id
stringlengths 20
32
| text
stringlengths 100
121k
| image_path
stringlengths 15
235
|
|---|---|---|
arxivcap_2303.06734_3 | Title: Environmental variability and network structure determine the optimal plasticity mechanisms in embodied agents | Caption: a. Schematic representation of two-states Markov model with transition probability $p_{tr}$ between two environments $E_1$ and $E_2$ defined by the ingredient value distributions. b. We vary the $E_2$ environment by changing the ingredient values linearly $E_2 = (1- 2 d_e) E_1$, the $d_e$ is indicated by the color. c. The evolved learning rate $\eta_p$ grows with the distance $d_e$ between the environments and decreases with the reward variance $\sigma$. d. The environment transition probability $p_{tr}$ (here for $d_e = 1$ and $\sigma = 0.25$) has a non-monotonous relationship with the evolved learning rate $\eta_p$. Up to a certain point, more rapid transitions lead to faster learning, but too rapid environmental transition leads to a reduction of the evolved learning rate. e. For slow environmental transition (top), the agent fully adapts to the environment after each transition. If the transitions happen fast (bottom), the agent maintains an intermediate position between the two environments and never fully adapts to either of them. | 2303.06734_3.jpg |
arxivcap_2206.11972_5 | Title: Task-Adaptive Few-shot Node Classification | Caption: An illustration of the overall process of TENT. We first sample a meta-task from the given graph. Then we construct subgraphs for node-level adaptions and utilize node embeddings in each class for class-level adaptations. We further maximize the mutual information between the support set and the query set during query matching for task-level adaptations. | 2206.11972_5.jpg |
arxivcap_2212.13126_3 | Title: Eliminating temporal correlation in quantum-dot entangled photon source by quantum interference | Caption: (color online) Schematics of principle and apparatus. (a) Diagram of energy levels for cascaded emission. $\gamma_\text{XX}$ and $\gamma_\text{X}$ indicate the decay rates of XX and X, respectively. (b) Calculated joint probability of X and XX in the frequency domain, which is the Fourier transform of the temporal correlation. Here, $\gamma_\text{XX}=3.22\gamma_\text{X}$, deduced from the lifetime measurement of the investigated single QD. (c) Schematic of the experiment. Different from the usual single PBS to project independent entangled photon sources, we interfere X and XX photons together to eliminate temporal correlation. (d) Experimental setup. See the maintext for details. A GaAs/AlGaAs QD embedded in a bull's-eye microcavity consecutively emits two entangled photon pairs. Then notch filter suppresses the excitation laser and DM routes X and XX to the reflection and transmission paths. Beam splitters are used to de-multiplex photons to unbalanced Mach-Zehnder interferometers and impinge on PBSs for quantum interference. After that, photons are projected to arbitrary basis using a quarter waveplate (QWP), half waveplate (HWP) and PBS, and then, measured in APDs. Correlation events are analyzed by the TDC module. | 2212.13126_3.jpg |
arxivcap_2108.09637_3 | Title: Graph-Convolutional Deep Learning to Identify Optimized Molecular Configurations | Caption: Left panels are the results from the global average pooling network: (a) The confusion matrix obtained from testing the untrained network using random weight matrices, (b) Training progress, and (c) The corresponding confusion matrix obtained from the trained global average pooling network. Right panels, (d-f), are the similar set of results from the global max pooling network. In each confusion matrix, the class-wise precision are the scores in the first row of the 'column summary' of the chart and the class-wise recall are the scores in the first column of the 'row summary' of the chart. | 2108.09637_3.jpg |
arxivcap_2005.06146_8 | Title: Neutron spin resonance in a quasi-two-dimensional iron-based superconductor | Caption: {Energy scans at $Q = (0.5, 0.5, L)$ ($L$ = 1, 2, 3, 4, 5, 6, 8) below (black) and above (red) $T_c$.} | 2005.06146_8.jpg |
arxivcap_1111.0430_4 | Title: Fermi-LAT spectral analysis of Fermi, Planck, Swift and radio selected samples of AGN | Caption: SED of PKS 1510-089 from the radio sample <cit.>. Red filled points: simultaneous multi-frequency data; green points: $\gamma$-ray data integrated over a period of 2 months centered on the Planck observing period, or ground-based data taken quasi-simultaneously; blue points: $Fermi$-LAT data integrated over 27 months; light gray points: literature or archival data. | 1111.0430_4.jpg |
arxivcap_1912.05084_12 | Title: Bayesian Copula Density Deconvolution for Zero-Inflated Data in Nutritional Epidemiology | Caption: =10pt Results for simulated data sets with sample size $n = 1000$, $q=2$ episodic components and $p=1$ regular components, each subject having $m_{i}=3$ replicates, for the data set corresponding to the 25th percentile $3$-dimensional ISE. The off-diagonal panels show the contour plots of the true two-dimensional marginals (upper triangular panels) and the corresponding estimates obtained by our method (lower triangular panels). The numbers $i,j$ at the top right corners indicate which marginal densities $f_{X_{i},X_{J-1}}$ are plotted in those panels. The diagonal panels show the true (in black) one dimensional marginal densities and the corresponding estimates (in blue) produced by our method. Compare with Figure <ref>. | 1912.05084_12.jpg |
arxivcap_2210.09435_1 | Title: Robot Learning Theory of Mind through Self-Observation: Exploiting the Intentions-Beliefs Synergy | Caption: The architecture utilised in the here reported studies, formed of a shared prediction net torso and subsequently of separate prediction heads. For the NoBeliefs architecture, the following prediction heads are considered: 1. Target position, 2. Actor's next action, and 3. Actor's next state. For the Beliefs architecture, the 4. belief prediction head (in red) is also considered | 2210.09435_1.jpg |
arxivcap_2206.01855_16 | Title: A possible surviving companion of the SN Ia in the Galactic SNR G272.2-3.2 | Caption: Degraded CARMENES VIS spectrum of star Karmn J05415+534 Normalized CARMENES VIS 1D spectrum of star Karmn J05415+534 (HD 233153), corrected for barycentric radial velocity, degraded to a resolving power of $R\sim28,000$, with a signal-to-noise ratio of $\sim$ 15 at 7400 Å, and normalized to unity using a running mean filter with a width of 200 pixels at 0.069 Å per pixel. We also display an interpolated SYNPLE synthetic spectrum with the stellar parameters $T_{\rm eff}=3825$ K, $\log g =4.80$ and and metallicity [Fe/H] $= -0.3$. The regions used to estimate the metallicity are shown in grey and the different lines used for chemical analysis are also highlighted. | 2206.01855_16.jpg |
