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
0.421132	7c10110a8c9b49728783c91f8c101522	Qubit trajectories in the Bloch sphere with the Zero-Noise Extrapolation (ZNE) technique. The accuracy of a noisy quantum computer can be improved with the ZNE error mitigation method. a) Here, one repeats a given calculation with different levels of noise. The green curve corresponds to a rotation in the Bloch sphere with a higher noise level than that leading to the red curve. b) Taking data from the red and green curves, ZNE can be used to estimate what the trajectory (blue) would be like in the absence of noise. Figure adapted from <cit.>.	2012.09265	Fig6.jpg
0.404493	5461a8aec640422a876abb8ce0d61d53	Collective enhancement of scattering. Steady-state distributions of mean concentration n(x_i,y_i) [panels (a) to (e)] illustrate enhanced scattering (blockade region growth) for heterogeneously fractured obstacle. Coordinates are in units of ℓ (lattice constant). A close view of impurity cluster inner structure for each case is shown at the top: (a) solid obstacle, (b) and (c) random clusters, (d) regular cluster, (e) uniform cluster. R=10.8ℓ for (a), R=20ℓ for (b), and R=40ℓ for (c)–(e). Number of constituent single-site impurities is N=362, n_0=0.37, \|\|=0.5 (stream is directed along the x-axis) for all calculated distributions. Plots (f) and (g) show dependencies of total drag force f≡\|\| (units of kT/ℓ) and √(ε) on cluster density ϕ; R∈(∞ 10.8ℓ], N=362, n_0=0.2.	2012.09266	fig1.jpg
0.430828	c0e866c160a84be3ba7a599755671e5e	Strong local fluctuations of scattered field δ n(x_i,y_i) inside a random impurity cluster. ϕ=0.0569 (N=362, R=45ℓ), n_0=0.2, \|\|=0.5.	2012.09266	fig2.jpg
0.348761	a7af51e09f1b44adbfd2fbc65f527e02	Spatial distribution of force magnitude \|f_x(x_i,y_i)\| calculated numerically for disordered impurity cluster shown on Fig. <ref>(c). The “lattice version” of the expression for the force has been used, following the approach from <cit.>.	2012.09266	fig3.jpg
0.459176	e526360bca334c0eb99af5712fbe7fb1	Probability density fitting for the force component f_x distribution. Kernel density estimator developed in <cit.> has been exploited. The inset shows the spatial distribution of impurities within the considered cluster.	2012.09266	fig4.jpg
0.408026	ddb6f4b1760c4a83883877e3ae715ddd	tr-ARPES snapshots of WS_2/graphene heterostructure before and after optical excitation: (a) Photocurrent measured along the ΓK-direction at negative pump-probe delay. Colored dashed lines indicate the position of the line profiles used to determine the transient band positions in Fig. <ref>c and e. The yellow box indicates the region where the Fermi-Dirac fits for Fig. <ref>c were performed. (b)-(d) Photocurrent for different pump-probe time delays after photoexcitation at ħω_pump=2 eV with a pump fluence of 2.85 mJ/cm^2. (e)-(h) Pump-induced changes of the photocurrent for different pump-probe delays. Red and blue indicate gain and loss of photoelectrons, respectively, with respect to the photocurrent measured at negative pump-probe delay. The colored boxes in (f) indicate the area of integration for the pump-probe traces shown in Fig. <ref>a and b. Thin white (a-d) and black (e-h) dashed lines represent the calculated band structures of graphene <cit.> and WS_2 <cit.>. A rigid band shift of -0.81 eV and -1.19 eV was applied to the WS_2 CB and VB, respectively. The graphene Dirac cone was shifted by +0.3 eV to account for the observed hole doping <cit.>. Thick white (a-d) and black (e-h) dashed lines mark the position of the Fermi level.	2012.09268	Figure1.jpg
0.434983	0a25de5f9e634e41be0692005d64d720	smoking gun evidence for ultrafast charge separation: (a) Gain in the WS_2 CB (orange) and loss in the WS_2 VB (green) as a function of pump-probe delay. (b) Gain above the Fermi level (red) and loss below the Fermi level (blue) inside the graphene Dirac cone as a function of pump-probe delay. (c) Position of the WS_2 CB (orange) and the upper WS_2 VB (green) as a function of pump-probe delay. The transient band gap of WS_2 (turquois) was obtained by subtracting the transient position of the VB from the transient position of the CB. (d) Charging-induced shift of WS_2 valence and conduction band obtained by adding (subtracting) \|Δ E_gap\|/2 to (from) the VB (CB) position in panel a. (e) Transient shift of the graphene Dirac cone. Black lines are single exponential fits to the data. Gray lines in panel c were calculated from the exponential fit for E_gap in panel c and the exponential fit in panel d. All data was obtained with a pump fluence of 2.85 mJ/cm^2.	2012.09268	Figure2.jpg
0.367992	965db54ced2c493caaf24bc44bbdab84	fluence dependence of ultrafast charge separation and recombination: (a) Lifetime of graphene gain from Fig. <ref>b as a function of fluence. The black dashed line represents our temporal resolution of 200 fs. (b) Lifetime of photoexcited electrons in the WS_2 CB from Fig. <ref>a as a function of fluence. (c) Lifetime of gain above the equilibrium position of the upper WS_2 VB obtained by integrating the photocurrent over the area marked by the pink box in Fig. <ref>f and by fitting the resulting population dynamics with a single exponential decay. Thick lines in a and b were obtained by inverting and rescaling the guides to the eye from Fig. <ref>e to match the data points. The thick line in c is a guide to the eye.	2012.09268	Figure3.jpg
0.438959	b1ff0a9a05674434b9a307a5364e615c	direct tunneling model: (a) Sketch of the band structure with tunneling barrier for holes (Δ E_h, red) and electrons (Δ E_e, orange). (b) Temporal evolution of the two barrier heights deduced from Fig. <ref> together with single-exponential fit. (c) Temporal evolution of the electronic temperature inside the Dirac cone together with double-exponential fit. (d) Temporal evolution of the electron and hole transfer rates deduced from panels b and c together with double-exponential fit. (e) Pump fluence dependence of the peak transfer rates for electrons and holes. Thick gray lines are guides to the eye.	2012.09268	Figure4.jpg
0.43077	0051014fded0434faf66ea03988fe88a	theory and final microscopic model: Momentum dependence of the tunneling matrix element calculated from tight-binding for CB (a) and VB (b). k_x=k_y=0 corresponds to the K-point of graphene. Calculated transfer time for electrons (c) and holes (d). (e) Final microscopic model including direct (yellow and red arrows) as well as defect-assisted tunneling (green arrows).	2012.09268	Figure5.jpg
0.449003	3e2bbad3190d488d933b0c14a4aaa3a3	Sample characterization: (a) Secondary electron microscopy image. (b) Raman spectrum. (c) Photoluminescence spectrum.	2012.09268	SFigure1.jpg
0.440464	ff3d26cb18a04dcabf465457226fb99d	Direct comparison of fluence dependence of gain above equilibrium position of upper WS_2 VB (purple) and WS_2 charging shift (green).	2012.09268	SFigure10.jpg
0.443261	20994388cf8a4160972a3d8c438fc0b8	Evidence for in-gap defect states. EDCs through the K-point of WS_2 in the conduction band region at negative delay and at the peak of the pump-probe signal at t=0.2 ps. Gray lines indicate the Gaussian peaks used to fit the spectrum at t=0.2 ps. The two Gaussians marked with black arrows are attributed to in-gap defect states due to Selenium vacancies.	2012.09268	SFigure11.jpg
0.420668	f8c81de3befe4f1daac5cd1f4ac53296	Space charge effects: (a) Position of the Fermi level at negative delays as a function of pump fluence. The gray line is used to extrapolate the position of the Fermi level to zero pump fluence. The data point in brackets was obtained on an area of the sample where the WS_2 coverage was particularly low. The error bars are smaller than the symbol size. (b) Energy resolution for different pump fluences.	2012.09268	SFigure2.jpg
0.439864	f5959c838ec7495ea8bd442defb39cda	Treatment of raw data: (a) Tr-ARPES snapshot at negative pump-probe delay taken with a pump fluence of 1.63 mJ/cm^2. The pink box is used to extract I_box(t) in panel (b). The orange box captures the carriers at the bottom of the conduction band shown in panel (d). (b) Photocurrent I_box(t) integrated over the area marked by the pink box in (a) together with its average for negative pump-probe delays <I_box>. (c) Normalization factor f=<I_box>/I_box(t) as a function of pump-probe delay. (d) Effect of the raw data correction on the time dependence of the photocurrent integrated over the area marked by the orange box in panel (a). (e) Ratio of the reference intensities I_ref,gr and I_ref,WS2 as defined in Fig. <ref>c and a, respectively, for the datasets obtained with different pump fluences.	2012.09268	SFigure3.jpg
