dedup-isc-ft-v107-score
float64 0.32
1
| uid
stringlengths 32
32
| text
stringlengths 0
32.5k
| paper_id
stringlengths 1
14
| original_image_filename
stringlengths 5
222
|
|---|---|---|---|---|
0.512813 | 0aa73067635e4ab58ac0f3efc2905607 | Π^4 5μνδ | 2012.14431 | 4d4.jpg |
0.44045 | 9c747c15159b443e8125d61b92a4910d | axial -A_α_2 -A_α_3 -e-e vertex | 2012.14431 | aV.jpg |
0.386229 | 489903e6202b472b8f24b6f04cd1c3c4 | axial-e-e vertex | 2012.14431 | av3.jpg |
0.442331 | ffd84b1c17c74807bfa0b97b69092b74 | axial -A_ν -e-e vertex | 2012.14431 | av4.jpg |
0.446117 | 122093395b154edea940d716aa56ada3 | Electron propagator | 2012.14431 | electron.jpg |
0.393639 | f1469fd59eed4314ac38707b69b19a1c | e-e-A_μ vertex | 2012.14431 | v3.jpg |
0.448773 | e75816f4ae9a4c52a42f7234a60d1f74 | Marangoni forces F̅_ M/π Pe and F̅_ M, tot/π Pe for constant flux boundary conditions as a function of U̅ in the presence of an anisotropy β̅ in the emission. All results are from numerical FEM solutions of the axisymmetric diffusion-advection equation in two-dimensional angular representation with ρ<R̅=30. The solid lines for U̅<1 are the analytical perturbative results (<ref>) and (<ref>). The solid lines for U̅>1 are the scaling results (<ref>) and (<ref>). For β̅=0, we recover the results from Fig. <ref>. | 2012.14433 | FMan.jpg |
0.48015 | 8adcacf834ad412fb474b130d365d925 | Marangoni forces F̅_ M/π Pe and F̅_ M, tot/π Pe for constant flux boundary conditions and F̅_ M, tot/π Pe for constant concentration boundary conditions as a function of U̅ in the decoupled limit Pe≪U̅. All results are from numerical FEM solutions of the axisymmetric diffusion-advection equation in two-dimensional angular representation with ρ<R̅=30. | 2012.14433 | FUAB3.jpg |
0.475449 | ffc854e491f045e0bc8a8d96d7c338ad | Left: Iterative three-dimensional FEM results for F̅_ M/π Pe (top) and F̅_ M,tot/π Pe (bottom) as a function of U̅ for Pe = 0-50 for a cubic system with -7 < y̅ <0, 0<x̅<7, -7<z̅<7. Blue open circles are results for Pe=0 from FEM solutions to the axisymmetric diffusion-advection equation in two-dimensional angular representation with R̅=30. The slope in the linear response regime for U̅≪ 1 is reduced according to Eq. (<ref>). Results for U̅≫ 1 are essentially not affected by strong Marangoni flows Pe≫U̅. Right: Corresponding slopes F̅_ M/U̅π Pe and F̅_ M,tot/U̅π Pe as a function of Pe calculated from the results for U̅=0.1. | 2012.14433 | FUPer7low.jpg |
0.418477 | 8b27f87019644ee9b6b921b8504a830f | Average Nusselt number as a function of U̅ for constant flux and constant concentration boundary conditions. We compare results from numerical FEM solutions of the axisymmetric diffusion-advection equation in two-dimensional angular representation with ρ<R̅=30 and from numerical solutions in Legendre representation with Legendre components up to n=61 on a larger domain ρ<R̅=300. | 2012.14433 | NusseltU3_Leg.jpg |
0.481653 | f1fa47add94d4f3ab05e3ed18edab583 | Normalized local Nusselt number Nu(θ)/ Nu (see text) as a function of θ for constant flux (left) and constant concentration (right) boundary conditions. Colored results are from numerical FEM solutions of the axisymmetric diffusion-advection equation in two-dimensional angular representation with ρ<R̅=30. Black lines are the exact analytical result (<ref>) and the approximate analytical result (<ref>). | 2012.14433 | Nusseltnorm_local.jpg |
0.410307 | 5a7e075126ac491caceb9d4da106ceb4 | Contour plots of the concentration c̅(ρ⃗) and the stream lines of the Marangoni flow field v̅⃗̅_ M(ρ⃗)/ Pe (in the comoving frame) for constant flux boundary conditions (A) and U̅=0.1,10 and Pe = 0.1,20,40 from numerical iterative three-dimensional FEM result for a half-cylindrical region (0<ρ<8, x̅>0, -6 < y̅ <0). After division by Pe the Marangoni flow field v̅⃗̅_ M(ρ⃗)/ Pe is rather independent of Pe suggesting that v̅⃗̅_ M(ρ⃗)/ Pe essentially depends on U̅. The Marangoni flow forms a roll. | 2012.14433 | figflows.jpg |
0.399623 | 459d96a698b04dbab9f5923a61554b2b | Concentration profiles in the z̅x̅-plane for (A) constant flux and (B) constant concentration boundary conditions. | 2012.14433 | snaps.jpg |