arxivcap_2303.14786_11 | Title: Detecting Low-Mass Perturbers in Cluster Lenses using Curved Arc Bases | Caption: Posterior probability distributions of the SMBH parameters for two different data sets with different masses. The green contours correspond to an SMBH with mass $10^{9}\,\msun$. The turquoise contours correspond to a SMBH with mass $10^{8}\,\msun$. Both are placed at the same optimal location as "perturber 1". | 2303.14786_11.jpg |
arxivcap_2305.05311_3 | Title: Structured Sentiment Analysis as Transition-based Dependency Parsing | Caption: Length of transition sequences predicted by our approach relative to the sentence length on development sets for the five SSA benchmarks. | 2305.05311_3.jpg |
arxivcap_1109.6459_2 | Title: Folding Kinetics of a Polymer | Caption: Chevron plots with folding and unfolding rates computed by FFS. The intersections give the transition temperatures $T_f^{\text{CD}}$ estimated by collision dynamics. The schematic for $\chi = 1.06$ shows that kinetic hindering of unfolding in CD simulations can explain the observation that $T_f^{\text{CD}}>T_f^{\text{MC}}$. The schematic assumes that the hindering involved in the folding process is negligible. | 1109.6459_2.jpg |
arxivcap_2206.11990_1 | Title: Equiformer: Equivariant Graph Attention Transformer for 3D Atomistic Graphs | Caption: Equivariant operations used in Equiformer. (a) Each gray line between input and output irreps features contains one learnable weight. (b) "RMS" denotes the root mean square value along the channel dimension. For simplicity, we have removed multiplying by $\gamma$ here. (c) Gate layers are equivariant activation functions where non-linearly transformed scalars are used to gate non-scalar irreps features. (d) The left two irreps features correspond to two input irreps features, and the rightmost one is the output irreps feature. The two gray lines connecting two vectors in the input irreps features and one vector in the output irreps feature form a path and contain one learnable weight. An alternative visualization of depth-wise tensor products can be found in Fig. <ref> in appendix. We show $SE(3)$-equivariant operations here, which can be generalized to $E(3)$-equivariant features. | 2206.11990_1.jpg |
arxivcap_1104.3706_6 | Title: Detection of a large massive circumstellar disk around a high-mass young stellar object in the Carina Nebula | Caption: SED (above) and near-infrared model images (below) computed with RADMC. The solid dots in the SED plot show the HAWK-I and Spitzer data points, the open squares the MSX data, the thick X symbol the TIMMI2 measurement, the open diamonds the IRAS fluxes, and the open square the LABOCA flux. A square-root intensity scaling is used for the model images. | 1104.3706_6.jpg |
arxivcap_2202.01503_9 | Title: Global sensitivity analysis based on Gaussian-process metamodelling for complex biomechanical problems | Caption: Summary of results for arterial growth and remodelling. A. Projected mean $m_{\textsf{GP}i}(X_i)$, projected 95% confidence interval (CI) and training samples $\datatrain$ for the diameter $\dmax$ (the y-axis labels apply to both figures). B. Nash-Sutcliffe efficiency $Q^2$ for different training sample set sizes. C. First-order and total-order Sobol indices and 95% confidence intervals (CI) for an increasing number of training samples. Monte-Carlo abbreviates as MC. | 2202.01503_9.jpg |
arxivcap_1906.05326_15 | Title: Magnetism and superconductivity in Fe$_{1+y}$Te$_{1-x}$Se$_x$ | Caption: From <cit.>. (a) Change in $a$ and $c$ lattice parameters, normalized to 300 K, as a function of temperature for the $x = 0.50$ sample. Statistical uncertainties are comparable to the symbol size. (b) Change in $a/c$, normalized to 300 K, as a function of temperature for FeTe$_{1-x}$Se$_x$; the values of $x$ are noted in the symbol legend. The average of in-plane lattice parameters was used for a in the low-temperature phase of $x = 0$. Reprinted with permission from Ref. <cit.>, 2016 by the American Physical Society. | 1906.05326_15.jpg |
arxivcap_2005.06146_10 | Title: Neutron spin resonance in a quasi-two-dimensional iron-based superconductor | Caption: {Constant-energy scans along $Q=[H, H, 6]$ and their differences for $E$ = 6, 10, 11, 12, 13, 14, 15, 16, 17, 17.5, 18, 20 and 22 meV}. The triangles mark the incommensurate peak of the magnetic scattering, and the solid line are gaussian fits by one or double functions. The incommensurability $\delta$ is defined by the distance between two peaks as shown in b5 - b12. | 2005.06146_10.jpg |
arxivcap_2202.00203_9 | Title: Constraining low redshift [CII] Emission by Cross-Correlating FIRAS and BOSS Data | Caption: The binned angular power spectrum, $C_b(z,z')$, for the CMASS galaxies. The left and right columns show the data and the best-fit model, respectively. The rows show the three angular bins used in this analysis, each of size $\Delta \ell=9$ and centered at $\ell=24$, 33, and 42. At the resolution of FIRAS, most of the cosmological clustering signal occurs on the diagonal, where $z=z'$, though there is also a small correlation between neighboring redshift bins, visible just off of the diagonal. The variation in amplitude along the diagonal is due to the redshift evolution of the growth factor, the change in physical scales being probed as a function of redshift, and changes in the CMASS galaxy density. | 2202.00203_9.jpg |
arxivcap_2102.12804_3 | Title: Compact Groups of Galaxies in Sloan Digital Sky Survey and LAMOST Spectral Survey. II. Dynamical Properties of Isolated and Embedded Groups | Caption: Symbols and shaded areas are the median and 16$\text{th}$ to 84$\text{th}$ percentiles of $M_{\text{dyn}}$ as a function of group luminosity ($L_{19.5}$) for isolated CGs (red), and noncompact group samples (black) on a logarithmic scale with bin size of $0.4$ dex. The intermediate ($\mu \lesssim 27.2$) and loose ($\mu \gtrsim 27.2$) subsamples of noncompact groups are also shown in green and blue, respectively. The solid lines are the linear fit for these samples. Thick dashed lines are $M_\text{dyn}-L_{19.5}$ relation based on virial equilibrium ($M_{\text{dyn}} \propto L_{\text{19.5}}$) with various $\frac{M_{\text{dyn}}}{L_{\text{19.5}}}$ values. The vertical error bars show the errors of the median $\log\left(M_{\text{dyn}}\right)$ and the horizontal error bars indicate the median absolute deviation of $\log\left(L_{19.5}\right)$ in each bin. Only the data bins with at least 10 groups are plotted. | 2102.12804_3.jpg |
