dedup-isc-ft-v107-score
float64 0.32
1
| uid
stringlengths 32
32
| text
stringlengths 0
32.5k
| paper_id
stringlengths 1
14
| original_image_filename
stringlengths 5
222
|
|---|---|---|---|---|
0.401846 | 6a5256579a2d453c938d469d12918c0e | Same as fig. <ref> for ring of radius R=5 a.u.. | 2012.15451 | ZandNucConstantR5Fig.jpg |
0.372926 | 115a363f927f4c71b5b449f5886a66da | Same as fig. <ref> for n alike spin particles on a ring of radius R=2 a.u.. | 2012.15451 | ZandNucConstantRFig.jpg |
0.42477 | b7c994d65f6d40a6995b75331c0dbc94 | QMC (symbols) and HF (lines) energy per particle ϵ for n alike spin particles on a r_s=1 ring with a repulsive Yukawa interaction potential exp(-ar)/r for a∈{0.0,0.1,0.5,1.0,∞}. | 2012.15451 | aConstantRhoFig.jpg |
0.404356 | 9b85270358ab4e539912b7498287803d | Diagram of our proposed framework. The ResDAVEnet-VQ model was trained using a {2}→{2,3} curriculum (in the notation given in <cit.>). | 2012.15454 | diagram.jpg |
0.499857 | c52719924a504f3891ca7646111de08d | M-SPICE F-score (same as Figure 5) and recall on the SpokenCOCO test set with different candidate proposal methods. | 2012.15454 | mspice_stage2_lg.jpg |
0.40498 | 5cde266596024a04b35e355fac8a8da5 | Word-based caption evaluation results of all five metrics on SAT models trained on different units, using both beam search decoding and sampling for unit sequence generation. | 2012.15454 | unit_comp_full.jpg |
0.450113 | b8cbb5277c6843adbe95bc03a2aa9025 | MSCOCO test set SPICE of various units and decoding methods. VQ3\RLE denotes VQ3 units without RLE. Top-k sampling considers the k-most probable units at each step. Other metrics are shown in Figure A3. | 2012.15454 | unit_comp_spice.jpg |
0.441763 | 92cf18ef55714e5abd478a1d3ffde459 | Utterance duration histograms for the three visually-grounded speech datasets. | 2012.15454 | vgs_dur_hist.jpg |
0.450105 | 1f6bba74dcf54122a1473de341a8f759 | Pendulum. | 2012.15458 | pendulum.jpg |
0.444828 | 1f92af4c84b3420381ca5a3eb591ee49 | Random-receiver quantum communication task. The sender A wants to transmit a quantum message to one of the n spatially separated receivers {B_i}_i=1^n. The identity of the targeted receiver x is unknown to the sender but known to the receivers. Receivers are allowed to collaborate through classical messages, but quantum communication among them is forbidden. | 2012.15459 | fig0.jpg |
0.449108 | 1e73ccabd93b41b79a1f12b8b1f5c50b | 2012.15459 | fig_1.jpg |
|
0.408504 | 3ad2625b9b244f3f984625c8ee0d6740 | 2012.15459 | fig_2.jpg |
|
0.452242 | 750cc7b8d4dc45ed92feab6003060021 | Random-receiver quantum communication task with controlled SWAP operations in space. | 2012.15459 | fig_3.jpg |
0.486365 | 28cf7d5e2e964141b5e24d44b7a73e76 | Two designs to introduce Re-ID into DETR. The left one uses a shared feature from a single cross-attention layer to train detection and re-identification. The right scheme uses two cross-attention layers to generate independent Re-ID features and detection features for the two sources of supervision. | 2012.15460 | reid.jpg |
0.471294 | 2e24cf3b8bde43ff8eb22a3fa120e044 | A schematic illustration of a directed K-order subgraph. | 2012.15462 | Fig_1.jpg |
0.537515 | 34f5832e7f6749dfa1f6a6d0c5158ad0 | An illustration for the Ethereum transaction network. Nodes are labeled by account addresses. Edges are attached by timestamp t and amount value w, and indexed in the increasing order of t. | 2012.15462 | Fig_2.jpg |
0.465217 | 7603489052a74143ada34a737b1c04ac | The distribution of nodes appearing in Ethereum transaction subgraph follows a power-law. | 2012.15462 | Fig_3.jpg |
0.42976 | ca70c175311746b99860815428d67939 | Illustration of an l-length temporal walk | 2012.15462 | Fig_4.jpg |
0.489332 | be5370750ebf4929b89e12d25d8efba5 | The illustration of Ethereum transaction network modeling and analysis. | 2012.15462 | Fig_5.jpg |