0.463835 | eb01657ef4604d2885a682bc084068ac | Iterative three-dimensional FEM results for Pe=40 and U̅=0.1 in a cubic system with -7 < y̅ <0, 0<x̅<7, -7<z̅<7. Top: Contour plots of the concentration c̅(ρ⃗) and the stream lines of the Marangoni flow field v⃗_ M(ρ⃗)/ Pe. Bottom: v⃗_ M as a function of ρ and c̅ as a function of ρ at the interface y̅=0 along with the predictions (<ref>) and (<ref>). | 2012.14433 | vMcPe.jpg |
0.404706 | 00d86ec98d2e4543b7709869e1677307 | The constraints on the present magnetic fields on the eddy turnover scale, Eqs. (<ref>) and (<ref>), are shown as functions of the present helicity fraction. The thick green line is the allowed parameter region for the delta-function model, while the thick yellow line is allowed for the power-law model with α=0, with the evolution parameter β=0. The navy dotted line shows the lower limit of the field strength extrapolated from the blazar observation <cit.>. The green arrows show the edges of the allowed regions for each β. For β≳1, only around B_0=1×10^-17G with ϵ_0=1 is allowed in both models. | 2012.14435 | fig0505.jpg |
0.496151 | 3c4155a6c555434e9666da2cfd92cbea | Comparison of the shapes of 𝒢(R) with the delta-function model and the power-law model with α=0 is shown. 𝒢(R) has terms proportional to 1/ϵ^2 originated from the non-helical part of the spectrum, and terms independent of ϵ originated from the helical part. In the delta-function model, the correlation function is exactly proportional to 1+1/ϵ^2. | 2012.14435 | graph0406.jpg |
0.436856 | fe1fadf5eee24b75a70b0fa59e647c47 | Kurtosis of the distribution of off-diagonal matrix elements computed for 200 eigenstates in the middle of the spectrum for systems with variable ranges of interactions without the middle-site impurity, Ĥ_γ^0 (top row), and with a middle-site impurity, Ĥ_γ^imp (bottom row), plotted for different number of excitations (different columns). In each panel, larger system sizes are represented by darker shades (see legends). | 2012.14436 | kurtosis_long_range.jpg |
0.459666 | 1c6df5bd112f44e283b5e2d89ee69e1f | Distribution of the off-diagonal elements of Ŝ_⌊ L/2⌋^z, computed for 200 eigenstates in the middle of the spectrum, for the integrable model Ĥ_∞^0 (top row) and the single-impurity model Ĥ_∞^imp (bottom row) for different number of excitations (different columns). In each panel, larger system sizes are represented by darker colors (see legends). The histograms are scaled by 𝒟^0.5L. The insets display log-log plots of the scaled variance σ^2/L^2 against the Hilbert space dimension and the black dashed lines correspond to σ^2/L^2∝𝒟^-1. | 2012.14436 | offdiag_dist_short_range.jpg |
0.436361 | ac6d11dd574949069f196300e2c57850 | Ratios Γ_Ŝ_⌊ L/2⌋^z(ω) for the integrable model Ĥ_∞^0 (top row) and the single-impurity model Ĥ_∞^imp (bottom row) for different number of excitations (different columns). In each panel, larger system sizes are represented by darker colors (see legends). The insets zoom in into areas of interest and are on a linear scale. See main text for the explanation of how Γ_Ŝ_⌊ L/2⌋^z(ω) was calculated. | 2012.14436 | offdiag_gamma_short_range.jpg |
0.447009 | 3e31aecb839b4e489b5998b5a79c2745 | The difference (R_max - R_min)/R_0 between the maximum and minimum values of the Ricci scalar on the geometries (<ref>) and (<ref>), normalized by the value on the round sphere R_0 = 2; solid orange, magenta, and blue correspond to the l = 2, 4, and 6 Type 1 geometries, while dashed orange, magenta, and blue correspond to the n = 1, 2, and 3 Type 2 geometries. | 2012.14437 | Combined_Ricci.jpg |
0.499154 | c7547d35719644a199f28398d0b44de3 | The Casimir energy for the conformal scalar, Dirac fermion, and holographic CFT on the type 1 geometries (<ref>); orange, magenta, and blue correspond to l = 2, 4, and 6, respectively. The energies are normalized by the perturbative behavior (<ref>), hence all the curves cross through 1 at x = 0 by design. Note that the data for the holographic CFT and the fermion cannot be distinguished by eye on this plot. Also note that for visual clarity we show only the restricted domain x ∈ (-0.6,0.7) (all other plots in this paper show the full domain x ∈ (-1,1)). | 2012.14437 | Type1_energies_rescaled.jpg |
