Datasets:
Dataset Viewer (First 5GB)
id
stringlengths 20
32
| text
stringlengths 100
121k
| image_path
stringlengths 15
235
|
|---|---|---|
arxivcap_1906.05427_1 | Title: (A) Data in the Life: Authorship Attribution of Lennon-McCartney Songs | Caption: Back-to-back histograms of the out-of-sample prediction probabilities of songs of known authorship. Bars to the left represent 39 songs or song portions known to be written by Lennon, and bars to the right represent 31 songs or song portions known to be written by McCartney. | 1906.05427_1.jpg |
arxivcap_1906.05427_2 | Title: (A) Data in the Life: Authorship Attribution of Lennon-McCartney Songs | Caption: ROC plot for out-of-sample song probability predictions based on 70 songs or song fragments with known authorship. | 1906.05427_2.jpg |
arxivcap_2209.11222_17 | Title: Concept Activation Regions: A Generalized Framework For Concept-Based Explanations | Caption: Linear separability implies concept smoothness but not the opposite. | 2209.11222_17.jpg |
arxivcap_1906.05255_1 | Title: A Simple Text Mining Approach for Ranking Pairwise Associations in Biomedical Applications | Caption: Visual example of KinderMiner, with contingency table and associated Fisher's Exact Test (FET) analysis of the key phrase "embryonic stem cell" and the target term "NANOG." Target terms are filtered by significance of co-occurrence with the key phrase and then sorted by the co-occurrence ratio. | 1906.05255_1.jpg |
arxivcap_2202.00211_1 | Title: GNNRank: Learning Global Rankings from Pairwise Comparisons via Directed Graph Neural Networks | Caption: Overview of GNNRank based on directed graph neural networks and the proximal gradient steps corresponding to Algo. <ref>: starting from an adjacency matrix $\mathbf{A}$ which encodes pairwise comparisons, input feature matrix $\mathbf{X}$ and embedding dimension $d,$ GNNRank first applies a directed graph neural network model to learn node embeddings $\mathbf{Z}$ for each competitor (node). Then it calculates the inner product or the similarity score with respect to a learnable vector to produce non-proximal outcomes for ranking scores ("innerproduct" or "dist"). Proximal variants start from a similarity matrix constructed from the learnable embeddings $\mathbf{Z}$, then utilize proximal gradient steps to output ranking scores. Depending on the initial guess score vector $\mathbf{r}'$, the proximal variants have names "proximal innerproduct", "proximal dist" or "proximal baseline". Ordering the scores in the score vector $\mathbf{r}$ induces the final ranking/ordering vector $\mathbf{R}\in\mathbb{R}^n.$ The loss function is applied to a variant's output score vector $\mathbf{r}$, given the input adjacency matrix $\mathbf{A},$ while the final evaluation is based on $\mathbf{R}$ and $\mathbf{A}.$ Red frames indicate trainable tensors/vectors/matrices. Grey squares correspond to fixed inputs. | 2202.00211_1.jpg |
arxivcap_1906.05246_2 | Title: Tensor train optimization for mathematical model of social networks | Caption: The exact and reconstruction functions $r(t)$ with parameters from the table <ref> and the relative error $E(r)$. | 1906.05246_2.jpg |
arxivcap_2303.14666_1 | Title: Generalization Matters: Loss Minima Flattening via Parameter Hybridization for Efficient Online Knowledge Distillation | Caption: The loss landscape visualization of four students (Ours-S1 and Ours-S2 are obtained by our method, and DML obtains DML-S1 and DML-S2), which are ResNet32 <cit.> trained by the same settings on CIFAR-10 <cit.>. Four students start from the initial point (Red points in the center) and converge to three basins along different trajectories. The x-axis and y-axis represent the values of model parameters that PCA <cit.> obtains. | 2303.14666_1.jpg |
arxivcap_1906.05246_4 | Title: Tensor train optimization for mathematical model of social networks | Caption: Reconstruction of the direct problem for model (<ref>) with parameters from the table <ref>. Solid width lines denote the real observations for the density of influenced users for a variety of time periods, dashed lines - the predicted density of influenced users. | 1906.05246_4.jpg |
arxivcap_2202.00276_1 | Title: Classical Tracking for Quantum Trajectories | Caption: Example section of a quantum trajectory (blue) and the classical state estimates (magenta), the classical state estimates are calculated at each time step but are only plotted every 100 time increments for clarity, parameter values are given in the text. | 2202.00276_1.jpg |
arxivcap_2202.00276_3 | Title: Classical Tracking for Quantum Trajectories | Caption: Three examples of Wigner quasi-probability distributions (left) and corresponding pdfs constructed from the particle weights (right) from one trajectory. The three examples show: classical-like state (top row); nonlinear quantum state with weak negative Wigner features (middle row); and nonlinear quantum state with strong negative Wigner features (bottom row). Nonlinearity $\gamma = 0.10$ with other parameter values given in the text. | 2202.00276_3.jpg |
arxivcap_1906.05491_1 | Title: A Computational Analysis of Natural Languages to Build a Sentence Structure Aware Artificial Neural Network | Caption: Languages tend to group together according to how similar their spellings are. The orange area highlights Germanic Languages where English (red arrow), whose spelling was influenced by Latin and French, is not included. Blue arrows show the interesting closeness of the Turkish language to the Basque Language. | 1906.05491_1.jpg |
arxivcap_2303.14666_3 | Title: Generalization Matters: Loss Minima Flattening via Parameter Hybridization for Efficient Online Knowledge Distillation | Caption: Framework of our OKDPH. Two students construct HWM by sampled convex parameter combinations in each training batch, and HWM's parameters are regularly fused with students. Two students and HWM's logits are obtained by feeding three types of data augmentations and are averaged to $\text{Logit}_{en}$. Each student's training loss consists of the classification loss $\mathcal{L}_{ce},\mathcal{L}_{ce}^{hwm}$ and the KD loss $\mathcal{L}_{kd}$. | 2303.14666_3.jpg |
arxivcap_2012.05423_8 | Title: Combining Transit and RV: A Synthesized Population Model | Caption: The number of accepted simulations with the <cit.> M-R relation and $F_{\text{rocky}} = 0.8$ as an an example, divided by the number with the <cit.> M-R relation (ratio of accepted simulations) as a function of acceptance distance threshold. Shown are results when calculating the overall distance (black), as well as the orbital period (blue), sample size (green), and mass (orange) distance components separately. Both models recover the period distribution and sample size of the observed RV dataset equally well, with roughly the same number of accepted simulations (ratio $\sim$ 1). However, the mass break component, and therefore the overall distance, show that the <cit.> model becomes increasingly favoured at small distances. Similar to Fig. <ref>, for distances less than $\sim 1$, it becomes more difficult for either model to be accepted. Only a few of the 100,000 draws satisfy the threshold for either model, so we do not believe the results in this regime are significant. | 2012.05423_8.jpg |
