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
0.514045	e7175813db3c4e50a49a277fdac746dc	The overall pipeline of our proposed approach. Note that the dashed arrows only active in RGC^†, as shown in the top-right corner.	2012.11294	pipeline.jpg
0.560015	6393c7afd47b42f2881ff704b1f1b5a6	Conceptual diagram of our proposed centralized information interaction strategy. Left: Typical U-shape structure which connects the corresponding stages in the bottom-up and top-down pathways directly; Right: Our proposed centralized strategy. The rounded gray rectangle refers to a shared module that can parallelly process all stages (scales) of features in a stage-wise manner.	2012.11294	teaser.jpg
0.429222	2806115adb504c56a6e9f5efdd609f97	Visualizing feature maps w/o and w/ CII. Every translucent gray rectangle refers to a sequence of two 3 × 3 convolution layers. As can be seen, CII can interact the complementary information encoded in feature maps of different scales (c,d). For comparison, w/o CII, the feature maps are roughly unchanged (a,b).	2012.11294	visual.jpg
0.452479	43a395dc9d1b4917b841e3063c84cdce	KPI distribution datset 1	2012.11299	KPI_distribution_DS1.jpg
0.447969	450232e1a0ae4d649966c86ca96091f9	KPI distribution dataset 2	2012.11299	KPI_distribution_DS2.jpg
0.489018	c53252788be844ab91df84b1ef4b473d	KPI prediction dataset 1 with scalar based meta-model	2012.11299	KPI_prediction_data_1.jpg
0.494671	01f8c4cce95e4a75b4815589526e4da6	KPI prediction dataset 2 with scalar based meta-model	2012.11299	KPI_prediction_data_2.jpg
0.477795	8a158de563884f21b35c9ae8eb5adb84	Parameter distribution dataset 1	2012.11299	Param_distribution_DS1.jpg
0.389469	6e390a80a57a4a88ba0577ba75d80ed9	Parameter distribution dataset 2	2012.11299	Param_distribution_DS2.jpg
0.462003	1cad76bf25be421595132fefb7f35e29	Cumulative accuracy plots of the KPI prediction with relative error ε_mre<5% dataset 1	2012.11299	Relative_error_distr_data_1.jpg
0.433644	9a0c35c9278e453e913368d92d1001ed	Cumulative accuracy plots of the KPI prediction with relative error ε_mre<5% dataset 2	2012.11299	Relative_error_distr_data_2.jpg
0.424201	180d0f73c48e4c02bfabe2406b5282da	Comparison between the 21st-order approximant of the ADM (dotted line) and the 4th iteration of the BLUES function method (dashed line) for α =1/2 and n=4. The numerically exact solution (red, full line) is also shown.	2012.11303	ADM_vs_BLUES_half.jpg
0.433833	8dc7e53470c14a369cd3ed844653ffc5		2012.11303	Fisher.jpg
0.482604	88072cf8303f43819218eb9eda6337f1	Wavefront values U_ψ^(n)(-1) at z = -1 versus order n for the Fisher equation (<ref>). The numerically exact value (red, dashed line) is also shown. Inset: A semi-log plot of the increments \|Δ U_ψ^(n,n-1)(-1)\| of the approximants versus n, and a linear fit. These increments are defined as the difference of the n-th and (n-1)-st approximant. Parameter values are k=1/3 and K/k = 1/2.	2012.11303	Fisher_maximum.jpg
0.533174	d7254928ddf6481f8f302b3f5a45cbfa	Residual function U_ψ^(n) of the approximants of order n = 0, 1, 2, 3 for the nonlinear Fisher DE with exponential corner source (<ref>). The residual operator applied to the numerical solution (red, full line) is also shown. Parameter values are k = 1/3 and K/k = 1/2.	2012.11303	Fisher_residuals.jpg
0.420582	73ef8a955ce441e28814a90577b6ca68		2012.11303	Fisher_zoom.jpg
0.469078	7e948b9c0c3049ab89cbcaed093edc57	Residual ℛ_t U(t) = -U^4(t) for α = 1/2 for different orders of iteration. The numerically exact residual ℛ_t U_num is also shown (red, full line).	2012.11303	fractional_alpha_0p50_residual.jpg
0.412458	f1883258e98e446dbdd85617c8e7c505	(a) The mechanical oscillator was driven by optical interface forces 𝐅_interface and tension 𝐅_tension of the four fibers, where the laser was propagating. These forces, illustrated for one of the fibers, were modulated by varying the laser intensity. The laser generated by a multimode laser diode was split into the four fibers by a 1x4 splitter and finally the laser exited from the ends of the four fibers, which were all bonded to the oscillator. The wavelength was either at 808 nm or 915 nm. (b) The cross-section of the 0.22 NA multimode fiber (Thorlabs, FG105LCA). (c) The nanoscale oscillation was detected by the Michelson interferometer utilizing a separate He-Ne laser. See the appendix for a more detailed description of the experimental setup.	2012.11308	fig1.jpg
0.501212	164b63f276354780971bbe8b87712834	The displacement amplitude of the mechanical oscillator is plotted as a function of the laser modulation frequency (a) for the first oscillator at the wavelength of 808 nm and (b) for the second oscillator at the wavelength of 915 nm. The net peak-to-peak power amplitude of the driving field in the four fibers together is P_0=5.125 W in both cases. The solid line represents the averaged frequency spectrum measured with a single modulation frequency. The peak at each modulation frequency is marked with a red dot. The oscillator response function is fitted and shown by the dashed line. The corresponding peak-to-peak force amplitudes of the two oscillators are plotted in (c) and (d) as a function of the peak-to-peak laser power amplitude of the fibers. The solid lines represent the regression lines and the dashed lines show the net theoretical force in Eq. (<ref>). The dash-dotted and dotted lines are the results of the Minkowski (F_M,0) and Abraham (F_A,0) momentum models, respectively, using Eq. (<ref>) with the corresponding momentum of light and excluding the tension in Eq. (<ref>).	2012.11308	fig2.jpg
0.422858	99aa22fa69f740f18eea3b4f52219039	Illustration of the fiber path and the related horizontal and vertical distance parameters a and h.	2012.11308	fig3.jpg
0.444229	c9e922aaadcb4a3a99ba0138d59816ba	Elasticity theory simulation of the total tension force of four fibers on the oscillator as a function of the total incident optical power propagating through the fibers. The dots represent the simulation results obtained by using the expression on the left hand side of Eq. (<ref>) and the dashed line is the theoretical line, given on the right hand side of Eq. (<ref>).	2012.11308	fig4.jpg
0.497802	be73ec07307b47dcb8f658100378ea5b	Qualitative results. Leftmost, we depict the template outfit of each row. Note how templates are smoothed as much as possible. Then, for each row we visualize the output for two unseen test poses. For each sample, we show the output dressed human (left) along with its corresponding deformed outfit in rest pose (right). As it can be seen, the learnt Pose Space Deformations generate folds and wrinkles to fulfill the energy balance requirement of the physical system. The last row depicts results of simulation with a different body shape (and its appropriately aligned template outfit) to show generalization to different bodies. All of the shown samples were obtained after just a few minutes of training.	2012.11310	PBNS_PSD.jpg
0.427121	8f8ae3e860a5494f9500b6735ece0f97	Methodology overview. Pose parameters are fed to a 4-layer Multi-Layer Perceptron with 32 dimensions in each layer. The output high-level pose embedding is multiplied with a PSD matrix 𝐃 to obtain deformations for the garment/outfit. As standard, deformations are applied in rest pose garments and skinned along with the body (note that garment blend weights are assigned by proximity w.r.t. the body in rest pose). Finally, we train by optimizing the potential energy of the output physical system represented by the body and the cloth. The parts of the model in red correspond to the trainable parameters.	2012.11310	PBNS_Pipeline.jpg
0.442317	a0beb1871a6a49fa8a2ec6b4f302a98d	Architecture ablation study. We perform three experiments to assess the effect of the proposed MLP: 1) PBNS as described, 2) θ as input with no MLP and 3) rotation matrices 𝐑_θ as input with no MLP. For experiment 2 we observe unrealistic V-shaped wrinkles around the hip. Also, for both models with no MLP, we observe artifacts where both legs merge. This effect is more evident on skirts. The merging point of left and right side of the skirt presents an important artifact for experiments with no MLP, despise being trained during hundreds of epochs (as opposed to 15 epochs for PBNS with MLP).	2012.11310	ablation.jpg
0.460932	aad780fe01934ba985171d56a8b9162a	PBNS is not limited to work with SMPL. Since it needs no PBS data, the engineering effort for the adaptation to custom avatars is minimum. Here we show different garments and outfits neurally simulated on top of custom avatars. For each, we show the avatar body in rest pose and two different viewpoints of a posed sample. PBNS can be used to enhance rigged 3D characters by animating their clothes.	2012.11310	avatar.jpg
0.435441	918e55582f424846821ea1658c080d6b	PBNS results on outfits with overlapping layers of cloth. Each row shows a sample. From left to right: template outfit, posed sample without outermost layer, posed sample with whole outfit and cross section. The proposed collision loss can handle multiple layers of cloth, even when the template outfit presents inter-penetrations. As we can observe on the cross section, PBNS is able to generate consistent predictions with minimal layer spacing. Also note how it is possible to neurally simulate different fabrics –different wrinkle count and size– within the same outfit by assigning per-vertex weights to cloth related losses.	2012.11310	layers.jpg
0.448594	cab79dc8f34b4eabaf7f9c284c2914a6	Textured qualitative samples. This rendering gives an approximate idea to the final application-level looks that our methodology yields. These results correspond to neural simulations of 3 additional outfits on top of 3 different bodies, further showing the generalization capabilities of our methodology.	2012.11310	qualitative_textured.jpg
0.476968	3237791ad3d24a939e553d2de3f6827b	Results obtained using PBNS for outfit resizing. Body shape changes from left to right. Outfit tightness changes from upper to lower rows. All the previous properties discussed for PBNS are also present when it is used for outfit resizing.	2012.11310	resize.jpg
