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
0.463401	2cf1491977a04cb584054021db1e9f1a	Structure of QRNN. (a) Structure of QRNN for a single time step. (b) QRNN in the training phase. (c) QRNN in the prediction phase. Note that the obtained prediction x_T is used as the input to the next time step.	2012.11242	QRNN_crop.jpg
0.433205	805e01cf06d64810896b01a267b1605c	(a) Schematic picture of a basic RNN. Blue circles represent elements of vectors x_t, h_t and x_t. The matrices V_in, V_out and W connecting layers are the parameters to be optimized during training of the model. The bias vectors b and c are abbreviated for simplicity. (b) Unfolded version of (a). (c) RNN for time-series prediction.	2012.11242	RNN_combined_new_crop.jpg
0.453777	07a5f9f67e444a6e9134ac76e5584845	The evolution part of the circuit utilized in the numerical simulations. α_i^(d), β_i^(d), and γ_i^(d) are rotation angles and act as independent parameters of the circuit, i.e., θ consists of {α_i^(d), β_i^(d), γ_i^(d)}. d=1,⋯, D is the layer index.	2012.11242	rotation_gate.jpg
0.439153	58b5ff5b19444b3b949995bb2f299c72	Results of numerical simulation of training the QRNN to predict (a) cosine wave, (b) triangular wave, and (c) three-spin Lindblad dynamics. The data points on the left of the vertical grey dashed line are used as the training dataset. The solid lines represent the function which the QRNN is trained to model. The parts of the data used in the training phase and the prediction phase are depicted as the black and orange lines, respectively.	2012.11242	simple-simulation_crop.jpg
0.429756	110d8e694a3d4d08910fbae8d4ac2884	Relation between the evolution time τ and the prediction error evaluated in terms of the mean squared error.	2012.11242	tau_vs_mse.jpg
0.519605	c17dc3b0b84948a6b4a0f0d3fb174d88	Top panel: The distribution of k^* for the LHCb acceptance with and without applying the efficiencies. Bottom panel: the ratio of -hadron from the same parent (SP) to all pairs.	2012.11250	hCKstarJpsiHadron_paper_1.jpg
0.565064	3fa3c4ab88dc411c8fa7fa9b8e719480	Top panel: The distribution of k^* for the STAR acceptance for J/ψ→ e^+e^- reconstruction with and without applying the efficiencies. Bottom panel: the ratio of -hadron from the same parent (SP) to all pairs.	2012.11250	hCKstarJpsiHadron_star.jpg
0.423184	85aab23eae2f4c5da83cd5956c9ca29b	The shuffled shot noises under mass and optimal weighting schemes (dotted lines). For comparison, the expected shot noises under mass and optimal weighting schemes (solid lines) are also presented.	2012.11251	shuffled_shot_noise.jpg
0.459529	d3e75f9ecdd04b70bc01577d69bdbe0f	(Left) This figure shows the “unfolded” multi-layered context network with increasing dimensionality that captures larger and more influencing contexts. (Right) Hierarchical context learning framework including geometric and semantic contexts. In this diagram, we only show four semantic neighbors for each image as an example (better to zoom the PDF version).	2012.11253	figure2.jpg
0.360482	f20f4f17dab64289a81af75461920485	The two DW configurations in 1x1x16 supercells used in order to determine the crystallographic plane where the oxygen vacancies are placed in the calculations. In a) BaO centered TT wall with vacancies, in b) TiO_2 centered TT wall with vacancies. The forces of both structures have been relaxed.	2012.11254	batenticen.jpg
0.582117	ac75320f24554a14bb3837cbeaf3dbfd	Properties of the unrelaxed CDWs in a 1×1×16 supercell. Top: schematic atomic structure with exagerated displacements. The blue arrows indicate the direction of polarization; green, blue and red spheres indicate Ba, Ti and O atoms, respectively. The dots indicate positions of Ti atoms in the individual unit cells. Second from top: polarization profile of the supercell. Second from bottom: bound ρ_b and free ρ_f charge density. Bottom: electrostatic potential energy v(z) averaged over the plane orthogonal to z and total charge density ρ obtained from a sliding window average.	2012.11254	fullfig1.jpg
