nopperl/arxiv-image-text · Datasets at Hugging Face

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
0.523534	f5714515a68a4231ae8dbd02d4bfe906	Results for scenarios 1 to 4 with noise σ = 3	2012.11573	beta-mse-boxplots_sigma3.jpg
0.480863	198f81741a1743df91f8a5f9393c4860	The first row presents 3 time-series of size 50, 300 and 500 that are obtained by the black hat-shaped signal with an additive Gaussian noise with σ = 12. For a range of values for b = β/σ^2log(n) we plot the average log(MSE) curve for algorithms FPOP, RFPOP and slopeOP for 10 simulated time series. The first two algorithms give performances far away from slopeOP. In the third row we show an example of an inferred signal by FPOP with a b value corresponding to the minimum in log(MSE) curves. In last row we plot the segments obtained by the biggest possible b value before having a unique segment with the same data. We observe that we fail to obtain only 2 segments and that we obviously obtain an unrealistic result.	2012.11573	fpop-justification.jpg
0.402854	e60f275eddca4898a95c28a56fecbf04	For a range of penalties from β = 0.5 σ^2 log(n) to β = 2.5 σ^2 log(n), 10 hat-shaped time series have been simulated with n=500 for each penalty value and we ran slopeOP with a smoothing option (minAngle = 130) and without constraint (std). On the left panel A, we compare the MSE between the inferred signal and the true one for the two methods; on the right panel B, we compare the number of inferred segments for the two methods. Results highlight the stability with respect to the penalty of the algorithm with minimal angle option.	2012.11573	minAngle_plot.jpg
0.399898	3e11b175a36245f4b1a66d63a51485ff	Without pruning the dynamic programming algorithm needs to scan m^2n(n-1) elements for finding mimima at each step. We compare the proportion of elements that the algorithm scans for channel-based and inequality-based prunings with a hat-shaped signal with different noise levels. For all noise levels, the channel method is much more efficient.	2012.11573	pruning.jpg
0.481011	dbf095581b184ff4922328fb096fd80e	Examples for the linear and sinusoidal time series with noise σ = 1 (left) and σ = 5 (right).	2012.11573	sigma_estimation-v2.jpg
0.493557	60538f8851fc458b9bf3f7bb712b00dd	The plain black line is the true signal. For each scenario we simulate time series with an additive Gaussian noise of increasing standard deviation (σ = 3, σ = 12, σ = 24). Each plot shows a realisation of this procedure with 500 data points. The number of segments for scenarios 1 to 4 is respectively 2, 7, 6 and 8.	2012.11573	simulated-signals.jpg
0.533147	2f58244079124070aac626218dbde718	Computational time of CPOP versus slopeOP for hat-shaped data with noise level σ =3 and σ =24 in log-log scale. The coefficient q in complexity O(n^q) is equal to 2.68 and 2.97 for CPOP for σ = 3 and σ = 24, respectively. For slopeOP we get 1.93 and 2.01 with the channel pruning option.	2012.11573	timePLOT3.jpg
0.435692	f8a6eaf7e8d240eca6c353bab050c863	The values of F(4,N) for 1≤ N≤ 100.	2012.11574	F_4_N.jpg
0.400357	64fe5210168e4e519a5187e5ff2777a5	Applying the proposed method to 7106 lists of the USHMM data. The model based on applying the Monte Carlo simulation to the German census of 1939 is shown for comparison. Note that the plot axes use the logarithmic scale.	2012.11574	all2_.jpg
0.488279	981de50846ba4d588183c98672337783	The probability density functions of the beta distribution and triangular distribution used in the experiments.	2012.11574	beta_triangular.jpg
0.407786	0a48639e7c1f43569d192280c25b4fbd	Same data for Jasenovac inmates as in Fig. <ref>, but with additional markings for the age heaping <cit.> artifacts.	2012.11574	jms_marked_.jpg
0.519931	be7aa515f9774abcb60af00bcef4e23c	The distribution of values d” calculated by the proposed method for birth years from the USHMM lists.	2012.11574	mc_scores.jpg
0.457904	034cc0e51fce45018891510a6e23a826	The dependence of the performance of the baseline and the proposed method on the random samples' size range when 100 inlier samples are drawn from 01, a single outlier sample is drawn from 00.9^2, the number of bins n is 15, c=5, and the size of the inlier and the outlier samples is randomly chosen to be between L and 10· L where L is the lower size bound that is shown on the x-axis.	2012.11574	sc_100_osc_1_b_5_std_0_9_bc_15_pr_10000.jpg
0.413884	017949aa2c454b829155eb0bdb2cc9ae	Birth year histogram of Jasenovac inmates of Serbian nationality with same markings for age heaping as in Fig. <ref>.	2012.11574	serbs_marked_.jpg
0.652915	a93aa4684a3a4d3393e78876c10578f8	The distribution of the sizes of the majority of the 7106 USHMM lists that are used for the experiments.	2012.11574	sizes.jpg
0.415562	5fa3240d0d034176a6e178018fae74df	The distribution of values d' calculated by the proposed method for birth years from the USHMM lists.	2012.11574	tvor_scores.jpg
0.397017	b8577d1eb1ed4dcabc5a3b520558b895	Data augmentation examples. Strong data augmentation is essential. We perform HSV-color augmentation and random horizontal flipping. For comparison, the top-left image shows an example that is not augmented.	2012.11575	data_augmentation.jpg
0.433863	6cd3fac3c3134c5d90e71b4bf49542ef	Overview of the proposed approach. Given a single RGB image, our model detects object centers as keypoints in a heatmap Y. The network directly predicts shape exemplars 𝐳 and 9-DoF bounding boxes jointly for all objects in the scene. The collision loss ℒ_coll favors non-intersecting reconstructions. Our method predicts lightweight, realistic and physically plausible reconstructions in a single pass.	2012.11575	model.jpg
0.45314	3591a0298c184215b1816d4debe2a0d4	Shape selection. We compare one-hot encoding (hard labels, left) for supervising the shape selection problem with soft labels (right) which allow for multiple shape predictions at the same time and are based on geometric similarity, specifically the Euclidean distance between SDF shape representations.	2012.11575	shape_selection.jpg
0.425089	051796af8613462987a6c70406bc932e	Computed spectral energy distributions of the IC cold pulsar wind emission, at periastron, for 11 different Γ_ p,0 values from Γ_ p,0=10 to 10^5, for cases A (dotted lines), B (dashed lines), and C (solid lines) (cases B' and C' are very similar; see main text). A schematic spectral energy distribution based on the observed X-ray to gamma-ray emission around superior conjunction (adapted from fig. 9 in <cit.>; see references therein), which is close to periastron, is also shown.	2012.11578	cc.jpg
0.458081	fbde123bcfff4834b18c3d50b02ee004	Two-point function. With periodic boundary conditions, (a) 𝒪_2(0,x,t=2^6) versus x and (b) 𝒪_2(0,x=L/2,t) versus t in the MBL (J=0.05) and ergodic (J=0.25) phases. L=8,10,12 as indicated. (c) L-scaling of 𝒪_2(0,L-1,t) with open boundary conditions, with t on the legend. Lines are fits to L ≥ 6. (d) Correlation length ξ(t) for various J, extracted from 𝒪_2(0,L-1,t) as in (c).	2012.11580	20201207_clockplot.jpg
0.452102	be1594b29b4b4c96ba46a7321704aab0	(a)-(c) Absolute values of all nonzero eigenvalues λ of the transfer matrix 𝒮⟨ℛ⟩ at times (a) t = 2 (b) t=3 and (c) t=4, with J=0.025m and m=0,1 … 12. The t leading eigenvalues at each J, some of which are degenerate, are shown black. For these times the largest is unique and in the ν=0 sector. (d) λ(0,t) calculated from the average SFF with open boundary conditions using Eq. (<ref>). (e) Ratio of ⟨ K(t) ⟩ with periodic boundary conditions, and with L=12, to the contribution of only the ν=0 sector. The black dashed line shows t.	2012.11580	20201209_spectralplot.jpg
0.462233	818c2c13305b4a719c1be88aaa325631	Clothed bodies (from ) automatically placed in real and synthetic scenes.	2012.11581	additional_rp_qualitatives_examples.jpg
0.432464	a6b5d73377d3462097b4c9a1fabfaa13	Qualitative examples of meshes automatically placed in real and synthetic scenes. The body shapes and poses were not used in training.	2012.11581	additional_smplx_qualitatives_examples.jpg
0.459887	e494262c5cc6459d9f05588c5b83f745	cVAE architecture. For each vertex on the body mesh, we concatenate the vertex positions x_i, y_i, z_i, the contact label , and the corresponding semantic scene label . The latent vector z is concatenated to the vertex positions, and the result passes to the decoder which reconstructs the input features , .	2012.11581	arch.jpg
0.464907	7b55d951ebd24d53ab76194b6cf22db2	Failure cases.	2012.11581	failure_cases_horizantal.jpg
0.45439	92bedc3c90e94e938f33b2466b0fb51b	Putting realistic people in scenes. Pipeline of affordance detection using meshes with clothing. acts as a proxy for the clothed scan. is used to sample features for this pose. These features are then used with the scene mesh to optimize the placement of the body. After convergence, we simply replace with the clothed scan.	2012.11581	pipeline.jpg
0.444717	6286ba1b748f4048bb6df42765751589	Random samples from our trained cVAE. For each example (image pair) we show from left to right: and . The color code is at the bottom. For , blue means contact, while pink means no contact. For , each scene category has a different color.	2012.11581	random_samples.jpg
0.477806	10c0ab4895bd4e64b62449b037ecc44e	Unmodified clothed bodies (from ) automatically placed in real scenes from the Replica dataset.	2012.11581	replica_rp.jpg
0.457901	cee2d9c72c7644558ea5c70cebff8610	Random samples from our trained cVAE for the same pose. For each example we show from left to right: and . The color code is at the bottom. For , blue means contact, while pink means no contact. For , each scene category has a different color.	2012.11581	supMat_random_samples_big.jpg
0.451179	3afe14d24540473392cad427e24fe455	Method overview. (a) The hypernetwork encoder based on an EfficientNet <cit.> backbone, b, with its final layer replaced by the context head, h. (b) The primary network decoder, d, and layers w_i of the weight mapping network, w, embedded in each meta block. The input to the decoder, d, are the input image and the features, F_i, concatenated with positional embedding. Its weights are determined dynamically for each patch in the image. Gray arrows represent skip connections, ×2 blocks are bilinear upsampling, and the blue '+' signs are concatenations. (c) The context head is designed as a nested U-Net. Please see Sec. <ref> for more details.	2012.11582	architecture_04.jpg
0.466793	4b3ae8f5cc62436eaef2f01e3e00d7ec	Qualitative results on CamVid test set <cit.>. The columns represent: input (left), prediction (center), and ground truth (right). The first four rows provide samples from different scenes, and the last two rows demonstrate failure cases.	2012.11582	camvid_test_qualitative_01.jpg