0.480326	c5d1467dbcf048b993dc08399909e4d1	Supervision vs PBNS. We compare results obtained with PBNS against a purely supervised approach and a hybrid approach (L2 plus Eq. <ref>). The same architecture and the same N = 3000 poses (2550 for training and 450 for validation) are used in these experiments. We additionally compare against PBS data used for supervised training. We show: a) L2 only, b) hybrid, c) PBNS and d) PBS data. The supervised approach has the lowest Euclidean error w.r.t. PBS data, but also lowest number of wrinkles and highest number of collisions. The hybrid approach has more visible wrinkles and fewer collisions, but generalizes poorly to extreme poses. PBNS has the highest Euclidean error, but realistic pose-dependant wrinkles and shows no collisions, even under extreme poses. We also observe how PBS is prone to failures when body presents self-collisions (upper-left sample right elbow and lower-right sample legs). These failures might be transferred to predictions if trained supervisedly. PBNS is more robust to collisions than PBS.	2012.11310	vs_supervised.jpg
0.481153	184ebbf41b0149ab89874baf8c440b08	PBNS vs. TailorNet. We show two different bodies and outfits in three different poses each. For each pose, from left to right: TailorNet raw predictions, TailorNet after post-processing and PBNS. As observed, TailorNet heavily relies on post-processing, hurting performance and removing model differentiability. TailorNet predictions are noisy, show a bias towards body geometry and is unable to model cloth-to-cloth interactions. On the other hand, PBNS predictions show physically cloth-consistent predictions without body or cloth-to-cloth interpenetrations (raw predictions).	2012.11310	vs_tailornet.jpg
0.436244	6d39b5eadcd64817a8d07b6071131571	First-principles spin-phonon relaxation times (τ_s) in graphene, graphene with 0.4 eV/nm electric field and graphene on hBN, as a function of carrier density at room temperature. (a) The out-of-plane to in-plane ratio deviates dramatically from 1/2 for the hBN substrate, partially captured by the DP formula due to the out-of-plane SO field (FIG. <ref>(b)). (b) Spin lifetime is appreciably inversely correlated with momentum lifetime (τ_p), as expected in the DP mechanism, only for graphene on hBN. Most importantly, both the (c) in-plane and (d) out-of-plane spin lifetime dramatically reduce due to inversion symmetry breaking by either electric field or hBN substrate, with the dependence on carrier density flipped relative to the free-standing case.	2012.11550	nCarrTauGrid.jpg
0.438827	e695bd48fef14dcc8c20f8bcdb0fbc3a	(a) Out-of-plane and (b) in-plane spin-phonon relaxation times in graphene on hBN can be further tuned by an applied electric field. (Experimental data from Ref. .) With a single gate, the electric field is dependent on the carrier density and cannot be varied independently. Variation of τ_s⊥/τ_s∥ ratio with (c) carrier density at specified electric fields, and with (d) electric field at a carrier density of 3.7× 10^12 cm-2, showing strong deviations from 1/2 in agreement with experiment.<cit.>	2012.11550	nCarrTauGridEfGrhBN.jpg
0.494858	fb96d351ca1d4618839bc54c58ea95e0	Band structure in the vicinity of the K point (left panels) and corresponding k-dependent effective SO magnetic field B⃗_k⃗ at a Fermi circle 0.1 eV above the conduction band edge (right panels) for (a) free-standing graphene, (b) graphene with electric field, E_z=0.4 eV/nm, and (c) graphene on hBN. The effect of hBN differs qualitatively from that of an electric field, splitting the valence band stronger than the conduction band and introducing an out-of-plane component to B⃗_k⃗.	2012.11550	schematic.jpg
0.561134	625c9380a21b4f109dc160b03a87833c	Illustration on why VAEs produce blurry reconstructions. Consider the example of a binary frontier in an image i and a latent code z which corresponds to the position of the frontier x_front. If q_θ_e(z\|i) = 𝒩(x_front, σ) then p_θ_e(x_front\| z) = 𝒩(z, σ) and the optimal reconstruction of the pixel at position x is 𝔼[pixel(x) \| z ] = 1 ×ℙ_θ_e(x>z) + 0 ×ℙ_θ_e(x<z) = 1/2(1+erf(x-z/√(2)σ)) which is a smooth transition between black and white instead of a sharp transition in the original binary image.	2012.11551	binary_front.jpg
0.54737	d5bb7fcf9d104b0c90248449495daece	Illustration of a toy example with two-dimensional data and a one-dimensional latent space. Points: data, dotted line: manifold of reconstructions from VAE, dashed line/density: manifold of reconstruction with our model. Color encodes the position in the one-dimensional latent space. Top: with a deterministic generator of the form G_θ_g(z). Bottom: with a probabilistic generator of the form G_θ_g(z, ξ). (best seen with zoom and color)	2012.11551	latent.jpg
0.440496	fcdf494bed4341b7bdaaeb3763d5e893	Dynamic BoW-prediction head. G(·) learns to quickly adapt to the visual words in the continuously refreshed vocabulary V. The outputs G(V) are in fact weights that are used for mapping the features () to the corresponding BoW vector y_().	2012.11552	bow_prediction_head_landscape.jpg
0.494136	70b0e118e5364087891c385741bf5ad9	Unsupervised learning with Bag-of-Words guidance. Two encoders and learn at different tempos by interacting and learning from each other. An image is passed through the encoder and its output feature maps ^l() are embedded into a BoW representation y_() over a vocabulary V of features from . The vocabulary V is updated at each step. The encoder aims to reconstruct y_() from data-augmented instances . A dynamic BoW-prediction head learns to leverage the continuously updated vocabulary V to compute the BoW representation from the features (). follows slowly the learning trajectory of via momentum updates.	2012.11552	overview_v2.jpg
0.41981	e89b5872a24d4bcea97f9b700c06168b	Vocabulary queue from randomly sampled local features. For each input image to , “local” features are pooled from ^l() by averaging over 3 × 3 sliding windows. One of the resulting vectors is selected randomly and added as visual word to the vocabulary queue, replacing the oldest word in the vocabulary.	2012.11552	vocabulary_queue.jpg
0.380766	79f2f2ae2c064401b32fbc37a1870666	Iso-contours of the 1σ sensitivity to C_W, in terms of the S parameter in eq. (<ref>), in the luminosity-energy plane. The FCC reach is also indicated, as well as the iso-contours of the relative statistical uncertainty of the cross-section measurement.	2012.11555	cW_S.jpg
0.440772	eba8bafcccbf459caf1d6f4db7221ac0	Transmission line model.	2012.11556	Line_parameters.jpg
0.401196	aaf01849a15e406c9c97c108f440d25e	Inverter filter circuit diagram, showing the notation and current direction convention used in this paper. The inverter itself is connected on the left, with v_iDQ representing the inverter voltage behind the filter.	2012.11556	filtertikzL.jpg
0.454655	aab50c4c8875468586292a69ac2e43aa	System response to load change: frequency.	2012.11556	freq5.jpg
0.387413	5894b10eccbf4f3db0f72279b13e4e90	The power system as an interconnection of bus and line dynamics.	2012.11556	ioDQ.jpg
0.477352	38c0e9fee47f4ba0868448e495bb3e96	Diagram of microgrid used for simulation study.	2012.11556	mgfig.jpg
0.419492	6477cc6e166a47568436941fdd98a81b	Plug-and-play response.	2012.11556	pp_pq.jpg
0.417944	125732c62872486c97e519545c32fce9	System response to load change: power.	2012.11556	pq5.jpg
0.433893	e87ba6ec90ce451699607cd8ca1c514e	System response to load change: voltage.	2012.11556	volt5.jpg
0.414075	fd1b33f35a0c48cea004ca8c851acecc	An example that shows up the intuition behinds the MIP-DoM approach. Consider two sets, P = {(2.0, 2.5), (3.0, 1.9)} and Q = {(2.2, 2.0), (3.0, 1.5)}. The DoM value indicates the total movement for P to weakly dominate Q. A typical vertical movement, zp_(1,2)=0.05 represents an improvement in f_2 objective generating p̂_1. However, p̂_1 does not dominate q_1. There is still a movement to be done in f_2 objective, represented by zpq_(1,1,2)=0.45, so that p'_1 will weakly dominate q_1. Observe that point p_2 does not weakly dominate q_2, so it is vertically moved obtaining p̂_2, in an attempt to weakly dominate q_2. With this vertical movement, indicated by zp_(2,2) = 0.40, p̂_2 dominates q_2, becoming p'_2. The DoM value is the total distance using the terms zp_(1,2), zp_(2,2), and zpq_(1,1,2). Limits on p' movements (lower (lp_(i,m)) and upper bounds (up_(i,m))) are also indicated.	2012.11557	model_sample.jpg
0.488515	7c4b83a755fc4ad3a1a9ddb1f2d013ef	Corner plot resulting from our MCMC analysis of the TESS photometry of 14 Peg. Red markers indicate the best-fit solution. Three vertical dashed lines superimposed on each marginalised 1D posterior distribution denote 16%, 50%, and 84% quantiles.	2012.11559	1334-pro_cornerplot.jpg
0.395083	419fb9fcd5e3404298894c95f900bba4	TESS light curves and frequency spectra of 14 Peg. (a) TESS light curve phased with the orbital period. Phase 0.5 corresponds to the epoch of periastron passage. (b) Residuals from the fit of Kumar's model. Red dots are median values in 0.01 phase bins. (c) Frequency spectrum of the light curve shown in panel b. In panels c – d orbital harmonics are marked with red vertical dashed lines (except for TEOs, which are marked with blue) and labeled with n. (d) The same as in panel c but after removing instrumental trend.	2012.11559	1334-teo.jpg
0.535908	c9adf52084fc4e39976144c380e57798	TESS light curves and frequency spectra of QX Car. (a) TESS light curve phased with the orbital period. Phase 0.5 corresponds to the epoch of periastron passage. (b) Zoom of the light curve after cutting out the eclipses and removing intrinsic variability (see text). Continuous dark blue line is the fitted Kumar's model. For comparison, the dashed line shows the fit of Kumarâs model with inclination fixed at the value derived by <cit.>. (c) Residuals from the fit of Kumar's model. The red dots are median values in 0.01 phase bins. Grey stripe marks phases in the vicinity of the heartbeat. (d) Frequency spectrum of the light curve shown in panel b. In panels d – f orbital harmonics are marked with red vertical dashed lines (except for TEOs, which are marked with blue) and labeled with n. (e) Frequency spectrum of the light curve shown in panel c. (f) Frequency spectrum of the light curve shown in panel c with data at phases marked by the grey stripe removed.	2012.11559	588-teo.jpg