0.393068	855459ad478c4842bdb0a0de144ef1cd	PDOS at the HH wall of the 1x1x16 supercell containing a HH and TT wall. Ti t_2g states, Ti non-t_2g states and the rest are differentiated with different colors. The Fermi level is set as the origin for the energies.	2012.11254	hhti.jpg
0.401585	457d73ba8e7444fb8b15df668463b89d	PDOS at the HH wall of the relaxed 1x1x16 supercell containing a HH and TT wall and an oxygen vacancy at the latter. Ti t_2g states, Ti non-t_2g states and the rest are differentiated with different colors. The Fermi level is set as the origin for the energies.	2012.11254	hhtivac.jpg
0.482519	28dedee96835440cb8c19247d1f5f6c3	Left: projected band structure of the unrelaxed 1×1×16 supercell with no vacancies. Right, the projected density of states resolved in individual unit cells.	2012.11254	novacband.jpg
0.39942	1ba6fdd59e5442778541b4c88cc1e7ef	Density of states of the relaxed 2x2x8 supercell with a HH and TT wall, with two vacancies at the latter, projected into the atomic orbitals of TiO_2 planes. The Fermi level is set as the origin for the energies.	2012.11254	pdostiolayers.jpg
0.502139	a54eb5b8d33f4440915b91f79f0ead8d	At the top, the reference 2×2×8 tetragonal BaTiO_3 supercell with one vacancy at a BaO plane. Below, the energy cost profiles for the i^th vacancy at α position (see text). On the top-right, the section of the supercell corresponding to the central BaO plane where static vacancies are set. Second from the bottom, the polarization profiles of the relaxed supercell for i= 0, 1, 2 and 3. At the bottom, bound and free charge density profiles for the relaxed 2×2×8 supercell with two vacancies at the central BaO plane.	2012.11254	vacprofiles.jpg
0.391881	aa39cb1106ce4a28abc0b4589fb20699	Mean and variance of difference rewards for a set of samples under different noise profiles.	2012.11258	Noise.jpg
0.376894	dcacbcd76262486eb3d46a43bbf3be8a	Mean and variance of difference rewards for a set of samples under different noise profiles.	2012.11258	Noise_Large.jpg
0.421188	9572e52f824b4893b04d5c64ca9b3ee9	Normalized mean prediction error and standard deviation for Dr.ReinforceR reward network R_ψ and COMA critic Q_ω on the on-policy dataset (first row) and the off-policy dataset (second row), for the two environments.	2012.11258	Statistics.jpg
0.45559	0b14aa1ac92a4ca599477754fe4e07f3	Illustration of static and dynamic adaptation of DOS under external perturbation by adding a small amount of electron dN. The effect of perturbing μ by a small amout dμ could be understood in a similar fashion. Note that the dynamic adaptation is closer to reality.	2012.11259	DOS-change-by-dN.jpg
0.414593	1003789162334b0d8d7fd54a4dc4330c	Demonstration of the correlation between the Fermi softness (S_F) and the reactivity of various top sites (tpf, tpc and tpl, as shown in the inset) on Pt nanoparticles as represented by the CO adsorption strength (upper panel). For comparison, correlation with the d-band center is also presented (lower panel). tpl79: top site on Pt atom in Pt nanoparticle made up of 79 Pt atoms.	2012.11259	E_COads_vs_sf_PtNP.jpg
0.379813	39c455e6bb6f449c8e97af37c117a00b	Correlation between the global Fermi softness (S_F) and the surface reactivity as indicated by the oxygen adsorption strength on Pd-derived surfaces (see the supplementary material of reference <cit.> for details). For comparison, the correlation of the d-band center for the same systems is also shown.	2012.11259	Eads-SF-vs-Eads-Ed.jpg