arxivcap_2303.14666_4 | Title: Generalization Matters: Loss Minima Flattening via Parameter Hybridization for Efficient Online Knowledge Distillation | Caption: The loss landscape visualization of three methods (Base, DML <cit.>, and KDCL <cit.> from left to right) compared with our method on two datasets (CIFAR-10 and CIFAR-100 <cit.> from top to bottom). Ours-S1 and Ours-S2 are the two students obtained by our method, and SOTA-S1 and SOTA-S2 are the two students obtained by other methods, both of which are ResNet32 <cit.> trained by the same settings. The x-axis and y-axis represent the values of model parameters by the PCA dimension reduction algorithm <cit.>. Each sub-diagram shows four students who start from the initial point (Red points in the center) and converge to three basins along different loss trajectories. | 2303.14666_4.jpg |
arxivcap_1906.05491_2 | Title: A Computational Analysis of Natural Languages to Build a Sentence Structure Aware Artificial Neural Network | Caption: Languages tend to group together according to how similar their structures are. Differently from Written Patterns Similarity, English structure make it group with Germanic languages (red arrow). Also in this case Basque and Turkish are close together. | 1906.05491_2.jpg |
arxivcap_2202.00276_4 | Title: Classical Tracking for Quantum Trajectories | Caption: Average values (with one standard deviation error bars) for KL divergences for the pdfs derived from the classical particle filter and the Wigner quasi-probability distributions for different values of the nonlinearity parameter $\gamma$, other parameter values are given in the text. | 2202.00276_4.jpg |
arxivcap_2012.05423_9 | Title: Combining Transit and RV: A Synthesized Population Model | Caption: ABC posterior distributions for the combined RV + transit fit. The transit component includes correction for reliability. Each dotted line indicates the 15.9th, 50th, and 84.1th percentiles. Note that all histograms span a subset of our prior range (see Table (<ref>)). | 2012.05423_9.jpg |
arxivcap_2303.14708_1 | Title: Exploring Multimodal Sentiment Analysis via CBAM Attention and Double-layer BiLSTM Architecture | Caption: Illustration of our Model overall framework diagram.To judge sentiment polarity, the proposed architecture employs supervised contrastive learning and a CNN-connected Transformer fusion. The proposed architecture adopts supervised comparative learning and transformer fusion of CNN and CBAM connections. Invariance occurs for graphic sample features when embeddings from the same category, such as $z_{1}$ and $z'_{1}$. Different categories of graphic features, such as $z_{1}$ and $z_{2}$, on the other hand, are far apart. | 2303.14708_1.jpg |
arxivcap_2202.00276_5 | Title: Classical Tracking for Quantum Trajectories | Caption: Average values (with one standard deviation error bars) for KL divergences for the pdfs derived from the classical particle filter and the Wigner quasi-probability distributions for different values of the measurement efficiency $\eta$, other parameter values are given in the text. | 2202.00276_5.jpg |
arxivcap_1912.05052_4 | Title: Stability of traveling waves in a driven Frenkel-Kontorova model | Caption: Floquet multipliers (dots) for traveling wave solutions at $c=0.1569$ (left panel) and $c=0.1572$ (right panel). The red curve marks the circle of $|\rho|=\exp[-\gamma/(2c)]$. Here $\mu=1$, $\gamma=0.01$, { and the multiplier $\rho=1$ is marked by a green dot}. | 1912.05052_4.jpg |
arxivcap_1906.05491_3 | Title: A Computational Analysis of Natural Languages to Build a Sentence Structure Aware Artificial Neural Network | Caption: The blue area shows Romance Languages with Greek. The orange area shows Germanic languages. The Red arrow shows the transition of English from Romance languages to Germanic languages as more sentence structure features are considered. | 1906.05491_3.jpg |
arxivcap_2303.14667_1 | Title: Constraining the nuclear symmetry energy with charge radii of the mirror pairs nuclei | Caption: (Color online) $\Delta{R}_{\mathrm{ch}}$ of the mirror partner nuclei $^{36}$Ca-$^{36}$S (a), $^{38}$Ca-$^{38}$Ar (b) and $^{54}$Ni-$^{54}$Fe (c) as a function of slope parameter $L$ at saturation density $\rho_{0}$. The experimental result is shown as a horizontal light blue band. The crosses are results of relativistic EDFs and the open circles are for the Skyrme EDFs calculations. The dashed lines indicate theoretical linear fits. | 2303.14667_1.jpg |
arxivcap_2202.00212_1 | Title: Strongly aperiodic SFTs on hyperbolic groups: where to find them and why we love them | Caption: Wang emulates the run of a Turing machine as a tiling problem. | 2202.00212_1.jpg |
arxivcap_1912.05052_7 | Title: Stability of traveling waves in a driven Frenkel-Kontorova model | Caption: Floquet multipliers (dots) for traveling wave solutions at velocities near the stability threshold $c \approx 0.157$ { corresponding to a local maximum of $\sigma(c)$}. Here $\mu=1$, $\gamma=0.01$, { and the multiplier $\rho=1$ is marked by a green dot}. | 1912.05052_7.jpg |
arxivcap_2012.05304_1 | Title: Competitive Simplicity for Multi-Task Learning for Real-Time Foggy Scene Understanding via Domain Adaptation | Caption: Segmentation and depth predictions on Foggy Zurich <cit.>, Foggy Driving <cit.> and Foggy Cityscapes <cit.> using our approach. | 2012.05304_1.jpg |
arxivcap_1906.05491_4 | Title: A Computational Analysis of Natural Languages to Build a Sentence Structure Aware Artificial Neural Network | Caption: Languages tend to group together according to both Written and Sentence Structure similarities. | 1906.05491_4.jpg |
arxivcap_2202.00258_1 | Title: Parking search in the physical world: Calculating the search time by leveraging physical and graph theoretical methods | Caption: Influence of the topology of the street network on (a,b) curbside and (c,d) off-street parking. A major destination (hot spot) is located at the extremity of a dead-end street in sketch (a) and at an intersection in sketch (b). Panel (c) displays two model off-street parking lots of $N=180$ spaces; the evolution of the search time $T_s$ in each, for drivers parking in the first vacant spot, is shown in panel (d) as a function of the average occupancy $\phi$. Inset: Relation between $\phi$ and the product of the injection rate $I$ with the parking time $D^{-1}$ [$D^{-1}=30\,\mathrm{min}$ for the triangles ($\triangle$), $60\,\mathrm{min}$ for the circles ($\bigcirc$)]. | 2202.00258_1.jpg |
arxivcap_2012.05304_2 | Title: Competitive Simplicity for Multi-Task Learning for Real-Time Foggy Scene Understanding via Domain Adaptation | Caption: A conceptual overview of our scene understanding approach via domain adaptation using <cit.> (Inference) and the detailed outline of the entire pipeline (Training). Our overall model consists of two main components: domain adaptation using <cit.> and an encoder-decoder sub-module for semantic segmentation and depth estimation. Foggy scenes from (domain $X$) are transformed to fine scenes (domain $Y$) and vice versa, resulting in $Y', X'$ (the desired domains), and cyclically mapping them back to their original domains, producing $X'', Y''$; $D_{X}, D_{Y}$: ground truth depths and $D', D''$: depth predictions; $S_{X}, S_{Y}$ semantic labels and $S', S''$: semantic segmentation predictions. | 2012.05304_2.jpg |