0.439623	ad1df3a9042342cf98c9de2285577154	TESS light curves and frequency spectra of p Vel A. (a) TESS light curve phased with the orbital period. Phase 0.5 corresponds to the epoch of periastron passage. (b) Residuals from the fit of Kumar's model. Red dots are median values in 0.01 phase bins. Grey stripe marks phases in the vicinity of the heartbeat. (c) Frequency spectrum of the light curve shown in panel b. In panels c and d orbital harmonics are marked with red vertical dashed lines (except for TEOs, which are marked with blue) and labeled with n. (d) Frequency spectrum of the light curve shown in panel b with data at phases marked by the grey stripe removed. A small trend in the light curve was also removed.	2012.11559	623-teo.jpg
0.413486	e2b0c63d138347aa96d92445b8c1b1bb	TESS light curves and frequency spectra of θ^1Cru. (a) TESS light curve phased with the orbital period. Phase 0.5 corresponds to the epoch of periastron passage. (b) Residuals from the fit of Kumar's model. Red dots are median values in 0.0075 phase bins. Grey stripe marks phases in the vicinity of the heartbeat. (c) Frequency spectrum of the light curve shown in panel b. In panels c – e orbital harmonics are marked with red vertical dashed lines (except for TEOs, which are marked with blue) and labeled with n. (d) Frequency spectrum of the light curve shown in panel b with data at phases marked by the grey stripe removed. (e) The same as in panel d, but after prewhitening TEOs with n=7 and 10.	2012.11559	701-teo.jpg
0.473304	923ddd31ced043489e3202f9c5f60781	TESS light curves and frequency spectra of ζ^1UMa. (a) TESS light curve phased with the orbital period. Phase 0.5 corresponds to the epoch of periastron passage. (b) Residuals from the fit of Kumar's model. Red dots are median values in 0.01 phase bins. Grey stripe marks phases in the vicinity of the heartbeat. (c) Frequency spectrum of the light curve shown in panel b. In panels c – e orbital harmonics are marked with red vertical dashed lines (except for TEOs, which are marked with blue) and labeled with n. (d) Frequency spectrum of the light curve shown in panel b with data at phases marked by the grey stripe removed. (e) The same as in panel d, but after prewhitening TEOs with n=3, 5, and 6.	2012.11559	764-teo.jpg
0.391961	206309d6682344bca27d538c945e4cea	TESS light curves and frequency spectrum of V1294 Sco. (a) TESS light curve phased with the orbital period. Phase 0.5 corresponds to the epoch of periastron passage. (b) Zoom of the light curve after cutting out the eclipses. Continuous dark blue line is the fitted Kumar's model. For comparison, the dashed line shows the fit of Kumarâs model with inclination fixed at the value derived by <cit.>. (c) Residuals from the fit of Kumar's model. The red dots are median values in 0.02 phase bins. Grey stripe marks phases in the vicinity of the heartbeat. (d) Frequency spectrum of the light curve shown in panel b. Orbital harmonics are marked with red vertical dashed lines (except for the TEO at n=7, which is marked with blue). Orbital harmonics and labeled with n.	2012.11559	925-teo.jpg
0.466719	559dd3c757d745e1a814540c9e5c174b	TESS light curves and frequency spectra of HD 158013. (a) TESS light curve phased with the orbital period. Phase 0.5 corresponds to the epoch of periastron passage. (b) Residuals from the fit of Kumar's model. Red dots are median values in 0.01 phase bins. (c) Frequency spectrum of the light curve shown in panel b. In panels c – d orbital harmonics are marked with red vertical dashed lines (except for TEOs, which are marked with blue) and labeled with n. (d) The same as in panel c but after removing trend in residuals and low-frequency terms.	2012.11559	964-teo.jpg
0.480497	50de24f8ae3b43f6b0a94180c05dd783	TESS data of HD 5980 freed from eclipses and long-period variability and phased with the period of 0.252527 d (top) and twice longer, 0.505055 d (bottom). Phase 0.0 corresponds to TBJD = 0.0. Red dots are averages in 0.05 (top) and 0.02 (bottom) intervals of phase.	2012.11559	Fig-HD5980-SP.jpg
0.441915	3fd3a98fa2ab4ad7af64f6b8aaf1cb52	ASAS-3 V-filter light curve of HD 5980 phased with the orbital period of 19.2656 d. The time of periastron passage corresponds to phase 0.0. Red circles denote averages of brightness calculated in the 0.03 phase bins. The heartbeat is clearly visible at the same phase (∼0.05) as in the TESS data (Fig. <ref>).	2012.11559	HD5980-asas.jpg
0.453151	5c03af748be145af9acd706f0b5d0b07	Top: TESS light curve of HD 5980. The continuous and dashed vertical lines mark times of periastron passage of the A – B, and C – D systems, respectively, both calculated from the ephemerides given by <cit.>. For the latter system, shaded band stands for ± 1σ error. Middle: PHOEBE 2 model of the TESS light curve of HD 5980. Details of modelling are given in the text. Bottom: Zoom of the model shown in the middle panel. For comparison, Kumar's model generated with the same orbital parameters is shown with the light red line.	2012.11559	HD5980-phoebe-model-comparison.jpg
0.441172	b7766101258446178302dad8a5bb77f7	TESS light curves of SW CMa (top), V1647 Sgr (middle), and V477 Cyg (bottom). Left: TESS light curves phased with the orbital periods. Phase 0.5 corresponds to the epoch of periastron passage. Right: Zoom of the light curve after cutting out the eclipses. Continuous dark blue line is the fitted Kumar's model. For comparison, the dashed line shows the fit of Kumarâs model with inclination fixed at the value given in Table <ref>.	2012.11559	eclipsing-hb.jpg
0.449652	7d16b731ac674cd688e8e51ce2e7ad6a	Frequency spectra of three O and early B-type HBSs (upper panels) and four A and F-type HBSs (lower panels) showing their intrinsic variability other than TEOs. Frequency spectra were calculated after subtracting fitted Kumar's models and TEOs (if present).	2012.11559	int-var.jpg
0.485792	1f23b7846b1047e1964e48fa6994e4e8	Relation between normalised tidal potential energy ε̃ and total mass M_total of heartbeat systems. Red circles stand for a sample of 14 Kepler HBSs which are red giants <cit.>. Green circles are 19 Kepler HBSs with radial-velocity measured by <cit.>. Known massive HBSs are marked by dark blue circles and labeled. Our sample of 20 massive HBSs is plotted with light blue circles. The eclipsing systems are encircled, those with detected TEOs are additionally marked with plus signs; see the text for explanation. The size of symbols depends on the peak-to-peak amplitude of heartbeats; the higher amplitude, the larger symbol.	2012.11559	mass-epsilon-conclusions.jpg
0.467488	0acaaef468ff456e91685770f9dc5b6d	TESS light curves of nine non-eclipsing HBSs without TEOs. Continuous dark blue lines are the fitted Kumarâs models.	2012.11559	modeled-hb.jpg
0.422072	38482023653c41ce988acee593365c59	Visual Schematic of the initial set up of the simulations for (a) the pillared surface and (b) the porous surface.	2012.11562	CI.jpg
0.4745	a7b604b7fd8e4c8b9a1527ba6c46ac19	(a) average acceptance ratio as a function of parameter T for a droplet of water (f_o=0) and a droplet of oil (f_o=1. (b) average number of neighbors of a liquid spin that have a gas neighbour.	2012.11562	Paper_AccViz.jpg
0.439619	aa5b24e90e8c489992708d86ef320acd	Green (coral) curve is the volume of the water (oil) component of the droplet divided by its target value for a drop with f_0=0.50.	2012.11562	Paper_alfaW.jpg
0.389115	ed0a5596838142929c6a9dc9620f1000	Theoretical wetting diagrams for (a) water droplet placed on a pillared surface, (b) water droplet placed on a porous surface, (c) oil droplet placed on a pillared surface and (d) oil droplet placed on a porous surface, as a function of two geometrical parameters of the surface: the height of the pillars h, and the interpillar/interpourous distance a. Pillars width are kept constant =5μ m. The solid line in (a) represents the predicted thermodynamic transition between the Cassie-Baxter (CB) and (W) states. Colors indicate the droplet's contact angle, in degrees.	2012.11562	ang_est.jpg
0.435584	ebbea89a0edb40c4bccbad31ef16afb5	Efficiency for pillared substrates. ξ^ a_ pil, ξ^ r_ pil and ξ_ pil as a function of interpillar distance for (a)f_o=0.10 and (b)f_o=0.90.	2012.11562	eff_pilares.jpg
0.431302	60a9187aac2c4f538a134690689f24fa	Results for porous substrates. ξ^ a_ por, ξ^ r_ por and ξ_ por as a function of interpillar distance for (a)f_o=0.10 and (b)f_o=0.90	2012.11562	eff_poro.jpg
0.471188	a70490963dbd4c77a67e7a17dccc959e	(a) red curve is Eq.(<ref>) for the CB state and in green is Eq.(<ref>) for the W state. (b) derivative of energy as a function of the angle parameter θ_i. Dotted line shows where Δ E^ CB and Δ E^ W have their minimum value and the contact angle associated with it.	2012.11562	minimo.jpg
0.473722	0fdff5a1cff5459ca8bd902e5b529157	Results for pillared substrates. Interpillar volume of water, V^w_p, and oil, V^o_p, as a function of the pillar distance for (a) f_o = 0.10 and (b) f_o = 0.90. Blue color represent water and orange represent oil. ξ^ a_ por, ξ^ r_ por and ξ_ por as a function of interpillar distance for (c)f_o=0.10 and (d)f_o=0.90	2012.11562	pilares_h_5.jpg
0.442533	ae3b9ecee61341d0bdfc4dd607079ca0	Results for pillared substrates for several geometric parameters. Above: Interpillar volume of water, V^w_p, and oil, V^o_p, as a function of the pillar distance for (a) f_o = 0.10 and (b) f_o = 0.90. Below each figure: Cross sections of the final droplet configuration of Monte Carlo simulations for a=2 μm and a=14 μm for each corresponding f_o. To build this figure, a mixed droplet with oil fraction given by f_o (value specified above the figures) and total volume given by V^T=4/3π R_0^3 with R_0 = 50 μm is simulated on a substrate with =5 μm, =10 μm and varying pillar distance . Blue color represent water and orange represent oil.	2012.11562	pilares_novo.jpg
0.439789	be0f3b7428b5490cae40fe93d551ecdf	Results for porous substrates. Interporous volume and reservoir volume of water, V^w_p and V^w_r , and oil, V^o_p and V^o_r , as a function of the porous distance for (a) f_o = 0.10 and (b) f_o = 0.90. Blue color represent water and orange represent oil. ξ^ a_ por, ξ^ r_ por and ξ_ por as a function of interporous distance for (c)f_o=0.10 and (d)f_o=0.90	2012.11562	poros_h_5.jpg