0.529162 | 606937b8caf04401b2ea985146c4bf40 | Performance in terms of Area Under Curve (AUC) under varying hyperparameters, when (a) fixing l=10, r=20, d=128, and varying k from 2 to 8; (b) fixing k=4, r=20, d=128, and varying l from 4 to 10; (c) fixing k=4, l=10, d=128, and varying r from 8 to 20; (d) fixing k=4, l=10, r=20, and varying d from 8 to 256. | 2012.15462 | Fig_6.jpg |
0.504485 | e8f791f0bf034fd88b355717e2cec614 | ODENet architecture | 2012.15465 | ODENet.jpg |
0.460596 | 1705f52b8a5a4fbb8ef430247106323d | ODEBlock design on FPGA | 2012.15465 | design.jpg |
0.473122 | f0863aa71e0346eb881bb38e1c641920 | Overview of PYNQ-Z2 board | 2012.15465 | pynq.jpg |
0.524107 | 686e3cd8b18b4108a47b3ab0f8324ad1 | ResNet architecture | 2012.15465 | resnet.jpg |
0.406413 | 837de3617f7e4360badb168006d23ab7 | Accuracy of four network architectures when N={20,32,44,56} | 2012.15465 | test_accuracy_for_raw_2021_0215.jpg |
0.431115 | 240c958e327e43a0ac29f6377d5097f7 | Bound for C_u | 2012.15467 | Cu.jpg |
0.508005 | 577ce5b0ee99425fa7d93eb7d9f138ea | Bound for L | 2012.15467 | L.jpg |
0.474568 | c91b7610b9304380a96020bda679bce8 | Matrix sensing with f(Z)=1/2T(Z)-y_2^2, n=100, r=5, m=2500. | 2012.15467 | fig_sec1_ms.jpg |
0.429876 | 874f27b7fd044152aa69e81aa6634c5d | Least squares function with F_1(Z)=1/2Z-X_F^2, n=200, r=10. | 2012.15467 | fig_sec1_original.jpg |
0.518079 | 6327c89d8dc645cfb06ad4e0a88c3c65 | Phase retrieval with f(Z)=1/2T(Z)-y_2^2, n=100, m=1000. | 2012.15467 | fig_sec1_pr.jpg |
0.500324 | f68b2adef92f4c53a1f51c783d24f1e1 | Magnitude of the gradient in the neighborhood of a spurious fixed point | 2012.15467 | fig_sec4_critical.jpg |
0.458166 | c97c5873275e421eb784763a3b8b8917 | Convergence guarantee on the majority of _r | 2012.15467 | fig_sec4_maj.jpg |
0.408312 | f8b671352f3248f9a56759a7bd5b0a0b | The evolution of h | 2012.15467 | pr_h.jpg |
0.421879 | 5480a35e80b0444e96f8bba06ccbade4 | Log-error | 2012.15467 | pr_logerror.jpg |
0.421018 | 1b2b0517930a42379ba2bcdd273d1785 | The evolution of ρ | 2012.15467 | pr_rho.jpg |
0.393295 | 9593ba60b83649c69d9ffbe57aa6f476 | The evolution of h | 2012.15467 | rank1_h.jpg |
0.505063 | 6a6bcd34b7d34ae09fe049b6b3550b9a | Log-error | 2012.15467 | rank1_logerror.jpg |
0.397526 | 423cc1940f354bc59bc8dd6658e7a31f | The evolution of ρ | 2012.15467 | rank1_rho.jpg |
0.557483 | de76a69e191449018ad318b01569f015 | Effect of trajectory length | 2012.15470 | acc_v_length_fullhome.jpg |
0.422519 | 41317abfa4b84c77810d37cd9e5abb80 | Additional Interior Map Visualizations: We present additional visualizations of the estimated Interior Maps. The circled areas indicate the locations that are misclassified by our proposed model. See text for discussion. | 2012.15470 | additional_interior_vis_full_occant.jpg |
0.430866 | dd5f14462a014f6395ce9f871fccf941 | Additional Room Map Visualizations | 2012.15470 | additional_room_cls.jpg |
0.424718 | 2c481507a2c348e3bd9e86612d9e11d1 | Room Class Confusion Matrix | 2012.15470 | confusion.jpg |
0.39132 | 267b9b9100c1419ebac85f73bc0711c6 | model overview: Our model has three stages: feature extraction, feature alignment, and a sequence encoder-decoder. At each time step, the feature extractor obtains independent top-down floorplan-aligned features for each modality (ambisonic audio and RGB). These features are aligned to a common coordinate frame using the relative motion of the camera. Entire sequences of audio and visual features are then processed by an encoder using a series of pixelwise self-attention operations and convolution layers. Finally, the two modalities are fused using a decoder architecture also including a series of self-attention and convolution layers. We depict training for three time steps here. No matter how many steps are used in training, the model is applicable to arbitrary length sequences at test time. | 2012.15470 | model.jpg |
