7.10332870e-01 1.19738966e-01 -4.10371870e-02 2.57985502e-01 5.88113070e-01 1.33451253e-01 -4.74639565e-01 -7.02907205e-01 5.78480065e-01 7.86248207e-01 -2.90644854e-01 2.45901138e-01 -1.16540515e+00 5.28072655e-01 2.69928612e-02 3.21568638e-01 -8.85964692e-01 -1.97737023e-01 3.53238583e-01 6.22469932e-02 5.81412971e-01 4.47398841e-01 -8.81932557e-01 -5.23690462e-01 -1.04577541e+00 3.08546484e-01 6.27246022e-01 2.59126902e-01 -2.80077010e-01 -8.15692544e-02 -5.93871415e-01 2.77850747e-01 4.53148365e-01 8.91835630e-01 8.17220867e-01 -1.00530064e+00 2.38891378e-01 1.02181983e+00 3.20654303e-01 -7.54728854e-01 -1.04528296e+00 -3.49920005e-01 -6.57311797e-01 1.87768098e-02 1.69950008e-01 -5.24707794e-01 -5.79423547e-01 1.28518736e+00 3.57313931e-01 1.47358432e-01 8.51889029e-02 6.96784496e-01 1.79892063e-01 -1.17886893e-01 2.63604611e-01 -8.56113374e-01 1.39286721e+00 1.00429989e-01 -1.07707262e+00 -6.96654664e-03 9.76736188e-01 -1.32833928e-01 8.86508107e-01 4.26031619e-01 -1.10123599e+00 1.26507625e-01 -5.00078022e-01 5.26433825e-01 -1.09042348e-02 1.34877302e-02 7.57336259e-01 8.36840570e-01 -8.51705134e-01 8.49595070e-01 -1.34505653e+00 -5.84774196e-01 7.08770633e-01 3.94714177e-01 -1.94692656e-01 1.06429644e-02 -9.25514936e-01 8.96964848e-01 1.44065931e-01 3.24361920e-01 -5.90813816e-01 -1.19282615e+00 -5.40306985e-01 1.56561300e-01 -2.14216903e-01 -1.01239729e+00 1.32108665e+00 -7.14152753e-01 -1.27028120e+00 6.30359948e-01 3.97835150e-02 -6.16900206e-01 8.51111591e-01 1.46190017e-01 -5.62363148e-01 8.24696049e-02 1.30164819e-02 1.50080308e-01 2.52542913e-01 -5.87718129e-01 -4.87686187e-01 -1.17069328e+00 -6.61658585e-01 1.29447377e-03 -2.20992014e-01 -5.20507544e-02 9.27532136e-01 -1.27978221e-01 6.24530986e-02 -7.77014077e-01 -5.43050528e-01 -9.77184102e-02 -1.13630272e-01 3.69208694e-01 6.01802230e-01 -6.96871042e-01 1.17377913e+00 -1.74839616e+00 -6.43911064e-01 8.06719884e-02 2.49569744e-01 3.54067892e-01 6.01459503e-01 6.33641779e-01 -2.48572290e-01 -3.78871374e-02 -1.76094294e-01 -5.91522716e-02 -2.68991977e-01 -1.66929364e-02 -2.16824606e-01 5.35954595e-01 1.22042939e-01 1.00657034e+00 -9.89918113e-01 2.60108784e-02 3.52790803e-01 4.42116737e-01 -5.12713611e-01 2.08312616e-01 -4.82147411e-02 6.87039614e-01 -3.17206204e-01 7.27533579e-01 3.84257585e-01 -3.46896678e-01 7.90576339e-01 3.27459335e-01 -1.33072302e-01 2.87786901e-01 -9.37503576e-01 1.13194871e+00 -1.76062241e-01 2.08833471e-01 -1.29093006e-01 -8.84197414e-01 8.15307736e-01 7.69700348e-01 7.87484705e-01 -6.56726122e-01 8.68537351e-02 2.01354221e-01 1.90433457e-01 -1.25936651e+00 -1.91304922e-01 -5.26611149e-01 2.99548745e-01 5.09239435e-01 -5.66285133e-01 5.69394231e-02 -5.66072702e-01 -2.30498970e-01 1.58879852e+00 -1.34613127e-01 7.78068841e-01 5.12227714e-02 -7.30260536e-02 -6.56479821e-02 7.07661748e-01 5.94052374e-01 -4.62147593e-01 1.70497611e-01 3.82097483e-01 -6.96552277e-01 -7.90302277e-01 -8.00534308e-01 -3.86360258e-01 2.73557067e-01 -7.27158248e-01 -1.64968427e-02 -5.56410015e-01 -2.36417726e-01 3.44936520e-01 8.94121349e-01 -5.24568915e-01 -4.19933826e-01 1.15723833e-01 -1.21514571e+00 4.47766304e-01 6.75587118e-01 -9.08225477e-02 -1.25975573e+00 -1.56381106e+00 6.44111335e-01 -1.04472585e-01 -8.49231482e-01 3.95963639e-01 -8.11675414e-02 -1.54715729e+00 -1.17186475e+00 -3.99783731e-01 1.55446365e-01 5.45468807e-01 -1.48453027e-01 6.70844078e-01 -2.12357938e-01 -4.63130027e-01 3.38205010e-01 -1.14391983e-01 -8.70666921e-01 -2.88437933e-01 -7.40067244e-01 3.53675604e-01 -9.87375006e-02 6.38773084e-01 -6.89937353e-01 -1.05744147e+00 -2.32961416e-01 -7.75574565e-01 -2.48461892e-03 4.35590088e-01 6.47352636e-01 3.52286071e-01 -4.21349555e-01 9.52454209e-01 -1.02217245e+00 1.07712579e+00 -9.04343545e-01 -2.81961650e-01 1.54392987e-01 -1.29779470e+00 -3.10603350e-01 4.94469166e-01 -2.00353190e-01 -1.14046371e+00 -2.42180601e-02 1.52547687e-01 -6.98941275e-02 -4.35116142e-01 6.98065102e-01 1.22609921e-01 6.50393128e-01 7.78356194e-01 -7.02662915e-02 2.05729365e-01 -3.64422023e-01 5.98989753e-03 1.15618277e+00 1.13504261e-01 1.40233144e-01 -1.15199469e-01 8.60394597e-01 6.45730719e-02 -4.03065026e-01 -2.97063679e-01 -3.76900762e-01 -3.59794676e-01 -3.96435648e-01 5.68260968e-01 -8.63155663e-01 -1.26162899e+00 1.17949434e-01 -8.89927685e-01 -3.70378524e-01 -3.06612045e-01 7.61727631e-01 -5.73306561e-01 -8.55054706e-02 -2.66605824e-01 -1.43260956e+00 -7.15494037e-01 -9.86790419e-01 8.91449094e-01 3.79445516e-02 -9.09563720e-01 -9.54668403e-01 2.70627588e-01 4.24859405e-01 3.52240622e-01 6.72759116e-01 6.39932394e-01 -9.61172640e-01 -2.17502993e-02 -4.31120157e-01 1.18172109e-01 -2.08690643e-01 5.18758297e-01 5.75040989e-02 -9.97498810e-01 -2.91444898e-01 4.04994845e-01 -7.79474750e-02 1.56746134e-01 8.06075871e-01 1.13969898e+00 -6.99249804e-01 -5.07406294e-01 1.57118708e-01 1.08560300e+00 6.94952190e-01 4.59509730e-01 2.30523899e-01 -1.27742011e-02 6.72284782e-01 5.50810277e-01 1.27009940e+00 6.19271278e-01 4.35066789e-01 5.26464939e-01 6.82154819e-02 4.71673489e-01 1.31884068e-01 1.82073310e-01 3.19634795e-01 -2.69458085e-01 2.96414971e-01 -1.23457253e+00 4.61685926e-01 -2.11039138e+00 -1.21855617e+00 -3.06889594e-01 2.34113026e+00 6.59306526e-01 -1.49580330e-01 2.83977482e-02 3.93526584e-01 3.77268285e-01 -8.15587521e-01 -6.97309911e-01 -7.50611424e-01 3.73504698e-01 -4.21535969e-02 3.08847159e-01 -7.50517324e-02 -4.03585166e-01 1.95755139e-02 6.71532059e+00 -5.12105286e-01 -1.14358485e+00 2.91026324e-01 8.81536543e-01 -5.10433495e-01 -3.21983546e-01 -2.59559929e-01 -2.36630768e-01 7.90767133e-01 1.66457570e+00 9.44652036e-02 2.70673096e-01 6.25441909e-01 1.60343874e+00 -1.44855991e-01 -1.10619628e+00 7.31773853e-01 -2.86438972e-01 -1.39522922e+00 -4.67567146e-01 2.21990675e-01 5.72648227e-01 -1.13277026e-01 -4.95940484e-02 -2.00287700e-01 2.20120400e-01 -9.17826056e-01 5.30504227e-01 1.22891462e+00 8.13494027e-01 -3.46192509e-01 1.05859768e+00 4.02728319e-01 -2.51452297e-01 -6.78542733e-01 7.53755197e-02 -7.80094206e-01 9.94842574e-02 1.04680860e+00 -1.69661784e+00 4.35948581e-01 6.42255068e-01 5.25066137e-01 -8.06913003e-02 1.05642366e+00 9.59644541e-02 7.78893411e-01 1.90303195e-02 1.80798069e-01 -1.13903992e-01 -1.74930483e-01 1.15563750e-01 1.02520144e+00 5.46435058e-01 5.44957876e-01 -2.18630537e-01 7.52478540e-01 3.67056280e-01 -3.32717528e-03 -8.08634460e-01 -1.27878770e-01 5.25887311e-01 1.02283478e+00 -7.22005844e-01 -4.28440958e-01 -2.36201331e-01 5.37603080e-01 5.93275838e-02 3.19592953e-02 -5.45694113e-01 2.77331561e-01 6.18604600e-01 6.73214555e-01 -4.28847522e-01 3.21536452e-01 -1.03354633e+00 -9.56889927e-01 -8.47995467e-03 -9.97869015e-01 5.58365822e-01 -7.17347383e-01 -9.10061717e-01 1.20900646e-01 -1.55414790e-02 -1.08158195e+00 -5.54768145e-01 -3.39510322e-01 -5.50966322e-01 7.29600549e-01 -1.17465281e+00 -9.01052654e-01 -3.70361358e-01 4.01562303e-01 7.85357207e-02 4.38751020e-02 1.22482049e+00 -4.77076694e-02 -6.63926959e-01 9.20792967e-02 3.73690754e-01 -4.72859651e-01 7.03761041e-01 -9.12165344e-01 -2.63646960e-01 2.90761262e-01 -5.97459078e-01 5.68296015e-01 6.75251186e-01 -9.86796916e-01 -1.04780209e+00 -1.04831100e+00 1.20842636e+00 -3.34001422e-01 5.03867447e-01 5.05624972e-02 -8.91421437e-01 7.23202527e-01 -2.15273768e-01 -1.88755855e-01 1.12403667e+00 8.84257257e-02 3.67989838e-01 -2.90193230e-01 -1.78645277e+00 5.06313086e-01 6.43856287e-01 -2.67276764e-01 -6.77106738e-01 5.90363741e-01 3.51204962e-01 -1.59431249e-01 -1.16125321e+00 4.71085936e-01 9.45722938e-01 -8.62056494e-01 5.46185613e-01 -1.17077553e+00 4.03652102e-01 2.60272056e-01 2.42342770e-01 -1.39546776e+00 -8.71455669e-02 -4.73053932e-01 -9.35138389e-02 7.68732965e-01 3.71608377e-01 -9.56412435e-01 5.07674873e-01 1.42872357e+00 -1.39123984e-02 -9.97312129e-01 -8.83244455e-01 -2.80527383e-01 -4.35975641e-01 -5.93608916e-01 8.96498024e-01 1.09717929e+00 9.11940932e-01 -3.95526230e-01 4.22469527e-03 2.98223615e-01 2.73831636e-01 -6.44455925e-02 8.42209756e-02 -1.41479671e+00 -6.52227253e-02 -2.58011464e-02 -5.66978276e-01 4.80242699e-01 -3.52392167e-01 -6.91444278e-01 -4.78424072e-01 -1.76719701e+00 3.95961255e-01 -2.96669573e-01 -2.99146980e-01 7.32002676e-01 -1.63837954e-01 -8.07126984e-02 -7.12463818e-03 2.71555632e-01 -4.81638871e-02 3.17646772e-01 1.10727203e+00 8.29069987e-02 -6.11289561e-01 1.79964811e-01 -8.06299746e-01 7.48102069e-01 1.00569129e+00 -6.47377849e-01 -4.47787851e-01 -1.16097957e-01 3.25819373e-01 3.95575047e-01 5.44492245e-01 -7.22834408e-01 8.15533847e-02 -5.25736213e-01 7.18036354e-01 -1.92012474e-01 2.00199634e-01 -9.87005830e-01 5.57482421e-01 1.15918612e+00 -3.30693245e-01 2.50673205e-01 1.79176852e-01 4.76081491e-01 3.21448445e-01 6.24792129e-02 4.85205203e-01 -2.68853128e-01 -1.91887431e-02 -7.01449960e-02 -7.10916340e-01 -4.63146299e-01 1.41891170e+00 -2.58416951e-01 -3.45655322e-01 -2.67769843e-01 -1.22085369e+00 -2.13522445e-02 6.92645460e-02 -2.63621937e-02 7.17379212e-01 -9.31539595e-01 -7.44735599e-01 1.57753229e-01 1.24415681e-01 -4.80108231e-01 5.02943397e-01 1.38539100e+00 -4.00288731e-01 7.41817176e-01 -3.76850784e-01 -5.51198900e-01 -1.17865789e+00 3.91090304e-01 3.72300804e-01 -1.66642502e-01 -7.11679935e-01 7.13747144e-02 -5.78597963e-01 -2.76448309e-01 -5.60244396e-02 -6.39876306e-01 -1.38050422e-01 1.77525237e-01 9.44489419e-01 4.97619420e-01 5.76133132e-01 7.72527903e-02 -5.73476136e-01 -4.67953414e-01 1.16673067e-01 -5.49046099e-02 1.67034376e+00 -1.66103661e-01 -3.33860852e-02 6.04111075e-01 3.01795185e-01 -4.95421559e-01 -1.10338187e+00 3.17353368e-01 2.51544476e-01 -3.12821746e-01 -6.76190406e-02 -1.39369321e+00 -5.01683593e-01 6.28132701e-01 1.18515944e+00 2.75043607e-01 1.27365232e+00 -2.78639585e-01 4.58153896e-02 1.35211289e-01 1.46795899e-01 -1.11564410e+00 -3.61053318e-01 -2.87123770e-01 8.17953110e-01 -1.19555604e+00 8.10867399e-02 1.55792624e-01 -7.85034478e-01 1.01594591e+00 8.61197263e-02 4.79222566e-01 5.51919222e-01 2.86507756e-01 2.05768630e-01 -4.54348862e-01 -1.27463400e+00 2.17306882e-01 -5.02145886e-01 8.43064487e-01 5.64650655e-01 8.65832090e-01 -6.99179471e-01 9.95600402e-01 -3.07846740e-02 7.68195331e-01 6.94456100e-01 9.40997303e-01 -2.12110326e-01 -8.25015724e-01 -6.51423991e-01 1.01363850e+00 -5.03402889e-01 -6.41126558e-02 -2.54730225e-01 3.96200538e-01 2.70785391e-01 1.15195036e+00 5.90207428e-02 -1.67155135e-02 5.90119421e-01 5.43264031e-01 1.77533597e-01 -6.53130233e-01 -6.96951389e-01 -1.84851930e-01 1.23620711e-01 -7.86023915e-01 -4.30856317e-01 -1.12899363e+00 -1.30392992e+00 -1.14680298e-01 1.19746715e-01 -2.04312712e-01 1.04032910e+00 1.20037150e+00 1.00944948e+00 8.94275665e-01 4.12043095e-01 -3.42516243e-01 -7.62730241e-01 -1.09906340e+00 -4.62028593e-01 2.31868297e-01 1.08106129e-01 -4.40354943e-01 1.44448886e-02 1.68922059e-02]

[8.272087097167969, 5.831803798675537]

058ca04e-54dd-4c00-8101-a1162f51a000

detecting-anomalies-within-time-series-using

2202.03944

https://arxiv.org/abs/2202.03944v2

https://arxiv.org/pdf/2202.03944v2.pdf

Detecting Anomalies within Time Series using Local Neural Transformations

We develop a new method to detect anomalies within time series, which is essential in many application domains, reaching from self-driving cars, finance, and marketing to medical diagnosis and epidemiology. The method is based on self-supervised deep learning that has played a key role in facilitating deep anomaly detection on images, where powerful image transformations are available. However, such transformations are widely unavailable for time series. Addressing this, we develop Local Neural Transformations(LNT), a method learning local transformations of time series from data. The method produces an anomaly score for each time step and thus can be used to detect anomalies within time series. We prove in a theoretical analysis that our novel training objective is more suitable for transformation learning than previous deep Anomaly detection(AD) methods. Our experiments demonstrate that LNT can find anomalies in speech segments from the LibriSpeech data set and better detect interruptions to cyber-physical systems than previous work. Visualization of the learned transformations gives insight into the type of transformations that LNT learns.

['Maja Rudolph', 'Stephan Mandt', 'Steffen Staab', 'Decky Aspandi Latif', 'Marius Kloft', 'Chen Qiu', 'Tim Schneider']

2022-02-08

['epidemiology']

['medical']

[ 1.37451753e-01 -1.24518998e-01 2.55271107e-01 -2.38611847e-01 -4.30343807e-01 -5.28318107e-01 7.87412167e-01 3.59380126e-01 -3.11681367e-02 1.06300190e-01 5.11249341e-03 -7.15210795e-01 -2.03637213e-01 -8.58331800e-01 -8.98721099e-01 -6.81836247e-01 -6.43026710e-01 2.01316942e-02 1.15140162e-01 -2.74652779e-01 9.66505855e-02 9.54817712e-01 -1.54640603e+00 1.87735587e-01 6.39071345e-01 1.14899635e+00 -5.48434079e-01 6.80943608e-01 -2.09560826e-01 8.53363931e-01 -8.99409890e-01 2.28567310e-02 3.22139233e-01 -4.48454320e-01 -5.77321708e-01 -1.46741465e-01 3.61374944e-01 -3.92646134e-01 -4.83898968e-01 1.04368854e+00 1.75637200e-01 1.58997431e-01 6.71862721e-01 -1.85750020e+00 -5.91717780e-01 2.81702608e-01 -4.39523906e-01 1.00243676e+00 2.43408382e-01 1.89222887e-01 7.08785295e-01 -4.94506091e-01 2.85905182e-01 1.24161398e+00 8.48806024e-01 3.20760548e-01 -9.90956843e-01 -8.55715930e-01 8.96487534e-02 7.34207392e-01 -9.05142725e-01 -9.92981866e-02 9.14777458e-01 -4.07455415e-01 1.02298796e+00 4.16417181e-01 6.49069786e-01 1.23904335e+00 5.78979909e-01 7.94803917e-01 6.65312409e-01 -2.56618172e-01 2.75538087e-01 -4.71231312e-01 -2.74072289e-02 5.26492894e-01 -2.16992185e-01 8.56335238e-02 -3.49139899e-01 -9.85060707e-02 6.74368978e-01 4.30877656e-01 1.22510232e-01 2.43933615e-03 -1.28472674e+00 7.76744843e-01 3.76799047e-01 6.31336391e-01 -5.23217022e-01 2.59864777e-01 7.73925900e-01 9.95828450e-01 8.31433713e-01 5.04976988e-01 -4.68199819e-01 -4.76633370e-01 -5.03676176e-01 8.18096846e-02 3.89776081e-01 4.44608808e-01 4.20982778e-01 5.44680655e-01 -2.02933803e-01 4.75617260e-01 -8.54551196e-02 5.86760163e-01 6.28588915e-01 -8.21768761e-01 5.28478548e-02 6.38980567e-01 -4.34940159e-01 -1.09477186e+00 -6.69769704e-01 -1.63476035e-01 -9.57533836e-01 2.96333104e-01 4.48426783e-01 1.33663593e-02 -1.00530052e+00 1.65528023e+00 2.62669146e-01 6.33588076e-01 -1.43847838e-01 3.86206955e-01 3.41094196e-01 7.51410127e-01 -2.47258127e-01 -2.22309202e-01 8.82636368e-01 -4.59738731e-01 -1.04096401e+00 1.05764270e-01 9.71299648e-01 -5.04754663e-01 1.07160401e+00 5.58258712e-01 -7.30254829e-01 -3.67375731e-01 -9.94347155e-01 3.12897235e-01 -8.68585825e-01 -5.14951169e-01 4.34175581e-01 5.10632396e-01 -1.11979473e+00 7.42286146e-01 -1.32223082e+00 -6.26543880e-01 5.61067045e-01 1.84982240e-01 -3.53595525e-01 2.85487652e-01 -1.16145098e+00 9.04162765e-01 -4.34855260e-02 -2.22263500e-01 -8.86083066e-01 -7.24689543e-01 -9.64052200e-01 -8.72237887e-03 1.37545541e-01 1.64520722e-02 1.36544442e+00 -8.69048357e-01 -1.18883073e+00 6.92144930e-01 -1.23350151e-01 -1.03733897e+00 4.03580666e-01 -1.60054401e-01 -1.02907145e+00 7.50279650e-02 1.13702856e-01 1.26058251e-01 1.25289559e+00 -4.59387332e-01 -6.89566731e-01 -4.15766507e-01 -2.46677190e-01 -4.51780856e-01 -6.82804465e-01 1.15346506e-01 6.12475798e-02 -7.66930282e-01 -5.88016631e-03 -6.83880627e-01 -1.71147570e-01 -5.19243479e-02 -3.61196369e-01 -5.00362515e-01 1.58111763e+00 -8.86886358e-01 1.16662478e+00 -2.37589025e+00 -3.51513267e-01 6.57097936e-01 4.71777618e-01 2.33300611e-01 -2.28225410e-01 4.64880288e-01 -7.20724761e-01 -6.30990416e-02 -3.93465489e-01 2.77957926e-03 1.22091379e-02 3.07285637e-01 -6.81821644e-01 6.65350795e-01 3.95029426e-01 8.50908995e-01 -8.05531859e-01 6.64594769e-02 5.30459285e-01 2.19613835e-01 -3.61934394e-01 2.19244495e-01 -9.98458117e-02 6.34001791e-01 -1.89963818e-01 5.31158566e-01 3.81432146e-01 1.38659496e-02 -6.52825177e-01 1.50227860e-01 -1.37103900e-01 1.75083056e-01 -5.59362710e-01 1.43477237e+00 -3.69775414e-01 1.11916137e+00 -5.40628433e-01 -1.61191630e+00 8.92100036e-01 3.74431640e-01 1.08777535e+00 -1.23494565e+00 1.16412111e-01 -3.86625119e-02 1.57701001e-01 -7.02126503e-01 -7.17960149e-02 3.10925663e-01 1.10616118e-01 7.43560195e-01 -6.05938472e-02 -1.86175108e-02 2.39356905e-01 6.34924844e-02 1.84223008e+00 -3.46059620e-01 1.85157523e-01 3.84959094e-02 4.58530992e-01 -1.44371554e-01 3.04244578e-01 6.91147506e-01 -3.25708568e-01 3.32567692e-01 8.09382141e-01 -9.30179477e-01 -1.28697658e+00 -1.20815361e+00 -5.29694110e-02 9.63759720e-01 -4.21132088e-01 -3.58223826e-01 -8.42747629e-01 -7.70444453e-01 -3.92853357e-02 8.45357358e-01 -7.49281347e-01 -7.27542102e-01 -6.45808756e-01 -4.71282184e-01 6.76530659e-01 6.70861542e-01 2.24114001e-01 -1.48243153e+00 -5.88133454e-01 9.74619389e-02 -6.20849766e-02 -1.18323231e+00 -5.42294562e-01 3.01966399e-01 -8.64729702e-01 -1.10310674e+00 -2.96307206e-01 -4.78041559e-01 6.08587682e-01 -9.32722352e-03 8.23088944e-01 -1.09080151e-01 -6.62387133e-01 8.90351951e-01 -3.31765175e-01 -1.06050956e+00 -7.64943361e-01 -2.27608621e-01 3.56111586e-01 1.60959050e-01 7.94826150e-01 -8.45151424e-01 -2.96424299e-01 1.84658736e-01 -1.16008031e+00 -5.73497474e-01 2.46672288e-01 5.47143638e-01 3.50002080e-01 3.13183159e-01 6.54455721e-01 -5.10889471e-01 8.94668698e-01 -7.04606950e-01 -6.66695595e-01 -8.13532099e-02 -5.66433609e-01 6.86737522e-02 7.49579430e-01 -5.08240700e-01 -4.93377060e-01 -1.99000940e-01 -3.63478750e-01 -7.55786121e-01 -5.00640154e-01 3.71305972e-01 2.00844720e-01 1.42653868e-01 9.00114179e-01 2.68482834e-01 3.38668853e-01 -2.39469364e-01 1.22386925e-01 3.44520837e-01 8.07170153e-01 -2.86493450e-01 1.04912555e+00 6.62225723e-01 9.32888314e-02 -1.03057694e+00 -4.36717421e-01 -4.96515423e-01 -6.07087910e-01 -3.42952520e-01 8.31523836e-01 -4.43197846e-01 -6.55873358e-01 6.03679061e-01 -1.00349748e+00 -3.38784337e-01 -8.20058167e-01 3.24570179e-01 -5.67877412e-01 3.37899059e-01 -4.40465599e-01 -7.05148995e-01 -9.91704240e-02 -6.43093824e-01 1.02638531e+00 -2.07912952e-01 -4.55640346e-01 -1.27243221e+00 3.16841602e-01 -3.71161908e-01 5.36355138e-01 6.74549401e-01 9.98742998e-01 -1.09405375e+00 -2.59848297e-01 -5.11339605e-01 1.25569016e-01 5.96046388e-01 5.22423267e-01 8.83988440e-02 -1.07074392e+00 -2.57131517e-01 2.84818441e-01 1.91625014e-01 5.04039943e-01 4.68303263e-01 1.92612779e+00 -3.75922531e-01 -1.23726614e-01 7.03591824e-01 5.57652831e-01 6.33712173e-01 7.55646706e-01 6.02341056e-01 7.87860036e-01 4.80265588e-01 3.62057805e-01 4.86144811e-01 4.93737906e-02 4.14285511e-01 6.47855818e-01 -4.95472908e-01 2.36031175e-01 4.92123254e-02 6.89103901e-01 8.18846643e-01 -3.76371630e-02 4.84308973e-02 -1.10773134e+00 6.87212110e-01 -1.87351942e+00 -1.11547315e+00 -1.24601163e-01 2.11077666e+00 4.05454934e-01 1.85356289e-01 4.15948570e-01 7.36910045e-01 5.78877926e-01 7.07271276e-03 -9.39391732e-01 -7.08131194e-01 -3.89409065e-02 4.93010253e-01 2.77091384e-01 -4.60288636e-02 -1.25662410e+00 5.33438683e-01 6.83495140e+00 4.77523685e-01 -1.26176167e+00 4.89990823e-02 5.08036911e-01 1.04836978e-01 -8.11034814e-02 -6.05520904e-01 1.66628510e-01 3.89610350e-01 1.43276429e+00 -3.86380166e-01 3.46057981e-01 7.28412330e-01 3.99676740e-01 5.49489737e-01 -1.41108251e+00 1.13420117e+00 -6.33354858e-02 -1.11292326e+00 -4.56925249e-03 1.08128965e-01 5.13711691e-01 8.25729668e-02 2.92702913e-01 2.39501029e-01 1.90871760e-01 -1.13693488e+00 3.23296338e-01 4.04494584e-01 4.78960961e-01 -9.74437892e-01 6.26701236e-01 1.27267271e-01 -9.88746107e-01 -3.25279266e-01 -7.45802745e-02 -1.51550382e-01 -1.21230848e-01 7.15921164e-01 -1.07142174e+00 2.53591955e-01 1.05616987e+00 1.13530302e+00 -5.86567104e-01 8.87077987e-01 9.70064551e-02 8.53771806e-01 -3.98005426e-01 4.47852045e-01 3.08171242e-01 -6.43230230e-02 7.78898954e-01 1.03140175e+00 7.79299676e-01 -3.03803742e-01 -6.42116740e-02 7.99620986e-01 6.41696677e-02 -6.67946041e-02 -1.22643888e+00 -3.32163006e-01 1.55354545e-01 9.70958054e-01 -7.22738564e-01 -5.08676954e-02 -5.14226794e-01 9.58332062e-01 -1.80215940e-01 4.24634904e-01 -9.23831403e-01 -5.15243411e-01 1.01746941e+00 1.01863213e-01 6.91348091e-02 -1.93822652e-01 -1.41912296e-01 -9.28074181e-01 2.10843012e-01 -1.08529830e+00 6.73171163e-01 -5.39684892e-01 -1.45818734e+00 5.10629475e-01 4.34523709e-02 -1.63629508e+00 -7.84102261e-01 -7.67642438e-01 -1.01415718e+00 3.35780591e-01 -1.12642670e+00 -9.08334255e-01 -2.77469814e-01 1.08603692e+00 4.91854489e-01 -5.61398387e-01 7.77487934e-01 3.70673478e-01 -5.38031459e-01 6.11488283e-01 1.06365554e-01 3.46851379e-01 5.96846938e-01 -1.48113346e+00 1.10392678e+00 1.36409402e+00 4.17956680e-01 1.96309641e-01 7.05389082e-01 -3.29835385e-01 -1.07436228e+00 -1.24917281e+00 4.78242069e-01 -6.80536807e-01 1.02489436e+00 -2.21192315e-01 -1.22733915e+00 8.73208225e-01 2.12686121e-01 1.30505875e-01 5.55405974e-01 -4.45046760e-02 -3.53101492e-01 -3.05960208e-01 -1.06483722e+00 5.88966310e-01 9.36598599e-01 -7.74328113e-01 -4.82208759e-01 5.65403581e-01 8.19556236e-01 -3.67164433e-01 -6.18601680e-01 3.31494749e-01 8.46636295e-02 -8.89239490e-01 1.03488767e+00 -7.62636781e-01 3.11239839e-01 -1.28669292e-01 2.24758655e-01 -1.62498403e+00 -9.04588997e-02 -8.74689460e-01 -5.68606019e-01 8.57705772e-01 1.35001108e-01 -1.10101402e+00 3.18558156e-01 3.23305070e-01 -5.17991722e-01 -2.87394911e-01 -1.05564737e+00 -1.10870850e+00 9.27334558e-03 -1.04886055e+00 9.49583769e-01 1.26718986e+00 -3.14025171e-02 -2.49305695e-01 -9.40677151e-02 3.76975894e-01 4.81766373e-01 -3.92776161e-01 8.06970835e-01 -1.41210794e+00 3.12427580e-01 -6.59881294e-01 -9.87219810e-01 -3.73003781e-01 2.39426032e-01 -7.14740455e-01 -5.34408279e-02 -1.14040172e+00 -5.36113143e-01 7.50129372e-02 -7.38946319e-01 7.68896639e-01 1.80880323e-01 1.83182091e-01 -3.95938009e-01 -4.22740318e-02 -3.29386741e-01 4.99621779e-01 9.76382971e-01 -3.74192953e-01 -2.01785758e-01 1.93234786e-01 -3.42978120e-01 8.88191640e-01 9.23760712e-01 -4.39588845e-01 -4.66639012e-01 -6.10781685e-02 7.48915225e-02 -3.66582483e-01 5.14006615e-01 -1.22141027e+00 1.97288290e-01 5.70544563e-02 3.36460829e-01 -7.16144443e-01 -1.27463743e-01 -1.02762127e+00 -3.24691296e-01 5.17425179e-01 -2.99396932e-01 7.80953169e-01 4.79903221e-01 5.52460372e-01 -4.45466250e-01 3.91708434e-01 4.57772970e-01 2.44946182e-01 -8.79027128e-01 6.36742651e-01 -6.21472836e-01 1.19232414e-02 1.07050240e+00 -1.10918716e-01 -1.59824312e-01 -9.04098153e-01 -7.33117998e-01 1.61937162e-01 -1.40292078e-01 8.86406302e-01 6.87116206e-01 -1.66678095e+00 -5.67121089e-01 7.33523726e-01 3.61464798e-01 -7.39468560e-02 7.55997002e-02 9.99461830e-01 -5.63004076e-01 3.06496948e-01 -4.39871997e-01 -1.03448832e+00 -1.09260964e+00 6.52347326e-01 2.89627403e-01 -2.67672464e-02 -1.14421737e+00 3.33692223e-01 1.91152617e-01 -4.57204759e-01 4.08453882e-01 -8.33117545e-01 -2.17185751e-01 -1.76691860e-01 6.81372046e-01 5.16469538e-01 4.48894262e-01 -2.86212951e-01 -2.54116267e-01 3.44343275e-01 -1.57228649e-01 5.82270371e-03 1.43164539e+00 1.46343365e-01 -2.17748955e-01 7.99521923e-01 1.27533078e+00 -3.92479062e-01 -1.10248232e+00 -3.29343289e-01 2.72378474e-01 -3.36054206e-01 -1.13417536e-01 -2.83923686e-01 -1.02613306e+00 9.64905143e-01 1.00129414e+00 9.41639841e-01 1.43643200e+00 -2.34237853e-02 1.05760407e+00 7.22944379e-01 -6.51059225e-02 -1.04351223e+00 4.64530170e-01 7.41446972e-01 8.94619524e-01 -1.21454716e+00 -5.76977789e-01 1.64198384e-01 -2.88431048e-01 1.24741471e+00 6.15386724e-01 -1.26537338e-01 8.04449618e-01 3.70634317e-01 2.74367243e-01 -4.80083585e-01 -4.90546674e-01 -1.17258005e-01 4.35221165e-01 8.26718211e-01 1.05723597e-01 -1.80855542e-01 3.85018259e-01 1.34071857e-01 -3.90874505e-01 -4.86940324e-01 4.08163190e-01 8.21228147e-01 -6.18426241e-02 -8.72250974e-01 -6.11063957e-01 6.66080832e-01 -4.87149119e-01 2.41224796e-01 -3.93518150e-01 8.36469710e-01 -5.13412207e-02 8.81505609e-01 6.34014130e-01 -3.90241832e-01 4.86233979e-01 3.59642148e-01 8.84814411e-02 -1.86686635e-01 -3.00990850e-01 -1.07460730e-01 -4.64611232e-01 -1.08591127e+00 -3.30748945e-01 -7.16380715e-01 -1.19927597e+00 -4.02034700e-01 2.50044942e-01 -5.83108291e-02 5.87780595e-01 1.29984283e+00 3.39505166e-01 1.07293093e+00 9.17765677e-01 -5.68412066e-01 -9.27613974e-02 -8.63071024e-01 -4.42533940e-01 7.98969150e-01 9.89846289e-01 -4.22282368e-01 -5.97835720e-01 1.23949274e-01]