0.420581	2f2aea20202b4a9b88965979013de682	Comparison of correlations between oxygen adsorption energy (Δ E_O) on Pt-related surfaces and S_F of surface Pt atom at two electronic temperatures: 0 eV (A, no spreading in dFDD) and 0.4 eV (B). How well Δ E_O is correlated with S_F is measured by the Spearman's r and larger absolute value indicates better correlation.	2012.11259	Eads-vs-sf-zero-and-finite-Tel.jpg
0.487747	6a28702d3d6a43f78af9b8f6f2ea004b	Plot of A) Fermi-Dirac distribution (FDD for short) function and B) its derivative (dFDD for short) at three representative electronic temperatures: 0.01 eV, 0.4 eV and 1 eV. One could also see from panel A the approximate equivalence of the chemical potential and the energy associated with Fermi level at these diverse T_el's, as indicated by the dashed vertical lines.	2012.11259	FDD-and-dFDD-Tel.jpg
0.407939	117c9cfbfa2242d9951f9d4b06b63e70	Spatially resolved chemical reactivity of a 1D MoS_2 edge. A) The geometrical structure of MoS_2, with three typical local sulfur environments (S#1, S#2, and S#3) and two typical molybdenum atom environments (Mo#1 and Mo#2). B) The projected density of states (g(ε)) of specific sulfur and molybdenum atoms. C) A 3D perspective of the local Fermi softness s_F(r), as represented by some isosurface (blue). D) The s_F(r) is plotted as projection onto two planes normal to the x direction: plane a intersects two S#1 and one Mo#2 atoms, and plane b contains one Mo#1, two S#2, and one S#3 atoms. E) Reaction energy profiles for the reaction 2H_ad→H_2 at two kinds of bridge sites (br-x or br-y, see the inset) on the S#1 dimer edge, obtained from nudged elastic band calculations.	2012.11259	SF-MoS2.jpg
0.476401	50f87e0dbf7b4763bdcd9676d8dd2cbb	Graphical illustration of S_F, defined as a weighted sum of the density of states ∫ g(ε)w(ε)dε, where the weight function w(ε) is chosen as the derivative of the FermiâDirac distribution function at non-zero electronic temperature.	2012.11259	Sf-concept.jpg
0.487786	8475e652d7ac4ae186c0359aca73475c	Schematic illustration of stabilization of total single electron energies upon shifting from ε_d (A) to a higher value (ε_d', B) of d-band center of transitional metal surface when interacting with an adsorbate.	2012.11259	demo-d-band-model.jpg
0.424796	6fb155c4df0442b1b955f6547ac617d9	Orbital interactions of surface band and three types of adsorbate orbital: half-occupied HOMO (A), doubly occupied HOMO (B) and empty LUMO (C). The five columns respectively correspond to five different states of surface band, being at Fermi level, slightly below ε_F, well below ε_F, slightly above ε_F and well above ε_F. Red bar: magnitude of energy gain after orbital interaction; Green bar: magnitude of energy penalty upon orbital interaction; Solid upward (downward) pointing arrow: electron with spin up (down) occupying some electronic state; Gray arrow: electron occupying some “intermediate” electronic state that is about to fill the unoccupied state at the Fermi level of surface band.	2012.11259	orbital-interactions.jpg
0.392249	85be5f44abaf4970844a23a1033c7946	Charging results of EVs controlled by PPC with tuning parameters β=2× 10^3 (top), 4× 10^3 (mid) and 6× 10^3 (bot) for selected days (with no less than 30 charging sessions, i.e., N≥ 30) from Dec. 2, 2019 to Jan. 1, 2020. Each bar represents a charging session.	2012.11261	battery.jpg
0.551064	2c5b7364fb454381ad72921192670196	Substation charging rates generated by the PPC (orange) in the closed-loop control shown in Algorithm <ref>, together with the MPC generated (blue) and global optimal (dashed black) charging rates.	2012.11261	charging_caltech.jpg
0.401171	f47e5a5b5d284cd5bc906120621dac01	Cost-energy curves for the offline optimization in (<ref>)-(<ref>) (for the example in Section <ref>), MPC (defined in (<ref>)-(<ref>)) and PPC (introduced in Section <ref>).	2012.11261	costs_energy.jpg