0.42138 | 05b99c86c4b148f297a1b39cff4cccf7 | vs. RGB-only: Here we compare the interior map predictions of and its RGB-only variant. In red, we show interior locations correctly captured by but not captured by the RGB-only model. In blue, we show locations captured by the RGB-only model but missed by AV-Map. Observe that anticipates regions that are not visible through the camera (cone), which the RGB-only model cannot capture. | 2012.15470 | qual_analysis_3d.jpg |
0.400131 | 9ea0f329bee54815a27cc4286032daf1 | Interior map predictions: Here we visualize reconstructed interior maps estimated by OccAnt <cit.> and our proposed audio-visual model in the device-generated audio setting. In the 3D examples, red cones are camera positions, cyan is the observed ground plane area, and transparent light green is the freespace inferred by our model. 2D output maps are color-coded against ground truth (see legend). The perfect 2D map would be all green and black. Our method “sees" substantially more area by using audio, and it produces more accurate maps than the state-of-the-art mapping method <cit.> that also attempts to extrapolate beyond the directly observed area. | 2012.15470 | qual_interior_moved_legend.jpg |
0.458157 | f67c865bb39b4da9b5e53f0d3c3c0ed1 | Audio-visual floorplan reconstruction: A short video walk through the house can reconstruct the visible portions of the floorplan, but is blind to many areas. We introduce audio-visual floorplan reconstruction, where sounds in the environment help infer both the geometric properties of the hidden areas as well as the semantic labels of the unobserved rooms (e.g., sounds of a person cooking behind a wall to the camera's left suggest the kitchen). | 2012.15470 | teaser.jpg |
0.436125 | 4fb3e54aa763417abc61cc2c1a4b4905 | Different space restriction strategies for acquisition function optimization visualized from left to right ellipse, convex hull and distance method. In the distance method the maximum distance the point can move is computed as a 90% of the maximum of the move distances of latent space means of the train data. The white regions indicate the area where the acquisition function is optimized. The red continuous line visualizes the region which contains 99% of the data used to train the VAE. | 2012.15471 | extrapolation_regions.jpg |
0.449341 | 307f7be867974cd48a6b9020d53b115e | Masses versus widths of the Y(4220) and Y(4360) obtained from the different final states by BESIII <cit.> and Y(4360) quoted from PDG <cit.>. Here the errors reflect both statistical and systematical uncertainties. | 2012.15473 | etajpsi-cmp.jpg |
0.499201 | ffa754819e36431c8e1c49d31de1ff59 | Top: Cross section and fits of e^+e^-→η J/ψ for XYZ data. Bottom: Same for the scan data. Dots with error bars are data. The solid (blue) curves represent the fit results of the following interfering amplitudes: Y(4040) (dashed red), Y(4220) (short-dashed pink), Y(4360) (short-dashed purple), and P-PHSP (long-dashed green). | 2012.15473 | etajpsi-cs.jpg |
0.484496 | f22940997e2640bdb0802f6cb3e9633e | Fit of Born cross sections of e^+e^-→η J/ψ with mass and width fixed Y(4160) and Y(4260). | 2012.15473 | etapjpsi-fit.jpg |
0.41378 | 201cb70052b84dfe9b0a977c3761ef10 | The e^+e^-→ωχ_c0 cross section as a function of the center-of-mass energy. The blue points are from Ref. <cit.>, the black square points are from Refs. <cit.>, and the red solid line is the fit result. | 2012.15473 | omegachic0-cs.jpg |
0.439987 | 4842f838bbf04042a6c7423d95ee045c | Fit to the dressed cross sections of e^+e^-→π^+ D^0D^*-, where the black dots with error bars are the measured cross sections and the blue line shows the fit result. The error bars are statistical only. The pink dashed triple-dot line describes the phase-space contribution, the green dashed double-dot line describes the recond resonance contribution, and the light blue dashed line describes the first resonance contribution. | 2012.15473 | pipD0Dstarm.jpg |