0.46755	94cedbb28c5044f0a0bf4872fb98b6ff	Results for porous substrates for several geometric parameters. Above: Interporous volume of water V^w_p and oil V^o_p and reservoir volume of water V^w_r and oil V^o_r as a function of the pillar distance for (a) f_o = 0.10 and (b) f_o = 0.90. Below each figure: Cross sections of the final droplet configuration of Monte Carlo simulations for a=2 μm and a=14 μm for each corresponding f_o. To build this figure, a mixed droplet with oil fraction given by f_o (value specified above the figures) and volume given by V^T=4/3π R_0^3 with R_0 = 50 μm is simulated on a substrate with =5 μm, =10 μm and varying porous distance . Blue color represent water and orange represent oil.	2012.11562	poros_novo.jpg
0.502261	7ddb257e3eac45138b251d8f13de4c40	Definition of the geometric parameters of the substrates and the droplet. (a): pillared surface with width , pillar distance and height . (b): porous surface with width , porous distance and height . (c): geometric parameters of the droplet. We consider a spherical cap with radius R, base radius B, height H, and contact angle θ_C.	2012.11562	surface_mod.jpg
0.42684	fa8d81ff28a24c7b99f21a033998ad58	Measured positron yield as a function of average laser intensity over the spot size (1/e^2). Diamond markers indicate current experiments. Scaling model (line) and data from other facilities (dots) from Ref. <cit.> and Ref. <cit.>, respectively.	2012.11563	PositronScaling_v3-eps-converted-to.jpg
0.384601	0d33c03598554198aa362d32218b65a7	The for each individual experiment. The experiment ID is listed on the abscissa and are identified in Tables <ref>, <ref>, <ref> and <ref>. The original DIS data used in the nCTEQ15 fit are in blue, the DY data in green, and the new JLab data in yellow. We also list the total for each fit inset in the plot.	2012.11566	chi2-trials.jpg
0.468139	8e6decc6e5aa454a8b898acbec61c357	We display the ratio of the modified (smeared/rescaled) PDF compared to the unmodified PDF for both the Gaussian smearing and the x-rescaling. For the Gaussian smearing (blue), we display S⊗ f(x) over f(x) with Δ=0.05, and δ=-0.007 (which resembles an A=12 nucleus). For the x-rescaling (red), we display f(x'_A)/f(x) for ε=0.02, κ=10, and A=12. While there is a small relative difference at intermediate to small x values, the rise at large x is quite similar for both methods.	2012.11566	comparesmear.jpg
0.409669	6d8a31cbda184c71954ccf7d428cf647	We display the F_2^A/F_2^D ratio of selected data sets sorted by nuclei. The data from the original nCTEQ15 DIS points are in blue, and the DY in green. The new JLab DIS data are in yellow.	2012.11566	dataquality.jpg
0.536915	485697f0acaa497d8a86f2ad9501e692	We show F_2^A/F_2^D for selected data sets in the large x region. The different curves show the impact of the variations of ϵ={0, 0.01,0.03,0.1 } (solid blue, dashed red, dot-dashed green, dotted purple) with κ=10 in the rescaling of Eq. (<ref>). The combined χ^2 values are displayed in the legend, both with and without the W>1.7 GeV cut. For reference, the bottom curve (blue, solid) in each plot corresponds to ϵ=0 which is the result. Points which are excluded from the fit by the W>1.7 GeV cut are indicated with a hollow (gray) symbol; this is evident for the Fe and Cu data sets. For reference, experiment Fe/D is ID=5131, Al/D is 5134, Cu/D is 9984, and C/D is 9990.	2012.11566	epsscan.jpg
0.447282	c314bd35bce44f3bbff7cb91031fec8a	We display the ratio F_2^TMC(x,Q)/F_2^leadingTMC(x,Q) for Q={1,1.3,2,3} GeV (blue, yellow, green, red) for neutral current DIS; the lowest Q values yield the largest corrections. Each colored band contains results for {D, C, Al, Fe, Au, Pb} nuclei, where D is the lowest and Pb is the highest within each band.	2012.11566	figure3acot.jpg
0.481095	66dd5d93ef16404aa58d8fc492d893eb	We display the F_2^A/F_2^D ·( F_2^D/F_2^p )_CJ ratio for selected data sets sorted by nuclei. We also overlay the theoretical prediction of in blue. The theory predictions have been calculated at averaged Q values where data sets overlap.	2012.11566	fitquality.jpg
0.391105	e4b0706e5b64448884035fe8baf637ff	We display the higher twist corrections as a function of x at Q=2 GeV for selected nuclei. The larger A values yield larger corrections since they scale as A^1/3. The upper frame shows the correction factor of Eq. (<ref>), and the lower frame shows the correction scaled to the proton.	2012.11566	htcorrections.jpg
0.39209	114a378a2f6642189d86500e3a328d9a	We compare our results to other nPDF sets from the literature including EPPS16 <cit.>, nNNPDF2.0 <cit.>, and TUJU19 <cit.>. We plot xf(x,Q) for carbon ^12C at Q=2 GeV on a log scale.	2012.11566	pdfabs_others.jpg
0.507594	a49f34bdd1c24d02ac2672ebff6d075a	Iron (^56Fe) PDFs ratio compared to nCTEQ15 at Q=2 GeV. We show the uncertainty bands for nCTEQ15 (blue) and (yellow) computed with the Hessian method.	2012.11566	pdfratios_iters_loglin_Fe.jpg
0.52704	b4f4d110c2ce44bfb19c222da9568e3a	Lead (^208Pb) PDFs ratios compared to nCTEQ15 at Q=2 GeV. We show the uncertainty bands for nCTEQ15 (blue) and (yellow) computed with the Hessian method.	2012.11566	pdfratios_iters_loglin_Pb.jpg
0.468019	345845e8547e42f4bfc4b77e11e662e6	We compare our results to other nPDF sets from the literature including EPPS16 <cit.>, nNNPDF2.0 <cit.>, and TUJU19 <cit.>. We plot the nPDF ratio for carbon ^12C at Q=2 GeV compared to on a log-linear scale.	2012.11566	pdfratios_others_loglin.jpg
0.417894	4a567833d586499cb32eeb3dde61c697	We display DIS and DY data entering our analysis in the {x,Q^2} space indicating the relevant kinematic cuts, where x and Q^2 are the usual DIS variables, and Q^2 for DY is the di-lepton mass squared. The more restrictive cuts of Q=2 GeV and W=3.5 GeV (black dashed line) are the cuts used in the original nCTEQ15 analysis. In the present work, we will relax the cuts to Q=1.3 GeV and W=1.7 GeV (red dashed line). This greatly expands the kinematic reach in the high-x region where much of the new JLab data is located.	2012.11566	q2x-space.jpg
0.400825	0509341223fd43ea958e50f4641a066d	We show the impact of the x-rescaling of Eq. (<ref>) for varying choices of ε and κ for a “toy” PDF, f(x)∼(1-x)^3/x^1.5. The plot shows the ratio f(x'_A)/f(x) for 2 groups of 5 curves for A=208. The largest 5 curves (solid) are for ε=0.01 with exponent κ={2,4,6,8,10} indicated by the colors {black, red, blue, green, purple}, respectively. The lower 5 curves (dashed) are for ε=0.001 with the same range of κ values.	2012.11566	toysmear.jpg
0.468828	aa870af357904482b20cff272c0d6d84	We display the classic F_2^A/F_2^D ratio for carbon illustrating the nuclear correction factor across the various x regions. The black points indicate the data used in the original nCTEQ15 fit, and the red points with the solid squares represent the additional data from this original set which are now included due to the relaxed Q and W cuts. The red open squares are the new JLab DIS data included in this analysis, and the blue points are those JLab DIS data which are excluded by the current kinematic cuts.	2012.11566	typical-f2ratio.jpg
0.565362	1610bde78f4c4f35a2ebd98cfee685da	Schematic representation of a Schwarzschild black hole and the parameters in scenario 2. The dashed circle is constant r and t hypersurface P. Distance ℓ_0 is from the point where the NEC starts being violated, and distance ℓ is from P to the singularity (pictured here at r=0).	2012.11569	blackhole.jpg
0.537658	ca02a29ee9cb43839fb46396fbbd9780	Schematic representation of a Schwarzschild black hole and the parameters in scenario 1. The dashed circle is constant r and t hypersurface P. Distance ℓ_0 is measured from the hypersurface P until the NEC starts being violated. Distance ℓ is from P to the singularity (pictured here at r=0).	2012.11569	blackhole2.jpg
0.520401	ed4da046432344aeaeeebd4d588e95e6	Qualitative plot of the right hand side of eq.(<ref>) and asymptotic behavior for large and small σ.	2012.11569	p1.jpg
0.578971	76166fc939844069949387c3e93d40f7	Required value of x to have a singularity for different values y. For Q_1=0.1 the minimum value is much smaller compared to Q_1=1.	2012.11569	scen2.jpg
0.478749	bf91e1da042c4a6590f30a9c2b00385a	Phase diagram of the total thermo-osmotic response coefficient M_, separating a thermophobic (M_ >0) or a thermophilic (M_<0) response, as a function of the Debye length and surface charge density Σ<0, for different wettings. One can observe a bigger effect of wetting for NaCl, while the change of sign is observed for KCl at higher values of \|Σ\|.	2012.11571	PAPERphasediag_Mtot.jpg
0.466176	5f895cb73706427a9bea3484ec57e883	(a) Effective slip length b^eff = b-z_s, with b the slip length and z_s the shear plane position. One can distinguish a slip situation when b^eff>0 and a stagnant layer situation when b^eff<0. (b) Modelled system for the measures of water enthalpy excess density and b^eff, constituted by water enclosed between two generic planar parallel walls. (c) Water enthalpy excess density δ h^wat profiles as a function of the wetting, controlled by the interaction energy between the liquid and the solid atoms ε_LS.	2012.11571	PAPERsystem_dh_beff.jpg
0.454486	b47f75bc7c3143b59e866246944c2321	In continuous line the total thermo-osmotic response coefficient \|M_^tot\|, and in discontinuous lines its electrostatic \|M_^el\|, solvation \|M_^sol\| and water \|M_^wat\| contributions, for θ∼ 120^∘ (corresponding to b^eff=1.20nm) and a surface charge density Σ = -0.01 C/m^2. The solvation contribution dominates for a wide range of Debye lengths , but a competition between the water and the solvation contribution appears for KCl at small . One can also note for KCl a change of sign in M_^tot as a consequence of the change of sign of the solvation contribution, with M_>0 for small and M_<0 for large . For NaCl, M_<0 in the whole range of .	2012.11571	PAPERtotalMto_comp_sC0.01.jpg
0.400935	716601f3359c4104bc858171af0654ce	Picture of a part of an antibiogram showing an antibiotic disk and its inhibition zone (left). On the right a plot of the intensity radial profile and the segmentation with slopeOP with isotonic constraint.	2012.11573	AST_radial_profile.jpg
0.530637	36818f3dcb8c44c28cfc1545ca951817	Results for scenarios 1 to 4 with noise σ = 12. Each point of the MSE curve is the mean over 30 independent simulations. For the smallest MSE in each of the 12 MSE curves, we present the 30 corresponding results in the boxplots. The last column shows the obtained segments for the best of the 30 results in boxplots for OP-2D and slopeOP.	2012.11573	beta-mse-boxplots_sigma12.jpg
0.497842	98d94c9ce9634e1f85ae14d55b8ce425	Results for scenarios 1 to 4 with noise σ = 24	2012.11573	beta-mse-boxplots_sigma24.jpg