[7.46690559387207, 2.567391872406006]

711ba897-64ce-44ad-9a24-caa0399a9f7e

fairr-faithful-and-robust-deductive-reasoning

https://openreview.net/forum?id=0DR74NlWR7u

https://openreview.net/pdf?id=0DR74NlWR7u

FaiRR: Faithful and Robust Deductive Reasoning over Natural Language

Transformers have been shown to be able to perform deductive reasoning on a logical rulebase containing rules and statements written in natural language. Recent works show that such models can also produce the reasoning steps (i.e., the proof graph) that emulate the model's logical reasoning process. Currently, these black-box models generate both the proof graph and intermediate inferences within the same model and thus may be unfaithful. In this work, we frame the deductive logical reasoning task by defining three modular components: rule selection, fact selection, and knowledge composition. The rule and fact selection steps select the candidate rule and facts to be used and then the knowledge composition combines them to generate new inferences. This ensures model faithfulness by assured causal relation from the proof step to the inference reasoning. To test our framework, we propose FaiRR (Faithful and Robust Reasoner) where the above three components are independently modeled by transformers. We observe that FaiRR is robust to novel language perturbations, and is faster at inference than previous works on existing reasoning datasets. Additionally, in contrast to black-box generative models, the errors made by FaiRR are more interpretable due to the modular approach.

['Anonymous']

2022-01-16

acl-arr-january-2022-1

['fact-selection']

['natural-language-processing']

[ 3.01948518e-01 1.01384687e+00 -2.48268303e-02 -9.24963132e-02 -2.94287711e-01 -8.47978771e-01 1.18967187e+00 8.84159580e-02 3.90408307e-01 9.32482660e-01 2.33377308e-01 -9.05891597e-01 -3.34004223e-01 -1.45247972e+00 -1.14873064e+00 -2.11976409e-01 8.07252973e-02 6.78325117e-01 5.96566617e-01 -3.09651434e-01 5.00011928e-02 2.40946457e-01 -1.60844612e+00 4.72486079e-01 1.05606616e+00 6.14727855e-01 -3.12830985e-01 6.56033039e-01 -1.46454453e-01 2.04039383e+00 -4.17900860e-01 -9.17608380e-01 6.86514899e-02 -6.60307884e-01 -1.22097945e+00 -3.33731949e-01 2.90451676e-01 -3.23705465e-01 -3.28332894e-02 1.23251009e+00 -1.10573553e-01 8.10608417e-02 4.76800501e-01 -1.73677897e+00 -7.53862023e-01 1.55862391e+00 1.87347725e-01 -3.72614682e-01 9.26357388e-01 1.65438414e-01 1.08758020e+00 -6.20283425e-01 9.42858279e-01 1.62369263e+00 5.53571999e-01 7.81874776e-01 -1.31149006e+00 -4.92333651e-01 8.71161968e-02 6.92056656e-01 -1.16331530e+00 -5.77108324e-01 7.73676515e-01 -2.43947208e-01 9.85279322e-01 6.68343186e-01 6.59531057e-01 1.26334357e+00 3.57605308e-01 5.07577538e-01 1.10800278e+00 -6.23417854e-01 6.25165761e-01 3.25894892e-01 1.36879593e-01 9.29351091e-01 6.24084473e-01 1.69585928e-01 -6.17470264e-01 -3.97102445e-01 3.95414323e-01 -3.97927612e-01 -2.47832313e-01 -3.75793695e-01 -1.06273556e+00 6.84967935e-01 2.40188297e-02 9.32917818e-02 -3.35264236e-01 3.24234664e-01 2.83249021e-01 4.45377707e-01 -3.44514072e-01 4.93009090e-01 -3.72548163e-01 -8.39825999e-03 -6.63168430e-01 7.20595241e-01 1.23973358e+00 1.03959036e+00 4.10165846e-01 -2.06415683e-01 -3.49608839e-01 2.03168795e-01 5.92611492e-01 7.19078183e-01 1.50652051e-01 -1.28985512e+00 2.73960173e-01 9.83063519e-01 1.54789954e-01 -1.05281425e+00 -1.01878047e-01 -9.60571542e-02 -5.36644876e-01 2.60737062e-01 3.69251281e-01 7.98282102e-02 -5.64384341e-01 1.92714155e+00 4.09991890e-01 1.44521862e-01 4.83911037e-01 4.26908970e-01 8.69256794e-01 4.76299256e-01 -1.00943349e-01 -2.39015222e-01 1.34195709e+00 -5.98294377e-01 -8.03771019e-01 -3.68890353e-02 4.02704418e-01 -2.43829653e-01 8.98497403e-01 4.29141313e-01 -1.58826375e+00 -1.04603536e-01 -1.05101395e+00 -2.90441979e-02 -2.52587199e-01 -5.53101838e-01 7.28488624e-01 2.15629458e-01 -9.92608964e-01 2.58218348e-01 -6.39975369e-01 -8.95996019e-02 1.93557084e-01 -1.70832783e-01 -3.00962597e-01 -2.24985868e-01 -1.68719292e+00 1.23501277e+00 5.51246941e-01 -1.60874818e-02 -1.29613280e+00 -7.31566131e-01 -1.03445637e+00 1.84483677e-01 9.12581205e-01 -1.14944112e+00 1.41736984e+00 -4.13221449e-01 -1.64845300e+00 5.81774414e-01 -2.61898220e-01 -8.95789921e-01 8.52110863e-01 1.23578161e-02 -4.36805487e-01 1.60287142e-01 1.02195658e-01 2.18533397e-01 5.17168939e-01 -1.50665450e+00 -4.36903328e-01 -1.53289184e-01 6.79608226e-01 -3.48145396e-01 4.75541145e-01 1.55541319e-02 -1.47046119e-01 -8.75875354e-02 1.47157490e-01 -8.31351936e-01 1.01446047e-01 -1.71862215e-01 -9.21943724e-01 -2.22921446e-01 4.93319988e-01 -4.13310260e-01 1.27335155e+00 -1.56222200e+00 2.59312809e-01 4.93998379e-01 2.27728993e-01 -5.94671741e-02 4.03125525e-01 6.24410391e-01 -2.31770992e-01 4.13402051e-01 -1.57486945e-01 1.91357076e-01 6.26810849e-01 3.94495070e-01 -9.14644361e-01 2.05678912e-03 4.39486094e-02 1.15137422e+00 -9.54971135e-01 -7.50078857e-01 2.29659691e-01 5.53931594e-02 -6.93724096e-01 8.71275291e-02 -8.45196903e-01 -9.31536108e-02 -1.97324663e-01 5.99439561e-01 5.72711885e-01 -2.12371856e-01 5.99464536e-01 -7.75141791e-02 2.42327183e-01 8.51092577e-01 -1.42590010e+00 1.22442591e+00 -4.38476801e-01 2.05869734e-01 -3.66976500e-01 -6.27978563e-01 7.86480188e-01 4.70184505e-01 -3.07171792e-01 -3.95938039e-01 -4.07883301e-02 -9.56774969e-03 1.36840701e-01 -6.82318330e-01 5.57813197e-02 -5.82511365e-01 -1.52345207e-02 7.34597147e-01 -6.20192848e-02 -5.52107692e-01 5.41144371e-01 7.23976552e-01 1.18822706e+00 4.99281824e-01 6.19743645e-01 -9.95512903e-02 6.83519900e-01 1.81045011e-01 7.26688266e-01 1.03509235e+00 2.31953546e-01 -7.56682009e-02 9.87091422e-01 -4.85350043e-01 -5.81731498e-01 -1.34454656e+00 1.72757939e-01 5.53551972e-01 -4.81448770e-02 -8.66887748e-01 -6.45281494e-01 -7.73964703e-01 -3.15634012e-02 1.71028769e+00 -6.52623951e-01 -6.55720770e-01 -3.73730928e-01 -9.13770031e-03 1.05720472e+00 4.25183833e-01 5.37395358e-01 -1.19770527e+00 -7.68589973e-01 9.44989547e-02 -5.10411739e-01 -1.06193805e+00 4.28315014e-01 -4.90630120e-02 -4.82329309e-01 -1.65100598e+00 6.90860808e-01 -1.75802112e-01 7.57352710e-01 -3.75313878e-01 1.28273940e+00 3.36918682e-01 2.26506561e-01 2.15191394e-01 -3.58644545e-01 -5.11620641e-01 -1.18797445e+00 -4.63589758e-01 -8.02648813e-02 -2.38480479e-01 7.20112920e-02 -7.76914775e-01 1.31135583e-01 -1.60241183e-02 -1.11658275e+00 4.98045206e-01 1.12522341e-01 5.62109649e-01 3.13858360e-01 6.51846051e-01 1.06495447e-01 -1.15071034e+00 7.42659628e-01 -2.85519063e-01 -7.25498557e-01 8.71114492e-01 -7.27675855e-01 5.58405459e-01 1.11629260e+00 5.06977551e-02 -1.41100442e+00 -4.35720444e-01 2.21358761e-01 -7.19624907e-02 -4.46344241e-02 7.49055862e-01 -3.89647126e-01 5.21495342e-01 8.45869780e-01 3.46777171e-01 -2.28777170e-01 5.72815239e-02 5.30938983e-01 5.35550453e-02 6.92917228e-01 -1.25402939e+00 1.39189672e+00 4.84221607e-01 4.09111828e-01 1.28372908e-01 -8.28873038e-01 6.80252850e-01 -1.90850124e-01 -1.59759998e-01 3.46245170e-01 -4.65956688e-01 -1.15850592e+00 4.33270633e-03 -1.20338070e+00 -5.35037935e-01 -6.23358071e-01 1.11591101e-01 -7.95913637e-01 2.76382387e-01 -4.75133538e-01 -1.13745594e+00 -2.52088606e-01 -1.02224422e+00 7.16448486e-01 -1.21899724e-01 -7.26513922e-01 -9.99518633e-01 1.51868060e-01 2.72653222e-01 4.60033976e-02 3.54692161e-01 1.43453920e+00 -6.25843287e-01 -6.13699913e-01 1.95984840e-02 1.32011443e-01 1.84556320e-01 8.30843672e-03 6.72816157e-01 -8.41930509e-01 3.77556205e-01 -5.67312725e-02 -2.31124789e-01 3.68814826e-01 -9.73739475e-02 7.62148559e-01 -7.96736598e-01 -1.90115690e-01 2.20224008e-01 1.16716325e+00 1.10780016e-01 9.50556755e-01 3.14109087e-01 2.32911095e-01 2.60545015e-01 3.02788287e-01 1.00420401e-01 7.00979590e-01 5.96376061e-01 3.44805181e-01 4.23750401e-01 -1.30092338e-01 -8.57933521e-01 5.73340058e-01 2.40400434e-01 -3.42508435e-01 1.29735144e-02 -1.20053041e+00 2.10505381e-01 -2.12133169e+00 -1.80212319e+00 -2.70448923e-01 1.83769298e+00 1.34265459e+00 4.12318885e-01 -3.34976017e-01 5.54696679e-01 2.35001862e-01 -2.36684605e-01 -2.26733163e-01 -6.97538674e-01 -2.42861331e-01 2.35644370e-01 -1.13273792e-01 1.07164013e+00 -3.13716561e-01 8.19546819e-01 6.39104795e+00 3.23188335e-01 -5.84559023e-01 -1.84054777e-01 -1.97724581e-01 -7.12806955e-02 -1.05072093e+00 6.90449893e-01 -4.55291510e-01 2.19623923e-01 8.19061935e-01 -4.89040196e-01 8.36305857e-01 7.26324260e-01 1.41844004e-01 -2.64535904e-01 -1.51824236e+00 2.58528233e-01 -2.93260626e-02 -1.64391637e+00 5.86321354e-01 -3.84100258e-01 5.38146973e-01 -6.51319861e-01 -3.83474886e-01 3.92915636e-01 1.22748661e+00 -9.35053468e-01 1.37895858e+00 7.94141829e-01 2.83481121e-01 -6.94514811e-01 6.30752981e-01 4.47344035e-01 -7.93803036e-01 -3.27052534e-01 4.71172296e-02 -3.43889713e-01 2.86489110e-02 9.23335850e-01 -1.04085279e+00 9.65506494e-01 4.30606425e-01 3.95383954e-01 -5.83882749e-01 4.43869203e-01 -1.26831794e+00 6.62370801e-01 -1.14301860e-01 1.48772582e-01 -2.50350982e-01 -4.88932729e-02 6.29587471e-01 8.77751112e-01 4.37858105e-02 -1.49292231e-01 -1.74914300e-01 1.51042223e+00 8.31985381e-03 -5.47370970e-01 -6.72906876e-01 2.08689243e-01 6.86783135e-01 8.33064616e-01 -4.16644126e-01 -8.23994994e-01 -3.51325050e-02 5.32156825e-01 4.17802036e-01 5.06305277e-01 -1.16552663e+00 -1.15833692e-01 4.32128280e-01 1.30577488e-02 -7.05247670e-02 1.56882390e-01 -3.98416460e-01 -1.30080438e+00 2.93380857e-01 -1.28043747e+00 3.98040175e-01 -1.35848475e+00 -8.98713171e-01 1.93874657e-01 4.34129119e-01 -6.06467962e-01 -6.55409575e-01 -3.82573187e-01 -8.03857207e-01 7.00681090e-01 -1.38001144e+00 -1.20890534e+00 -1.04234971e-01 7.66278625e-01 -5.02860695e-02 2.80952722e-01 1.09579098e+00 -2.53744692e-01 -5.17623186e-01 2.56064504e-01 -8.89532387e-01 1.01081863e-01 3.37150156e-01 -1.42794311e+00 1.93834096e-01 1.44798803e+00 3.49773169e-02 1.36023581e+00 1.13444638e+00 -8.19113314e-01 -1.75656307e+00 -1.04949439e+00 1.24103415e+00 -7.28314042e-01 8.64791930e-01 -3.66967738e-01 -6.70114636e-01 1.28265786e+00 7.79678226e-02 -2.53595889e-01 3.44456732e-01 1.64281338e-01 -1.02465880e+00 -1.99305981e-01 -1.31522393e+00 1.14034307e+00 1.23308456e+00 -5.95050812e-01 -1.32689834e+00 2.27803200e-01 6.26210153e-01 -3.28026533e-01 -6.19300902e-01 3.54029536e-01 6.11285150e-01 -1.15027761e+00 7.95909345e-01 -8.64924729e-01 9.43500519e-01 -9.36919451e-01 -2.08665520e-01 -1.17987669e+00 -3.35424960e-01 -8.19699407e-01 -6.88453734e-01 1.35325277e+00 7.00902641e-01 -8.39848280e-01 1.05891307e-03 9.81362343e-01 1.97447650e-02 -4.17979926e-01 -8.55261683e-01 -7.55998552e-01 -1.16722062e-01 -8.09531748e-01 1.06024373e+00 8.08465779e-01 7.84398317e-01 4.21692133e-01 -3.33584920e-02 2.48410836e-01 5.63466489e-01 6.58292532e-01 8.81523609e-01 -1.24010527e+00 -5.98104775e-01 -4.53338206e-01 -6.89715147e-02 -1.69424921e-01 4.76273865e-01 -1.17075253e+00 9.21636894e-02 -1.98642659e+00 3.31205845e-01 -1.00626223e-01 2.12788343e-01 1.11340439e+00 -1.80342644e-01 -4.01736259e-01 2.19940618e-01 -4.70085703e-02 -5.12408495e-01 2.08299160e-01 1.23536015e+00 -3.37563664e-01 1.35882080e-01 -3.00411642e-01 -9.83503163e-01 8.84365022e-01 6.50234699e-01 -3.54478002e-01 -8.69692445e-01 1.59331374e-02 1.11693764e+00 7.87586644e-02 9.73787427e-01 -8.24424446e-01 3.65725517e-01 -5.47755659e-01 -7.19151273e-02 -2.43158499e-03 -9.83171314e-02 -7.24326968e-01 7.22274899e-01 6.61064506e-01 -6.17722273e-01 -1.27088144e-01 6.52122200e-02 4.76522036e-02 -3.89829651e-02 -1.64327174e-01 4.81792510e-01 -2.09968373e-01 -2.96245635e-01 -3.18997681e-01 -4.93454933e-01 1.41292930e-01 9.12716627e-01 1.61413968e-01 -7.15372443e-01 -3.70866865e-01 -6.62351787e-01 9.40183476e-02 5.09330988e-01 1.65509224e-01 5.75425625e-01 -1.18790865e+00 -7.04546750e-01 6.77098110e-02 1.16286293e-01 1.56288624e-01 -6.70640543e-02 9.18901622e-01 -3.40265214e-01 2.66194969e-01 -5.88236749e-02 -7.28488714e-02 -9.87172544e-01 8.56937349e-01 5.63857615e-01 -4.84521240e-01 -5.35459161e-01 4.38321590e-01 -2.69510269e-01 -5.65060318e-01 -6.87577426e-02 -8.11288476e-01 1.38178542e-01 -3.31372976e-01 7.39163995e-01 2.91563362e-01 5.02479933e-02 -9.24945250e-02 -6.66769385e-01 1.18441001e-01 2.67769337e-01 -2.45268837e-01 1.10366511e+00 1.84636459e-01 -6.89668715e-01 4.92906302e-01 2.22322792e-01 4.58272994e-01 -6.06662214e-01 -8.39299038e-02 -1.67533636e-01 -7.14626303e-03 -3.77275884e-01 -1.29046226e+00 -5.09093404e-01 5.10325015e-01 -5.22009790e-01 6.94239497e-01 8.67204905e-01 1.30457699e-01 2.06257880e-01 5.82187593e-01 5.70483148e-01 -8.73606741e-01 -5.44945240e-01 5.21714330e-01 1.12983119e+00 -5.55311084e-01 5.23549318e-02 -6.62362099e-01 -4.40595329e-01 9.94361997e-01 4.11909312e-01 2.42156759e-01 3.36980671e-02 6.96264148e-01 -8.67852271e-02 -2.88361579e-01 -1.39412796e+00 1.16462879e-01 -7.93757215e-02 7.07713187e-01 4.36693877e-02 2.09905207e-01 -7.15018287e-02 8.14893305e-01 -6.67302608e-01 5.53691447e-01 7.06194103e-01 8.71023655e-01 -1.44885555e-01 -1.04966915e+00 -6.94092751e-01 2.81448569e-02 -8.91981050e-02 -1.73323646e-01 -7.28644788e-01 7.74441779e-01 1.62425086e-01 1.37149155e+00 -5.63804984e-01 -2.89416283e-01 3.25804830e-01 5.10116935e-01 8.27130497e-01 -4.50488806e-01 -4.45388287e-01 -8.89624059e-01 6.47920549e-01 -7.16285229e-01 -3.24086577e-01 -4.70470399e-01 -1.68203413e+00 -7.01668859e-01 -2.92410981e-02 3.74243647e-01 3.85509767e-02 1.23578537e+00 2.58723795e-01 7.33102918e-01 1.65545747e-01 7.05692321e-02 -6.54060125e-01 -5.94626427e-01 -1.93252191e-01 5.75521946e-01 1.21078815e-03 -6.09635592e-01 -4.04784828e-01 4.42995697e-01]

[9.177820205688477, 7.169439315795898]

9ad72de5-c5b2-488e-823f-fab00f7ec856

data-driven-deep-supervision-for-skin-lesion

2209.01527

https://arxiv.org/abs/2209.01527v1

https://arxiv.org/pdf/2209.01527v1.pdf

Data-Driven Deep Supervision for Skin Lesion Classification

Automatic classification of pigmented, non-pigmented, and depigmented non-melanocytic skin lesions have garnered lots of attention in recent years. However, imaging variations in skin texture, lesion shape, depigmentation contrast, lighting condition, etc. hinder robust feature extraction, affecting classification accuracy. In this paper, we propose a new deep neural network that exploits input data for robust feature extraction. Specifically, we analyze the convolutional network's behavior (field-of-view) to find the location of deep supervision for improved feature extraction. To achieve this, first, we perform activation mapping to generate an object mask, highlighting the input regions most critical for classification output generation. Then the network layer whose layer-wise effective receptive field matches the approximated object shape in the object mask is selected as our focus for deep supervision. Utilizing different types of convolutional feature extractors and classifiers on three melanoma detection datasets and two vitiligo detection datasets, we verify the effectiveness of our new method.

