image
imagewidth (px) 44
1.64k
| file_name_index
stringlengths 26
33
| text
stringlengths 16
2.81k
| class
stringclasses 5
values | super_class
stringclasses 2
values | sub_class
stringclasses 5
values | split
stringclasses 1
value |
|---|---|---|---|---|---|---|
$2305.00001v1-Figure1-1.png | Fig. 1. The projection of x onto convex set A is the unique point in A which is closest to x and is denoted as y. | fig_illustration | fig | illustration | train |
|
$2305.00001v1-Figure2-1.png | Fig. 2. Graphical interpretation of the parallel POCS for disjoint convex sets. | fig_illustration | fig | illustration | train |
|
$2305.00001v1-Figure3-1.png | Fig. 3. Face image samples of Ben Affleck from the 5 Celebrity Faces dataset. | fig_illustration | fig | illustration | train |
|
$2305.00001v1-Figure4-1.png | Fig. 4. Typical diagram of an autoencoder network. | fig_illustration | fig | illustration | train |
|
$2305.00001v1-Figure5-1.png | Fig. 5. Description of the AE model used in this study: (a) encoder, and (b) decoder. | fig_result | fig | result_fig | train |
|
$2305.00001v1-Figure6-1.png | Fig. 6. Training curves of the used AE model on the MNIST dataset. | fig_result | fig | result_fig | train |
|
$2305.00001v1-Figure7-1.png | Fig. 7. Reconstructed images on MNIST dataset using the autoencoder model described in the paper (top: input image, bottom: reconstructed image). | fig_illustration | fig | illustration | train |
|
$2305.00001v1-Table1-1.png | Table 1. Comparison between the POCS-based clustering and the K-Means++ algorithms in terms of clustering error on different feature embedding sets. | table_result | table | result_tab | train |
|
$2305.00001v1-Table2-1.png | Table 2. Comparison between the POCS-based clustering and the K-Means++ algorithms in terms of execution time (ms) on different feature embedding sets. | table_result | table | result_tab | train |
|
$2305.00001v1-Table3-1.png | Table 3. Comparison between the POCS-based clustering and the K-Means++ algorithms in terms of classification accuracy on different feature embedding sets. | table_result | table | result_tab | train |
|
$2305.00001v1-Table4-1.png | Table 4. Comparison of the K-Means, FCM, and POCSbased clustering algorithms in terms of clustering error on different feature embedding sets. | table_result | table | result_tab | train |
|
$2305.00001v1-Table5-1.png | Table 5. Comparison of the K-Means, FCM, and POCSbased clustering algorithms in terms of execution time (ms) on different feature embedding sets. | table_result | table | result_tab | train |
|
$2305.00001v1-Table6-1.png | Table 6. Comparison of the K-Means, FCM, and POCSbased clustering algorithms in terms of classification accuracy on different feature embedding sets. | table_result | table | result_tab | train |
|
$2305.00003v2-Figure1-1.png | Figure 1: Schematic of the contribution of this study. Data-driven surrogate model is developed to replace the physics-based simulator on the process-structure-property problem. | fig_result | fig | result_fig | train |
|
$2305.00003v2-Figure2-1.png | Figure 2: Finite element discretization of the orientation space for face-centered cubic (FCC) microstructures. The red-colored nodal points show the independent ODF values while the bluecolored nodes indicate the dependent ODFs as a result of the crystallographic symmetries. | fig_result | fig | result_fig | train |
|
$2305.00003v2-Figure3-1.png | Figure 3: The comparison between the predicted textures by the neural network model (a, b and c) and finite element crystal plasticity model (d, e and f) at time steps t=0.3 sec (a and d), t=0.8 sec (b and e), and t=1 sec (c and f). | fig_result | fig | result_fig | train |
|
$2305.00003v2-Figure4-1.png | Figure 4: Training error and testing error (left) on the synthetic dataset. Average stiffness error amount 31 modes (right). | fig_result | fig | result_fig | train |
|
$2305.00003v2-Figure5-1.png | Figure 5: Texture evolution prediction through the optimum processing path by the neural network surrogate model. The figure shows the different steps of deformation processing from an initial texture to a final optimum texture which maximizes an objective function defined for the homogenized elastic stiffness constants of a Copper microstructure. T and C stand for tension and compression, respectively, and XY, XZ & YZ represent the corresponding shear processes. | fig_result | fig | result_fig | train |
|
$2305.00003v2-Figure6-1.png | Figure 6: Texture evolution through the optimum processing path by the physics-based simulator. The figure shows the different steps of deformation processing from an initial texture to a final optimum texture which maximizes an objective function defined for the homogenized elastic stiffness constants of a Copper microstructure. T and C stand for tension and compression, respectively, and XY, XZ & YZ represent the corresponding shear processes. | fig_result | fig | result_fig | train |
|
$2305.00003v2-Figure7-1.png | Figure 7: A single crystal optimum texture is obtained using linear programming to maximize the homogenized elastic stiffness constants without considering processing. | fig_result | fig | result_fig | train |
|
$2305.00003v2-Table1-1.png | Table 1: The average relative L2 error/stiffness error of neural network predictions compared with FE simulator on each deformation mode. We use binary strings of length 5 to represent different surrogate networks. Each digit represents one of the deformation processing modes (tension, compression, and xy, xz, yz shear). 1 if it is applied, otherwise 0 (e.g., 10010 represents the network that learns the deformation process when tension and xz shear are simultaneously applied). | table_result | table | result_tab | train |