0.55711 | c4630c345ae4435f96729b6a158f6ffd | Top: Fit to the Bron cross sections of e^+e^-→π^0π^0J/ψ, where points with error bars are data, the red solid line is the total fit result, the blue dotted line is the nonresonant component, while the red dashed and dot-dashed lines represent the contributions from Y(4220) and Y(4320), respectively. Bottom: Cross-section ratio of e^+e^-→π^0π^0 J/ψ to e^+e^-→π^+π^- J/ψ, where the black dashed line corresponds to the average. | 2012.15473 | pizpizjpsi-cs.jpg |
0.501738 | ccabae69b3074813bdfeebe6b8005148 | (Left column) Dalitz plots of M(π^0J/ψ) versus M(π^0π^0), invariant-mass projections (middle column) M(π^0J/ψ) and (right column) M(π^0π^0) of the results of the nominal PWA for data samples √(s) = 4.226 - 4.258 GeV. Points with errors are data, red solid curves are the total fit results, the blue dashed (magenta long-dashed) curves represent Z_c(3900)^0 (ππ S-wave) components, and green shaded histograms represent the estimated backgrounds. Each event appears twice in the Dalitz plots and M(π^0J/ψ) distributions. The χ^2/ndf is calculated by merging those bins with less than 10 events in the Dalitz plots. | 2012.15473 | pizpizjpsi-pwa.jpg |
0.481886 | a8ce926b91874a68a259fb99ca848d03 | M(γ_L D^0D̅^0) with M(γ_L D^0) (a) in or (b) below the D^*0 mass window. (c) M(π^0D^0D̅^0) with M(π^0D^0) in the D^*0 mass window. Simultaneous fit results for X(3872) → D^*0D̅^0 with (d) D^*0→γ D^0 and (e) D^*0→π^0D^0 mode. (f) Fit results for X(3872) →γ_LD^+D^-. The points with error bars are from data, the red curves are the best fit, and the blue dashed curves are the background components. | 2012.15473 | x3872-dd.jpg |
0.436739 | b7df6f2b52fc47d38650a9c7e40cb996 | (a) Fit results for X(3872) →γ J/ψ for the μ^+μ^- (top) and e^+ e^- (bottom) mode. (b) Fit results for X(3872) →γψ(3686) for the π^+π^-J/ψ (top) and μ^+μ^- (bottom) mode. The points with error bars are from data, the red curves are the best fit. In (b), the rose-red dotted line represents the fit with the signal constrained to the expectation using X(3872)→π^+π^-J/ψ based on the relative ratios taken from a global fit <cit.>; the green dash-dotted lines are using X(3872) →γ J/ψ as the reference based on the LHCb measurement <cit.>, and the gray long dashed lines are using X(3872) →γ J/ψ as the reference based on the Belle measurement <cit.>. | 2012.15473 | x3872-gamjpsi.jpg |
0.463519 | 8ee1ebe3c4ee432baf0546d6fa06e45c | Schematic plot of an atom-waveguide interface. A periodic array of atoms aligning close to the nanofiber couple with each other with one-dimensional dipole-dipole interactions. This long-range interaction emerges via the guided modes supported by the waveguide along ẑ. A zoom-in plot of a dashed box of three atoms presents the effect of dipole-dipole interactions which couple the nearest-neighbor atoms (thick double arrows) and beyond (dashed double arrows). Each atom is a two-level quantum emitter (|g⟩ and |e⟩ as the ground and the excited states, respectively) which decays with an intrinsic rate of γ to either sides of the waveguide. | 2012.15474 | fig1.jpg |
0.411795 | 1eac2c63d7934950bc7f70e125df8a77 | Atomic excitation (⟨ P_j(t)⟩) localization. The central atom is initially excited at the 51th site of the periodic atomic array with N = 101 and ξ = π/8. From the left to the right panels, a crossover from the excitation delocalization to localization is presented as the disorder strengths increase from w/π = 0, 0.01, 0.025 to 0.1. The presented results here and throughout the paper are converged after averaging over 200 realizations of disorders. | 2012.15474 | fig2.jpg |