arxivcap_2108.09598_1 | Title: SERF: Towards better training of deep neural networks using log-Softplus ERror activation Function | Caption: Output landscapes of a randomly initialized 6-layered neural network with ReLU (Left) and Serf (Right) activations. | 2108.09598_1.jpg |
arxivcap_2108.09598_2 | Title: SERF: Towards better training of deep neural networks using log-Softplus ERror activation Function | Caption: Activation functions (Left), first derivatives (Middle) and second derivatives (Right) for Swish, Mish and Serf. | 2108.09598_2.jpg |
arxivcap_2212.13138_2 | Title: Large Language Models Encode Clinical Knowledge | Caption: Scaling plots for Flan-PaLM with few-shot and Flan-PaLM few-shot + chain-of-thought (CoT) + self-consistency (SC) on MedQA and MedMCQA | 2212.13138_2.jpg |
arxivcap_1111.0261_1 | Title: Extended Lagrangian free energy molecular dynamics | Caption: (Color online) The fluctuations in the total free energy, $E^{\rm tot}_F = E_{\rm K} + U - T_e{\cal S}$, of a water molecule using regular free energy molecular dynamics for a conventional density matrix based Born-Oppenheimer method with a linear extrapolation of the electronic state, Eq. (<ref>), in comparison to the density matrix extended Lagrangian free energy molecular dynamics. | 1111.0261_1.jpg |
arxivcap_1806.08658_7 | Title: Privacy-Preserving Identification via Layered Sparse Code Design:
Distributed Servers and Multiple Access Authorization | Caption: The relation between normalized similarity and: a) sparsity ratio, b) encoding rate. | 1806.08658_7.jpg |
arxivcap_2210.09435_2 | Title: Robot Learning Theory of Mind through Self-Observation: Exploiting the Intentions-Beliefs Synergy | Caption: Visualisation of example 11x11 grid world maps and observed trajectories, which varied in the location of walls, columns and free cells to move around the map. Colour code - Black: walls; White: empty cells; Yellow: target; Green: distractor objects; Blue: current actor's position; Pink: past actor's positions. | 2210.09435_2.jpg |
arxivcap_1109.6459_3 | Title: Folding Kinetics of a Polymer | Caption: Phase diagram with freezing temperatures determined from chevron plots ($T_f^{\text{CD}}$ , FFS and BF) and by Maxwell construction ($T_f^{\text{MC}}$, WL). The errors are smaller than the symbol sizes. | 1109.6459_3.jpg |
arxivcap_2108.09598_3 | Title: SERF: Towards better training of deep neural networks using log-Softplus ERror activation Function | Caption: Ablations for MNIST dataset. Top: Testing Accuracies vs Dense Units (Left), Dropout Rates (Middle) and Initializers (Right) for Swish, Mish and Serf. Bottom: Testing Accuracies vs Learning Rates (Left), Optimizers (Middle) and Number of Layers (Right) for Swish, Mish and Serf. | 2108.09598_3.jpg |
arxivcap_2206.01855_17 | Title: A possible surviving companion of the SN Ia in the Galactic SNR G272.2-3.2 | Caption: Normalized degraded CARMENES VIS 1D spectrum of star Karmn J00183+440 (GX And), corrected for barycentric radial velocity, degraded to a resolving power of $R\sim28,000$, with a signal-to-noise ratio of $\sim$ 15 at 7400 Å, and normalized to unity using a running mean filter with a width of 200 pixels at 0.069 Å per pixel. We also display an interpolated SYNPLE synthetic spectrum with the stellar parameters $T_{\rm eff}=3600$ K, $\log g =4.85$ and and metallicity [Fe/H] $= -0.6$. The regions used to estimate the metallicity are shown in grey and the different lines used for chemical analysis are also highlighted. | 2206.01855_17.jpg |
arxivcap_2206.11990_2 | Title: Equiformer: Equivariant Graph Attention Transformer for 3D Atomistic Graphs | Caption: Architecture of Equiformer. {Add detailed explanation of each submodule of Attention to figures in Appendix.} {Add dot product attention in Appendix.} {Update variables: $x_{ij}, a_{ij}, v_{ij}$} Here "$\otimes$" denotes multiplication, "$\oplus$" denotes addition, and "DTP" stands for depth-wise tensor product. $\sum$ within a circle denotes summation over all neighbors. | 2206.11990_2.jpg |
arxivcap_2303.06734_4 | Title: Environmental variability and network structure determine the optimal plasticity mechanisms in embodied agents | Caption: The evolved parameters $ {\theta} = (\theta_1, \dots, \theta_8)$ of the plasticity rule for the reward prediction (a.) and the decision (b.) tasks, for a variety of parameters ($p_{tr} = 0.01$, $d_e \in {0, 0.1, \dots, 1}$, and $\sigma \in {0, 0.1, \dots, 1}$ in all 100 combinations). Despite the relatively small difference between the tasks, the evolved learning rules differ considerably. For visual guidance, the lines connect $\theta$s from the same run. | 2303.06734_4.jpg |
arxivcap_2210.09455_1 | Title: Track Targets by Dense Spatio-Temporal Position Encoding | Caption: The illustration of associating targets within a video clip. For general cases, we use bounding boxes to represent the targets of interest while we can further replace the detector with a segmentation model to do the attention in a more fine-grained mask area. Positional encoding is added to CNN feature maps to encode position information. | 2210.09455_1.jpg |
arxivcap_1503.07004_6 | Title: Impact of symmetry breaking in networks of globally coupled oscillators | Caption: (a)Space-time plot and (b) Snapshot of the variables $x_j$ for $N=100$ oscillators which confirm the existence of chimera death state for coupling strength $\epsilon=3.5$ and nonisochronicity parameter $c=2.5$. | 1503.07004_6.jpg |
arxivcap_2303.14786_13 | Title: Detecting Low-Mass Perturbers in Cluster Lenses using Curved Arc Bases | Caption: Posterior probability distributions for the NFW model parameters applied on a SMBH with mass $10^{8}\,\msun$ placed at the same optimal location as "perturber 1". | 2303.14786_13.jpg |
arxivcap_1902.04007_6 | Title: Scaling laws for harmonically trapped two-species mixtures at thermal equilibrium | Caption: Scaling of the total interaction energy with system size ($N_A=N_B=N$) for both repulsive (left) and attractive (right) cases. The strength $\gamma$ is specified while $m_A=m_B=1.0$, $\lambda=0.1$, $\omega_A= 1.0,\omega_B=144/89$, $\beta_A=0.2$ and $\beta_B=2.0$. | 1902.04007_6.jpg |
arxivcap_2012.05329_3 | Title: Know Your Limits: Uncertainty Estimation with ReLU Classifiers Fails at Reliable OOD Detection | Caption: (a) Uncertainty of a neural classifier with ReLU activations measured by predictive entropy on synthetic data, illustrated by increasing shades of purple with white denoting absolute certainty. (b) Polytopal, linear regions in the feature space induced by the same classifier (as introduced by <cit.>). (c) Norm of the gradient of the predictive entropy plotted by increasing shades of green, showing how small perturbations in the input have a decreasing influence on the uncertainty of the network as we stray away from the training data, creating large areas in which uncertainty levels are overgeneralized. Exceptions to this are the model's decision boundaries, which is discussed in Section <ref>. Polytopes are drawn using the code of <cit.>. | 2012.05329_3.jpg |