0.481955 | e25629204d0443f9a7783f4bf156a4f2 | The Casimir energy for the conformal scalar, Dirac fermion, and holographic CFT on the type 2 geometries (<ref>); orange, magenta, and blue correspond to n = 1, 2, and 3, respectively. The energies are normalized by the perturbative behavior (<ref>), hence all the curves cross through 1 at x = 0 by design; note also an additional normalization factor of (1-x)^-1, introduced to clarify the behavior near x = 1. We emphasize that the data for the holographic CFT and the fermion cannot be distinguished by eye on this plot. We note that the scalar theory has less extent in the positive x sense than the other theories as it ceases to exist when the conformal Laplacian -^2 + R/8 fails to be positive. | 2012.14437 | Type2_energies.jpg |
0.516116 | d7ae73558d184624b4dcb5db5b53511f | A four qubit bQCNN containing 111 trainable parameters (see Appendix <ref> for details on gates). Note the dotted box connected to the measurements by classical double-lined rails. This dotted box indicates the presence of multiple (in this case four) distinct convolutional layers, which are chosen based on the mid-circuit measurement outcomes. The corresponding standard QCNN contains 66 parameters, and looks the same, though it lacks any mid-circuit measurements, and has only a single branch after the first pooling layer. | 2012.14439 | 4_qubit_bqcnn_diagram.jpg |
0.416743 | be03b1c0d769403faa183348855e014c | The correctness (one minus the mean absolute error) of the best circuit in a generation vs. training generation for a 4 qubit bQCNN (blue) and standard QCNN (orange) when training on bQCNN-generated data. To generate the data, a bQCNN circuit was created with a set of 111 random parameters. The complete set of 16 binary basis states (|0000⟩, |0001⟩, |0010⟩, etc.) was then passed through the circuit in reverse, and the resulting 16 states were used as training data. The binary label of the zeroth qubit before being passed through the inverted bQCNN, of course, served as the states' corresponding training labels. The above data was averaged over three training sessions on different randomly generated datasets. | 2012.14439 | 4qubit_training-on_bQCNN_data.jpg |
0.521396 | 1eff0400f2af41b89488cb0c4ac70cda | Training a 4 qubit, 111 parameter bQCNN (blue) and 66 parameter QCNN (orange) to classify a set of states generated by running a standard QCNN backwards. Here we see that, while initially slower to train, the bQCNN eventually catches up to the QCNN in performance. | 2012.14439 | 4qubit_training_bQCNN_on_QCNN.jpg |
0.540322 | 76f7b8cfce694356a85f0f559a285a3f | Gate decomposition for the general 2-qubit SU(4) rotation applied to qubits i and j using trapped-ion native gates where s=±1 depending on the laser tuning of the device. | 2012.14439 | MS_decomp.jpg |
0.601017 | 1b48039696154f18ac5659a98b04fcb2 | Gate decomposition for the general 2-qubit SU(4) rotation applied to qubits i and j using cnot gates. | 2012.14439 | SU4_decomp.jpg |
0.500445 | aaecef7e658248c8a067a3f1dcc04309 | An example of an L-qubit bQCNN. The blue boxes denote parameterized SU(4) 2-qubit gates gates (though one could use many-qubit parameterized gates depending on the problem at hand). The green boxes are controlled single qubit SU(2) operations, and, along with the measurements, constitute the pooling layers. At each pooling layer, the controlled rotations are applied to the qubits that will move onto the next layer, a fraction of the qubits are measured and discarded (2/3 of them in this case), And the remaining qubits proceed to the subsequent convolutional layer. Which convolutional layer we proceed to (i.e. which branch) is determined by the measurement outcomes of the previous pooling layer. In this example, since we measure 2L/3 qubits at the first pooling layer, there is a choice of up to 2^2L/3 distinct convolutional layers that can be applied next, each containing its own set of parameterized gates. This pattern of convolution, pooling, branching, convolution is repeated until we are left with a small number of qubits, at which point we execute a small, fully-connected layer of gates, as in <cit.> and measure the designated output qubit(s). | 2012.14439 | bQCNN_diagram1.jpg |