0.492933	9eddfa5e51154bb8a6b5a9a37bd0f2fb	Average rewards (defined in (<ref>)) in the training stage with a tuning parameter β=6× 10^3. Shadow region measures the variance.	2012.11261	learning_curve.jpg
0.466523	bcff564a23ab4d31aec83b91f3ffaaa1	Feasible trajectories of power signals and the computed maximum entropy feedback in Example <ref>.	2012.11261	toy_example.jpg
0.464716	5dcbc1069d6b4eb7bc66dd8b6ce1e601	Trade-offs of cost and charging performance. The dashed curve in the left figure corresponds to offline optimal cost. The tested days are selected (with no less than 30 charging sessions, i.e., N≥ 30) from Dec. 2, 2019 to Jan. 1, 2020.	2012.11261	tuning.jpg
0.41819	767cb8a5e7474ecc8d95d7f45e3b2a16	(Bottom) Pulsed peak power dependence of emission spectra for 4 diameter microdisks with Sn contents between 7 % to 10.5 % with corresponding L-L curves at 25 (Top). The continuous curves in the figures on top are calculated curves using the laser rate equations with the indicated thresholds and beta factors. The insets show typical spectra above lasing threshold.	2012.11262	FIG3.jpg
0.412057	edc2b4f786c14e37ba096bd6d3c980a7	a L-L curve obtained at 25 K from a 10 diameter microdisk with 10.5 % of Sn underetched with a CF_4 plasma. The dashed lines are drawn to guide the eyes. The inset shows a SEM image of such microdisks b L-L curve obtained at 25 K from a 8 diameter microdisk fabricated from the same epitaxial layer with 10.5 % of Sn but underetched with a SF_6 plasma.	2012.11262	FIG3bis.jpg
0.383255	bc7a49ef05664b069611e2949801a0dd	a 3D Scanning Electron Microscopy image of a GeSn microdisk b Bright Field (BF) TEM image of a GeSn microdisk. The lamella was milled by Focused Ion Beam through a microdisk with a 10 diameter c Electromagnetic field \|E_z\|/\|E_z-inc\| of a TM whispering gallery mode (L=26, N=1) with an energy of 0.5352 eV in the Oxz plane. d Same as c) in the Oxy plane, the dotted circle shows the edges of the Ge pillar. e 2D map of Raman shift measured on a 8 diameter microdisk. The Raman position of TO vibration mode of the as-grown layer is taken as a reference.	2012.11262	Fig1n.jpg
0.429087	a169cfb5b27a4692b24ea49db9ec50bf	a-bottom, PL spectra from as-grown GeSn layers under 1 cw pump power (0.9 □ power density), except for the x_Sn=7 % layer pumped with a 14 pump power (12 □ power density) and a signal multiplied by a factor of 5. Middle, PL spectra from layers patterned into 4 diameter microdisks, under cw excitation of 1 . Top, same as middle but excitation is turned into a pulsed mode with a 44 peak power, (38 □ power density). Spectra has been offset and divided by a factor of 500. b Experimental values of fundamental band gap obtained from PL spectra under cw pumping of microdisks. Orange lines correspond to modeled direct and indirect band gap energies E_GΓ and E_GL of relaxed GeSn layers with bowing parameters b_L=0.89 eV and b_Γ=2.77 eV. The arrow shows the indirect-direct bandgap crossover. c same as (b) but for as-grown layers assuming 30 % of residual compressive strain due to the lattice parameter mismatch (between GeSn and Ge).	2012.11262	Fig2n.jpg
0.448253	5efd39f0b14242c2bca55db16241d2ee	a Carrier density threshold to reach transparency as a function of the directness parameter Δ E_L-Γ at 20 K in strain-free material. The black circles show the corresponding relaxed GeSn microdisks with the Sn contents probed in this work b Review of low temperature lasing thresholds in GeSn devices as a function of the Sn content in the active media (FP stands for Fabry-Perot). We show only threshold below 25 K. Relaxed GeSn disk cavities are shown as blue circles,<cit.> while results obtained with compressive GeSn ridge waveguides appear as pink diamonds <cit.>, including complex multilayers. More data and details on experimental conditions used for previous works can be found in the comparison table reported in ref.<cit.>. Black, grey and light blue squares are current values. The dashed line is drawn to show the trend.	2012.11262	Fig5n.jpg