0.443656 | 53415a66bfa0469299d1f4f91663fe45 | Time-evolved localization lengths and power-law scalings of the entropy of entanglement. (a) The total excitation population P_ tot(t) = ⟨∑_j=1^N P_j(t)⟩ (blue-solid line) decays as time evolves for the disorder strength w/π = 0.01. The case without disorder (red-dashed line) is plotted as a comparison, which is close to the lower bound of the shaded area formed by 1σ deviation (dotted line) of the disordered case. Three insets are population distributions P_j(t) at γ t/100 = 5, 10, 15, respectively. (b) The fitted power-law scaling of a function t^-β_w in the von Neumann entropy of entanglement for the time range between γ t/100 = 5-15. The quotient β_w is normalized by the one without disorder and is plotted in logarithmic scales of w. For both plots, we consider N = 101 and ξ = π/8 as an example. | 2012.15474 | fig3.jpg |
0.435476 | 7ca7b96ca7334caca9b34e663119d60e | The mean and intrasample variance of gap ratios for various disorder strengths. The mean gap ratio (a) decreases while the intrasample variance (b) increases as w increases for N = 101 (solid line), 201 (dash-dotted line), 301 (dotted line), 401 (□), and 501 (×) with ξ/π = 1/8. (c) The mean and rescaled intrasample variance of gap ratios for N = 501 with ξ/π = 1/8 (solid line), 1/4 (dash-dotted line), and 1/2 (dotted line) are calculated, which indicates an interaction-driven excitation delocalization at low w/π ≲ 0.02. | 2012.15474 | fig4.jpg |
0.401061 | 1dc3a94d4300477daf2a8807f4805c04 | Model used for predicting spatiotemporal sequence follows a sequential approach with BatchNormalization layer corresponding to every ConvLSTM layer. The model presented here is for Input shape 60× 60× 3. | 2012.15476 | model.jpg |
0.476403 | 1415c2f42f064b23bcca20385b19c9d8 | PDA with general ℓ_p regularizer (<ref>) in learning target function with varying sparsity s. | 2012.15477 | sparsity_comparison_errorbar.jpg |
0.509307 | c938dff9e6694d1cbfe5e85bcaa5c114 | Test error of mean field neural networks (α=1) trained with SGD (red) and PDA (blue), and network in the kernel regime (α=1/2) optimized by SGD (green). | 2012.15477 | test_loss_MSE_new.jpg |
0.462484 | 734a2c3dcc054b259dba5e16e6644cfa | Chernoff Information of the graphs of ψ(β) and η_β = d/dβψ(β). | 2012.15480 | alpha_star.jpg |
0.48971 | 3e25c9621be04ff1b66b36b5cb376378 | Chernoff point on ψ(β) | 2012.15480 | jensen_gap.jpg |
0.459702 | 12806e230ee84032b08e1dee9b58f529 | Sanov's Theorem. Each exponential arc intersects the manifold ℳ_η̅ with a right angle, due to a generalized Pythagorean Theorem for the kl divergence (see Sec. 1.6, 2.8 of <cit.>). | 2012.15480 | sanov.jpg |
0.510443 | 3ab85fae010043658ab40f552527270e | Chernoff point on η_β = ∇ψ(β). | 2012.15480 | tvo_chernoff2.jpg |
0.446412 | ba659286d37c41a6b55615aefd76c290 | Fusion-in-Decoder architecture for rationale prediction. Each sentence from the passage is marked with sentence markers S1 ... SN. The passage is broken up into C contexts/chunks, which are passed to the encoder. The decoder then attends to the C concatenated and encoded passages to generate the output sequence. The output sequence is the classification token followed by rationale sentence markers. | 2012.15482 | fid_qa_architecture.jpg |
0.397513 | a31abcd0697f43cd9aa119d4718038f9 | Example questions, answers, and corresponding passages from the BoolQ and MultiRC datasets from the ERASER benchmark <cit.>. Annotated rationales are highlighted. Note that rationales can be multi-sentence and non-contiguous. | 2012.15482 | qa_examples.jpg |
0.460699 | 14ba77502b4a40e997e3601cd0ab0856 | Dominance probabilities | 2012.15483 | cifar_dominance.jpg |
0.412938 | 7e5407515442423386152f5ca23789f9 | Events defined by triplets | 2012.15483 | cifar_triples.jpg |
0.50284 | c8b3a2176adc4eae8c0df1f6848f1fae | Dominance probabilities | 2012.15483 | dominance.jpg |
0.51596 | 0ab2a500dafa4d99800d0a39cdaaba39 | Dominance probabilities | 2012.15483 | dominance_fisher.jpg |
0.48944 | 7c60692bebf24801807443bdb5d176c3 | Guarantee after applying (<ref>) once. | 2012.15483 | hypotheticalv1.jpg |
0.404034 | 06e0bbc44c394ce9b2246be77ecb8ca7 | Guarantee after applying (<ref>) thrice. | 2012.15483 | hypotheticalv2.jpg |
0.454977 | 8f078fdf0b5d451da449eedd22ba6901 | Guarantees with δ_1 = 0.0, δ_2 = 0.25, and ν_1 = ν_2 = 0.005. | 2012.15483 | imagenet-probit-eps-converted-to.jpg |