0.452665	94e0997a4f9241bebcd532df33aff194	Determination of transient band positions: (a) EDCs extracted along the dashed green line in Fig. 1a at t<0 fs (light green) and at t=200 fs (dark green) together with Gaussian fit (black) in the WS_2 VB region. The shaded areas indicate the three individual Gaussians used to fit the data. The area of the dark-shaded Gaussian I_ref,WS2 was used to turn the measured photocurrent independent of the WS_2 coverage. (b) EDCs extracted along the dashed green line in Fig. 1a from the differential tr-ARPES data shown in the lower panel of Fig. 1 in the WS_2 CB region for t<0 fs (dark orange) and t=200 fs (light orange) together with Gaussian fit (black). (c) MDCs extracted along the dashed red line in Fig. 1a for t<0 fs (dark red) and t=200 fs (light red) together with Lorentzian fit. The area of the dark-shaded Lorentzian I_ref,gr was used to turn the measured photocurrent independent of the WS_2 coverage. The data was obtained for a pump fluence of 2.85 mJ/cm^2.	2012.09268	SFigure4.jpg
0.482561	c3a0ecdfc79946c48793acb0f239a007	Fermi-Dirac fits: EDCs obtained by integrating the counts over the momentum width of the yellow box in Fig. 1a at t<0 fs (color) and t=200 fs (color) together with Fermi-Dirac fits (black).	2012.09268	SFigure5.jpg
0.462177	7dddc0b6343845d9933cc389dd418f63	Fluence dependence of the population dynamics: (a) Gain in the CB (orange) and loss in the VB (green) of WS_2. (b) Graphene dynamics: Gain above the Fermi level (red) and loss below the Fermi level (blue). (c) Gain above the equilibrium position of the upper WS_2 VB. Thin black lines are single exponential fits to the data.	2012.09268	SFigure6.jpg
0.492969	11f796503f0541129a6f3f367fdc027c	Fluence dependence of the transient band positions: (a) Position of the WS_2 CB. (b) Position of the WS_2 VB. (c) Transient band gap of WS_2. (d) Shift of the WS_2 CB due to charging of the WS_2 layer. (e) Shift of the upper WS_2 VB due to charging of the WS_2 layer. (f) Shift of the Dirac cone.	2012.09268	SFigure7.jpg
0.437439	29c232b068d449288999f6a61ae04807	Fluence dependence of hot carrier dynamics in graphene: (a) Electronic temperature, (b) chemical potential referenced with respect to the vacuum level, (c) chemical potential referenced with respect to the Dirac point, and (d) number of holes transferred from WS_2 to graphene as a function of pump-probe delay for different pump fluences.	2012.09268	SFigure8.jpg
0.477864	6a0072e37f0c4e119dcb9161faae0d3f	Fluence dependence of different parameters at the peak of the pump-probe signal (t=0.2 ps): (a) Pump-induced change of photocurrent inside colored boxes from Fig. 1f. (b) Change of the WS_2 band gap from Fig. 2a. (c) Number of holes transferred into the graphene layer from Fig. <ref>c. (d) Charging-induced WS_2 and graphene band shifts from Figs. Fig. 2e and f. (e) Peak electronic temperature of Dirac carriers from Fig. <ref>a. Light gray lines are guides to the eye. The datapoints in brackets were obtained on areas of the sample with lower WS_2 coverage.	2012.09268	SFigure9.jpg
0.45671	c84ebb3bdad340979b2dedb4a8cf30b6	Comparison of rankings by CoachScore while using different damping factors. (a) We report the rank position of all soccer coaches of our combined dataset. The ranking on the x-axis is for CoachScore with damping factor α=0.85, and the ranking on the y-axis with α=0.95. Spearman correlation coefficient is ρ=0.999, while Kendall correlation coefficient is τ=0.972. (b) Same as in panel a, but basketball coaches in NBA. Spearman correlation coefficient is ρ=0.996, while Kendall correlation coefficient is τ=0.954.	2012.09270	damping_factor.jpg
0.414143	4929304f403348e68e31d2360027a3c5	Statistics of the soccer dataset. (a) Total number of matches in our dataset per season. We combined together games from all the national leagues and European cups. (b) Number of coaches in our dataset per season. Each season includes every coach that has managed at least one game in that season. Data combine all games of the national leagues and European cups. (c) Complementary cumulative distribution functions (ccdf) of the number of matches managed, won, lost, and tied by coaches in our dataset. The stretched exponential ccdfs that best fit the empirical data are displayed as dashed curves. The stretched exponential ccdf is defined as f(x)=exp(-x^β). The parameter of the best fits are β=0.695 for total matches, β=0.592 for wins, β=0.791 for losses, and β=0.676 for ties. (d) Ccdf for the number of head-to-head matches played between pairs of coaches. The best fit curve to the empirical data is also shown. In this case, we used the lognormal ccdf g(x)=1 - 1/2 erf( log (x) - μ/√(2) σ). The parameters for the best fit are μ=0.452 and σ=0.945.	2012.09270	fig1.jpg
0.429798	ad02f80db17c4e19b62707c1454bfe40	Comparing career performance of basketball coaches. We display career trajectories for five selected coaches: Pat Riley, Phil Jackson, Gregg Popovich, Rick Carlisle, and Erik Spoelstra.	2012.09270	fig10.jpg
0.488916	f4cf97698ed8415e930623592769d005	Statistics of the basketball dataset. (a) Total number of matches per season. (b) Number of coaches per season. Each season includes all coaches that have managed at least one game in that season. (c) Complementary cumulative distribution functions (ccdf) of the number of matches managed, won, and lost by coaches in our dataset. The stretched exponential ccdf is defined as f(x)=exp(-x^β). The parameter of the best fits are β=0.746 for total matches, β=0.664 for wins, and β=0.821 for losses. (d) Ccdf for the number of head-to-head matches played between pairs of coaches. The best fit curve to the empirical data is also shown. In this case, we used the lognormal ccdf g(x)=1 - 1/2 erf( log (x) - μ/√(2) σ). The parameters for the best fit are μ=1.567 and σ=1.098.	2012.09270	fig2.jpg
0.46285	1a64c6149406447095b81a76fa7dbf65	Comparison of rankings by CoachScore and number of wins. (a) We report the rank position of all soccer coaches in our combined dataset (see Table <ref>). Rankings are performed by either relying on the CoachScore values or the number of wins. Each point in the plot is a coach. Spearman correlation coefficient is ρ=0.989, while Kendall correlation coefficient is τ=0.923. (b) Similar to panel a, but for basketball coaches. Rank correlation coefficients are ρ=0.979 and τ=0.888.	2012.09270	fig4.jpg
0.407662	3ed82277240c45529c88cc75e712c54f	CoachScore vs. number of points. We report the rank positions for the 20 soccer coaches who managed teams in the 2018/19 English Premier League. We rank coaches on the basis of their CoachScore and compare those ranks with those obtained with the number of points they gathered in the league. Spearman rank correlation coefficient is ρ=0.973, and Kendall rank correlation coefficient is τ=0.896. The line stands for perfect agreement between the two rankings.	2012.09270	fig5.jpg
0.489632	62b778457a1e47bb9081920b1b9f4e22	Monitoring the career performance of soccer coaches. (a) We visualize the dynamic rank of Pep Guardiola throughout his career. Rank positions are updated after each game day. Red lines indicate when a coach takes over a new team, dashed green lines indicate the start of a new season with his current team, and grey shaded areas represent periods of time when the coach is not managing any team. FCB stands for FC Barcelona, BM for FC Bayern Munich, and MC for Manchester City FC. (b) Same as in panel a, but for JosÃ© Mourinho. CHE stands for Chelsea FC, INT for FC Internazionale Milano, RM for Real Madrid CF, MU for Manchester United FC, and TOT for Tottenham Hotspur FC.	2012.09270	fig7.jpg
0.495882	f3756f0c05494ea085a5f236b7c3c739	Monitoring the career performance of basketball coaches. (a) We visualize the dynamic rank of Phil Jackson throughout his career. Rank positions are updated after each game day. Red lines indicate when a coach takes over a new team, dashed green lines indicate the start of a new season with his current team, and grey shaded areas represent periods of time when the coach is not managing any team. CHI stands for Chicago Bulls, and LAL for Los Angeles Lakers. (b) Same as in panel a, but for Steve Kerr. GSW stands for Golden State Warriors.	2012.09270	fig8.jpg
0.415024	db764d74cd914c6f97584629a13561ee	Comparing career performance of soccer coaches. We display career trajectories for five selected coaches: Jupp Heynckes, Sir Alex Ferguson, JosÃ© Mourinho, JÃ¼rgen Klopp, and Pep Guardiola.	2012.09270	fig9.jpg
0.512086	d324045861ef44b0a95860792e79e519	Network of contact among top coaches. (a) We display the network of contact among soccer coaches that have won the UEFA Champions League at least twice after 1980. The network is built setting β =0 in Eq. (<ref>) and including all games in our dataset. In the visualization, the size of the nodes is proportional to their in-strength, and the width of the lines connecting pairs of nodes is proportional to the weight of the corresponding edge. (b) Same as in panel a, but for top basketball coaches. We visualize the network of contacts among coaches that have won at least 2 NBA titles.	2012.09270	nws.jpg