0.425196	79c5372b12fb448d9678879a195a0e8a	a L-L curves at temperatures from 25 to 105, for GeSn microdisks with a 4 diameter and Sn contents x_Sn of 7 %, 8.1 % and 10.5 %. No lasing is observed at temperatures above 100. b Emission spectra above lasing threshold for temperatures varying from 25 to 105.c Integrated PL signal from microdisks with different Sn contents as a function of temperature measured under cw excitation below lasing threshold.d Calculated electron population ratio in the Γ valley as a function of temperature for different conduction band barrier energies Δ E_L-Γ=E_L-E_Γ.e Maximal lasing temperature as a function of investigated % of Sn and corresponding directness parameter.	2012.11262	Figure4new.jpg
0.429428	c8fb0a33736d4e018ac667689fb65622	Typical trajectory of x(t) given by Eq. (<ref>) with D_+=10 (blue regions), and D_-=0 (red regions). For this trajectory exponential waiting times with ⟨τ⟩_+ =1 and ⟨τ⟩_- =5 were used.	2012.11265	Fig1.jpg
0.467007	5e170cbbbd674910aa4b4a49e475420c	Alternating process for the diffusivity, starting from the state `+' and N=2k+1. For the case of equilibrium initial conditions exposed in section <ref>, for N=1 τ_1 works as the forward recurrence time with PDF Eq. (<ref>).	2012.11265	Fig2.jpg
0.50121	e15a22b9452f4593a8e9c6d07d79f34f	For a system with, ⟨τ⟩_+=1 and ⟨τ⟩_-=5, P(z,t) in semi-log scale, with z=x/√(t). For short times t=0.1 (red circles) and t=0.5 (blue crosses) P(z,t) is represented by Eq. (<ref>) ( black solid line) with a tent like shape. For large times t=20 (orange circles) and t=30 (green diamonds) P(z,t) converges to the Gaussian statistics Eq. (<ref>) (magenta solid line). In all the cases D_+=10 and D_-=0 were used. Compared with Figure <ref> in this case the Gaussian curve is above the tent curve, contrary to the case with equal mean waiting times. This is because the coefficient of the Gaussian curve Eq. (<ref>) is bigger compared with the weight of the delta peak in Eq. (<ref>). In Figure <ref> we have the opposite, the weight of the corresponding delta function in Eq.(<ref>) is bigger compared with the Gaussian Eq. (<ref>).	2012.11265	Fig7.jpg
0.412161	adc9bb9cc90d41f5914107b59bb4a90a	Distribution of diffusion coefficients P(D) obtained via TAMSD analysis of simulated trajectories of a two state system with D_+=10, D_-=0 and exponentially distributed waiting times. From the linear plots of the TAMSD versus the lag time estimates of D were extracted. We show two cases, the first for a system with same mean waiting times ⟨τ⟩_+ = ⟨τ⟩_-= ⟨τ⟩ =1 (red boxes). And the PDF of D for a system with different mean waiting times with ⟨τ⟩_+ =1 and ⟨τ⟩_-=5 is also shown (blue boxes). For the system with same mean waiting times the average diffusivity found in the simulations is ⟨ D ⟩=4.98 and for the case of different mean waiting times we have ⟨ D ⟩ = 1.69. In both cases we used t=1000 and 1000 trajectories.	2012.11265	Fig8.jpg
0.526269	28da972911ac47a9b8a2d05451212e7f	Distribution of displacements P(x,t) obtained by simulations of a two state system with D_+> D_->0 and gamma distributed waiting times τ∼ Gamma(3,1) at D_+ and τ∼ Gamma(6,1) at D_- following Eq. (<ref>). We compare with Eq. (<ref>) (solid lines) with t=0.5, ⟨τ⟩_+=3, ⟨τ⟩_-=6, D_+=10. For D_-=0.1 (red triangles), D_-=5 (cyan squares) and D_-=9 (magenta circles). Exponential like decaying is present at small values for x, when D_+=10>>D_-=0.1 (red solid line). In the cases when D_-⟶ D_- (cyan and magenta solid lines) P(x,t) follows a full Gaussian distribution.	2012.11265	FigA1.jpg