0.506221 | 39f66452f7154680ae91197e950d4fb5 | Guarantees with δ_1 = 0.31, δ_2 = 0.38 and ν_1 = 0.005, ν_2 = 0.008. | 2012.15483 | imagenet-probit-pess-eps-converted-to.jpg |
0.4373 | 3ff1f02222cc498bb7e41a15fef1d203 | Piecewise linear fit of 66 ImageNet models with two pieces. The switch point between pieces is set to be the 6-th least accurate model. | 2012.15483 | imagenet-switch-eps-converted-to.jpg |
0.407001 | 87ec96ed746444ef9eafc28ca5dd7a8d | Events defined by triplets | 2012.15483 | triples.jpg |
0.405049 | 3585431a07c544abab24926f124d52d7 | Events defined by triplets | 2012.15483 | triples_with_fisher.jpg |
0.418373 | b586e603f1494e189ac909e246864c98 | Seeing is Knowing architecture. (1) Scenes, objects, & actions are detected in the image. (2) For detected entities, we retrieve their KG embedding representations. The span of these embedding vectors represents the `image-as-knowledge'. (3) Lexical semantic vectors for each word in the query are accumulated via an LSTM. (4) The joint image-question encoding is derived using a CoAttention mechanism described in Fig. <ref>, then (5) passed through a multi-layer perceptron, whose (6) last layer is used as a pair of queries that are (7) gated, and (8) used to retrieve the entity to answer the question. | 2012.15484 | Architecture.jpg |
0.461163 | 99a009780df340c99685ff4e3baa73d6 | Image and query are fused using the CoAttention mechanism depicted here. First, self-attention computes a weighted summary of the query (bottom orange circle). Second, the query is used to compute an attention-weighted summary of the concepts in the image (top orange circle). The resulting image query is a vector drawn from the span of the entities present in the image. | 2012.15484 | CoAttention.jpg |
0.388023 | 835a33d03c4d4d8e98e11aa27fce7e57 | An illustration of the proposed scalable Bayesian gradient aggregator, which can be implemented as a simple neural network. | 2012.15486 | Fig_SBFL.jpg |
0.389866 | 5521402b00b342518615b4194da7d641 | A comparison of test accuracy according to different learning rate γ^t∈{10^-2,10^-3} when the downlink compression is used. | 2012.15486 | Fig_convergence_DC2.jpg |
0.407892 | 995f3c07b27341e1bed948ad662fcd0b | A comparison of training loss for homogeneous and heterogeneous synthetic datasets. Training losses after converged are shown in magnified box in each figure. | 2012.15486 | Fig_convergence_syn.jpg |
0.390596 | a043ec970be344b5867f76ae9969d60d | The effect of quantization levels of ν_k^t or (λ_k^t) according to SBFL with different priors. | 2012.15486 | Fig_quantization2.jpg |
0.4524 | d5d255017bea4a38bb066bf23232e167 | Heterogeneities in the simulation setup. | 2012.15486 | Fig_simulation2.jpg |
0.413536 | 0a776f8e980d42928da1cba751c2f64e | Statistics of the logarithm of local exports. (a, b) Distribution of the logarithm of the exports from each country c (a) to different destinations and (b) of different product categories in year 1980. Color varies with the out-degree of c. (c, d) The mean and standard deviation of the logarithmic exports from country c (c) to different destinations and (d) of different products versus the out-degree of c. (e, f) Distribution of the standardized logarithmic exports from country c (e) to different destinations and (f) of different products. The dashed lines in (e) and (f) are P̃^ (Gaussian)(ṽ) = e^-ṽ^2 2√(2π). (g, h) The skewness of the logarithmic exports from c (g) to different destinations and (h) of different products versus the out-degree of c. Shades in (g) and (f) represent the middle 90% range of the random-version skewness. | 2012.15487 | Figure1.jpg |