0.494204	633260fc93b14853b2588340ba5a6fc4	Comparison of rankings by CoachScore while using different weights for win in soccer. We report the rank position of all soccer coaches of our combined dataset. The ranking on the x-axis relies on the network where a win has twice the weight of a tie. The ranking on the y-axis relies on a network where a win has three times the weight of a tie. Spearman correlation coefficient is ρ=0.999, while Kendall correlation coefficient is τ=0.983.	2012.09270	win2_vs_win3.jpg
0.46667	5d7d6b3c2449473985b6c39c9e1f6795	Picture of a regular {5,5} tessellation in in the Poincaré disc model. All line segments are geodesics with respect to the hyperbolic geometry.	2012.09271	55PD.jpg
0.405759	5a2e3ebfc7d24513a512dd5b320b9b42	The second largest eigenvalues of hyperbolic surfaces of small size labeled by their Schläfli symbol {r,s}. The dashed lines indicate the asymptotic lower bound 2√(s-1) of <Ref>.	2012.09271	hyperbolic_spectrum.jpg
0.466659	642f35d95d0e46d2977854f43b7702a0	A hyperbolic genus 3 surface supporting the {3,7} tessellation. It is called the Klein quartic. The dotted line indicates the translation τ = (ρ σ^-2)^4 under which points are being identified. The symmetry group of the surface is Isom(ℍ^2/N_τ)=PSL(2,7)⋊ℤ_2.	2012.09271	klein_quartic.jpg
0.487849	bd691c91b1b74e8686dc2ff365b2dd86	Applying DBSCAN to the t-SNE output, noise shown in black (Automobile).	2012.09272	clustering.jpg
0.493152	21cde0c9bc98481ca5dc521a40df515f	Large batch training accuracy for Full CIFAR-10, Randomly filtered and Network filtered data (Avg. of 4 runs). Network filtered data gives at least the same accuracy as full data and outperforms randomly filtered data.	2012.09272	comparison.jpg
0.446409	a3f509a33e5b44ae90ed36033cce711a	Large batch size vs. Accuracy for different lr. Accuracy increases until it reaches an optimum batch size and then starts to gradually decrease.	2012.09272	lbtr.jpg
0.460934	e7404a452998484a8c8808838a3df433	Learning rate vs. Preferred batch size for single GPU	2012.09272	optimum_batch_size.jpg
0.402811	fa4a7173166346588b666700230e45e2	Steps showing the process of applying data optimized strategy by (1) fine tuning a pretrained network, (2) Extracting and reducing dimensionality by PCA / t-SNE, (3) Clustering using DBSCAN to remove noise and (4) Training another network faster and at improved accuracy.	2012.09272	process.jpg
0.428562	c919c951c71f4283b53dab522e32521f	Small batch size vs. Accuracy for network filtered, randomly filtered and full CIFAR-10 data sets.	2012.09272	small_batch_single_GPU.jpg
0.424957	ea4c6a837fa142379284c33156d545f8	Visualizing output of extracted features from the fine-tuned ResNet-152 using t-SNE.	2012.09272	t-sne.jpg
0.436008	ffac5929004641e687b1f15f6fed7612	Schematic view of a possible TIARA design: the pixels are placed around the target at fixed positions. Pixels coordinates can be established patient-wise at treatment planning in order to clear one or more entrance windows for the proton beam (red line). TIARA will be read in time coincidence with a beam monitor (light blue).	2012.09275	figure_1.jpg
0.394033	d3d79ad4b97e4c4a9dad85cb12c468ad	Left: correlation between the detectors y_COG calculated under treatment conditions and the actual beam displacement imposed in the MC simulation. Right: correlation between the calculated x_COG and the actual cavity thickness considered in the MC simulation. Both datasets correspond to a monitoring statistics of 10^8 incident protons. The 1σ (orange) and 2σ (blue) errors are displayed in both graphs. A linear fit was carried out for each dataset.	2012.09275	figure_10.jpg
0.436175	690990a9d28b488cb1cf660184bf5d60	Left: geometry simulated with Geant4 to validate the analytical reconstruction algorithm. The impinging 100 MeV proton beam is represented in blue. In red, a PG created along the beam path and recorded in the phase space is also shown. A diamond based beam monitor (represented in black) placed upstream of the target was considered. Right: the same geometry is displayed, with the insertion of an air cavity of variable size and fixed position at (-5.5,0,0).	2012.09275	figure_2.jpg
0.381793	d4685c24e954444faaf1ec10cf8bd3e5	Energy (left) and TOF (right) spectra of each particle type scored in the phase space. "Others" emissions mainly include 511 keV gamma-rays, but also PGs from secondary neutrons and protons.	2012.09275	figure_3.jpg
0.478887	1b6a05c7ed8842df8a2e1b1be97ac498	Reconstructed PG profiles for the 1 cm thick air cavity geometry obtained with different monitoring statistics and at TPS statistic (1.5×10^9 incident protons). The Ecut was applied to all datasets.	2012.09275	figure_5.jpg
0.406984	9652532097974ebeb176e3e873ca03cc	The reconstructed PG vertex profile obtained at TPS statistics and shown in figure <ref> after the Ecut was applied, is decomposed to show the separate contributions of the different secondary particle types detected in the phase space. Slower particles as neutrons (red curve) are naturally time-discriminated from the other particles.	2012.09275	figure_6.jpg
0.416034	36886ed2861c4564be817c1b11fc117a	Comparison of PG vertex profiles reconstructed in TPS (blue curve) and treatment (orange curve) conditions in the case of a 1 cm thick air cavity. In both graphs, a high statistics of 1.5 × 10^9 incident protons was employed for the TPS profile. On the left, the treatment profile is reconstructed assuming a 100 ps (rms) CTR and a statistics of 10^8 incident protons, corresponding to ≈ 3×10^4 PG detected, whereas on the right a CTR of 1 ns (rms) was considered, with a statistics of 10^9 incident protons, corresponding to ≈ 3×10^5 PG detected. The Ecut (E > 2 MeV) was applied to data in both the left and right graphs.	2012.09275	figure_7.jpg
0.484782	2c37d0b77ba14a0f9c8bab60caf078c7	Comparison of PG vertex profiles (left) and integrated PG vertex profile (right) obtained for the reference simulation in TPS conditions (1 cm air cavity) (blue) and for a treatment geometry (orange) presenting a 1.5 cm thick air cavity. In both plots, the vertical dotted line (in red) locates the TPS PG profile peak at x_TPS. In the right plot, the horizontal dashed line (in green) in the inset represents the calculated profile distance d_5.	2012.09275	figure_8.jpg
0.451755	7e8d1b279c634cacbe119f29cb85c04b	Correlation between the actual air cavity thickness and the corresponding measured profile distance d_i in the case of a 100 ps (rms) CTR and 10^8 incident protons (left), and in the case of a 1 ns (rms) CTR and a statistics of 10^9 incident protons (right). The 1σ (orange) and 2σ (blue) errors are displayed for each data point. A linear fit was carried out for each dataset.	2012.09275	figure_9.jpg
0.421111	8eefbe95a9794696b652d69d8abbcc67	Cars3D	2012.09276	cars3d_kendall.jpg
0.417795	84a141c2af8c45738290b42bd259a76c	Mean metric scores as the representation becomes less modular and less compact.	2012.09276	compactness_modularity.jpg
0.424028	5bd454e5e5d54aa3b6af3bdc09441ee3	Metric scores for perfectly disentangled representations under increasing noise level (α).	2012.09276	noise.jpg
0.395223	330924e4a87943a4b476099973627f8d	Metric scores for perfectly disentangled representations with increasingly nonlinear factor-code relation (α).	2012.09276	non_linearity.jpg
0.419194	5c1dc207e89041b1bb3963a9b76e74c5	Scores for perfectly disentangled representations. The abscissa indicates how many of the 8 factors are measured by the metrics.	2012.09276	partial_factors.jpg
0.410188	a51f6c35428a44d1b13d5aec7a208461	SmallNORB	2012.09276	smallnorb_kendall.jpg
0.514931	7154b85e78e44633b2c8b0c35842ae9a	Shape of the factor-code relation as parameter α increases.	2012.09276	tan_curve.jpg
0.412327	4103573f4cc74f16a84ced696451b906	Effect of nonlinearity on discretization bins population.	2012.09276	tan_hist.jpg
0.436849	b639f9381d184792b7299f05b0104757	Semi-analytical Slonczewski field H_S (a) and velocity v_S (b) compared to the limiting analytical expressions Eq. (<ref>) resp. (<ref>). The ratio of the two expressions is plotted versus δ=H_D/H_DW for different values of the damping parameter α. The curves for α=0 are analytic (see text).	2012.09277	Figure3-512.jpg
0.404942	173f2648f32848f5985c919bd937e19b	Attention vector learnt by the proposed method. Y-axis is the slice number of the CT image, and the X-axis is the slice number (depth) of the 3D feature map F_i^2D.	2012.09279	attention.jpg
0.370367	d24d9e5f5d264f56a760c454977cc06a	A CT image from ABD-110 dataset. The first and second columns are the ground truth and predicted mask overlayed on the original CT image slice, respectively. The third column shows the comparison of contours of the ground truth and predicted mask on the same slice.	2012.09279	visualization_WACV.jpg