['Danny Z. Chen', 'X. Sharon Hu', 'Tianyu Zhang', 'Li Zhang', 'Yizhe Zhang', 'Suraj Mishra']

2022-09-04

['skin-lesion-classification']

['medical']

[ 6.85979307e-01 3.82471317e-03 -1.64316863e-01 -4.11383569e-01 -3.28804344e-01 -4.08021986e-01 4.95719761e-01 -4.16796245e-02 -3.57650310e-01 5.87699115e-01 -2.22785324e-02 -1.75784156e-01 -4.45086211e-02 -4.72112775e-01 -4.77098882e-01 -1.00533080e+00 2.37664118e-01 -5.31573474e-01 2.07157075e-01 -7.30309188e-02 5.44501364e-01 6.90284967e-01 -1.48073387e+00 6.76723063e-01 1.09807920e+00 1.19704628e+00 1.21102497e-01 5.99461496e-01 1.31622329e-01 5.25661469e-01 -7.18516588e-01 -1.14263088e-01 1.51555121e-01 -4.73077893e-01 -6.81082428e-01 3.70463550e-01 6.54325724e-01 -2.50406593e-01 -1.36103570e-01 1.22157419e+00 5.84392071e-01 -2.00555533e-01 9.16194916e-01 -6.88264430e-01 -4.20015633e-01 1.78555250e-02 -8.09856296e-01 4.49570149e-01 -5.72569929e-02 4.10525531e-01 4.27753478e-01 -5.77634871e-01 7.67500401e-01 7.11501598e-01 5.19061208e-01 7.65873194e-01 -9.43601370e-01 -2.32397273e-01 -1.43369928e-01 -9.43693146e-03 -1.27173948e+00 -4.72357571e-01 7.18228996e-01 -4.44692999e-01 5.69974244e-01 3.96838546e-01 7.68289804e-01 9.93431866e-01 4.68513608e-01 8.20409656e-01 1.66945732e+00 -6.43347621e-01 1.30371600e-01 3.77483010e-01 -1.87641501e-01 9.13411319e-01 6.73232004e-02 5.18559404e-02 -3.06432843e-01 1.19426958e-01 8.51392925e-01 -6.36853501e-02 -4.52908784e-01 6.78218082e-02 -7.52938807e-01 4.22825336e-01 8.02301347e-01 1.46945477e-01 -3.26819509e-01 -3.38374600e-02 7.29908571e-02 1.63195981e-03 3.24341685e-01 4.25346076e-01 -1.64295748e-01 3.46824914e-01 -6.82747900e-01 -2.15303525e-01 1.45320550e-01 2.72334814e-01 5.95335603e-01 -2.06663713e-01 -2.49160275e-01 9.20418859e-01 6.36029318e-02 1.04841270e-01 4.99469429e-01 -3.81568789e-01 9.44174454e-02 1.17899096e+00 -2.03492954e-01 -5.79299808e-01 -5.40051162e-01 -3.27019155e-01 -8.26589644e-01 5.16023397e-01 6.39949977e-01 -2.70013392e-01 -1.38264823e+00 1.08720922e+00 3.92555118e-01 -2.65615821e-01 6.55589625e-02 1.21974051e+00 6.09731853e-01 2.00954556e-01 6.87063038e-02 1.60061549e-02 1.34241235e+00 -8.56792569e-01 -2.74959326e-01 -2.87931085e-01 6.25976741e-01 -6.87276185e-01 8.99978995e-01 2.95081168e-01 -7.74917245e-01 -3.32455277e-01 -9.97806966e-01 -6.80911094e-02 -4.43288326e-01 6.48744285e-01 6.64702177e-01 5.49394429e-01 -8.56939435e-01 4.38388050e-01 -6.28449559e-01 -4.79884744e-01 8.07710230e-01 4.88880903e-01 -5.15389681e-01 -7.79046714e-02 -6.26073480e-01 7.49178052e-01 1.90576434e-01 2.93360919e-01 -7.44745135e-01 -5.62866747e-01 -7.05684602e-01 -2.59548724e-01 -8.19462240e-02 -5.37124276e-01 8.09918582e-01 -1.55126894e+00 -1.44720924e+00 9.96119380e-01 -1.14111848e-01 -1.96222410e-01 2.77703464e-01 2.96878070e-01 -4.09492224e-01 4.39704329e-01 -1.91460952e-01 7.02094793e-01 1.21098506e+00 -1.11877012e+00 -7.81371772e-01 -6.42883599e-01 -3.51571455e-03 2.64811099e-01 -4.28064972e-01 1.97470292e-01 -1.91572919e-01 -3.38773519e-01 -9.69244912e-02 -6.95975363e-01 -3.75537097e-01 2.42475554e-01 -8.46835375e-01 -6.96082190e-02 6.28382444e-01 -6.54359400e-01 9.21971858e-01 -2.40403461e+00 -4.79035340e-02 5.22520363e-01 2.52285063e-01 3.71022463e-01 -1.18046440e-01 1.13818742e-01 -4.97955829e-02 1.56256571e-01 -2.34298125e-01 2.83945762e-02 -4.88628149e-01 -2.22920775e-01 8.43557417e-02 5.67070782e-01 7.91948020e-01 8.19595993e-01 -5.38414240e-01 -5.22737086e-01 3.00685108e-01 5.36511421e-01 -1.01165362e-01 3.10871582e-02 -1.99750528e-01 3.48184615e-01 -5.20968139e-01 1.15647638e+00 6.86939776e-01 -2.36121103e-01 1.09539129e-01 -5.03564715e-01 -1.67531356e-01 -2.20694363e-01 -6.30511165e-01 1.25642407e+00 -5.09636700e-01 9.76616859e-01 1.05926692e-01 -4.88719046e-01 9.20978904e-01 -1.62351131e-01 3.21074754e-01 -8.46253216e-01 2.75071740e-01 3.19811314e-01 3.93084288e-01 -7.76079118e-01 2.25527287e-01 1.29541695e-01 5.18232822e-01 2.04504788e-01 -2.11378634e-01 3.03239286e-01 7.17250779e-02 -2.96313882e-01 9.18892622e-01 -6.34714141e-02 4.15677279e-01 -1.95031315e-01 4.96138453e-01 1.00086294e-02 2.10952818e-01 2.96992958e-01 -2.51763195e-01 6.64699733e-01 7.92626977e-01 -6.03028655e-01 -8.88410270e-01 -6.67347789e-01 -5.65492988e-01 7.38991678e-01 1.85020223e-01 5.19230440e-02 -9.22548652e-01 -9.68271911e-01 -4.16021422e-03 1.44893795e-01 -1.01666057e+00 -1.90326229e-01 -3.76676202e-01 -1.01832438e+00 4.91656959e-01 3.42678368e-01 5.72236657e-01 -1.16912031e+00 -8.38599443e-01 -8.01099539e-02 3.33538920e-01 -5.91457248e-01 -3.16175461e-01 2.00844705e-01 -6.09757364e-01 -1.42427087e+00 -7.24813938e-01 -7.92354465e-01 1.48107314e+00 2.11596593e-01 5.75941622e-01 2.00305328e-01 -1.13836014e+00 -5.39065115e-02 -2.21878842e-01 -4.65086937e-01 -3.51238489e-01 5.21096289e-02 -3.30880165e-01 5.10014176e-01 3.71060193e-01 1.22589432e-03 -1.05062544e+00 2.63192356e-01 -1.15590477e+00 7.84807429e-02 9.98732328e-01 8.58217478e-01 5.52995145e-01 8.39270428e-02 4.49272424e-01 -7.42101252e-01 7.68002927e-01 -2.43502721e-01 -4.92060274e-01 3.10576588e-01 -1.62569791e-01 -1.92894623e-01 5.66615582e-01 -3.27592313e-01 -1.01008141e+00 3.15528274e-01 4.69256146e-03 -7.23468438e-02 -4.74424750e-01 3.65424663e-01 -6.86053000e-03 -4.68347281e-01 1.08695722e+00 3.13446224e-01 4.18604255e-01 -1.16299152e-01 -1.20310083e-01 9.33517218e-01 1.77798048e-01 5.70971072e-02 4.98287261e-01 7.62609661e-01 1.05731443e-01 -9.24253881e-01 -8.86023760e-01 -3.02687079e-01 -6.03628516e-01 -3.77141178e-01 6.96989596e-01 -4.35057014e-01 -6.45952046e-01 7.81235576e-01 -8.69828701e-01 -3.31645072e-01 -1.16742164e-01 3.28040630e-01 -2.09151581e-01 3.87602113e-02 -2.59852558e-01 -7.36046612e-01 -5.26572585e-01 -1.16640186e+00 1.13478327e+00 8.33015084e-01 9.04766843e-02 -1.13075733e+00 -3.81353162e-02 3.45372587e-01 3.82933587e-01 4.99532431e-01 9.40748751e-01 -2.28054404e-01 -4.50391084e-01 -2.62417287e-01 -5.99727750e-01 4.76371676e-01 3.85383934e-01 3.71678978e-01 -1.29112422e+00 -1.83030084e-01 -4.78302509e-01 -4.06330556e-01 1.22170472e+00 7.52197742e-01 1.40672266e+00 -2.40560453e-02 -5.09318054e-01 9.32409465e-01 1.49908483e+00 -3.84319648e-02 8.00996482e-01 2.84271359e-01 5.65724492e-01 9.13097501e-01 4.87824410e-01 1.64355904e-01 -1.83694698e-02 3.15415323e-01 6.68442965e-01 -8.18954229e-01 -3.86583865e-01 1.05049796e-01 2.93900758e-01 -7.35419244e-02 -3.16585869e-01 1.28726438e-02 -7.38821626e-01 6.04878068e-01 -1.34465122e+00 -4.50189203e-01 2.89263371e-02 2.13832259e+00 8.42885315e-01 -2.89796051e-02 1.37065545e-01 -1.43042549e-01 7.19204247e-01 4.57570516e-02 -8.06669950e-01 -3.79763335e-01 -1.72869369e-01 2.40993023e-01 4.61070538e-01 2.71251082e-01 -1.07881284e+00 8.61517906e-01 5.54361486e+00 7.03283191e-01 -1.62856483e+00 -4.39117461e-01 8.82944763e-01 -4.10417616e-02 -1.20983541e-01 -2.69011110e-01 -8.58116567e-01 4.12372977e-01 3.32941353e-01 3.91719043e-01 2.93765485e-01 4.30236399e-01 1.37663409e-01 -4.07326967e-01 -9.20195162e-01 7.27624595e-01 1.67232528e-01 -1.36503971e+00 -9.82193798e-02 3.45573932e-01 8.06343019e-01 -1.05466507e-01 4.64948386e-01 -2.63214469e-01 -2.79147148e-01 -1.37962878e+00 -9.63597149e-02 8.57853770e-01 1.19799352e+00 -5.90114355e-01 8.45071673e-01 5.10051362e-02 -6.25555694e-01 -2.78674990e-01 -3.82105440e-01 3.06865096e-01 -4.90839571e-01 6.36821330e-01 -1.26195359e+00 1.91654325e-01 4.38322693e-01 5.72220445e-01 -1.00049341e+00 1.29420483e+00 -1.33315995e-01 5.23156285e-01 -1.54056191e-01 -2.00905964e-01 2.52474874e-01 9.06557366e-02 2.05613017e-01 1.08025753e+00 3.68749388e-02 -3.44713122e-01 -2.27747560e-01 1.01032209e+00 7.50290900e-02 2.45668232e-01 -5.29736340e-01 -6.43234178e-02 3.43826562e-02 1.61273289e+00 -7.13495374e-01 2.66064018e-01 -2.06370622e-01 8.41644049e-01 3.23513150e-01 4.77155268e-01 -4.17753547e-01 -5.73462546e-01 5.29918253e-01 3.29960048e-01 1.37443766e-02 2.18629539e-01 -2.60816485e-01 -8.34829330e-01 9.05537009e-02 -8.31122696e-01 2.12821051e-01 -5.05550504e-01 -1.14619315e+00 5.36967158e-01 -5.03566086e-01 -9.93629098e-01 5.90829514e-02 -8.93633306e-01 -1.04202902e+00 1.10720062e+00 -1.80105484e+00 -1.40320706e+00 -7.32779264e-01 5.14158309e-01 2.92134047e-01 -3.88652205e-01 7.55875647e-01 -1.14541031e-01 -9.17118430e-01 6.42078459e-01 -1.21740609e-01 1.32652506e-01 7.52307773e-01 -1.17410886e+00 1.23509124e-01 6.65201187e-01 -2.09438786e-01 4.47519094e-01 1.00979224e-01 -5.48637271e-01 -1.19378126e+00 -1.16664910e+00 2.47402534e-01 -2.19060868e-01 5.49748421e-01 -2.31525421e-01 -5.99806905e-01 6.28966168e-02 2.12576523e-01 2.54622161e-01 6.34933829e-01 -2.23615468e-01 -1.27509147e-01 -4.32990104e-01 -1.20499754e+00 7.38790810e-01 5.42162597e-01 -2.81363279e-01 6.09636940e-02 6.17774546e-01 5.82526773e-02 -4.43343431e-01 -7.33305156e-01 4.26058412e-01 5.20043254e-01 -9.83908057e-01 5.83875895e-01 -7.30398178e-01 7.66087174e-01 -1.70004666e-01 2.15281069e-01 -1.28536153e+00 -1.19560473e-01 -5.17432511e-01 2.18924761e-01 7.17604756e-01 4.69895095e-01 -5.48807800e-01 1.05139935e+00 1.42257884e-01 -1.89664617e-01 -1.36018300e+00 -6.68356240e-01 9.12548974e-03 -3.62368673e-02 3.11118275e-01 2.49591008e-01 6.69578135e-01 -1.99546680e-01 1.40185580e-02 4.24806476e-02 1.81877479e-01 5.19405961e-01 1.89601198e-01 5.72695196e-01 -9.37882304e-01 1.09026961e-01 -4.63798404e-01 -5.74002802e-01 -5.31816244e-01 -1.30469814e-01 -9.03360069e-01 -2.32310668e-01 -1.55797970e+00 1.61404639e-01 -4.78799701e-01 -4.00649041e-01 7.53877044e-01 -1.80521399e-01 5.99416375e-01 -2.16236204e-01 2.79555116e-02 -2.29202539e-01 2.55704261e-02 1.73523831e+00 -2.97249138e-01 -3.82478148e-01 2.99198598e-01 -8.01062107e-01 6.76905930e-01 1.11453652e+00 5.69995753e-02 -2.51312166e-01 -3.39736700e-01 3.30973901e-02 -3.34041089e-01 6.57974899e-01 -9.15607512e-01 3.84863853e-01 -3.17008555e-01 9.91248012e-01 -2.60519356e-01 2.94508040e-01 -6.00417197e-01 -4.61194158e-01 4.29551750e-01 -3.71209979e-01 -6.44037604e-01 1.89564332e-01 3.35042834e-01 -1.22569732e-01 -4.99845594e-01 1.02897525e+00 -9.54699218e-02 -6.88344598e-01 1.54824123e-01 -3.30125123e-01 -2.18911335e-01 1.26709926e+00 -4.88077432e-01 -7.16662586e-01 2.16594130e-01 -6.03043795e-01 -5.44725955e-02 5.38062692e-01 1.87253550e-01 8.40129018e-01 -9.72440481e-01 -8.02689731e-01 6.01444900e-01 4.49279100e-01 9.80299786e-02 4.01394367e-01 1.04685712e+00 -6.05176151e-01 2.18775585e-01 -5.91786444e-01 -7.61750162e-01 -1.27656138e+00 -5.96881052e-03 8.79304051e-01 1.18981719e-01 -4.41876769e-01 9.61904824e-01 2.09566429e-02 -1.44639537e-01 4.22858372e-02 -3.04943711e-01 -2.62190670e-01 -1.19098790e-01 5.33005834e-01 1.18728459e-01 1.50769979e-01 -2.57136762e-01 -1.43586233e-01 6.09846234e-01 -6.11451387e-01 4.29881752e-01 9.98253644e-01 2.37306923e-01 -3.05432647e-01 -8.00511166e-02 9.88402724e-01 8.66766553e-03 -1.54110312e+00 -5.51014964e-04 -3.12988669e-01 -6.13899112e-01 2.84100920e-01 -1.15680063e+00 -1.11569715e+00 8.78556967e-01 9.22446787e-01 2.00676724e-01 1.28875566e+00 -1.72572166e-01 4.45468187e-01 1.35619506e-01 -1.06130220e-01 -1.09953380e+00 4.33249027e-02 -5.97155318e-02 7.44237065e-01 -1.41725087e+00 -7.06205424e-03 -3.73382270e-01 -6.00386739e-01 1.28212655e+00 9.95350003e-01 -2.75400393e-02 4.79108751e-01 2.67431676e-01 3.99276137e-01 -3.06009829e-01 -6.36590242e-01 -3.66672188e-01 6.57539487e-01 8.82866263e-01 3.19101691e-01 4.91546318e-02 -8.52784142e-04 2.02325091e-01 5.39090559e-02 -2.22303703e-01 5.63118994e-01 6.87067807e-01 -5.55865705e-01 -7.35116839e-01 -1.26516387e-01 9.03399348e-01 -3.31781179e-01 -1.68882590e-02 -8.37141275e-01 7.09989250e-01 2.99426109e-01 5.41701794e-01 1.93800628e-01 -2.45425358e-01 -1.63172912e-02 -2.83082008e-01 6.13072276e-01 -6.97629094e-01 -7.04808354e-01 1.42855644e-01 -1.38233574e-02 -3.57641727e-01 -2.74849921e-01 -3.01902175e-01 -1.11072326e+00 2.76916236e-01 -3.57297003e-01 -5.04090190e-01 1.02826977e+00 7.23213613e-01 2.60468602e-01 4.46554571e-01 8.43094110e-01 -5.12321770e-01 -5.27575970e-01 -9.68058288e-01 -6.33389115e-01 2.80613720e-01 5.18912554e-01 -3.58104765e-01 -2.11062685e-01 1.17847823e-01]

[15.662674903869629, -2.9580752849578857]

f6b210dd-36ff-4f3a-98d7-07f9bab44eef

benchmarking-the-robustness-of-lidar-semantic

2301.00970

https://arxiv.org/abs/2301.00970v2

https://arxiv.org/pdf/2301.00970v2.pdf

Benchmarking the Robustness of LiDAR Semantic Segmentation Models

When using LiDAR semantic segmentation models for safety-critical applications such as autonomous driving, it is essential to understand and improve their robustness with respect to a large range of LiDAR corruptions. In this paper, we aim to comprehensively analyze the robustness of LiDAR semantic segmentation models under various corruptions. To rigorously evaluate the robustness and generalizability of current approaches, we propose a new benchmark called SemanticKITTI-C, which features 16 out-of-domain LiDAR corruptions in three groups, namely adverse weather, measurement noise and cross-device discrepancy. Then, we systematically investigate 11 LiDAR semantic segmentation models, especially spanning different input representations (e.g., point clouds, voxels, projected images, and etc.), network architectures and training schemes. Through this study, we obtain two insights: 1) We find out that the input representation plays a crucial role in robustness. Specifically, under specific corruptions, different representations perform variously. 2) Although state-of-the-art methods on LiDAR semantic segmentation achieve promising results on clean data, they are less robust when dealing with noisy data. Finally, based on the above observations, we design a robust LiDAR segmentation model (RLSeg) which greatly boosts the robustness with simple but effective modifications. It is promising that our benchmark, comprehensive analysis, and observations can boost future research in robust LiDAR semantic segmentation for safety-critical applications.

['Dengxin Dai', 'Shuguang Cui', 'Zhen Li', 'Chaoda Zheng', 'Xu Yan']

2023-01-03

['lidar-semantic-segmentation']

['computer-vision']

[ 3.09778363e-01 -2.46553481e-01 -1.35112539e-01 -5.32795966e-01 -8.93794298e-01 -5.56320727e-01 3.86881351e-01 9.41221267e-02 -2.20182016e-01 4.97519255e-01 -9.61361974e-02 -4.00880605e-01 -3.48101765e-01 -9.20208931e-01 -8.94268930e-01 -5.33479810e-01 1.34396791e-01 2.06303060e-01 5.04460514e-01 -3.54564786e-01 4.13718343e-01 8.91683400e-01 -2.01384735e+00 -3.38649154e-01 1.30409324e+00 1.14181995e+00 -4.98788357e-02 1.99215874e-01 -1.74557753e-02 1.86857894e-01 -6.42611504e-01 -3.13671321e-01 4.90748376e-01 2.01750651e-01 -4.95889276e-01 -1.34339735e-01 5.76055586e-01 -8.48427117e-02 -2.56640851e-01 1.27983975e+00 5.26568055e-01 2.74236143e-01 7.09953606e-01 -1.44819987e+00 -5.80552042e-01 3.69367272e-01 -3.38556170e-01 4.58501354e-02 -1.74138904e-01 4.77307558e-01 8.46590400e-01 -9.14401889e-01 2.73731709e-01 1.42607820e+00 8.57374668e-01 2.93939263e-01 -1.10147154e+00 -9.97751415e-01 4.55261081e-01 2.56479561e-01 -1.42254913e+00 -3.69274437e-01 7.66955853e-01 -4.19718593e-01 6.89989090e-01 2.57065982e-01 1.65629908e-01 1.17069185e+00 3.20055395e-01 6.34010553e-01 1.09696245e+00 2.58319348e-01 3.77365232e-01 -5.53058744e-05 5.15363097e-01 2.88775533e-01 6.79228127e-01 4.84925181e-01 -4.13445503e-01 1.33237898e-01 1.56233549e-01 -3.27453762e-01 -1.87146328e-02 -4.22536165e-01 -8.05463910e-01 8.81190121e-01 5.68027616e-01 -9.13028270e-02 6.64297640e-02 2.35538453e-01 4.81078327e-01 9.17206109e-02 4.28602993e-01 2.76735038e-01 -3.30694139e-01 8.58064815e-02 -8.39810610e-01 2.45653704e-01 3.74611646e-01 1.21950579e+00 9.93013203e-01 4.63724077e-01 2.90830317e-03 8.45112801e-01 3.07010651e-01 9.38298702e-01 1.70206755e-01 -1.00808704e+00 6.28226459e-01 4.29415852e-01 -1.74776629e-01 -9.49204147e-01 -5.68835378e-01 -4.13559139e-01 -9.08703864e-01 3.07904631e-01 4.93773725e-03 1.01469636e-01 -1.17328119e+00 1.66183722e+00 1.66192695e-01 3.64497542e-01 2.56068915e-01 7.95591712e-01 1.01130855e+00 2.42997095e-01 1.46349192e-01 1.25564799e-01 1.25026977e+00 -5.33752143e-01 -5.80308199e-01 -5.11795342e-01 3.03552389e-01 -6.44756913e-01 1.05587208e+00 3.45283121e-01 -6.98476553e-01 -9.34342325e-01 -1.35833812e+00 -1.96174026e-01 -7.27878273e-01 -1.20899297e-01 4.97445017e-01 8.42635512e-01 -8.12299192e-01 6.71891987e-01 -8.30777705e-01 -4.22468781e-01 3.94575566e-01 3.67621362e-01 -1.49816975e-01 -9.50207487e-02 -1.20179462e+00 9.35565174e-01 3.06082726e-01 2.31104657e-01 -8.73080730e-01 -7.34006584e-01 -1.09589565e+00 -2.94231027e-01 5.63456118e-01 -6.57081366e-01 1.04343128e+00 -1.62874460e-01 -1.21874106e+00 5.45937359e-01 -2.95538247e-01 -6.20418787e-01 4.84511882e-01 -3.66509199e-01 -4.24700230e-01 4.06535678e-02 3.73452336e-01 8.34272504e-01 9.03009057e-01 -1.70241284e+00 -5.46091318e-01 -5.63447118e-01 -1.89726070e-01 -1.10235490e-01 1.05559774e-01 -1.76082239e-01 -3.88913274e-01 -5.48066854e-01 4.04857844e-01 -1.05193782e+00 -3.19159776e-01 -2.37996951e-01 -6.79473281e-01 -3.91287021e-02 9.63024378e-01 -2.07543939e-01 8.70895505e-01 -2.31059051e+00 -2.50758052e-01 2.85220116e-01 -1.00458987e-01 3.39329004e-01 -1.86824873e-01 8.67260993e-02 -1.71016023e-01 5.02215624e-01 -6.93241239e-01 -4.30460781e-01 1.19019575e-01 5.51588118e-01 -7.32258081e-01 3.79504055e-01 6.17871284e-01 1.07642043e+00 -6.07166886e-01 -3.65270644e-01 5.07566333e-01 4.66124684e-01 -4.40607607e-01 -7.58804902e-02 6.05857465e-03 6.89712942e-01 -7.73192406e-01 9.48674202e-01 1.12410998e+00 4.11759257e-01 -5.22734106e-01 -2.96305299e-01 -1.02599636e-01 2.18002513e-01 -1.28939176e+00 1.48438048e+00 -3.16857874e-01 5.23233712e-01 5.28798774e-02 -9.74426031e-01 1.12449908e+00 -2.13104784e-01 4.21803236e-01 -9.50245857e-01 1.22972369e-01 2.70453751e-01 -2.79321522e-01 -3.42124999e-01 1.00187111e+00 -1.73836187e-01 -3.63245964e-01 3.01720686e-02 -2.42140636e-01 -7.59581387e-01 -6.48930855e-03 -4.35775965e-02 7.34645844e-01 -3.65655534e-02 -2.60777641e-02 -2.30666026e-01 4.43554431e-01 -8.41758475e-02 7.48938739e-01 8.85083914e-01 -4.44042265e-01 9.06051993e-01 3.46470326e-01 -7.84952715e-02 -5.89588344e-01 -1.37097514e+00 -5.53333938e-01 6.53896034e-01 5.59863150e-01 -2.09631890e-01 -6.65455937e-01 -5.13181448e-01 5.03043056e-01 9.48696911e-01 -4.06163961e-01 -6.12398446e-01 -6.65829778e-01 -7.58529365e-01 9.30453718e-01 8.59423995e-01 6.85870349e-01 -7.22778797e-01 -5.64236641e-01 -1.57468081e-01 -1.45061418e-01 -1.52224576e+00 1.00307666e-01 2.47508436e-01 -1.11122870e+00 -1.04652643e+00 1.26780570e-01 -2.32392535e-01 2.05372348e-01 7.96735346e-01 1.02642238e+00 2.51158653e-03 -2.03159198e-01 3.41209203e-01 -2.02862054e-01 -6.32302761e-01 -2.36286521e-01 7.87637159e-02 3.04890007e-01 -2.75282174e-01 2.61725396e-01 -6.08545899e-01 -3.53734463e-01 6.20422721e-01 -1.08886933e+00 -5.67248106e-01 4.38896567e-01 4.70675766e-01 8.33821297e-01 2.78416395e-01 3.62508863e-01 -6.90217912e-01 5.59441805e-01 -4.69308048e-01 -8.28655541e-01 -9.14909616e-02 -6.59587562e-01 8.94588679e-02 4.97217834e-01 -3.64995562e-02 -7.89416313e-01 -1.05083182e-01 -3.81632954e-01 -5.34543276e-01 -4.29987371e-01 3.17748815e-01 -3.99661809e-01 -2.22405702e-01 6.41553879e-01 -2.12595635e-03 -1.81788921e-01 -4.38015908e-01 6.07994497e-01 5.91803253e-01 8.08447063e-01 -9.11449432e-01 1.26493776e+00 7.58957565e-01 2.46508658e-01 -1.06437564e+00 -7.80069828e-01 -5.02296269e-01 -6.93925142e-01 9.59766284e-02 7.96669304e-01 -1.04937840e+00 -5.40130854e-01 5.06964505e-01 -9.94798720e-01 -1.17288969e-01 -4.35195059e-01 2.28611022e-01 -6.88207924e-01 5.74160337e-01 -1.11450933e-01 -8.59527707e-01 -1.28366444e-02 -1.70503402e+00 1.51738036e+00 7.33999833e-02 6.19154749e-03 -5.86152136e-01 -3.51099283e-01 5.13381183e-01 2.80927330e-01 3.55704725e-01 8.71980488e-01 -5.92268825e-01 -9.11763489e-01 4.18670103e-02 -4.08217400e-01 7.04305470e-01 -5.08174114e-02 2.68855572e-01 -1.31714094e+00 -1.85243189e-01 1.74099922e-01 -1.50618270e-01 1.21156466e+00 3.42834264e-01 1.54787159e+00 2.00921640e-01 -2.15588957e-01 9.13367450e-01 1.20904183e+00 -5.63665852e-02 7.77351677e-01 2.48623252e-01 7.38789856e-01 7.54372835e-01 1.06662893e+00 -2.36819554e-02 5.13302743e-01 5.20359993e-01 9.00953114e-01 -5.64120710e-02 -8.04847553e-02 -2.10180044e-01 2.62525022e-01 6.26048803e-01 1.04200408e-01 -1.60035118e-01 -8.61483514e-01 4.43738788e-01 -1.70658743e+00 -5.74337780e-01 -2.52968937e-01 2.23230863e+00 3.03319722e-01 4.38707203e-01 -1.34422034e-01 4.01099175e-01 5.34338295e-01 4.24014002e-01 -8.26366544e-01 -3.63275051e-01 -2.42578015e-01 4.13792908e-01 9.25059319e-01 4.18305397e-01 -1.20919764e+00 1.28895390e+00 6.57887554e+00 9.17839706e-01 -1.13702631e+00 1.21372603e-01 4.18470055e-01 1.66103512e-01 -3.91770214e-01 -1.31540418e-01 -1.01573515e+00 4.45636153e-01 9.56003010e-01 1.09591536e-01 1.45575315e-01 8.34169090e-01 3.74776125e-01 -2.74029374e-01 -9.12452400e-01 7.32519090e-01 1.62973572e-02 -9.43826973e-01 1.32231489e-01 3.08878291e-02 5.66475093e-01 4.03097034e-01 3.48479897e-01 4.04488623e-01 3.92106324e-01 -1.17339575e+00 8.44144106e-01 2.96689332e-01 6.84941113e-01 -7.79054046e-01 7.10701108e-01 2.61817455e-01 -1.35402679e+00 -2.32425332e-01 -5.54605901e-01 1.34555221e-01 2.22859934e-01 7.85279989e-01 -4.16502476e-01 1.00223827e+00 8.57511401e-01 7.93820977e-01 -8.01976800e-01 9.00725663e-01 -5.06359398e-01 5.68003237e-01 -3.88668150e-01 5.30374408e-01 2.23535597e-01 -3.06677967e-01 6.35456085e-01 1.16644263e+00 4.47166324e-01 -6.75989548e-03 9.24661011e-02 1.14405620e+00 1.53201327e-01 -3.19144964e-01 -9.72239912e-01 2.83030570e-01 6.26020908e-01 1.02015638e+00 -6.23397410e-01 -1.14357486e-01 -1.93086207e-01 3.49568397e-01 -1.03500038e-02 5.83199203e-01 -9.93788600e-01 -1.30822420e-01 1.35260141e+00 -4.46456708e-02 2.25363240e-01 -6.31028175e-01 -9.71782386e-01 -8.50537479e-01 2.17019007e-01 -6.75300479e-01 1.15992822e-01 -8.25646639e-01 -1.26121330e+00 3.03368658e-01 2.85831720e-01 -1.35414660e+00 1.19846672e-01 -7.18165576e-01 -5.65535247e-01 6.27987683e-01 -2.16987920e+00 -1.17089403e+00 -5.04688263e-01 4.69918489e-01 5.60736775e-01 9.64848623e-02 3.51693809e-01 2.43521363e-01 -7.86019802e-01 4.99776185e-01 -2.01215655e-01 -1.17040388e-01 6.31097496e-01 -9.53611851e-01 9.07481372e-01 1.06697190e+00 -6.81725144e-02 6.87306762e-01 6.99731648e-01 -7.29250252e-01 -1.29578590e+00 -1.67803836e+00 4.43537027e-01 -6.76583767e-01 7.41942048e-01 -4.00354892e-01 -1.11001480e+00 5.50860405e-01 -2.63171405e-01 8.83918479e-02 2.85530210e-01 -6.56380281e-02 -5.06308973e-01 -3.51719677e-01 -1.20104611e+00 6.10392451e-01 1.34616530e+00 -6.38555706e-01 -6.06524885e-01 2.15022296e-01 1.12062681e+00 -5.29079735e-01 -7.40662396e-01 1.16537690e+00 1.47635564e-01 -1.17363870e+00 1.39552915e+00 -3.80677462e-01 1.03734277e-01 -4.49935049e-01 -5.03124535e-01 -1.21200466e+00 2.74511091e-02 -3.27333927e-01 3.03570449e-01 1.17762864e+00 2.36731172e-01 -1.06068754e+00 5.39157093e-01 4.79530275e-01 -6.32460952e-01 -5.03039837e-01 -1.33302319e+00 -1.17905879e+00 4.94124979e-01 -1.22979391e+00 9.62632120e-01 4.90937859e-01 -7.10756779e-01 7.65960962e-02 -8.76614917e-03 6.65879190e-01 5.87171733e-01 -1.53898855e-03 9.40979421e-01 -1.52424538e+00 4.67758507e-01 -4.34617311e-01 -5.53895772e-01 -1.00405478e+00 4.69673365e-01 -7.46673286e-01 3.76252264e-01 -1.25382626e+00 -3.08365643e-01 -7.61077642e-01 -3.50448936e-01 3.98793638e-01 -3.17841828e-01 3.31921786e-01 1.57808036e-01 1.41546384e-01 -3.34768564e-01 7.79972494e-01 1.02433515e+00 -4.04993117e-01 8.72194208e-03 3.35662037e-01 -9.26668227e-01 7.40367770e-01 9.89379764e-01 -5.07805049e-01 -4.85493481e-01 -5.78789949e-01 4.11464907e-02 -3.95823956e-01 6.36999011e-01 -1.20217752e+00 5.33882007e-02 -3.17167252e-01 -1.64929166e-01 -9.59638536e-01 5.89346230e-01 -8.06210816e-01 -1.03616156e-01 2.29219139e-01 1.93469405e-01 1.17871568e-01 5.13973773e-01 6.43265009e-01 -3.40430170e-01 -1.61544532e-01 9.47848320e-01 3.62510271e-02 -8.69324684e-01 2.92800248e-01 -2.73905933e-01 2.11318254e-01 8.90658259e-01 -5.95158100e-01 -4.42221284e-01 -1.02058999e-01 -3.61247510e-01 4.45834637e-01 4.82993364e-01 7.29982436e-01 6.68290854e-01 -1.20044720e+00 -4.93583620e-01 4.93736267e-01 3.35601628e-01 4.91159558e-01 1.78022534e-01 7.46169269e-01 -6.35589585e-02 4.73731220e-01 1.10516511e-01 -1.14148271e+00 -9.54592586e-01 4.72616345e-01 1.78817019e-01 1.61483005e-01 -4.67870325e-01 6.92127883e-01 2.60823429e-01 -8.41964483e-01 2.32748792e-01 -9.88767922e-01 3.92639451e-03 1.56208262e-01 -9.98454820e-03 5.31061471e-01 5.00905693e-01 -9.95442212e-01 -4.44301605e-01 1.21888554e+00 4.05922204e-01 2.06553057e-01 9.44344997e-01 -8.76320899e-02 1.64376810e-01 4.00768876e-01 9.32584822e-01 -2.80743130e-02 -1.23533821e+00 1.41959335e-03 -1.76801234e-02 -5.09284675e-01 -4.83318754e-02 -4.30726379e-01 -1.29253983e+00 1.22672164e+00 5.98484695e-01 9.49333757e-02 1.08458734e+00 -1.94984168e-01 8.70642662e-01 3.40109795e-01 4.69010502e-01 -1.19773805e+00 -1.00717582e-01 7.74841487e-01 8.26686859e-01 -1.33960092e+00 1.28037050e-01 -8.46155763e-01 -6.65969014e-01 6.45569205e-01 4.63159293e-01 3.54131907e-02 6.25289679e-01 2.60600507e-01 1.74349383e-01 -2.22920477e-01 -2.15455577e-01 -6.09142184e-01 2.78006405e-01 8.72678041e-01 -2.23685265e-01 1.65451735e-01 1.76346704e-01 4.86538529e-01 -6.09430909e-01 -3.63430887e-01 3.42864454e-01 6.58183694e-01 -5.97343743e-01 -8.94423485e-01 -5.15007615e-01 1.85258210e-01 3.91708575e-02 3.53660733e-02 -4.61495548e-01 9.66060221e-01 3.44747990e-01 1.33363819e+00 -7.87369013e-02 -5.28131127e-01 8.46669197e-01 -9.90135595e-02 -1.65137723e-02 -6.05024457e-01 -4.61234838e-01 -4.13580954e-01 1.11989193e-01 -9.03717041e-01 -2.70803988e-01 -7.17716336e-01 -1.31343675e+00 -2.12527439e-01 -3.90046448e-01 -2.19292581e-01 8.87370169e-01 9.31486309e-01 4.50542778e-01 6.95774317e-01 5.34207761e-01 -9.20722187e-01 -7.08694041e-01 -6.24245226e-01 -4.39967632e-01 2.72392213e-01 3.72493088e-01 -1.12288010e+00 -5.18903673e-01 -4.86849308e-01]