arxivcap_2303.14667_2 | Title: Constraining the nuclear symmetry energy with charge radii of the mirror pairs nuclei | Caption: (Color online) " Data-to-data" relation between $\Delta{R_{\mathrm{ch}}}$ of $A=36$, $38$ and $54$ mirror-partner nuclei and the neutron skin thickness $\Delta{R_{\mathrm{np}}}$ of $^{48}$Ca. The same marks and color coding are used as in Fig. <ref>. | 2303.14667_2.jpg |
arxivcap_2202.01566_10 | Title: Unified theory of atom-centered representations and message-passing machine-learning schemes | Caption: {Learning curves for the prediction of molecular dipole models using a linear equivariant model trained on QM7 structures (restricted to CHNO composition), using different types of representations. (top) Validation error on 800 hold-out QM7 structures. (bottom) Error for extrapolative prediction on 1000 larger molecules taken from QM9. } | 2202.01566_10.jpg |
arxivcap_1912.05052_8 | Title: Stability of traveling waves in a driven Frenkel-Kontorova model | Caption: { Floquet multipliers (dots) for traveling wave solutions at velocities near the stability threshold $c \approx 0.16919$ corresponding to a local minimum of $\sigma(c)$. Here $\mu=1$, $\gamma=0.01$, and $\rho=1$ is marked by a green dot. The velocity is below the threshold in the upper panel and above it in the two lower panels.} | 1912.05052_8.jpg |
arxivcap_2202.00258_2 | Title: Parking search in the physical world: Calculating the search time by leveraging physical and graph theoretical methods | Caption: Influence of the inhomogeneous attractiveness of parking spots in a small `toy' network. Cars have equal turning probabilities at each intersection. Numerically (a) and analytically (b) derived average occupancy of parking spaces (represented as squares) in the steady state, when drivers are most attracted to free spots ($p_i=100\%$) and only have a $p_i=1\%$ probability to park at every other vacant spot that they drive by. Panel (c) compares the resulting mean search times with the situation in which all spots are equally attractive ($p_i=100\%$). | 2202.00258_2.jpg |
arxivcap_1906.05323_2 | Title: Specifying Weight Priors in Bayesian Deep Neural Networks with Empirical Bayes | Caption: Accuracy vs Confidence curves: Networks trained on MNIST and tested on both MNIST and the NotMNIST (out-of-distribution) test sets. | 1906.05323_2.jpg |
arxivcap_2303.14667_3 | Title: Constraining the nuclear symmetry energy with charge radii of the mirror pairs nuclei | Caption: (Color online) Symmetry energy $E_{\mathrm{s}}$ and the slope of symmetry energy $L$ are limited by the $\Delta{R_{\mathrm{ch}}}$ of $A=36$ (yellow plane), $38$ (light purple panel) and $54$ (light blue plane) mirror-partner nuclei. The plane covered by the shadow slash represents the result of theoretical prediction in this work. | 2303.14667_3.jpg |
arxivcap_2202.00258_3 | Title: Parking search in the physical world: Calculating the search time by leveraging physical and graph theoretical methods | Caption: Map of the city of Lyon with its 84k on-street parking spots (as of 2019) and the injection points and destinations implemented in our model (the symbol sizes represent the injection rates of cars associated with these points). | 2202.00258_3.jpg |
arxivcap_2012.05304_3 | Title: Competitive Simplicity for Multi-Task Learning for Real-Time Foggy Scene Understanding via Domain Adaptation | Caption: A detailed outline of the encoder-decoder architecture for semantic segmentation and depth estimation. The network consists of two sub-encoders taking two types of inputs: RGB and luminance L and/or depth D images (depending on the task); two decoders and two discriminators <cit.> for semantic segmentation and depth estimation. | 2012.05304_3.jpg |
arxivcap_1906.05323_4 | Title: Specifying Weight Priors in Bayesian Deep Neural Networks with Empirical Bayes | Caption: Benchmarking MOPED_MFVI with state-of-art Bayesian deep learning techniques on diabetic retinopathy diagnosis task using BDL-benchmarks. Accuracy and area under the receiver-operating characteristic curve (AUC-ROC) plots for varied percentage of retained data based on predictive uncertainty. MOPED_MFVI performs better than the other baselines from BDL-benchmarks suite. Shading shows the standard error. | 1906.05323_4.jpg |
arxivcap_1912.05052_9 | Title: Stability of traveling waves in a driven Frenkel-Kontorova model | Caption: (a) displacement $q_n$ and (b) particle velocity $p_n$ for the unstable eigenmode at $c=0.16$ corresponding to $\rho=1.2591$ and (c) $\sigma(c)$ curve with the initial unstable (U) and final stable (S) states marked. The stable wave was obtained by solving (<ref>) with the initial conditions given by the unstable traveling wave solution perturbed along the unstable eigendirection (with perturbation amplitude $0.01$). It has velocity $c_f=0.1562$ and the same $\sigma=0.0039$ as the unstable wave. Here $\mu=1$ and $\gamma=0.01$. | 1912.05052_9.jpg |
arxivcap_2202.01442_1 | Title: PERSPECTIVE: Emergent phases in rare earth nickelate heterostructures | Caption: Ni-O bond lengths (left axis) and phase transition temperatures (right axis) have been plotted as a function of Ni-O-Ni bond angle of bulk $RE$NiO$_3$ series. All data have been adapted from Ref. {Zhou:2004p153105}. | 2202.01442_1.jpg |
arxivcap_2212.13206_4 | Title: Efficient Graph Reconstruction and Representation Using Augmented Persistence Diagrams | Caption: The splitting of edge arc $\eAVar$ into $\eAVar_{\ell}$ and $\eAVar_r$, as in {split-arc}. The large gray region is the region containing all edges of $bigedges$. That is, all edges whose angle with the positive $e_1$-axis is at least $\eAVar.\alpha_1$. On {split:countell} of the algorithm, we compute the number of edges in $\eAVar_{\ell}$ by first computing the indegree of $\eAVar.\thevert$ in direction $s$ from the diagram in direction $s$, then we subtract the number of edges in $bigedges$ that are below the height of $\eAVar.\thevert$ in direction $\dir$ (i.e., below the blue line). By the pigeonhole principal, we find $\eAVar_r.\eC = \eAVar.\eC-\eAVar_{\ell}.\eC$. | 2212.13206_4.jpg |
arxivcap_2303.14667_4 | Title: Constraining the nuclear symmetry energy with charge radii of the mirror pairs nuclei | Caption: (Color online) Comparison between the values of $L$ extracted in this work and those from the existing literature. We partly compare the values extracted from various models: Carbone $et~al$. <cit.>, Chen $et~al$. <cit.>, Steiner $et~al$. <cit.>, Roca-Maza $et~al$. <cit.>, Zhang $et~al$ <cit.>, Mondal $et~al$. <cit.>, Raithel $et~al$. <cit.>, Brown $et~al$. <cit.>, Malik $et~al$. <cit.>, Zhang $et~al$. <cit.>, Pineda $et~al$. <cit.>, Newton $et~al$. <cit.>, Tagami, $et~al$. <cit.>. | 2303.14667_4.jpg |