0.451998	c083bcdcca4149579bf9f58f1a74a266	Sensitivity of a ^76Ge double-β decay experiment searching for massive sterile neutrinos with a mass of 500 keV, given as a function of exposure and background level. The sensitivity is expressed in terms of the median 90% C.L. upper limit on the squared of the mixing angle (sin^2θ), computed assuming no signal. The markers indicate the value computed using a full frequentist analysis, while the lines show their best fit with the analytical expression discussed in the text. The plot shows the impact of additional backgrounds with a rate R_ other which is expressed w.r.t. the irreducible decay rate R_.	2012.09281	Sensitivity_vs_exposure_m500keV_Ge76.jpg
0.452741	a55b279888934e96af8f942f9aacb446	Comparison of the algorithm and QuTiP's over an evolution of 20 Rabi oscillations, with the same parameters as in Fig <ref> of the main text. Left panel: Frobenius norm distance between the algorithm with different subpixel numbers and QuTiP's with (circles) and (x's) methods. The squares are the norm distance between and with 10^4 subpixels. The dashed line indicates the error between with the and methods. Right panel: Computational time of and for results of the left panel. The dashed lines refer to the computational time for with and with the highly accurate settings described in the text.	2012.09282	Fig-2panels-DysonvsPropag-Dyson1to4-1drives-Hilbert25-density0.20-500ns.npz.jpg
0.492081	77c790b2c83443b5bd32af9bad7d26ae	benchmarks. First row: Frobenius norm distance from the propagator Û_ref(0,T) for the algorithm for T=500 ns. Second row: Contraction time of the algorithm as a function of the subpixel number N_s. The slope of the data is precisely 1, meaning that the computational time scales linearly with the number of subpixels. Preparation stage computation time: (35 ms, 1.9 s, 6.8 s) for (-2, 3, 4) and 1 input drive, (0.35 s, 13 s, 87 s) for 2 input drives and (0.78 s, 47 s, 408 s) for 3 input drives.	2012.09282	Fig-6panels-benchmarks.jpg
0.464242	59c661d50b864fb6968e7daff885947d	Infidelity of the CR gate as a function of the chosen gate time when the gate is operated at the qubit-qubit detuning Δ/2π = (ω_c - ω_t)/2π = 210 MHz. The red curve indicates the best obtainable infidelity from a constant amplitude pulse, with the orange curve reporting the infidelity up to arbitrary single qubit rotations on both qubits, applied before and after the cross-resonance gate channel. The blue curve reports the infidelity when optimized under the GRAPE algorithm until convergence.	2012.09282	Fig-GRAPE-vs-flatpulse-10subpix.jpg
0.469097	b475a16537554959b74d8a4fb4a0bf2f	Infidelity of the CR gate as a function of the detuning of the qubits when optimized under GRAPE for 5000 iterations. The control qubit frequency is fixed to ω_c/2π=5.10 GHz.	2012.09282	Fig2-Detunings-GRAPE-TransTrans-300ns.jpg
0.518366	d0ed5f5ccff942a39a14d350a529bd07	Linear interpolation of the pulse sequences (yellow) with two subpixels per pixel. The black bars indicate the original Gaussian filtering, with the blue line the `true' filtered pulse.	2012.09282	Interpolation-new.jpg
0.465413	31acc6275a5c49f8b11ea9e558d594ff	Tree diagram showing the different branches of the time ordered integral, with two example branches highlighted.	2012.09282	PathDiagramTeststars.jpg
0.464479	a4584907c6054809b595fc9c125070f1	Superconducting circuit for cross-resonance gates between transmon qubits. The leftmost transmon, of frequency ω_c/2π, plays the role of the control qubit and is strongly driven by the voltage source V_c at the frequency ω_t/2π of the target qubit (rightmost transmon). The target qubit is also driven by V_t using a relatively weak tone that serves to give additional control. φ̂_c and φ̂_t correspond to the phase degree of freedom associated with the control and target qubits, respectively.	2012.09282	cr.jpg
0.457594	c1c42b3bab3c43aa80c4158f96ea0c66	Cryptographic information flow between two agents and one SO.	2012.09285	Encryption_flow_2.jpg
0.427638	5a59626cb0ce4f2897c5dde3d4837ee9	Encryption error and optimality gap (precision level σ = 3).	2012.09285	congestion_convergence_errors.jpg
0.483331	f31008255dd44d5aadda51d4bb6218d4	Convergence of the primal and dual variables.	2012.09285	congestion_primal_dual.jpg
0.435422	28957206e29b431c8a62c40063f174aa	Encryption error and optimality gap (precision level σ = 3).	2012.09285	convergence_error.jpg
0.408228	e0ea6bc6b037472aa93705b3e195d769	Secure information exchange between agents and the SO.	2012.09285	information_exchange_all_rc.jpg
0.467081	d083dc35a4154398aeb6e178ceb041d7	Convergence of the primal and dual variables.	2012.09285	primal_dual_convergence.jpg
0.506112	cfd988da1da040048ba97b3bf77581a3	A transportation network with 9 links and 5 agents <cit.>.	2012.09285	transportation_network_f.jpg
0.458471	a83ea6eaf86f488eb6a14f8647c0e8cb	Model performance on CelebA during training.	2012.09290	ablation_chart.jpg
0.415399	42fd4cede0714b779ff23765542d8c18	In each panel, the second row shows the images from AE, and the third row shows the GAN revisions.	2012.09290	ae-to-gan-sample.jpg
0.453876	3efe23289b264d0ebb20aeb087f7318a	Qualitative comparison of the Auto-encoder without and with the proposed self-supervision objectives. In each panel, the first row are the referential style images, the first column is the input sketch, the second row are synthesis results from a plain AE trained with only reconstruction loss, the last row are the results from the proposed AE with self-supervision objectives: ℒ^s_tri, ℒ^c_tri, ℒ^s_cls, ℒ^c_cls.	2012.09290	ae_noss_compare_art.jpg
0.435088	63166d25d75d41869202458c84361a6d	Qualitative comparison of the Auto-encoder without and with the proposed self-supervision objectives. The image arrangement is the same as Figure <ref>	2012.09290	ae_noss_compare_face.jpg
0.484324	3ee2a700c98847e58e3a120c0c968b56	Uncurated synthesis results of our model on WikiArt. Note that the images are compressed due to file size limit.	2012.09290	art_quality_1_compressed.jpg
0.430207	f628ece3d8f84505929569ac974c22fb	Uncurated synthesis results of our model on WikiArt. Note that the images are compressed due to file size limit.	2012.09290	art_quality_2_compressed.jpg
0.487831	ed75bceae8764990903d57e974bb7901	Uncurated synthesis results of our model on CelebA. Note that the images are compressed due to file size limit.	2012.09290	face_quality_1_compressed.jpg
0.472949	d5e893750a9b4b1ea45d04d9e865fc3d	Uncurated synthesis results of our model on CelebA. Note that the images are compressed due to file size limit.	2012.09290	face_quality_2-compressed.jpg
0.457278	9811f5018a814a89996f0cb1695645bc	Exemplar-based sketch-to-image synthesis from our model on varied image domains in 1024^2 resolution.	2012.09290	main-samples_compressed.jpg
0.454973	e012c8c4e50a41548fb6b2f442e08244	Overview of the proposed model.	2012.09290	model_overview.jpg
0.474955	b68ed3baaf0a4cf1bf6da68a05809c33	Synthesises from TOM. TOM generalizes well across multiple image domains, from photo-realistic to artistic, and from human portrait to nature landscape.	2012.09290	skt_sample.jpg
0.530023	d5f533a8a03b4e199fc575d2e3a3b5f7	Example sketches we used to train TOM. The bottom panel are line-sketches from artists, and the top panel are sketches we randomly searched from the internet.	2012.09290	skt_target.jpg
0.431286	7e0a1b43b29a4db8bda0afa5feef3761	Synthesis by mixing multiple style images.	2012.09290	style-mix-example.jpg
0.409181	c697fbfeba8e4f299764376f99594328	Synthesis from out-domain style images.	2012.09290	style-outdomain-example.jpg
0.400487	02a22247b59545439d7ef5b79843da7b	Works as a feed-forward style-transfer model.	2012.09290	style-transfer-example.jpg
0.472118	6a793e7956364659a9319a046fa14184	Illustration of our TOM. Dashed arrows in red indicate the gradient flow to train the sketch generator.	2012.09290	tom_overview.jpg
0.420657	bfe55f1ad5c9400aac96ef8f8e22a251	Hubble diagram of quasars. Magenta dashed line is the best-fit flat ΛCDM model with = 0.670 from the QSO-z < 1.479 QSO data. Cyan points are the determined QSO distance moduli and uncertainties <cit.> and red points are the means and uncertainties of these distance moduli, in narrow redshift bins which roughly represent a fifth order cosmographic fit of the whole QSO data <cit.>. The black solid line shows a flat ΛCDM model with = 0.30.	2012.09291	QSO_hubble_2421.jpg
0.437314	bf2a041eddc74eb8b66f26d2022cf7c2	Redshift distribution of the 2038 QSO that we use in our analyses.	2012.09291	Redshift_distribution_2400.jpg
0.466448	e29f1d2560584b1eb4a28155c8c34a9e	AppIS: runtime behavior.	2012.09292	AppIS-scheme-runtime.jpg
0.444738	4025f86a031841dbab58edc2f7a0c74f	The AppIS protection scheme.	2012.09292	AppIS-scheme.jpg
0.436712	495b598594014c0da58cb5fd41a279cc	The BombDroid protection scheme.	2012.09292	BombDroid-scheme.jpg
0.435353	8f399f8159084d518f5c4b9f79cee4c8	Runtime behavior of a nested logic bomb.	2012.09292	LB_runtime_example.jpg
0.477864	e3b00d546459444eadce0b86cdee960d	Transformation process of a qualified condition into a logic bomb.	2012.09292	QC_to_LB.jpg