[7.9230828285217285, -2.359726667404175]

c0fca77e-7729-449f-98a5-8f6bb05a0e1f

self-supervised-learning-from-images-with-a

2301.08243

https://arxiv.org/abs/2301.08243v3

https://arxiv.org/pdf/2301.08243v3.pdf

Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture

This paper demonstrates an approach for learning highly semantic image representations without relying on hand-crafted data-augmentations. We introduce the Image-based Joint-Embedding Predictive Architecture (I-JEPA), a non-generative approach for self-supervised learning from images. The idea behind I-JEPA is simple: from a single context block, predict the representations of various target blocks in the same image. A core design choice to guide I-JEPA towards producing semantic representations is the masking strategy; specifically, it is crucial to (a) sample target blocks with sufficiently large scale (semantic), and to (b) use a sufficiently informative (spatially distributed) context block. Empirically, when combined with Vision Transformers, we find I-JEPA to be highly scalable. For instance, we train a ViT-Huge/14 on ImageNet using 16 A100 GPUs in under 72 hours to achieve strong downstream performance across a wide range of tasks, from linear classification to object counting and depth prediction.

['Nicolas Ballas', 'Yann Lecun', 'Michael Rabbat', 'Pascal Vincent', 'Piotr Bojanowski', 'Ishan Misra', 'Quentin Duval', 'Mahmoud Assran']

2023-01-19

http://openaccess.thecvf.com//content/CVPR2023/html/Assran_Self-Supervised_Learning_From_Images_With_a_Joint-Embedding_Predictive_Architecture_CVPR_2023_paper.html

http://openaccess.thecvf.com//content/CVPR2023/papers/Assran_Self-Supervised_Learning_From_Images_With_a_Joint-Embedding_Predictive_Architecture_CVPR_2023_paper.pdf

cvpr-2023-1

['object-counting']

['computer-vision']

[ 4.67743874e-01 4.10388917e-01 -1.84388578e-01 -3.39736491e-01 -8.77012730e-01 -3.11540812e-01 7.96021223e-01 -8.99061635e-02 -3.66062790e-01 3.83416384e-01 4.27860767e-01 -2.93005854e-01 3.21151227e-01 -8.74118507e-01 -1.16641152e+00 -7.45562732e-01 -9.54029709e-03 5.38725793e-01 3.33249360e-01 1.34394109e-01 3.22094738e-01 3.71969014e-01 -1.78715110e+00 5.39932787e-01 3.11038047e-01 1.02666616e+00 6.74965143e-01 9.51667726e-01 -7.42693841e-02 9.42103565e-01 -3.64457816e-01 -8.97095725e-02 1.99834943e-01 -4.32448745e-01 -7.78413892e-01 3.07732403e-01 3.59754920e-01 -3.60921115e-01 -3.97980690e-01 7.16842532e-01 2.71107197e-01 9.81094167e-02 9.18126404e-01 -1.14754510e+00 -7.32555866e-01 3.29107285e-01 -6.40951753e-01 2.40912318e-01 -1.93091720e-01 5.03699124e-01 1.09002447e+00 -9.31568503e-01 5.52805960e-01 1.26573133e+00 4.73661393e-01 6.06658578e-01 -1.56167078e+00 -4.03513432e-01 1.25766397e-01 3.83938104e-02 -1.06157267e+00 -4.32924360e-01 5.30153275e-01 -4.24977779e-01 1.28930128e+00 8.38348791e-02 6.82307720e-01 1.20925379e+00 1.05165176e-01 8.83176267e-01 1.16295552e+00 -4.84401584e-01 4.64059711e-01 6.16775043e-02 2.49549329e-01 8.36801171e-01 6.91577792e-02 1.07779138e-01 -6.33917511e-01 -1.42621800e-01 8.11933994e-01 -5.50881438e-02 1.43453464e-01 -5.23861110e-01 -1.10973680e+00 9.69923198e-01 7.32004702e-01 3.71257737e-02 -3.23753625e-01 7.22539186e-01 4.76434052e-01 1.10472562e-02 5.26780903e-01 3.01759571e-01 -3.53682578e-01 -1.92294139e-02 -8.93777251e-01 2.16172159e-01 4.63306397e-01 1.02388740e+00 1.08928430e+00 1.47408023e-01 -2.56532818e-01 6.24714017e-01 1.85339317e-01 3.05421382e-01 5.17256975e-01 -9.96852458e-01 2.18101397e-01 3.03678632e-01 -7.53610209e-02 -7.39612639e-01 -1.43151462e-01 -3.46367657e-01 -6.26025021e-01 2.83247441e-01 2.42280126e-01 1.04379684e-01 -1.24796748e+00 1.65225410e+00 2.16513187e-01 3.62091869e-01 1.46915525e-01 7.23936677e-01 6.88238204e-01 9.03664470e-01 4.00153816e-01 3.55892956e-01 1.46829116e+00 -1.30250323e+00 1.43304840e-01 -8.12572718e-01 6.96224511e-01 -3.88798028e-01 1.18509758e+00 2.21927136e-01 -1.00030112e+00 -8.23776126e-01 -9.49412704e-01 -4.20473754e-01 -2.78772861e-01 2.58859187e-01 7.83757806e-01 3.82493734e-01 -1.24082971e+00 5.85993350e-01 -8.57531965e-01 -2.74159431e-01 7.34322250e-01 2.93046743e-01 -3.91112357e-01 -3.06621820e-01 -4.38781142e-01 7.18987167e-01 4.65794772e-01 -3.33816648e-01 -1.20785630e+00 -6.61086202e-01 -9.56725657e-01 2.41653502e-01 -9.97803360e-03 -8.56257141e-01 9.54936206e-01 -1.21050370e+00 -1.18229032e+00 1.14305687e+00 -3.33174914e-01 -6.20493054e-01 1.47617921e-01 -1.37271628e-01 2.07326934e-01 2.99617648e-01 3.18440586e-01 1.39249158e+00 1.22430873e+00 -1.35496557e+00 -4.73088622e-01 -5.35933018e-01 -1.92035839e-01 1.95825145e-01 -3.78380835e-01 -1.44061670e-01 -7.01815784e-01 -6.37148082e-01 -8.01614001e-02 -9.61476564e-01 -3.92170608e-01 7.45064318e-02 -2.56535083e-01 -2.70160258e-01 7.77133226e-01 -6.12382531e-01 3.70510757e-01 -2.38984513e+00 1.40039384e-01 -2.55902223e-02 3.26154381e-01 1.94154277e-01 -3.35792094e-01 2.05299556e-01 -1.44454241e-01 -2.77571231e-01 -1.63423717e-01 -7.32604325e-01 -8.01005289e-02 4.30793822e-01 -5.68659961e-01 3.14403772e-01 5.22434235e-01 1.33235323e+00 -9.16970730e-01 -3.67569566e-01 4.28024501e-01 5.17538667e-01 -7.48076797e-01 2.50930607e-01 -5.82697034e-01 2.93532699e-01 -3.49635214e-01 4.54363048e-01 4.71915692e-01 -6.39102638e-01 7.91600570e-02 -1.49897873e-01 1.48498282e-01 3.13568175e-01 -5.87246299e-01 1.88362443e+00 -7.02946186e-01 8.41435552e-01 -1.80737048e-01 -1.27230167e+00 7.59968996e-01 -6.64858744e-02 2.19919831e-01 -6.48623705e-01 1.25262752e-01 -2.08498575e-02 -3.06431681e-01 -2.28762463e-01 5.11170745e-01 7.23607540e-02 1.33056656e-01 6.05759978e-01 3.46642017e-01 3.64273340e-02 -8.62241387e-02 2.72119284e-01 1.28095973e+00 3.41254860e-01 1.83305874e-01 -3.46583366e-01 7.99188390e-02 2.84005791e-01 1.92659289e-01 7.15585053e-01 -1.12618171e-02 8.82820189e-01 5.55943072e-01 -6.22724891e-01 -1.39364910e+00 -1.07340443e+00 1.40634794e-02 1.34322214e+00 -4.48124111e-02 -1.87654927e-01 -6.85065269e-01 -6.13262355e-01 9.07208696e-02 7.37585425e-01 -8.53159368e-01 -1.84081510e-01 -3.78509998e-01 -6.33071661e-01 3.91432166e-01 9.54465449e-01 5.18158495e-01 -1.08843160e+00 -8.93352032e-01 2.60238498e-01 1.03046469e-01 -1.11955369e+00 -3.42149079e-01 6.27388358e-01 -9.35971737e-01 -7.88300157e-01 -6.85412169e-01 -9.27130699e-01 8.35084319e-01 5.86565852e-01 1.26450562e+00 2.25537885e-02 -5.91258228e-01 4.08867538e-01 -2.07796305e-01 -1.87384889e-01 -2.92511344e-01 -8.15470517e-02 -3.97614777e-01 -1.25596046e-01 4.24199522e-01 -5.92230320e-01 -8.38853776e-01 -3.74578647e-02 -7.51702726e-01 3.08675587e-01 9.37842965e-01 9.91057932e-01 8.19755435e-01 -3.71435910e-01 2.46028349e-01 -8.03732872e-01 1.24271877e-01 -5.95223010e-01 -4.49122727e-01 1.48272634e-01 -4.57776248e-01 2.71344483e-01 5.27703226e-01 -4.96889710e-01 -6.93697572e-01 3.81067157e-01 -1.56051770e-01 -7.86250412e-01 -2.89323300e-01 -4.34964187e-02 9.70807821e-02 1.76231086e-01 8.10573101e-01 7.91793466e-01 2.10152492e-01 -2.27264404e-01 7.43297458e-01 4.91254181e-01 6.73888862e-01 -6.86819553e-01 4.80865926e-01 6.93066120e-01 2.33482588e-02 -9.19272959e-01 -8.33950162e-01 -4.99054581e-01 -4.55620259e-01 3.04490011e-02 1.16566658e+00 -1.30458391e+00 -4.40157473e-01 1.33042306e-01 -1.10227323e+00 -6.97035491e-01 -4.82400239e-01 2.67685503e-01 -7.70453393e-01 1.00626953e-01 -5.08208871e-01 -6.09969079e-01 -5.36677301e-01 -1.19592118e+00 1.53464568e+00 3.44663449e-02 -2.49433935e-01 -9.03606951e-01 -4.77930158e-02 4.93086874e-01 3.10510099e-01 7.40591250e-03 9.64071572e-01 -5.30443966e-01 -8.84912193e-01 -6.71408325e-02 -6.86571419e-01 4.18596447e-01 -2.06972539e-01 -3.51935089e-01 -1.34585381e+00 -3.36815953e-01 -3.41903090e-01 -8.83406341e-01 1.15706456e+00 4.10537869e-01 1.55822074e+00 -1.99738145e-01 -4.75252032e-01 8.13571990e-01 1.50067317e+00 -3.28923643e-01 7.95761049e-01 3.01493973e-01 6.61043286e-01 4.47536707e-01 3.23149443e-01 2.68817931e-01 5.10258734e-01 5.12701452e-01 4.70351011e-01 -6.17243052e-02 -4.72043693e-01 -5.34295976e-01 2.51312375e-01 4.15485561e-01 1.45282954e-01 2.56683491e-02 -6.55976176e-01 8.08103144e-01 -1.58873653e+00 -7.39133298e-01 2.60605142e-02 1.89212239e+00 6.77392662e-01 1.63296342e-01 8.57472643e-02 -2.64604509e-01 4.52156484e-01 2.60545552e-01 -6.79460466e-01 -4.09216166e-01 1.69478968e-01 5.34157038e-01 6.51122212e-01 2.62840748e-01 -9.92502213e-01 1.12042534e+00 6.44749260e+00 8.91645789e-01 -1.05972087e+00 3.20290834e-01 1.05249810e+00 -4.26213108e-02 -3.17868054e-01 8.89778063e-02 -8.76082540e-01 4.65009660e-01 1.02445734e+00 2.47152746e-01 3.04650724e-01 1.33546472e+00 -2.65766084e-01 -3.17270130e-01 -1.17754471e+00 1.03410470e+00 1.58094674e-01 -1.65175200e+00 9.38416198e-02 2.50089705e-01 7.19388783e-01 4.27180529e-01 7.74668455e-02 3.90151203e-01 5.51167428e-01 -1.15719283e+00 6.05947435e-01 1.54021725e-01 9.45492804e-01 -5.78615785e-01 2.82922983e-01 3.09954673e-01 -8.41818452e-01 -3.64684910e-02 -8.13613296e-01 2.72544250e-02 -2.13711768e-01 5.57371914e-01 -7.59162128e-01 9.98256579e-02 7.26703763e-01 6.27308309e-01 -5.69477677e-01 6.12560153e-01 -2.46358037e-01 7.14430511e-01 -2.92405486e-01 1.01146385e-01 4.86170739e-01 4.85697575e-02 6.00598454e-02 1.19012833e+00 2.61148304e-01 3.07915229e-02 -1.03195220e-01 1.07799792e+00 -1.22568063e-01 -2.14894459e-01 -8.24508011e-01 4.08067182e-02 3.54162455e-01 1.20252502e+00 -9.52984393e-01 -5.12541175e-01 -4.70328569e-01 1.30766511e+00 6.15789235e-01 1.24404833e-01 -6.71381295e-01 1.32089570e-01 6.25480831e-01 2.78012687e-03 7.68763781e-01 -1.51898161e-01 -6.31570518e-01 -1.02849293e+00 -1.53163612e-01 -5.36614776e-01 1.65678814e-01 -9.93459642e-01 -1.09849882e+00 5.49098969e-01 -3.32503229e-01 -9.78448391e-01 -3.21215749e-01 -7.27666020e-01 -7.02536523e-01 8.02984774e-01 -1.56940043e+00 -1.45628226e+00 -4.32954371e-01 5.29818177e-01 7.67685533e-01 -1.30640045e-01 1.08121526e+00 -1.64141059e-01 -3.07442456e-01 4.63271618e-01 1.14721566e-01 -7.47858547e-03 2.53951281e-01 -1.33779800e+00 8.11995804e-01 6.06274128e-01 5.34730613e-01 3.62265915e-01 6.15878582e-01 -4.39613551e-01 -1.46692240e+00 -1.39908421e+00 5.73713422e-01 -6.72751009e-01 4.99144614e-01 -5.69319725e-01 -7.93167055e-01 8.82273912e-01 7.46000037e-02 4.46162164e-01 4.66575414e-01 1.44770378e-02 -6.87807560e-01 4.56607826e-02 -8.57409537e-01 5.58381796e-01 9.84422624e-01 -6.28035486e-01 -6.50691509e-01 5.78900158e-01 6.53514683e-01 -2.06000611e-01 -4.21192974e-01 1.42410863e-02 3.78441125e-01 -1.03234065e+00 1.24295986e+00 -5.55507958e-01 9.83531952e-01 1.23958010e-03 -2.96266794e-01 -1.17313802e+00 -3.31990153e-01 -2.03161582e-01 -2.20040411e-01 7.91189969e-01 1.71268150e-01 -4.07263488e-01 1.08962333e+00 4.71438974e-01 -9.03069302e-02 -8.85333240e-01 -6.96550071e-01 -7.87088037e-01 1.96070075e-01 -4.99222696e-01 2.66100824e-01 5.60420811e-01 -3.58049095e-01 4.20650870e-01 -3.29159796e-01 2.00605184e-01 6.99370921e-01 2.57596314e-01 1.05330658e+00 -7.52677500e-01 -6.50754750e-01 -2.72581607e-01 -6.30150795e-01 -1.26033044e+00 1.78402767e-01 -9.09914196e-01 1.57374948e-01 -1.48959041e+00 4.32094634e-01 -5.09700298e-01 -1.34775862e-01 6.44318342e-01 -1.90243721e-01 5.11135995e-01 -1.09266331e-02 4.43841636e-01 -5.41247606e-01 6.50078297e-01 9.67207253e-01 -2.38980372e-02 1.70288548e-01 -2.56044954e-01 -7.49838531e-01 4.10736978e-01 5.67155063e-01 -3.80416274e-01 -6.33620262e-01 -6.29567742e-01 -1.00240491e-01 -1.82459608e-01 7.29505062e-01 -1.02200484e+00 -1.63112637e-02 1.68449253e-01 6.99250996e-01 -5.40244102e-01 5.46842158e-01 -5.11678696e-01 -1.72140092e-01 4.78882343e-01 -3.92313153e-01 -1.30849659e-01 2.19980717e-01 7.97622442e-01 -3.58799212e-02 -2.12283924e-01 8.27313364e-01 -3.29394907e-01 -1.14788139e+00 2.16929391e-01 -1.76188141e-01 1.20620150e-02 1.11056721e+00 -2.53659546e-01 -3.32615435e-01 -1.77903637e-01 -5.22063673e-01 -4.83328477e-02 5.41627765e-01 3.01225692e-01 6.49618924e-01 -1.05706346e+00 -4.35693353e-01 4.34572160e-01 4.52824503e-01 1.52989984e-01 2.06378102e-01 3.11031461e-01 -5.58081329e-01 4.99442995e-01 -2.46111095e-01 -7.75818288e-01 -9.53116059e-01 6.41362309e-01 -3.38068083e-02 -1.37983575e-01 -1.06860054e+00 1.16013122e+00 5.93917310e-01 -1.65591761e-01 1.27030373e-01 -1.06125735e-01 9.09631774e-02 -2.00653106e-01 6.83013260e-01 -9.30339992e-02 6.41618520e-02 -4.30537462e-01 -3.66399251e-02 3.10703248e-01 -2.44611755e-01 -1.06749631e-01 1.54670608e+00 9.50485021e-02 4.44050319e-03 2.65889734e-01 1.41376030e+00 -5.57751834e-01 -1.79247570e+00 -3.43409330e-01 -1.68366209e-01 -5.42097688e-01 3.97957623e-01 -4.28889066e-01 -1.03432739e+00 9.96037900e-01 5.05906999e-01 -8.64281058e-02 9.19044256e-01 3.67150843e-01 7.54258513e-01 2.72693843e-01 3.97686481e-01 -9.54626918e-01 4.56657976e-01 2.74601161e-01 6.95222497e-01 -1.33200836e+00 -5.31285219e-02 -1.30336836e-01 -8.73159587e-01 9.53237712e-01 6.59648418e-01 -5.71778774e-01 4.75112498e-01 2.18401015e-01 -1.86731473e-01 -3.79317462e-01 -9.91331458e-01 -3.98787737e-01 1.82250246e-01 7.43356526e-01 1.68820038e-01 4.71377224e-02 3.83247495e-01 1.93393499e-01 -7.32884929e-02 -1.07707366e-01 1.91576034e-01 8.28756332e-01 -4.83703464e-01 -8.71237993e-01 -1.91967934e-01 6.78677917e-01 -1.13542251e-01 -2.26964563e-01 -2.12214962e-02 6.64833188e-01 -6.67206943e-02 4.93055791e-01 5.33584893e-01 -3.59433740e-01 -2.01736271e-01 1.78600147e-01 5.64635932e-01 -9.36290264e-01 -2.49138057e-01 -7.02751875e-02 -1.61602601e-01 -7.07239747e-01 -3.20022941e-01 -6.36037827e-01 -1.12028515e+00 -1.98155746e-01 4.07865867e-02 -4.06372517e-01 8.51548195e-01 9.81526196e-01 5.47119796e-01 4.21287864e-01 4.74616379e-01 -1.13025331e+00 -4.38728988e-01 -9.78486896e-01 -3.76807094e-01 3.49548191e-01 2.75717437e-01 -5.25411546e-01 -2.50166327e-01 4.93070781e-02]

[9.766432762145996, 1.4235448837280273]

845a2504-951c-4e0e-9da8-85f1a9ee08c0

automatically-bounding-the-taylor-remainder

2212.11429

https://arxiv.org/abs/2212.11429v2

https://arxiv.org/pdf/2212.11429v2.pdf

Automatically Bounding the Taylor Remainder Series: Tighter Bounds and New Applications

We present a new algorithm for automatically bounding the Taylor remainder series. In the special case of a scalar function $f: \mathbb{R} \to \mathbb{R}$, our algorithm takes as input a reference point $x_0$, trust region $[a, b]$, and integer $k \ge 1$, and returns an interval $I$ such that $f(x) - \sum_{i=0}^{k-1} \frac {1} {i!} f^{(i)}(x_0) (x - x_0)^i \in I (x - x_0)^k$ for all $x \in [a, b]$. As in automatic differentiation, the function $f$ is provided to the algorithm in symbolic form, and must be composed of known atomic functions. At a high level, our algorithm has two steps. First, for a variety of commonly-used elementary functions (e.g., $\exp$, $\log$), we derive sharp polynomial upper and lower bounds on the Taylor remainder series. We then recursively combine the bounds for the elementary functions using an interval arithmetic variant of Taylor-mode automatic differentiation. Our algorithm can make efficient use of machine learning hardware accelerators, and we provide an open source implementation in JAX. We then turn our attention to applications. Most notably, we use our new machinery to create the first universal majorization-minimization optimization algorithms: algorithms that iteratively minimize an arbitrary loss using a majorizer that is derived automatically, rather than by hand. Applied to machine learning, this leads to architecture-specific optimizers for training deep networks that converge from any starting point, without hyperparameter tuning. Our experiments show that for some optimization problems, these hyperparameter-free optimizers outperform tuned versions of gradient descent, Adam, and AdaGrad. We also show that our automatically-derived bounds can be used for verified global optimization and numerical integration, and to prove sharper versions of Jensen's inequality.