arxivcap_2202.00258_5 | Title: Parking search in the physical world: Calculating the search time by leveraging physical and graph theoretical methods | Caption: Dependence of the mean total driving time, including the curbside parking search time, for two categories of drivers (i.e., two destinations, irrespective of the entry point) on the global injection rate. Cyan-filled symbols represent destination 12 (Ainay), whereas yellow-filled symbols represent destination 18 (Part-Dieu). | 2202.00258_5.jpg |
arxivcap_2012.05339_1 | Title: Neural Rate Control for Video Encoding using Imitation Learning | Caption: Overview of our method. (a) A policy network (<ref>) is trained on a teacher dataset generated by evolution strategies (<ref>) and refined by hindsight experience replay (<ref>). (b) At inference, low-probability QPs are truncated (<ref>), and the policy is augmented by feedback control to have precise bitrate control (<ref>). | 2012.05339_1.jpg |
arxivcap_1906.05323_5 | Title: Specifying Weight Priors in Bayesian Deep Neural Networks with Empirical Bayes | Caption: Density histograms obtained from in- and out-of-distribution samples. Bayesian DNN model uncertainty estimates indicate higher uncertainty for out-of-distribution samples as compared to the in-distribution samples. | 1906.05323_5.jpg |
arxivcap_1912.05052_10 | Title: Stability of traveling waves in a driven Frenkel-Kontorova model | Caption: Same as the previous figure but now for $c=0.0801$ and for the unstable eigendirection with $\rho=1.0984$. As a result of the perturbation, in this case the evolution dynamics selects to move to the right stable branch portion corresponding to the higher speed $c_f=0.0838$ for the same value of $\sigma=0.0012$. | 1912.05052_10.jpg |
arxivcap_2202.01442_3 | Title: PERSPECTIVE: Emergent phases in rare earth nickelate heterostructures | Caption: (a) Schematics of deposition sequence for n uc ENO/ n uc LNO superlattice. Cyan and orange color represent ENO and LNO, respectively. (b) Temperature dependent electrical resistivity of n ENO/ n LNO SLs. (c) $L$ scan around (1/4 1/4 1/4) resonant soft x-ray magnetic reflection with the photon energy tuned to the Ni $L_3$ edge for the n=4 SL. The absence of any satellite along $L$ signifies no magnetic contrast between ENO and LNO layers i.e. the $E^\prime$-type antiferromagnetic spin arrangement is present throughout the entire film. $L$ scans for the n=4 sample through (d) (1/2 1/2 1/2) and (e) (-1/2 1/2 1/2) reflections. The (1/2 1/2 1/2) peak is allowed for both orthorhombic and monoclinic symmetry but forbidden for rhombohedral symmetry. The absence of any satellite established no significant structural contrast between ENO and LNO. The (-1/2 1/2 1/2) reflection is allowed for monoclinic and forbidden for both orthorhombic & rhombohedral symmetry. The presence of a satellite peak in this case, signifies contrast in the breathing mode distortion between the ENO and LNO layers. LNO layers have become orthorhombic without (or much suppressed compared to ENO) breathing mode distortion. The data shown in panel (b)-(e) have been adapted from Refs. <cit.>). | 2202.01442_3.jpg |
arxivcap_2303.14766_1 | Title: Detection Rate of <50-meter Interstellar Objects with LSST | Caption: The maximum distance at which an ISO with an albedo of 0.06 can be detected as a function of the phase angle between the Sun, object, and Earth for various ISO diameters. | 2303.14766_1.jpg |
arxivcap_2212.13249_1 | Title: Statistical equilibrium principles in 2D fluid flow: from geophysical fluids to the solar tachocline | Caption: Strip and annular (or disc if $R_2 = 0$) geometries for which, respectively, a conserved linear momentum (<ref>) or angular momentum (<ref>) exists. The strip has periodic boundary conditions along $x$ and Dirichlet boundary conditions on the lower and upper boundaries $\Gamma_{1,2}$. The annulus has Dirichlet boundary conditions on both boundaries. The latter lead to two independent circulation integrals (<ref>) for each domain (which are seen to actually have the same topology). | 2212.13249_1.jpg |
arxivcap_2202.00258_6 | Title: Parking search in the physical world: Calculating the search time by leveraging physical and graph theoretical methods | Caption: Map of Lyon comparing the the simulated mean total driving time and its analytically derived counterpart for all possible destinations (drivers' category), irrespective of the entry point, for a global injection rate of 24 cars/min. | 2202.00258_6.jpg |
arxivcap_1906.05472_3 | Title: On Feasibility and Flexibility Operating Regions of Virtual Power Plants and TSO/DSO interfaces | Caption: Actual (red) and assumed (grey) PQ capabilities of resources considered in this study. | 1906.05472_3.jpg |
arxivcap_2303.14766_2 | Title: Detection Rate of <50-meter Interstellar Objects with LSST | Caption: Cross-sectional area of the detection region for ISOs with an albedo of 0.06 in the ecliptic plane for various diameters. The plot is centered on Earth where the radial direction (x-axis) points away from the Sun and the perpendicular direction (y-axis) is tangential to the Earth's position vector relative to the Sun. | 2303.14766_2.jpg |
arxivcap_2303.14766_3 | Title: Detection Rate of <50-meter Interstellar Objects with LSST | Caption: The probability per logarithmic bin in velocity, where $v$ is the velocity relative to the Earth for the sample population of ISOs. The minimum velocity relative to the Earth is equivalent to the escape speed from the Earth. We assume the velocity distribution is ISOs is similar to local stars around the LSR and is independent of other variables, such as albedo or diameter. | 2303.14766_3.jpg |
arxivcap_2212.13249_2 | Title: Statistical equilibrium principles in 2D fluid flow: from geophysical fluids to the solar tachocline | Caption: Notional illustration of the equilibrium state of an overall neutral set of point-like vortices or charges as a function of internal energy <cit.>. The low energy state on the left corresponds to a molecular dipole state with strongly bound charges. The middle state corresponds to a higher energy plasma-like state with unbounded charges but that continue to obey local charge neutrality. The state on the right exhibits large scale structure obtained by increasing the energy even further, forcing the charges to segregate into separate nonneutral regions. This negative temperature state is accessible in fluid dynamics because the charges are not conventional momentum and kinetic energy carrying particles. In the vortex field description, charges carry only potential energy of interaction. | 2212.13249_2.jpg |
arxivcap_1912.05052_15 | Title: Stability of traveling waves in a driven Frenkel-Kontorova model | Caption: { Displacement profiles $u_n(0)=\phi(n)$ (a) along the first branch at $c=\hat{c}_1$; (b) along the second, third and fourth branches, at $c=\hat{c}_2$, $c=\hat{c}_3$ and $c=\hat{c}_4$, respectively. Inset in (b) shows the profile along the fourth branch with the equilibrium state $3\pi-\text{arcsin}(\sigma)$ marked by the dashed line for the corresponding $\sigma \approx 0.54$. Here $\mu=1$ and $\gamma=0.1$.} | 1912.05052_15.jpg |