0.421132

7c10110a8c9b49728783c91f8c101522

Qubit trajectories in the Bloch sphere with the Zero-Noise Extrapolation (ZNE) technique. The accuracy of a noisy quantum computer can be improved with the ZNE error mitigation method. a) Here, one repeats a given calculation with different levels of noise. The green curve corresponds to a rotation in the Bloch sphere with a higher noise level than that leading to the red curve. b) Taking data from the red and green curves, ZNE can be used to estimate what the trajectory (blue) would be like in the absence of noise. Figure adapted from <cit.>.

2012.09265

Fig6.jpg

0.404493

5461a8aec640422a876abb8ce0d61d53

Collective enhancement of scattering. Steady-state distributions of mean concentration n(x_i,y_i) [panels (a) to (e)] illustrate enhanced scattering (blockade region growth) for heterogeneously fractured obstacle. Coordinates are in units of ℓ (lattice constant). A close view of impurity cluster inner structure for each case is shown at the top: (a) solid obstacle, (b) and (c) random clusters, (d) regular cluster, (e) uniform cluster. R=10.8ℓ for (a), R=20ℓ for (b), and R=40ℓ for (c)–(e). Number of constituent single-site impurities is N=362, n_0=0.37, ||=0.5 (stream is directed along the x-axis) for all calculated distributions. Plots (f) and (g) show dependencies of total drag force f≡|| (units of kT/ℓ) and √(ε) on cluster density ϕ; R∈(∞ 10.8ℓ], N=362, n_0=0.2.

2012.09266

fig1.jpg

0.430828

c0e866c160a84be3ba7a599755671e5e

Strong local fluctuations of scattered field δ n(x_i,y_i) inside a random impurity cluster. ϕ=0.0569 (N=362, R=45ℓ), n_0=0.2, ||=0.5.

2012.09266

fig2.jpg

0.348761

a7af51e09f1b44adbfd2fbc65f527e02

Spatial distribution of force magnitude |f_x(x_i,y_i)| calculated numerically for disordered impurity cluster shown on Fig. <ref>(c). The “lattice version” of the expression for the force has been used, following the approach from <cit.>.

2012.09266

fig3.jpg

0.459176

e526360bca334c0eb99af5712fbe7fb1

Probability density fitting for the force component f_x distribution. Kernel density estimator developed in <cit.> has been exploited. The inset shows the spatial distribution of impurities within the considered cluster.

2012.09266

fig4.jpg

0.408026

ddb6f4b1760c4a83883877e3ae715ddd

tr-ARPES snapshots of WS_2/graphene heterostructure before and after optical excitation: (a) Photocurrent measured along the ΓK-direction at negative pump-probe delay. Colored dashed lines indicate the position of the line profiles used to determine the transient band positions in Fig. <ref>c and e. The yellow box indicates the region where the Fermi-Dirac fits for Fig. <ref>c were performed. (b)-(d) Photocurrent for different pump-probe time delays after photoexcitation at ħω_pump=2 eV with a pump fluence of 2.85 mJ/cm^2. (e)-(h) Pump-induced changes of the photocurrent for different pump-probe delays. Red and blue indicate gain and loss of photoelectrons, respectively, with respect to the photocurrent measured at negative pump-probe delay. The colored boxes in (f) indicate the area of integration for the pump-probe traces shown in Fig. <ref>a and b. Thin white (a-d) and black (e-h) dashed lines represent the calculated band structures of graphene <cit.> and WS_2 <cit.>. A rigid band shift of -0.81 eV and -1.19 eV was applied to the WS_2 CB and VB, respectively. The graphene Dirac cone was shifted by +0.3 eV to account for the observed hole doping <cit.>. Thick white (a-d) and black (e-h) dashed lines mark the position of the Fermi level.

2012.09268

Figure1.jpg

0.434983

0a25de5f9e634e41be0692005d64d720

smoking gun evidence for ultrafast charge separation: (a) Gain in the WS_2 CB (orange) and loss in the WS_2 VB (green) as a function of pump-probe delay. (b) Gain above the Fermi level (red) and loss below the Fermi level (blue) inside the graphene Dirac cone as a function of pump-probe delay. (c) Position of the WS_2 CB (orange) and the upper WS_2 VB (green) as a function of pump-probe delay. The transient band gap of WS_2 (turquois) was obtained by subtracting the transient position of the VB from the transient position of the CB. (d) Charging-induced shift of WS_2 valence and conduction band obtained by adding (subtracting) |Δ E_gap|/2 to (from) the VB (CB) position in panel a. (e) Transient shift of the graphene Dirac cone. Black lines are single exponential fits to the data. Gray lines in panel c were calculated from the exponential fit for E_gap in panel c and the exponential fit in panel d. All data was obtained with a pump fluence of 2.85 mJ/cm^2.

2012.09268

Figure2.jpg

0.367992

965db54ced2c493caaf24bc44bbdab84

fluence dependence of ultrafast charge separation and recombination: (a) Lifetime of graphene gain from Fig. <ref>b as a function of fluence. The black dashed line represents our temporal resolution of 200 fs. (b) Lifetime of photoexcited electrons in the WS_2 CB from Fig. <ref>a as a function of fluence. (c) Lifetime of gain above the equilibrium position of the upper WS_2 VB obtained by integrating the photocurrent over the area marked by the pink box in Fig. <ref>f and by fitting the resulting population dynamics with a single exponential decay. Thick lines in a and b were obtained by inverting and rescaling the guides to the eye from Fig. <ref>e to match the data points. The thick line in c is a guide to the eye.

2012.09268

Figure3.jpg

0.438959

b1ff0a9a05674434b9a307a5364e615c

direct tunneling model: (a) Sketch of the band structure with tunneling barrier for holes (Δ E_h, red) and electrons (Δ E_e, orange). (b) Temporal evolution of the two barrier heights deduced from Fig. <ref> together with single-exponential fit. (c) Temporal evolution of the electronic temperature inside the Dirac cone together with double-exponential fit. (d) Temporal evolution of the electron and hole transfer rates deduced from panels b and c together with double-exponential fit. (e) Pump fluence dependence of the peak transfer rates for electrons and holes. Thick gray lines are guides to the eye.

2012.09268

Figure4.jpg

0.43077

0051014fded0434faf66ea03988fe88a

theory and final microscopic model: Momentum dependence of the tunneling matrix element calculated from tight-binding for CB (a) and VB (b). k_x=k_y=0 corresponds to the K-point of graphene. Calculated transfer time for electrons (c) and holes (d). (e) Final microscopic model including direct (yellow and red arrows) as well as defect-assisted tunneling (green arrows).

2012.09268

Figure5.jpg

0.449003

3e2bbad3190d488d933b0c14a4aaa3a3

Sample characterization: (a) Secondary electron microscopy image. (b) Raman spectrum. (c) Photoluminescence spectrum.

2012.09268

SFigure1.jpg

0.440464

ff3d26cb18a04dcabf465457226fb99d

Direct comparison of fluence dependence of gain above equilibrium position of upper WS_2 VB (purple) and WS_2 charging shift (green).

2012.09268

SFigure10.jpg

0.443261

20994388cf8a4160972a3d8c438fc0b8

Evidence for in-gap defect states. EDCs through the K-point of WS_2 in the conduction band region at negative delay and at the peak of the pump-probe signal at t=0.2 ps. Gray lines indicate the Gaussian peaks used to fit the spectrum at t=0.2 ps. The two Gaussians marked with black arrows are attributed to in-gap defect states due to Selenium vacancies.

2012.09268

SFigure11.jpg

0.420668

f8c81de3befe4f1daac5cd1f4ac53296

Space charge effects: (a) Position of the Fermi level at negative delays as a function of pump fluence. The gray line is used to extrapolate the position of the Fermi level to zero pump fluence. The data point in brackets was obtained on an area of the sample where the WS_2 coverage was particularly low. The error bars are smaller than the symbol size. (b) Energy resolution for different pump fluences.

2012.09268

SFigure2.jpg

0.439864

f5959c838ec7495ea8bd442defb39cda

Treatment of raw data: (a) Tr-ARPES snapshot at negative pump-probe delay taken with a pump fluence of 1.63 mJ/cm^2. The pink box is used to extract I_box(t) in panel (b). The orange box captures the carriers at the bottom of the conduction band shown in panel (d). (b) Photocurrent I_box(t) integrated over the area marked by the pink box in (a) together with its average for negative pump-probe delays <I_box>. (c) Normalization factor f=<I_box>/I_box(t) as a function of pump-probe delay. (d) Effect of the raw data correction on the time dependence of the photocurrent integrated over the area marked by the orange box in panel (a). (e) Ratio of the reference intensities I_ref,gr and I_ref,WS2 as defined in Fig. <ref>c and a, respectively, for the datasets obtained with different pump fluences.