['Joshua V. Dillon', 'Matthew Streeter']

2022-12-22

['numerical-integration']

['miscellaneous']

[-2.84459412e-01 3.02381031e-02 -1.76562667e-01 -3.47553045e-01 -9.66390789e-01 -6.62439346e-01 -1.15003794e-01 1.27545163e-01 -7.29361594e-01 9.60510731e-01 -6.73661470e-01 -7.69974232e-01 -1.38395578e-01 -8.32738519e-01 -8.81628990e-01 -7.73235977e-01 -6.91943586e-01 3.05481195e-01 -2.18190506e-01 -5.15924037e-01 2.59806126e-01 6.76646471e-01 -1.16847694e+00 -2.90238976e-01 8.30013394e-01 1.60885155e+00 -5.04539847e-01 9.50014055e-01 -9.13673043e-02 4.78861898e-01 -7.59204328e-01 -4.66029644e-01 6.84831679e-01 -3.08168143e-01 -8.85620356e-01 -4.96117413e-01 1.94583133e-01 -2.62844294e-01 1.31279111e-01 1.19033873e+00 2.11546272e-01 3.82024825e-01 5.66795111e-01 -1.17805672e+00 -3.17216992e-01 6.17184222e-01 -7.69915879e-01 -2.01721583e-03 -1.24205321e-01 1.13257408e-01 9.03044999e-01 -6.04874134e-01 1.71573117e-01 9.05274272e-01 1.07267725e+00 4.61688042e-01 -1.35066831e+00 -7.04384267e-01 1.24542810e-01 -3.03211123e-01 -1.62119555e+00 -1.89689875e-01 5.50747156e-01 -3.00290138e-01 1.21780658e+00 4.97994959e-01 6.15584910e-01 -1.25735030e-01 4.37718987e-01 3.70063901e-01 8.05165291e-01 -5.90626478e-01 4.24086571e-01 8.20183530e-02 3.27789307e-01 1.22029102e+00 -1.20516665e-01 -4.50367816e-02 4.30562720e-03 -3.63326013e-01 8.52047265e-01 -1.77064449e-01 -4.78496999e-01 1.35924682e-01 -7.26636410e-01 1.21817935e+00 3.78091842e-01 7.63203427e-02 -1.76284626e-01 7.48073339e-01 4.18829232e-01 4.12673563e-01 5.18248379e-01 5.15834033e-01 -9.13975120e-01 -1.92935124e-01 -1.00581777e+00 4.01986897e-01 1.26697862e+00 7.55967736e-01 1.12606072e+00 2.74154186e-01 2.85686672e-01 7.37528324e-01 2.74100471e-02 4.24153090e-01 3.45443964e-01 -1.46363652e+00 1.81317061e-01 1.69358224e-01 1.21603213e-01 -7.69518793e-01 -4.92063403e-01 -4.36046660e-01 -8.73884022e-01 5.84825873e-01 5.78416765e-01 -4.09021944e-01 -4.83509272e-01 1.86148512e+00 2.13068202e-01 -2.13245332e-01 -1.19375654e-01 6.74060106e-01 1.15434527e-01 8.10173392e-01 -2.45989934e-01 -5.75384796e-01 1.05138063e+00 -7.92171597e-01 -2.20557511e-01 1.49393454e-01 8.70852351e-01 -6.70661211e-01 1.22986066e+00 6.01275206e-01 -1.67954969e+00 -2.48639241e-01 -1.12744224e+00 -7.45022669e-02 -4.34552789e-01 2.04479679e-01 8.92453313e-01 6.40683591e-01 -1.39686084e+00 9.85680163e-01 -8.04736078e-01 5.58790267e-01 2.09362313e-01 8.62523079e-01 -6.30667135e-02 5.83704054e-01 -1.15748560e+00 7.31323719e-01 1.49789512e-01 2.35660717e-01 -4.25141692e-01 -1.16634166e+00 -8.51671636e-01 6.78541735e-02 9.26949084e-02 -5.28284311e-01 1.41710567e+00 -1.21589816e+00 -1.69627881e+00 8.37295473e-01 -6.77234754e-02 -8.59911799e-01 3.78898531e-01 6.35636300e-02 1.19017467e-01 4.50901501e-02 -1.10401154e-01 4.46778119e-01 8.69675636e-01 -7.44828165e-01 -5.98042786e-01 -4.28082317e-01 2.30345711e-01 -1.13008283e-01 -2.67072201e-01 1.62570015e-01 -2.34551102e-01 -4.14334238e-01 -1.71691537e-01 -8.47989917e-01 -4.83284801e-01 9.57699865e-02 -1.97264940e-01 -1.10672206e-01 2.63332069e-01 -6.35497570e-01 1.41696215e+00 -2.02956796e+00 -7.19110221e-02 6.17616594e-01 3.29849064e-01 2.59901822e-01 4.49748129e-01 -4.62903269e-02 -2.78015554e-01 2.44660124e-01 -5.78395069e-01 -1.10661723e-01 8.98070484e-02 5.32657690e-02 -3.54292005e-01 7.02969670e-01 -9.65511501e-02 5.95925510e-01 -5.40789306e-01 -3.79519701e-01 -1.09734133e-01 6.41319811e-01 -8.75633657e-01 -1.61175057e-01 -4.78183895e-01 -1.81881160e-01 -3.01252365e-01 2.90269077e-01 6.90270424e-01 -2.81912625e-01 -1.59201413e-01 -1.62734374e-01 -4.10758853e-01 2.50971884e-01 -1.37769699e+00 1.28908861e+00 -7.52710938e-01 3.84719878e-01 7.12402225e-01 -1.41569388e+00 7.10734367e-01 -1.18212037e-01 6.91039622e-01 -2.81154752e-01 3.19201529e-01 4.31691140e-01 -1.83783948e-01 1.57264441e-01 3.14816535e-01 -2.92838126e-01 -1.14107013e-01 4.64627117e-01 -1.90181404e-01 -4.63984400e-01 3.10361415e-01 -5.38814478e-02 1.02593255e+00 -6.99096322e-02 1.49213478e-01 -4.44172770e-01 7.48277485e-01 1.68731943e-01 3.62611085e-01 5.58670878e-01 -1.40741885e-01 3.13135356e-01 8.55488241e-01 -5.20419300e-01 -9.13763344e-01 -7.60049641e-01 -4.21086103e-01 1.43671000e+00 -2.49645516e-01 -3.43657136e-01 -1.06576419e+00 -3.46627384e-01 2.97400616e-02 7.16406465e-01 -6.18798375e-01 -4.03930284e-02 -9.60264683e-01 -8.98545325e-01 6.26991153e-01 6.20008290e-01 6.52795017e-01 -7.44242489e-01 -7.38737643e-01 2.99064904e-01 4.42338347e-01 -4.65338737e-01 -8.12081277e-01 7.74614036e-01 -1.00988853e+00 -7.82666385e-01 -6.42131984e-01 -7.04021573e-01 7.38974631e-01 -5.08006692e-01 1.15926754e+00 2.34555051e-01 -4.68787670e-01 3.65796119e-01 3.04902852e-01 -2.32194588e-01 -2.67593145e-01 -3.07298787e-02 -5.97278848e-02 -3.78476113e-01 -1.36543483e-01 -5.92291594e-01 -6.74366295e-01 1.10953838e-01 -7.40015090e-01 -3.88441801e-01 5.97297177e-02 1.03240526e+00 9.54921722e-01 1.04256831e-01 2.46475369e-01 -4.63356078e-01 8.22327495e-01 -7.35765398e-02 -1.40315354e+00 1.35247886e-01 -7.26405859e-01 4.69842494e-01 1.35718584e+00 -4.37657803e-01 -3.85045737e-01 2.50152331e-02 -4.09753174e-01 -6.31784976e-01 4.55664992e-01 5.07415175e-01 1.82945848e-01 -4.93423164e-01 8.38530660e-01 6.99673817e-02 2.80416310e-01 -2.51179188e-01 4.94471699e-01 3.02741021e-01 7.44889200e-01 -1.06503332e+00 3.25392276e-01 3.79321337e-01 3.03899080e-01 -6.14368856e-01 -5.93308210e-01 6.20236024e-02 -3.47236767e-02 2.03441575e-01 5.63336849e-01 -5.10439634e-01 -1.42371523e+00 1.35346785e-01 -1.17454231e+00 -6.20422482e-01 -6.65532649e-01 2.93759853e-01 -6.74781799e-01 1.61142036e-01 -6.70080245e-01 -8.17659557e-01 -8.05702806e-01 -1.34111762e+00 7.59078681e-01 2.12087274e-01 -1.46624431e-01 -1.19943631e+00 -1.58910528e-01 -2.12929606e-01 4.62870270e-01 1.05399944e-01 1.04660237e+00 -6.07594907e-01 -2.94391006e-01 -3.13825667e-01 -9.01973546e-02 9.27274823e-01 -3.61035556e-01 2.74839520e-01 -5.28534353e-01 -4.85792495e-02 5.11546850e-01 -1.84760213e-01 6.59898102e-01 5.90731442e-01 1.52714658e+00 -8.94298136e-01 -1.50269061e-01 1.28333223e+00 1.37591279e+00 2.15409577e-01 3.53542447e-01 3.70948553e-01 2.22251534e-01 3.16960365e-02 2.30868205e-01 6.38697803e-01 2.82824282e-02 3.66459072e-01 2.63632566e-01 1.59327723e-02 5.78243732e-01 4.00379062e-01 3.90862077e-01 4.17713672e-01 -1.68013930e-01 4.20066208e-01 -6.82229221e-01 2.50034451e-01 -1.50174963e+00 -6.84531510e-01 -5.80393896e-03 2.46932864e+00 1.36235988e+00 2.35601604e-01 1.73178032e-01 1.69526428e-01 4.77418929e-01 -1.93250656e-01 -6.45996094e-01 -1.04725051e+00 2.70542115e-01 1.20479906e+00 1.07271135e+00 9.17487144e-01 -8.49651992e-01 7.15521455e-01 6.24240351e+00 9.96429086e-01 -1.33606875e+00 1.26645369e-02 9.24022436e-01 -3.70560050e-01 -3.06137979e-01 7.71037713e-02 -1.00031042e+00 4.97013450e-01 9.65632796e-01 -2.08619595e-01 9.67242837e-01 1.40331972e+00 -8.47101063e-02 2.81768311e-02 -1.08378458e+00 9.41629589e-01 -2.58355647e-01 -1.35339665e+00 -6.07923865e-01 2.00713966e-02 6.24456644e-01 -2.00024292e-01 3.17400873e-01 4.66365486e-01 5.86205482e-01 -1.36286736e+00 5.87572515e-01 -1.33379977e-02 8.50782573e-01 -1.16155028e+00 2.80983061e-01 2.96096295e-01 -1.14092970e+00 4.07051593e-02 -3.57313484e-01 -1.15140304e-01 -1.77810807e-02 8.45566273e-01 -5.75982213e-01 1.21840008e-01 5.12858331e-01 -1.32806182e-01 9.73691046e-02 6.05177701e-01 3.64629645e-03 4.75731939e-01 -9.72697675e-01 -2.18720406e-01 4.07550812e-01 -5.56504786e-01 2.03467235e-01 1.18773925e+00 1.72677770e-01 4.89160925e-01 6.77764043e-02 1.00063384e+00 -1.44441828e-01 3.38614762e-01 -1.97662175e-01 2.44744718e-01 2.39866406e-01 1.24384069e+00 -6.02038264e-01 -3.64657044e-01 -1.64051488e-01 5.88230729e-01 4.92632538e-01 4.71305609e-01 -1.21621573e+00 -1.01231396e+00 9.66589332e-01 -3.22030362e-04 4.02377695e-01 -5.23114502e-01 -6.43177569e-01 -9.36671436e-01 2.32695535e-01 -9.38003004e-01 4.68143940e-01 -4.76685226e-01 -7.38085270e-01 5.60529947e-01 6.14304580e-02 -6.11549199e-01 -6.51800871e-01 -8.85560989e-01 -4.98434842e-01 1.31545854e+00 -1.21201217e+00 -4.50628847e-01 3.34862947e-01 7.06838429e-01 -3.86469215e-02 -7.61682987e-02 8.70295286e-01 1.04373008e-01 -4.95637119e-01 1.11930883e+00 2.21769363e-01 1.71430901e-01 1.85702667e-01 -1.10953546e+00 -2.13945471e-02 5.87505996e-01 -2.38602996e-01 9.72356498e-01 6.77543402e-01 -2.64910281e-01 -1.55494022e+00 -7.50901222e-01 5.32838404e-01 4.63529415e-02 8.36573780e-01 1.95982400e-02 -8.15729022e-01 9.18516576e-01 -5.53217940e-02 3.17433774e-01 5.78600466e-01 -4.51527312e-02 -2.69241720e-01 -5.37005723e-01 -1.49524188e+00 6.29878402e-01 5.16373873e-01 -2.17730910e-01 -1.39059564e-02 3.81180227e-01 6.71123445e-01 -8.44377220e-01 -1.24202466e+00 3.61517191e-01 4.86146659e-01 -9.49540377e-01 1.20731902e+00 -5.49992859e-01 2.67744929e-01 -1.42105654e-01 -2.30483681e-01 -1.03890014e+00 2.34644517e-01 -1.35436773e+00 -2.12640390e-01 6.53653145e-01 5.19918621e-01 -1.06693840e+00 5.83074331e-01 9.73703563e-01 -3.31116319e-01 -1.23329341e+00 -1.08937824e+00 -6.30429208e-01 7.47007966e-01 -6.35869861e-01 6.01418793e-01 5.11785805e-01 7.91250318e-02 -1.91182628e-01 2.21788540e-01 5.24164364e-02 4.66520220e-01 1.72953367e-01 5.21671116e-01 -6.34581923e-01 -7.95428574e-01 -1.02476859e+00 -1.88849881e-01 -1.05945444e+00 3.36279184e-01 -8.71886790e-01 -5.14482148e-02 -7.92257905e-01 -3.69955093e-01 -8.93011391e-01 -5.76004200e-02 6.64427161e-01 1.25355601e-01 8.69935527e-02 -1.53107196e-01 -7.67830163e-02 -1.44234136e-01 2.27569461e-01 9.73366499e-01 -1.02348477e-02 -4.16376621e-01 2.45723370e-02 -7.78826296e-01 1.11788344e+00 7.21825957e-01 -1.75070524e-01 -9.83875170e-02 -3.45772356e-01 4.61188108e-01 3.94468397e-01 2.95658946e-01 -7.69860029e-01 9.40274298e-02 -1.94408193e-01 1.97882816e-01 -1.60701901e-01 2.67534286e-01 -3.98900181e-01 -1.18357249e-01 3.88004035e-01 -3.51368099e-01 3.45006764e-01 3.57437104e-01 -3.05788845e-01 7.36234412e-02 -6.27378523e-01 1.10352945e+00 -1.82903573e-01 1.10237785e-02 3.40100318e-01 4.16280478e-02 3.24387491e-01 7.97462046e-01 2.56751597e-01 -1.08763807e-01 -3.06492686e-01 -6.35331094e-01 1.59374446e-01 3.98241788e-01 -5.56124806e-01 2.70260394e-01 -1.14884663e+00 -4.39729512e-01 3.06905717e-01 -6.81968391e-01 4.82471466e-01 -1.77848086e-01 8.89243841e-01 -9.50571120e-01 1.66333362e-01 2.74129272e-01 -5.00920832e-01 -7.30138838e-01 3.92315924e-01 7.91200519e-01 -4.62588012e-01 -1.71553656e-01 1.18151593e+00 -4.69163358e-02 -6.09465912e-02 4.06513005e-01 -8.48343909e-01 5.37637234e-01 -3.02351773e-01 6.16011798e-01 4.52821672e-01 2.38551214e-01 -1.31287262e-01 -3.20894420e-01 5.50398290e-01 -3.38572897e-02 -5.25043607e-01 1.23123837e+00 4.19508785e-01 -5.52894354e-01 1.37117654e-01 1.87918782e+00 3.82393748e-02 -1.13661528e+00 3.56289178e-01 -6.27963841e-01 -1.83549442e-03 -2.07828671e-01 -5.68946004e-01 -1.43292677e+00 8.14686120e-01 3.47470522e-01 3.09826463e-01 1.40286314e+00 -3.16234946e-01 7.80510306e-01 6.00962818e-01 3.24846685e-01 -1.20719481e+00 -4.40574706e-01 7.54949927e-01 7.03878701e-01 -7.92302489e-01 1.08500667e-01 3.03999819e-02 -1.01017706e-01 1.38439250e+00 3.16544980e-01 -5.30843735e-01 9.41785157e-01 7.86268175e-01 -1.16694987e-01 1.09993927e-01 -4.63343710e-01 2.17096582e-01 1.30213007e-01 8.08079988e-02 6.78917170e-01 -1.21203542e-01 -7.12592363e-01 8.80399346e-01 -7.85492599e-01 6.03050403e-02 2.54788339e-01 9.02633727e-01 -5.81664205e-01 -1.01546609e+00 -6.13296270e-01 3.62288028e-01 -7.95199156e-01 -3.05416644e-01 2.06294268e-01 9.44979310e-01 7.47574344e-02 4.21281695e-01 2.64246594e-02 5.09305345e-03 1.55620314e-02 2.96614856e-01 7.18688130e-01 -2.60145515e-02 -8.42318654e-01 9.57707968e-03 -1.58885047e-01 -5.60231209e-01 2.26820365e-01 -4.75492805e-01 -2.05780339e+00 -6.36858821e-01 -2.46015087e-01 4.05374318e-01 1.12236297e+00 7.83341706e-01 1.25252753e-01 2.58743584e-01 6.84042156e-01 -7.12982178e-01 -1.06471145e+00 -3.89124990e-01 -5.05585909e-01 -5.74328266e-02 3.76479954e-01 -2.51612544e-01 -7.61974633e-01 -1.66970771e-02]

[7.338155269622803, 3.673576593399048]

2f9d3a2e-f5aa-47c2-b71a-4ab1ec6f1532

utterance-level-aggregation-for-speaker

1902.10107

https://arxiv.org/abs/1902.10107v2

https://arxiv.org/pdf/1902.10107v2.pdf

Utterance-level Aggregation For Speaker Recognition In The Wild

The objective of this paper is speaker recognition "in the wild"-where utterances may be of variable length and also contain irrelevant signals. Crucial elements in the design of deep networks for this task are the type of trunk (frame level) network, and the method of temporal aggregation. We propose a powerful speaker recognition deep network, using a "thin-ResNet" trunk architecture, and a dictionary-based NetVLAD or GhostVLAD layer to aggregate features across time, that can be trained end-to-end. We show that our network achieves state of the art performance by a significant margin on the VoxCeleb1 test set for speaker recognition, whilst requiring fewer parameters than previous methods. We also investigate the effect of utterance length on performance, and conclude that for "in the wild" data, a longer length is beneficial.

['Weidi Xie', 'Joon Son Chung', 'Arsha Nagrani', 'Andrew Zisserman']

2019-02-26

utterance-level-aggregation-for-speaker-1

['text-independent-speaker-verification']

['speech']

[ 2.25628857e-02 -9.96335000e-02 8.16264674e-02 -7.97751129e-01 -8.61515343e-01 -3.63067478e-01 5.43049037e-01 -3.07103485e-01 -5.27619123e-01 4.05919731e-01 4.19210017e-01 -3.63335878e-01 1.45760819e-01 -2.45485276e-01 -5.13535202e-01 -7.87970841e-01 -2.98459500e-01 8.81781131e-02 9.15080160e-02 -3.16697866e-01 -2.86840677e-01 6.34289861e-01 -1.55358291e+00 3.49051982e-01 1.23599246e-01 1.15640604e+00 -2.27103338e-01 9.43790376e-01 -5.43422028e-02 1.06848788e+00 -1.06246972e+00 -3.10152411e-01 9.58080292e-02 -3.81608844e-01 -8.12617958e-01 1.45390034e-01 8.49996924e-01 -3.82800877e-01 -5.58219910e-01 6.87788010e-01 1.02398920e+00 4.50950354e-01 3.21465760e-01 -9.62541044e-01 -1.86391383e-01 9.33823466e-01 -1.27163470e-01 6.96807444e-01 2.68804699e-01 2.45966360e-01 8.25425982e-01 -8.88879299e-01 5.02900422e-01 1.47408378e+00 8.01828086e-01 8.92745316e-01 -1.31641054e+00 -8.06132793e-01 4.43968743e-01 1.98574871e-01 -1.51676977e+00 -1.50463331e+00 7.58355379e-01 -2.69599646e-01 1.27403641e+00 3.90129894e-01 1.90837294e-01 1.68289459e+00 7.14132860e-02 7.66569614e-01 5.76474786e-01 -3.53248924e-01 3.85890633e-01 -1.44815341e-01 3.55975151e-01 4.55887645e-01 -7.26648211e-01 3.04430246e-01 -9.20043230e-01 5.56413867e-02 5.53084552e-01 -3.77399951e-01 -1.97904840e-01 2.61546433e-01 -9.45560396e-01 8.43245387e-01 1.69274092e-01 4.87913072e-01 -2.91119277e-01 2.21736148e-01 7.37015128e-01 7.99851835e-01 7.86288559e-01 3.34204659e-02 -5.26682973e-01 -2.77201295e-01 -1.34620988e+00 3.40874940e-01 8.88853192e-01 6.81165874e-01 2.14773148e-01 7.02610254e-01 -2.25121394e-01 1.16671932e+00 2.17678189e-01 1.88560188e-01 6.31627679e-01 -8.65378857e-01 3.87644917e-01 -1.16621532e-01 -1.80262625e-01 -4.58783001e-01 -5.42850375e-01 -6.48479760e-01 -8.51761162e-01 1.56916544e-01 4.41667348e-01 -5.52834928e-01 -1.12428260e+00 2.07641125e+00 1.91568345e-01 4.34950531e-01 1.57322422e-01 9.07534063e-01 1.14862299e+00 7.09955335e-01 3.03722378e-02 -2.66630411e-01 1.25537181e+00 -7.96714902e-01 -8.16565096e-01 -2.51931194e-02 5.89476347e-01 -6.27364814e-01 8.50575328e-01 3.80594373e-01 -1.20019186e+00 -7.19565868e-01 -9.83471155e-01 -9.86036882e-02 -2.20263362e-01 -9.23219472e-02 2.56343931e-01 8.34982097e-01 -1.63324761e+00 6.37339652e-01 -8.42787445e-01 -3.07635754e-01 2.08234817e-01 6.27993882e-01 -2.14778006e-01 3.32320750e-01 -1.14991534e+00 7.95709312e-01 9.06187594e-02 2.13454485e-01 -1.14018226e+00 -7.80129492e-01 -8.29056501e-01 1.40469491e-01 3.00173033e-02 -3.60031605e-01 1.61382985e+00 -1.21237004e+00 -1.75090694e+00 5.31258464e-01 -5.51170945e-01 -8.68300200e-01 5.66754580e-01 7.19319135e-02 -7.88968801e-01 1.20610557e-03 -3.19425434e-01 7.82599747e-01 1.08798850e+00 -7.03459918e-01 -5.79936326e-01 -3.74587208e-01 -4.29846942e-02 -2.33599972e-02 -2.57109016e-01 3.32988322e-01 -2.12558225e-01 -8.86382043e-01 -6.91314191e-02 -7.39410818e-01 -7.94639736e-02 -3.30407768e-01 -2.22634688e-01 -5.90482533e-01 1.03854120e+00 -9.10864413e-01 9.40786004e-01 -2.59164810e+00 1.05830720e-02 1.10928476e-01 1.72330573e-01 1.24635272e-01 -4.83766615e-01 1.86594829e-01 -3.09029728e-01 -3.21299955e-02 1.91564173e-01 -8.22778702e-01 5.59336990e-02 8.83421972e-02 -4.67280030e-01 6.08646870e-01 1.61297068e-01 6.64533079e-01 -4.24159199e-01 -2.45892495e-01 1.48263320e-01 7.52439082e-01 -3.58088315e-01 2.69214392e-01 -2.11502209e-01 3.66366982e-01 2.97991503e-02 4.08474386e-01 5.33527911e-01 2.39478305e-01 -9.61741135e-02 -1.25645265e-01 -2.29475185e-01 7.90204227e-01 -1.08725798e+00 1.79188967e+00 -5.69163442e-01 1.16500390e+00 5.02980709e-01 -6.95023119e-01 8.57350707e-01 8.90865088e-01 3.48610193e-01 -7.60555327e-01 2.94065058e-01 -2.54273806e-02 1.00400142e-01 -2.91699678e-01 3.98725897e-01 -1.98893040e-01 8.33851621e-02 3.15918654e-01 3.31395715e-01 3.22885543e-01 9.55628976e-02 2.53519341e-02 1.33765173e+00 -4.70481813e-01 -3.26202452e-01 -4.16542292e-01 4.11887795e-01 -6.97337985e-01 6.08404219e-01 7.62698472e-01 -3.00041348e-01 5.74636579e-01 4.38598692e-01 -5.64674437e-01 -8.45039070e-01 -9.15461659e-01 -3.19579214e-01 1.62662876e+00 -3.44290465e-01 -8.94613117e-02 -7.31764674e-01 -4.78550851e-01 -1.80479109e-01 7.48841047e-01 -5.62780201e-01 1.18063502e-01 -7.87707329e-01 -5.34497797e-01 1.12070227e+00 6.60056591e-01 3.26724142e-01 -1.13448799e+00 -2.88105696e-01 6.27202392e-01 -9.21627358e-02 -1.28064811e+00 -7.29469359e-01 5.57620764e-01 -6.42721355e-01 -4.00063276e-01 -6.53670609e-01 -8.22389662e-01 2.13446870e-01 -7.95370638e-02 1.11976910e+00 -2.08205998e-01 7.99798686e-03 1.76384494e-01 -3.28105569e-01 -3.57634783e-01 -4.73179549e-01 2.38378599e-01 2.04619870e-01 1.52067274e-01 3.07183743e-01 -5.66400707e-01 -4.02735591e-01 1.59594551e-01 -6.64130270e-01 -2.40441784e-01 1.26229614e-01 9.35989559e-01 -2.98034474e-02 8.56022071e-03 9.41380918e-01 -5.03110647e-01 7.04261839e-01 -8.10409635e-02 -5.34254611e-01 -9.50208008e-02 1.38963647e-02 1.62111104e-01 5.74883163e-01 -4.62106317e-01 -9.63752627e-01 -4.07597423e-02 -7.88245678e-01 -6.03998542e-01 -3.01088214e-01 3.05451483e-01 -2.02196643e-01 -8.88900310e-02 6.71801150e-01 1.65144354e-01 1.40053943e-01 -6.26681268e-01 1.71529129e-01 7.10821092e-01 5.44790983e-01 -3.48018050e-01 1.63638666e-01 2.93292403e-01 -4.62192386e-01 -1.32819200e+00 -5.28576791e-01 -4.26698387e-01 -4.22918975e-01 -1.90746129e-01 7.02141523e-01 -1.18462968e+00 -9.03057754e-01 5.32650650e-01 -1.34574795e+00 -6.65041208e-01 -4.58514661e-01 3.79265249e-01 -2.42768750e-01 7.72252679e-02 -9.36592579e-01 -9.48681295e-01 -4.14605767e-01 -1.19403028e+00 1.00760090e+00 1.51148736e-02 -3.13843369e-01 -8.83679211e-01 -1.95623152e-02 1.32317349e-01 5.80870926e-01 -4.70642522e-02 4.33915436e-01 -8.62312436e-01 -3.09946984e-01 2.31983718e-02 -1.11478955e-01 4.62057441e-01 -1.19877666e-01 -6.71591833e-02 -1.61039138e+00 -4.81765300e-01 2.48700753e-01 -1.78547978e-01 1.07588089e+00 7.20909595e-01 1.05784476e+00 -5.38027704e-01 7.76243657e-02 7.08955109e-01 9.63191330e-01 3.91677797e-01 5.23596883e-01 -1.94284096e-01 5.77507019e-01 6.74581051e-01 -2.43475258e-01 4.20726359e-01 3.22011828e-01 9.27141607e-01 1.35151535e-01 -1.07075654e-01 -3.87326688e-01 3.62538546e-01 7.94469476e-01 1.05587065e+00 1.86356425e-01 -5.19440949e-01 -6.99937403e-01 6.83208048e-01 -1.63731313e+00 -1.31147802e+00 2.88617402e-01 1.94082284e+00 6.07256413e-01 9.23394486e-02 4.71204251e-01 2.48417392e-01 6.27714753e-01 5.30544698e-01 -5.04675806e-01 -7.66011477e-01 -2.68895119e-01 4.65567559e-01 1.80198595e-01 7.43923843e-01 -1.16671634e+00 9.25527573e-01 6.82528162e+00 7.54841447e-01 -1.63929331e+00 3.55149657e-01 5.91494441e-01 -4.94220793e-01 1.25370637e-01 -5.27481318e-01 -1.01112056e+00 4.72770691e-01 1.67812467e+00 2.84273416e-01 3.62498373e-01 5.42350590e-01 4.42605942e-01 3.89747620e-01 -1.38216138e+00 9.78439510e-01 8.87165964e-02 -1.32964063e+00 -4.17309582e-01 7.86041189e-03 3.48790288e-01 4.24953908e-01 1.59124196e-01 4.58382905e-01 4.06521440e-01 -1.03834128e+00 9.46886241e-01 1.03695430e-01 6.91839159e-01 -7.49331474e-01 6.89593136e-01 1.44015774e-01 -1.34995258e+00 -1.68020353e-02 -1.30720451e-01 1.79431915e-01 1.36586577e-01 4.35403705e-01 -1.00966370e+00 1.93334147e-01 6.54515386e-01 5.26560366e-01 -4.27944869e-01 7.07960248e-01 1.55953735e-01 8.48269820e-01 -5.27121425e-01 -8.42271969e-02 3.00273269e-01 5.39997160e-01 5.93672335e-01 1.72588050e+00 2.38555416e-01 -1.03189781e-01 1.09103203e-01 4.72161323e-01 -2.40642011e-01 -2.03489751e-01 -4.83717382e-01 2.55691230e-01 5.21111667e-01 7.78462470e-01 -4.11181957e-01 -3.48775417e-01 -2.54114091e-01 7.01712728e-01 1.36792988e-01 6.28772438e-01 -6.71111226e-01 -3.27393889e-01 8.35572541e-01 2.36033164e-02 6.71095788e-01 -1.65920988e-01 -1.21641695e-01 -1.25542569e+00 4.89346795e-02 -1.00471187e+00 3.56291503e-01 -3.88450891e-01 -1.29350114e+00 1.01865184e+00 -3.61966878e-01 -8.47941577e-01 -6.89737499e-01 -3.82891566e-01 -7.19288886e-01 8.19334924e-01 -1.16988575e+00 -9.32692587e-01 -2.64349077e-02 8.91145706e-01 9.56961930e-01 -3.31032246e-01 7.34624803e-01 5.04808247e-01 -7.21368134e-01 1.06259739e+00 1.23853505e-01 4.01444435e-01 4.49622124e-01 -1.03530061e+00 7.00694442e-01 9.82490003e-01 3.60659957e-01 4.54828143e-01 8.11524332e-01 -1.58735409e-01 -1.31774771e+00 -1.13408029e+00 9.45013940e-01 -2.92244971e-01 3.58370245e-01 -7.54946887e-01 -9.04067874e-01 6.81598365e-01 4.82535869e-01 -6.74800351e-02 5.82836509e-01 5.36359787e-01 -6.56310797e-01 -4.00065958e-01 -1.04650557e+00 5.87822676e-01 9.47883606e-01 -8.67342591e-01 -4.95971769e-01 1.99815527e-01 9.04201925e-01 -4.89407271e-01 -6.80843294e-01 1.98046878e-01 7.29216456e-01 -1.02850664e+00 9.96502221e-01 -4.56190675e-01 -1.12172790e-01 2.53137257e-02 -2.55730569e-01 -1.27928329e+00 -1.49966031e-01 -7.57813156e-01 -9.44532976e-02 1.36477697e+00 5.28866887e-01 -4.78433669e-01 9.27789152e-01 5.89388907e-01 -3.92304540e-01 -3.35986227e-01 -1.53296399e+00 -9.86289859e-01 -1.29276276e-01 -9.01249051e-01 6.75501585e-01 7.83433974e-01 -2.19905362e-01 6.29652739e-01 -4.74113613e-01 2.19547942e-01 5.87817371e-01 -3.38017732e-01 5.67316115e-01 -1.19000864e+00 -1.64734513e-01 -5.77591896e-01 -5.79140902e-01 -1.11856043e+00 4.63911802e-01 -6.69378996e-01 2.29308143e-01 -8.18330646e-01 -3.70387793e-01 -1.67672142e-01 -4.03281659e-01 3.84735733e-01 3.19507986e-01 1.48787880e-02 2.47603431e-01 -7.59362802e-02 -4.55724597e-01 4.14204687e-01 6.28548205e-01 -2.49401435e-01 -5.39516568e-01 -1.21308431e-01 -3.25452954e-01 5.07101893e-01 5.87313712e-01 -3.18465024e-01 -2.44532287e-01 -5.85640311e-01 -5.15333891e-01 3.92498702e-01 4.02965218e-01 -1.04803097e+00 4.87751544e-01 2.97689497e-01 3.10036927e-01 -7.20363736e-01 8.39840591e-01 -5.52690804e-01 7.47992024e-02 1.51369840e-01 -7.85949349e-01 1.40239522e-01 3.22869450e-01 1.71006322e-01 -2.98748523e-01 -2.41436791e-02 9.19373453e-01 2.30813935e-01 -5.99024355e-01 3.51459920e-01 -4.89936382e-01 4.16720212e-02 3.51148427e-01 -2.79877195e-03 -1.62278637e-01 -5.13524771e-01 -9.40311015e-01 -1.33552164e-01 -3.88762541e-02 4.41093802e-01 5.14148235e-01 -1.27734447e+00 -9.52710330e-01 4.92380917e-01 -1.99983895e-01 -1.62941501e-01 4.62367028e-01 6.95781589e-01 -2.47043744e-01 4.44836348e-01 1.67741776e-01 -8.09069633e-01 -1.35677969e+00 4.86333966e-02 5.26685536e-01 -6.59561157e-02 -6.74915373e-01 1.24060512e+00 2.29431912e-01 -2.32683003e-01 9.72918689e-01 -4.21297461e-01 -7.02766404e-02 3.71199399e-01 8.02450538e-01 1.59941688e-01 4.53157097e-01 -8.54835212e-01 -5.03993928e-01 -5.25128208e-02 -4.81455058e-01 -5.47475278e-01 1.44495761e+00 -1.00258298e-01 2.49179006e-01 5.90689063e-01 1.43762553e+00 -2.61827707e-01 -1.23447037e+00 -3.81587893e-01 -1.94355324e-01 -8.68134275e-02 3.84747684e-01 -5.37246644e-01 -1.24168074e+00 1.03741813e+00 8.82309437e-01 4.66866761e-01 9.94967401e-01 -9.93207395e-02 7.20723093e-01 2.46716917e-01 7.76497871e-02 -9.12139177e-01 -7.10424110e-02 7.25676656e-01 9.37317431e-01 -1.05423141e+00 -5.12353480e-01 1.90690562e-01 -3.62857878e-01 1.11288774e+00 3.33660245e-01 3.81721407e-02 8.04094911e-01 6.68520331e-01 3.06020498e-01 -1.25292495e-01 -1.16261184e+00 -8.80815834e-03 1.86564237e-01 5.08523405e-01 7.40672231e-01 5.05681382e-03 3.41525257e-01 3.65074784e-01 -4.74112600e-01 -2.52549589e-01 2.31987670e-01 6.39349341e-01 -1.78945571e-01 -9.58617330e-01 -4.06258702e-01 4.01502192e-01 -7.17329383e-01 -1.84554800e-01 -2.23326609e-01 4.58547354e-01 3.51366866e-03 1.28409958e+00 3.27951401e-01 -4.97431844e-01 4.23818171e-01 3.04476291e-01 2.50393420e-01 -3.78946334e-01 -1.06589234e+00 4.20696110e-01 3.52299064e-01 -4.71610516e-01 -3.85414213e-01 -8.02071273e-01 -7.71117151e-01 -7.52603889e-01 -3.90746713e-01 5.35141453e-02 7.07826793e-01 1.01781511e+00 4.01307851e-01 8.87763500e-01 6.94967806e-01 -9.48284924e-01 -5.36009848e-01 -1.13002443e+00 -6.02754951e-01 1.04055069e-01 9.89432633e-01 -2.40413636e-01 -4.94640529e-01 2.41715550e-01]