|
$2305.00004v1-Figure1-1.png | Figure 1: A schematic experimental layout including optical diagnostics and the laminar flow reactor. | fig_result | fig | result_fig | train |
|
$2305.00004v1-Figure2-1.png | Figure 2: A time-resolved sequence of particle ignition with tign given by the ground truth (manual labeling). (a) OH-LIF raw images. (b) binary OH-LIF images. | fig_result | fig | result_fig | train |
|
$2305.00004v1-Figure3-1.png | Figure 3: Comparison of ignition delay times by the SAS method ti,SAS and the manual label ti,gt for two particle sizes A and B in seven atmospheres. | fig_result | fig | result_fig | train |
|
$2305.00004v1-Figure4-1.png | Figure 4: Ignition time difference ti,RN - ti,gt by using ResNet-18 with the amount of particle events (a) Nev = 14, (b) Nev = 56, (c) Nev = 140, and (d) Nev = 462. | fig_result | fig | result_fig | train |
|
$2305.00004v1-Figure5-1.png | Figure 5: Ignition time difference ti,FPN - ti,gt using different ResNet models in the bottom-up pathway of FPN networks. | fig_result | fig | result_fig | train |
|
$2305.00004v1-Figure6-1.png | Figure 6: Relative probability distributions of ignition time differences ti,det - ti,gt, in which ti,det represents predicted ignition delay times by different approaches. | fig_result | fig | result_fig | train |
|
$2305.00005v2-Table1-1.png | Table 1. Study population demographics of the Río Hortega University Hospital Glioblastoma dataset (RHUH-GBM) | table_result | table | result_tab | train |
|
$2305.00009v2-Figure1-1.png | FIG. 1. Example of the experimental data used for the reconstruction experiments. (a) Epicardial snapshot and (b) endocardial snapshot at time t = 5734 ms, with (c) time traces of the recording at positions labeled by the + (epicardial) and × (endocardial) markers. The thin black curve in (a) and (b) designates the boundary of the opposing surface recording, and the thick black curve designates the boundary of their mutual intersection. The inset dashed square has side-length 3 cm, aligned to the horizontal and vertical axes, and shows the region selected to extract observations for assimilation. | fig_result | fig | result_fig | train |
|
$2305.00009v2-Figure10-1.png | FIG. 10. Uncertain wavefront experiment results, depicting the (a) CRPSa(t) (line) and CRPSo(t) (band), (b) RMSEb(t) (line) plus and minus one standard deviation (band), and (c) SPRDb(t, z) (color) for the reconstruction using the Barone et al. parameter set. | fig_result | fig | result_fig | train |
|
$2305.00009v2-Figure11-1.png | FIG. 11. (a) RMSEb(t) and (b) SPRDb u(t) for the Autonomous, Wave Uncertainty, and Synthetic observations experiments, with the distribution of (right, top) surface errors and (right, bottom) ensemble spread over time. | fig_result | fig | result_fig | train |
|
$2305.00009v2-Figure3-1.png | FIG. 3. Autonomous experiment results, depicting the (a) CRPSa(t) (line) and CRPSo(t) (band), (b) RMSEb(t) (line) plus and minus one standard deviation (band), and (c) SPRDb(t, z) (color) for the reconstruction using the Barone et al. parameter set. | fig_result | fig | result_fig | train |
|
$2305.00009v2-Figure6-1.png | FIG. 6. (a) RMSEb(t) and (b) depth-average SPRDb u(t) for the Free-Run, Autonomous, and Stimulus experiments, with the distribution of (right, top) surface errors and (right, bottom) ensemble spread over time. | fig_result | fig | result_fig | train |
|
$2305.00009v2-Figure7-1.png | FIG. 7. (a) Temporal trace for the Autonomous experiment and associated observations, sampled on the epicardium (z/d = 0) at (x, y) = (0.825, 0.42) cm, and (b) corresponding action potential durations (APD) and amplitudes (APA) (uthr = 0.1). | fig_result | fig | result_fig | train |
|
$2305.00009v2-Figure8-1.png | FIG. 8. Stochastic experiment results, depicting the (a) CRPSa(t) (line) and CRPSo(t) (band), (b) RMSEb(t) (line) plus and minus one standard deviation (band), and (c) SPRDb(t, z) (color) for the reconstruction using the Barone et al. parameter set. | fig_result | fig | result_fig | train |
|
$2305.00009v2-Figure9-1.png | FIG. 9. Synthetic Observations experiment results, depicting the (a) CRPSa(t) (line) and CRPSo(t) (band), (b) RMSEb(t) (line) plus and minus one standard deviation (band), and (c) SPRDb(t, z) (color) for the reconstruction using the Barone et al. parameter set. | fig_result | fig | result_fig | train |
|
$2305.00009v2-TableI-1.png | TABLE I. Fenton-Karma model parameter values used in dynamical model for the reconstruction of experimental data. ∗Note: as τ− v1 ≡ τ− v2, the switching parameter uv is unspecified in Ref. 1. | table_result | table | result_tab | train |
|
$2305.00011v1-Figure1-1.png | Fig. 1. Problem setup diagram where speech privacy is violated while being sent to a cloud. | fig_result | fig | result_fig | train |
|
$2305.00011v1-Figure2-1.png | Fig. 2. Schematic diagram of the proposed method. F , C, D, and D′ are neural networks and L shows different loss terms in our method. The solid lines illustrate the forward pass. The dashed line shows the forward pass which is active only after P number of epochs. Finally, the dotted line shows the backpropagation of each specific error w.r.t the relevant weights. | fig_result | fig | result_fig | train |
|
$2305.00011v1-Figure3-1.png | Fig. 3. The effect of parameter P on the performance of RDAL on the test data for SAD and SED tasks. | fig_result | fig | result_fig | train |