0.418304 | 533d2808f9b44192beaa9d426625c7c5 | Statistics of local in-degrees in the EI and the EP network. The in-degree of a product or a country node means the number of exporter countries exporting the product or to the country. (a, b) Distribution of the in-degrees of (a) the country nodes in the EI network and (b) the product nodes in the EP network receiving links from each exporter country c in year 1980. Color varies with the out-degree of c. Inset: The mean and standard deviation of local in-degrees versus the out-degree of exporter. (c, d) The skewness of the in-degrees of (c) the country in the EI network and (d) the product nodes in the EP network receiving links from c versus the out-degree of the exporter c. | 2012.15487 | Figure2.jpg |
0.442294 | 85805a642e824b8eb4889c8f87c50715 | Correlation between local in-degrees and local logarithmic exports and their skewness. (a) The joint probability density function (PDF) P^ (EI) (k̃,ṽ) = ∑_c,c^' A_cc^'δ(k̃_cc^'-k̃)δ(ṽ_cν-ṽ)/∑_c,c^' A_cc^' of standardized local in-degrees and local logarithmic exports in the EI network. The orientation of the red (high density) region to the upper right indicates a positive correlation. (b) The joint PDF P^ (EP) (k̃,ṽ) = ∑_c,p A_cpδ(k̃_cp-k̃)δ(ṽ_cp-ṽ)/∑_c,p A_cp. (c) Scatter plot of the skewness of local logarithmic exports to different destinations γ_v;c^ (EI) versus the skewness γ_k;c^ (EI) of local in-degrees in the EI network. Color varies with the out-degree of c as in Fig. <ref>. (d) Scatter plot of the skewness of local logarithmic exports of different products γ_v;c^ (EP) versus the skewness γ_k;c^ (EP) of local in-degrees in the EP network. | 2012.15487 | Figure3.jpg |
0.380132 | 0e97423dc9184a5db325af2722b4357b | Formulating local exports as functions of local in-degrees for given exporter. (a) Standardized logarithmic exports {ṽ_cc^'|c^'∈𝙸_c} to different destinations c^''s from a selected country c versus their standardized in-degrees {k̃_cc^'|c^'∈𝙸_c} in 1980. The slope ρ_c of the fitting line is equal to their Pearson correlation coefficient. The deviation of ṽ_cc^' from the prediction by the in-degree, ρ_c k̃_cc^', is set to √(1-ρ^2)ξ_cc^', by which the random noise ξ_cc^' is defined. (b) {ṽ_cp} versus {k̃_p} for a selected country c in 1980. (c) Pearson correlation coefficients ρ_c^ (EI) between the logarithmic exports to different destinations from an exporter c and their in-degrees versus the out-degree of c. (d) Pearson correlation coefficients ρ_c^ (EP) between the logarithmic exports of different products from an exporter c and their in-degrees versus the out-degree of c. | 2012.15487 | Figure4.jpg |
0.386999 | 3d404e66f41749c798cf1af24c0bf6a0 | Third-order cumulants of local in-degrees and random noises and their relative magnitudes. (a-f) are obtained from the data of year 1980. (a) The skewness of random noises ⟨ξ^3⟩_c^ (EI) in the local exports from each exporter c to different destinations versus the out-degree q_c^ (EI). (b) The skewness of random noises ⟨ξ^3⟩_c^ (EP) in the local exports from exporter c of different products versus the out-degree q_c^ (EP). (c) Third-order joint cumulants ⟨k̃^2 ξ⟩_c^ (EI) and ⟨k̃ξ^2⟩_c^ (EI) versus q_c^ (EI). (d) ⟨k̃^2 ξ⟩_c^ (EP) and ⟨k̃ξ^2⟩_c^ (EP) versus q_c^ (EP). (e) Relative magnitudes of the four third-order cumulants of local in-degrees and random noises in the EI network. (f) The same plot as (e) for the EP network. (g) The averages of the relative magnitudes of the third-order cumulants in the EI networks as functions of time. Errorbars represents the standard deviations. (h) The same plot as (g) for the EP networks. | 2012.15487 | Figure5.jpg |