[14.430797576904297, 6.077672004699707]

c85e8d79-4822-4f0c-9d64-3f4be0bf218a

correspondence-on-acmg-statement-acmg-sf-v3-0

2203.04684

https://arxiv.org/abs/2203.04684v1

https://arxiv.org/pdf/2203.04684v1.pdf

Correspondence on ACMG STATEMENT: ACMG SF v3.0 list for reporting of secondary findings in clinical exome and genome sequencing: a policy statement of the American College of Medical Genetics and Genomics (ACMG) by Miller et al

We were interested to read the recent update on recommendations for reporting of secondary findings in clinical sequencing1, and the accompanying updated list of genes in which secondary findings should be sought (ACMG SF v3.0)2. Though the authors discuss challenges around incomplete penetrance in considerable detail, we are concerned that the recommendations do not fully convey the degree of uncertainty regarding the penetrance of variants in genes associated with inherited cardiomyopathies, which make up almost a quarter of the list. Since penetrance is incomplete and age-related, individuals found to carry variants will often require surveillance, rather than a one-off definitive diagnostic assessment. There is a lack of evidence regarding benefits, harms, and healthcare costs associated with opportunistic screening.

['James S. Ware', 'Declan P. O Regan', 'Thomas R. Lumbers', 'Angharad Roberts', 'Nicola Whiffin', 'Antonio de Marvao', 'Matthew Edwards', 'Katherine Josephs', 'Albert Henry', 'Sean L. Zheng', 'Kathryn A. McGurk']

2022-03-09

['medical-genetics']

['miscellaneous']

[ 6.43147469e-01 3.20576727e-01 -3.76562685e-01 -3.56521875e-01 -7.00889289e-01 -7.92087018e-01 -4.98939008e-02 5.66533804e-01 -4.16446477e-01 1.06118274e+00 4.15148824e-01 -8.47212434e-01 -4.93469745e-01 -2.56196916e-01 -4.73846346e-01 -3.56384397e-01 -2.02135623e-01 4.09699351e-01 1.08628504e-01 2.41313159e-01 -7.29252864e-03 1.18441939e-01 -1.01943016e+00 1.77324980e-01 1.17675161e+00 1.73929706e-01 -9.63123664e-02 7.15010166e-01 -2.84169975e-04 3.52301180e-01 -6.64241910e-01 -5.73472261e-01 -2.06925392e-01 -9.29915130e-01 -8.71661305e-01 3.82727152e-03 -4.28229794e-02 -7.25840330e-01 1.96217686e-01 8.37123454e-01 7.77817190e-01 -4.30979878e-01 2.91565597e-01 -6.03353918e-01 -4.61592197e-01 6.76080287e-01 -3.17453116e-01 6.35324180e-01 4.54978377e-01 3.52334559e-01 7.99587250e-01 -1.73701584e-01 8.80225003e-01 8.46238494e-01 8.03803205e-01 7.25618124e-01 -1.04719198e+00 -2.53017932e-01 6.34482726e-02 -9.38052908e-02 -1.22202945e+00 -5.11426747e-01 -8.46609175e-02 -4.60976213e-01 1.08911777e+00 6.24992371e-01 1.31692851e+00 6.76346123e-01 9.11058625e-04 5.93899451e-02 6.62023485e-01 -2.07783654e-01 -8.47887620e-02 -4.47655201e-01 -6.68243691e-02 5.28351128e-01 9.67161238e-01 -3.14531699e-02 -5.45608699e-01 -7.72621572e-01 4.53919291e-01 -4.02983755e-01 -3.17014843e-01 -6.04941137e-02 -8.75424147e-01 4.64663684e-01 -6.93083763e-01 2.06315532e-01 -1.84522703e-01 5.27038202e-02 8.43531668e-01 1.08583011e-01 1.90311968e-01 -2.68101040e-02 -5.85344255e-01 -6.29026234e-01 -7.67512679e-01 1.64965346e-01 4.41116422e-01 6.50961578e-01 2.26227399e-02 1.32800825e-02 -1.46248922e-01 8.75288248e-01 3.37201089e-01 2.77972281e-01 3.06908548e-01 -9.76051211e-01 8.77126902e-02 4.42862958e-01 1.51404813e-01 -2.81371266e-01 -6.22711778e-01 -3.97708952e-01 -4.22757179e-01 -1.16732195e-01 2.28779763e-01 -7.53546298e-01 -5.92924774e-01 1.85092592e+00 2.08095953e-01 -2.13563845e-01 6.73293397e-02 6.14553094e-01 6.94670677e-01 -9.82042551e-02 3.97650719e-01 -4.41237479e-01 1.28985083e+00 -1.87621698e-01 -5.43517411e-01 2.53837444e-02 6.31947100e-01 -5.70146382e-01 2.65317589e-01 4.46867615e-01 -1.29375076e+00 2.22170085e-01 -9.03330624e-01 1.21232137e-01 2.19044477e-01 8.07401445e-03 6.74742818e-01 1.31883109e+00 -1.03746676e+00 6.83774710e-01 -9.50570583e-01 -8.96287978e-01 2.61658758e-01 2.77286947e-01 -6.63319752e-02 2.50081360e-01 -1.14476192e+00 8.15825701e-01 4.56169903e-01 8.45712051e-02 -6.22675180e-01 -9.70632374e-01 -3.82930845e-01 -3.13117892e-01 4.28085953e-01 -8.38841677e-01 8.85774434e-01 -1.62505865e-01 -9.68485057e-01 1.06932998e+00 -1.64761677e-01 -1.94988564e-01 5.18934906e-01 -1.61208779e-01 -8.22173178e-01 3.84523660e-01 1.90442130e-01 1.76388800e-01 1.13877274e-01 1.33441687e-02 -8.59881043e-01 -5.06045997e-01 -3.29979844e-02 2.53795981e-01 2.03595027e-01 5.72288632e-01 -1.63263708e-01 -6.81235194e-01 -2.10582033e-01 -6.87764227e-01 -1.15999490e-01 1.10822447e-01 -5.84113002e-01 -1.50054827e-01 5.31699955e-01 -8.56443167e-01 8.57312083e-01 -1.91195762e+00 -3.73975128e-01 -1.66270301e-01 2.08693221e-01 3.92012626e-01 1.57487482e-01 6.31742120e-01 -1.19422190e-01 8.24045360e-01 -7.77782425e-02 6.33478463e-01 -2.54100878e-02 -1.64955854e-01 1.47762522e-01 8.06147099e-01 4.86672997e-01 8.03219616e-01 -9.62786674e-01 3.85221280e-02 8.29209667e-03 5.00754952e-01 -4.44801688e-01 -2.69014239e-01 1.97588690e-02 6.35362446e-01 -6.49130523e-01 9.65051889e-01 4.92948264e-01 -4.77373779e-01 9.34068024e-01 3.69345844e-01 -2.66057014e-01 6.85482860e-01 -8.23638380e-01 1.50863051e+00 6.24577999e-01 8.19002613e-02 1.10561915e-01 -6.71774089e-01 5.37625074e-01 6.68399751e-01 1.57296345e-01 6.21744804e-02 -3.51077944e-01 2.96716362e-01 7.66891599e-01 -6.00712776e-01 -1.85947850e-01 -2.65933365e-01 1.96483180e-01 4.43165958e-01 -3.44603330e-01 3.95577759e-01 4.35307384e-01 -2.00844184e-02 1.56837344e+00 3.98331702e-01 5.79410374e-01 -2.89620638e-01 2.12668210e-01 1.16964661e-01 1.17758346e+00 8.29030216e-01 -1.07520409e-01 6.09263003e-01 8.67837012e-01 -1.52954295e-01 -9.67886746e-01 -7.34738350e-01 -5.52030027e-01 3.10038865e-01 -4.16003704e-01 -7.33826101e-01 -3.30770791e-01 -4.20642942e-01 -5.15362695e-02 4.33416218e-01 -4.18417454e-01 -1.09842479e-01 -1.57979771e-01 -1.35067213e+00 1.17166567e+00 4.54847127e-01 4.62842099e-02 -1.83834419e-01 -1.10558891e+00 2.79842675e-01 4.80484925e-02 -6.55585170e-01 -2.12596580e-01 9.84059051e-02 -9.15141225e-01 -1.58481491e+00 -1.15423918e+00 -1.54961050e-01 4.34190452e-01 -4.61031526e-01 6.53636634e-01 5.88275731e-01 -7.23586500e-01 4.66981053e-01 -5.45088351e-01 -3.15214038e-01 -5.39295673e-01 -3.77322435e-01 5.00982068e-02 -7.31124699e-01 3.38256210e-01 -5.71501374e-01 -7.52556503e-01 -1.52363449e-01 -4.07362759e-01 -3.57436925e-01 6.64284348e-01 6.41076565e-01 2.59453893e-01 -2.82079041e-01 7.49894142e-01 -1.22916770e+00 5.59303463e-01 -4.58578020e-01 -3.99047226e-01 3.09732556e-01 -8.99858117e-01 -5.31563580e-01 -1.11685600e-02 -5.93816340e-02 -8.74671102e-01 -4.23403263e-01 -3.60628873e-01 4.24278349e-01 -3.23657811e-01 5.47864735e-01 -1.59579813e-02 1.17884725e-01 1.59106106e-01 2.14966331e-02 8.35004598e-02 -6.39153957e-01 -1.95804074e-01 2.91609377e-01 1.07310027e-01 -2.50491202e-01 5.21284491e-02 3.17007959e-01 4.92726527e-02 -9.29809988e-01 -2.15514407e-01 -4.13110644e-01 4.59719151e-02 1.98101193e-01 7.65793264e-01 -6.42973006e-01 -1.04588377e+00 2.07348675e-01 -7.15301991e-01 -2.29847625e-01 -1.40298948e-01 8.31838131e-01 -3.61433536e-01 7.46722996e-01 -7.52019346e-01 -8.36799324e-01 -4.65236455e-01 -1.13867748e+00 5.42766154e-01 -1.19670764e-01 -8.20998192e-01 -6.93257689e-01 3.28626722e-01 3.55619073e-01 2.80890226e-01 3.48388582e-01 1.42079985e+00 -7.35536873e-01 -1.41777217e-01 -1.89214393e-01 5.77075221e-02 -5.31822592e-02 5.34948230e-01 1.05490528e-01 -4.24067974e-01 2.73418031e-03 -4.41012830e-01 -2.91948915e-01 6.01976037e-01 6.42255723e-01 7.51187861e-01 -1.71511173e-02 -6.31283820e-01 2.76626885e-01 1.13254511e+00 4.90456104e-01 4.16984618e-01 5.66652194e-02 1.19154699e-01 4.58831489e-01 4.37528908e-01 6.63715005e-01 1.81916654e-02 5.63215733e-01 1.06759537e-02 1.43476129e-01 -2.08656490e-01 -2.71325082e-01 1.29385190e-02 2.37244427e-01 -3.51874471e-01 -4.01033700e-01 -7.24607944e-01 5.35354793e-01 -1.17961061e+00 -7.69950032e-01 -4.82797831e-01 2.33811998e+00 5.69586456e-01 -7.00989664e-02 7.13675737e-01 -7.76051953e-02 1.03670180e+00 -9.64635983e-02 -9.22708750e-01 -4.39881772e-01 -2.23225102e-01 -7.70965666e-02 3.15713465e-01 1.70803875e-01 -4.24212337e-01 3.23108047e-01 9.26912594e+00 7.73425251e-02 -8.18632066e-01 -4.41781767e-02 7.44898200e-01 -1.69916928e-01 -6.26875520e-01 5.37287354e-01 -7.18677819e-01 4.70259845e-01 1.03046679e+00 -2.03972146e-01 -1.39720008e-01 1.61917791e-01 4.42596138e-01 -1.74540862e-01 -6.88673615e-01 3.91973972e-01 -3.47426414e-01 -1.38642991e+00 -2.65900493e-01 3.76673102e-01 2.28997976e-01 1.56586617e-01 -1.51413858e-01 -3.10223460e-01 9.79624540e-02 -7.72286892e-01 4.59458739e-01 5.25592625e-01 1.25261354e+00 -6.09335601e-01 5.25837183e-01 -1.73692107e-01 -6.00829065e-01 1.72693804e-01 -3.81802917e-01 -4.96051647e-02 6.91523492e-01 9.40404534e-01 -8.84354889e-01 3.94168645e-01 7.29715586e-01 2.31741831e-01 -1.18814573e-01 1.08965635e+00 -4.14238304e-01 9.57123220e-01 -2.01808691e-01 -2.22976223e-01 -2.15598658e-01 -2.06009895e-01 7.29855239e-01 8.52934957e-01 3.69321287e-01 3.70431930e-01 -5.48945785e-01 8.09719563e-01 2.09583715e-01 2.35912010e-01 -1.42787471e-01 -7.84508467e-01 3.53566587e-01 8.25739622e-01 -7.47185290e-01 -3.03441525e-01 -9.42907929e-01 9.97183442e-01 1.86152101e-01 1.49038732e-01 -1.98369205e-01 -2.34650373e-01 6.25255764e-01 2.45537713e-01 2.75714427e-01 5.82464397e-01 1.83332771e-01 -5.70222080e-01 1.09856166e-01 -9.88127649e-01 9.23593760e-01 -4.31978315e-01 -6.55522108e-01 -8.23482722e-02 -2.10229196e-02 -7.09412932e-01 -4.49703306e-01 -2.72256196e-01 -5.36712825e-01 9.91464972e-01 -7.93098629e-01 -4.97214705e-01 6.35560036e-01 -3.37705165e-01 -4.06390019e-02 2.15965733e-01 1.05230069e+00 -3.02713625e-02 -5.57702243e-01 3.48433942e-01 1.95160676e-02 -2.78145492e-01 4.97792065e-01 -1.11926544e+00 4.62114990e-01 5.78333497e-01 -8.71341825e-01 1.05874062e+00 7.37103939e-01 -1.24855149e+00 -1.16482973e+00 -8.34077835e-01 8.30294549e-01 -1.25013247e-01 3.19999963e-01 2.14413494e-01 -5.44560313e-01 9.82312202e-01 1.40070498e-01 -5.39635003e-01 1.20862317e+00 5.02371013e-01 1.98844448e-01 2.96371818e-01 -1.64390802e+00 4.84085083e-01 1.10053444e+00 -2.07555875e-01 -4.56994802e-01 4.35380667e-01 5.17354369e-01 -2.25597695e-01 -1.31848121e+00 4.60246176e-01 9.49194431e-01 -9.98002231e-01 6.01635814e-01 -6.14604950e-01 -1.32688448e-01 -1.04552887e-01 1.51497692e-01 -8.69526625e-01 -2.94123471e-01 -8.89129043e-01 4.70302671e-01 1.24000978e+00 7.77142704e-01 -7.56424963e-01 8.70798707e-01 7.29158580e-01 -1.51491016e-01 -4.62548584e-01 -8.13571930e-01 -3.73301119e-01 -6.37075230e-02 -1.05256043e-01 6.86688006e-01 8.73253703e-01 6.67260170e-01 -1.73430387e-02 -3.49916458e-01 2.70401333e-02 4.46139246e-01 -2.75652885e-01 1.05506286e-01 -9.69864488e-01 -5.89964509e-01 -1.64082285e-03 -7.38819242e-01 -3.42794120e-01 -6.24980807e-01 -6.72127724e-01 -6.08556807e-01 -1.63424027e+00 7.93911278e-01 -2.09879562e-01 3.05494126e-02 2.74229318e-01 -2.30615467e-01 8.11066478e-02 -4.04329956e-01 -2.97411442e-01 -2.58092046e-01 -3.11071351e-02 8.80435109e-01 7.63854310e-02 -5.54088056e-02 5.60331345e-02 -1.31000459e+00 7.70523012e-01 9.91036713e-01 -2.72990286e-01 -6.14342093e-01 -2.87297279e-01 6.38695359e-01 2.47871965e-01 2.67831624e-01 -3.97808731e-01 -3.32923532e-01 -2.67591417e-01 4.73557949e-01 -8.33143413e-01 2.80134767e-01 -6.47287294e-02 6.81397438e-01 1.00826180e+00 -3.47003460e-01 1.61324814e-01 1.98966101e-01 4.16481674e-01 4.01991814e-01 -6.58196986e-01 3.15264821e-01 -4.99037862e-01 -5.01133263e-01 1.56716228e-01 -1.04837489e+00 3.09051573e-01 6.81074023e-01 -1.62673235e-01 -4.62248266e-01 -4.64332730e-01 -1.29589653e+00 -4.60145213e-02 7.46621966e-01 -7.86651000e-02 3.95050287e-01 -7.30796099e-01 -9.32711720e-01 -3.41581285e-01 2.55000412e-01 -4.67081726e-01 7.53511608e-01 1.42343295e+00 -7.21407890e-01 7.83774197e-01 1.79309770e-01 -3.58267963e-01 -1.31015623e+00 5.68457544e-01 3.11097682e-01 4.42224860e-01 -1.00381732e+00 7.38967896e-01 -8.36095065e-02 -1.26511991e-01 -7.95696974e-02 -6.27018791e-03 1.16654910e-01 -5.61663628e-01 8.57990861e-01 6.83754802e-01 1.66813210e-01 -3.38402271e-01 -7.09038913e-01 2.63823599e-01 -1.70324475e-01 -1.82566851e-01 1.03381598e+00 -3.91094863e-01 -4.23292935e-01 3.81429106e-01 6.46096528e-01 4.16440606e-01 -5.74069262e-01 4.73089486e-01 -1.17530130e-01 -2.60125041e-01 -2.72068709e-01 -9.67215180e-01 -5.93479335e-01 1.93850562e-01 6.68963611e-01 -1.44929066e-01 9.17367458e-01 2.37314746e-01 2.96463251e-01 -7.58868903e-02 3.30533832e-01 -9.40964043e-01 -9.88176823e-01 -3.50171804e-01 6.49880290e-01 -3.69663894e-01 3.17250222e-01 -7.52582133e-01 -2.53348827e-01 6.21910095e-01 1.10819660e-01 4.04838532e-01 1.16199002e-01 2.51765221e-01 7.58144706e-02 -1.45428166e-01 -1.18400693e+00 4.38582823e-02 -2.60417730e-01 1.28262246e+00 8.96430016e-01 4.11138460e-02 -1.53678834e+00 5.33756137e-01 -1.38061019e-02 1.08504891e-01 5.72009563e-01 9.63328600e-01 -2.44245917e-01 -1.67359781e+00 -3.85325521e-01 1.13426709e+00 -1.13917994e+00 -1.23249389e-01 -7.62372494e-01 6.13581061e-01 3.59330475e-01 1.05883527e+00 -5.03247045e-02 2.07320243e-01 1.21135287e-01 2.71786749e-01 5.99991381e-01 -4.70634550e-01 -1.30147293e-01 5.20164192e-01 1.05657005e+00 -4.00631875e-01 -4.65782017e-01 -1.26442242e+00 -1.15556085e+00 3.21946256e-02 -2.81789720e-01 3.11256468e-01 3.10593277e-01 1.02727151e+00 3.59251291e-01 6.77209914e-01 -2.84820646e-01 3.24800074e-01 -5.33158779e-01 -8.24774325e-01 -1.15608215e+00 -2.99884528e-01 1.65721819e-01 -3.65612984e-01 1.50754191e-02 -2.68805861e-01]

[7.952182769775391, 5.639594554901123]

2a4a2cb9-0ecf-4ffe-a143-2bcfa0531cfe

safe-imitation-learning-of-nonlinear-model

2212.02941

https://arxiv.org/abs/2212.02941v1

https://arxiv.org/pdf/2212.02941v1.pdf

Safe Imitation Learning of Nonlinear Model Predictive Control for Flexible Robots

Flexible robots may overcome the industry's major problems: safe human-robot collaboration and increased load-to-mass ratio. However, oscillations and high dimensional state space complicate the control of flexible robots. This work investigates nonlinear model predictive control (NMPC) of flexible robots -- for simultaneous planning and control -- modeled via the rigid finite element method. Although NMPC performs well in simulation, computational complexity prevents its deployment in practice. We show that imitation learning of NMPC with neural networks as function approximator can massively improve the computation time of the controller at the cost of slight performance loss and, more critically, loss of safety guarantees. We leverage a safety filter formulated as a simpler NMPC to recover safety guarantees. Experiments on a simulated three degrees of freedom flexible robot manipulator demonstrate that the average computational time of the proposed safe approximate NMPC controller is 3.6 ms while of the original NMPC is 11.8 ms. Fast and safe approximate NMPC might facilitate the industry's adoption of flexible robots and new solutions for similar problems, e.g., deformable object manipulation and soft robot control.