arxivcap_1210.7665_1 | Title: Graph Estimation From Multi-attribute Data | Caption: Average hamming distance plotted against the rescaled sample size. Results are averaged over 100 independent runs. Off-diagonal blocks are full matrices. | 1210.7665_1.jpg |
arxivcap_1912.05084_13 | Title: Bayesian Copula Density Deconvolution for Zero-Inflated Data in Nutritional Epidemiology | Caption: =10pt Results for simulated data sets with sample size $n = 1000$, $q=2$ episodic components and $p=1$ regular components, each subject having $m_{i}=3$ replicates, for the data set corresponding to the 25th percentile $3$-dimensional ISE. The off-diagonal panels show the contour plots of the true two-dimensional marginals (upper triangular panels) and the corresponding estimates obtained by the method of <cit.> (lower triangular panels). The numbers $i,j$ at the top right corners indicate which marginal densities $f_{X_{i},X_{J-1}}$ are plotted in those panels. The diagonal panels show the true (in black) one dimensional marginal densities and the corresponding estimates (in red) produced by the method of <cit.>. Compare with Figure <ref>. | 1912.05084_13.jpg |
arxivcap_1906.05326_16 | Title: Magnetism and superconductivity in Fe$_{1+y}$Te$_{1-x}$Se$_x$ | Caption: Schematic summary of the changes in magnetic correlations in with composition and temperature. | 1906.05326_16.jpg |
arxivcap_2202.01525_6 | Title: Reliable Community Search in Dynamic Networks | Caption: Index Construction Time on All Datasets Index Construction Time by Maintenance | 2202.01525_6.jpg |
arxivcap_1111.0261_2 | Title: Extended Lagrangian free energy molecular dynamics | Caption: (Color online) The fluctuations in the nuclear kinetic and potential energy, $E_{\rm K} + U$, in comparison to the total free energy, $E^{\rm tot}_F = E_{\rm K} + U - T_e{\cal S}$, for a density matrix extended Lagrangian free energy molecular dynamics simulation of a single water molecule using Hartree-Fock (RHF) or hybrid density functional theory (PBE0 <cit.>). The average nuclear temperature was approximately room temperature, i.e. $T_{\rm ion}\approx 300$ K. | 1111.0261_2.jpg |
arxivcap_1109.6459_4 | Title: Folding Kinetics of a Polymer | Caption: Configurations at $\lambda_{n-1}^B$ with largest eigenvalue $\gamma$ of the dynamical matrix lower than the critical value ($\gamma^c = 11.9$) have almost no chance to crystallize. Insets: (a) Unimodal distribution $a(\lambda^A;\gamma)$ at the surface $\lambda^A = -235$ sampled by pathways started in $A$. (b) Bimodal distribution of $b(\lambda^B;\gamma)$ at the same surface ($\lambda_{n-1}^B = -235$) but sampled by pathways started in $B$. | 1109.6459_4.jpg |
arxivcap_2005.06146_11 | Title: Neutron spin resonance in a quasi-two-dimensional iron-based superconductor | Caption: {Constant-energy scans along $Q=[H, H, 3]$ and $[H, H, 6]$ for $E$ = 12 meV.} a, Raw data of constant-energy scans below and above $T_c$ along $Q=[H, H, 3]$. b, Comparison of incommensurate resonant peaks distribution for $[H, H, 3]$ and $[H, H, 6]$ scans. | 2005.06146_11.jpg |
arxivcap_1503.07004_7 | Title: Impact of symmetry breaking in networks of globally coupled oscillators | Caption: (a) Phase diagram for the globally coupled system (<ref>) with symmetry breaking by varying the values of $c$ and $\epsilon$ for $N=100$ oscillators. Green color shows the synchronized state, yellow color shows the state corresponding to desynchronized region (DS), blue color shows amplitude chimera state (AC), violet color shows the frequency chimera state (FC) and brown color shows the amplitude cluster states (ACS) and frequency chimera states (FCS) and red color shows the chimera death region. Region `A': DS $\rightarrow$ACS$\rightarrow$ SY $\rightarrow$ CD, region `B': DS $\rightarrow$ACS$\rightarrow$ AC $\rightarrow$ACS $\rightarrow$ SY $\rightarrow$ CD, and region `C' : DS $\rightarrow$FCS$\rightarrow$ FC $\rightarrow$FCS $\rightarrow$ SY $\rightarrow$ CD. (b) shows the analytical plot for global coupling with symmetry breaking. Regions ($I$) and ($IV$) correspond to synchronized states, region ($II$) and ($III$) are for chimera and cluster states and the region ($V$) shows the chimera death states. The red color dot represents the Takens-Bogdanov point. | 1503.07004_7.jpg |
arxivcap_2205.02281_3 | Title: Three-Body Problem in Modified Dynamics | Caption: The time evolution ($200~\mathcal{T}$) of three equal mass galaxies ( all three with mass $\mathcal{M}$ ) in two dimensions and according to Newtonian and MOD models with exactly the same initial conditions derived from <cit.>. The plots show that the fate of galaxies in a group could entirely change under modified dynamics. The coordinates $x$ and $y$ are in units of $\mathcal{D}$. | 2205.02281_3.jpg |
arxivcap_2212.13138_3 | Title: Large Language Models Encode Clinical Knowledge | Caption: Overview of our contributions We curated MultiMedQA, a benchmark for medical question answering spanning medical exam, medical research, and consumer medical questions. We evaluated PaLM and its instructed-tuned variant, Flan-PaLM, on MultiMedQA. With a combination of prompting strategies, Flan-PaLM exceeded SOTA performance on MedQA (USMLE), MedMCQA, PubMedQA, and MMLU clinical topics. In particular, it improved over the previous SOTA on MedQA (USMLE) by over 17%. We next proposed instruction prompt tuning to further align Flan-PaLM to the medical domain, producing Med-PaLM. Med-PaLM's answers to consumer medical questions compared favorably with clinician-generated answers under our human evaluation framework, demonstrating the effectiveness of instruction prompt tuning. | 2212.13138_3.jpg |
arxivcap_2202.13869_1 | Title: Query Expansion and Entity Weighting for Query Reformulation Retrieval in Voice Assistant Systems | Caption: The overview of the workflow: 1) entity catalog data is first used to build the Entity Expansion Knowledge Base; 2) an entity weight prediction model is trained with annotated Alexa <query, reformulation> pairs. | 2202.13869_1.jpg |
arxivcap_2210.09455_2 | Title: Track Targets by Dense Spatio-Temporal Position Encoding | Caption: A deeper look at the component in our method. (a): how we generate the feature representations for both trajectories and single-frame objects. The representation of a trajectory is the accumulated positional encoding of all contained historical locations and the appearance feature of the last snapshot of the object. (b): for the video tracking and segmentation task, we use the semantic occupancy map onto object RoI to obtain more fine-grained RoI features where both position and semantics are encoded. | 2210.09455_2.jpg |