0.429925 | b2ebe773c47b4a1bb882a4c69cac4b3f | Trade networks in 1970 and their global degree and link-weight distributions. (a) A subnetwork of the EI network in 1970, consisting of 30 countries having the largest (out+in)-degrees connected by 141 links of standardized local weight (logarithmic export value) larger than 1, i.e., ṽ> 1. (b) A subnetwork of the EP network of 15 countries having the largest out-degrees and 15 products having the largest in-degrees connected by 133 links of ṽ>1. (c) Global distribution of the logarithm of exports P^ (EI) (v) = ∑_c,c^' A_cc^'δ(v_cc^' - v) ∑_c,c^' A_cc^' in the EI network with A_cc^' the adjacency matrix. (d) Global distribution of the logarithm of exports P^ (EP) (v) in the EP network. (e) Global out-degree distribution P^ (EI)(q) = ∑_c δ(q_c^ (EI) - q) ∑_c 1 and the plot of the rank R_c^ (EI) versus the out-degree q_c^ (EI) of each exporter c in the EI network. (f) Global out-degree distribution P^ (EP)(q) and the plot of the rank R_c^ (EP) versus the out-degree q_c^ (EP) of exporter c in the EP network. (g) Global in-degree distribution P^ (EI)(k) = ∑_cδ(k_c^ (EI) - k) ∑_c 1 and the plot of the rank R_c^ (EI) versus the in-degree k_c^ (EI) of importer c in the EI network. (f) Global in-degree distribution P^ (EP)(k) = ∑_pδ(k_p^ (EP) - k) ∑_p 1 and the plot of the rank R_p^ (EP) versus the in-degree k_p^ (EP) of product p in the EP network. | 2012.15487 | FigureA1.jpg |
0.392696 | fbe984b57d9f4041ae01e00593e409c9 | Significance of skewness of local logarithmic exports. (a) The fraction of exporters whose true skewness remain in the middle x(%) range of the random-version skewness in 1980. Squares and circles are for the EI and the EP network. (b) The same plot as (a) in year 2010. | 2012.15487 | FigureA2.jpg |
0.471468 | 9d032832055946fcad62369335b7a8ae | Time-dependent correlation of the skewness with the average of local logarithmic exports and with the out-degree of exporter. The Pearson correlation coefficient between the skewness of local logarithmic exports γ_v;c^ (EI,EP) and the out-degree q_c^ (EI,EP) of exporter or the average local logarithmic export ⟨ v⟩_c^ (EI,EP) as a function of time t. | 2012.15487 | FigureA3.jpg |
0.371829 | 95438d76c5b04125941e4e2ea853326f | Plot of the skewness versus the average of local logarithmic exports. (a) Plot of the skewness of local logarithmic exports γ_v;c^ (EI) versus the average of local logarithmic exports ⟨ v⟩_c^ (EI) in the EI network in year 1980. (b) The same plot as (a) in the EP network. | 2012.15487 | FigureA4.jpg |
0.456209 | 5987e0c7097b493b9932534639b4baa1 | Correlation of skewness of local in-degrees with the out-degree of exporter and the average local in-degree. The Pearson correlation coefficient between the skewness of local in-degrees γ_k;c^ (EI,EP) and the out-degree q_c^ (EI,EP) of exporter or the average local in-degree ⟨ k⟩_c^ (EI,EP) as a function of time t. Correlations are observed robustly in the studied period, except for the correlations of γ_k;c^ (EP) with q_c^ (EP) and with ⟨ k⟩_c^ (EP) which change sign in the period 1984-1999 presumably owing to the virtual product categories added artificially [See Appendix <ref>]. | 2012.15487 | FigureA5.jpg |
0.465306 | 19719b62387848db948c255bb5fff11b | Robustness against time of the sign of the correlation coefficient and third-order cumulants. (a) The portions of exporters showing a non-negative correlation coefficient between local logarithmic exports and local in-degrees (ρ_c^ (EI)≥ 0), a non-negative cumulant ⟨k̃^2ξ⟩_c^ (EI)≥ 0, and another non-positive cumulant ⟨k̃ξ^2⟩_c^ (EI)≠ 0 of local in-degrees and random noises are shown respectively for the EI network as functions of time. (b) The same plots as (a) for the EP network. | 2012.15487 | FigureA6.jpg |
0.502931 | 119da1f22cc649a3b7f3ba13bca84b07 | Schematic distribution of {(k̃, ṽ)} corresponding to positive or negative third-order cumulants. Each panel shows an example of the distribution of the data points of standardized local exports and in-degrees {(k̃, ṽ)} exhibiting a given positive or negative cumulant. The solid line represents ṽ=ρk̃ with ρ=1 for simplicity. Therefore the deviation of the data points from the line represents the random nooise ξ. Note that ⟨k̃⟩=⟨ξ⟩=0 and ⟨k̃^2⟩ = ⟨ξ^2⟩=1. | 2012.15487 | FigureA7.jpg |
0.42001 | 840574c0ac474e179d2d4df246458f25 | An overview of framework TransRegex for regex synthesis. | 2012.15489 | framework5.jpg |