0.40431	7c57e3a14eaa4baca8c71e44f3c3a121	GFLOPs vs. accuracy trade-off on Cityscapes <cit.>. Our models (in orange) attain a significantly better trade-off than previous methods.	2012.11582	cityscapes_test_gflops_tradeoff_02.jpg
0.44123	cf09b9d2e5b747249ecc10adca3cd144	Run-time / accuracy trade-off comparison on the Cityscapes <cit.> test set. Our models (in orange) achieve the best accuracy and the best run-time vs. accuracy trade-off relative to all previous real-time methods.	2012.11582	cityscapes_test_tradeoff_02.jpg
0.415258	dd4ea0ecf7074743a117b25a66dd9351	Qualitative results on Cityscapes <cit.> validation set images. Left to right: input, our result, and ground truth. The first four rows showcase our model's performance in diverse scenes. The last two rows provide sample failures. Please note that the reflective car hood region is ignored in evaluation.	2012.11582	cityscapes_val_qualitative_02.jpg
0.496888	5f2384912b8845e68990d6befd18250f	(a) The meta block based on the inverted residual block <cit.>. Each purple layer represents a dynamic patch-wise convolution with weights generated by the orange layer, w_i. (b) Visualization of the dynamic patch-wise convolution operation. Each color represents weights corresponding to a specific patch and '*' is the convolution operation. Please see Sec. <ref> for more details.	2012.11582	meta_block_03.jpg
0.449996	24ddc03d98de46b89da08cc8c73e84a1	Qualitative results on PASCAL VOC 2012 validation set <cit.>. First and 4th columns: input image, 2nd and 5th columns: our predictions, and 3rd and final column: ground truth. The first four rows demonstrate how our model performs on different classes. The last two rows present failure cases of our model.	2012.11582	pascal_voc_val_qualitative_01.jpg
0.421041	5f15739163ce45fb80c626b6a268a046	Proposed task: Semantic audio-visual navigation. The agent must navigate to a sounding object in the 3D environment. Since the sound may stop while the agent searches for the object, the agent is incentivized to learn the association between how objects look and sound, and to build contextual models for where different semantic sounds are more likely to occur (e.g., water in the bathroom).	2012.11583	intro_v2.jpg
0.414836	e3a21cbc3d7b4f46bf3d8487f5700b1e	In our model, the agent first encodes input observations and stores their features in memory M. Then it uses the acoustic cues to dynamically infer and update a goal descriptor D_t of the target object, which contains both location L_t and object category C_t information about the goal. By conditioning the agent's scene memory on the goal descriptor, the learned state representation s_t preserves information most relevant to the goal. Our transformer-based policy network attends to the encoded observations in M with self-attention to reason about the 3D environment seen so far, and it attends to M_e with D_t to capture possible associations between the hypothesized goal and the visual and acoustic observations to predict the state s_t. Then, s_t is fed to an actor-critic network, which predicts the next action a_t. The agent receives its reward from the environment based on how close to the goal it moves and whether it succeeds in reaching it.	2012.11583	model_fig_v2.jpg
0.450459	064c4b76a485420eae4d88ba729f23cc	Navigation trajectories for our SAVi model from the test data. In the first episode (top/magenta) the agent hears a water dripping sound and in the second episode (bottom/orange) a sound of opening and closing a door. For each episode, we show three egocentric visual views (right) sampled from the agent's trajectory at the start location .5pt-.9pt1, when the sound stops .5pt-.9pt2, and at the end location .5pt-.9pt3. We see that for the top episode, the acoustic event lasts for two thirds of the trajectory and when the sound stops the agent has an accurate estimate of the object location that helps it find the sounding object (the sink). The second episode (bottom) has a much shorter acoustic event. The agent's estimate of the object location is inaccurate when the sound stops but still helps the agent as a general directional cue. The agent leverages this spatial cue and the semantic cue from its estimate of the object category, a cabinet, to attend to its multimodal memory and find the object in the kitchen and end the episode successfully.	2012.11583	qual_trajs_figv1.jpg
0.442642	3ae6016c02cb4e3a8b45a59ca836ec56	Cumulative success rate vs. silence percentage.	2012.11583	success_vs_silence_cumulative.jpg
0.462024	b3d67ca50aa342d6bce93c69ff2d14f9	Evolution of the comoving axion energy density for θ=1. Around a=a_osc, the axion starts oscillating and scaling like matter ρ_ϕ≈ a^-3m^2f^2. Without particle production, this scaling would persist (blue dot-dashed line) yielding the standard abundance from misalignment. For α=60, the backreaction of dark photon production becomes strong around a/ a_osc∼ 9. The thin gray line shows the result from the linear analysis, while the solid orange line gives the lattice result. The lattice result shows a suppression of the final axion abundance by ≈ 10^-2 compared to the case with no particle production, in stark contrast to the linear analysis which suggests a much stronger suppresion. The dotted lines show possible further suppression in case where the final mass is adiabatically reduced, while the brown dashed line corresponds to a time dependent potential that vanishes around a/a_osc=100 (see Sec. <ref> for details).	2012.11584	Axion_Supression.jpg
0.469317	5877da3c32464e0e930cc291b5573ab7	Suppression of the axion relic abundance for different values of α and fixed θ=1 compared to the standard misalignment case where α=0 and there is no dark photon production. We see that θα≳30 is required for efficient dark photon production. For values of θα≳200, friction from particle production causes the axion to slow-roll and behave as vacuum energy, thus it will quickly come to dominate the energy density of the universe. As we ignore the effect of the axion-dark photon system on the gravitational background, this regime is beyond the scope of our simulation, and we simply sketch the expected sharp loss of suppression in this region with the dashed line.	2012.11584	Axion_Supression_alphaScan.jpg
0.549841	354fa8c6651d4b2c85bfa969a5b09bec	Gravitational wave spectra computed on the lattice for different values of α with θ=1 held fixed. The light dashed lines show the two polarizations (red, blue) when ρ_X=ρ_ϕ/2 (roughly the end of the perturbative regime). The solid lines are the final spectra taken at a/a_osc=40 when the GW spectrum has fully converged. The solid black line gives the sum of the two polarizations in the final spectrum and green crosses mark the NDA scaling relation from Eq. <ref> with c_ eff=1. The source material includes the final spectra in tabulated form.	2012.11584	GWspec_alphaScan.jpg
0.476371	bb47c9342c3e41caa725f00e843cac7d	Same as Fig. <ref> except α=40 is held fixed while θ is varied. In the case of θ=3 we chose a smaller sized box L=π/(3ma_osc) to better resolve the UV part of the spectrum.	2012.11584	GWspec_thetaScan.jpg
0.464597	8cd858b2fa334d37bf31b369f4208678	Evolution of the relative error in the Gauss constraint <ref> for different choices of the number of lattice sites along each direction N and the side length of the simulated box L with fixed α=60, θ=1. The number of iterations used in the implicit scheme is fixed at 2 for the left panel while it is varied in the right panel with N=256, L=π/(2· m) held fixed. In all cases, the error stays close to machine precision ≈10^-16 up to a/a_osc≈8, when the dark photon production backreacts on the axion. Thereafter, the error is minimized for small lattice spacings dx=L/N and a high number of iterations.	2012.11584	GaussConstraint_LatScan.jpg
0.515684	075266895e134c9b8ea55e182ebb404a	Evolution of the spectra in the non-linear regime. The model parameters are α=60 and θ=1. The “+" polarization is the first to experience tachyonic instability. The dark photon and axion panels correspond to those in <ref> but within a much smaller range of energy densities.	2012.11584	NonPertEvolution.jpg
0.444435	0e8f7f6641f24b4a888d69ecfd705dc4	Comoving axion (solid) and dark photon (dotted) number densities for different choices of the lattice parameters with α = 60 and θ =1 held fixed. In the left panel, L and N are varied while the number of iterations in the implicit scheme is held fixed at 2. Similarly, the right panel fixes N=256, L=π/(2m) and varies the number of iterations in the implicit scheme. The different choices agree to within ∼ 10% except in the case of the smallest length L=π/(4m).	2012.11584	NumDensities_LatScan.jpg
0.496165	a1557ff5cd924b2990b6eea7dee34e93	Early evolution of the dark photon (left) and axion (right) spectra. The model parameters are α=60 and θ=1. The “+" polarization is the first to experience tachyonic instability.	2012.11584	PertEvolution.jpg
0.416141	824f6f5aca694563a29458de56684035	ALP parameter space in the mass vs. inverse decay constant plane with α=100 and θ=1 held fixed. The parameter space below the bright colored curves could be probed by future GW experiments, such as pulsar timing arrays (SKA) as well as space- (LISA, DECIGO, BBO, μAres) and Earth-based (ET) interferometers. The filled orange region corresponds to the present limits from Planck+BICEP2+Keck and the dashed line shows the possible improvement by the LiteBird mission. The blue curve is the limit on CMB spectral distortions that could be probed by the Voyage2050 mission. The purple region is where the model could account for the recently reported NANOGrav signal. The gray region is excluded in case of a relativistic dark photon by bounds on N_ eff, while in the green region a massive dark photon can be a viable DM candidate. The solid diagonal lines refer to axion dark matter scenarios in which, from left to right, there is no particle production (standard misalignment), only the suppression from particle production ≈10^-2 (PP only), or further suppression η from model extensions (PP + η). In the blue shaded area, the axion is cool enough to be DM, assuming sufficient suppression of the relic abundance.	2012.11584	reference_plot_Lattice.jpg
0.453793	f466cb079ffc414588745b99d135fd90	Examples with no crosswalks. 1) ground camera, 2) ground LiDAR, 3) predicted inverse distance transform, 4) predicted segmentation, 5) predicted crosswalk polygons, 6) gt polygons.	2012.11585	generalization.jpg
0.468851	4edd5b0d66654c90ba07cab8d328a5fe	Overview of our prediction network. Here we use MaxPool2d(kernel width, kernel height, stride, padding) and Conv2d(kernel width, kernel height, out channels, stride, padding, dilation).	2012.11585	model.jpg