['Jan Swevers', 'Moritz Diehl', 'Rudolf Reiter', 'Shamil Mamedov']

2022-12-06

['deformable-object-manipulation']

['robots']

[ 9.15084183e-02 7.17743397e-01 -2.44200855e-01 3.42701524e-01 -2.41655529e-01 -6.39861763e-01 2.88793683e-01 -4.07595247e-01 -2.78521955e-01 7.94983566e-01 -5.51217794e-01 -2.67795116e-01 -5.17145634e-01 -3.61548543e-01 -9.83896613e-01 -1.02274120e+00 -1.10617213e-01 6.39252603e-01 5.09383567e-02 -3.16179097e-01 4.18581180e-02 6.87987030e-01 -7.76149690e-01 -5.34635842e-01 9.55960274e-01 7.65016794e-01 4.43883806e-01 2.06706479e-01 9.65424478e-01 4.04598296e-01 -4.97369543e-02 1.92046449e-01 6.89924777e-01 4.33741689e-01 -6.83012366e-01 2.36875683e-01 -4.06698018e-01 -4.93380636e-01 -5.22271514e-01 1.04025257e+00 3.71217728e-01 8.50097716e-01 4.97104168e-01 -1.56667256e+00 -5.57719946e-01 6.07489526e-01 -3.75540584e-01 -7.78215289e-01 -5.91078363e-02 4.08592552e-01 2.36193433e-01 -6.68324232e-01 7.49183953e-01 1.50444233e+00 7.18329728e-01 7.02727735e-01 -1.08720362e+00 -4.27672803e-01 3.51476558e-02 -5.82199037e-01 -1.42216110e+00 -1.10558622e-01 3.83660257e-01 -4.60982412e-01 9.97346342e-01 1.37507081e-01 2.96541929e-01 1.01872802e+00 1.04667938e+00 4.06111717e-01 5.13593376e-01 6.35953993e-02 1.14507163e-02 -2.67828643e-01 -4.04404290e-02 8.03780854e-01 5.53539872e-01 3.76610309e-01 6.48123205e-01 -3.10294658e-01 1.31410515e+00 3.87629360e-01 -3.30048829e-01 -5.02652347e-01 -1.25154448e+00 7.10761547e-01 4.55551177e-01 -3.21222574e-01 -4.27911401e-01 4.02488083e-01 4.26389068e-01 3.92599225e-01 -1.54146835e-01 7.40202129e-01 -5.27017772e-01 2.13265911e-01 -2.12061659e-01 5.97651958e-01 1.12690282e+00 1.59665859e+00 -1.68287560e-01 4.82803613e-01 3.28422248e-01 5.43311179e-01 5.58671474e-01 6.35582447e-01 -7.60505944e-02 -1.55645990e+00 4.04484540e-01 2.68800557e-01 7.14924932e-01 -8.94376934e-01 -7.14292705e-01 -3.24661545e-02 -9.56828892e-01 6.43954098e-01 1.92511544e-01 -4.53927815e-01 -6.41361594e-01 1.42559063e+00 3.20323020e-01 -5.71318388e-01 1.26403524e-02 1.14115036e+00 -4.42841381e-01 8.61187935e-01 -2.98562199e-01 -5.39521456e-01 7.20161796e-01 -1.04091930e+00 -5.92913091e-01 9.54930633e-02 2.09070191e-01 -5.55175364e-01 7.30254292e-01 5.95320106e-01 -1.27220535e+00 -4.38544452e-01 -9.53248560e-01 2.13284895e-01 4.52816874e-01 2.25225419e-01 1.91544637e-01 -1.66808844e-01 -5.07652879e-01 1.24124277e+00 -1.35403049e+00 -2.30974898e-01 -2.69714534e-01 1.00160980e+00 -5.48582859e-02 4.57526058e-01 -8.34706962e-01 1.33280373e+00 5.65711021e-01 3.43249559e-01 -8.03689301e-01 -5.95667779e-01 -5.56551099e-01 -1.99348897e-01 9.34313536e-01 -6.73939884e-01 1.35337162e+00 -2.67652869e-01 -2.23167229e+00 -1.44172832e-02 6.43236339e-01 -3.82776737e-01 7.51267076e-01 -7.07309544e-01 -7.44506717e-02 6.33160621e-02 -1.39806464e-01 4.61859375e-01 1.01943696e+00 -1.17760587e+00 -1.76922739e-01 1.37067884e-01 -8.30441788e-02 3.78519185e-02 1.09374210e-01 -3.00227195e-01 -8.97294581e-02 -6.43842876e-01 -4.30075116e-02 -2.07095885e+00 -7.44151235e-01 4.60071206e-01 -5.01886427e-01 -3.89347762e-01 9.28063452e-01 -3.92708123e-01 6.15111768e-01 -1.90740466e+00 5.87025940e-01 4.82073009e-01 -2.41413504e-01 3.81534666e-01 1.18034683e-01 6.01190090e-01 1.79284319e-01 -2.91232765e-01 9.94273368e-03 4.07806158e-01 1.28712475e-01 6.11599684e-01 -5.88659883e-01 8.23747814e-01 1.92839235e-01 6.45617664e-01 -7.04852402e-01 4.22683050e-04 2.60976553e-01 2.93118775e-01 -5.40294826e-01 8.06249455e-02 -3.34738970e-01 5.01721501e-01 -8.19303513e-01 5.93580127e-01 5.47213018e-01 4.32328805e-02 2.37897858e-01 -2.66173244e-01 -4.24444973e-01 -3.80387098e-01 -1.15822625e+00 1.22018421e+00 -2.69091040e-01 3.60316262e-02 9.26560342e-01 -8.47453833e-01 8.41265142e-01 3.95293444e-01 4.87187177e-01 -1.68405712e-01 5.32446563e-01 5.51688135e-01 8.19189101e-02 -5.21042347e-01 5.06886065e-01 -2.22321868e-01 -1.19432859e-01 -6.89272396e-03 -2.71450222e-01 -6.63661599e-01 -1.84567899e-01 -3.70786279e-01 9.14152563e-01 4.85271662e-01 1.80624902e-01 -7.11342812e-01 5.00308156e-01 1.81948155e-01 6.41011775e-01 2.77017117e-01 -8.92771110e-02 1.24138724e-02 -9.26801935e-03 -1.33805543e-01 -1.30068374e+00 -1.06556952e+00 8.26340392e-02 9.40564156e-01 5.51536381e-01 2.30000257e-01 -4.68914598e-01 -5.95359458e-03 5.60950875e-01 7.22307622e-01 5.01870513e-02 -4.16721910e-01 -1.13031733e+00 2.37434448e-04 2.98306406e-01 7.87164390e-01 -3.18556130e-01 -6.96849585e-01 -9.06701565e-01 5.93930781e-01 3.09959918e-01 -9.99938428e-01 -5.97649992e-01 2.39629894e-01 -9.63914156e-01 -1.04406369e+00 -5.99750340e-01 -9.28649485e-01 9.29528594e-01 2.83676833e-01 1.96631238e-01 -8.64434466e-02 -2.28085518e-01 4.15548682e-01 9.83658154e-03 -3.18305433e-01 -6.75634563e-01 -2.90772766e-01 9.32049334e-01 -7.78262496e-01 -9.58573520e-01 -3.53692144e-01 -3.36196184e-01 1.00598288e+00 -5.24894893e-01 -1.30539209e-01 7.38371551e-01 9.10149932e-01 7.36865997e-01 2.04661325e-01 4.70365256e-01 -2.60317951e-01 6.30341113e-01 -1.91396296e-01 -1.16237795e+00 2.70233750e-02 -6.22731507e-01 1.70523629e-01 1.08480597e+00 -1.02252734e+00 -1.09813654e+00 3.59538049e-01 1.65474862e-01 -8.89692426e-01 4.68681961e-01 3.54457974e-01 2.88577974e-01 -4.73445714e-01 3.59464437e-01 -4.94678691e-02 7.67696679e-01 -3.88352305e-01 3.86430711e-01 4.81633872e-01 7.45359719e-01 -8.64652514e-01 1.05627549e+00 1.98502794e-01 5.68531215e-01 -5.87015450e-01 -1.06312536e-01 -1.93478584e-01 -2.82778889e-01 -4.29856628e-02 3.15519363e-01 -8.42478693e-01 -1.34954429e+00 1.36439830e-01 -1.10791755e+00 -4.96967256e-01 -2.03250617e-01 6.43653750e-01 -1.14142823e+00 3.97792518e-01 -9.98540521e-01 -9.90969896e-01 -6.72993720e-01 -1.19084001e+00 8.33853662e-01 -2.68230867e-02 -4.60114390e-01 -5.26708782e-01 -2.01044232e-01 -7.26259425e-02 5.06509900e-01 6.12725914e-01 6.19237542e-01 -1.63699299e-01 -3.52233261e-01 -4.24206465e-01 3.18662405e-01 2.57476360e-01 -4.99074832e-02 1.31175026e-01 -1.69831917e-01 -1.06528127e+00 3.67459953e-01 -2.97835529e-01 8.94281864e-02 4.49967444e-01 7.21037209e-01 -9.68702555e-01 -7.30486274e-01 5.72527982e-02 1.38700628e+00 5.89689255e-01 1.23520486e-01 2.55283028e-01 6.19647026e-01 5.91199160e-01 1.13883924e+00 5.56709647e-01 -2.65642524e-01 6.25614405e-01 8.50888848e-01 6.69518948e-01 4.94522035e-01 1.41208366e-01 7.67098725e-01 7.75377452e-01 -6.07026160e-01 -8.15165788e-02 -6.42118454e-01 2.64914334e-01 -2.10124278e+00 -5.00466704e-01 -2.45357648e-01 2.10609031e+00 7.14000642e-01 1.44032881e-01 -5.08750491e-02 -6.54065609e-02 9.32359576e-01 -6.89948678e-01 -7.17450440e-01 -5.10769427e-01 5.72567582e-01 -4.40269440e-01 8.54548991e-01 4.64518666e-01 -9.72127080e-01 4.48940128e-01 5.01148987e+00 8.02370846e-01 -1.13311481e+00 -2.77897865e-01 -3.14637989e-01 -2.23811790e-01 2.48617709e-01 -2.80715764e-01 -6.02180898e-01 4.87565994e-01 9.64332402e-01 -4.84335065e-01 7.57991850e-01 1.46609032e+00 4.25980747e-01 2.12493867e-01 -1.09437072e+00 5.30280471e-01 -5.56157172e-01 -1.02731931e+00 -2.03364983e-01 1.23909183e-01 7.57395208e-01 3.62204909e-02 3.50480117e-02 3.47606570e-01 2.37541035e-01 -6.72840357e-01 1.20476997e+00 4.90533233e-01 4.83358979e-01 -9.31440890e-01 5.67585945e-01 7.70208001e-01 -1.12025428e+00 -5.53336501e-01 -6.57601893e-01 -1.70140728e-01 7.73332477e-01 1.34247243e-01 -6.75603926e-01 5.53288758e-01 1.14736117e-01 2.68172413e-01 4.80547071e-01 8.05518270e-01 1.38904393e-01 -6.97001517e-02 -6.56129420e-01 -2.43975714e-01 3.35789144e-01 -4.20466185e-01 1.18440926e+00 7.42227793e-01 5.06923616e-01 2.14363635e-01 7.44237363e-01 8.89674246e-01 2.96205282e-01 -4.43048716e-01 -6.66142941e-01 -2.09522739e-01 3.64595979e-01 1.16449916e+00 -4.69062209e-01 1.52677789e-01 1.32266909e-01 8.17498088e-01 -8.70184228e-02 -3.60127911e-02 -1.27017713e+00 -5.67181110e-01 4.47014302e-01 3.40868086e-02 1.14809364e-01 -9.06657875e-01 2.76784436e-03 -6.18279397e-01 8.15456547e-03 -5.41706026e-01 -2.27155581e-01 -6.59974635e-01 -1.24753833e+00 2.22098485e-01 8.37059468e-02 -1.83250129e+00 -6.65781975e-01 -9.13256824e-01 -3.59581441e-01 4.78790760e-01 -8.92267287e-01 -9.62958038e-01 2.96671629e-01 4.27681327e-01 7.68755853e-01 1.56935304e-01 7.18016624e-01 -1.22866839e-01 -5.54012001e-01 1.64180368e-01 5.56515098e-01 -3.02540779e-01 7.09663987e-01 -7.57458329e-01 -4.40800190e-02 5.27738869e-01 -1.14472270e+00 9.63330925e-01 1.06172037e+00 -8.54601502e-01 -2.26775432e+00 -1.49644351e+00 7.52050281e-02 -2.40555732e-03 1.02978230e+00 3.30293685e-01 -8.69892955e-01 7.51061440e-01 -2.69251261e-02 2.81232260e-02 -2.68552363e-01 -7.72936165e-01 4.78570670e-01 2.52758265e-01 -1.30416691e+00 9.15171206e-01 8.04768741e-01 4.30858321e-02 -6.41235709e-01 3.98799181e-01 1.16500413e+00 -8.28620493e-01 -1.42930627e+00 5.42341232e-01 7.80289590e-01 2.59040058e-01 1.06019127e+00 -3.52687865e-01 4.35978360e-02 -4.48582441e-01 1.78032350e-02 -1.33545566e+00 -6.65097773e-01 -1.38384342e+00 -5.70707619e-01 7.81040788e-01 1.03918448e-01 -6.22363567e-01 3.07218373e-01 9.56891894e-01 -6.84279144e-01 -9.44499671e-01 -1.04330099e+00 -1.64667416e+00 1.75795361e-01 7.10594431e-02 -3.94858807e-01 7.29582012e-01 4.13223594e-01 1.91886827e-01 -7.27258205e-01 6.29099488e-01 5.67659199e-01 7.44711086e-02 5.19308746e-01 -1.16175616e+00 -3.80672336e-01 -6.93810657e-02 1.44434255e-02 -7.20424235e-01 3.29488754e-01 -7.04356194e-01 6.16376698e-01 -1.23094034e+00 -1.75708279e-01 -3.62611890e-01 1.34005457e-01 5.66348672e-01 3.63656223e-01 -4.80617732e-01 3.76782566e-01 3.80579531e-01 -3.67577016e-01 6.79297090e-01 1.66475201e+00 -2.27672800e-01 -5.33271968e-01 4.21669513e-01 -1.53031768e-02 9.32548642e-01 9.43369925e-01 -2.86800712e-01 -5.32861650e-01 -4.03451115e-01 -1.32662132e-01 5.65590084e-01 2.63407737e-01 -1.04549217e+00 3.12544882e-01 -7.80253530e-01 -1.30227759e-01 -3.42778474e-01 2.21714333e-01 -1.35312176e+00 7.30250180e-01 1.41252673e+00 -1.17360614e-01 3.26600112e-02 2.46448129e-01 9.55896735e-01 2.87335247e-01 -2.50743538e-01 1.18691146e+00 1.14247575e-01 -3.50775838e-01 1.55670032e-01 -7.11485445e-01 -5.98837197e-01 1.22101533e+00 -1.26242861e-02 -2.04469919e-01 8.00378546e-02 -8.15435350e-01 5.27529657e-01 5.68662882e-01 5.12901008e-01 4.76990491e-01 -1.07100666e+00 -1.46670237e-01 -2.00235292e-01 -6.39909923e-01 6.13089353e-02 -7.33488146e-03 1.18738282e+00 -4.97073680e-01 6.67741954e-01 -5.98205447e-01 -4.29730058e-01 -1.07104540e+00 8.49574506e-01 1.48672417e-01 -2.05473714e-02 -8.70739818e-01 5.25116742e-01 -5.08366637e-02 -5.50609052e-01 6.67296797e-02 -8.52704704e-01 4.26486880e-01 -6.14589214e-01 -1.59371257e-01 1.08535075e+00 -3.34651619e-01 -2.75218278e-01 -3.33761185e-01 6.86848223e-01 8.57858062e-02 1.20812014e-01 1.57119143e+00 -8.69785324e-02 -1.44445911e-01 4.04335000e-02 8.80028903e-01 -3.63536894e-01 -1.77291024e+00 2.66045213e-01 -1.39746130e-01 -8.16577002e-02 -4.25003409e-01 -5.08461714e-01 -7.17220008e-01 2.19232917e-01 4.10156578e-01 -1.89720059e-03 5.11595786e-01 -4.15412664e-01 9.51774299e-01 1.22033775e+00 9.49196994e-01 -1.46393847e+00 -2.58156117e-02 9.70408022e-01 1.46952379e+00 -6.90975785e-01 2.23812416e-01 -8.22130620e-01 -6.00905836e-01 1.47598755e+00 6.10035777e-01 -9.23633873e-01 4.40558016e-01 5.88778436e-01 -5.12128174e-01 3.71009648e-01 -5.96938550e-01 6.76990688e-01 2.14069381e-01 3.03962976e-01 -1.42839074e-01 1.65496334e-01 -4.21675563e-01 8.26792300e-01 3.88866544e-01 2.61764497e-01 5.92928827e-01 1.42004859e+00 -7.34707713e-01 -8.30640912e-01 -5.11803269e-01 8.40390399e-02 -2.43461639e-01 7.00091541e-01 1.96900070e-01 1.17163754e+00 -2.61502564e-01 7.80965149e-01 -1.19395673e-01 -3.34496379e-01 6.99656963e-01 -2.98857301e-01 6.28923655e-01 -3.52552652e-01 -2.88356751e-01 3.83268863e-01 1.57940194e-01 -8.40494275e-01 -8.50511640e-02 -3.04657727e-01 -1.92264402e+00 -3.30912590e-01 -7.38910973e-01 -6.03391081e-02 4.50636208e-01 6.69411898e-01 4.35166746e-01 3.03822547e-01 5.80507278e-01 -1.62424970e+00 -1.73906016e+00 -6.73459470e-01 -6.31694019e-01 8.41463916e-03 2.50598013e-01 -1.11679769e+00 -6.51862398e-02 1.62992850e-02]

[4.882218837738037, 1.5986902713775635]

78f324f9-e8dd-41d6-8610-e2913393f69b

kernel-functional-optimisation

http://proceedings.neurips.cc/paper/2021/hash/251e16a2aac0ca4847adf561483381bf-Abstract.html

http://proceedings.neurips.cc/paper/2021/file/251e16a2aac0ca4847adf561483381bf-Paper.pdf

Kernel Functional Optimisation

Traditional methods for kernel selection rely on parametric kernel functions or a combination thereof and although the kernel hyperparameters are tuned, these methods often provide sub-optimal results due to the limitations induced by the parametric forms. In this paper, we propose a novel formulation for kernel selection using efficient Bayesian optimisation to find the best fitting non-parametric kernel. The kernel is expressed using a linear combination of functions sampled from a prior Gaussian Process (GP) defined by a hyperkernel. We also provide a mechanism to ensure the positive definiteness of the Gram matrix constructed using the resultant kernels. Our experimental results on GP regression and Support Vector Machine (SVM) classification tasks involving both synthetic functions and several real-world datasets show the superiority of our approach over the state-of-the-art.

['Svetha Venkatesh', 'Sunil Gupta', 'Santu Rana', 'Alistair Shilton', 'Arun Kumar Anjanapura Venkatesh']

2021-12-01

https://openreview.net/forum?id=zDtFO9vohmF

https://openreview.net/pdf?id=zDtFO9vohmF

neurips-2021-12

['bayesian-optimisation']

['methodology']