0.514045

e7175813db3c4e50a49a277fdac746dc

The overall pipeline of our proposed approach. Note that the dashed arrows only active in RGC^†, as shown in the top-right corner.

2012.11294

pipeline.jpg

0.560015

6393c7afd47b42f2881ff704b1f1b5a6

Conceptual diagram of our proposed centralized information interaction strategy. Left: Typical U-shape structure which connects the corresponding stages in the bottom-up and top-down pathways directly; Right: Our proposed centralized strategy. The rounded gray rectangle refers to a shared module that can parallelly process all stages (scales) of features in a stage-wise manner.

2012.11294

teaser.jpg

0.429222

2806115adb504c56a6e9f5efdd609f97

Visualizing feature maps w/o and w/ CII. Every translucent gray rectangle refers to a sequence of two 3 × 3 convolution layers. As can be seen, CII can interact the complementary information encoded in feature maps of different scales (c,d). For comparison, w/o CII, the feature maps are roughly unchanged (a,b).

2012.11294

visual.jpg

0.452479

43a395dc9d1b4917b841e3063c84cdce

KPI distribution datset 1

2012.11299

KPI_distribution_DS1.jpg

0.447969

450232e1a0ae4d649966c86ca96091f9

KPI distribution dataset 2

2012.11299

KPI_distribution_DS2.jpg

0.489018

c53252788be844ab91df84b1ef4b473d

KPI prediction dataset 1 with scalar based meta-model

2012.11299

KPI_prediction_data_1.jpg

0.494671

01f8c4cce95e4a75b4815589526e4da6

KPI prediction dataset 2 with scalar based meta-model

2012.11299

KPI_prediction_data_2.jpg

0.477795

8a158de563884f21b35c9ae8eb5adb84

Parameter distribution dataset 1

2012.11299

Param_distribution_DS1.jpg

0.389469

6e390a80a57a4a88ba0577ba75d80ed9

Parameter distribution dataset 2

2012.11299

Param_distribution_DS2.jpg

0.462003

1cad76bf25be421595132fefb7f35e29

Cumulative accuracy plots of the KPI prediction with relative error ε_mre<5% dataset 1

2012.11299

Relative_error_distr_data_1.jpg

0.433644

9a0c35c9278e453e913368d92d1001ed

Cumulative accuracy plots of the KPI prediction with relative error ε_mre<5% dataset 2

2012.11299

Relative_error_distr_data_2.jpg

0.424201

180d0f73c48e4c02bfabe2406b5282da

Comparison between the 21st-order approximant of the ADM (dotted line) and the 4th iteration of the BLUES function method (dashed line) for α =1/2 and n=4. The numerically exact solution (red, full line) is also shown.

2012.11303

ADM_vs_BLUES_half.jpg

0.433833

8dc7e53470c14a369cd3ed844653ffc5

2012.11303

Fisher.jpg

0.482604

88072cf8303f43819218eb9eda6337f1

Wavefront values U_ψ^(n)(-1) at z = -1 versus order n for the Fisher equation (<ref>). The numerically exact value (red, dashed line) is also shown. Inset: A semi-log plot of the increments |Δ U_ψ^(n,n-1)(-1)| of the approximants versus n, and a linear fit. These increments are defined as the difference of the n-th and (n-1)-st approximant. Parameter values are k=1/3 and K/k = 1/2.

2012.11303

Fisher_maximum.jpg

0.533174

d7254928ddf6481f8f302b3f5a45cbfa

Residual function U_ψ^(n) of the approximants of order n = 0, 1, 2, 3 for the nonlinear Fisher DE with exponential corner source (<ref>). The residual operator applied to the numerical solution (red, full line) is also shown. Parameter values are k = 1/3 and K/k = 1/2.

2012.11303

Fisher_residuals.jpg

0.420582

73ef8a955ce441e28814a90577b6ca68

2012.11303

Fisher_zoom.jpg

0.469078

7e948b9c0c3049ab89cbcaed093edc57

Residual ℛ_t U(t) = -U^4(t) for α = 1/2 for different orders of iteration. The numerically exact residual ℛ_t U_num is also shown (red, full line).

2012.11303

fractional_alpha_0p50_residual.jpg

0.412458

f1883258e98e446dbdd85617c8e7c505

(a) The mechanical oscillator was driven by optical interface forces 𝐅_interface and tension 𝐅_tension of the four fibers, where the laser was propagating. These forces, illustrated for one of the fibers, were modulated by varying the laser intensity. The laser generated by a multimode laser diode was split into the four fibers by a 1x4 splitter and finally the laser exited from the ends of the four fibers, which were all bonded to the oscillator. The wavelength was either at 808 nm or 915 nm. (b) The cross-section of the 0.22 NA multimode fiber (Thorlabs, FG105LCA). (c) The nanoscale oscillation was detected by the Michelson interferometer utilizing a separate He-Ne laser. See the appendix for a more detailed description of the experimental setup.

2012.11308

fig1.jpg

0.501212

164b63f276354780971bbe8b87712834

The displacement amplitude of the mechanical oscillator is plotted as a function of the laser modulation frequency (a) for the first oscillator at the wavelength of 808 nm and (b) for the second oscillator at the wavelength of 915 nm. The net peak-to-peak power amplitude of the driving field in the four fibers together is P_0=5.125 W in both cases. The solid line represents the averaged frequency spectrum measured with a single modulation frequency. The peak at each modulation frequency is marked with a red dot. The oscillator response function is fitted and shown by the dashed line. The corresponding peak-to-peak force amplitudes of the two oscillators are plotted in (c) and (d) as a function of the peak-to-peak laser power amplitude of the fibers. The solid lines represent the regression lines and the dashed lines show the net theoretical force in Eq. (<ref>). The dash-dotted and dotted lines are the results of the Minkowski (F_M,0) and Abraham (F_A,0) momentum models, respectively, using Eq. (<ref>) with the corresponding momentum of light and excluding the tension in Eq. (<ref>).

2012.11308

fig2.jpg

0.422858

99aa22fa69f740f18eea3b4f52219039

Illustration of the fiber path and the related horizontal and vertical distance parameters a and h.

2012.11308

fig3.jpg

0.444229

c9e922aaadcb4a3a99ba0138d59816ba

Elasticity theory simulation of the total tension force of four fibers on the oscillator as a function of the total incident optical power propagating through the fibers. The dots represent the simulation results obtained by using the expression on the left hand side of Eq. (<ref>) and the dashed line is the theoretical line, given on the right hand side of Eq. (<ref>).

2012.11308

fig4.jpg

0.497802

be73ec07307b47dcb8f658100378ea5b

Qualitative results. Leftmost, we depict the template outfit of each row. Note how templates are smoothed as much as possible. Then, for each row we visualize the output for two unseen test poses. For each sample, we show the output dressed human (left) along with its corresponding deformed outfit in rest pose (right). As it can be seen, the learnt Pose Space Deformations generate folds and wrinkles to fulfill the energy balance requirement of the physical system. The last row depicts results of simulation with a different body shape (and its appropriately aligned template outfit) to show generalization to different bodies. All of the shown samples were obtained after just a few minutes of training.

2012.11310

PBNS_PSD.jpg

0.427121

8f8ae3e860a5494f9500b6735ece0f97

Methodology overview. Pose parameters are fed to a 4-layer Multi-Layer Perceptron with 32 dimensions in each layer. The output high-level pose embedding is multiplied with a PSD matrix 𝐃 to obtain deformations for the garment/outfit. As standard, deformations are applied in rest pose garments and skinned along with the body (note that garment blend weights are assigned by proximity w.r.t. the body in rest pose). Finally, we train by optimizing the potential energy of the output physical system represented by the body and the cloth. The parts of the model in red correspond to the trainable parameters.

2012.11310

PBNS_Pipeline.jpg

0.442317

a0beb1871a6a49fa8a2ec6b4f302a98d

Architecture ablation study. We perform three experiments to assess the effect of the proposed MLP: 1) PBNS as described, 2) θ as input with no MLP and 3) rotation matrices 𝐑_θ as input with no MLP. For experiment 2 we observe unrealistic V-shaped wrinkles around the hip. Also, for both models with no MLP, we observe artifacts where both legs merge. This effect is more evident on skirts. The merging point of left and right side of the skirt presents an important artifact for experiments with no MLP, despise being trained during hundreds of epochs (as opposed to 15 epochs for PBNS with MLP).

2012.11310

ablation.jpg

0.460932

aad780fe01934ba985171d56a8b9162a

PBNS is not limited to work with SMPL. Since it needs no PBS data, the engineering effort for the adaptation to custom avatars is minimum. Here we show different garments and outfits neurally simulated on top of custom avatars. For each, we show the avatar body in rest pose and two different viewpoints of a posed sample. PBNS can be used to enhance rigged 3D characters by animating their clothes.

2012.11310

avatar.jpg

0.435441

918e55582f424846821ea1658c080d6b

PBNS results on outfits with overlapping layers of cloth. Each row shows a sample. From left to right: template outfit, posed sample without outermost layer, posed sample with whole outfit and cross section. The proposed collision loss can handle multiple layers of cloth, even when the template outfit presents inter-penetrations. As we can observe on the cross section, PBNS is able to generate consistent predictions with minimal layer spacing. Also note how it is possible to neurally simulate different fabrics –different wrinkle count and size– within the same outfit by assigning per-vertex weights to cloth related losses.

2012.11310

layers.jpg

0.448594

cab79dc8f34b4eabaf7f9c284c2914a6

Textured qualitative samples. This rendering gives an approximate idea to the final application-level looks that our methodology yields. These results correspond to neural simulations of 3 additional outfits on top of 3 different bodies, further showing the generalization capabilities of our methodology.

2012.11310

qualitative_textured.jpg

0.476968

3237791ad3d24a939e553d2de3f6827b

Results obtained using PBNS for outfit resizing. Body shape changes from left to right. Outfit tightness changes from upper to lower rows. All the previous properties discussed for PBNS are also present when it is used for outfit resizing.

2012.11310

resize.jpg

0.480326

c5d1467dbcf048b993dc08399909e4d1

Supervision vs PBNS. We compare results obtained with PBNS against a purely supervised approach and a hybrid approach (L2 plus Eq. <ref>). The same architecture and the same N = 3000 poses (2550 for training and 450 for validation) are used in these experiments. We additionally compare against PBS data used for supervised training. We show: a) L2 only, b) hybrid, c) PBNS and d) PBS data. The supervised approach has the lowest Euclidean error w.r.t. PBS data, but also lowest number of wrinkles and highest number of collisions. The hybrid approach has more visible wrinkles and fewer collisions, but generalizes poorly to extreme poses. PBNS has the highest Euclidean error, but realistic pose-dependant wrinkles and shows no collisions, even under extreme poses. We also observe how PBS is prone to failures when body presents self-collisions (upper-left sample right elbow and lower-right sample legs). These failures might be transferred to predictions if trained supervisedly. PBNS is more robust to collisions than PBS.