0.46779 | 0eb146f1c9194389a666e236040fee78 | The training correctness (1 minus the mean absolute error cost function) for 500 generations of training of a 4 qubit bQCNN (blue) and 4 qubit QCNN on the diverse dataset. This training dataset features independent X and XXX perturbations in the Hamiltonian <ref> as a means of driving the phase transition. The relatively long-range perturbation introduced by the XXX terms, as well as the diversity of the perturbed states in the training set allow for superior training performance in the bQCNN, which very rapidly approaches its ultimate maximum correctness of 0.743. Meanwhile the QCNN slowly reaches a final correctness value of 0.706, approximately five percent less than the bQCNN correctness, and after many more iterations of the genetic algorithm. | 2012.14439 | diverse_training_temp.jpg |
0.499957 | 7739156341fa43c496ac4edf9fb07cfa | Cross correlation (top) and power spectrum (bottom) of the density field produced by the approximations of the displacement field considered in the text. Surprisingly, T2LPT approach performs as well as ALPT in these statistics, despite its poor modeling of the overdense regions. An attempt to reintroduce the SC information with AT2LPT (the only dashed line), performs almost identically to ALPT and T2LPT. and labels refer to approaches where the displacement field is modelled by <Ref>. In these cases the small scale component has been set to ψ=-3 in correspondence of halo regions detected by these halo finders on the reference simulation (and so they are not predictive). This suggests that an improvement of Lagrangian approximations schemes should come from a better detection of overdense regions. "Perfect ψ" refers to a realization whose density field was generated by the curl-free displacement as measured from the N-body. | 2012.14446 | Fig1.jpg |
0.487683 | 242e41c84aa048d59d581e1f5db07864 | Eulerian positions of particles in a one-particle-thick Lagrangian slice of the simulations. One can notice how T2LPT traces well the cosmic web, but is not able to reproduce collapses or voids. The augmentation of T2LPT with SC (Augmented Truncated LPT) recovers the voids more correctly, but loses the benefit of truncation for filaments and haloes. The approach based on detecting halo particles through is the one that looks more similar to the N-body. | 2012.14446 | Fig2.jpg |
0.427372 | d5f4bef5dd8b4c91a3558c8d23f52881 | 1 -thick Lagrangian slice of the initial conditions. On the left we show halo particles color coded randomly based on their halo-id, identified by . On the right we show halo particles found by . Their agreement is rather high, considering the difference in their baseline process. | 2012.14446 | Fig3.jpg |
0.431089 | cb2860e93574448189e7b940f74a224c | One-particle-thick Lagrangian slice of the displacement field of the various approximation schemes. Particles have been colour-coded according to the value of ψ. ALPT and can trace well the innermost halo voxels when compared to and N-body. Our method of expanding them further according to <Ref> is shown in the bottom left panel. It increases the similarity to the N-body, with the limit that there is no fragmentation between the halo patches. In <Ref> we explain how the fragmentation can be reintroduced with the addition of a halo model, as shown in the bottom-right panel. | 2012.14446 | Fig4.jpg |
0.454454 | b9b13250b0244d70a52b5d1ed6633907 | Difference in the power spectrum of a density field realized through with Gaussian and top-hat filters. The power is directly affected by the number of particles that satisfies the SC criterion, a number that increases with the top-hat filter with respect to the Gaussian filter. When the density field is generated through <Ref> by setting ψ=-3 around collapsing voxels, that is by expanding the proto-halo patches, there is a further increase of power. This offset can be correct by implementing an interpolation in order to match the large-scale power to perturbation theory predictions, as prescribed in <Ref>. | 2012.14446 | Fig5.jpg |
0.441859 | efd3f8705c20401dbc945fba6ce757ce | In the top panel we show the HMFs obtained in various scenarios at z=0. EPS (Gauss) refers to an EPS approach with a Gaussian window function, when no percolation occurs. Analogously for the EPS (top-hat) result. In we implement the scheme discussed in the text based on percolating Gaussian proto-haloes to match a target HMF. The label corresponds to the HMF measured directly from the reference simulation. In the middle panel we show the relative residuals of the lines in the top panel with respect to the fit provided by <cit.>, which we chose as the target HMF and whose implementation is found in colossus <cit.>. In the bottom panel, we show the relative residuals of HMFs at higher redshifts with respect to the corresponding target fits of <cit.>. | 2012.14446 | Fig6b.jpg |