[-3.75959836e-02 -4.92487907e-01 -3.06106955e-01 -3.06642711e-01 -6.16772950e-01 -4.75205868e-01 4.40154940e-01 -7.13735968e-02 -4.51642334e-01 9.11185801e-01 -3.81515265e-01 -2.48103738e-01 -5.38144052e-01 -6.62809134e-01 -4.42368090e-01 -1.01560879e+00 -1.55061379e-01 3.95309329e-01 5.41845977e-01 1.75456494e-01 4.87810522e-01 6.50497794e-01 -1.23247552e+00 -2.10881829e-01 1.38520789e+00 9.15529311e-01 -1.12581290e-01 5.86556137e-01 2.18081191e-01 2.83045590e-01 -3.70447695e-01 -3.27865392e-01 2.89515734e-01 -1.42962961e-02 -3.29111606e-01 1.38618564e-02 -5.01093939e-02 -5.57214767e-02 -2.27043569e-01 1.03702271e+00 3.01631719e-01 4.21397597e-01 1.37645590e+00 -1.31972098e+00 -6.73539042e-01 1.32605851e-01 -4.07713681e-01 -4.94553186e-02 4.37398665e-02 -8.97497833e-02 6.48391247e-01 -8.65706325e-01 -4.53619622e-02 9.23127770e-01 7.79824376e-01 8.51797014e-02 -1.81729412e+00 -5.67446172e-01 -3.96385998e-01 7.45226145e-02 -1.82074594e+00 -3.31992418e-01 5.51850021e-01 -6.02586389e-01 8.91897380e-01 2.59746760e-01 4.57617342e-01 7.22645760e-01 3.70406032e-01 4.82581258e-01 1.28701043e+00 -5.72900832e-01 5.49067140e-01 6.30490899e-01 4.87611085e-01 8.02175760e-01 3.65484864e-01 1.00059584e-01 -2.54915804e-01 -1.30396807e+00 1.15301669e+00 -3.33984762e-01 -5.32984734e-01 -9.72962797e-01 -9.91214216e-01 1.24462259e+00 -3.39133084e-01 -1.14375085e-01 -3.82432878e-01 5.93783371e-02 4.42375898e-01 2.03464568e-01 4.39872682e-01 4.92608815e-01 -5.17059207e-01 -7.24438727e-02 -8.48323166e-01 1.55927449e-01 1.28996360e+00 6.43387377e-01 7.96151042e-01 1.47410622e-02 -3.08492184e-01 1.08542240e+00 5.46074152e-01 4.89462584e-01 2.66464293e-01 -3.86688381e-01 4.30500880e-03 3.69372666e-01 5.39508998e-01 -6.55978560e-01 -6.61199242e-02 -9.74991545e-02 -5.37745297e-01 4.02122706e-01 6.55958831e-01 -1.08909257e-01 -7.84682393e-01 1.25070441e+00 4.19366330e-01 3.95184964e-01 9.93142202e-02 5.65833807e-01 2.11264953e-01 7.69801080e-01 9.29687619e-02 -5.22069335e-02 1.08514619e+00 -7.72386014e-01 -4.82811600e-01 3.08372438e-01 4.89467651e-01 -8.82115364e-01 1.01777720e+00 5.99549055e-01 -4.37774956e-01 -3.27121466e-01 -1.11150455e+00 6.94809735e-01 -1.68088123e-01 6.64481223e-01 8.82917047e-01 1.12877786e+00 -8.79520118e-01 5.24205208e-01 -7.60539353e-01 -1.23753518e-01 1.57991797e-02 6.84021473e-01 -4.91620004e-01 3.63667130e-01 -1.10137236e+00 1.05024552e+00 7.80066609e-01 -3.14490311e-02 -3.80022079e-01 -6.60676479e-01 -7.94463933e-01 -4.01981249e-02 2.15439230e-01 -5.68045735e-01 9.08553421e-01 -6.39996648e-01 -2.17562461e+00 4.93776262e-01 2.52131760e-01 -3.78243089e-01 3.64473462e-01 -3.47827584e-01 -2.30186075e-01 1.71545193e-01 -4.10473913e-01 -5.38165718e-02 1.46094441e+00 -8.90923858e-01 -4.13358569e-01 1.03002071e-01 -5.00852466e-01 -1.00081615e-01 -3.20877850e-01 3.39516282e-01 -3.43002498e-01 -5.80934465e-01 -2.89647579e-01 -1.22665644e+00 -1.62642509e-01 -3.60649318e-01 -5.73617816e-01 -3.98762017e-01 7.67892241e-01 -6.65217161e-01 1.52735674e+00 -2.22086787e+00 9.42351893e-02 7.62234747e-01 -3.08636814e-01 3.19561332e-01 2.41092265e-01 4.74136442e-01 -2.63492733e-01 -1.61921993e-01 -2.86333025e-01 3.21775787e-02 1.86447829e-01 3.73353697e-02 -5.77190042e-01 9.54103291e-01 6.57140166e-02 3.84542346e-01 -6.28694892e-01 -3.37310255e-01 3.82805079e-01 6.62710190e-01 5.44373840e-02 3.01477313e-01 1.95436776e-01 -9.46493745e-02 -6.46777689e-01 5.12545109e-01 6.06847942e-01 -1.37092605e-01 -2.30523497e-01 -2.09365353e-01 1.97455168e-01 -1.98869646e-01 -1.58044195e+00 9.45746541e-01 -5.48268147e-02 2.32504919e-01 -1.44705772e-01 -9.48981822e-01 1.00395620e+00 2.85570979e-01 2.23614782e-01 4.06183362e-01 5.47557399e-02 2.69759655e-01 -9.04615223e-02 -2.71952778e-01 1.27187237e-01 -1.13490187e-02 6.61627948e-02 2.46967882e-01 1.65694758e-01 -1.85021594e-01 8.35578367e-02 -2.49283075e-01 1.00013661e+00 2.03493714e-01 8.25177431e-01 -5.98437667e-01 7.45733202e-01 -1.26166660e-02 2.83441663e-01 7.95675695e-01 -1.64250001e-01 2.86456466e-01 6.02063775e-01 -2.09738418e-01 -8.93286049e-01 -1.21872294e+00 -6.70757294e-01 8.87171626e-01 -2.73751736e-01 -2.29740798e-01 -6.75221264e-01 -7.32160747e-01 2.90340096e-01 1.00028169e+00 -6.31268919e-01 -3.45451355e-01 -2.71371454e-01 -1.28807271e+00 7.53220975e-01 3.38107735e-01 2.28511587e-01 -6.92064822e-01 -4.35255438e-01 2.12680250e-01 5.18994093e-01 -1.00640666e+00 -4.94912326e-01 4.14934963e-01 -8.77641797e-01 -9.72257793e-01 -6.06993973e-01 -5.37518919e-01 6.89308405e-01 -8.48512352e-02 5.25228679e-01 -4.93588388e-01 -2.07637474e-01 3.56817901e-01 -2.56799489e-01 -1.84533849e-01 -3.88404518e-01 1.96788814e-02 1.51523232e-01 2.76442766e-01 2.23225996e-01 -3.05788487e-01 -3.04449350e-01 4.87332374e-01 -8.81405771e-01 -3.26784134e-01 6.52565420e-01 1.11253905e+00 4.91594434e-01 5.56440771e-01 5.10325015e-01 -6.96663618e-01 1.00073087e+00 -3.64166498e-01 -1.12479162e+00 4.87813175e-01 -6.67776167e-01 4.38960165e-01 5.20030677e-01 -1.02620351e+00 -1.03958285e+00 3.74687284e-01 5.98041415e-01 -5.47340572e-01 -1.97403684e-01 4.00028259e-01 -4.53491658e-02 -6.27645671e-01 8.78561735e-01 3.98689508e-01 7.59679750e-02 -2.56822377e-01 3.60208541e-01 5.67010939e-01 2.21425101e-01 -9.73096013e-01 8.81951571e-01 3.29902500e-01 1.01702832e-01 -9.16875660e-01 -3.30249041e-01 -6.22578919e-01 -6.67078972e-01 1.89479098e-01 6.38976753e-01 -4.28852290e-01 -5.59030116e-01 5.78956366e-01 -7.81763256e-01 -3.57774079e-01 -2.18249578e-02 8.99216354e-01 -8.13405395e-01 5.39658964e-01 -5.77884793e-01 -1.14649904e+00 -4.58213121e-01 -1.18785167e+00 8.62026453e-01 2.52547473e-01 -2.90884703e-01 -1.17679417e+00 3.27301234e-01 -6.57209158e-02 3.58097523e-01 3.55900973e-02 1.26763785e+00 -1.06640685e+00 -2.14925140e-01 -6.84537768e-01 -2.84679145e-01 6.81243002e-01 2.03801543e-01 4.86341685e-01 -7.27148890e-01 -3.17199975e-01 -8.56531113e-02 -1.51847869e-01 7.46869683e-01 5.70091188e-01 8.67233157e-01 -1.40017852e-01 -5.32382667e-01 6.85838819e-01 1.44651902e+00 9.65798423e-02 6.66907072e-01 5.37126899e-01 4.93821412e-01 4.62473959e-01 7.24553585e-01 5.63490927e-01 -1.67855456e-01 5.94932318e-01 -1.05308942e-01 1.93108961e-01 7.02411592e-01 1.48264736e-01 4.81287092e-01 2.49091864e-01 3.55491065e-03 2.14665085e-01 -9.13296044e-01 2.87236512e-01 -2.19646454e+00 -7.14626312e-01 -1.85179427e-01 2.94200420e+00 1.16076183e+00 -5.29318396e-03 2.64320970e-01 -1.17223330e-01 6.96348131e-01 -4.30506289e-01 -3.39989811e-01 -3.17517996e-01 5.74623384e-02 5.10118365e-01 9.12692606e-01 4.23468411e-01 -1.54598045e+00 9.61423635e-01 7.20111895e+00 1.41251326e+00 -1.07383311e+00 -2.49158099e-01 2.07984224e-01 3.62049758e-01 1.74482137e-01 3.01690370e-01 -1.09040475e+00 4.36453074e-01 8.17377925e-01 -1.34277746e-01 2.58840173e-01 1.07353032e+00 -6.95049018e-02 -5.98823130e-01 -8.81353676e-01 1.08966780e+00 -1.47480518e-01 -7.40742683e-01 -2.40914747e-01 1.64462075e-01 5.93309760e-01 -3.33721578e-01 1.20530024e-01 3.95378828e-01 3.78583223e-01 -1.20042431e+00 3.00597221e-01 7.42080569e-01 5.04725337e-01 -9.97024059e-01 6.56040132e-01 2.52704293e-01 -9.14623976e-01 1.65476561e-01 -6.50967300e-01 4.21583295e-01 -1.74152911e-01 9.61922348e-01 -1.24953389e+00 3.64978045e-01 3.31952393e-01 4.05996181e-02 -5.43390691e-01 1.58520114e+00 -2.31000692e-01 1.03874075e+00 -6.04505301e-01 -1.85611814e-01 -7.27003515e-02 -7.44175494e-01 6.47241592e-01 1.41410232e+00 3.30097139e-01 -1.86424300e-01 2.60798067e-01 7.67674744e-01 6.86439395e-01 5.44727087e-01 -4.96970206e-01 -1.47551373e-01 4.14487511e-01 1.42839193e+00 -8.41434181e-01 -2.29060784e-01 -3.82125884e-01 1.09731758e+00 4.16513324e-01 7.22898722e-01 -9.89941359e-01 -7.00602174e-01 8.28296363e-01 -8.60837176e-02 4.83176261e-01 -3.96608472e-01 -9.80510190e-02 -1.14614403e+00 -9.84058157e-02 -6.21344388e-01 3.05759490e-01 -3.60669613e-01 -1.78398955e+00 3.68557841e-01 5.27117133e-01 -8.98634970e-01 -2.60233968e-01 -9.36430454e-01 -5.69823623e-01 1.18711627e+00 -1.07548237e+00 -9.93318200e-01 2.57759571e-01 5.24991095e-01 -1.50202692e-01 -3.22557181e-01 1.14046431e+00 -1.72889367e-01 -5.08374572e-01 5.11901498e-01 5.27343094e-01 -1.21453889e-01 1.07323349e+00 -1.41690981e+00 3.71902920e-02 4.12487537e-01 -2.54710317e-01 8.37785244e-01 7.49316931e-01 -8.53753626e-01 -1.28957045e+00 -7.65723586e-01 2.87332863e-01 -3.61010909e-01 9.85890508e-01 -2.00159863e-01 -1.13280797e+00 3.54366034e-01 -1.62978366e-01 3.58365960e-02 1.15462160e+00 1.28646627e-01 -4.00923312e-01 1.54732719e-01 -1.24459934e+00 5.32960951e-01 3.14978212e-02 -4.10386592e-01 -4.85538036e-01 1.97302923e-01 1.88923120e-01 -1.14237718e-01 -1.09088135e+00 6.61509931e-01 6.17767036e-01 -6.82041526e-01 1.12178826e+00 -5.86431503e-01 -4.98277217e-01 -6.72292352e-01 -1.21027730e-01 -1.48778009e+00 -5.01408279e-01 -6.84713125e-01 -3.23106855e-01 1.06931400e+00 2.93407112e-01 -1.05841577e+00 6.18580937e-01 7.62116849e-01 4.11890984e-01 -1.05057418e+00 -8.84926736e-01 -9.97183323e-01 -2.08906502e-01 -2.83070803e-01 2.25854471e-01 9.79011476e-01 -5.09420745e-02 2.04851866e-01 -4.29573148e-01 3.07751358e-01 9.11055863e-01 -1.76100999e-01 8.36067319e-01 -1.26936507e+00 -6.19999468e-01 -4.08531904e-01 -6.15936697e-01 -5.04683733e-01 4.79664296e-01 -5.16681433e-01 -2.82418225e-02 -9.35207963e-01 1.94595456e-01 -6.32896841e-01 -2.19452053e-01 4.13111299e-01 -3.97666454e-01 -1.93578780e-01 -3.62052321e-01 1.65549174e-01 -1.83077291e-01 5.73345006e-01 7.29573071e-01 3.31784904e-01 -7.17063725e-01 3.76074374e-01 -2.20847845e-01 7.32248008e-01 9.30880964e-01 -4.95752722e-01 -5.47812998e-01 5.85678101e-01 5.58803640e-02 -1.34030506e-01 3.84878039e-01 -8.58206391e-01 1.26688018e-01 -3.45231682e-01 6.03935003e-01 -4.25898254e-01 4.46653634e-01 -6.59142554e-01 2.04474047e-01 9.12971199e-02 5.98719269e-02 -2.54719496e-01 2.43482947e-01 7.62691915e-01 6.40202090e-02 -6.81271672e-01 1.02343178e+00 4.53921676e-01 -4.36563671e-01 9.03417766e-02 -5.68657577e-01 -3.41870487e-01 1.24542844e+00 -3.91057044e-01 -1.58085618e-02 -1.13663390e-01 -5.90621471e-01 -1.36596456e-01 6.21187150e-01 -3.26811559e-02 6.34621322e-01 -1.23274708e+00 -5.86115062e-01 2.59430379e-01 1.76616147e-01 -5.31313002e-01 -2.85288483e-01 1.04419398e+00 -5.42772949e-01 4.89266574e-01 -2.25187838e-02 -4.49433684e-01 -1.27609062e+00 3.96129578e-01 2.89319783e-01 -3.95475745e-01 -3.62231761e-01 7.36318111e-01 2.32872248e-01 -4.70852852e-01 1.41436040e-01 -3.27965885e-01 -3.60355005e-02 -2.45263502e-01 4.75376874e-01 5.86501896e-01 -1.13778681e-01 -5.04432797e-01 -1.98128685e-01 4.25317228e-01 2.31720451e-02 -2.64350057e-01 8.80774438e-01 4.35498148e-01 -3.78041506e-01 4.14348423e-01 1.06111908e+00 6.62608668e-02 -1.18511569e+00 -1.94748148e-01 8.34156424e-02 -5.61907470e-01 -5.54782245e-03 -4.65057999e-01 -2.42853880e-01 2.15774268e-01 4.92391795e-01 8.97904783e-02 8.10449660e-01 -2.12038472e-01 3.26162875e-01 5.79221785e-01 4.35838044e-01 -1.12100494e+00 -3.21498454e-01 4.33860451e-01 6.31233096e-01 -9.78568137e-01 1.74394310e-01 -7.92196453e-01 -7.32015252e-01 1.44293869e+00 3.92812282e-01 -5.49949050e-01 1.15255284e+00 2.18161523e-01 -1.63700894e-01 1.89119458e-01 -5.33408403e-01 -1.33266062e-01 7.59708643e-01 7.92008817e-01 2.62048364e-01 2.30873793e-01 -2.72215992e-01 5.45697153e-01 5.65517396e-02 -6.97946697e-02 4.89447750e-02 8.23725879e-01 -5.29526174e-01 -1.28686702e+00 -8.43834937e-01 6.20087862e-01 -4.28241521e-01 -1.64319888e-01 -1.42477334e-01 7.61153102e-01 -4.66534644e-01 7.41893768e-01 -4.85050648e-01 -2.04119444e-01 1.07983410e-01 3.61354083e-01 7.65804708e-01 -6.34595454e-01 -3.73564661e-01 2.53693938e-01 8.74729678e-02 -1.37831256e-01 2.14275524e-01 -6.14854097e-01 -1.10678875e+00 6.14366233e-02 -7.98324883e-01 3.88114631e-01 6.41365469e-01 7.13286340e-01 1.54128060e-01 -3.86313647e-02 4.05998290e-01 -6.95351303e-01 -1.21514714e+00 -9.52731371e-01 -1.05279672e+00 7.78740197e-02 -1.28681958e-01 -1.02499747e+00 -5.10433078e-01 2.75512375e-02]

[7.661070346832275, 4.178031921386719]

814c48d2-d115-4f55-8ea8-7f4244440534

what-makes-a-language-easy-to-deep-learn

2302.12239

https://arxiv.org/abs/2302.12239v1

https://arxiv.org/pdf/2302.12239v1.pdf

What makes a language easy to deep-learn?

Neural networks drive the success of natural language processing. A fundamental property of natural languages is their compositional structure, allowing us to describe new meanings systematically. However, neural networks notoriously struggle with systematic generalization and do not necessarily benefit from a compositional structure in emergent communication simulations. Here, we test how neural networks compare to humans in learning and generalizing a new language. We do this by closely replicating an artificial language learning study (conducted originally with human participants) and evaluating the memorization and generalization capabilities of deep neural networks with respect to the degree of structure in the input language. Our results show striking similarities between humans and deep neural networks: More structured linguistic input leads to more systematic generalization and better convergence between humans and neural network agents and between different neural agents. We then replicate this structure bias found in humans and our recurrent neural networks with a Transformer-based large language model (GPT-3), showing a similar benefit for structured linguistic input regarding generalization systematicity and memorization errors. These findings show that the underlying structure of languages is crucial for systematic generalization. Due to the correlation between community size and linguistic structure in natural languages, our findings underscore the challenge of automated processing of low-resource languages. Nevertheless, the similarity between humans and machines opens new avenues for language evolution research.

['Limor Raviv', 'Yoav Ram', 'Lukas Galke']

2023-02-23

['memorization', 'systematic-generalization']

['natural-language-processing', 'reasoning']

[ 1.94574758e-01 3.78834419e-02 2.16360062e-01 -1.19061351e-01 5.45801640e-01 -7.52247691e-01 1.01673841e+00 3.05456489e-01 -7.31242299e-01 5.83811224e-01 4.88845021e-01 -4.39909607e-01 -9.16432962e-02 -1.05681944e+00 -5.93262613e-01 -6.28793418e-01 -4.61842597e-01 4.71612573e-01 -8.01007673e-02 -6.26745462e-01 8.93305913e-02 3.08707505e-01 -1.54218650e+00 2.29522899e-01 8.96442890e-01 1.77807629e-01 4.08959270e-01 5.46054184e-01 -4.95518781e-02 7.13775635e-01 -5.07727981e-01 1.84647664e-02 2.93474585e-01 -8.61739218e-01 -9.34227169e-01 -1.58273429e-01 1.90587044e-01 2.61189044e-01 -2.13186979e-01 6.93771124e-01 4.75925744e-01 2.15861499e-01 7.10435569e-01 -5.31244636e-01 -1.10247910e+00 1.06118536e+00 -1.19318657e-01 5.60919642e-01 1.13424897e-01 7.37625539e-01 1.05264604e+00 -4.60064381e-01 8.48197162e-01 1.55643833e+00 8.39233220e-01 6.28322423e-01 -1.57732117e+00 -5.09951830e-01 2.02742264e-01 -3.17750484e-01 -1.26927149e+00 -5.40437400e-01 3.55088741e-01 -6.65651023e-01 1.53219938e+00 -2.76322603e-01 1.22054720e+00 1.04813111e+00 5.24763584e-01 2.44685858e-01 1.28646970e+00 -3.31892550e-01 2.46161595e-01 7.82403816e-03 -2.53364407e-02 7.42011189e-01 4.51760471e-01 3.93282324e-01 -6.47248626e-01 -1.60041869e-01 7.03593254e-01 -1.21327490e-01 -2.77799368e-01 9.78157967e-02 -1.43076551e+00 7.99176097e-01 7.34092057e-01 8.58160615e-01 -6.13740146e-01 2.14328021e-01 5.25799990e-01 7.38897562e-01 3.25096548e-01 1.04931903e+00 -3.89567614e-01 -3.22441757e-02 -6.63399160e-01 6.14084862e-02 7.93524504e-01 2.46460870e-01 6.95687711e-01 3.47462595e-01 2.68095225e-01 7.64000654e-01 8.13576207e-02 5.87772787e-01 9.94334280e-01 -9.11068618e-01 -1.49928212e-01 6.80306733e-01 -6.73408806e-01 -1.12192011e+00 -5.71585596e-01 -8.25861514e-01 -1.06540990e+00 5.58962487e-02 2.89070129e-01 -1.48034647e-01 -2.93332756e-01 2.43904614e+00 -1.82134792e-01 -6.18621230e-01 3.78109038e-01 5.70971370e-01 1.55829623e-01 7.15178728e-01 1.56586081e-01 -1.64696500e-01 1.20673895e+00 -4.56455797e-01 -1.14786066e-01 -5.64414084e-01 6.86101556e-01 -2.28241608e-01 1.36723149e+00 1.14027470e-01 -1.01867700e+00 -4.41234589e-01 -7.56546915e-01 2.46452093e-01 -4.40516800e-01 -6.77336812e-01 7.70360053e-01 5.56290567e-01 -1.61924839e+00 5.81275582e-01 -6.14964724e-01 -1.14498270e+00 3.01777691e-01 2.80897588e-01 -2.16918692e-01 4.22512472e-01 -1.28019977e+00 1.13330829e+00 5.98123789e-01 -1.43622488e-01 -7.47402191e-01 -4.69089150e-01 -5.10043919e-01 7.13101029e-02 -8.81845728e-02 -1.18271005e+00 1.08756816e+00 -1.54784656e+00 -1.41280532e+00 1.11436987e+00 -1.46476462e-01 -7.53933132e-01 -3.96941602e-02 3.69409949e-01 1.24826087e-02 -6.83206841e-02 -1.05436318e-01 8.60873103e-01 4.83176351e-01 -1.05610585e+00 -2.36532331e-01 -3.83429468e-01 -3.94074433e-02 2.84118652e-01 -2.84758717e-01 -1.04390785e-01 6.82334781e-01 -7.67672002e-01 -1.52089521e-01 -8.35306644e-01 -2.31528983e-01 -1.57777399e-01 8.02076310e-02 -3.81626904e-01 1.93366051e-01 -3.51135850e-01 1.04145169e+00 -2.05383444e+00 6.13910317e-01 2.55488396e-01 5.37521660e-01 -2.19377279e-02 -4.98531818e-01 8.68845761e-01 1.25696093e-01 4.35619026e-01 -4.53599155e-01 1.05061427e-01 1.03546381e-02 4.02611107e-01 -5.14926612e-01 3.49133909e-01 1.93912879e-01 1.28996146e+00 -1.07361567e+00 2.18769014e-02 -4.04569298e-01 3.79443824e-01 -8.00548732e-01 -7.52017945e-02 -2.32112929e-01 3.91136616e-01 1.05272487e-01 2.39009276e-01 3.68113443e-02 -5.75099111e-01 4.34748203e-01 6.16855085e-01 -3.14949006e-01 6.94113910e-01 -4.40923452e-01 1.43495464e+00 -6.07230008e-01 1.02564549e+00 1.60397980e-02 -9.06222165e-01 8.06599617e-01 9.26190466e-02 -2.61517167e-01 -8.64456952e-01 2.02661827e-01 2.84422249e-01 1.23051441e+00 -2.38953546e-01 3.12782377e-01 -5.75337112e-01 1.03274450e-01 1.36487448e+00 1.34205773e-01 -4.45357412e-01 2.49909028e-01 3.96180093e-01 8.94985974e-01 -4.19378936e-01 4.25850421e-01 -1.00590825e+00 2.71711886e-01 -1.31769121e-01 3.46143991e-01 1.06698895e+00 -2.46797949e-02 -1.80606917e-01 4.24491853e-01 -6.61973298e-01 -1.31669104e+00 -1.04705667e+00 4.87322733e-03 1.50135076e+00 -3.77328426e-01 -3.60074908e-01 -6.56855583e-01 1.77230269e-01 -6.74070865e-02 7.45632470e-01 -8.47924352e-01 -7.21809566e-01 -5.96488655e-01 -7.20031083e-01 9.36445892e-01 1.34954557e-01 4.32622135e-01 -1.54990757e+00 -1.27497530e+00 1.74025491e-01 1.01862743e-01 -6.10696495e-01 -3.11613649e-01 1.87268838e-01 -1.07853520e+00 -4.44279104e-01 -5.21954775e-01 -8.54890764e-01 6.88014448e-01 9.93923023e-02 1.36281538e+00 5.88925183e-01 -2.61154950e-01 4.37800825e-01 -1.29687399e-01 -2.73175925e-01 -1.03149080e+00 5.41701317e-01 7.12161362e-01 -5.02245784e-01 1.10764373e-02 -9.61467028e-01 -3.87857109e-01 8.50637853e-02 -1.23664963e+00 8.25596675e-02 6.52093172e-01 9.46783185e-01 -2.42002174e-01 -6.40518814e-02 6.96962297e-01 -4.98295695e-01 1.38956380e+00 -5.08554399e-01 -3.34875464e-01 3.59025180e-01 -6.96120322e-01 5.33259034e-01 7.50106752e-01 -6.11403108e-01 -8.10603142e-01 -5.56647062e-01 2.36123666e-01 6.03403151e-01 -4.85313199e-02 8.20668459e-01 6.83455646e-01 3.18584070e-02 1.13356435e+00 6.50381863e-01 4.17046100e-01 4.98725520e-03 3.84201527e-01 4.35429186e-01 7.06631914e-02 -8.58057201e-01 6.98698521e-01 3.20693105e-01 -2.14356601e-01 -1.27675402e+00 -2.53121525e-01 4.44734275e-01 -7.33913779e-01 -7.53715560e-02 6.34319484e-01 -7.96835244e-01 -8.33627820e-01 5.88728964e-01 -1.22356915e+00 -8.73281062e-01 -5.69440484e-01 2.80458659e-01 -3.91894817e-01 2.26646036e-01 -8.35968018e-01 -7.55571246e-01 -5.66934109e-01 -8.53501916e-01 5.60063004e-01 8.03339761e-03 -6.90747261e-01 -1.41206098e+00 6.00045502e-01 -3.93032730e-01 1.11986327e+00 -8.08897316e-02 1.45145881e+00 -6.71644688e-01 -3.42296332e-01 4.83211845e-01 7.88133219e-02 -6.08034083e-04 1.70270383e-01 -1.93415601e-02 -6.67538524e-01 -3.79662395e-01 -2.72575095e-02 -5.79080164e-01 1.05684316e+00 9.12093222e-02 1.78940475e-01 -5.01554728e-01 -4.73069176e-02 4.80335504e-01 9.93632674e-01 1.34569854e-01 4.38409075e-02 4.09184009e-01 2.83284746e-02 1.07264256e+00 -5.85265875e-01 1.35625377e-01 3.96034807e-01 2.02052042e-01 -2.25333780e-01 1.11846194e-01 5.99632934e-02 -3.48791897e-01 7.88267553e-01 1.25546229e+00 -5.13346680e-02 -1.81702688e-01 -1.38538969e+00 4.30699617e-01 -1.49047077e+00 -1.14456487e+00 4.41791177e-01 1.96789110e+00 1.10932326e+00 -5.94495330e-04 2.94414699e-01 -3.36820453e-01 4.17291433e-01 9.57133994e-02 -6.00506067e-01 -7.77387500e-01 -8.33929956e-01 2.02630937e-01 1.88821238e-02 6.71349883e-01 -8.87523293e-02 1.25187433e+00 7.41342735e+00 3.05744201e-01 -1.34205008e+00 -1.46870524e-01 5.29551566e-01 -1.30625546e-01 -5.75881243e-01 -6.60176575e-02 -2.52473295e-01 3.57907191e-02 1.18522966e+00 -6.87516987e-01 1.19451547e+00 1.49697540e-02 1.25198260e-01 -1.03630558e-01 -1.26101542e+00 5.32332122e-01 9.21291336e-02 -1.35331571e+00 5.04895508e-01 6.37729317e-02 6.66900992e-01 5.46055019e-01 2.79706120e-01 2.66290367e-01 6.43482685e-01 -1.40359962e+00 7.31047928e-01 4.80518728e-01 3.37815940e-01 -1.97584406e-01 3.06075931e-01 6.24744833e-01 -7.86648512e-01 -2.45760396e-01 -4.76008922e-01 -8.49972546e-01 1.45539939e-01 2.64739901e-01 -7.28657544e-01 -3.07762384e-01 6.59061909e-01 6.23499036e-01 -8.77929449e-01 3.66225719e-01 -2.22584412e-01 5.85253835e-01 -4.70447004e-01 -4.67944682e-01 2.21311495e-01 -1.45281360e-01 6.01842523e-01 1.27433610e+00 4.73270789e-02 5.31467572e-02 -1.57332242e-01 1.38233590e+00 2.37896182e-02 1.93616286e-01 -1.04879260e+00 -6.21439099e-01 4.54947919e-01 7.60116756e-01 -9.53125238e-01 -4.49586928e-01 1.43534150e-02 8.84983063e-01 5.80793142e-01 4.46796238e-01 -9.39442664e-02 -5.15767224e-02 5.00813007e-01 2.91366037e-02 -6.07146658e-02 -6.64735079e-01 -2.50617176e-01 -1.18582952e+00 -4.21665967e-01 -1.20233274e+00 1.98242124e-02 -5.84204674e-01 -1.48284423e+00 9.63629723e-01 -1.20121956e-01 -2.64031023e-01 -4.64199722e-01 -4.98429120e-01 -5.55790484e-01 8.70142579e-01 -8.85580659e-01 -9.18537259e-01 1.50610685e-01 4.57592249e-01 3.67153883e-01 -4.14172351e-01 8.18370461e-01 -3.13996673e-01 -4.58265156e-01 4.55387294e-01 2.31890213e-02 -5.22032604e-02 2.98881382e-01 -8.28438342e-01 8.53366971e-01 6.72411501e-01 2.81024307e-01 1.54201341e+00 6.79699063e-01 -6.17674172e-01 -1.23585057e+00 -6.43719673e-01 9.10730720e-01 -4.31294531e-01 8.58995616e-01 -7.52188861e-01 -9.73592699e-01 6.50521100e-01 8.59736085e-01 -6.82510614e-01 8.09744418e-01 3.39540750e-01 -6.39946580e-01 2.68131584e-01 -7.93488741e-01 9.72761273e-01 1.50711155e+00 -9.67747927e-01 -8.93799126e-01 1.23765118e-01 1.04228783e+00 5.85299790e-01 -5.58149695e-01 8.39976445e-02 6.94672465e-01 -9.06380951e-01 6.68652296e-01 -8.32391918e-01 4.54360962e-01 -1.04758210e-01 -6.17358051e-02 -1.64094460e+00 -6.45278513e-01 -5.23676634e-01 5.21514654e-01 8.00225616e-01 7.48941362e-01 -1.30125499e+00 7.93573260e-02 2.79098719e-01 3.40023071e-01 -4.60198820e-01 -8.13856900e-01 -7.83605814e-01 7.84191191e-01 4.70193140e-02 3.47377092e-01 1.14534199e+00 3.75168979e-01 6.87689483e-01 1.31859973e-01 -3.16914082e-01 3.57708365e-01 -5.09831570e-02 5.58221817e-01 -1.37222040e+00 -5.82647204e-01 -1.15775692e+00 -3.05477053e-01 -7.16923892e-01 4.65730876e-01 -1.38583779e+00 -1.20040871e-01 -1.17853212e+00 3.58739048e-01 -1.47727460e-01 1.74666848e-02 4.19744730e-01 4.87642623e-02 -1.04226582e-01 3.95913035e-01 5.10734260e-01 -3.79145175e-01 6.35101676e-01 7.97298253e-01 1.56524554e-02 -3.89206052e-01 -5.53043246e-01 -9.34152365e-01 6.46273434e-01 9.82302129e-01 -3.04451644e-01 -3.56160611e-01 -8.05183411e-01 1.02288508e+00 -6.01928711e-01 2.83046365e-01 -8.71240735e-01 3.06295216e-01 -9.51451287e-02 3.47800478e-02 3.81931305e-01 -2.55665332e-01 -4.71679240e-01 7.62174129e-02 1.32896030e+00 -6.92922831e-01 7.34685004e-01 3.71531695e-01 2.77392060e-01 2.24907920e-01 1.00189872e-01 7.47735858e-01 -5.04451156e-01 -4.18078363e-01 -9.98477414e-02 -1.41023159e+00 2.74199069e-01 5.75993240e-01 -2.20132887e-01 -4.66592610e-01 -3.96804065e-01 -4.55721974e-01 1.92042086e-02 9.05789971e-01 5.23833811e-01 4.78842437e-01 -8.61133814e-01 -8.64113271e-01 2.86690831e-01 -2.90261284e-02 -4.39890295e-01 -1.72686905e-01 6.91243887e-01 -6.07799768e-01 3.30431700e-01 -5.75704396e-01 -5.82504809e-01 -6.67609453e-01 4.76394385e-01 8.05361331e-01 4.93389852e-02 -2.75626242e-01 8.97685230e-01 5.16169727e-01 -8.43402207e-01 -2.20964253e-01 -5.13809323e-01 1.00078098e-01 8.74800086e-02 3.09196621e-01 1.18683325e-02 -5.18665433e-01 -6.37649655e-01 -3.18849087e-01 5.74856877e-01 -2.89206598e-02 -4.41506684e-01 1.38286293e+00 -1.00860476e-01 -8.98224592e-01 9.64681923e-01 7.43914366e-01 -1.92871429e-02 -5.44265091e-01 -6.08237803e-01 1.77270740e-01 1.19399920e-01 -4.20015991e-01 -6.58753633e-01 -6.37385190e-01 9.56325114e-01 3.54127735e-01 2.82747835e-01 8.73170078e-01 2.09406301e-01 3.78234535e-01 1.11480916e+00 5.31754911e-01 -8.81757796e-01 2.83407927e-01 1.05508626e+00 9.62732434e-01 -8.42081070e-01 -2.14332193e-01 5.33838928e-01 -4.10468936e-01 9.77611542e-01 5.41299284e-01 -1.59410641e-01 4.12715852e-01 1.01300500e-01 -6.38898239e-02 -3.99630845e-01 -1.28367972e+00 -8.57973248e-02 -1.16747290e-01 5.62846661e-01 7.66847730e-01 -8.53910390e-03 -3.89178634e-01 1.10283852e-01 -7.86295891e-01 -3.56376410e-01 5.29576778e-01 6.53090119e-01 -7.15782881e-01 -8.73646557e-01 -1.44255057e-01 1.58264056e-01 7.49081955e-05 -5.81291914e-01 -9.61179495e-01 5.86411059e-01 5.07097021e-02 6.02805376e-01 2.99867362e-01 -3.97532016e-01 -3.76016237e-02 3.12446147e-01 4.50971633e-01 -8.23113441e-01 -1.11651981e+00 -5.12558043e-01 -3.15096647e-01 -7.94087723e-02 -3.52833182e-01 -8.13414216e-01 -1.25171340e+00 -5.92288673e-01 -5.86464331e-02 1.92757845e-01 2.01425746e-01 7.75494277e-01 5.43867528e-01 3.91452193e-01 1.81163400e-01 -6.60616577e-01 -5.02936780e-01 -1.03460634e+00 -3.60919535e-01 3.71351659e-01 3.98423105e-01 -8.69809762e-02 -4.81803328e-01 -2.48889159e-02]

[4.529013633728027, 1.6490544080734253]