0.504619	8f87379c69634e51bc3a4f27d3dbcf09	Distribution of displacements P(x,t) in semi-log scale, obtained from simulations, of a two state system with uniform and gamma distributed waiting times within the short time limit and displaying the whole span of x. P(x,t) for uniformly distributed waiting times is shown in red triangles. And the case of gamma distributed waiting times is shown in blue squares. We employed the same set of parameters as those used in Fig. <ref> at the left panel. Both cases fit with Eq. (<ref>) (red and blue solid lines).	2012.11265	FigA2.jpg
0.508386	fa1f7b94ab444732aeb429e551113d02	Fragment of Ch(, 𝔾^8(a_0, a_8)) (please, print in colors!)	2012.11269	dav-chase.jpg
0.466694	662f490f86a34d8c81d5b11f323b35db	Poncelet 3-periodic families in the various concentric ellipse pairs studied in the article. Properties and loci of the confocal pair (elliptic billiard) were studied in <cit.>. For each family the particular triangle center which is stationary is indicated.	2012.11270	0100_all_3b.jpg
0.412162	d475d71ebff442aca3ed91732d5cea59	Diagram of transformations that take one 3-periodic family into another. The families are specified in each box while the transformations label the arrows. The second (resp. third) line in each box lists the stationary point(s) (resp. main invariants) in the family.	2012.11270	0160_transf_overview.jpg
0.488299	6982bf4be2ed477e8b92718774b7b774	Defect in Poincaé _3	2012.11274	Wilsonline.jpg
0.422626	a6513e2b7d6c4028a98e059092afefed	Riemann sphere	2012.11274	nonreplica.jpg
0.479954	fb81694a556241a8839ef7b0663d90f2	Source at the boundary.	2012.11280	gamma_boundary.jpg
0.469732	e72c3b2457084d1097446139fb25d691	Interior source.	2012.11280	gamma_interior.jpg
0.495247	9cd1e165e1774670b2deb8a192b4219b	Inverse solution, α = 0.3α̅.	2012.11280	inv_03.jpg
0.52248	abb5fca6453542ab9914e0c25c505237	Inverse solution, α = 0.3α̅.	2012.11280	inv_03.jpg
0.566466	ebd9a46e3bee43b6bee42ef2c1884733	Inverse solution, α = 0.3α̅.	2012.11280	inv_03.jpg
0.55768	aa5ec9c01faa4c09a2a4e4d92979e561	Inverse solution, α = 3α̅.	2012.11280	inv_3.jpg
0.572433	5ee0b6bb2f6c49d3b8839cddbb197ae7	Inverse solution, α = 3α̅.	2012.11280	inv_3.jpg
0.577355	8f4c48f363654f4b88f87df68a8f8574	Inverse solution, α = 3α̅.	2012.11280	inv_3.jpg
0.499945	936da36af5df4794b40470e955d70394	Re-scaling of the solution depicted in Figure <ref>a).	2012.11280	inv_3_scaled.jpg
0.540498	b2fe06951e4c4f9eaaf39608e804a7de	Re-scaling of the solution depicted in Figure <ref>a).	2012.11280	inv_3_scaled.jpg
0.56029	e160d475daf34269b0ad43d19ddc77bd	Re-scaling of the solution depicted in Figure <ref>a).	2012.11280	inv_3_scaled.jpg
0.417103	544d48117bc54b9fad571fa388764ba8	Inverse solution.	2012.11280	inverse.jpg
0.61733	cab8b52ce24745a696b9296bd8c2af15	Inverse solution.	2012.11280	inverse_boundary.jpg
0.427865	112165e160cf4f3dafb959686b0fc3d8	Inverse solution.	2012.11280	inverse_eightsource.jpg
0.481774	1022c4b4d9074c118ef5c496b7fbdd8a	Inverse solution.	2012.11280	inverse_foursource.jpg
0.463099	343e144b4b004b7586721672f38290ab	Inverse solution.	2012.11280	inverse_interior.jpg
0.467379	b80ebbe8b39d4edd92b876658114bb6a	Inverse solution.	2012.11280	inverse_source.jpg
0.462634	1c83461df62b41ac855474daee7e2fc8	Tikhonov. α = 10^-6.	2012.11280	inverse_sourceAlpha0000001.jpg
0.434832	57e75e055cb341bb85ff51389f0345da	Truncated SVD. α = 10^-6.	2012.11280	inverse_sourceAlpha0000001k.jpg
0.42099	f610bfbeac5c47908ba36921c7f191d1	Tikhonov. α = 10^-4.	2012.11280	inverse_sourceAlpha00001.jpg
0.479344	87bb3176e1d14722a7261424804ad215	Truncated SVD. α = 10^-4.	2012.11280	inverse_sourceAlpha00001k.jpg