|
$2305.00011v1-Figure4-1.png | Fig. 4. Estimated density curves using Gaussian kernel to represent continuous probability densities of the predicted probabilities for the test data using baseline (left figure) and RDAL (right figure) methods. | fig_result | fig | result_fig | train |
|
$2305.00011v1-Figure5-1.png | Fig. 5. t-SNE illustration of latent features of the test data obtained by F in RDAL (right figure) compared to the features from F when it is trained in supervised manner for sound events and speech(left figure). The dog barking, glass breaking and gun shot are red, green and blue, respectively. The samples containing speech are marked by “o” and the non-speech samples by “x”. | fig_result | fig | result_fig | train |
|
$2305.00011v1-Table1-1.png | Table 1. Number of one-second sound event samples in each split, number of male/female speakers, and the duration of speech (in seconds) in our dataset. | table_parameter | table | parameter | train |
|
$2305.00011v1-Table2-1.png | Table 2. Results of baseline, naive adversarial learning, and RDAL. | table_result | table | result_tab | train |
|
$2305.00013v1-Figure12-1.png | FIG. 12: Best constraint on the branching fraction BR(h → γDγD) using the Higgs diphoton resonance search, four-photon resonant search, Higgs mixing angle limits, and Higgs branching ratio to the unknown. The resonance searches require 1% detector efficiency. The contours are upper bounds on BR(h → γDγD) while the heat map is for log10BR(h → γDγD). We assume BR(a→ γγ) = BR(γD → aγ) = 1. | fig_result | fig | result_fig | train |
|
$2305.00013v1-Figure2-1.png | FIG. 2: Maximum branching ratio of the Higgs to two dark photons from (dashed) Higgs fits to the branching fraction of Higgs to the unknown and from (solid) Higgs mixing assuming sin θmax = 0.1 with (black) vD = mγD , (red) vD = 10mγD , (blue) vD = 25mγD , (yellow) vD = 50mγD , and (green) vD = 100mγD . | fig_result | fig | result_fig | train |
|
$2305.00013v1-Figure4-1.png | FIG. 4: Regions with the largest probability of ending in n observed photon final states after merging photons into γobs: (blue) two, (orange) three, (green) four, (purple) five, and (red) six observed photons. No isolation requirements are imposed. The black corner indicates the kinematically forbidden area (mγD < ma). | fig_architecture | fig | architecture | train |
|
$2305.00013v1-Figure5-1.png | FIG. 5: Regions with the largest probabilities into n isolated photons and m ξ-jet final states. The light gray (dark gray) region indicates the dominant final states have one (two) ξ-jets. Color coding for other regions: (blue) 2γiso + 0ξ-jet, (green) 4γiso + 0ξ-jet, and (red) 6γiso + 0ξ-jet. | fig_result | fig | result_fig | train |
|
$2305.00013v1-Figure6-1.png | FIG. 6: Estimated detector efficiencies for n photon triggers applied to n isolated photons and zero ξ-jet signals for both (a) ATLAS and (b) CMS. Efficiencies are estimated by placing rapidity [Eqs. (27)-(28)] and transverse momentum requirements on the isolated photons (Tabs. I, II). | fig_result | fig | result_fig | train |
|
$2305.00013v1-Figure7-1.png | FIG. 7: (a) Probability of a signal appearing as a 2γiso + 0ξ-jet final state. (b) Upper bound on the branching fraction BR(h → γDγD) from Higgs diphoton signal strength. A detector efficiency of Eff2(6γ → 2γiso + 0ξ) ≥ 1% is required. In (b) the contours are upper bounds on BR(h → γDγD) while the heat map is for log10BR(h → γDγD). We assume BR(a→ γγ) = BR(γD → aγ) = 1. | fig_result | fig | result_fig | train |
|
$2305.00013v1-Figure8-1.png | FIG. 8: (a) Probability of four-isolated photon + zero-ξ-jet final state. (b) Constraint (dark area) on the branching fraction BR(h→ γDγD) using the Higgs four-photon resonant search from CMS [27] and requiring 1% detector efficiency for the CMS four-photon trigger. In (b) the contours are upper bounds on BR(h→ γDγD) while the heat map is for log10BR(h→ γDγD). We assume BR(a→ γγ) = BR(γD → aγ) = 1. | fig_result | fig | result_fig | train |
|
$2305.00014v1-Figure1-1.png | Figure 1: Left panel: The parameter space excluded by searches for the LFV decays of Higgs boson. The dark gray area is excluded by the h → eµ searches [1], the medium gray area is excluded by the h → eτ searches [5] and the light gray area by the h → µτ searches [5]. Right panel: The parameter regions excluded by the measurements of BR(h→ µµ) (blue area) and BR(h→ ττ) (red area). | fig_result | fig | result_fig | train |
|
$2305.00014v1-Figure2-1.png | Figure 2: Left panel: Cross-section σ(pp → H2 → eµ) as a function of flavon VEV at 13 TeV LHC for different Higgs-flavon mixing angles. Right panel: H2 → eµ branching ratio as a function of flavon VEV. In both panels solid lines correspond to numerical benchmark and the dashed lines correspond to analytical estimate with maximal eµ coupling. In both panels the red area is excluded by BR(h → ττ) and BR(h → µµ) measurements for sin θ = 0.1. For sin θ = 0.2 red and green areas are excluded, and for sin θ = 0.3 all colored areas are excluded. | fig_result | fig | result_fig | train |
|
$2305.00014v1-Figure3-1.png | Figure 3: Left and middle panels: The 1- and 2-loop contributions to µ→ eγ in the case of additional quark couplings. Right panel: The tree-level contribution to µ↔ e conversion in nuclei in the case of additional quark couplings. | fig_result | fig | result_fig | train |
|