0.422107	6bf9ff2d05d34732abc60017dc30c9ef	Statistics of failure cases for the augmented NS-VQA model on validation set. Left is the distribution on last operator and right is on intermediate operation.	2012.11587	error-distribution-enlarged.jpg
0.400667	28edcfd04e774597a14b56ec6be03a6b	Examples showing step-wise visual reasoning for our graph reasoning machine. The top row shows three samples that our model predicts the correct answers while the bottom row shows three failure cases due to different reasons.	2012.11587	gqa_examples_v1.jpg
0.416572	66529dc260524bfe96e22c1a97fb1952	Our graph reasoning machine consists of three components: a) probabilistic scene graph generation; b) question program generation and c) probabilistic program execution.	2012.11587	graph_reasoning_machine_v2.jpg
0.384567	1098bbdbf26a42579feaa82332bd728a	Interpreting grounded regions in different models. From top to bottom, we show ground-truth target objects, grounding results for different models. The red boxes in images represent grounded objects and the green boxes represent attended objects. We show at most five detected bounding boxes from the images for clarity. Below each image, we show the answer to the specific question for each model, with the corresponding grounding scores. The last row is from the proposed Graph Reasoning Machine, which we will introduce in Sec. <ref>.	2012.11587	groundingness-new.jpg
0.447902	d45a0991749e4893a9018910ecc0c352	Examples showing answers flipped by removing a single irrelevant objects from images. From left to right, the methods are Bottom-Up <cit.>, LCGN <cit.>, MMN <cit.>, and NS-VQA <cit.>. The red boxes show ground-truth grounded objects for the question below, while the green box is the one we remove from the image.	2012.11587	robustness-2.jpg
0.375191	4850f2abf9794de68fa6d3a661494d8b	Two simple probing tests show that current VQA system fails to grounding on the correct objects and robust to perturbations: at top row, model gives the same answer after removing foreground objects “houses”; At bottom row, however, covering a background object inverts the answer.	2012.11587	teaser-figure-pami-v3.jpg
0.486541	eb728d4181114bdeafec0f05461fb9f0	Numerical verification of <ref>. The blue boundary marks the location of the inequality [ M] ≥ 2 √(Det[ M]) for a two dimensional square matrix M. The red diamonds are results from the non-equilibrium KMC simulations with the parameters used in Fig. <ref> and the red line is the theoretical mean field prediction for those same parameters. Gray dots are computed by constructing the matrices δμ, D, and L^-1 using the the master equation results and computing the eigenvalues of (δμ- D- L^-1)/J_ tot using Mathematica <cit.> for randomly selected parameters from L_β, eq = [1,90000], f_ density,β=[1,100] and k_ grow=[0.001,100] nm/s, with all other parameters the same as the red line. The inset shows these two quantities for a wider range, with both axes in logarithmic scale. We do not consider k_ grow < 0.001 nm/s due to limitations of numerical precision. <ref>/<ref> provides strong constraints between the non-equilibrium forcing, morphology, and speed of growth.	2012.11588	eigenvalue-gray.jpg
0.423772	cbedb6b0677742e496d2311f5cca94d7	Schematic of adding one ABP at the tip of a growing actin bundle. Here, α and β represent α-actinin and fascin, which are 35 nm and 8 nm in size, respectively. The energetic cost of bending actin disfavors the binding a fascin after an α-actinin, or vice versa, resulting in domains of consecutive α or β types of ABPs. k^f,1_ij and k^b,1_ij (k^f,2_ij and k^b,2_ij) are the forward and backward rates for the first (second) site of an ABP binding, where i and j are the types of the last two ABPs at the tip; k^f,1_αα =k^f,1_βα because both rates represent the binding of the first site of an α ABP. An analogous schematic can be drawn for the case that the second-to-last ABP is α, and this case introduces four additional pairs of forward and backward rates with corresponding constraints.	2012.11588	fig1.jpg
0.426357	56e028b2151f496c84344b0c60d1e06c	Domain length of α and β ABPs as a function of polymerization rate k_grow. The distance between neighboring ABPs is assumed to be 0.037 μm in computing domain lengths <cit.>. Blue and red triangles are domain lengths L_α and L_β measured from KMC simulations. Up and down triangles represent the domain lengths measured in simulations with initial configurations composed of either all α or all β types of ABPs, respectively. Gray lines (<ref>) are the domain lengths computed by self-consistently solving the master equation (<ref>). The parameters for both KMC simulations and the master equation are k^f,1_αα,eq = 6, k^f,1_ββ,eq = 2, f_ density,α=f_ density,β = 100, f_ molecular,α = 0.4, f_ molecular,β = 1, L_α, eq = 900 (33.3 μm), L_β, eq = 300 (11.1 μm) and τ = 1 s. The inset shows the domain lengths over a wider range of polymerization rates with the same symbols. The plateaus toward the left of the inset represent the domain lengths approaching their equilibrium values.	2012.11588	fig2.jpg
0.45346	94338cbaec46474fa262d154b503bccd	Schematic of the N^th cross-linker binds to the actin pairs. The states between two pink dashed lines describe the N^th ABP binding. Arrows link states between which the transitions are allowed. k^f,m_ij and k^b,m_ij are the forward and backward rates, where i and j are the types of the last two ABPs at the tip and * represents a half bound state. The rates used in the KMC simulations and in the master equation are compared in <ref>.	2012.11588	fig3.jpg
0.394331	888d63ebb7164a17b148d599839d12cf	Validation of the backward rate expressions used in the master equation. Domain lengths ratio (L_α-1)/L_α and (L_β-1/L_β (blue and red circles) are computed from full bundle configurations in KMC simulations. The effective backward rates k^b',2_αα (cyan) and k^b',2_ββ (black) in KMC simulations are computed as k^b',2_αα = N^b,2_αα/N^b,2_αα+N^b,2_βα and k^b',2_ββ = N^b,2_ββ/N^b,2_αβ+N^b,2_ββ. N^b,2_ij is the number of occurrences of unbinding the second head of an ABP, where i and j are the types of the last two ABPs at the tip.	2012.11588	figS2.jpg
0.394829	93eba8e9039e42e697caa7fb0689a93c	Comparison of currents and domain length ratios from KMC simulations at various actin polymerization rates. Both currents and domain lengths are measured from KMC simulations of 10^6 steps. The simulation parameters are the same as those used in generating <ref>.	2012.11588	figS3.jpg
0.449921	a4ed8ecfece349fa97fe1e23b22beb24	Coefficient tanϕ at various actin polymerization rates k_ grow. The parameters are the same as in <ref>.	2012.11588	figS5.jpg
0.481123	84c3a351e362449d8e970a6556c31a2b	The ratio between the two currents J_α/J_β computed from KMC simulations (blue points) and predicted by the master equation (black curve).	2012.11588	figS6.jpg
0.459568	bdbed1516fda40499d86d84133abce5f	Normalized variance and covariance computed from KMC simulations. Blue and red points are the variance normalized by the corresponding current (⟨δJ_β^2⟩/J_β^2 and ⟨δJ_α^2⟩/J_α^2) computed from simulations. Green points are the covariance normalized by the magnitude of the two currents (-⟨δJ_αδJ_β⟩/J_αJ_β).	2012.11588	figS7.jpg
0.443648	30c50efceaea4280855fa4106ce96c0d	Schematic showing all the possible transitions of the actin bundle system. We index the graph edges from 1 to 8.	2012.11588	figS8.jpg
0.421715	79d76e2754ec4e9191ed2d248741398b	Comparison of bounds on the non-equilibrium driving force Δμ. The black line(<ref>)is the actual driving force predicted by the master equation. The blue curve(<ref>) is the driving force required for morphology change. The brown (<ref>) and red (<ref>) lines are the thermodynamic bounds computed from KMC simulations using TUR and MTUR. Each of the brown and red data points is generated with 500 independent KMC simulations, each of 10^7 steps. All parameters of the KMC simulations are the same as in <ref>.	2012.11588	reconstruct-deltamu.jpg
0.410092	94cecae7824f40eb9acf4be46f71be79	Initial departure of the instantaneous polaron energy computed with the variational approach (red dashed line) against the experimental measurements (orange points) <cit.> at unitarity. The blue region is the QMC prediction for the polaron stationary energy <cit.>.	2012.11591	9.jpg
0.425892	7ad5d1afc75c480ca9890d6d72cc4639	The fundamental domain of the spatial torus.	2012.11592	Fundamental_Domain.jpg
0.44175	84689c0b9a1f469e852f329d05faa8a3	The surface integral over the twisted torus in terms of the x and y integrals. The contour of the integral ∮ dy runs along the blue lines, and the ∫ dx integral runs over the red segment of length .	2012.11592	Integration.jpg
0.390219	3780cd25d1534f37a22c00ad03554c96	The X and X̃ cycles.	2012.11592	XXtcycles.jpg
0.52367	a4d4d7231a7348e998578ab248d8101d	Radiative corrections to the three point function.	2012.11600	AppliedKernel.jpg
0.505637	c95a958eff70486db3a661c62cb238ec	Diagrammatic representation of equation (<ref>).	2012.11600	AppliedKernelMomenta.jpg
0.445509	0803ab74beb7458ea3c17905ee765ff7	Diagrammatic representation of the DSE. The line with the full ball represents the two-point function while the line alone represents the propagator.	2012.11600	DSE.jpg
0.58691	1f380409dfc2408b945257c530b39f23	Melonic diagram contributing to the 2-point function in a tensor model theory with three indexes and quartic interaction.	2012.11600	MelonDiagram.jpg
0.500054	aeda3f56e7124cf98bf0f163e032c40d	Radiative contributions to the three point function in the melonic limit.	2012.11600	RadiativeContributions3pt.jpg
0.445518	586817db1d594891b443a02740d213be	Solutions of eigenvalue problem for scalar bilinears in various representations in d=2.8 dimensions. The black dots corresponds the the global symmetry current multiplet. The vertical dashed line corresponds to the unitarity bound.	2012.11600	plotAllIrrepN2.jpg
0.454865	be66bfeeabbd406098f7e840453fb0ba	Solution of eigenvalue problem g_1(Δ_𝒪)=1 in d=2.9 dimensions. The vertical dashed line corresponds to the unitarity bound.	2012.11600	plotSingletN1.jpg
0.505644	bf98de6b30b243d595d4ab212076ef9c	Solutions of eigenvalue problem for scalar bilinears of spin-1 in d=2.8 dimensions. The vertical dashed line corresponds to the unitarity bound. The black point represents the stress tensor multiplet.	2012.11600	plotSpin1N2.jpg
0.428029	4c41b2d8e3c045a98a4b8323adbec496	Schematic diagram of our non-equilibrium setup. The spin chain is connected at both ends to thermal baths at temperatures T_L and T_R. Each bath acts on two boundary spins via Lindblad operators L_k. In NESS, a homogeneous current j flows through the bulk.	2012.11601	Fig1.jpg