0.421522 | 44f0c7816a25481eaf5d0ed13f9f6a8e | Eulerian positions of the particles in a one-particle-thick Lagrangian slice of the simulations. We plot the displacement without the fragmentation into haloes in the upper right panel, where one can notice the improvement over ALPT in collapsing filaments and haloes, which we attribute to the increased size of halo patches that make the result more similar to what we obtained thorough . Some of the particles, however, are not collapsed at all. This issue can be corrected by incorporating a halo model (bottom right panel), where these particles are redistributed into NFW haloes as found in the initial conditions. | 2012.14446 | Fig7.jpg |
0.43572 | b205fb5340d4412ab5bec53186e3380c | Eulerian PDF of the density field generated through various approximation schemes. The density field has been measured from the 256^3 particles on a mesh grid of 128^3 cells, with a cloud-in-cell mass assignment scheme. The density field has been corrected in Fourier space for the effect of the mass assignment with the related cloud-in-cell kernel <cit.>. | 2012.14446 | Fig8.jpg |
0.406881 | 6e1bfd9aa79d420e8ffb285795c0cc8d | Power spectra and cross-correlations of the density fields generated through ALPT, and 2LPT, compared to the density field of an N-body simulation run with the same initial conditions. The label no frag refers to the realization where particles have not been fragmented into haloes, but only displaced according to <Ref>. | 2012.14446 | Fig9.jpg |
0.539401 | ec8ea21a5ac14ea9bf8e5b52f9b6c8a4 | Plots of the correlation between the displacement field and the linear theory eigenvalues. The value on top is the Spearman's rank correlation coefficient. | 2012.14446 | FigA1.jpg |
0.443852 | b8b99ebf926f495cbea25ec0dcd0df60 | Correlation of the cosmic web invariants with the displacement field ψ. The value on top is the Spearman`s rank correlation coefficient. Apart from I_1=δ_ℓ, only I_5 seems to be correlated with the displacement. Unfortunately the Spearman`s rank correlation between I_5 and δ_ℓ is about 0.95, which shows how it does not provide new information with respect to the linear density. | 2012.14446 | FigA2.jpg |
0.399378 | da53053336124fa29d975f339d9460f1 | Comparison of lidar processing time across the different lidar odometry algorithms. The times are the duration for processing a single scan. Top - Urban Beta dataset, Bottom - Tunnel Safety Research dataset. A processing time of 0.1 s indicates realtime performance (10 Hz scans). | 2012.14447 | LidarProcTime_2.jpg |
0.394899 | 1317425014ed4bae9cb108aa0553347e | Testing of the proposed lidar odometry system, LOCUS, in the DARPA Subterranean Challenge. a) Urban Circuit test environment with a Husky wheeled robot, showing unavoidable rubble. b) Tunnel Circuit test environment with a Husky wheeled robot carrying two lidars, showing self-similar areas with low lidar observability. c) Tunnel dataset ground truth map. d) Example Urban Circuit test with a legged robot carrying one lidar. Both robots run LOCUS. e) Map produced by LOCUS on one Husky robot in the Urban Beta course. f) Urban Alpha ground truth map. | 2012.14447 | OpeningLOCUSImage_5.jpg |
0.520118 | 0b02d68d777c48c894bbec4651272519 | Evolution of the Absolute Position Error (APE) of the proposed method for different processing configurations in the Alpha course of the SubT Challenge. The inset gives more detail on the four best configurations. baseline: all features in Sec. <ref>. imu_int: no WIO integration, only IMU integration, no_int: neither WIO or IMU integration, loam_feat: using LOAM feature extraction instead of filtering, fga_off: no FGA, rdf_off: no random downsample filter, vgf_off: no voxel-grid filter, mdc_off: no MDC. | 2012.14447 | ablation_study.jpg |
0.486194 | 43351972c53c4255afe94b15c95a157e | Architecture of the proposed lidar odometry system. | 2012.14447 | final_architecture.jpg |