$2305.00014v1-Figure4-1.png | Figure 4: Left panel : Regions of the (vφ, mA) parameter space excluded by searches for µ → eγ in the leptophilic model for different values of the mixing angle. Right panel : The LFV constraints in the case of additional quark-flavon couplings for ct = 1.9 and three different values of the charm coupling, cc = −1.9, 0 and 1. The allowed white wedge shaped area continues until the corner of the orange area, but is too fine to be visible by eye. | fig_result | fig | result_fig | train |
|
$2305.00014v1-Figure5-1.png | Figure 5: Left panel: Cross-section σ(pp→ H2 → eµ) as a function of the flavon VEV at the √ s = 13 TeV LHC for different quark couplings as indicated in the figure. The gray shaded region is excluded by µ ↔ e conversion for all values of mA. The horizontal line corresponds to 5.77 fb. Right panel: The relevant branching ratios for the same case. The solid lines correspond to ct = −cc = 1.9, dashed line to 1.7 and dot dashed line to 1.5. | fig_result | fig | result_fig | train |
|
$2305.00015v1-Figure1-1.png | FIG. 1. Neutron skin and collective flow in relativistic nuclear collisions. a: Two ions collide with impact parameter b = 8 fm. Both ions are Lorentz-contracted by a factor γ ≈ 2500, and the relevant dynamics hence effectively takes place in the transverse plane, x⊥ = (x, y). b: The collision deposits energy in the interaction region depending on the extent of the neutron skin of the 208Pb nuclei. We consider ∆rnp = 0.086 fm (top) and ∆rnp = 0.384 (bottom). The neutron skin is varied by keeping the half-width neutron radius, Rn, constant while changing the neutron diffuseness, as displayed by the dotted lines (see also Eqn. (2) below). A larger neutron skin leads to a considerably larger total hadronic cross section, σtot, and the resulting QGP is in addition more diffuse and less elliptical. c: We show a single QGP evolving hydrodynamically and being converted into particles (marked in the figure with their respective symbols) as it cools, while expanding both in z and in the transverse plane. The observation of millions such events leads to characteristic azimuthal anisotropies in the momentum distribution of the produced particles, the most important of which is quantified by the rms value of its second Fourier component, the elliptic flow v2{2}, which reflects the ellipticity of the QGP. | fig_result | fig | result_fig | train |
|
$2305.00015v1-Figure2-1.png | FIG. 2. Signature of the neutron skin on bulk particle production in ultrarelativisitic 208Pb+208Pb collisions. Varying only the neutron skin size at our optimal parameter settings we show the charged particle multiplicity (left), the mean transverse momentum (middle) and the elliptic flow as measured by v2{k} (right) with a comparison to ALICE data [19, 20]. A larger neutron skin leads to more collisions, but per collision the multiplicity is lower at larger centralities. The larger size of the QGP leads to a reduced transverse momentum on average. Smearing of the elliptical shape leads to reduced elliptic flow as measured by v2{2} and v2{4}. Theoretical error bars are statistical only, experimental uncertainties include systematics as well. | fig_result | fig | result_fig | train |
|
$2305.00015v1-Figure4-1.png | FIG. 4. State-of-the-art determinations of the neutron skin of 208Pb. We show the final likelihood distribution of the neutron skin as determined from the LHC data as compared to the values obtained experimentally by the PREX collaboration (including both experimental and theoretical uncertainties in the extraction) [6] and the estimate of ab initio nuclear theory (with an error bar corresponding to a 68% credibility interval) [7]. | fig_result | fig | result_fig | train |
|
$2305.00015v1-Figure5-1.png | FIG. 5. Complete correlation matrix among all 26 model parameters. For detailed information about the prior ranges, we refer the reader to Ref. [18]. The only new parameters of this analysis are an, whose prior can be inferred from Fig. 4, aEOS, whose | fig_illustration | fig | illustration | train |
|
$2305.00017v1-Figure1-1.png | FIG. 1. Left : Number of biνos produced at SHiP assuming 5 years of operation and ΛM = 1 TeV. (This messenger scale is taken as a benchmark for comparison to other relevant models.) Right : Branching ratios of the biνo into different final states as a function of the biνo mass MB̃ . | fig_result | fig | result_fig | train |
|
$2305.00017v1-Figure2-1.png | FIG. 2. SHiP sensitivity to biνos assuming 5 years of operation using our conservative method (solid black curve) and the HNL@SHiP method (dashed black curve). The greater sensitivity with the conservative method below MB̃ ∼ 400 MeV is due to the biνo production from kaons, which is not included in the HNL@SHiP package. Existing limits from µ→ eγ [31] are depicted by the gray shaded region below the dashed gray line labelled “µ → eγ”. We also show the projected exclusion limit from the proposed Mu2e experiment [32], which will look for µ→ e conversion in nuclei. BBN constraints are shown by the orange shaded regions. See text for further details. | fig_result | fig | result_fig | train |
|