arxivcap_2108.09598_4 | Title: SERF: Towards better training of deep neural networks using log-Softplus ERror activation Function | Caption: Ablations for CIFAR-10 dataset. Testing Accuracies vs Batch Sizes (Left), Optimizers (Middle) and Learning Rates (Right) for Swish, Mish and Serf. | 2108.09598_4.jpg |
arxivcap_2202.13869_2 | Title: Query Expansion and Entity Weighting for Query Reformulation Retrieval in Voice Assistant Systems | Caption: Illustration of the building process of the entity connection knowledge base. First, each entity in one entity catalog entry will be categorized based on its occurrence in this entry's query and response. Second, every two entities in the same entry will be connected with each other. Third, for each edge, the "overall relevance score" is calculated based on these two entities' respective scores in all times they appear together. | 2202.13869_2.jpg |
arxivcap_2102.12804_4 | Title: Compact Groups of Galaxies in Sloan Digital Sky Survey and LAMOST Spectral Survey. II. Dynamical Properties of Isolated and Embedded Groups | Caption: The dynamic mass of groups scaled to their true mass ($\frac{M_{\text{dyn}}}{M_{0}}$) versus crossing time ($t_{\text{cr}}$) scaled to the age of universe $t_0$ for isolated CGs (red) and noncompact groups (gray), respectively. The error bar represents the typical error of these samples. The solid lines are the evolutionary track of a galaxy system adapted from <cit.> based on the softened potential assumption and the arrows indicate the evolutionary direction. | 2102.12804_4.jpg |
arxivcap_2303.06734_5 | Title: Environmental variability and network structure determine the optimal plasticity mechanisms in embodied agents | Caption: a. The trajectory of an agent (blue line) in the 2D environment. A well-trained agent will approach and consume food with positive values (green dots) and avoid negative food (red dots). b. The learning rate of the plastic sensory network $eta_p$ grows with the distance between environments $d_e$ c. and decreases with the frequency of environmental change. d. The fitness of an agent (measured as the total food consumed over its lifetime) increases over generations of the EA for both the scalar and binary readouts in the sensory network. e. The Pearson correlation coefficient of an evolved agent's weights with the ingredient value vector of the current environment ($E_1$ - blue, $E_2$ - red). In this example, the agent's weights are anti-correlated with its environment, which is not an issue for performance since the motor network can interpret the inverted signs of food. | 2303.06734_5.jpg |
arxivcap_2012.05329_4 | Title: Know Your Limits: Uncertainty Estimation with ReLU Classifiers Fails at Reliable OOD Detection | Caption: Uncertainty on the half-moon dataset, including the binary classification AUC-ROC. (Top row) The uncertainty surface is represented with increasingly darker shades of purple, with white being the lowest uncertainty. Open-ended regions of static certainty appear across different models and metrics, bein extrapolated to unseen data (see <ref>-<ref>); this phenomenon is less apparent in some instances (<ref>). (Bottom row) Increasing shades of green indicate the magnitude of the gradient of the uncertainty score w.r.t. the input. All metrics show open ended regions where the magnitude approaches zero. | 2012.05329_4.jpg |
arxivcap_2206.11990_3 | Title: Equiformer: Equivariant Graph Attention Transformer for 3D Atomistic Graphs | Caption: Architecture of Equiformer. We embed input 3D graphs with atom and edge-degree embeddings and process them with Transformer blocks, consisting of equivariant graph attention and feed forward networks. In this figure, "$\otimes$" denotes multiplication, "$\oplus$" denotes addition, and "DTP" stands for depth-wise tensor product. $\sum$ within a circle denotes summation over all neighbors. Gray cells indicate intermediate irreps features. | 2206.11990_3.jpg |
arxivcap_2202.00203_10 | Title: Constraining low redshift [CII] Emission by Cross-Correlating FIRAS and BOSS Data | Caption: The binned angular power spectrum, $C_{b}(z,z')$, in the LOWZ redshift range for three angular bins of width $\Delta \ell = 9$, centered at $\ell=24$, 33, and 42. The left column shows the data, and the right column shows the best-fit model, of the form of Equation <ref>. The structure on the diagonal of the plots is due to thermal noise, with small just-off-diagonal correlations due to the FIRAS frequency window function, $A(\nu)$. The thermal noise increases with higher $\ell$, roughly as the inverse of the FIRAS beam and scan window function squared. The foregrounds are visible in the $\ell=24$ bin as a roughly constant offset to all $z, z'$ combinations. The foreground amplitude is comparable to the thermal noise at $\ell=24$, but it drops at higher $\ell$ with a power-law index of $\gamma \approx -2.3$. The foregrounds are negligible compared to the thermal noise at $\ell=33$ and $\ell=42$. | 2202.00203_10.jpg |
arxivcap_2005.06146_14 | Title: Neutron spin resonance in a quasi-two-dimensional iron-based superconductor | Caption: {DFT calculation results for the band structure and Fermi surfaces.} a, Electronic band structure and its orbital distributions. b, Partial-density-of-state (PDOS) for each orbital of Fe$^{2+}$. c, Fermi surfaces and their sizes. {d - h}, Orbital characters of the Fermi surfaces shown with different color codes. | 2005.06146_14.jpg |
arxivcap_1802.01289_4 | Title: On Distributed Algorithms for Cost-Efficient Data Center Placement in Cloud Computing | Caption: Total response time ratio of DLM to the greedy algorithm for Internet topologies. | 1802.01289_4.jpg |
arxivcap_2206.01855_18 | Title: A possible surviving companion of the SN Ia in the Galactic SNR G272.2-3.2 | Caption: Fits of a {SOAR} spectrum of the high-velocity star {Gaia} EDR3 5323871314998012928. Upper panel: fit to a M1V template, for solar metallicity, after dereddening by {E(B - V)} = 0.532 mag. Lower panel: fit to a M2V template, also for solar metallicity, without dereddening. In this case, a negligible reddening is favored. | 2206.01855_18.jpg |
arxivcap_2303.14786_14 | Title: Detecting Low-Mass Perturbers in Cluster Lenses using Curved Arc Bases | Caption: Posterior probability distributions of the lens model parameters listed in Table <ref>. The best fits are shown as dashed lines. Parameters determining the relative offsets between the images and the source model parameters are omitted for brevity. They are uncorrelated with the parameters shown here. | 2303.14786_14.jpg |