0.499786 | ee74c05a47064a429eae7d5725d5a0c3 | Boxplot visualization of the Absolute Position Error (APE) computed for the different methods on the test datasets. For clarity, only the best six algorithms in each dataset are shown. | 2012.14447 | odometry_accuracy_boxplots.jpg |
0.539571 | a1fc308a534447698294d8afc8ac7759 | Absolute Position Error (APE) of the trajectories estimated by the different methods against ground-truth in Beta course, including the performance when running live in the SubT Challenge (LOCUS_REALTIME). | 2012.14447 | real_time.jpg |
0.416062 | 753ea3edf7634ce6992d2189df773be3 | Robustness test in Beta course: a) results on WIO/IMU failure, b) results on WIO failure, c) results on Lidar failure. The failure locations are circled in all cases. | 2012.14447 | robustness.jpg |
0.471134 | 68cd9e2e125046e8bd84e3b31dc81b46 | (a): TL proton generalized FF as a function of p from the data of BaBar, Ref. <cit.> (black circles), BESIII-ISR <cit.> (blue squares) and BESIII-SC <cit.> (green triangles), with the regular background fit with Eq. (<ref>) (black solid line ); (b): data after subtraction, fitted with Eq. (<ref>) (black solid line). For comparison the fit from Ref. <cit.> (red dashed lines) is also shown. | 2012.14656 | Alldif.jpg |
0.424633 | f815ef03565f435eb6c53c7dc0cb9cc9 | Born cross section for e^++e^-→ p + p̅ as a function of the momentum p. The data are from CMD-3 <cit.> (red stars), BaBar <cit.> (black circles) BESIII-ISR <cit.> (blue squares) and BESIII-SC <cit.> (green triangles) are shown together with the six-parameter fit from Eqs. (<ref>,<ref>,<ref>) and Table <ref>(black solid line), compared to the fit from Ref. <cit.> (red dashed line). The blue dash-dotted line corresponds to a constant, fitted in the range 0.1 <√(s)< 0.9 GeV. | 2012.14656 | FigCS.jpg |
0.414647 | 98360a7b41df4635bb4cb713b78b210e | Same as Fig. <ref> but for the TL proton generalized FF. The blue dash-dotted line is the expectation for a constant cross section σ=0.87 nb. | 2012.14656 | Fp.jpg |
0.438618 | 84031d84f2ef4affbe9aa052ae397088 | |G_E| (red circles) and |G_M| (blue squares) from BESIII. The dashed red (dash-dotted blue) line is the calculation of |G_E| and |G_M| from the fits of the effective FF F_p and the ratio R. | 2012.14656 | GEGM.jpg |
0.410456 | 388950ca98974ff68d9b38d53c576988 | (a) |G_E| (red circles) and (b) |G_M| (blue squares) from the BESIII Collaboration. The dashed blue (solid green) curve is the calculation of |G_E| and |G_M| from the QCD extrapolation <cit.> and from the VMD model of Ref. <cit.>. | 2012.14656 | Models.jpg |
0.473704 | 484d168fb6624400a1cf55888c3079f5 | Output of the L^2-Boosting algorithm in the vanishing learning rate asymptotic with input (Y_i,X_i)_1≤ i≤ 100 generated from model (<ref>) at different values of t. The black dots represent the observed data set (Y_i,X_i)_1≤ i≤ 100. | 2012.14657 | Estimates.jpg |
0.464116 | a120661402d64acd8346a76aaef79ea4 | First eigenvectors of the linear base learner interpreted as signal. | 2012.14657 | eigenvectors.jpg |
0.477593 | 59a83e6b6600416891750a4ae893bca3 | Last eigenvectors of the linear base learner interpreted as noise. | 2012.14657 | eigenvectors2.jpg |
0.486856 | ab19d25d561d4a3bbbff55e617e4b443 | Coefficient appearing in Equation (<ref>), the interpretation of the boosting linear operator as a smoothed projection. | 2012.14657 | smoothed-projection.jpg |
0.444447 | 46f20d2524304dd9b53a3509bbe81fcb | Training and test error of the predictors F̂_t(·) and F̂^λ_[t/λ](·) for different values of λ. | 2012.14657 | test-training.jpg |
0.448464 | 5026287416054e3f9c314f45675c9f1a | Decay of the base learner eigenvalues in logarithmic scale - only the 60 largest eigenvalues are plotted due to numerical instability for smaller ones. | 2012.14657 | vp.jpg |
0.493202 | 6b497242f03b4f389a75af1286c320e8 | Correlation of ARP and repetition metrics. | 2012.14660 | arp-correlation.jpg |
0.501243 | 291578c63c264f258de588ac4c31fff3 | Illustration of loops starting and ending with the same word “I”. On average, each word transits to three subsequent words. | 2012.14660 | h.jpg |
0.369691 | 7fcd531461c2498ab29f544a9a532d4e | Influence of high inflow pair. x axis is the high inflow pair count. | 2012.14660 | reason-is-high-inflow.jpg |