$2305.00019v1-Figure1-1.png | Figure 1. Diagram illustrating the general processes that occur during and after lobe inflation. Panel A: A fast jet drives into the ambient medium, forms a bow shock and inflates a hot lobe that expands into the ICM. The lobe morphology can depend sensitively on the injected jet properties (e.g., content, velocity, geometry). The expanding shock wave results in a layer of shocked ICM material surrounding the jet lobe. Panel B: As the lobe expansion slows and the jet ceases, the shock driven into the ICM broadens into a sound wave that can detach from the lobes and continue to buoyantly rise through the ICM. Panel C: As the lobe rises dense, low entropy material can be entrained and pulled up in the wake and instabilities can lead to mixing of the lobe and ICM material. Sound waves generated by the lobe expansion can continue to propagate to large distances depending on the ICM viscosity. Panel D: This process continues at late times, with mixing continuing to dilute the lobe material, although the rate at which this occurs can depend sensitively on the ICM physical processes including magnetic fields, viscosity and cluster weather. | fig_result | fig | result_fig | train |
|
$2305.00019v1-Figure2-1.png | Figure 2. Thin temperature projections illustrate how jet injection parameters impact jet and lobe morphologies. All jets are 100 Myr old with each row illustrating the effect of changing one parameter, which from top to bottom are jet power, half opening angle, velocity and resolution, respectively. These quantities take fiducial values of 1046 erg s−1, 10◦, 15000 km s−1 and 1.81× 105 M , unless being varied. (Figure 6 from Huško and Lacey [47], CC BY 4.0) | fig_result | fig | result_fig | train |
|
$2305.00019v1-Figure3-1.png | Figure 3. An illustration of jet-driven shocks within the central 100 kpc of a simulated cluster. The left panel shows the energy dissipation rate while the right-hand panel shows shock Mach numbers, illustrating that stronger shocks and hence higher dissipation rates are seen closer to the jet but that overall the shocks are typically quite weak. (Figure 1 from Li et al. [72], ©AAS, reproduced with permission) | fig_result | fig | result_fig | train |
|
$2305.00019v1-Figure4-1.png | Figure 4. Volume rendering of jet lobes (green) and cold material (red) within a cosmologically evolved galaxy cluster. The top, middle and bottom rows show low, medium and high-power jets, respectively. The small panels show the evolution of the jet lobes for two different viewing angles, while the large panels additionally overlay the gas velocity field. Overall, jet lobes can be displaced, disrupted and mixed by cluster weather and cold structures, with lower-power jets being more susceptible. (Figure 2 from Bourne and Sijacki [173], CC BY 4.0) | fig_result | fig | result_fig | train |
|
$2305.00019v1-Figure5-1.png | Figure 5. Comparisons of simulation results with varied jet composition and assumptions for modeling CR transport. Rows from top to bottom show the results from kinetic-energy dominated jets (KIN), CR dominated jets (CR), and CR dominated jets with diffusion and heating (CRdh). The morphology of jet-inflated lobes tend to be elongated for the KIN case, whereas bubbles inflated by CR dominated jets are wider. Comparisons between the bottom two panels show that CR heating is more efficient and the amount of cold gas is less for the CRdh case, as the CRs can diffuse outside the bubbles and heat the ICM due to hadronic and Coulomb interactions. (Figure 3 from Yang et al. [66], ©AAS, reproduced with permission) | fig_result | fig | result_fig | train |
|
$2305.00019v1-Figure6-1.png | Figure 6. Impact of assumptions about ICM viscosity on the evolution of AGN jet-inflated bubbles. Cases A-D show simulations with no viscosity, isotropic viscosity with full Spitzer values, anisotropic viscosity with full Braginskii values, and anisotropic viscosity limited by microinstabilities, respectively. While the hydrodynamic instabilities are suppressed by viscosity in cases B and C, when the parallel viscosity along magnetic field lines is suppressed by microinstabilities (case D), the viscosity is strongly limited and the bubbles are deformed as in the inviscid case (A). This illustrates the importance of modelling the ICM microphysics in AGN simulations. (Figure 1 from Kingsland et al. [93], ©AAS, reproduced with permission) | fig_result | fig | result_fig | train |
|
$2305.00019v1-Figure7-1.png | Figure 7. Illustration of sound waves generated by jets. The left-hand panel shows the jet and cocoon temperature and density structure, with key features labelled. The right-hand panel shows jet entropy and the acoustic flux density, with structure in the latter illustrating the production of sound waves within shocked ICM material. (Figure 1 from Bambic and Reynolds [184], ©AAS, reproduced with permission) | fig_result | fig | result_fig | train |
|
$2305.00020v2-Figure1-1.png | Fig. 1: Intensity-weighted mean velocity (first moment) maps of H2CO (30,3 − 20,2) showing the large-scale kinematics of the full CORE sample. The contours correspond to the continuum maps imaged with uniform weighting as presented in Beuther et al. (2018). The outermost three contours correspond to 5, 10, and 20σ levels, then increasing in steps of 15σ. For IRAS23151, NGC7538 IRS1, NGC7538 IRS9 and AFGL2591 the outermost three contours correspond to 5, 15, and 40σ levels, then increasing in steps of 25σ (see Table 3 for σ values). The synthesised beam is shown in the bottom left corner and a scale bar in the bottom | fig_result | fig | result_fig | train |
|
$2305.00020v2-Figure10-1.png | Fig. 10: Median Q plotted against gas mass (left) and stellar mass (right) for 13 candidate disks within the CORE survey, coloured according to the luminosity of the regions within which they reside. The Toomre-stable disks are marked by triangles. | fig_result | fig | result_fig | train |
|