2012.09268

SFigure3.jpg

0.452665

94e0997a4f9241bebcd532df33aff194

Determination of transient band positions: (a) EDCs extracted along the dashed green line in Fig. 1a at t<0 fs (light green) and at t=200 fs (dark green) together with Gaussian fit (black) in the WS_2 VB region. The shaded areas indicate the three individual Gaussians used to fit the data. The area of the dark-shaded Gaussian I_ref,WS2 was used to turn the measured photocurrent independent of the WS_2 coverage. (b) EDCs extracted along the dashed green line in Fig. 1a from the differential tr-ARPES data shown in the lower panel of Fig. 1 in the WS_2 CB region for t<0 fs (dark orange) and t=200 fs (light orange) together with Gaussian fit (black). (c) MDCs extracted along the dashed red line in Fig. 1a for t<0 fs (dark red) and t=200 fs (light red) together with Lorentzian fit. The area of the dark-shaded Lorentzian I_ref,gr was used to turn the measured photocurrent independent of the WS_2 coverage. The data was obtained for a pump fluence of 2.85 mJ/cm^2.

2012.09268

SFigure4.jpg

0.482561

c3a0ecdfc79946c48793acb0f239a007

Fermi-Dirac fits: EDCs obtained by integrating the counts over the momentum width of the yellow box in Fig. 1a at t<0 fs (color) and t=200 fs (color) together with Fermi-Dirac fits (black).

2012.09268

SFigure5.jpg

0.462177

7dddc0b6343845d9933cc389dd418f63

Fluence dependence of the population dynamics: (a) Gain in the CB (orange) and loss in the VB (green) of WS_2. (b) Graphene dynamics: Gain above the Fermi level (red) and loss below the Fermi level (blue). (c) Gain above the equilibrium position of the upper WS_2 VB. Thin black lines are single exponential fits to the data.

2012.09268

SFigure6.jpg

0.492969

11f796503f0541129a6f3f367fdc027c

Fluence dependence of the transient band positions: (a) Position of the WS_2 CB. (b) Position of the WS_2 VB. (c) Transient band gap of WS_2. (d) Shift of the WS_2 CB due to charging of the WS_2 layer. (e) Shift of the upper WS_2 VB due to charging of the WS_2 layer. (f) Shift of the Dirac cone.

2012.09268

SFigure7.jpg

0.437439

29c232b068d449288999f6a61ae04807

Fluence dependence of hot carrier dynamics in graphene: (a) Electronic temperature, (b) chemical potential referenced with respect to the vacuum level, (c) chemical potential referenced with respect to the Dirac point, and (d) number of holes transferred from WS_2 to graphene as a function of pump-probe delay for different pump fluences.

2012.09268

SFigure8.jpg

0.477864

6a0072e37f0c4e119dcb9161faae0d3f

Fluence dependence of different parameters at the peak of the pump-probe signal (t=0.2 ps): (a) Pump-induced change of photocurrent inside colored boxes from Fig. 1f. (b) Change of the WS_2 band gap from Fig. 2a. (c) Number of holes transferred into the graphene layer from Fig. <ref>c. (d) Charging-induced WS_2 and graphene band shifts from Figs. Fig. 2e and f. (e) Peak electronic temperature of Dirac carriers from Fig. <ref>a. Light gray lines are guides to the eye. The datapoints in brackets were obtained on areas of the sample with lower WS_2 coverage.

2012.09268

SFigure9.jpg

0.45671

c84ebb3bdad340979b2dedb4a8cf30b6

Comparison of rankings by CoachScore while using different damping factors. (a) We report the rank position of all soccer coaches of our combined dataset. The ranking on the x-axis is for CoachScore with damping factor α=0.85, and the ranking on the y-axis with α=0.95. Spearman correlation coefficient is ρ=0.999, while Kendall correlation coefficient is τ=0.972. (b) Same as in panel a, but basketball coaches in NBA. Spearman correlation coefficient is ρ=0.996, while Kendall correlation coefficient is τ=0.954.

2012.09270

damping_factor.jpg

0.414143

4929304f403348e68e31d2360027a3c5

Statistics of the soccer dataset. (a) Total number of matches in our dataset per season. We combined together games from all the national leagues and European cups. (b) Number of coaches in our dataset per season. Each season includes every coach that has managed at least one game in that season. Data combine all games of the national leagues and European cups. (c) Complementary cumulative distribution functions (ccdf) of the number of matches managed, won, lost, and tied by coaches in our dataset. The stretched exponential ccdfs that best fit the empirical data are displayed as dashed curves. The stretched exponential ccdf is defined as f(x)=exp(-x^β). The parameter of the best fits are β=0.695 for total matches, β=0.592 for wins, β=0.791 for losses, and β=0.676 for ties. (d) Ccdf for the number of head-to-head matches played between pairs of coaches. The best fit curve to the empirical data is also shown. In this case, we used the lognormal ccdf g(x)=1 - 1/2 erf( log (x) - μ/√(2) σ). The parameters for the best fit are μ=0.452 and σ=0.945.

2012.09270

fig1.jpg

0.429798

ad02f80db17c4e19b62707c1454bfe40

Comparing career performance of basketball coaches. We display career trajectories for five selected coaches: Pat Riley, Phil Jackson, Gregg Popovich, Rick Carlisle, and Erik Spoelstra.

2012.09270

fig10.jpg

0.488916

f4cf97698ed8415e930623592769d005

Statistics of the basketball dataset. (a) Total number of matches per season. (b) Number of coaches per season. Each season includes all coaches that have managed at least one game in that season. (c) Complementary cumulative distribution functions (ccdf) of the number of matches managed, won, and lost by coaches in our dataset. The stretched exponential ccdf is defined as f(x)=exp(-x^β). The parameter of the best fits are β=0.746 for total matches, β=0.664 for wins, and β=0.821 for losses. (d) Ccdf for the number of head-to-head matches played between pairs of coaches. The best fit curve to the empirical data is also shown. In this case, we used the lognormal ccdf g(x)=1 - 1/2 erf( log (x) - μ/√(2) σ). The parameters for the best fit are μ=1.567 and σ=1.098.

2012.09270

fig2.jpg

0.46285

1a64c6149406447095b81a76fa7dbf65

Comparison of rankings by CoachScore and number of wins. (a) We report the rank position of all soccer coaches in our combined dataset (see Table <ref>). Rankings are performed by either relying on the CoachScore values or the number of wins. Each point in the plot is a coach. Spearman correlation coefficient is ρ=0.989, while Kendall correlation coefficient is τ=0.923. (b) Similar to panel a, but for basketball coaches. Rank correlation coefficients are ρ=0.979 and τ=0.888.

2012.09270

fig4.jpg

0.407662

3ed82277240c45529c88cc75e712c54f

CoachScore vs. number of points. We report the rank positions for the 20 soccer coaches who managed teams in the 2018/19 English Premier League. We rank coaches on the basis of their CoachScore and compare those ranks with those obtained with the number of points they gathered in the league. Spearman rank correlation coefficient is ρ=0.973, and Kendall rank correlation coefficient is τ=0.896. The line stands for perfect agreement between the two rankings.

2012.09270

fig5.jpg

0.489632

62b778457a1e47bb9081920b1b9f4e22

Monitoring the career performance of soccer coaches. (a) We visualize the dynamic rank of Pep Guardiola throughout his career. Rank positions are updated after each game day. Red lines indicate when a coach takes over a new team, dashed green lines indicate the start of a new season with his current team, and grey shaded areas represent periods of time when the coach is not managing any team. FCB stands for FC Barcelona, BM for FC Bayern Munich, and MC for Manchester City FC. (b) Same as in panel a, but for JosÃ© Mourinho. CHE stands for Chelsea FC, INT for FC Internazionale Milano, RM for Real Madrid CF, MU for Manchester United FC, and TOT for Tottenham Hotspur FC.

2012.09270

fig7.jpg

0.495882

f3756f0c05494ea085a5f236b7c3c739

Monitoring the career performance of basketball coaches. (a) We visualize the dynamic rank of Phil Jackson throughout his career. Rank positions are updated after each game day. Red lines indicate when a coach takes over a new team, dashed green lines indicate the start of a new season with his current team, and grey shaded areas represent periods of time when the coach is not managing any team. CHI stands for Chicago Bulls, and LAL for Los Angeles Lakers. (b) Same as in panel a, but for Steve Kerr. GSW stands for Golden State Warriors.

2012.09270

fig8.jpg

0.415024

db764d74cd914c6f97584629a13561ee

Comparing career performance of soccer coaches. We display career trajectories for five selected coaches: Jupp Heynckes, Sir Alex Ferguson, JosÃ© Mourinho, JÃ¼rgen Klopp, and Pep Guardiola.

2012.09270

fig9.jpg

0.512086

d324045861ef44b0a95860792e79e519

Network of contact among top coaches. (a) We display the network of contact among soccer coaches that have won the UEFA Champions League at least twice after 1980. The network is built setting β =0 in Eq. (<ref>) and including all games in our dataset. In the visualization, the size of the nodes is proportional to their in-strength, and the width of the lines connecting pairs of nodes is proportional to the weight of the corresponding edge. (b) Same as in panel a, but for top basketball coaches. We visualize the network of contacts among coaches that have won at least 2 NBA titles.

2012.09270

nws.jpg

0.494204

633260fc93b14853b2588340ba5a6fc4

Comparison of rankings by CoachScore while using different weights for win in soccer. We report the rank position of all soccer coaches of our combined dataset. The ranking on the x-axis relies on the network where a win has twice the weight of a tie. The ranking on the y-axis relies on a network where a win has three times the weight of a tie. Spearman correlation coefficient is ρ=0.999, while Kendall correlation coefficient is τ=0.983.