2012.11310

vs_supervised.jpg

0.481153

184ebbf41b0149ab89874baf8c440b08

PBNS vs. TailorNet. We show two different bodies and outfits in three different poses each. For each pose, from left to right: TailorNet raw predictions, TailorNet after post-processing and PBNS. As observed, TailorNet heavily relies on post-processing, hurting performance and removing model differentiability. TailorNet predictions are noisy, show a bias towards body geometry and is unable to model cloth-to-cloth interactions. On the other hand, PBNS predictions show physically cloth-consistent predictions without body or cloth-to-cloth interpenetrations (raw predictions).

2012.11310

vs_tailornet.jpg

0.436244

6d39b5eadcd64817a8d07b6071131571

First-principles spin-phonon relaxation times (τ_s) in graphene, graphene with 0.4 eV/nm electric field and graphene on hBN, as a function of carrier density at room temperature. (a) The out-of-plane to in-plane ratio deviates dramatically from 1/2 for the hBN substrate, partially captured by the DP formula due to the out-of-plane SO field (FIG. <ref>(b)). (b) Spin lifetime is appreciably inversely correlated with momentum lifetime (τ_p), as expected in the DP mechanism, only for graphene on hBN. Most importantly, both the (c) in-plane and (d) out-of-plane spin lifetime dramatically reduce due to inversion symmetry breaking by either electric field or hBN substrate, with the dependence on carrier density flipped relative to the free-standing case.

2012.11550

nCarrTauGrid.jpg

0.438827

e695bd48fef14dcc8c20f8bcdb0fbc3a

(a) Out-of-plane and (b) in-plane spin-phonon relaxation times in graphene on hBN can be further tuned by an applied electric field. (Experimental data from Ref. .) With a single gate, the electric field is dependent on the carrier density and cannot be varied independently. Variation of τ_s⊥/τ_s∥ ratio with (c) carrier density at specified electric fields, and with (d) electric field at a carrier density of 3.7× 10^12 cm-2, showing strong deviations from 1/2 in agreement with experiment.<cit.>

2012.11550

nCarrTauGridEfGrhBN.jpg

0.494858

fb96d351ca1d4618839bc54c58ea95e0

Band structure in the vicinity of the K point (left panels) and corresponding k-dependent effective SO magnetic field B⃗_k⃗ at a Fermi circle 0.1 eV above the conduction band edge (right panels) for (a) free-standing graphene, (b) graphene with electric field, E_z=0.4 eV/nm, and (c) graphene on hBN. The effect of hBN differs qualitatively from that of an electric field, splitting the valence band stronger than the conduction band and introducing an out-of-plane component to B⃗_k⃗.

2012.11550

schematic.jpg

0.561134

625c9380a21b4f109dc160b03a87833c

Illustration on why VAEs produce blurry reconstructions. Consider the example of a binary frontier in an image i and a latent code z which corresponds to the position of the frontier x_front. If q_θ_e(z|i) = 𝒩(x_front, σ) then p_θ_e(x_front| z) = 𝒩(z, σ) and the optimal reconstruction of the pixel at position x is 𝔼[pixel(x) | z ] = 1 ×ℙ_θ_e(x>z) + 0 ×ℙ_θ_e(x<z) = 1/2(1+erf(x-z/√(2)σ)) which is a smooth transition between black and white instead of a sharp transition in the original binary image.

2012.11551

binary_front.jpg

0.54737

d5bb7fcf9d104b0c90248449495daece

Illustration of a toy example with two-dimensional data and a one-dimensional latent space. Points: data, dotted line: manifold of reconstructions from VAE, dashed line/density: manifold of reconstruction with our model. Color encodes the position in the one-dimensional latent space. Top: with a deterministic generator of the form G_θ_g(z). Bottom: with a probabilistic generator of the form G_θ_g(z, ξ). (best seen with zoom and color)

2012.11551

latent.jpg

0.440496

fcdf494bed4341b7bdaaeb3763d5e893

Dynamic BoW-prediction head. G(·) learns to quickly adapt to the visual words in the continuously refreshed vocabulary V. The outputs G(V) are in fact weights that are used for mapping the features () to the corresponding BoW vector y_().

2012.11552

bow_prediction_head_landscape.jpg

0.494136

70b0e118e5364087891c385741bf5ad9

Unsupervised learning with Bag-of-Words guidance. Two encoders and learn at different tempos by interacting and learning from each other. An image is passed through the encoder and its output feature maps ^l() are embedded into a BoW representation y_() over a vocabulary V of features from . The vocabulary V is updated at each step. The encoder aims to reconstruct y_() from data-augmented instances . A dynamic BoW-prediction head learns to leverage the continuously updated vocabulary V to compute the BoW representation from the features (). follows slowly the learning trajectory of via momentum updates.

2012.11552

overview_v2.jpg

0.41981

e89b5872a24d4bcea97f9b700c06168b

Vocabulary queue from randomly sampled local features. For each input image to , “local” features are pooled from ^l() by averaging over 3 × 3 sliding windows. One of the resulting vectors is selected randomly and added as visual word to the vocabulary queue, replacing the oldest word in the vocabulary.

2012.11552

vocabulary_queue.jpg

0.380766

79f2f2ae2c064401b32fbc37a1870666

Iso-contours of the 1σ sensitivity to C_W, in terms of the S parameter in eq. (<ref>), in the luminosity-energy plane. The FCC reach is also indicated, as well as the iso-contours of the relative statistical uncertainty of the cross-section measurement.

2012.11555

cW_S.jpg

0.440772

eba8bafcccbf459caf1d6f4db7221ac0

Transmission line model.

2012.11556

Line_parameters.jpg

0.401196

aaf01849a15e406c9c97c108f440d25e

Inverter filter circuit diagram, showing the notation and current direction convention used in this paper. The inverter itself is connected on the left, with v_iDQ representing the inverter voltage behind the filter.

2012.11556

filtertikzL.jpg

0.454655

aab50c4c8875468586292a69ac2e43aa

System response to load change: frequency.

2012.11556

freq5.jpg

0.387413

5894b10eccbf4f3db0f72279b13e4e90

The power system as an interconnection of bus and line dynamics.

2012.11556

ioDQ.jpg

0.477352

38c0e9fee47f4ba0868448e495bb3e96

Diagram of microgrid used for simulation study.

2012.11556

mgfig.jpg

0.419492

6477cc6e166a47568436941fdd98a81b

Plug-and-play response.

2012.11556

pp_pq.jpg

0.417944

125732c62872486c97e519545c32fce9

System response to load change: power.

2012.11556

pq5.jpg

0.433893

e87ba6ec90ce451699607cd8ca1c514e

System response to load change: voltage.

2012.11556

volt5.jpg

0.414075

fd1b33f35a0c48cea004ca8c851acecc

An example that shows up the intuition behinds the MIP-DoM approach. Consider two sets, P = {(2.0, 2.5), (3.0, 1.9)} and Q = {(2.2, 2.0), (3.0, 1.5)}. The DoM value indicates the total movement for P to weakly dominate Q. A typical vertical movement, zp_(1,2)=0.05 represents an improvement in f_2 objective generating p̂_1. However, p̂_1 does not dominate q_1. There is still a movement to be done in f_2 objective, represented by zpq_(1,1,2)=0.45, so that p'_1 will weakly dominate q_1. Observe that point p_2 does not weakly dominate q_2, so it is vertically moved obtaining p̂_2, in an attempt to weakly dominate q_2. With this vertical movement, indicated by zp_(2,2) = 0.40, p̂_2 dominates q_2, becoming p'_2. The DoM value is the total distance using the terms zp_(1,2), zp_(2,2), and zpq_(1,1,2). Limits on p' movements (lower (lp_(i,m)) and upper bounds (up_(i,m))) are also indicated.

2012.11557

model_sample.jpg

0.488515

7c4b83a755fc4ad3a1a9ddb1f2d013ef

Corner plot resulting from our MCMC analysis of the TESS photometry of 14 Peg. Red markers indicate the best-fit solution. Three vertical dashed lines superimposed on each marginalised 1D posterior distribution denote 16%, 50%, and 84% quantiles.

2012.11559

1334-pro_cornerplot.jpg

0.395083

419fb9fcd5e3404298894c95f900bba4

TESS light curves and frequency spectra of 14 Peg. (a) TESS light curve phased with the orbital period. Phase 0.5 corresponds to the epoch of periastron passage. (b) Residuals from the fit of Kumar's model. Red dots are median values in 0.01 phase bins. (c) Frequency spectrum of the light curve shown in panel b. In panels c – d orbital harmonics are marked with red vertical dashed lines (except for TEOs, which are marked with blue) and labeled with n. (d) The same as in panel c but after removing instrumental trend.

2012.11559

1334-teo.jpg

0.535908

c9adf52084fc4e39976144c380e57798

TESS light curves and frequency spectra of QX Car. (a) TESS light curve phased with the orbital period. Phase 0.5 corresponds to the epoch of periastron passage. (b) Zoom of the light curve after cutting out the eclipses and removing intrinsic variability (see text). Continuous dark blue line is the fitted Kumar's model. For comparison, the dashed line shows the fit of Kumarâs model with inclination fixed at the value derived by <cit.>. (c) Residuals from the fit of Kumar's model. The red dots are median values in 0.01 phase bins. Grey stripe marks phases in the vicinity of the heartbeat. (d) Frequency spectrum of the light curve shown in panel b. In panels d – f orbital harmonics are marked with red vertical dashed lines (except for TEOs, which are marked with blue) and labeled with n. (e) Frequency spectrum of the light curve shown in panel c. (f) Frequency spectrum of the light curve shown in panel c with data at phases marked by the grey stripe removed.

2012.11559

588-teo.jpg

0.439623

ad1df3a9042342cf98c9de2285577154

TESS light curves and frequency spectra of p Vel A. (a) TESS light curve phased with the orbital period. Phase 0.5 corresponds to the epoch of periastron passage. (b) Residuals from the fit of Kumar's model. Red dots are median values in 0.01 phase bins. Grey stripe marks phases in the vicinity of the heartbeat. (c) Frequency spectrum of the light curve shown in panel b. In panels c and d orbital harmonics are marked with red vertical dashed lines (except for TEOs, which are marked with blue) and labeled with n. (d) Frequency spectrum of the light curve shown in panel b with data at phases marked by the grey stripe removed. A small trend in the light curve was also removed.