0.460394	8b87fdd41e5c4a7b9e8b868b689e9ef7	NESS transport properties of the XZ model at different bath temperatures T_b. (a) Spatial energy profiles for a spin chain of length L=100, showing a constant gradient ∇ E. (b) Scaled energy current j/Δ E as a function of system size. Symbols are numerical values and lines represent fits to the scaling j/Δ E=-D/L^γ. These values of γ indicate diffusion according to Fourier's law. Analogous results for the Ising model in a tilted field can be found in Ref. <cit.>.	2012.11601	Fig2.jpg
0.504593	20330b77ec3c4cfe9a475eac7e8072b3	Comparison between local expectation values in the XZ chain at different bath temperatures T_b. The two-point functions ⟨σ_i^x σ_i+1^x ⟩ measured directly in NESS (circles) match with the ones computed using the reduced density matrix ρ_2 of a thermal state (solid lines). Inset shows the difference in expectation values between our method introduced in sec:thermal and the one described in sec:appendixC. Similar results also hold for the tilted-field Ising model.	2012.11601	Fig3.jpg
0.486313	92be500e49a74d0ba0f0086dee2de4d2	Local temperature in NESS of the XZ model at different bath temperatures T_b. (a) Spatial temperature profiles for a spin chain of length L=100 and the same NESS as in fig:ness(a). Inset shows the difference in the local temperatures extracted via our two methods. (b) Local temperature T in the middle of the chain as a function of bath temperature T_b for the two models. Symbols are numerical values and lines represent linear fits.	2012.11601	Fig4.jpg
0.452062	9a517a015dc34511b28dc9a8a042a555	Temperature dependence of the energy diffusion constants D for (a) tilted-field Ising model and (b) XZ model. Symbols represent numerical values and solid lines are fits to (<ref>) and (<ref>) with k=3 respectively. Insets show linear extrapolations to infinite temperature. All inverse temperatures are scaled by the corresponding energy gap Δ.	2012.11601	Fig5.jpg
0.508189	e4dfc6749f4b49119315bc9ae072d891	Schmidt spectrum for the NESS of a boundary driven XZ chain of size L=100 with a bond dimension χ=200 at different bath temperatures T_b. Dashed lines represent fits to a power law decay λ_k ∼ k^-p. Similar results for the Ising model in a tilted field can be found in Ref. <cit.>.	2012.11601	Fig6.jpg
0.439516	f2b69c059b254d0ebdd69d3cf7bb9f29	Diffusion constant D as a function of average energy density E=⟨ H_2⟩/(L-1) estimated using Lindblad dynamics (dashed line) compared to the results from DMT calculations (dots, Ref. <cit.>). Both results are for an XZ chain of size L=100.	2012.11601	Fig7.jpg
0.407804	4f9d7b385ba34cf786ccff1bc38d0218	Temperature dependence of the heat capacity C_v for the tilted-field Ising model (blue) and the XZ model (red). The inset shows a quadratic fit to the heat capacity C_v∼ T^-2 at high temperatures.	2012.11601	Fig8.jpg
0.455871	525b76f4bd4544e58d6dc8187355cbbd	LO topologies for 0νββ: the thick solid lines denote nucleons and the oriented ones leptons (internal neutrino and external electrons). The squares denote LNV vertices. The diagrams for the electromagnetic current are obtained by replacing the internal neutrinos by photons, omitting the external electrons, and adding an additional topology with the internal neutrinos replaced by pions. In the full theory, the EFT vertices, here denoted by gray circles and diamonds, are supplemented by the appropriate form factors and scattering amplitudes that capture the momentum dependence of the elastic NN intermediate-state contributions. Iterations of the NN strong Yukawa and short-range interactions (diamonds) are not shown as they are irrelevant for the matching analysis.	2012.11602	amplitudes.jpg
0.49422	45ff28680eca43bebe5b2a4afb61de3b	Integrand functions defined in Eq. (<ref>), appearing in the matching relations (<ref>) and (<ref>) for the LNV coupling C̃_1: a_χ (\|\|) (solid red line extending to \|\| = 4 m_π), a_< (\|\|) (solid blue line), and a_> (\|\|) (dark-green thin band obtained by taking the range g̅_1^NN∈ [-10,10]). The dashed red/blue lines and light green band correspond to the integrands a_χ,< ,>(\|\|) entering the matching relation for the CIB coupling C̃_1+ C̃_2. The light green band corresponds to the range g̅_LR^NN∈ [-10,10].	2012.11602	integrand.jpg
0.506772	a76c88e7955b454fa8afaf2dab7e04d4	Forward scattering amplitude (left) and self-energy contraction (right). The solid line refers to the hadronic states (pion, nucleon, two-nucleon), the gray blob to the nonperturbative amplitude, and the wiggly lines to the massless mediator attached to the currents (photon or neutrino).	2012.11602	self_energy.jpg
0.455743	c9abfa5a7f0d45edad08d2def06f20fe	Diagrammatic representations of the actions of the U and η symbols. Anyon lines are black and symmetry defect sheets are orange.	2012.11603	Fig_U_eta.jpg
0.425635	037ecb6274a94102a10599a162343c66	Example illustrating the type of symmetry constraint imposed by the Grabner–Swanson equation (Eq. <ref>) on systems in an external magnetic or electric field. For a square lattice with four-fold (90^∘) rotational symmetry with respect to the z-axis (C_4^z) in the presence of an external magnetic field 𝐇 applied along the x-axis, even though the magnetic field breaks the system's C_4^z symmetry, we can nevertheless use this symmetry to state that for a field-dependent tensor T_ij(𝐇), we must have , where T'_ij(𝐇̃) is the tensor expressed in the transformed coordinates (passive transformation) and 𝐇̃ is the transformed field (active transformation). For the element T_xz(H𝐱̂), for example, this gives .	2012.11604	Grabner-Swanson_symmetry_constraint_example.jpg
0.397516	a2e416ddb7fc4de6a8c398867546a610	Honeycomb lattice for a monoclinic crystal (point groups 2 or 2/m). This lattice lacks three-fold (120^∘) rotational symmetry because z-bonds do not have the same length as x- and y-bonds, and it has two-fold (180^∘) rotational symmetry because x- and y-bonds have the same length. The high-symmetry armchair axis (shown in orange) along z-bonds therefore has two-fold rotational symmetry, whereas the other two armchair axes along x- and y-bonds lack this symmetry.	2012.11604	Kitaev_honeycomb_lattice-squished.jpg
0.560158	355fa4951569431b9af9595d20323b0c	Honeycomb lattice of edge-sharing octahedra and its possible point group generating symmetries. The circles are magnetic metal ions, and the violet and pink circles surrounding them are ligands above and below the honeycomb plane, respectively, that form octahedra. Each metal ion interacts with its three neighboring metal ions through superexchange mediated by their shared non-magnetic ligands. This interaction is generally bond-dependent, so we can label the three different types of bonds as x-, y-, and z-bonds. The three possible generating symmetries of the crystallographic point groups associated with this lattice are shown in orange. C_2^e_2 is the two-fold (180^∘) rotational symmetry with respect to each z-bond axis, C_3^e_3 is the three-fold (120^∘) rotational symmetry with respect to the out-of-plane axis through each site, and ℐ is the inversion symmetry with respect to each bond center.	2012.11604	Kitaev_honeycomb_lattice-with_symmetries.jpg
0.452878	ca7d5ea9bcf544b99bfac6728b72e913	Illustration of the two sets of unusual equalities (Eqs. <ref> and <ref>) for zero-field tensors describing crystals with C_3^e_3 symmetry (i.e., belonging to the trigonal point groups 3, 3̅, 32, or 3̅m). The changes in the thermal conductivity that results from applying a small magnetic field δ𝐇 in scenarios all have the same magnitude, i.e. , and similarly for scenarios . Reversing the direction of any one of these vectors simply reverses the sign of the change in the thermal conductivity. These results hold when the system is either not magnetized, or magnetized along the out-of-plane direction (e_3). Even though these illustrations describe the thermomagnetic susceptibility tensor for concreteness, they more generally apply to any general tensor describing a system in the absence of external magnetic or electric fields by simply replacing the heat current 𝐉_Q, temperature gradient ∇ T, and small magnetic field δ𝐇 with any three vector quantities as long as they are not large external fields (however, small magnetic or electric fields that are used to probe the system, i.e. δ𝐇 and δ𝐄, are allowed).	2012.11604	thermomagnetic_susceptibility-all_predictions.jpg
0.399117	801b8513df0941e7a36323d9d3e597d4	Example illustrating one of the unusual equalities for tensors describing crystals with C_3^e_3 symmetry (i.e., belonging to the trigonal point groups 3, 3̅, 32, or 3̅m). In the absence of external magnetic or electric fields, (a) the change in the thermal Hall conductivity δκ_e_1 e_2 that result from applying a small magnetic field δ𝐇 along the zigzag direction e_1 will be the same as (b) the change in the longitudinal thermal conductivity δκ_e_2 e_2 that results from applying a small magnetic field δ𝐇 along the armchair direction e_2 (Eq. <ref>). 𝐉_Q is the heat current and ∇ T is the temperature gradient. Reversing the direction of any one of the three vectors simply reverses the sign of the change in the thermal conductivity. These results hold when the system is either not magnetized, or magnetized along the out-of-plane direction (e_3). Applying Eqs. <ref> and <ref> using the thermomagnetic susceptibility tensor yields several other similar equalities, illustrated in Fig. <ref>.	2012.11604	thermomagnetic_susceptibility-one_example_of_fully_longitudinal_equals_partly_transverse.jpg
0.503359	44d2f382c2aa414ba56ccc12343e6f39	A function which is smooth and real along the unit circle \|ζ\| = 1 is “inversion symmetric” and analytic within an annulus 𝒜 (the white region) with radii R_out = 1/R_in. These radii are determined by the location of the nearest singularities, pictured here as black lines emanating from the origin and infinity. The function's Fourier/Laurent coefficients can be found by integrating along a contour 𝒞 contained within the annulus. Their asymptotic behavior is thus determined by the size of this annulus.	2012.11605	Figure1.jpg
0.459267	ab494d99da9244efbc3d8b74a3bb5246	Predicted values of the spectral index n_s and the tensor-to-scalar ratio r for the potential (<ref>) with t_* = 0 and various values of α, natural inflation, and m^2 φ^2 at a pivot scale N_* = 60, compared with the Planck 2018 1σ and 2 σ exclusion limits. The predictions change as we vary the decay constant f and we mark each curve at f/M_pl = 5, 6, …, 9, connecting each set with a black line. As f →∞, all predictions approach that of m^2 φ^2 inflation. Similarly, the predictions from (<ref>) replicate those of natural inflation when α≫ 1.	2012.11605	Figure10.jpg