$2305.00020v2-Figure11-1.png | Fig. 11: Median Q as a function of ( H r ) ( M∗+Mdisk Mdisk ) . The Toomrestable disks are marked by triangles. | fig_result | fig | result_fig | train |
|
$2305.00020v2-Figure12-1.png | Fig. 12: Summary plot showing the kinematics and derived properties for one source in the CORE sample (G75.78). Top left panel: intensity-weighted mean velocity (first moment) map of CH3CN (125 − 115) in colour with 1.37 mm continuum contours. The blue and red arrows correspond to the estimated directions of bipolar blueshifted and redshifted molecular outflows, respectively. The dotted line indicates the position of the strongest velocity gradient tracing the disk, i.e. perpendicular to the rotation axis. The edges of the assumed disk extent are marked with an ×. Top right panel: PV plot of CH3CN (125 − 115) along the cut in the direction of rotation as depicted by the dotted line in the top left panel. The contours correspond to the 6σ level increasing in steps of 24σ. Yellow lines show the Keplerian rotation curves for an enclosed mass of 19 M . The cross in the bottom right corner corresponds to the spatial and spectral resolutions. Bottom left panel: Rotational temperature maps obtained by fitting CH3CN (12K−11K) K = 0−6 and CH3 13CN (12K −11K) K = 0−3 lines with XCLASS in colour and 1.37 mm continuum contours. Bottom right panel: Toomre Q map assuming a protostar is located at the position of the continuum peak as depicted by a star and accounting for the self-gravity of the disk, with 1.37 mm continuum contours. Regions outside of 6σ continuum contours are masked out. The synthesised beam and a scale bar are shown in the bottom of the panels. | fig_result | fig | result_fig | train |
|
$2305.00020v2-Figure2-1.png | Fig. 2: Intensity-weighted mean velocity (first moment) maps of CH3CN (123 − 113) showing the dense gas kinematics for 15 of the 20 sources in the CORE survey. The contours correspond to the 1.37 mm continuum as described in Fig. 1. The blue and red arrows correspond to the estimated directions of bipolar blueshifted and redshifted molecular outflows, respectively (see Sect. 4.3). The dotted lines indicate the position of the strongest velocity gradient tracing the disk, i.e. perpendicular to the rotation axis. The edges of the assumed disk extent are marked with an ×. The synthesised beam is shown in the bottom left corner and a scale bar in | fig_result | fig | result_fig | train |
|
$2305.00020v2-Figure3-1.png | Fig. 3: Rotational temperature maps obtained by fitting CH3CN (12K − 11K) K = 0 − 6 and CH3 13CN (12K − 11K) K = 0 − 3 lines with XCLASS. The contours correspond to the 1.37 mm continuum as described in Fig. 1. For NGC7538 IRS1, only the region outside the continuum to the south-west is modelled by XCLASS and was scaled towards the continuum peak position following a temperature power-law distribution T ∝ r−0.4. The blue and red arrows correspond to the estimated directions of bipolar blueshifted and redshifted molecular outflows, respectively. The synthesised beam is shown in the bottom left corner and a scale bar in the bottom right corner of each panel. Maps of column density, velocity offset, linewidth, and source size are shown in Appendix E. | fig_result | fig | result_fig | train |
|
$2305.00020v2-Figure4-1.png | Fig. 4: Position–velocity (PV) plots for best fitted transitions listed in Table 5 along cuts in the direction of rotation as depicted by dotted lines in Fig. 2. The width of the cut is the size of a synthesised beam to increase the signal-to-noise ratio. The PV plots of W3(H2O) E and W3(H2O) W make use of the A-array only observations in order to gain more resolution and disentangle the two cores from the circumbinary material. PV plots for all other sources make use of the full dataset observed in the CORE survey. The contours correspond to the 6σ level increasing in steps of 24σ. Yellow lines show the Keplerian rotation curves for enclosed masses, MPV, listed in Table 5 for each source. The inner radii that were masked out (i.e. excluded) in the fitting routine are shown as vertical dotted lines. The fitted central velocities and positions are shown as dashed lines. The zero offset corresponds to the continuum peak position. The crosses in the bottom right corners correspond to the spatial and spectral resolutions. | fig_result | fig | result_fig | train |
|
$2305.00020v2-Figure5-1.png | Fig. 5: Ratio of free-fall to rotational timescale as a function of disk gas mass, showing most of the sources with CH3CN velocity gradients are rotationally supported. The blue and red dots correspond to the values calculated based on the rotational velocities at the edges of the disks on the blueshifted and redshifted sides, as depicted by crosses in Fig. 2 and listed in Table 3. The black dashed line corresponds to theoretical curve of tff/trot for a disk of mass Mgas in Keplerian rotation about a star with mass M∗ = 10 M (see Eq. 4). | fig_illustration | fig | illustration | train |
|
$2305.00020v2-Figure6-1.png | Fig. 6: Specific angular momentum radial profiles calculated using Eq. 7 along the cut with the strongest velocity gradient (dotted lines in Fig. 4) (circles), a cut with position angle +10◦ (triangles pointing up), and a cut with position angle −10◦ (triangles pointing down) with respect to the strongest velocity gradient, for the redshifted (red) and blueshifted (blue) sides of the emission. The blue and red filled regions show the error in j sin(i) for the circles. Note that the inclinations are not corrected for as they are not known. The blue and red dashed lines show least-squared power-law fits to the first two data points and extrapolated to larger radii. The values of the fitted slopes are noted in the legends and listed in Table 6 along with the intercepts. The blue and red dotted vertical lines correspond to radii associated with the edges of the disks depicted with × markers in Fig. 2. The black curve shows the maximum specific angular momentum calculated using Eq. 8 for M∗ listed in the sub-panels with the grey region corresponding to M∗ ± 5 M . The grey dashed line shows the j sin(i) ∝ r1.7 slope. | fig_result | fig | result_fig | train |