2012.11559

623-teo.jpg

0.413486

e2b0c63d138347aa96d92445b8c1b1bb

TESS light curves and frequency spectra of θ^1Cru. (a) TESS light curve phased with the orbital period. Phase 0.5 corresponds to the epoch of periastron passage. (b) Residuals from the fit of Kumar's model. Red dots are median values in 0.0075 phase bins. Grey stripe marks phases in the vicinity of the heartbeat. (c) Frequency spectrum of the light curve shown in panel b. In panels c – e orbital harmonics are marked with red vertical dashed lines (except for TEOs, which are marked with blue) and labeled with n. (d) Frequency spectrum of the light curve shown in panel b with data at phases marked by the grey stripe removed. (e) The same as in panel d, but after prewhitening TEOs with n=7 and 10.

2012.11559

701-teo.jpg

0.473304

923ddd31ced043489e3202f9c5f60781

TESS light curves and frequency spectra of ζ^1UMa. (a) TESS light curve phased with the orbital period. Phase 0.5 corresponds to the epoch of periastron passage. (b) Residuals from the fit of Kumar's model. Red dots are median values in 0.01 phase bins. Grey stripe marks phases in the vicinity of the heartbeat. (c) Frequency spectrum of the light curve shown in panel b. In panels c – e orbital harmonics are marked with red vertical dashed lines (except for TEOs, which are marked with blue) and labeled with n. (d) Frequency spectrum of the light curve shown in panel b with data at phases marked by the grey stripe removed. (e) The same as in panel d, but after prewhitening TEOs with n=3, 5, and 6.

2012.11559

764-teo.jpg

0.391961

206309d6682344bca27d538c945e4cea

TESS light curves and frequency spectrum of V1294 Sco. (a) TESS light curve phased with the orbital period. Phase 0.5 corresponds to the epoch of periastron passage. (b) Zoom of the light curve after cutting out the eclipses. Continuous dark blue line is the fitted Kumar's model. For comparison, the dashed line shows the fit of Kumarâs model with inclination fixed at the value derived by <cit.>. (c) Residuals from the fit of Kumar's model. The red dots are median values in 0.02 phase bins. Grey stripe marks phases in the vicinity of the heartbeat. (d) Frequency spectrum of the light curve shown in panel b. Orbital harmonics are marked with red vertical dashed lines (except for the TEO at n=7, which is marked with blue). Orbital harmonics and labeled with n.

2012.11559

925-teo.jpg

0.466719

559dd3c757d745e1a814540c9e5c174b

TESS light curves and frequency spectra of HD 158013. (a) TESS light curve phased with the orbital period. Phase 0.5 corresponds to the epoch of periastron passage. (b) Residuals from the fit of Kumar's model. Red dots are median values in 0.01 phase bins. (c) Frequency spectrum of the light curve shown in panel b. In panels c – d orbital harmonics are marked with red vertical dashed lines (except for TEOs, which are marked with blue) and labeled with n. (d) The same as in panel c but after removing trend in residuals and low-frequency terms.

2012.11559

964-teo.jpg

0.480497

50de24f8ae3b43f6b0a94180c05dd783

TESS data of HD 5980 freed from eclipses and long-period variability and phased with the period of 0.252527 d (top) and twice longer, 0.505055 d (bottom). Phase 0.0 corresponds to TBJD = 0.0. Red dots are averages in 0.05 (top) and 0.02 (bottom) intervals of phase.

2012.11559

Fig-HD5980-SP.jpg

0.441915

3fd3a98fa2ab4ad7af64f6b8aaf1cb52

ASAS-3 V-filter light curve of HD 5980 phased with the orbital period of 19.2656 d. The time of periastron passage corresponds to phase 0.0. Red circles denote averages of brightness calculated in the 0.03 phase bins. The heartbeat is clearly visible at the same phase (∼0.05) as in the TESS data (Fig. <ref>).

2012.11559

HD5980-asas.jpg

0.453151

5c03af748be145af9acd706f0b5d0b07

Top: TESS light curve of HD 5980. The continuous and dashed vertical lines mark times of periastron passage of the A – B, and C – D systems, respectively, both calculated from the ephemerides given by <cit.>. For the latter system, shaded band stands for ± 1σ error. Middle: PHOEBE 2 model of the TESS light curve of HD 5980. Details of modelling are given in the text. Bottom: Zoom of the model shown in the middle panel. For comparison, Kumar's model generated with the same orbital parameters is shown with the light red line.

2012.11559

HD5980-phoebe-model-comparison.jpg

0.441172

b7766101258446178302dad8a5bb77f7

TESS light curves of SW CMa (top), V1647 Sgr (middle), and V477 Cyg (bottom). Left: TESS light curves phased with the orbital periods. Phase 0.5 corresponds to the epoch of periastron passage. Right: Zoom of the light curve after cutting out the eclipses. Continuous dark blue line is the fitted Kumar's model. For comparison, the dashed line shows the fit of Kumarâs model with inclination fixed at the value given in Table <ref>.

2012.11559

eclipsing-hb.jpg

0.449652

7d16b731ac674cd688e8e51ce2e7ad6a

Frequency spectra of three O and early B-type HBSs (upper panels) and four A and F-type HBSs (lower panels) showing their intrinsic variability other than TEOs. Frequency spectra were calculated after subtracting fitted Kumar's models and TEOs (if present).

2012.11559

int-var.jpg

0.485792

1f23b7846b1047e1964e48fa6994e4e8

Relation between normalised tidal potential energy ε̃ and total mass M_total of heartbeat systems. Red circles stand for a sample of 14 Kepler HBSs which are red giants <cit.>. Green circles are 19 Kepler HBSs with radial-velocity measured by <cit.>. Known massive HBSs are marked by dark blue circles and labeled. Our sample of 20 massive HBSs is plotted with light blue circles. The eclipsing systems are encircled, those with detected TEOs are additionally marked with plus signs; see the text for explanation. The size of symbols depends on the peak-to-peak amplitude of heartbeats; the higher amplitude, the larger symbol.

2012.11559

mass-epsilon-conclusions.jpg

0.467488

0acaaef468ff456e91685770f9dc5b6d

TESS light curves of nine non-eclipsing HBSs without TEOs. Continuous dark blue lines are the fitted Kumarâs models.

2012.11559

modeled-hb.jpg

0.422072

38482023653c41ce988acee593365c59

Visual Schematic of the initial set up of the simulations for (a) the pillared surface and (b) the porous surface.

2012.11562

CI.jpg

0.4745

a7b604b7fd8e4c8b9a1527ba6c46ac19

(a) average acceptance ratio as a function of parameter T for a droplet of water (f_o=0) and a droplet of oil (f_o=1. (b) average number of neighbors of a liquid spin that have a gas neighbour.

2012.11562

Paper_AccViz.jpg

0.439619

aa5b24e90e8c489992708d86ef320acd

Green (coral) curve is the volume of the water (oil) component of the droplet divided by its target value for a drop with f_0=0.50.

2012.11562

Paper_alfaW.jpg

0.389115

ed0a5596838142929c6a9dc9620f1000

Theoretical wetting diagrams for (a) water droplet placed on a pillared surface, (b) water droplet placed on a porous surface, (c) oil droplet placed on a pillared surface and (d) oil droplet placed on a porous surface, as a function of two geometrical parameters of the surface: the height of the pillars h, and the interpillar/interpourous distance a. Pillars width are kept constant =5μ m. The solid line in (a) represents the predicted thermodynamic transition between the Cassie-Baxter (CB) and (W) states. Colors indicate the droplet's contact angle, in degrees.

2012.11562

ang_est.jpg

0.435584

ebbea89a0edb40c4bccbad31ef16afb5

Efficiency for pillared substrates. ξ^ a_ pil, ξ^ r_ pil and ξ_ pil as a function of interpillar distance for (a)f_o=0.10 and (b)f_o=0.90.

2012.11562

eff_pilares.jpg

0.431302

60a9187aac2c4f538a134690689f24fa

Results for porous substrates. ξ^ a_ por, ξ^ r_ por and ξ_ por as a function of interpillar distance for (a)f_o=0.10 and (b)f_o=0.90

2012.11562

eff_poro.jpg

0.471188

a70490963dbd4c77a67e7a17dccc959e

(a) red curve is Eq.(<ref>) for the CB state and in green is Eq.(<ref>) for the W state. (b) derivative of energy as a function of the angle parameter θ_i. Dotted line shows where Δ E^ CB and Δ E^ W have their minimum value and the contact angle associated with it.

2012.11562

minimo.jpg

0.473722

0fdff5a1cff5459ca8bd902e5b529157

Results for pillared substrates. Interpillar volume of water, V^w_p, and oil, V^o_p, as a function of the pillar distance for (a) f_o = 0.10 and (b) f_o = 0.90. Blue color represent water and orange represent oil. ξ^ a_ por, ξ^ r_ por and ξ_ por as a function of interpillar distance for (c)f_o=0.10 and (d)f_o=0.90

2012.11562

pilares_h_5.jpg

0.442533

ae3b9ecee61341d0bdfc4dd607079ca0

Results for pillared substrates for several geometric parameters. Above: Interpillar volume of water, V^w_p, and oil, V^o_p, as a function of the pillar distance for (a) f_o = 0.10 and (b) f_o = 0.90. Below each figure: Cross sections of the final droplet configuration of Monte Carlo simulations for a=2 μm and a=14 μm for each corresponding f_o. To build this figure, a mixed droplet with oil fraction given by f_o (value specified above the figures) and total volume given by V^T=4/3π R_0^3 with R_0 = 50 μm is simulated on a substrate with =5 μm, =10 μm and varying pillar distance . Blue color represent water and orange represent oil.

2012.11562

pilares_novo.jpg

0.439789

be0f3b7428b5490cae40fe93d551ecdf

Results for porous substrates. Interporous volume and reservoir volume of water, V^w_p and V^w_r , and oil, V^o_p and V^o_r , as a function of the porous distance for (a) f_o = 0.10 and (b) f_o = 0.90. Blue color represent water and orange represent oil. ξ^ a_ por, ξ^ r_ por and ξ_ por as a function of interporous distance for (c)f_o=0.10 and (d)f_o=0.90

2012.11562

poros_h_5.jpg

0.46755

94cedbb28c5044f0a0bf4872fb98b6ff

Results for porous substrates for several geometric parameters. Above: Interporous volume of water V^w_p and oil V^o_p and reservoir volume of water V^w_r and oil V^o_r as a function of the pillar distance for (a) f_o = 0.10 and (b) f_o = 0.90. Below each figure: Cross sections of the final droplet configuration of Monte Carlo simulations for a=2 μm and a=14 μm for each corresponding f_o. To build this figure, a mixed droplet with oil fraction given by f_o (value specified above the figures) and volume given by V^T=4/3π R_0^3 with R_0 = 50 μm is simulated on a substrate with =5 μm, =10 μm and varying porous distance . Blue color represent water and orange represent oil.