0.523534

f5714515a68a4231ae8dbd02d4bfe906

Results for scenarios 1 to 4 with noise σ = 3

2012.11573

beta-mse-boxplots_sigma3.jpg

0.480863

198f81741a1743df91f8a5f9393c4860

The first row presents 3 time-series of size 50, 300 and 500 that are obtained by the black hat-shaped signal with an additive Gaussian noise with σ = 12. For a range of values for b = β/σ^2log(n) we plot the average log(MSE) curve for algorithms FPOP, RFPOP and slopeOP for 10 simulated time series. The first two algorithms give performances far away from slopeOP. In the third row we show an example of an inferred signal by FPOP with a b value corresponding to the minimum in log(MSE) curves. In last row we plot the segments obtained by the biggest possible b value before having a unique segment with the same data. We observe that we fail to obtain only 2 segments and that we obviously obtain an unrealistic result.

2012.11573

fpop-justification.jpg

0.402854

e60f275eddca4898a95c28a56fecbf04

For a range of penalties from β = 0.5 σ^2 log(n) to β = 2.5 σ^2 log(n), 10 hat-shaped time series have been simulated with n=500 for each penalty value and we ran slopeOP with a smoothing option (minAngle = 130) and without constraint (std). On the left panel A, we compare the MSE between the inferred signal and the true one for the two methods; on the right panel B, we compare the number of inferred segments for the two methods. Results highlight the stability with respect to the penalty of the algorithm with minimal angle option.

2012.11573

minAngle_plot.jpg

0.399898

3e11b175a36245f4b1a66d63a51485ff

Without pruning the dynamic programming algorithm needs to scan m^2n(n-1) elements for finding mimima at each step. We compare the proportion of elements that the algorithm scans for channel-based and inequality-based prunings with a hat-shaped signal with different noise levels. For all noise levels, the channel method is much more efficient.

2012.11573

pruning.jpg

0.481011

dbf095581b184ff4922328fb096fd80e

Examples for the linear and sinusoidal time series with noise σ = 1 (left) and σ = 5 (right).

2012.11573

sigma_estimation-v2.jpg

0.493557

60538f8851fc458b9bf3f7bb712b00dd

The plain black line is the true signal. For each scenario we simulate time series with an additive Gaussian noise of increasing standard deviation (σ = 3, σ = 12, σ = 24). Each plot shows a realisation of this procedure with 500 data points. The number of segments for scenarios 1 to 4 is respectively 2, 7, 6 and 8.

2012.11573

simulated-signals.jpg

0.533147

2f58244079124070aac626218dbde718

Computational time of CPOP versus slopeOP for hat-shaped data with noise level σ =3 and σ =24 in log-log scale. The coefficient q in complexity O(n^q) is equal to 2.68 and 2.97 for CPOP for σ = 3 and σ = 24, respectively. For slopeOP we get 1.93 and 2.01 with the channel pruning option.

2012.11573

timePLOT3.jpg

0.435692

f8a6eaf7e8d240eca6c353bab050c863

The values of F(4,N) for 1≤ N≤ 100.

2012.11574

F_4_N.jpg

0.400357

64fe5210168e4e519a5187e5ff2777a5

Applying the proposed method to 7106 lists of the USHMM data. The model based on applying the Monte Carlo simulation to the German census of 1939 is shown for comparison. Note that the plot axes use the logarithmic scale.

2012.11574

all2_.jpg

0.488279

981de50846ba4d588183c98672337783

The probability density functions of the beta distribution and triangular distribution used in the experiments.

2012.11574

beta_triangular.jpg

0.407786

0a48639e7c1f43569d192280c25b4fbd

Same data for Jasenovac inmates as in Fig. <ref>, but with additional markings for the age heaping <cit.> artifacts.

2012.11574

jms_marked_.jpg

0.519931

be7aa515f9774abcb60af00bcef4e23c

The distribution of values d” calculated by the proposed method for birth years from the USHMM lists.

2012.11574

mc_scores.jpg

0.457904

034cc0e51fce45018891510a6e23a826

The dependence of the performance of the baseline and the proposed method on the random samples' size range when 100 inlier samples are drawn from 01, a single outlier sample is drawn from 00.9^2, the number of bins n is 15, c=5, and the size of the inlier and the outlier samples is randomly chosen to be between L and 10· L where L is the lower size bound that is shown on the x-axis.

2012.11574

sc_100_osc_1_b_5_std_0_9_bc_15_pr_10000.jpg

0.413884

017949aa2c454b829155eb0bdb2cc9ae

Birth year histogram of Jasenovac inmates of Serbian nationality with same markings for age heaping as in Fig. <ref>.

2012.11574

serbs_marked_.jpg

0.652915

a93aa4684a3a4d3393e78876c10578f8

The distribution of the sizes of the majority of the 7106 USHMM lists that are used for the experiments.

2012.11574

sizes.jpg

0.415562

5fa3240d0d034176a6e178018fae74df

The distribution of values d' calculated by the proposed method for birth years from the USHMM lists.

2012.11574

tvor_scores.jpg

0.397017

b8577d1eb1ed4dcabc5a3b520558b895

Data augmentation examples. Strong data augmentation is essential. We perform HSV-color augmentation and random horizontal flipping. For comparison, the top-left image shows an example that is not augmented.

2012.11575

data_augmentation.jpg

0.433863

6cd3fac3c3134c5d90e71b4bf49542ef

Overview of the proposed approach. Given a single RGB image, our model detects object centers as keypoints in a heatmap Y. The network directly predicts shape exemplars 𝐳 and 9-DoF bounding boxes jointly for all objects in the scene. The collision loss ℒ_coll favors non-intersecting reconstructions. Our method predicts lightweight, realistic and physically plausible reconstructions in a single pass.

2012.11575

model.jpg

0.45314

3591a0298c184215b1816d4debe2a0d4

Shape selection. We compare one-hot encoding (hard labels, left) for supervising the shape selection problem with soft labels (right) which allow for multiple shape predictions at the same time and are based on geometric similarity, specifically the Euclidean distance between SDF shape representations.

2012.11575

shape_selection.jpg

0.425089

051796af8613462987a6c70406bc932e

Computed spectral energy distributions of the IC cold pulsar wind emission, at periastron, for 11 different Γ_ p,0 values from Γ_ p,0=10 to 10^5, for cases A (dotted lines), B (dashed lines), and C (solid lines) (cases B' and C' are very similar; see main text). A schematic spectral energy distribution based on the observed X-ray to gamma-ray emission around superior conjunction (adapted from fig. 9 in <cit.>; see references therein), which is close to periastron, is also shown.

2012.11578

cc.jpg

0.458081

fbde123bcfff4834b18c3d50b02ee004

Two-point function. With periodic boundary conditions, (a) 𝒪_2(0,x,t=2^6) versus x and (b) 𝒪_2(0,x=L/2,t) versus t in the MBL (J=0.05) and ergodic (J=0.25) phases. L=8,10,12 as indicated. (c) L-scaling of 𝒪_2(0,L-1,t) with open boundary conditions, with t on the legend. Lines are fits to L ≥ 6. (d) Correlation length ξ(t) for various J, extracted from 𝒪_2(0,L-1,t) as in (c).

2012.11580

20201207_clockplot.jpg

0.452102

be1594b29b4b4c96ba46a7321704aab0

(a)-(c) Absolute values of all nonzero eigenvalues λ of the transfer matrix 𝒮⟨ℛ⟩ at times (a) t = 2 (b) t=3 and (c) t=4, with J=0.025m and m=0,1 … 12. The t leading eigenvalues at each J, some of which are degenerate, are shown black. For these times the largest is unique and in the ν=0 sector. (d) λ(0,t) calculated from the average SFF with open boundary conditions using Eq. (<ref>). (e) Ratio of ⟨ K(t) ⟩ with periodic boundary conditions, and with L=12, to the contribution of only the ν=0 sector. The black dashed line shows t.

2012.11580

20201209_spectralplot.jpg

0.462233

818c2c13305b4a719c1be88aaa325631

Clothed bodies (from ) automatically placed in real and synthetic scenes.

2012.11581

additional_rp_qualitatives_examples.jpg

0.432464

a6b5d73377d3462097b4c9a1fabfaa13

Qualitative examples of meshes automatically placed in real and synthetic scenes. The body shapes and poses were not used in training.

2012.11581

additional_smplx_qualitatives_examples.jpg

0.459887

e494262c5cc6459d9f05588c5b83f745

cVAE architecture. For each vertex on the body mesh, we concatenate the vertex positions x_i, y_i, z_i, the contact label , and the corresponding semantic scene label . The latent vector z is concatenated to the vertex positions, and the result passes to the decoder which reconstructs the input features , .

2012.11581

arch.jpg

0.464907

7b55d951ebd24d53ab76194b6cf22db2

Failure cases.

2012.11581

failure_cases_horizantal.jpg

0.45439

92bedc3c90e94e938f33b2466b0fb51b

Putting realistic people in scenes. Pipeline of affordance detection using meshes with clothing. acts as a proxy for the clothed scan. is used to sample features for this pose. These features are then used with the scene mesh to optimize the placement of the body. After convergence, we simply replace with the clothed scan.

2012.11581

pipeline.jpg

0.444717

6286ba1b748f4048bb6df42765751589

Random samples from our trained cVAE. For each example (image pair) we show from left to right: and . The color code is at the bottom. For , blue means contact, while pink means no contact. For , each scene category has a different color.

2012.11581

random_samples.jpg

0.477806

10c0ab4895bd4e64b62449b037ecc44e

Unmodified clothed bodies (from ) automatically placed in real scenes from the Replica dataset.

2012.11581

replica_rp.jpg

0.457901

cee2d9c72c7644558ea5c70cebff8610

Random samples from our trained cVAE for the same pose. For each example we show from left to right: and . The color code is at the bottom. For , blue means contact, while pink means no contact. For , each scene category has a different color.

2012.11581

supMat_random_samples_big.jpg

0.451179

3afe14d24540473392cad427e24fe455

Method overview. (a) The hypernetwork encoder based on an EfficientNet <cit.> backbone, b, with its final layer replaced by the context head, h. (b) The primary network decoder, d, and layers w_i of the weight mapping network, w, embedded in each meta block. The input to the decoder, d, are the input image and the features, F_i, concatenated with positional embedding. Its weights are determined dynamically for each patch in the image. Gray arrows represent skip connections, ×2 blocks are bilinear upsampling, and the blue '+' signs are concatenations. (c) The context head is designed as a nested U-Net. Please see Sec. <ref> for more details.