0.455843 | a962b2d03c224593b00a5777ed9557eb | Loop and repetition subsequence. | 2012.14660 | task.jpg |
0.468047 | 573b2c8e9f4a4a7799286a8ff4d181be | Evaluation results of Soloist on different levels of language variation corpus. | 2012.14666 | language_variation_noise_env_v2.jpg |
0.451 | c2f7adc309b0474cb6f7efda45461484 | An h-tiling of the annulus. | 2012.14667 | Cylinder.jpg |
0.51035 | c2285758404545f882ca697e89b9db43 | An h-tiling of the triangle. | 2012.14667 | Exotic.jpg |
0.43312 | 6b813145c9884edb9f3de04ef8ec61c6 | A shelled h-tiling. | 2012.14667 | stellar.jpg |
0.424984 | 922dc13940484024b603a36908e7b16e | Building blocks of RL and equivalent control-system elements | 2012.14668 | rlvc_1.jpg |
0.44918 | 770be597d502408ea57f4e6bc3fc4e7f | Basic block components | 2012.14668 | rlvc_13.jpg |
0.443749 | 05796115ab5049038bb290081fe149b4 | Simulink valve model | 2012.14668 | rlvc_18.jpg |
0.4515 | 05a276a0300543cfae99b7d304a761fd | Actor-Critic architecture | 2012.14668 | rlvc_2.jpg |
0.484028 | 6d10de09cbeb43d9b46aa008c34708f8 | FOPTD process model | 2012.14668 | rlvc_20.jpg |
0.44667 | af086565904242cf88bbc6469e8fd179 | PID control setup | 2012.14668 | rlvc_21.jpg |
0.446285 | 99afa17e1ee14048be2f5a88c4db44e5 | RL agent training setup | 2012.14668 | rlvc_22.jpg |
0.436448 | 4dcd3fa2cacc4f4cb2f1423397c01e29 | RL DDPG agent details | 2012.14668 | rlvc_23.jpg |
0.537287 | 187eb2c4df7a420fb9dd74bd181e9716 | Error integral | 2012.14668 | rlvc_24.jpg |
0.464213 | ad40f33bf5884f69bc69e9377bf00261 | RL observations vector | 2012.14668 | rlvc_25.jpg |
0.417719 | 0b48290706d04551bc5cc589670cdf35 | RL rewards computation block | 2012.14668 | rlvc_26.jpg |
0.468931 | e62af0bc1d744f69b045ef775659b837 | Unified setup for a comparative evaluation of RL and PID control strategies | 2012.14668 | rlvc_29.jpg |
0.472299 | 60d950c1137041ab810ddd5a5da02662 | Actual valve movement trajectory <cit.> | 2012.14668 | rlvc_3.jpg |
0.45656 | c4e50d60567a4d4aa6bff0f7e0a062c6 | Block diagram of a single-loop control system | 2012.14668 | rlvc_39.jpg |
0.472403 | 4dbbf22b45be439292b054aea7a86d5c | Nonlinear valve operating characteristics, with stiction <cit.> | 2012.14668 | rlvc_4.jpg |
0.435091 | bde20cbe63234da0913907db742b7b98 | Setup for transfer-function estimation | 2012.14668 | rlvc_40.jpg |
0.47184 | d234db5eb3064f6890caa462451f437f | Stability analysis: Comparing plant responses using the estimated transfer-functions | 2012.14668 | rlvc_42.jpg |
0.524211 | 33b7c53df99845aaa2a017a92b1ab4cb | RL controller: Step response | 2012.14668 | rlvc_43.jpg |
0.438814 | 4476d58282ef47d08180427e48768cde | RL controller: Open-loop Bode loop | 2012.14668 | rlvc_44.jpg |
0.453235 | a62a936007ae4338bc4cdf901179bd57 | Expt.-1: Constant reference signal | 2012.14668 | rlvc_45.jpg |
0.516174 | ba2db209b3bf4fc3a858f824e16ac751 | Expt.-2: Benchmark waveform | 2012.14668 | rlvc_50.jpg |
0.491646 | 940853bb11b24f028db70d081afe97cb | Expt.-3.b: Noise at plant input (μ=0, σ=3.0, 1 Hz) | 2012.14668 | rlvc_57.jpg |
0.428827 | 8f1ec0159f474252a7031634590a8041 | Asymmetric cost-function to avoid overshoots <cit.> | 2012.14668 | rlvc_6.jpg |
0.437002 | 60d616df84d244ec852ffde6bcf07179 | No signs of exploration decay for a decay-rate=3×10^-4 | 2012.14668 | rlvc_75.jpg |
0.451782 | d4e4e96a792443d98e474e394396ad67 | HVAC control <cit.> | 2012.14668 | rlvc_9.jpg |
0.395513 | edfb7045868d44728d8f0bacbf74db31 | Left: Comparison between the values of Y_0.75' measured for 25 clusters that are detected both in the SPT-SZ <cit.> and SPTpol 100d <cit.> surveys. Right: Histograms showing the values of Y_0.75' measured for all clusters in the SPT-SZ (top) and SPTpol 100d (down) catalogs. We also show the Y_0.75' values of the 17 SPTpol clusters considered in this work (red). In all panels, the limit below which we consider that Eddington bias is significant is shown with a vertical dash-dotted line. | 2012.14669 | Eddington_analysis.jpg |
0.384393 | 2277e1d3b18d4dd4b28a044a0410958d | Location of each cluster of the considered sample in the mass-redshift plane. We sub-divided the sample into a high-S/N sample (purple stars) and a low-S/N one (red stars) based on the number of available counts. Diagonal lines show the predicted growth for halos of different final masses <cit.>. The 8 high redshift SPT clusters analyzed by <cit.> are shown with grey triangles for comparison along with the , ACT, and SPT clusters (light grey symbols). We also indicated the 90% confidence level upper limit on the halo mass at each redshift given the considered cosmological model (black line). | 2012.14669 | M_z_plane.jpg |