0.502324	f34ffe7f5c8d46cd9805cc45e7a1c24f	Tikhonov. α = 10^-2.	2012.11280	inverse_sourceAlpha001.jpg
0.496589	9d555e536ca24c85823b45829cb7668b	Truncated SVD. α = 10^-2.	2012.11280	inverse_sourceAlpha001k.jpg
0.426661	fabde517c85c43aaafd2c41f1654154c	Inverse solution.	2012.11280	inverse_twosource.jpg
0.487916	a7fed92e71744262bb3e9a6332e239c8	k=15, α = 10^-2.	2012.11280	k15a100.jpg
0.512656	d83048ce6abd4dbb8c66cc0ca85713fe	k=15, α = 10^-3.	2012.11280	k15a1000.jpg
0.550797	40cb9f0124a047ea811be5d8f08b9b5c	k=15, α = 10^-4.	2012.11280	k15a10000.jpg
0.464048	d2f3d8a6c11841578619d3b94a7c369a	k=3, α = 10^-2.	2012.11280	k3a100.jpg
0.465318	e68447f91fb04ada88962bafc118d54b	k=3, α = 10^-3.	2012.11280	k3a1000.jpg
0.541553	c932822ba0f34baf8914774b805e12e3	k=3, α = 10^-4.	2012.11280	k3a10000.jpg
0.535079	c356fd1866394e5c946ea91147e241e8	k=5, α = 10^-2.	2012.11280	k5a100.jpg
0.426469	995dd03b615f483aa17dce15ad6aa30e	k=5, α = 10^-3.	2012.11280	k5a1000.jpg
0.439425	98b1d400810f4946bdd03ac10f3a93d4	Solution of (<ref>) with ζ = 10^-4.	2012.11280	tikh.jpg
0.57456	09fee4a8104443508f6273938171b8c1	Example 2, true source.	2012.11280	true.jpg
0.464358	0624f2e5c5f04892afc768b733ecd52e	Example 2, true source.	2012.11280	true.jpg
0.540997	46c4789a204c48ed9b6c70f5c0260321	True source.	2012.11280	true_boundary.jpg
0.532658	9cbcf0f9d059460cbe61bb379918aed2	True sources.	2012.11280	true_eightsource.jpg
0.554223	a233adb16b564ecaa7a8befdd0680baf	True sources.	2012.11280	true_foursource.jpg
0.485527	854a011e282f45e290676ac041c82a5d	True source.	2012.11280	true_interior.jpg
0.414676	feaf10781e134c3a8b608f719c2d0763	True source.	2012.11280	true_source.jpg
0.553396	2fa1f25b367741b6bb42e3383115bb2c	True source.	2012.11280	true_sourceR.jpg
0.58299	0a75de8587684b789c65adf03fab0c81	True sources.	2012.11280	true_twosource.jpg
0.492258	9e0a644756fc4035b10e1ba777beffd8	k = 7.	2012.11280	weightsK7.jpg
0.45418	7887363a930c4f37af866a026c7b8baf	k = 70.	2012.11280	weightsK70.jpg
0.522236	4e4eec17b68545abbd3cc53554c98ce5	Bruhat interval W/W_P_1 for E_6	2012.11290	e6.jpg
0.483109	9942ea6965a043d5b6c02c410c79d0c2	Bruhat interval W/W_P_7 for E_7	2012.11290	e7.jpg
0.444363	ad15fffee26d4450b801da0427bfe1e1	Best molecule generated on the different tasks. Tasks from top to bottom: QED, penalized logP, Similarity. Left: Best molecule generated by the curious agent. Right: Best molecule generated by a non curious agent. Note that no chemical stability or synthesizabitly filters were employed on this work, so the structures may look a bit strange to organic chemists.	2012.11293	6mols.jpg
0.426469	00866619b35c43649f67e3d6f760d04d	An illustration of curiosity: The agent generates molecules and gives them to the property prediction network. Initially, the predictions are wrong everywhere in the chemical space, but over time the network learns to predict the properties of molecules it has already encountered well. By using the prediction error as an intrinsic reward, the agent is incentivised to move to regions, it has not yet explored, because the prediction network will make more errors there. The combination of the intrinsic and extrinsic rewards is then used as feedback to train the agent.	2012.11293	drawing-1_with_arrows.jpg
0.418756	f1db71e5a45a4fd6ba0d9eed44f55231	An example case showing the impact of spatial interpolation. (a) is the source image and (d) is the corresponding output feature map of ResNet-18's stage. (b) is obtained by directly 4 × bilinear up-sampling and then 4 × bilinear down-sampling on (d), while (e) is down-sampled first and then up-sampled. (c,f) are the difference maps of (b,d) and (e,d), respectively. It can be seen that even the simplest spatial interpolation operations can cause obvious differences.	2012.11294	interpolation.jpg