arxivcap_2108.09588_3 | Title: Decomposition Multi-Objective Evolutionary Optimization: From State-of-the-Art to Future Opportunities | Caption: Development trajectories of MOEA/D from the past to the state-of-the-art. | 2108.09588_3.jpg |
arxivcap_1503.07222_3 | Title: Exponential Convergence Bounds using Integral Quadratic Constraints | Caption: Modified feedback diagram with additional multipliers and inputs. For appropriately chosen $e$ and $f$ and with zero initial condition, we show how this diagram is equivalent to that of Fig. <ref>. | 1503.07222_3.jpg |
arxivcap_1806.08917_1 | Title: Halide-Perovskite Resonant Nanophotonics | Caption: Schematic illustration of major properties of halide perovskites emerged recently as promising novel materials for many applications in optoelectronics and photonics. | 1806.08917_1.jpg |
arxivcap_1109.6747_2 | Title: Deterministic qubit transfer between orbital and spin angular momentum of single photons | Caption: (a) Conversion efficiency of the q-plate QP1 as a function of the applied voltage. Above the threshold voltage (around 2.2 V), the conversion efficiency of the q-plate can be adjusted, achieving its maximum around 4.5 V. (b) Photo of the q-plate. | 1109.6747_2.jpg |
arxivcap_1802.01289_6 | Title: On Distributed Algorithms for Cost-Efficient Data Center Placement in Cloud Computing | Caption: Two simulation instances for grid network when $n=400$, $k=0.5\%\cdot n=2$. | 1802.01289_6.jpg |
arxivcap_1906.05326_17 | Title: Magnetism and superconductivity in Fe$_{1+y}$Te$_{1-x}$Se$_x$ | Caption: From <cit.>. Square of the effective magnetic moment obtained by integrating the $S(E)$, as a function of temperature. Upper (blue) symbols show the total response, bottom (red) symbols are the Bragg contribution, green symbols are the quasielastic contribution. 2011 American Physical Society. | 1906.05326_17.jpg |
arxivcap_1709.03144_24 | Title: The 12C(a,g)16O reaction and its implications for stellar helium burning | Caption: (Color online) The reaction rate integrand as a function of CM energy for $T$ = 1, 2, 4, and 10 GK. At larger temperatures above $T$ = 1 GK several resonance contributions begin to dominate the rate. Above $T \approx$ 4 GK it is estimated that higher lying resonance contributions (at $E_{\text{c.m.}} >$ 6.5 MeV) not included in the present analysis could have a significant contribution compared to the quoted uncertainty. While the ground state transition has been shown to be fairly weak at these higher energies, limited information is available for the cascade transitions, and they may make significant contributions. For comparison, the Gaussian Gamow energy windows described by Eqs. <ref> and (<ref>) are indicated by the horizontal error bars. | 1709.03144_24.jpg |
arxivcap_2212.13138_5 | Title: Large Language Models Encode Clinical Knowledge | Caption: Comparison of our method and prior SOTA We achieve state-of-the-art performance on MedQA (4 options), MedMCQA and PubMedQA datasets with our Flan-PaLM 540B model. SOTA results come from Galactica (MedMCQA) <cit.>, PubMedGPT, and BioGPT <cit.> | 2212.13138_5.jpg |
arxivcap_1401.8053_12 | Title: Hallucinating optimal high-dimensional subspaces | Caption: The effects of additive zero-mean Gaussian noise, isotropic in the image space, applied to low resolution images before the construction of the corresponding subspaces. Shown is the change in the observed class separation which is for the sake of visualization clarity measured relative to the separation achieved using the original, un-corrupted images; see Figure <ref>(a). The results are for the bilinear projection model. Notice the remarkable robustness of the proposed model: even for noise with the pixel-wise standard deviation of 30 greyscale levels (approximately 12% of the entire greyscale intensity range), which corresponds to the average signal-to-noise ratio of 1.7, class separation is decreased by less than 1.5%. Note that this means that even when the proposed method performs matching in the presence of extreme noise, its performance exceeds that of the naïve approach applied to un-corrupted data. | 1401.8053_12.jpg |
arxivcap_2108.09588_7 | Title: Decomposition Multi-Objective Evolutionary Optimization: From State-of-the-Art to Future Opportunities | Caption: MOEA/D obtains evenly distributed solutions by using evenly distributed weight vectors merely when the PF is a simplex. Otherwise, some weight vectors do not have Pareto-optimal solutions like PF2; or the distribution of the corresponding Pareto-optimal solutions is highly biased like PF3 and PF4. | 2108.09588_7.jpg |
arxivcap_1111.0261_3 | Title: Extended Lagrangian free energy molecular dynamics | Caption: (Color online) The fluctuations in the nuclear kinetic and potential energy, $E_{\rm K} + U$, in comparison to the total free energy, $E^{\rm tot}_F = E_{\rm K} + U - T_e{\cal S}$, for a density matrix extended Lagrangian free energy molecular dynamics simulation of a single water molecule embedded within a $C_{60}$ Bucky ball using Hartree-Fock theory. The average nuclear temperature was approximately $T_{\rm ion}\approx 1000$ K. | 1111.0261_3.jpg |
arxivcap_1802.01257_1 | Title: Promoting cooperation by punishing minority | Caption: (Color online) The fraction of cooperators $\rho_{c}$ as a function of the multiplication factor $r$ for different values of the punishment fine $\alpha$. For small values of $\alpha$ (e.g., $\alpha=0$ or $\alpha=0.3$), the dependence of $\rho_{c}$ on $r$ displays a continuous phase transition. However, for large values of $\alpha$ (e.g., $\alpha=0.8$), the phase transition becomes discontinuous. | 1802.01257_1.jpg |
arxivcap_1902.03999_1 | Title: KTBoost: Combined Kernel and Tree Boosting | Caption: Test mean square error (MSE) versus the number of boosting iteration for KTBoost in comparison with tree and kernel boosting for one data set (wine). | 1902.03999_1.jpg |
arxivcap_1503.07222_6 | Title: Exponential Convergence Bounds using Integral Quadratic Constraints | Caption: Upper bounds on the exponential convergence rate $\rho$ for the system $G_1(z)$ given in (<ref>) in feedback as in Fig. <ref>. A tight bound is achieved using two $\rho$-IQCs. | 1503.07222_6.jpg |
arxivcap_2202.13833_3 | Title: Formally verified asymptotic consensus in robust networks | Caption: Illustration of the tube of convergence bounded above by $A_M + \epsilon$ and bounded below by $A_m - \epsilon$. We observe the behavior of functions $M(t)$ and $m(t)$ inside this tube of convergence $\forall t \geq t_\epsilon$. We prove that $M(t)$ and $m(t)$ are monotonous $\forall t \geq t_\epsilon$, and they approach the limits $A_M$ and $A_m$, respectively. We start by assuming that $A_M \neq A_m$, but later prove that $A_M = A_m$ by contradiction, thereby proving asymptotic consensus. {Nice figure!} | 2202.13833_3.jpg |