2012.11582

architecture_04.jpg

0.466793

4b3ae8f5cc62436eaef2f01e3e00d7ec

Qualitative results on CamVid test set <cit.>. The columns represent: input (left), prediction (center), and ground truth (right). The first four rows provide samples from different scenes, and the last two rows demonstrate failure cases.

2012.11582

camvid_test_qualitative_01.jpg

0.40431

7c57e3a14eaa4baca8c71e44f3c3a121

GFLOPs vs. accuracy trade-off on Cityscapes <cit.>. Our models (in orange) attain a significantly better trade-off than previous methods.

2012.11582

cityscapes_test_gflops_tradeoff_02.jpg

0.44123

cf09b9d2e5b747249ecc10adca3cd144

Run-time / accuracy trade-off comparison on the Cityscapes <cit.> test set. Our models (in orange) achieve the best accuracy and the best run-time vs. accuracy trade-off relative to all previous real-time methods.

2012.11582

cityscapes_test_tradeoff_02.jpg

0.415258

dd4ea0ecf7074743a117b25a66dd9351

Qualitative results on Cityscapes <cit.> validation set images. Left to right: input, our result, and ground truth. The first four rows showcase our model's performance in diverse scenes. The last two rows provide sample failures. Please note that the reflective car hood region is ignored in evaluation.

2012.11582

cityscapes_val_qualitative_02.jpg

0.496888

5f2384912b8845e68990d6befd18250f

(a) The meta block based on the inverted residual block <cit.>. Each purple layer represents a dynamic patch-wise convolution with weights generated by the orange layer, w_i. (b) Visualization of the dynamic patch-wise convolution operation. Each color represents weights corresponding to a specific patch and '*' is the convolution operation. Please see Sec. <ref> for more details.

2012.11582

meta_block_03.jpg

0.449996

24ddc03d98de46b89da08cc8c73e84a1

Qualitative results on PASCAL VOC 2012 validation set <cit.>. First and 4th columns: input image, 2nd and 5th columns: our predictions, and 3rd and final column: ground truth. The first four rows demonstrate how our model performs on different classes. The last two rows present failure cases of our model.

2012.11582

pascal_voc_val_qualitative_01.jpg

0.421041

5f15739163ce45fb80c626b6a268a046

Proposed task: Semantic audio-visual navigation. The agent must navigate to a sounding object in the 3D environment. Since the sound may stop while the agent searches for the object, the agent is incentivized to learn the association between how objects look and sound, and to build contextual models for where different semantic sounds are more likely to occur (e.g., water in the bathroom).

2012.11583

intro_v2.jpg

0.414836

e3a21cbc3d7b4f46bf3d8487f5700b1e

In our model, the agent first encodes input observations and stores their features in memory M. Then it uses the acoustic cues to dynamically infer and update a goal descriptor D_t of the target object, which contains both location L_t and object category C_t information about the goal. By conditioning the agent's scene memory on the goal descriptor, the learned state representation s_t preserves information most relevant to the goal. Our transformer-based policy network attends to the encoded observations in M with self-attention to reason about the 3D environment seen so far, and it attends to M_e with D_t to capture possible associations between the hypothesized goal and the visual and acoustic observations to predict the state s_t. Then, s_t is fed to an actor-critic network, which predicts the next action a_t. The agent receives its reward from the environment based on how close to the goal it moves and whether it succeeds in reaching it.

2012.11583

model_fig_v2.jpg

0.450459

064c4b76a485420eae4d88ba729f23cc

Navigation trajectories for our SAVi model from the test data. In the first episode (top/magenta) the agent hears a water dripping sound and in the second episode (bottom/orange) a sound of opening and closing a door. For each episode, we show three egocentric visual views (right) sampled from the agent's trajectory at the start location .5pt-.9pt1, when the sound stops .5pt-.9pt2, and at the end location .5pt-.9pt3. We see that for the top episode, the acoustic event lasts for two thirds of the trajectory and when the sound stops the agent has an accurate estimate of the object location that helps it find the sounding object (the sink). The second episode (bottom) has a much shorter acoustic event. The agent's estimate of the object location is inaccurate when the sound stops but still helps the agent as a general directional cue. The agent leverages this spatial cue and the semantic cue from its estimate of the object category, a cabinet, to attend to its multimodal memory and find the object in the kitchen and end the episode successfully.

2012.11583

qual_trajs_figv1.jpg

0.442642

3ae6016c02cb4e3a8b45a59ca836ec56

Cumulative success rate vs. silence percentage.

2012.11583

success_vs_silence_cumulative.jpg

0.462024

b3d67ca50aa342d6bce93c69ff2d14f9

Evolution of the comoving axion energy density for θ=1. Around a=a_osc, the axion starts oscillating and scaling like matter ρ_ϕ≈ a^-3m^2f^2. Without particle production, this scaling would persist (blue dot-dashed line) yielding the standard abundance from misalignment. For α=60, the backreaction of dark photon production becomes strong around a/ a_osc∼ 9. The thin gray line shows the result from the linear analysis, while the solid orange line gives the lattice result. The lattice result shows a suppression of the final axion abundance by ≈ 10^-2 compared to the case with no particle production, in stark contrast to the linear analysis which suggests a much stronger suppresion. The dotted lines show possible further suppression in case where the final mass is adiabatically reduced, while the brown dashed line corresponds to a time dependent potential that vanishes around a/a_osc=100 (see Sec. <ref> for details).

2012.11584

Axion_Supression.jpg

0.469317

5877da3c32464e0e930cc291b5573ab7

Suppression of the axion relic abundance for different values of α and fixed θ=1 compared to the standard misalignment case where α=0 and there is no dark photon production. We see that θα≳30 is required for efficient dark photon production. For values of θα≳200, friction from particle production causes the axion to slow-roll and behave as vacuum energy, thus it will quickly come to dominate the energy density of the universe. As we ignore the effect of the axion-dark photon system on the gravitational background, this regime is beyond the scope of our simulation, and we simply sketch the expected sharp loss of suppression in this region with the dashed line.

2012.11584

Axion_Supression_alphaScan.jpg

0.549841

354fa8c6651d4b2c85bfa969a5b09bec

Gravitational wave spectra computed on the lattice for different values of α with θ=1 held fixed. The light dashed lines show the two polarizations (red, blue) when ρ_X=ρ_ϕ/2 (roughly the end of the perturbative regime). The solid lines are the final spectra taken at a/a_osc=40 when the GW spectrum has fully converged. The solid black line gives the sum of the two polarizations in the final spectrum and green crosses mark the NDA scaling relation from Eq. <ref> with c_ eff=1. The source material includes the final spectra in tabulated form.

2012.11584

GWspec_alphaScan.jpg

0.476371

bb47c9342c3e41caa725f00e843cac7d

Same as Fig. <ref> except α=40 is held fixed while θ is varied. In the case of θ=3 we chose a smaller sized box L=π/(3ma_osc) to better resolve the UV part of the spectrum.

2012.11584

GWspec_thetaScan.jpg

0.464597

8cd858b2fa334d37bf31b369f4208678

Evolution of the relative error in the Gauss constraint <ref> for different choices of the number of lattice sites along each direction N and the side length of the simulated box L with fixed α=60, θ=1. The number of iterations used in the implicit scheme is fixed at 2 for the left panel while it is varied in the right panel with N=256, L=π/(2· m) held fixed. In all cases, the error stays close to machine precision ≈10^-16 up to a/a_osc≈8, when the dark photon production backreacts on the axion. Thereafter, the error is minimized for small lattice spacings dx=L/N and a high number of iterations.

2012.11584

GaussConstraint_LatScan.jpg

0.515684

075266895e134c9b8ea55e182ebb404a

Evolution of the spectra in the non-linear regime. The model parameters are α=60 and θ=1. The “+" polarization is the first to experience tachyonic instability. The dark photon and axion panels correspond to those in <ref> but within a much smaller range of energy densities.

2012.11584

NonPertEvolution.jpg

0.444435

0e8f7f6641f24b4a888d69ecfd705dc4

Comoving axion (solid) and dark photon (dotted) number densities for different choices of the lattice parameters with α = 60 and θ =1 held fixed. In the left panel, L and N are varied while the number of iterations in the implicit scheme is held fixed at 2. Similarly, the right panel fixes N=256, L=π/(2m) and varies the number of iterations in the implicit scheme. The different choices agree to within ∼ 10% except in the case of the smallest length L=π/(4m).

2012.11584

NumDensities_LatScan.jpg

0.496165

a1557ff5cd924b2990b6eea7dee34e93

Early evolution of the dark photon (left) and axion (right) spectra. The model parameters are α=60 and θ=1. The “+" polarization is the first to experience tachyonic instability.

2012.11584

PertEvolution.jpg

0.416141

824f6f5aca694563a29458de56684035

ALP parameter space in the mass vs. inverse decay constant plane with α=100 and θ=1 held fixed. The parameter space below the bright colored curves could be probed by future GW experiments, such as pulsar timing arrays (SKA) as well as space- (LISA, DECIGO, BBO, μAres) and Earth-based (ET) interferometers. The filled orange region corresponds to the present limits from Planck+BICEP2+Keck and the dashed line shows the possible improvement by the LiteBird mission. The blue curve is the limit on CMB spectral distortions that could be probed by the Voyage2050 mission. The purple region is where the model could account for the recently reported NANOGrav signal. The gray region is excluded in case of a relativistic dark photon by bounds on N_ eff, while in the green region a massive dark photon can be a viable DM candidate. The solid diagonal lines refer to axion dark matter scenarios in which, from left to right, there is no particle production (standard misalignment), only the suppression from particle production ≈10^-2 (PP only), or further suppression η from model extensions (PP + η). In the blue shaded area, the axion is cool enough to be DM, assuming sufficient suppression of the relic abundance.

2012.11584

reference_plot_Lattice.jpg

0.453793

f466cb079ffc414588745b99d135fd90

Examples with no crosswalks. 1) ground camera, 2) ground LiDAR, 3) predicted inverse distance transform, 4) predicted segmentation, 5) predicted crosswalk polygons, 6) gt polygons.

2012.11585

generalization.jpg

0.468851

4edd5b0d66654c90ba07cab8d328a5fe

Overview of our prediction network. Here we use MaxPool2d(kernel width, kernel height, stride, padding) and Conv2d(kernel width, kernel height, out channels, stride, padding, dilation).

2012.11585

model.jpg

0.422107

6bf9ff2d05d34732abc60017dc30c9ef

Statistics of failure cases for the augmented NS-VQA model on validation set. Left is the distribution on last operator and right is on intermediate operation.