2012.09270

win2_vs_win3.jpg

0.46667

5d7d6b3c2449473985b6c39c9e1f6795

Picture of a regular {5,5} tessellation in in the Poincaré disc model. All line segments are geodesics with respect to the hyperbolic geometry.

2012.09271

55PD.jpg

0.405759

5a2e3ebfc7d24513a512dd5b320b9b42

The second largest eigenvalues of hyperbolic surfaces of small size labeled by their Schläfli symbol {r,s}. The dashed lines indicate the asymptotic lower bound 2√(s-1) of <Ref>.

2012.09271

hyperbolic_spectrum.jpg

0.466659

642f35d95d0e46d2977854f43b7702a0

A hyperbolic genus 3 surface supporting the {3,7} tessellation. It is called the Klein quartic. The dotted line indicates the translation τ = (ρ σ^-2)^4 under which points are being identified. The symmetry group of the surface is Isom(ℍ^2/N_τ)=PSL(2,7)⋊ℤ_2.

2012.09271

klein_quartic.jpg

0.487849

bd691c91b1b74e8686dc2ff365b2dd86

Applying DBSCAN to the t-SNE output, noise shown in black (Automobile).

2012.09272

clustering.jpg

0.493152

21cde0c9bc98481ca5dc521a40df515f

Large batch training accuracy for Full CIFAR-10, Randomly filtered and Network filtered data (Avg. of 4 runs). Network filtered data gives at least the same accuracy as full data and outperforms randomly filtered data.

2012.09272

comparison.jpg

0.446409

a3f509a33e5b44ae90ed36033cce711a

Large batch size vs. Accuracy for different lr. Accuracy increases until it reaches an optimum batch size and then starts to gradually decrease.

2012.09272

lbtr.jpg

0.460934

e7404a452998484a8c8808838a3df433

Learning rate vs. Preferred batch size for single GPU

2012.09272

optimum_batch_size.jpg

0.402811

fa4a7173166346588b666700230e45e2

Steps showing the process of applying data optimized strategy by (1) fine tuning a pretrained network, (2) Extracting and reducing dimensionality by PCA / t-SNE, (3) Clustering using DBSCAN to remove noise and (4) Training another network faster and at improved accuracy.

2012.09272

process.jpg

0.428562

c919c951c71f4283b53dab522e32521f

Small batch size vs. Accuracy for network filtered, randomly filtered and full CIFAR-10 data sets.

2012.09272

small_batch_single_GPU.jpg

0.424957

ea4c6a837fa142379284c33156d545f8

Visualizing output of extracted features from the fine-tuned ResNet-152 using t-SNE.

2012.09272

t-sne.jpg

0.436008

ffac5929004641e687b1f15f6fed7612

Schematic view of a possible TIARA design: the pixels are placed around the target at fixed positions. Pixels coordinates can be established patient-wise at treatment planning in order to clear one or more entrance windows for the proton beam (red line). TIARA will be read in time coincidence with a beam monitor (light blue).

2012.09275

figure_1.jpg

0.394033

d3d79ad4b97e4c4a9dad85cb12c468ad

Left: correlation between the detectors y_COG calculated under treatment conditions and the actual beam displacement imposed in the MC simulation. Right: correlation between the calculated x_COG and the actual cavity thickness considered in the MC simulation. Both datasets correspond to a monitoring statistics of 10^8 incident protons. The 1σ (orange) and 2σ (blue) errors are displayed in both graphs. A linear fit was carried out for each dataset.

2012.09275

figure_10.jpg

0.436175

690990a9d28b488cb1cf660184bf5d60

Left: geometry simulated with Geant4 to validate the analytical reconstruction algorithm. The impinging 100 MeV proton beam is represented in blue. In red, a PG created along the beam path and recorded in the phase space is also shown. A diamond based beam monitor (represented in black) placed upstream of the target was considered. Right: the same geometry is displayed, with the insertion of an air cavity of variable size and fixed position at (-5.5,0,0).

2012.09275

figure_2.jpg

0.381793

d4685c24e954444faaf1ec10cf8bd3e5

Energy (left) and TOF (right) spectra of each particle type scored in the phase space. "Others" emissions mainly include 511 keV gamma-rays, but also PGs from secondary neutrons and protons.

2012.09275

figure_3.jpg

0.478887

1b6a05c7ed8842df8a2e1b1be97ac498

Reconstructed PG profiles for the 1 cm thick air cavity geometry obtained with different monitoring statistics and at TPS statistic (1.5×10^9 incident protons). The Ecut was applied to all datasets.

2012.09275

figure_5.jpg

0.406984

9652532097974ebeb176e3e873ca03cc

The reconstructed PG vertex profile obtained at TPS statistics and shown in figure <ref> after the Ecut was applied, is decomposed to show the separate contributions of the different secondary particle types detected in the phase space. Slower particles as neutrons (red curve) are naturally time-discriminated from the other particles.

2012.09275

figure_6.jpg

0.416034

36886ed2861c4564be817c1b11fc117a

Comparison of PG vertex profiles reconstructed in TPS (blue curve) and treatment (orange curve) conditions in the case of a 1 cm thick air cavity. In both graphs, a high statistics of 1.5 × 10^9 incident protons was employed for the TPS profile. On the left, the treatment profile is reconstructed assuming a 100 ps (rms) CTR and a statistics of 10^8 incident protons, corresponding to ≈ 3×10^4 PG detected, whereas on the right a CTR of 1 ns (rms) was considered, with a statistics of 10^9 incident protons, corresponding to ≈ 3×10^5 PG detected. The Ecut (E > 2 MeV) was applied to data in both the left and right graphs.

2012.09275

figure_7.jpg

0.484782

2c37d0b77ba14a0f9c8bab60caf078c7

Comparison of PG vertex profiles (left) and integrated PG vertex profile (right) obtained for the reference simulation in TPS conditions (1 cm air cavity) (blue) and for a treatment geometry (orange) presenting a 1.5 cm thick air cavity. In both plots, the vertical dotted line (in red) locates the TPS PG profile peak at x_TPS. In the right plot, the horizontal dashed line (in green) in the inset represents the calculated profile distance d_5.

2012.09275

figure_8.jpg

0.451755

7e8d1b279c634cacbe119f29cb85c04b

Correlation between the actual air cavity thickness and the corresponding measured profile distance d_i in the case of a 100 ps (rms) CTR and 10^8 incident protons (left), and in the case of a 1 ns (rms) CTR and a statistics of 10^9 incident protons (right). The 1σ (orange) and 2σ (blue) errors are displayed for each data point. A linear fit was carried out for each dataset.

2012.09275

figure_9.jpg

0.421111

8eefbe95a9794696b652d69d8abbcc67

Cars3D

2012.09276

cars3d_kendall.jpg

0.417795

84a141c2af8c45738290b42bd259a76c

Mean metric scores as the representation becomes less modular and less compact.

2012.09276

compactness_modularity.jpg

0.424028

5bd454e5e5d54aa3b6af3bdc09441ee3

Metric scores for perfectly disentangled representations under increasing noise level (α).

2012.09276

noise.jpg

0.395223

330924e4a87943a4b476099973627f8d

Metric scores for perfectly disentangled representations with increasingly nonlinear factor-code relation (α).

2012.09276

non_linearity.jpg

0.419194

5c1dc207e89041b1bb3963a9b76e74c5

Scores for perfectly disentangled representations. The abscissa indicates how many of the 8 factors are measured by the metrics.

2012.09276

partial_factors.jpg

0.410188

a51f6c35428a44d1b13d5aec7a208461

SmallNORB

2012.09276

smallnorb_kendall.jpg

0.514931

7154b85e78e44633b2c8b0c35842ae9a

Shape of the factor-code relation as parameter α increases.

2012.09276

tan_curve.jpg

0.412327

4103573f4cc74f16a84ced696451b906

Effect of nonlinearity on discretization bins population.

2012.09276

tan_hist.jpg

0.436849

b639f9381d184792b7299f05b0104757

Semi-analytical Slonczewski field H_S (a) and velocity v_S (b) compared to the limiting analytical expressions Eq. (<ref>) resp. (<ref>). The ratio of the two expressions is plotted versus δ=H_D/H_DW for different values of the damping parameter α. The curves for α=0 are analytic (see text).

2012.09277

Figure3-512.jpg

0.404942

173f2648f32848f5985c919bd937e19b

Attention vector learnt by the proposed method. Y-axis is the slice number of the CT image, and the X-axis is the slice number (depth) of the 3D feature map F_i^2D.

2012.09279

attention.jpg

0.370367

d24d9e5f5d264f56a760c454977cc06a

A CT image from ABD-110 dataset. The first and second columns are the ground truth and predicted mask overlayed on the original CT image slice, respectively. The third column shows the comparison of contours of the ground truth and predicted mask on the same slice.

2012.09279

visualization_WACV.jpg

0.451998

c083bcdcca4149579bf9f58f1a74a266

Sensitivity of a ^76Ge double-β decay experiment searching for massive sterile neutrinos with a mass of 500 keV, given as a function of exposure and background level. The sensitivity is expressed in terms of the median 90% C.L. upper limit on the squared of the mixing angle (sin^2θ), computed assuming no signal. The markers indicate the value computed using a full frequentist analysis, while the lines show their best fit with the analytical expression discussed in the text. The plot shows the impact of additional backgrounds with a rate R_ other which is expressed w.r.t. the irreducible decay rate R_.