arxivcap_1912.05084_14 | Title: Bayesian Copula Density Deconvolution for Zero-Inflated Data in Nutritional Epidemiology | Caption: =10pt Results for the EATS data sets with sample size $n = 965$, $q=2$ episodic components, milk and whole grains, and $p=2$ regular components, sodium and energy, each subject having $m_{i}=4$ replicates. The off-diagonal panels show the contour plots of two-dimensional marginals estimated by our method (upper triangular panels) and the method of <cit.> (lower triangular panels). The numbers $i,j$ at the top right corners indicate which marginal densities $f_{X_{i},X_{J-1}}$ are plotted in those panels. The diagonal panels show the one dimensional marginal densities estimated by our method (in blue) and the method of <cit.> (in red). | 1912.05084_14.jpg |
arxivcap_2005.06243_1 | Title: Detecting and analyzing collusive entities on YouTube | Caption: An example of anomalous pattern in videos detected by for collusive (in red) and other videos (in green). The x-axis displays the time span (in days) starting from when the first comment was posted and y-axis determines an anomaly score calculated by . Here, the anomaly score is the Mahalanobis distance computed using Equation <ref> as mentioned in Section <ref>. The peak width (horizontal black line) corresponding to every peak indicates the duration of the peak. | 2005.06243_1.jpg |
arxivcap_1906.05326_19 | Title: Magnetism and superconductivity in Fe$_{1+y}$Te$_{1-x}$Se$_x$ | Caption: From <cit.>. Fe-Fe (a) and Fe-Te (b) bond length in Fe$_{1.09(1)}$Te obtained by fitting the $(1,0,0)$ and $(1,1,0)$ intensities. Resistivity (c) and magnetic susceptibility (d) measured on cooling (closed) and warming (open symbols) in the same Fe$_{1+y}$Te, $y = 0.09(1)$ sample. Formation of Fe-Fe zigzag chains manifests itself by the concomitant hysteretic decrease in both quantities. 2014 American Physical Society. | 1906.05326_19.jpg |
arxivcap_2108.09713_1 | Title: Robustness-via-Synthesis: Robust Training with Generative Adversarial Perturbations | Caption: We generate 100 adversarial images for each natural sample in the CIFAR10 test set. The adversarial images are obtained from 100 different random vectors. Then, the Euclidean distances between the normalized latent representations of the natural sample and 100 adversarial samples are computed. The average distances for 174 checkpoints are plotted. The curve represents the change in the average distance during training. The shaded region around the curve represents the average standard deviation of the distances between the natural sample and its 100 adversarial samples. While the diversity in the adversarial directions is preserved during the robust training, the average distance between the latent representations decreases as expected. | 2108.09713_1.jpg |
arxivcap_2206.11990_6 | Title: Equiformer: Equivariant Graph Attention Transformer for 3D Atomistic Graphs | Caption: Error distributions of different Equiformer models on different sub-splits of OC20 IS2RE validation set. | 2206.11990_6.jpg |
arxivcap_2202.01525_8 | Title: Reliable Community Search in Dynamic Networks | Caption: Index Construction Time with Different Sampling Ratios | 2202.01525_8.jpg |
arxivcap_1902.03999_2 | Title: KTBoost: Combined Kernel and Tree Boosting | Caption: An example of a random function with five random jumps in $[0,0.5]$ and corresponding observed data (left plot) and pointwise mean square error (MSE) for tree and kernel boosting as well as the combined KTBoost algorithm (right plot). | 1902.03999_2.jpg |
arxivcap_2303.06734_6 | Title: Environmental variability and network structure determine the optimal plasticity mechanisms in embodied agents | Caption: The evolved parameters of moving agents' plasticity rule for the $g(s) = x$, identity (a.) and the step function (Eq. <ref>) (b.) sensory networks (the environmental parameters here are $d_e \in [0, 1], \ \sigma = 0$ and $p_{tr} = 0.001$). The step function (binary output) network evolved a more structured plasticity rule (e.g., $\theta_3>0$ for all realizations) than the linear network. Moreover, the learned weights for the identity network (c.) have higher variance and correlate significantly less with the environment's ingredient distribution compared to the learned weights for the thresholded network (d.) | 2303.06734_6.jpg |
arxivcap_1109.6297_2 | Title: Low-rank data modeling via the Minimum Description Length principle | Caption: Results for the "Lobby" sequence (see text for a description of the above pictures and graphs). The rank of the approximation decomposition for this case is $\rank=10$. The moment where the lights are turned off is clearly seen here as the "square pulse" in the middle of the first two right-eigenvectors (bottom-right figure). Also note how $\vec{u}_2$ (top-right) compensates for changes in shadows. | 1109.6297_2.jpg |
arxivcap_2206.01855_19 | Title: A possible surviving companion of the SN Ia in the Galactic SNR G272.2-3.2 | Caption: Sky trajectory of the fastest moving star ({Gaia} EDR3 5323871314998012928) in the {Gaia} EDR3 sample of stars located within 1 degree from the centroid of the G272.2-3.2 SNR and with parallaxes corresponding to distances 1 kpc $\leq d \leq$ 3 kpc. The red dot marks the current position (labelled $N$) while the black dot marks that 10,000 yr ago (labelled $T$). The central crowded region is the area explored in the main survey, within 11 arcmin from the centroid. We easily see that this star can not have originated in the SN Ia that produced the remnant. | 2206.01855_19.jpg |
arxivcap_2212.13138_6 | Title: Large Language Models Encode Clinical Knowledge | Caption: Comparison of SOTA LLMs on MMLU clinical topics Flan-PaLM achieves state-of-the-art performance on MMLU clinical topics. | 2212.13138_6.jpg |
arxivcap_1503.07222_8 | Title: Exponential Convergence Bounds using Integral Quadratic Constraints | Caption: Upper bounds on the exponential convergence rate $\rho$ for the system $G_2(z)$ given in (<ref>) in feedback as in Fig. <ref>. As we include more $\rho$-IQCs, we can certify tighter bounds. | 1503.07222_8.jpg |
arxivcap_1806.08917_2 | Title: Halide-Perovskite Resonant Nanophotonics | Caption: Harmonics generation from halide perovskites. Spectra of photoluminescence (PL), second-harmonic (SHG) and third-harmonic generation (THG) from CsPbBr$_3$ at different excitation wavelengths (of a femtosecond laser): (a) 1700 nm and (b) 1600 nm. {Insets schematically show level diagrams and processes of SHG, THG, and PL at corresponding pulsed laser optical excitation. Adapted with permission from <cit.>. Copyright 2018 by the American Physical Society.} | 1806.08917_2.jpg |