arxivcap_2012.05339_5 | Title: Neural Rate Control for Video Encoding using Imitation Learning | Caption: Ablation of HER and feedback control. Results of ES and 's policy are included as baselines. | 2012.05339_5.jpg |
arxivcap_1906.05472_5 | Title: On Feasibility and Flexibility Operating Regions of Virtual Power Plants and TSO/DSO interfaces | Caption: VPP feasibility regions (FOR) in Case I, along with the resulting potential regions for FCAS raise (green) and lower (red) services. | 1906.05472_5.jpg |
arxivcap_2202.00258_7 | Title: Parking search in the physical world: Calculating the search time by leveraging physical and graph theoretical methods | Caption: Variations of the driving and cruising time in the `toy' network of Fig. <ref>(a) (with free spots) with the car injection rate. The outcome of the simulation (circles) is compared with the analytical predictions (triangles), both in the case of unbound search times and when search times are capped to $180\,\mathrm{s}$. | 2202.00258_7.jpg |
arxivcap_1702.06296_8 | Title: Gaia eclipsing binary and multiple systems. Supervised classification and self-organizing maps | Caption: Composition of ANN-estimated system morphology classes versus the LD1 scores (top panel) and versus the originally estimated morphology parameter (bottom panel) on the {Kepler} eclipsing binaries. For ease of comparison, the order of the LD1 coefficients is reversed. | 1702.06296_8.jpg |
arxivcap_2303.14766_5 | Title: Detection Rate of <50-meter Interstellar Objects with LSST | Caption: The probability per logarithmic bin in chord length, where $l$ is the chord length through the observable region, for the distribution of chord lengths for ISO populations with various diameters. Two albedo distributions are depicted: the albedo distribution from equation (2) on the left and a population with a constant albedo of 0.06 on the right. | 2303.14766_5.jpg |
arxivcap_2012.05339_7 | Title: Neural Rate Control for Video Encoding using Imitation Learning | Caption: Illustrative example of the rate control problem and the metrics for encoding efficiency. | 2012.05339_7.jpg |
arxivcap_1906.05472_7 | Title: On Feasibility and Flexibility Operating Regions of Virtual Power Plants and TSO/DSO interfaces | Caption: Case I VPP's potential for FCAS participation resulting from the dispatch point $S^\lambda$. | 1906.05472_7.jpg |
arxivcap_2212.13249_3 | Title: Statistical equilibrium principles in 2D fluid flow: from geophysical fluids to the solar tachocline | Caption: Highly schematic illustration of the turbulent mixing process that begins here with a well defined though irregular region of finite, fixed vorticity $\omega = q_0$, surrounded by a vorticity free (potential flow) region, $\omega = 0$. Over time the vortex region stretches and folds to give rise as $t \to \infty$ to a fully mixed smoothly varying macroscale steady state. However, the macro-view obscures the continuing microscale dynamics (illustrated in Fig. <ref>) where restriction to values $\omega = 0,q$ is preserved, consistent with the Casimir constraints. | 2212.13249_3.jpg |
arxivcap_2202.01442_4 | Title: PERSPECTIVE: Emergent phases in rare earth nickelate heterostructures | Caption: (a) and (b) Charge transfer between metallic LaNiO$_3$ and insulating LaTiO$_3$ resulting in the formation of a new Mott state at the interface. Adapted from Ref. {Cao:2016p10418} | 2202.01442_4.jpg |
arxivcap_1912.05027_6 | Title: SpineNet: Learning Scale-Permuted Backbone for Recognition and Localization | Caption: Building scale-permuted network by permuting ResNet. From (a) to (d), the computation gradually shifts from ResNet-FPN to scale-permuted networks. (a) The R50-FPN model, spending most computation in ResNet-50 followed by a FPN, achieves 37.8% AP; (b) R23-SP30, investing 7 blocks in a ResNet and 10 blocks in a scale-permuted network, achieves 39.6% AP; (c) R0-SP53, investing all blocks in a scale-permuted network, achieves 40.7% AP; (d) The SpineNet-49 architecture achieves 40.8% AP with 10% fewer FLOPs (85.4B 95.2B) by learning additional block adjustments. Rectangle block represent bottleneck block and diamond block represent residual block. Output blocks are indicated by red border. | 1912.05027_6.jpg |
arxivcap_1906.05497_1 | Title: Deep Network Approximation Characterized by Number of Neurons | Caption: A summary of existing and our new results on the approximation rate of ReLU FNNs for continuous functions. Existing results <cit.> are applicable in the areas in {color1}, {color2}, and {color3}; our new result is suitable for almost all areas when $L\geq 2$. | 1906.05497_1.jpg |
arxivcap_2303.14766_6 | Title: Detection Rate of <50-meter Interstellar Objects with LSST | Caption: Probability density functions of albedo distributions of entire sample interstellar object population (left) and detected interstellar object populations for various diameters (right). The populations are constructed from albedo distributions which resemble that of near-earth asteroids. | 2303.14766_6.jpg |
arxivcap_1702.06296_9 | Title: Gaia eclipsing binary and multiple systems. Supervised classification and self-organizing maps | Caption: Relation between the LD1 coefficient and the morphology parameter (large panel) and a few examples from the upper left region. Green circles: detached, violet diamonds: semi-detached (including near-contact and contact, cf. Section <ref>), red triangles: overcontact systems. | 1702.06296_9.jpg |
arxivcap_2202.01562_3 | Title: Doubly Robust Off-Policy Evaluation for Ranking Policies under the Cascade Behavior Model | Caption: Comparison of user behavior assumptions and Markov Decision Process | 2202.01562_3.jpg |
arxivcap_2212.13249_4 | Title: Statistical equilibrium principles in 2D fluid flow: from geophysical fluids to the solar tachocline | Caption: Illustration of separation of scales entering the exact thermodynamic solution. The vortex self-advection is dominated by the large-scale flow, while the small scale fluctuations asymptotically obey a simple $a$-cell permutation rule generating the microscale entropy (<ref>) characterizing each intermediate scale $l$-cell. Within each $l$-cell one may define the local vorticity distribution $n_0({\bf r}_l,\sigma)$ which has a well defined continuum limit $a,l \to 0$ but in such a way that $l/a \to \infty$. Its first moment defines the equilibrium vorticity (<ref>) and its area integral is constrained by the Casimir function (<ref>). This illustrates the formal limiting process by which, e.g., a discrete set of ($a$-scale) vorticity levels controlled by the Casimirs produces a smooth ($l$-scale) average. | 2212.13249_4.jpg |
arxivcap_2202.00243_1 | Title: Adversarial Imitation Learning from Video using a State Observer | Caption: State Observer Analysis Average $L^2$ norm of proprioceptive state prediction error of the state observer on the expert's demonstration sequence, per update iteration. We see that in all environments, the error has a downward trend which shows that as the learning progresses, the state observer tends towards predicting the true proprioceptive states of the demonstrator (unseen data) as well as the imitator (training data). | 2202.00243_1.jpg |