2012.09281

Sensitivity_vs_exposure_m500keV_Ge76.jpg

0.452741

a55b279888934e96af8f942f9aacb446

Comparison of the algorithm and QuTiP's over an evolution of 20 Rabi oscillations, with the same parameters as in Fig <ref> of the main text. Left panel: Frobenius norm distance between the algorithm with different subpixel numbers and QuTiP's with (circles) and (x's) methods. The squares are the norm distance between and with 10^4 subpixels. The dashed line indicates the error between with the and methods. Right panel: Computational time of and for results of the left panel. The dashed lines refer to the computational time for with and with the highly accurate settings described in the text.

2012.09282

Fig-2panels-DysonvsPropag-Dyson1to4-1drives-Hilbert25-density0.20-500ns.npz.jpg

0.492081

77c790b2c83443b5bd32af9bad7d26ae

benchmarks. First row: Frobenius norm distance from the propagator Û_ref(0,T) for the algorithm for T=500 ns. Second row: Contraction time of the algorithm as a function of the subpixel number N_s. The slope of the data is precisely 1, meaning that the computational time scales linearly with the number of subpixels. Preparation stage computation time: (35 ms, 1.9 s, 6.8 s) for (-2, 3, 4) and 1 input drive, (0.35 s, 13 s, 87 s) for 2 input drives and (0.78 s, 47 s, 408 s) for 3 input drives.

2012.09282

Fig-6panels-benchmarks.jpg

0.464242

59c661d50b864fb6968e7daff885947d

Infidelity of the CR gate as a function of the chosen gate time when the gate is operated at the qubit-qubit detuning Δ/2π = (ω_c - ω_t)/2π = 210 MHz. The red curve indicates the best obtainable infidelity from a constant amplitude pulse, with the orange curve reporting the infidelity up to arbitrary single qubit rotations on both qubits, applied before and after the cross-resonance gate channel. The blue curve reports the infidelity when optimized under the GRAPE algorithm until convergence.

2012.09282

Fig-GRAPE-vs-flatpulse-10subpix.jpg

0.469097

b475a16537554959b74d8a4fb4a0bf2f

Infidelity of the CR gate as a function of the detuning of the qubits when optimized under GRAPE for 5000 iterations. The control qubit frequency is fixed to ω_c/2π=5.10 GHz.

2012.09282

Fig2-Detunings-GRAPE-TransTrans-300ns.jpg

0.518366

d0ed5f5ccff942a39a14d350a529bd07

Linear interpolation of the pulse sequences (yellow) with two subpixels per pixel. The black bars indicate the original Gaussian filtering, with the blue line the `true' filtered pulse.

2012.09282

Interpolation-new.jpg

0.465413

31acc6275a5c49f8b11ea9e558d594ff

Tree diagram showing the different branches of the time ordered integral, with two example branches highlighted.

2012.09282

PathDiagramTeststars.jpg

0.464479

a4584907c6054809b595fc9c125070f1

Superconducting circuit for cross-resonance gates between transmon qubits. The leftmost transmon, of frequency ω_c/2π, plays the role of the control qubit and is strongly driven by the voltage source V_c at the frequency ω_t/2π of the target qubit (rightmost transmon). The target qubit is also driven by V_t using a relatively weak tone that serves to give additional control. φ̂_c and φ̂_t correspond to the phase degree of freedom associated with the control and target qubits, respectively.

2012.09282

cr.jpg

0.457594

c1c42b3bab3c43aa80c4158f96ea0c66

Cryptographic information flow between two agents and one SO.

2012.09285

Encryption_flow_2.jpg

0.427638

5a59626cb0ce4f2897c5dde3d4837ee9

Encryption error and optimality gap (precision level σ = 3).

2012.09285

congestion_convergence_errors.jpg

0.483331

f31008255dd44d5aadda51d4bb6218d4

Convergence of the primal and dual variables.

2012.09285

congestion_primal_dual.jpg

0.435422

28957206e29b431c8a62c40063f174aa

Encryption error and optimality gap (precision level σ = 3).

2012.09285

convergence_error.jpg

0.408228

e0ea6bc6b037472aa93705b3e195d769

Secure information exchange between agents and the SO.

2012.09285

information_exchange_all_rc.jpg

0.467081

d083dc35a4154398aeb6e178ceb041d7

Convergence of the primal and dual variables.

2012.09285

primal_dual_convergence.jpg

0.506112

cfd988da1da040048ba97b3bf77581a3

A transportation network with 9 links and 5 agents <cit.>.

2012.09285

transportation_network_f.jpg

0.458471

a83ea6eaf86f488eb6a14f8647c0e8cb

Model performance on CelebA during training.

2012.09290

ablation_chart.jpg

0.415399

42fd4cede0714b779ff23765542d8c18

In each panel, the second row shows the images from AE, and the third row shows the GAN revisions.

2012.09290

ae-to-gan-sample.jpg

0.453876

3efe23289b264d0ebb20aeb087f7318a

Qualitative comparison of the Auto-encoder without and with the proposed self-supervision objectives. In each panel, the first row are the referential style images, the first column is the input sketch, the second row are synthesis results from a plain AE trained with only reconstruction loss, the last row are the results from the proposed AE with self-supervision objectives: ℒ^s_tri, ℒ^c_tri, ℒ^s_cls, ℒ^c_cls.

2012.09290

ae_noss_compare_art.jpg

0.435088

63166d25d75d41869202458c84361a6d

Qualitative comparison of the Auto-encoder without and with the proposed self-supervision objectives. The image arrangement is the same as Figure <ref>

2012.09290

ae_noss_compare_face.jpg

0.484324

3ee2a700c98847e58e3a120c0c968b56

Uncurated synthesis results of our model on WikiArt. Note that the images are compressed due to file size limit.

2012.09290

art_quality_1_compressed.jpg

0.430207

f628ece3d8f84505929569ac974c22fb

Uncurated synthesis results of our model on WikiArt. Note that the images are compressed due to file size limit.

2012.09290

art_quality_2_compressed.jpg

0.487831

ed75bceae8764990903d57e974bb7901

Uncurated synthesis results of our model on CelebA. Note that the images are compressed due to file size limit.

2012.09290

face_quality_1_compressed.jpg

0.472949

d5e893750a9b4b1ea45d04d9e865fc3d

Uncurated synthesis results of our model on CelebA. Note that the images are compressed due to file size limit.

2012.09290

face_quality_2-compressed.jpg

0.457278

9811f5018a814a89996f0cb1695645bc

Exemplar-based sketch-to-image synthesis from our model on varied image domains in 1024^2 resolution.

2012.09290

main-samples_compressed.jpg

0.454973

e012c8c4e50a41548fb6b2f442e08244

Overview of the proposed model.

2012.09290

model_overview.jpg

0.474955

b68ed3baaf0a4cf1bf6da68a05809c33

Synthesises from TOM. TOM generalizes well across multiple image domains, from photo-realistic to artistic, and from human portrait to nature landscape.

2012.09290

skt_sample.jpg

0.530023

d5f533a8a03b4e199fc575d2e3a3b5f7

Example sketches we used to train TOM. The bottom panel are line-sketches from artists, and the top panel are sketches we randomly searched from the internet.

2012.09290

skt_target.jpg

0.431286

7e0a1b43b29a4db8bda0afa5feef3761

Synthesis by mixing multiple style images.

2012.09290

style-mix-example.jpg

0.409181

c697fbfeba8e4f299764376f99594328

Synthesis from out-domain style images.

2012.09290

style-outdomain-example.jpg

0.400487

02a22247b59545439d7ef5b79843da7b

Works as a feed-forward style-transfer model.

2012.09290

style-transfer-example.jpg

0.472118

6a793e7956364659a9319a046fa14184

Illustration of our TOM. Dashed arrows in red indicate the gradient flow to train the sketch generator.

2012.09290

tom_overview.jpg

0.420657

bfe55f1ad5c9400aac96ef8f8e22a251

Hubble diagram of quasars. Magenta dashed line is the best-fit flat ΛCDM model with = 0.670 from the QSO-z < 1.479 QSO data. Cyan points are the determined QSO distance moduli and uncertainties <cit.> and red points are the means and uncertainties of these distance moduli, in narrow redshift bins which roughly represent a fifth order cosmographic fit of the whole QSO data <cit.>. The black solid line shows a flat ΛCDM model with = 0.30.

2012.09291

QSO_hubble_2421.jpg

0.437314

bf2a041eddc74eb8b66f26d2022cf7c2

Redshift distribution of the 2038 QSO that we use in our analyses.

2012.09291

Redshift_distribution_2400.jpg

0.466448

e29f1d2560584b1eb4a28155c8c34a9e

AppIS: runtime behavior.

2012.09292

AppIS-scheme-runtime.jpg

0.444738

4025f86a031841dbab58edc2f7a0c74f

The AppIS protection scheme.

2012.09292

AppIS-scheme.jpg

0.436712

495b598594014c0da58cb5fd41a279cc

The BombDroid protection scheme.

2012.09292

BombDroid-scheme.jpg

0.435353

8f399f8159084d518f5c4b9f79cee4c8

Runtime behavior of a nested logic bomb.

2012.09292

LB_runtime_example.jpg

0.477864

e3b00d546459444eadce0b86cdee960d

Transformation process of a qualified condition into a logic bomb.

2012.09292

QC_to_LB.jpg