2012.11587

error-distribution-enlarged.jpg

0.400667

28edcfd04e774597a14b56ec6be03a6b

Examples showing step-wise visual reasoning for our graph reasoning machine. The top row shows three samples that our model predicts the correct answers while the bottom row shows three failure cases due to different reasons.

2012.11587

gqa_examples_v1.jpg

0.416572

66529dc260524bfe96e22c1a97fb1952

Our graph reasoning machine consists of three components: a) probabilistic scene graph generation; b) question program generation and c) probabilistic program execution.

2012.11587

graph_reasoning_machine_v2.jpg

0.384567

1098bbdbf26a42579feaa82332bd728a

Interpreting grounded regions in different models. From top to bottom, we show ground-truth target objects, grounding results for different models. The red boxes in images represent grounded objects and the green boxes represent attended objects. We show at most five detected bounding boxes from the images for clarity. Below each image, we show the answer to the specific question for each model, with the corresponding grounding scores. The last row is from the proposed Graph Reasoning Machine, which we will introduce in Sec. <ref>.

2012.11587

groundingness-new.jpg

0.447902

d45a0991749e4893a9018910ecc0c352

Examples showing answers flipped by removing a single irrelevant objects from images. From left to right, the methods are Bottom-Up <cit.>, LCGN <cit.>, MMN <cit.>, and NS-VQA <cit.>. The red boxes show ground-truth grounded objects for the question below, while the green box is the one we remove from the image.

2012.11587

robustness-2.jpg

0.375191

4850f2abf9794de68fa6d3a661494d8b

Two simple probing tests show that current VQA system fails to grounding on the correct objects and robust to perturbations: at top row, model gives the same answer after removing foreground objects “houses”; At bottom row, however, covering a background object inverts the answer.

2012.11587

teaser-figure-pami-v3.jpg

0.486541

eb728d4181114bdeafec0f05461fb9f0

Numerical verification of <ref>. The blue boundary marks the location of the inequality [ M] ≥ 2 √(Det[ M]) for a two dimensional square matrix M. The red diamonds are results from the non-equilibrium KMC simulations with the parameters used in Fig. <ref> and the red line is the theoretical mean field prediction for those same parameters. Gray dots are computed by constructing the matrices δμ, D, and L^-1 using the the master equation results and computing the eigenvalues of (δμ- D- L^-1)/J_ tot using Mathematica <cit.> for randomly selected parameters from L_β, eq = [1,90000], f_ density,β=[1,100] and k_ grow=[0.001,100] nm/s, with all other parameters the same as the red line. The inset shows these two quantities for a wider range, with both axes in logarithmic scale. We do not consider k_ grow < 0.001 nm/s due to limitations of numerical precision. <ref>/<ref> provides strong constraints between the non-equilibrium forcing, morphology, and speed of growth.

2012.11588

eigenvalue-gray.jpg

0.423772

cbedb6b0677742e496d2311f5cca94d7

Schematic of adding one ABP at the tip of a growing actin bundle. Here, α and β represent α-actinin and fascin, which are 35 nm and 8 nm in size, respectively. The energetic cost of bending actin disfavors the binding a fascin after an α-actinin, or vice versa, resulting in domains of consecutive α or β types of ABPs. k^f,1_ij and k^b,1_ij (k^f,2_ij and k^b,2_ij) are the forward and backward rates for the first (second) site of an ABP binding, where i and j are the types of the last two ABPs at the tip; k^f,1_αα =k^f,1_βα because both rates represent the binding of the first site of an α ABP. An analogous schematic can be drawn for the case that the second-to-last ABP is α, and this case introduces four additional pairs of forward and backward rates with corresponding constraints.

2012.11588

fig1.jpg

0.426357

56e028b2151f496c84344b0c60d1e06c

Domain length of α and β ABPs as a function of polymerization rate k_grow. The distance between neighboring ABPs is assumed to be 0.037 μm in computing domain lengths <cit.>. Blue and red triangles are domain lengths L_α and L_β measured from KMC simulations. Up and down triangles represent the domain lengths measured in simulations with initial configurations composed of either all α or all β types of ABPs, respectively. Gray lines (<ref>) are the domain lengths computed by self-consistently solving the master equation (<ref>). The parameters for both KMC simulations and the master equation are k^f,1_αα,eq = 6, k^f,1_ββ,eq = 2, f_ density,α=f_ density,β = 100, f_ molecular,α = 0.4, f_ molecular,β = 1, L_α, eq = 900 (33.3 μm), L_β, eq = 300 (11.1 μm) and τ = 1 s. The inset shows the domain lengths over a wider range of polymerization rates with the same symbols. The plateaus toward the left of the inset represent the domain lengths approaching their equilibrium values.

2012.11588

fig2.jpg

0.45346

94338cbaec46474fa262d154b503bccd

Schematic of the N^th cross-linker binds to the actin pairs. The states between two pink dashed lines describe the N^th ABP binding. Arrows link states between which the transitions are allowed. k^f,m_ij and k^b,m_ij are the forward and backward rates, where i and j are the types of the last two ABPs at the tip and * represents a half bound state. The rates used in the KMC simulations and in the master equation are compared in <ref>.

2012.11588

fig3.jpg

0.394331

888d63ebb7164a17b148d599839d12cf

Validation of the backward rate expressions used in the master equation. Domain lengths ratio (L_α-1)/L_α and (L_β-1/L_β (blue and red circles) are computed from full bundle configurations in KMC simulations. The effective backward rates k^b',2_αα (cyan) and k^b',2_ββ (black) in KMC simulations are computed as k^b',2_αα = N^b,2_αα/N^b,2_αα+N^b,2_βα and k^b',2_ββ = N^b,2_ββ/N^b,2_αβ+N^b,2_ββ. N^b,2_ij is the number of occurrences of unbinding the second head of an ABP, where i and j are the types of the last two ABPs at the tip.

2012.11588

figS2.jpg

0.394829

93eba8e9039e42e697caa7fb0689a93c

Comparison of currents and domain length ratios from KMC simulations at various actin polymerization rates. Both currents and domain lengths are measured from KMC simulations of 10^6 steps. The simulation parameters are the same as those used in generating <ref>.

2012.11588

figS3.jpg

0.449921

a4ed8ecfece349fa97fe1e23b22beb24

Coefficient tanϕ at various actin polymerization rates k_ grow. The parameters are the same as in <ref>.

2012.11588

figS5.jpg

0.481123

84c3a351e362449d8e970a6556c31a2b

The ratio between the two currents J_α/J_β computed from KMC simulations (blue points) and predicted by the master equation (black curve).

2012.11588

figS6.jpg

0.459568

bdbed1516fda40499d86d84133abce5f

Normalized variance and covariance computed from KMC simulations. Blue and red points are the variance normalized by the corresponding current (⟨δJ_β^2⟩/J_β^2 and ⟨δJ_α^2⟩/J_α^2) computed from simulations. Green points are the covariance normalized by the magnitude of the two currents (-⟨δJ_αδJ_β⟩/J_αJ_β).

2012.11588

figS7.jpg

0.443648

30c50efceaea4280855fa4106ce96c0d

Schematic showing all the possible transitions of the actin bundle system. We index the graph edges from 1 to 8.

2012.11588

figS8.jpg

0.421715

79d76e2754ec4e9191ed2d248741398b

Comparison of bounds on the non-equilibrium driving force Δμ. The black line(<ref>)is the actual driving force predicted by the master equation. The blue curve(<ref>) is the driving force required for morphology change. The brown (<ref>) and red (<ref>) lines are the thermodynamic bounds computed from KMC simulations using TUR and MTUR. Each of the brown and red data points is generated with 500 independent KMC simulations, each of 10^7 steps. All parameters of the KMC simulations are the same as in <ref>.

2012.11588

reconstruct-deltamu.jpg

0.410092

94cecae7824f40eb9acf4be46f71be79

Initial departure of the instantaneous polaron energy computed with the variational approach (red dashed line) against the experimental measurements (orange points) <cit.> at unitarity. The blue region is the QMC prediction for the polaron stationary energy <cit.>.

2012.11591

9.jpg

0.425892

7ad5d1afc75c480ca9890d6d72cc4639

The fundamental domain of the spatial torus.

2012.11592

Fundamental_Domain.jpg

0.44175

84689c0b9a1f469e852f329d05faa8a3

The surface integral over the twisted torus in terms of the x and y integrals. The contour of the integral ∮ dy runs along the blue lines, and the ∫ dx integral runs over the red segment of length .

2012.11592

Integration.jpg

0.390219

3780cd25d1534f37a22c00ad03554c96

The X and X̃ cycles.

2012.11592

XXtcycles.jpg

0.52367

a4d4d7231a7348e998578ab248d8101d

Radiative corrections to the three point function.

2012.11600

AppliedKernel.jpg

0.505637

c95a958eff70486db3a661c62cb238ec

Diagrammatic representation of equation (<ref>).

2012.11600

AppliedKernelMomenta.jpg

0.445509

0803ab74beb7458ea3c17905ee765ff7

Diagrammatic representation of the DSE. The line with the full ball represents the two-point function while the line alone represents the propagator.

2012.11600

DSE.jpg

0.58691

1f380409dfc2408b945257c530b39f23

Melonic diagram contributing to the 2-point function in a tensor model theory with three indexes and quartic interaction.

2012.11600

MelonDiagram.jpg

0.500054

aeda3f56e7124cf98bf0f163e032c40d

Radiative contributions to the three point function in the melonic limit.

2012.11600

RadiativeContributions3pt.jpg

0.445518

586817db1d594891b443a02740d213be

Solutions of eigenvalue problem for scalar bilinears in various representations in d=2.8 dimensions. The black dots corresponds the the global symmetry current multiplet. The vertical dashed line corresponds to the unitarity bound.

2012.11600

plotAllIrrepN2.jpg

0.454865

be66bfeeabbd406098f7e840453fb0ba

Solution of eigenvalue problem g_1(Δ_𝒪)=1 in d=2.9 dimensions. The vertical dashed line corresponds to the unitarity bound.

2012.11600

plotSingletN1.jpg

0.505644

bf98de6b30b243d595d4ab212076ef9c

Solutions of eigenvalue problem for scalar bilinears of spin-1 in d=2.8 dimensions. The vertical dashed line corresponds to the unitarity bound. The black point represents the stress tensor multiplet.

2012.11600

plotSpin1N2.jpg

0.428029

4c41b2d8e3c045a98a4b8323adbec496

Schematic diagram of our non-equilibrium setup. The spin chain is connected at both ends to thermal baths at temperatures T_L and T_R. Each bath acts on two boundary spins via Lindblad operators L_k. In NESS, a homogeneous current j flows through the bulk.