arxivcap_2202.01562_4 | Title: Doubly Robust Off-Policy Evaluation for Ranking Policies under the Cascade Behavior Model | Caption: Estimators' performance comparison with different data size $n$ Estimators' performance comparison with different slate size $L$ Estimators' performance comparison with different policy similarity $\lambda$ | 2202.01562_4.jpg |
arxivcap_1906.05497_4 | Title: Deep Network Approximation Characterized by Number of Neurons | Caption: An illustration of $f$, $f_p$, $\phi$, $x_\beta$, $Q_{\beta}$, and the trifling region $\Omega([0,1]^d,K,\delta)$ in the one-dimensional case for $\beta\in \{0,1,\cdots,K-1\}^d$, where $K=N^2L^2$ and $d=1$ with $N=2$ and $L=2$. $f$ is the target function; $f_p$ is the piecewise constant function approximating $f$; $\phi$ is a function, implemented by a ReLU FNN, approximating $f$; and $x_\beta$ is a representative of $Q_\beta$. The measure of the trifling region $\Omega([0,1]^d,K,\delta)$ can be arbitrarily small as we shall see in the proof of Theorem <ref>. | 1906.05497_4.jpg |
arxivcap_2303.14766_7 | Title: Detection Rate of <50-meter Interstellar Objects with LSST | Caption: Detection rates of ISOs of various diameters given the limiting magnitude (higher numbers are fainter) at which an object can be detected. LSST has a minimum magnitude of $m=24$ (shown in green). Our plot shows estimates for the sample population where all objects have an albedo of 0.06. | 2303.14766_7.jpg |
arxivcap_2202.00243_2 | Title: Adversarial Imitation Learning from Video using a State Observer | Caption: Demonstrating the need for more sample efficiency and improved performance in imitation learning from video-only demonstrations. Here we show the learning curves of two representative algorithms, - that learns to imitate an expert with privileged access to proprioceptive-state-only demonstrations and - that learns to imitate an expert from video-only demonstrations, without access to expert's proprioceptive state information. The x-axis shows number of timesteps of interactions for the imitator agents with the environment and the y-axis shows the task reward function from OpenAI gym <cit.> used only for evaluation. Evident from the performance gap in their respective learning curves, we notice that is more sample efficient than in imitation learning. In this work, we seek to improve both sample efficiency and imitation learning performance by proposing a new algorithm called . | 2202.00243_2.jpg |
arxivcap_1104.3616_3 | Title: Strategies used as spectroscopy of financial markets reveal new stylized facts | Caption: Performance comparison of strategic trading (real data) and random trading using the average values of return $R$ versus trading frequency $J$. We exclude the sell transactions without any preceding matching buys. The simulations for the random strategies are repeated for 2000 times. We show the results for individuals in A-share (a), institutions in A-share (b), individuals in B-share (c) and institutions in B-share (d), respectively. In each plot, the colorful symbols ({{$\circ$}}, {{$\vartriangle$}}, {{$\triangledown$}}) correspond to strategic trading, the continuous lines correspond to random trading, and the dashed line indicates the base line of zero return ($R=0$). | 1104.3616_3.jpg |
arxivcap_2012.05446_1 | Title: Visual Perception Generalization for Vision-and-Language Navigation via Meta-Learning | Caption: Illustration of visual perception generalization problem. In the VLN task, the agent usually uses a fixed configuration camera, such as 1.5 meters high, panoramic view or 90 degree HFOV. However, in real life, robots with different functions and different forms apparently have very distinct camera configurations. The discrepancy in visual information obtained by different visual perceptions is very obvious, which makes it impossible to directly use the learned navigation skills on other robots. | 2012.05446_1.jpg |
arxivcap_2212.13249_5 | Title: Statistical equilibrium principles in 2D fluid flow: from geophysical fluids to the solar tachocline | Caption: Schematic illustration of the entropy function $S(E)$ associated with the two level system (<ref>), and also of the point vortex system pictured in Fig. <ref>. As described in the text, the Casimir constraints on the vorticity allow for both positive and negative temperatures, and corresponding entropy limited to a finite energy interval, vanishing with infinite slope at both ends. This general picture will hold for any $g(\sigma)$ with bounded support. The dashed line corresponds to conventional particle systems in which the momentum degree of freedom can absorb unbounded energy. | 2212.13249_5.jpg |
arxivcap_2303.14737_1 | Title: Growing Convex Collision-Free Regions in Configuration Space using Nonlinear Programming | Caption: Three configuration that lie in the same convex collision-free region of configuration space generated by IRIS-NP. Convex regions that contain both approaching grasp and grasping configurations allow users to plan all the way up to grasp efficiently, without requiring the heuristic pre-grasp configuration used by other planners. | 2303.14737_1.jpg |
arxivcap_1702.06296_13 | Title: Gaia eclipsing binary and multiple systems. Supervised classification and self-organizing maps | Caption: Two examples, AA UMa (period: 0.4681258 days) and V346 Cen (period: 6.3219350 days) for the successive time series preprocessing steps (the black dots indicate the data taken from CALEB). The magnitudes for AA UMa are $V$ magnitudes from <cit.>, for V346 Cen, StroΜmgren $y$ from <cit.>. Green solid line: the double Gaussian fit, blue solid line: double Gaussian + smooth spline fit, dashed lines: principal component reconstructions up to different orders. Gold: PC1-PC4 (95% of the total variation), orange: PC1-PC11 (99%), red: PC1-PC21 (99.9%), violetred: PC1-PC36 (99.999%). The light curves reconstructed from the PCA decomposition are scaled back from between [0,1] to the original scale of the data. | 1702.06296_13.jpg |
arxivcap_1906.05497_5 | Title: Deep Network Approximation Characterized by Number of Neurons | Caption: An illustration of the desired function $\phi=\phi_2\circ\bmPhi_1$. Note that $\phi\approx f$ on $[0,1]^d\backslash \Omega([0,1]^d,K,\delta)$, since $\phi(\bmx)=\phi_2\circ\bmPhi_1(\bmx)=\phi_2(\bmbeta)\approx f(\bmx_\bmbeta)$ for any $\bmx\in Q_\bmbeta$ and each $\bmbeta\in \{0,1,\cdots,K-1\}^d$. | 1906.05497_5.jpg |
arxivcap_2303.14737_2 | Title: Growing Convex Collision-Free Regions in Configuration Space using Nonlinear Programming | Caption: The counterexample search consists of finding the first configuration on a uniform expansion of the ellipse that results in collision. | 2303.14737_2.jpg |