2012.11562

poros_novo.jpg

0.502261

7ddb257e3eac45138b251d8f13de4c40

Definition of the geometric parameters of the substrates and the droplet. (a): pillared surface with width , pillar distance and height . (b): porous surface with width , porous distance and height . (c): geometric parameters of the droplet. We consider a spherical cap with radius R, base radius B, height H, and contact angle θ_C.

2012.11562

surface_mod.jpg

0.42684

fa8d81ff28a24c7b99f21a033998ad58

Measured positron yield as a function of average laser intensity over the spot size (1/e^2). Diamond markers indicate current experiments. Scaling model (line) and data from other facilities (dots) from Ref. <cit.> and Ref. <cit.>, respectively.

2012.11563

PositronScaling_v3-eps-converted-to.jpg

0.384601

0d33c03598554198aa362d32218b65a7

The for each individual experiment. The experiment ID is listed on the abscissa and are identified in Tables <ref>, <ref>, <ref> and <ref>. The original DIS data used in the nCTEQ15 fit are in blue, the DY data in green, and the new JLab data in yellow. We also list the total for each fit inset in the plot.

2012.11566

chi2-trials.jpg

0.468139

8e6decc6e5aa454a8b898acbec61c357

We display the ratio of the modified (smeared/rescaled) PDF compared to the unmodified PDF for both the Gaussian smearing and the x-rescaling. For the Gaussian smearing (blue), we display S⊗ f(x) over f(x) with Δ=0.05, and δ=-0.007 (which resembles an A=12 nucleus). For the x-rescaling (red), we display f(x'_A)/f(x) for ε=0.02, κ=10, and A=12. While there is a small relative difference at intermediate to small x values, the rise at large x is quite similar for both methods.

2012.11566

comparesmear.jpg

0.409669

6d8a31cbda184c71954ccf7d428cf647

We display the F_2^A/F_2^D ratio of selected data sets sorted by nuclei. The data from the original nCTEQ15 DIS points are in blue, and the DY in green. The new JLab DIS data are in yellow.

2012.11566

dataquality.jpg

0.536915

485697f0acaa497d8a86f2ad9501e692

We show F_2^A/F_2^D for selected data sets in the large x region. The different curves show the impact of the variations of ϵ={0, 0.01,0.03,0.1 } (solid blue, dashed red, dot-dashed green, dotted purple) with κ=10 in the rescaling of Eq. (<ref>). The combined χ^2 values are displayed in the legend, both with and without the W>1.7 GeV cut. For reference, the bottom curve (blue, solid) in each plot corresponds to ϵ=0 which is the result. Points which are excluded from the fit by the W>1.7 GeV cut are indicated with a hollow (gray) symbol; this is evident for the Fe and Cu data sets. For reference, experiment Fe/D is ID=5131, Al/D is 5134, Cu/D is 9984, and C/D is 9990.

2012.11566

epsscan.jpg

0.447282

c314bd35bce44f3bbff7cb91031fec8a

We display the ratio F_2^TMC(x,Q)/F_2^leadingTMC(x,Q) for Q={1,1.3,2,3} GeV (blue, yellow, green, red) for neutral current DIS; the lowest Q values yield the largest corrections. Each colored band contains results for {D, C, Al, Fe, Au, Pb} nuclei, where D is the lowest and Pb is the highest within each band.

2012.11566

figure3acot.jpg

0.481095

66dd5d93ef16404aa58d8fc492d893eb

We display the F_2^A/F_2^D ·( F_2^D/F_2^p )_CJ ratio for selected data sets sorted by nuclei. We also overlay the theoretical prediction of in blue. The theory predictions have been calculated at averaged Q values where data sets overlap.

2012.11566

fitquality.jpg

0.391105

e4b0706e5b64448884035fe8baf637ff

We display the higher twist corrections as a function of x at Q=2 GeV for selected nuclei. The larger A values yield larger corrections since they scale as A^1/3. The upper frame shows the correction factor of Eq. (<ref>), and the lower frame shows the correction scaled to the proton.

2012.11566

htcorrections.jpg

0.39209

114a378a2f6642189d86500e3a328d9a

We compare our results to other nPDF sets from the literature including EPPS16 <cit.>, nNNPDF2.0 <cit.>, and TUJU19 <cit.>. We plot xf(x,Q) for carbon ^12C at Q=2 GeV on a log scale.

2012.11566

pdfabs_others.jpg

0.507594

a49f34bdd1c24d02ac2672ebff6d075a

Iron (^56Fe) PDFs ratio compared to nCTEQ15 at Q=2 GeV. We show the uncertainty bands for nCTEQ15 (blue) and (yellow) computed with the Hessian method.

2012.11566

pdfratios_iters_loglin_Fe.jpg

0.52704

b4f4d110c2ce44bfb19c222da9568e3a

Lead (^208Pb) PDFs ratios compared to nCTEQ15 at Q=2 GeV. We show the uncertainty bands for nCTEQ15 (blue) and (yellow) computed with the Hessian method.

2012.11566

pdfratios_iters_loglin_Pb.jpg

0.468019

345845e8547e42f4bfc4b77e11e662e6

We compare our results to other nPDF sets from the literature including EPPS16 <cit.>, nNNPDF2.0 <cit.>, and TUJU19 <cit.>. We plot the nPDF ratio for carbon ^12C at Q=2 GeV compared to on a log-linear scale.

2012.11566

pdfratios_others_loglin.jpg

0.417894

4a567833d586499cb32eeb3dde61c697

We display DIS and DY data entering our analysis in the {x,Q^2} space indicating the relevant kinematic cuts, where x and Q^2 are the usual DIS variables, and Q^2 for DY is the di-lepton mass squared. The more restrictive cuts of Q=2 GeV and W=3.5 GeV (black dashed line) are the cuts used in the original nCTEQ15 analysis. In the present work, we will relax the cuts to Q=1.3 GeV and W=1.7 GeV (red dashed line). This greatly expands the kinematic reach in the high-x region where much of the new JLab data is located.

2012.11566

q2x-space.jpg

0.400825

0509341223fd43ea958e50f4641a066d

We show the impact of the x-rescaling of Eq. (<ref>) for varying choices of ε and κ for a “toy” PDF, f(x)∼(1-x)^3/x^1.5. The plot shows the ratio f(x'_A)/f(x) for 2 groups of 5 curves for A=208. The largest 5 curves (solid) are for ε=0.01 with exponent κ={2,4,6,8,10} indicated by the colors {black, red, blue, green, purple}, respectively. The lower 5 curves (dashed) are for ε=0.001 with the same range of κ values.

2012.11566

toysmear.jpg

0.468828

aa870af357904482b20cff272c0d6d84

We display the classic F_2^A/F_2^D ratio for carbon illustrating the nuclear correction factor across the various x regions. The black points indicate the data used in the original nCTEQ15 fit, and the red points with the solid squares represent the additional data from this original set which are now included due to the relaxed Q and W cuts. The red open squares are the new JLab DIS data included in this analysis, and the blue points are those JLab DIS data which are excluded by the current kinematic cuts.

2012.11566

typical-f2ratio.jpg

0.565362

1610bde78f4c4f35a2ebd98cfee685da

Schematic representation of a Schwarzschild black hole and the parameters in scenario 2. The dashed circle is constant r and t hypersurface P. Distance ℓ_0 is from the point where the NEC starts being violated, and distance ℓ is from P to the singularity (pictured here at r=0).

2012.11569

blackhole.jpg

0.537658

ca02a29ee9cb43839fb46396fbbd9780

Schematic representation of a Schwarzschild black hole and the parameters in scenario 1. The dashed circle is constant r and t hypersurface P. Distance ℓ_0 is measured from the hypersurface P until the NEC starts being violated. Distance ℓ is from P to the singularity (pictured here at r=0).

2012.11569

blackhole2.jpg

0.520401

ed4da046432344aeaeeebd4d588e95e6

Qualitative plot of the right hand side of eq.(<ref>) and asymptotic behavior for large and small σ.

2012.11569

p1.jpg

0.578971

76166fc939844069949387c3e93d40f7

Required value of x to have a singularity for different values y. For Q_1=0.1 the minimum value is much smaller compared to Q_1=1.

2012.11569

scen2.jpg

0.478749

bf91e1da042c4a6590f30a9c2b00385a

Phase diagram of the total thermo-osmotic response coefficient M_, separating a thermophobic (M_ >0) or a thermophilic (M_<0) response, as a function of the Debye length and surface charge density Σ<0, for different wettings. One can observe a bigger effect of wetting for NaCl, while the change of sign is observed for KCl at higher values of |Σ|.

2012.11571

PAPERphasediag_Mtot.jpg

0.466176

5f895cb73706427a9bea3484ec57e883

(a) Effective slip length b^eff = b-z_s, with b the slip length and z_s the shear plane position. One can distinguish a slip situation when b^eff>0 and a stagnant layer situation when b^eff<0. (b) Modelled system for the measures of water enthalpy excess density and b^eff, constituted by water enclosed between two generic planar parallel walls. (c) Water enthalpy excess density δ h^wat profiles as a function of the wetting, controlled by the interaction energy between the liquid and the solid atoms ε_LS.

2012.11571

PAPERsystem_dh_beff.jpg

0.454486

b47f75bc7c3143b59e866246944c2321

In continuous line the total thermo-osmotic response coefficient |M_^tot|, and in discontinuous lines its electrostatic |M_^el|, solvation |M_^sol| and water |M_^wat| contributions, for θ∼ 120^∘ (corresponding to b^eff=1.20nm) and a surface charge density Σ = -0.01 C/m^2. The solvation contribution dominates for a wide range of Debye lengths , but a competition between the water and the solvation contribution appears for KCl at small . One can also note for KCl a change of sign in M_^tot as a consequence of the change of sign of the solvation contribution, with M_>0 for small and M_<0 for large . For NaCl, M_<0 in the whole range of .

2012.11571

PAPERtotalMto_comp_sC0.01.jpg

0.400935

716601f3359c4104bc858171af0654ce

Picture of a part of an antibiogram showing an antibiotic disk and its inhibition zone (left). On the right a plot of the intensity radial profile and the segmentation with slopeOP with isotonic constraint.

2012.11573

AST_radial_profile.jpg

0.530637

36818f3dcb8c44c28cfc1545ca951817

Results for scenarios 1 to 4 with noise σ = 12. Each point of the MSE curve is the mean over 30 independent simulations. For the smallest MSE in each of the 12 MSE curves, we present the 30 corresponding results in the boxplots. The last column shows the obtained segments for the best of the 30 results in boxplots for OP-2D and slopeOP.

2012.11573

beta-mse-boxplots_sigma12.jpg

0.497842

98d94c9ce9634e1f85ae14d55b8ce425

Results for scenarios 1 to 4 with noise σ = 24

2012.11573

beta-mse-boxplots_sigma24.jpg