arxivcap_2102.12804_5 | Title: Compact Groups of Galaxies in Sloan Digital Sky Survey and LAMOST Spectral Survey. II. Dynamical Properties of Isolated and Embedded Groups | Caption: Example SDSS images of cCGs where their members are in one-to-one correspondence with the members of Y07 groups: (a) isolated CGs with no external galaxy host in the same halo. (b) Predominant CGs with other fainter galaxies sharing the same halo. (c) Embedded CGs with brighter galaxies occupying the same halo as nondominant subsystems. (d) Split CGs whose members belong to at least two different halos of Y07. The inner white dashed circles represent the smallest enclosed circles $\theta_G$, the outer white dashed circles represent the concentric circles 3$\theta_G$. Green dashed circles represent the smallest enclosed circles for the Y07 groups, which are manually enlarged for clarity. Solid circles mark the member galaxies of cCGs (white) or their corresponding Y07 groups (green). The ID of the cCG and its corresponding Y07 group are labeled at the top-right corner of each image. | 2102.12804_5.jpg |
arxivcap_2005.06243_3 | Title: Detecting and analyzing collusive entities on YouTube | Caption: Distribution of temporal properties characterizing propagation dynamics - (a) initial delays and (b) lifetimes of Youtube collusive videos. | 2005.06243_3.jpg |
arxivcap_2303.14739_1 | Title: Geometry-Aware Attenuation Field Learning for Sparse-View CBCT Reconstruction | Caption: CBCT Imaging: (a) CBCT scanning would produce a series of (b) X-ray projections, which will be used to solve (c) a 3D CBCT image through CBCT reconstruction. | 2303.14739_1.jpg |
arxivcap_1802.01257_2 | Title: Promoting cooperation by punishing minority | Caption: Full punishment fine-multiplication factor ($\alpha-r$) phase diagram. There are three regions: full cooperators ($C$), full defectors ($D$), and the coexistence of cooperators and defectors ($C+D$). Solid line denotes continuous phase transition, while dashed line marks discontinuous phase transition. The region of $C+D$ disappears when $\alpha>0.45$. | 1802.01257_2.jpg |
arxivcap_1111.0261_4 | Title: Extended Lagrangian free energy molecular dynamics | Caption: (Color online) The fluctuations in the total nuclear temperature, $T_{\mathrm{total}}$, in comparison with the temperature of only the Bucky ball, $T_{\mathrm{C60}}$ and the temperature of the water molecule, $T_{\mathrm{water}}$. We start our simulation with the Bucky ball at $T_{\mathrm{C60}} = 0$ K and a water molecule at the center inside the Bucky ball with a non-zero initial velocity. While the total temperature remains fairly constant, the energy transfer from the water molecule to the Bucky ball is clearly visible in the rising $T_{\mathrm{C60}}$ and the falling $T_{\mathrm{water}}$. | 1111.0261_4.jpg |
arxivcap_2108.09713_2 | Title: Robustness-via-Synthesis: Robust Training with Generative Adversarial Perturbations | Caption: Robustness across varying attack budgets for (a) CIFAR10, (b) CIFAR100, and (c) SVHN. Adversarial accuracies against the PGD-20 attack of the proposed and Madry's models <cit.> are plotted for varying $\epsilon \in \left[\frac{0}{255}, \frac{20}{255}\right]$ values. The step size of the PGD attack is set to $\frac{2}{255}$. | 2108.09713_2.jpg |
arxivcap_2202.00203_11 | Title: Constraining low redshift [CII] Emission by Cross-Correlating FIRAS and BOSS Data | Caption: The binned angular power spectrum, $C_{b}(z,z')$, in the CMASS redshift range with layout and general properties similar to the LOWZ region described in Figure <ref>. The vertical and horizontal lines in the $\ell=24$ data at $z\sim 0.67$ are due to a spurious correlation between thermal noise and Galactic foregrounds. | 2202.00203_11.jpg |
arxivcap_1401.8257_1 | Title: Online Clustering of Bandits | Caption: A true underlying graph $G = (V,E)$ made up of $n = |V| = 11$ nodes, and $m = 4$ true clusters $V_1 = \{1,2,3\}$, $V_2 = \{4,5\}$, $V_3 = \{6,7,8,9\}$, and $V_4 = \{10,11\}$. There are $m_t = 2$ current clusters $\hat V_{1,t} $ and $\hat V_{2,t} $. The black edges are the ones contained in $E$, while the red edges are those contained in $E_t\setminus E$. The two current clusters also correspond to the two connected components of graph $G_t = (V,E_t)$. Since aggregate vectors $\bbw_{j,t}$ are build based on current cluster membership, if for instance, $i_t = 3$, then $\hj_t = 1$, so $\bM_{1,t-1} = I + \sum_{i = 1}^5 (M_{i,t-1}-I)$, $\bbb_{1,t-1} = \sum_{i = 1}^5 \bb_{i,t-1}$, and $\bbw_{1,t-1} = \bM_{1,t-1}^{-1}\bbb_{1,t-1}$. | 1401.8257_1.jpg |
arxivcap_2005.06243_4 | Title: Detecting and analyzing collusive entities on YouTube | Caption: Genre-wise distribution of likes, dislikes and comments of collusive videos. Full forms of the labels in the x-axis are mentioned in Section <ref>. | 2005.06243_4.jpg |
arxivcap_2202.01477_2 | Title: Unsourced Random Access with a Massive MIMO Receiver Using Multiple Stages of Orthogonal Pilots | Caption: {The required $E_b/N_0$ as a function of the number of active users in the proposed scheme and the method in <cit.> for $M=100$ and $L\approx 320, 200$.} | 2202.01477_2.jpg |
arxivcap_2202.13870_1 | Title: Simulating Network Paths with Recurrent Buffering Units | Caption: $MMD^2$ (with $\zeta = 1$) vs starting packet number (of chunks) on the dataset for (left) the training protocol Cubic and (right) the test protocol Vegas. | 2202.13870_1.jpg |
arxivcap_1709.03144_25 | Title: The 12C(a,g)16O reaction and its implications for stellar helium burning | Caption: (Color online) Comparison of the reaction rate and uncertainty calculated in this work (orange band, dash central line) and that from {0004-637X-567-1-643} (blue band, dash-dotted central line) normalized to the adopted value from {nacre} (NACRE compilation) (gray band, solid central line). The deviations at higher temperature are the result of the different narrow resonance and cascade transitions that were considered in the different works. Arrows at the bottom indicate temperature ranges for different Carbon burning scenarios. | 1709.03144_25.jpg |
arxivcap_1503.07205_1 | Title: The Observer Strikes Back | Caption: Two histories of the simple box model universe described in the text. The boxes model Hubble volumes. Their color models an observable like the CMB. An `E' means that an observer is in the box observing its color. A blank means there is no observer in the box. The third person probabilities for these histories to occur are at the right. | 1503.07205_1.jpg |
arxivcap_2205.02281_4 | Title: Three-Body Problem in Modified Dynamics | Caption: The time evolution ($100~\mathcal{T}$) of Pythagorean three-body problem <cit.> according to Newtonian and MOD models. The coordinates $x$ and $y$ are in units of $\mathcal{D}$. | 2205.02281_4.jpg |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.