arxivcap_2212.13249_8 | Title: Statistical equilibrium principles in 2D fluid flow: from geophysical fluids to the solar tachocline | Caption: Example numerically generated long-time (near-equilibrium) behavior of freely decaying 2D magnetohydrodynamics on the sphere. The zonal velocity field (above) and zonal magnetic field (below) undergo coupled dynamics according to (<ref>), reducing to (<ref>) and (<ref>) in the 2D solar tachocline model <cit.>. In the $(\psi,A)$ representation (<ref>) of the statistical functional, where ${\bf v}$ is defined by the level curves of $\psi$ and ${\bf B}$ is defined by the level curves of $A$, the model is that of two gradient-coupled membranes in an external potential which, among other things, tends to preferentially align the two vector fields. | 2212.13249_8.jpg |
arxivcap_2202.00243_3 | Title: Adversarial Imitation Learning from Video using a State Observer | Caption: A diagrammatic representation of the algorithm proposed in this work. Different from existing methods, we propose here to utilize a novel state observer module, which serves to simplify discriminator training. More specifically, the state observer learns to map high-dimensional visual observations to low-dimensional proprioceptive states of the agent. While traditional methods like directly utilize the high-dimensional observations in optimizing the discriminator's objective, in this work, we instead use the low-dimensional proprioceptive state predictions of the state observer to optimize the discriminator objective. | 2202.00243_3.jpg |
arxivcap_1906.05497_9 | Title: Deep Network Approximation Characterized by Number of Neurons | Caption: A illustration of the desired network based on Equation (<ref>) and (<ref>), and the fact $x=\sigma(x)-\sigma(-x)$ for any $x\in \R$. We omit the activation function ($\sigma$) if the input is non-negative. | 1906.05497_9.jpg |
arxivcap_2012.05446_2 | Title: Visual Perception Generalization for Vision-and-Language Navigation via Meta-Learning | Caption: Framework of the VLN model. We divide the complete VLN model into three modules: (1) a visual perception module; (2) a language understanding module; (3) a navigation reasoning module, where the upper right represents a simple sequence-to-sequence navigator, and the lower right represents a more complex cross-modal attention navigation. Our visual perception generalization strategy only modifies the visual perception module and fixes the other two parts. | 2012.05446_2.jpg |
arxivcap_2303.14737_3 | Title: Growing Convex Collision-Free Regions in Configuration Space using Nonlinear Programming | Caption: To separate obstacles that are concave in configuration space from the interior of the polytope with finitely many hyperplanes, the configuration-space margin backs the hyperplane away from the surface of the obstacle. Increasing this margin reduces the number of faces in the final polytope at the expense of a more conservative region. | 2303.14737_3.jpg |
arxivcap_2202.00243_5 | Title: Adversarial Imitation Learning from Video using a State Observer | Caption: Performance of different algorithms with varied numbers of expert demonstration trajectories, trained for two million timesteps of interaction with the environment. We see that <cit.> performs the best due to its privileged access to expert's proprioceptive state information. , the algorithm introduced in this work, outperforms <cit.> and <cit.> on all six environments and achieves performance similar to in InvertedPendulum, InvertedDoublePendulum and Hopper, without access to the expert's true proprioceptive states. | 2202.00243_5.jpg |
arxivcap_1906.05497_11 | Title: Deep Network Approximation Characterized by Number of Neurons | Caption: A illustration of the target ReLU FNN implementing $\phi$ to output $\sum_{j=1}^{L} z_{j,\ell}=\sum_{j=1}^{\ell}\theta_j=\phi(\xi_1,\ell)$ given the input $(\xi_1,\ell)=(\bin 0.\theta_1\theta_2\cdots \theta_L,\ell)$ for $\ell\in \{1,2,\cdots,L\}$ and $\theta_1,\theta_2,\cdots,\theta_L\in\{0,1\}$. The construction is mainly based on Equation (<ref>), (<ref>), (<ref>), and (<ref>). The numbers above the architecture indicate the order of hidden layers. It builds a whole iteration step for every two layers. We output both $\sigma(\ell-j)$ and $\sigma(j-\ell)$ in the hidden layers for $j=1,2,\cdots,L$ because of the fact $x=\sigma(x)-\sigma(-x)$ for any $x\in\R$. We omit the activation function ($\sigma$) if the input of a neuron is non-negative. Note that all parameters of this network are essentially determined by Equation (<ref>) and (<ref>), which are valid no matter what $\theta_1,\theta_2,\cdots,\theta_L\in \{0,1\}$ are. Thus, the desired function $\phi$ implemented by this network is independent of $\theta_1,\theta_2,\cdots,\theta_L\in \{0,1\}$. | 1906.05497_11.jpg |
arxivcap_2108.09597_5 | Title: Hierarchical Summarization for Longform Spoken Dialog | Caption: Summarization Generation Pipeline. Our system enables the conversion of audio files to multiple tiers of summarization. In the first stage, we convert the audio file into a speaker-segmented and punctuated transcript and process the transcript, split by speaker turns. In the second stage, we take each speaker turn and cluster conceptually similar summaries via semantic segmentation. Each cluster's summaries are joined (concatenated) based off of semantic similarity. We remove small summarizations and then repeat the summarization and merging process to obtain the Medium and Short summaries. | 2108.09597_5.jpg |
arxivcap_2303.14737_4 | Title: Growing Convex Collision-Free Regions in Configuration Space using Nonlinear Programming | Caption: The environment used for testing the collision pair ordering. The arm is a seven degree of freedom KUKA LBR iiwa with welded gripper, in a forest of randomly placed pillars. | 2303.14737_4.jpg |
arxivcap_1702.06296_14 | Title: Gaia eclipsing binary and multiple systems. Supervised classification and self-organizing maps | Caption: Error rates of the ensemble members of Random forest (left) and LDA (right) for the classification into classes EA/EB/EW using Hipparcos, as a function of the number of light curve shape attributes. The ensemble median global error rate is plotted in black (with superimposed black squares), and the range of its (0.1, 0.9) quantiles in grey, highlighted by a thin black line. The same quantities on the three classes are shown in orange (EA), blue (EB) and pink (EW). The models contain period and peak-to-peak amplitude beside the PCs, whose number is indicated in the x-axis of the plot. | 1702.06296_14.jpg |
arxivcap_2012.05446_3 | Title: Visual Perception Generalization for Vision-and-Language Navigation via Meta-Learning | Caption: Overview of the generalization with affine transformation layers. For the pseudo-seen task, we add the AT layers to the two ResNet50 networks to imitate the visual features of the image information obtained by agents with different configuration sensors and update the parameters $ \theta $, $ \mu $ of the ResNets. For the pseudo-unseen task, we remove the AT layers to measure the generalization performance of the visual perception module, and update the AT parameters $ \theta_f $, $ \mu_f $. | 2012.05446_3.jpg |