2012.11601

Fig1.jpg

0.460394

8b87fdd41e5c4a7b9e8b868b689e9ef7

NESS transport properties of the XZ model at different bath temperatures T_b. (a) Spatial energy profiles for a spin chain of length L=100, showing a constant gradient ∇ E. (b) Scaled energy current j/Δ E as a function of system size. Symbols are numerical values and lines represent fits to the scaling j/Δ E=-D/L^γ. These values of γ indicate diffusion according to Fourier's law. Analogous results for the Ising model in a tilted field can be found in Ref. <cit.>.

2012.11601

Fig2.jpg

0.504593

20330b77ec3c4cfe9a475eac7e8072b3

Comparison between local expectation values in the XZ chain at different bath temperatures T_b. The two-point functions ⟨σ_i^x σ_i+1^x ⟩ measured directly in NESS (circles) match with the ones computed using the reduced density matrix ρ_2 of a thermal state (solid lines). Inset shows the difference in expectation values between our method introduced in sec:thermal and the one described in sec:appendixC. Similar results also hold for the tilted-field Ising model.

2012.11601

Fig3.jpg

0.486313

92be500e49a74d0ba0f0086dee2de4d2

Local temperature in NESS of the XZ model at different bath temperatures T_b. (a) Spatial temperature profiles for a spin chain of length L=100 and the same NESS as in fig:ness(a). Inset shows the difference in the local temperatures extracted via our two methods. (b) Local temperature T in the middle of the chain as a function of bath temperature T_b for the two models. Symbols are numerical values and lines represent linear fits.

2012.11601

Fig4.jpg

0.452062

9a517a015dc34511b28dc9a8a042a555

Temperature dependence of the energy diffusion constants D for (a) tilted-field Ising model and (b) XZ model. Symbols represent numerical values and solid lines are fits to (<ref>) and (<ref>) with k=3 respectively. Insets show linear extrapolations to infinite temperature. All inverse temperatures are scaled by the corresponding energy gap Δ.

2012.11601

Fig5.jpg

0.508189

e4dfc6749f4b49119315bc9ae072d891

Schmidt spectrum for the NESS of a boundary driven XZ chain of size L=100 with a bond dimension χ=200 at different bath temperatures T_b. Dashed lines represent fits to a power law decay λ_k ∼ k^-p. Similar results for the Ising model in a tilted field can be found in Ref. <cit.>.

2012.11601

Fig6.jpg

0.439516

f2b69c059b254d0ebdd69d3cf7bb9f29

Diffusion constant D as a function of average energy density E=⟨ H_2⟩/(L-1) estimated using Lindblad dynamics (dashed line) compared to the results from DMT calculations (dots, Ref. <cit.>). Both results are for an XZ chain of size L=100.

2012.11601

Fig7.jpg

0.407804

4f9d7b385ba34cf786ccff1bc38d0218

Temperature dependence of the heat capacity C_v for the tilted-field Ising model (blue) and the XZ model (red). The inset shows a quadratic fit to the heat capacity C_v∼ T^-2 at high temperatures.

2012.11601

Fig8.jpg

0.455871

525b76f4bd4544e58d6dc8187355cbbd

LO topologies for 0νββ: the thick solid lines denote nucleons and the oriented ones leptons (internal neutrino and external electrons). The squares denote LNV vertices. The diagrams for the electromagnetic current are obtained by replacing the internal neutrinos by photons, omitting the external electrons, and adding an additional topology with the internal neutrinos replaced by pions. In the full theory, the EFT vertices, here denoted by gray circles and diamonds, are supplemented by the appropriate form factors and scattering amplitudes that capture the momentum dependence of the elastic NN intermediate-state contributions. Iterations of the NN strong Yukawa and short-range interactions (diamonds) are not shown as they are irrelevant for the matching analysis.

2012.11602

amplitudes.jpg

0.49422

45ff28680eca43bebe5b2a4afb61de3b

Integrand functions defined in Eq. (<ref>), appearing in the matching relations (<ref>) and (<ref>) for the LNV coupling C̃_1: a_χ (||) (solid red line extending to || = 4 m_π), a_< (||) (solid blue line), and a_> (||) (dark-green thin band obtained by taking the range g̅_1^NN∈ [-10,10]). The dashed red/blue lines and light green band correspond to the integrands a_χ,< ,>(||) entering the matching relation for the CIB coupling C̃_1+ C̃_2. The light green band corresponds to the range g̅_LR^NN∈ [-10,10].

2012.11602

integrand.jpg

0.506772

a76c88e7955b454fa8afaf2dab7e04d4

Forward scattering amplitude (left) and self-energy contraction (right). The solid line refers to the hadronic states (pion, nucleon, two-nucleon), the gray blob to the nonperturbative amplitude, and the wiggly lines to the massless mediator attached to the currents (photon or neutrino).

2012.11602

self_energy.jpg

0.455743

c9abfa5a7f0d45edad08d2def06f20fe

Diagrammatic representations of the actions of the U and η symbols. Anyon lines are black and symmetry defect sheets are orange.

2012.11603

Fig_U_eta.jpg

0.425635

037ecb6274a94102a10599a162343c66

Example illustrating the type of symmetry constraint imposed by the Grabner–Swanson equation (Eq. <ref>) on systems in an external magnetic or electric field. For a square lattice with four-fold (90^∘) rotational symmetry with respect to the z-axis (C_4^z) in the presence of an external magnetic field 𝐇 applied along the x-axis, even though the magnetic field breaks the system's C_4^z symmetry, we can nevertheless use this symmetry to state that for a field-dependent tensor T_ij(𝐇), we must have , where T'_ij(𝐇̃) is the tensor expressed in the transformed coordinates (passive transformation) and 𝐇̃ is the transformed field (active transformation). For the element T_xz(H𝐱̂), for example, this gives .

2012.11604

Grabner-Swanson_symmetry_constraint_example.jpg

0.397516

a2e416ddb7fc4de6a8c398867546a610

Honeycomb lattice for a monoclinic crystal (point groups 2 or 2/m). This lattice lacks three-fold (120^∘) rotational symmetry because z-bonds do not have the same length as x- and y-bonds, and it has two-fold (180^∘) rotational symmetry because x- and y-bonds have the same length. The high-symmetry armchair axis (shown in orange) along z-bonds therefore has two-fold rotational symmetry, whereas the other two armchair axes along x- and y-bonds lack this symmetry.

2012.11604

Kitaev_honeycomb_lattice-squished.jpg

0.560158

355fa4951569431b9af9595d20323b0c

Honeycomb lattice of edge-sharing octahedra and its possible point group generating symmetries. The circles are magnetic metal ions, and the violet and pink circles surrounding them are ligands above and below the honeycomb plane, respectively, that form octahedra. Each metal ion interacts with its three neighboring metal ions through superexchange mediated by their shared non-magnetic ligands. This interaction is generally bond-dependent, so we can label the three different types of bonds as x-, y-, and z-bonds. The three possible generating symmetries of the crystallographic point groups associated with this lattice are shown in orange. C_2^e_2 is the two-fold (180^∘) rotational symmetry with respect to each z-bond axis, C_3^e_3 is the three-fold (120^∘) rotational symmetry with respect to the out-of-plane axis through each site, and ℐ is the inversion symmetry with respect to each bond center.

2012.11604

Kitaev_honeycomb_lattice-with_symmetries.jpg

0.452878

ca7d5ea9bcf544b99bfac6728b72e913

Illustration of the two sets of unusual equalities (Eqs. <ref> and <ref>) for zero-field tensors describing crystals with C_3^e_3 symmetry (i.e., belonging to the trigonal point groups 3, 3̅, 32, or 3̅m). The changes in the thermal conductivity that results from applying a small magnetic field δ𝐇 in scenarios all have the same magnitude, i.e. , and similarly for scenarios . Reversing the direction of any one of these vectors simply reverses the sign of the change in the thermal conductivity. These results hold when the system is either not magnetized, or magnetized along the out-of-plane direction (e_3). Even though these illustrations describe the thermomagnetic susceptibility tensor for concreteness, they more generally apply to any general tensor describing a system in the absence of external magnetic or electric fields by simply replacing the heat current 𝐉_Q, temperature gradient ∇ T, and small magnetic field δ𝐇 with any three vector quantities as long as they are not large external fields (however, small magnetic or electric fields that are used to probe the system, i.e. δ𝐇 and δ𝐄, are allowed).

2012.11604

thermomagnetic_susceptibility-all_predictions.jpg

0.399117

801b8513df0941e7a36323d9d3e597d4

Example illustrating one of the unusual equalities for tensors describing crystals with C_3^e_3 symmetry (i.e., belonging to the trigonal point groups 3, 3̅, 32, or 3̅m). In the absence of external magnetic or electric fields, (a) the change in the thermal Hall conductivity δκ_e_1 e_2 that result from applying a small magnetic field δ𝐇 along the zigzag direction e_1 will be the same as (b) the change in the longitudinal thermal conductivity δκ_e_2 e_2 that results from applying a small magnetic field δ𝐇 along the armchair direction e_2 (Eq. <ref>). 𝐉_Q is the heat current and ∇ T is the temperature gradient. Reversing the direction of any one of the three vectors simply reverses the sign of the change in the thermal conductivity. These results hold when the system is either not magnetized, or magnetized along the out-of-plane direction (e_3). Applying Eqs. <ref> and <ref> using the thermomagnetic susceptibility tensor yields several other similar equalities, illustrated in Fig. <ref>.

2012.11604

thermomagnetic_susceptibility-one_example_of_fully_longitudinal_equals_partly_transverse.jpg

0.503359

44d2f382c2aa414ba56ccc12343e6f39

A function which is smooth and real along the unit circle |ζ| = 1 is “inversion symmetric” and analytic within an annulus 𝒜 (the white region) with radii R_out = 1/R_in. These radii are determined by the location of the nearest singularities, pictured here as black lines emanating from the origin and infinity. The function's Fourier/Laurent coefficients can be found by integrating along a contour 𝒞 contained within the annulus. Their asymptotic behavior is thus determined by the size of this annulus.

2012.11605

Figure1.jpg

0.459267

ab494d99da9244efbc3d8b74a3bb5246

Predicted values of the spectral index n_s and the tensor-to-scalar ratio r for the potential (<ref>) with t_* = 0 and various values of α, natural inflation, and m^2 φ^2 at a pivot scale N_* = 60, compared with the Planck 2018 1σ and 2 σ exclusion limits. The predictions change as we vary the decay constant f and we mark each curve at f/M_pl = 5, 6, …, 9, connecting each set with a black line. As f →∞, all predictions approach that of m^2 φ^2 inflation. Similarly, the predictions from (<ref>) replicate those of natural inflation when α≫ 1.

2012.11605

Figure10.jpg