|
$2305.00020v2-Figure7-1.png | Fig. 7: Toomre Q maps for the best disk candidates in the CORE survey assuming a protostar is located at the position of the continuum peak as depicted by a star and accounting for the self-gravity of the disk (protostellar and gas mass values are listed in Table 7). The contours correspond to the 1.37 mm continuum as described in Fig. 1. Regions outside of 6σ continuum contours are masked out. Fragments detected by ancillary observations are marked by plus symbols in green for AFGL2591 (Suri et al. 2021), IRAS23385 (Cesaroni et al. 2019), and NGC7538 IRS1 (Beuther et al. 2017). The synthesised beam is shown in the bottom left corner and a scale bar in the bottom right corner of each panel. | fig_result | fig | result_fig | train |
|
$2305.00020v2-Figure8-1.png | Fig. 8: Distribution of median and minimum Q as a function of radius shown in solid and dashed blue lines. The dotted horizontal line corresponds to the global median Q computed over the entire disk and listed in Table 7. The dash-dotted red horizontal line shows the critical Q = 2 threshold. | fig_result | fig | result_fig | train |
|
$2305.00020v2-Figure9-1.png | Fig. 9: Box-plot showing the Q distribution versus the ratio of gas to stellar mass, coloured according to the luminosity of the regions within which they reside. The boxes extend from the first to the third quartiles, with a red line at the median Q value. The whiskers extend from the box by ±1.5 times the box size. Disks with Q . 2 are Toomre unstable (i.e. all but Cep A & NGC7538 IRS9). | fig_result | fig | result_fig | train |
|
$2305.00020v2-FigureA.1-1.png | Fig. A.1: Integrated intensity (zeroth moment) maps of CH3CN (123 − 113) showing the dense gas distribution for 15 of the 20 sources in the CORE survey. The contours and features are as described in Fig. 2. | fig_result | fig | result_fig | train |
|
$2305.00020v2-FigureA.2-1.png | Fig. A.2: Intensity-weighted velocity dispersion (second moment) maps of CH3CN (123 − 113) showing the dense gas kinematics for 15 of the 20 sources in the CORE survey. The contours and features are as described in Fig. 2. | fig_illustration | fig | illustration | train |
|
$2305.00020v2-FigureB.1-1.png | Fig. B.1: Peak intensity (amplitude) maps of CH3CN (123 − 113) showing the dense gas kinematics for 15 of the 20 sources in the CORE survey obtained by fitting Gaussian profiles to their spectra. The contours and features are as described in Fig. 2. | fig_illustration | fig | illustration | train |
|
$2305.00020v2-FigureB.2-1.png | Fig. B.2: Peak velocity maps of CH3CN (123 − 113) showing the dense gas kinematics for 15 of the 20 sources in the CORE survey obtained by fitting Gaussian profiles to their spectra. The contours and features are as described in Fig. 2. | fig_illustration | fig | illustration | train |
|
$2305.00020v2-FigureB.3-1.png | Fig. B.3: Linewidth (FWHM) maps of CH3CN (123 − 113) showing the dense gas kinematics for 15 of the 20 sources in the CORE survey obtained by fitting Gaussian profiles to their spectra. The contours and features are as described in Fig. 2. | fig_illustration | fig | illustration | train |
|
$2305.00020v2-FigureC.1-1.png | Fig. C.1: Intensity maps of CO (2–1) emission from IRAM 30-m telescope integrated over the blueshifted and redshifted wings of emission, showing the outflow structure. The position of the strongest source in the field is depicted by a star. The blue and red arrows correspond to the estimated directions of bipolar blueshifted and redshifted molecular outflows, respectively. A scale-bar is shown in the bottom right corner of each panel. The map size of the IRAM 30-m observations is 1.5′ by 1.5′ with a half-power beam width of ∼11′′ at this frequency. | fig_result | fig | result_fig | train |
|
$2305.00020v2-FigureC.2-1.png | Fig. C.2: Intensity maps of 13CO (2 − 1) emission from IRAM 30-m telescope integrated over the blueshifted and redshifted wings of emission, showing the outflow structure. The position of the strongest source in the field is depicted by a star. The blue and red arrows correspond to the estimated directions of bipolar blueshifted and redshifted molecular outflows, respectively. A scale-bar is shown in the bottom right corner of each panel. The map size of the IRAM 30-m observations is 1.5′ by 1.5′ with a half-power beam width of ∼11′′ at this frequency. | fig_result | fig | result_fig | train |
|
$2305.00020v2-FigureC.3-1.png | Fig. C.3: Intensity maps of 13CO (2 − 1) emission from merged NOEMA and IRAM 30-m data integrated over the blueshifted and redshifted wings of emission, showing the outflow structure. The position of the strongest source in the field is depicted by a star. The blue and red arrows correspond to the estimated directions of bipolar blueshifted and redshifted molecular outflows, respectively. A scale-bar is shown in the bottom right corner of each panel. We did not merge the data for the pilot sources as we did not observe this target with the most compact NOEMA configuration (D-array), hence the missing panels. | fig_result | fig | result_fig | train |