arxivcap_2202.00243_6 | Title: Adversarial Imitation Learning from Video using a State Observer | Caption: Learning curves of the three adversarial algorithms , and (ours). The learning curves here show that has better sample efficiency than in imitation learning from video-only demonstrations and achieves performance close to the state-of-the-art algorithm that has privileged access to expert's proprioceptive states. The x-axis shows timesteps of interactions of the agents with the environment and the y-axis shows the task reward from OpenAI gym <cit.> (used only for evaluation). | 2202.00243_6.jpg |
arxivcap_1912.05088_1 | Title: An Introduction to Complex Systems Science and its Applications | Caption: From ref. <cit.>. Each column contains three examples of systems consisting of the same components (from left to right: molecules, cells, people) but with different relations between them. Each row contains systems representing a certain kind of relationship between components. For random systems, the behavior of each component is independent from the behavior of all other components. For coherent systems, all components exhibit the same behavior; for example, the behavior (location, orientation, and velocity) of one part of the cannonball completely determines the behavior of the other parts. Correlated systems lie between these two extremes, such that the behaviors of the system's components do depend on one another, but not so strongly that every component acts in the same way; for example, the shape of one part of a snowflake is correlated with but does not completely determine the shape of the other parts. (Implicit in these descriptions is the necessity of specifying the set of behaviors under consideration, as discussed in <ref>.) | 1912.05088_1.jpg |
arxivcap_2303.14737_5 | Title: Growing Convex Collision-Free Regions in Configuration Space using Nonlinear Programming | Caption: As the number of consecutive infeasible counterexample searches required before continuing increases, the percent of samples within the region that are in collision heads to zero. This trend allows the user to trade off the run time of region generation versus probabilistic certification of the region. Note that this specific environment and seed configuration was explicitly designed to make it very difficult for IRIS-NP to generate completely collision-free regions. In most practical environments, we find that a single infeasible sample is sufficient to eliminate all collisions from the region. | 2303.14737_5.jpg |
arxivcap_1702.06296_15 | Title: Gaia eclipsing binary and multiple systems. Supervised classification and self-organizing maps | Caption: Error rates of the Random forest classifier using the {randomized} training sets in the light curve morphology classification using Hipparcos. The error rates are measured on objects only in the test sets for each training set/test set partition. Grey histograms: period, amplitude, all PCs; blue: period, amplitude, PC1-PC8 (best performing model). | 1702.06296_15.jpg |
arxivcap_2202.00274_1 | Title: Infinite ergodic theory for three heterogeneous stochastic models with application to subrecoil laser cooling | Caption: A typical trajectory of momentum $p(t)$ in the HRW model, where $R(p)=|p|^2$, $p_0=1$, $p_{\max}=2$, $\sigma^2=0.01$, and $p_{\rm trap} \cong 0.035$ is shown for reference. The blue and the yellow region are the trapping and the recycling region, respectively. The inset is a schematic illustration of the jump rate $R(p)$. | 2202.00274_1.jpg |
arxivcap_2212.13249_9 | Title: Statistical equilibrium principles in 2D fluid flow: from geophysical fluids to the solar tachocline | Caption: Example equilibrium vorticity profiles $\omega_0({\bf r})$ for the two level system on the unit disk for a sequence of inverse temperatures $-\infty \leq \beta \leq \infty$, obtained by numerically solving the nonlinear Laplace equation (<ref>). Vorticity level $q = 1$ occupies fractional area $\alpha = 0.2$, hence total vorticity $\Omega_0 = \pi\alpha$. For each temperature, the Lagrange multiplier $\mu_q(\beta)$ must be determined iteratively to satisfy this constraint. As seen the $\beta = -\infty$ ($T = 0^-$) the maximum energy solution gathers all vorticity near the disc center, while the $\beta = +\infty$ ($T = 0^+$) solution compacts all vorticity against the disc boundary. The $\beta = 0$ ($T = \pm \infty$) maximum entropy solution distributes the vorticity uniformly (center panel). | 2212.13249_9.jpg |
arxivcap_1702.06296_16 | Title: Gaia eclipsing binary and multiple systems. Supervised classification and self-organizing maps | Caption: Histogram of the error rates of the classifiers in the system morphology classification on the 500 {randomized} training set/test set partitions, using CALEB. The error rates are measured for both Random Forest (top row) and for LDA (bottom row) on objects only in the test sets for each partition. Grey: period, amplitude, all PCs (only for Random Forest), red: period, amplitude, PC1-PC11, blue: period, amplitude, PC1-PC4. | 1702.06296_16.jpg |
arxivcap_1906.05497_13 | Title: Deep Network Approximation Characterized by Number of Neurons | Caption: Illustrations of two sub-networks implementing the desired function $\phi=\phi_2\circ\phi_1$ based Equation (<ref>) and the fact $\min\{x_1,x_2\}=\tfrac{x_1+x_2-|x_1-x_2|}{2}=\tfrac{\sigma(x_1+x_2)-\sigma(-x_1-x_2)-\sigma(x_1-x_2)-\sigma(-x_1+x_2)}{2}$. $y_{\tn{max}}$ is given by $\max\{ y_{m,\ell}: m=0,1,\cdots,M-1 \tn{ and } \ell=0,1,\cdots,L-1\}$. "$\psi_1$","$\psi_2$", and "$\psi_3$" near "$\longrightarrow$" represent the respective ReLU FNN implementing itself. We omit the activation function ReLU if the input of a neuron is non-negative. | 1906.05497_13.jpg |
End of preview. Expand
in Data Studio
Vis-IR: Unifying Search With Visualized Information Retrieval
Overview
VIRA (Vis-IR Aggregation), a large-scale dataset comprising a vast collection of screenshots from diverse sources, carefully curated into captioned and questionanswer formats.
Statistics
There are three types of data in VIRA: caption data, query-to-screenshot (q2s) data, and screenshot+query-to-screenshot (sq2s) data. The table below provides a detailed breakdown of the data counts for each domain and type.
Organization Structure
The dataset is organized in the following structure:
Domain/
βββ caption.jsonl: a screenshot image path and its corresponding caption
βββ q2s.jsonl: a query, a positive screenshot and eight negative screenshots
βββ sq2s.jsonl: a query, a query screenshot, a positive screenshot and eight negative screenshots
βββ images/
βββ image1.jpg
βββ image2.jpg
...
Due to the large number of images, uploading all of them takes time. Currently, the upload is not yet complete, and we will continue the process.
License
VIRA is licensed under the MIT License.
Citation
If you find this dataset useful, please cite:
@article{liu2025any,
title={Any Information Is Just Worth One Single Screenshot: Unifying Search With Visualized Information Retrieval},
author={Liu, Ze and Liang, Zhengyang and Zhou, Junjie and Liu, Zheng and Lian, Defu},
journal={arXiv preprint arXiv:2502.11431},
year={2025}
}
- Downloads last month
- 0