0.451594 | 4b827841241746fb84caf2bbe5715fd8 | Left: Best-fit surface brightness profile (dark purple line) estimated from the joint X-ray/SZ analysis (see §<ref>) of the data (grey points) measured on SPT-CLJ2343-5024 at z = 0.88. The 1σ and 2σ confidence levels on the best-fit profile are shown with the dark and light magenta regions, respectively. The red and blue regions give the 1σ intervals around the best-fit models of the ICM-induced and background X-ray emissions, respectively. The lower panel shows the difference between the data and the model, weighted by the measurement errors. Right: Posterior distribution of the integrated Compton parameter models of SPT-CLJ2343-5024 (grey) along with the best-fit value (red line). The measured SPT value and its associated uncertainties are shown with the solid and dash-dotted blue lines, respectively. | 2012.14669 | SPTCLJ23295831_xsz.jpg |
0.527193 | 92b028adeead4e759b0588c682b894a0 | X-ray spectrum of SPT-CLJ0304-4921 extracted from a down-sampled event file (see §<ref>). The best-fit model (see §<ref>) is shown in red. It is the combination of a background model (grey) and an ICM emission model (orange). The weighted difference between the data and the model is shown in the lower panel. The significance of these residuals is also lower than 3σ in all energy bins if we consider the bakground model only as a best-fit model. | 2012.14669 | Spectrum_180.jpg |
0.403828 | 492995182bf24b0c94f109991918210c | Density profiles of SPT-CLJ0304-4921 obtained with a standard X-ray analysis based on 2298 counts (left), with the same analysis based on 167 counts (middle), and with the joint X-ray/SZ analysis described in Sect. <ref> based on 167 counts and the SPT integrated Compton parameter (right). In each panel, the solid line shows the best-fit estimate of the ICM density profile and the dark and light colored regions give the 1σ and 2σ confidence levels, respectively. | 2012.14669 | ne_comparison.jpg |
0.401698 | 97902ef597a444be99130eabe2f66114 | Left: Standard deviation of the angular distances measured between the centroid position obtained with all the counts and the ones estimated with down-sampled event files of 1000 counts as a function of the standard deviation obtained by considering the X-ray peak instead of the centroid for the 30 clusters in the high-S/N sub-sample with the highest S/N. The grey shaded region shows the Gaussian kernel-density estimate associated with the 2D distribution of points in the figure. Iso-density contours are shown in black at 10, 30, and 50% of the peak amplitude of the distribution. Right: Four realizations of the image of SPT-CLJ0235-5121 obtained with 1000 counts out of the 4685 available in the 0.7-7 keV band. The images have been adaptively smoothed. The cyan cross shows the X-ray peak position obtained with all the counts while the ones measured in each image is shown with a blue circle. | 2012.14669 | peak_vs_cent.jpg |
0.442189 | ce5ef6bc47514252804e30865b6106fe | Standard deviation of the angular distances measured between the cluster center position obtained with all the counts and the ones estimated with down-sampled event files of N_counts counts as a function of N_counts for three different proxies of the cluster center based on the 30 clusters with the highest S/N in the sample. The results obtained for the X-ray peak estimated by smoothing the map with a 5 arcsec (30 arcsec) FWHM Gaussian kernel are shown in yellow (red). The results obtained for the X-ray centroid are shown in magenta. The computed values are fit with power laws shown with plane lines. | 2012.14669 | sig_center_Ncounts.jpg |
0.456206 | 4e6ca10579564bb7900bb9b21bf96ba3 | For λ∈{0.25, 0.5, 0.75} and μ∈0.01, 0.9, evolution of the quantity ℬ given by <ref> | 2012.14670 | BoundB1.jpg |
0.395874 | c6f42b1ee18741c1bd09a237cf19ed87 | For λ∈{0.25, 0.5, 0.75} and μ∈0.01, 0.9, evolution of the quantity ℬ given by <ref> | 2012.14670 | BoundB2.jpg |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.