|
$2305.00020v2-FigureC.4-1.png | Fig. C.4: The greyscale corresponds to the 1.37 mm continuum while the blue and red contours correspond to the NOEMA intensity maps of 13CO (2 − 1) integrated over the blueshifted and redshifted wings of emission, tracing either outflows or disk winds.The blue and red arrows correspond to the estimated directions of bipolar blueshifted and redshifted molecular outflows, respectively. A scale-bar is shown in the bottom right corner of each panel. Note that most of the emission is filtered out by the interferometer.Article number, page 36 of 46 | fig_result | fig | result_fig | train |
|
$2305.00020v2-FigureC.5-1.png | Fig. C.5: Intensity maps of C18O (2 − 1) emission from IRAM 30-m telescope integrated over the blueshifted and redshifted wings of emission, showing the outflow structure. The position of the strongest source in the field is depicted by a star. The blue and red arrows correspond to the estimated directions of bipolar blueshifted and redshifted molecular outflows, respectively. A scale-bar is shown in the bottom right corner of each panel. The map size of the IRAM 30-m observations is 1.5′ by 1.5′ with a half-power beam width of ∼11′′ at this frequency. C18O is not a good outflow tracer for all sources, hence the lack of contours in some targets. | fig_result | fig | result_fig | train |
|
$2305.00020v2-FigureC.6-1.png | Fig. C.6: Intensity maps of SO (65 − 54) emission from IRAM 30-m telescope integrated over the blueshifted and redshifted wings of emission, showing the outflow structure. The position of the strongest source in the field is depicted by a star. The blue and red arrows correspond to the estimated directions of bipolar blueshifted and redshifted molecular outflows, respectively. A scale-bar is shown in the bottom right corner of each panel. The map size of the IRAM 30-m observations is 1.5′ by 1.5′ with a half-power beam width of ∼11′′ at this frequency. | fig_result | fig | result_fig | train |
|
$2305.00020v2-FigureC.7-1.png | Fig. C.7: The distribution of absolute difference between the disk position angles and the assumed outflow position angles for our sample of 13 disk candidates in the CORE survey. | fig_result | fig | result_fig | train |
|
$2305.00020v2-Table1-1.png | Table 1: Positions and properties of the CORE sample, grouped in track-sharing pairs, adapted from Beuther et al. (2018). | table_result | table | result_tab | train |
|
$2305.00020v2-Table2-1.png | Table 2: Frequency setup of the narrow-band correlator and important lines covered. | table_result | table | result_tab | train |
|
$2305.00020v2-Table3-1.png | Table 3: Observational parameters for the disk candidates within the CORE survey. | table_result | table | result_tab | train |
|
$2305.00020v2-Table4-1.png | Table 4: Gas mass estimates. | table_result | table | result_tab | train |
|
$2305.00020v2-Table5-1.png | Table 5: KeplerFit parameters and dynamical mass estimates. | table_result | table | result_tab | train |
|
$2305.00020v2-Table6-1.png | Table 6: Fit parameters to the specific angular momentum radial profiles shown in Fig. 6 (dashed lines). | fig_result | fig | result_fig | train |
|
$2305.00020v2-Table7-1.png | Table 7: Overview of mass estimates and Toomre Q results. | table_result | table | result_tab | train |
End of preview. Expand
in Dataset Viewer.
Dataset Card for "SciMMIR_dataset"
SciMMIR
This is the repo for the paper SciMMIR： Benchmarking Scientific Multi-modal Information Retrieval.
In this paper, we propose a novel SciMMIR benchmark and a corresponding dataset designed to address the gap in evaluating multi-modal information retrieval (MMIR) models in the scientific domain.
It is worth mentioning that we define a data hierarchical architecture of "Two subsets, Five subcategories" and use human-created keywords to classify the data (as shown in the table below).
As shown in the table below, we conducted extensive baselines (both fine-tuning and zero-shot) within various subsets and subcategories.
For more detailed experimental results and analysis, please refer to our paper SciMMIR.
Dataset
Our SciMMIR benchmark dataset used in this paper contains 537K scientific image-text pairs which are extracted from the latest 6 months' papers in Arxiv (2023.05 to 2023.10), and we will continue to expand this data by extracting data from more papers in Arxiv and provide larger versions of the dataset.
The datasets can be obtained from huggingface Datasets m-a-p/SciMMIR, and the following codes show how to use it:
import datasets
ds_remote = datasets.load_dataset("m-a-p/SciMMIR")
test_data = ds_remote['test']
caption = test_data[0]['text']
image_type = test_data[0]['class']
image = test_data[0]['image']
Codes
The codes of this paper can be found in our Github
Potential TODOs before ACL
TODO: case study table
TODO: statistics of the paper fields (perhaps in appendix)
TODO: See if it's possible to further divide the "Figure Results" subsets.
Citation
@misc{wu2024scimmir,
title={SciMMIR: Benchmarking Scientific Multi-modal Information Retrieval},
author={Siwei Wu and Yizhi Li and Kang Zhu and Ge Zhang and Yiming Liang and Kaijing Ma and Chenghao Xiao and Haoran Zhang and Bohao Yang and Wenhu Chen and Wenhao Huang and Noura Al Moubayed and Jie Fu and Chenghua Lin},
year={2024},
eprint={2401.13478},
archivePrefix={arXiv},
primaryClass={cs.IR}
}
- Downloads last month
- 550