0.463401

2cf1491977a04cb584054021db1e9f1a

Structure of QRNN. (a) Structure of QRNN for a single time step. (b) QRNN in the training phase. (c) QRNN in the prediction phase. Note that the obtained prediction x_T is used as the input to the next time step.

2012.11242

QRNN_crop.jpg

0.433205

805e01cf06d64810896b01a267b1605c

(a) Schematic picture of a basic RNN. Blue circles represent elements of vectors x_t, h_t and x_t. The matrices V_in, V_out and W connecting layers are the parameters to be optimized during training of the model. The bias vectors b and c are abbreviated for simplicity. (b) Unfolded version of (a). (c) RNN for time-series prediction.

2012.11242

RNN_combined_new_crop.jpg

0.453777

07a5f9f67e444a6e9134ac76e5584845

The evolution part of the circuit utilized in the numerical simulations. α_i^(d), β_i^(d), and γ_i^(d) are rotation angles and act as independent parameters of the circuit, i.e., θ consists of {α_i^(d), β_i^(d), γ_i^(d)}. d=1,⋯, D is the layer index.

2012.11242

rotation_gate.jpg

0.439153

58b5ff5b19444b3b949995bb2f299c72

Results of numerical simulation of training the QRNN to predict (a) cosine wave, (b) triangular wave, and (c) three-spin Lindblad dynamics. The data points on the left of the vertical grey dashed line are used as the training dataset. The solid lines represent the function which the QRNN is trained to model. The parts of the data used in the training phase and the prediction phase are depicted as the black and orange lines, respectively.

2012.11242

simple-simulation_crop.jpg

0.429756

110d8e694a3d4d08910fbae8d4ac2884

Relation between the evolution time τ and the prediction error evaluated in terms of the mean squared error.

2012.11242

tau_vs_mse.jpg

0.519605

c17dc3b0b84948a6b4a0f0d3fb174d88

Top panel: The distribution of k^* for the LHCb acceptance with and without applying the efficiencies. Bottom panel: the ratio of -hadron from the same parent (SP) to all pairs.

2012.11250

hCKstarJpsiHadron_paper_1.jpg

0.565064

3fa3c4ab88dc411c8fa7fa9b8e719480

Top panel: The distribution of k^* for the STAR acceptance for J/ψ→ e^+e^- reconstruction with and without applying the efficiencies. Bottom panel: the ratio of -hadron from the same parent (SP) to all pairs.

2012.11250

hCKstarJpsiHadron_star.jpg

0.423184

85aab23eae2f4c5da83cd5956c9ca29b

The shuffled shot noises under mass and optimal weighting schemes (dotted lines). For comparison, the expected shot noises under mass and optimal weighting schemes (solid lines) are also presented.

2012.11251

shuffled_shot_noise.jpg

0.459529

d3e75f9ecdd04b70bc01577d69bdbe0f

(Left) This figure shows the “unfolded” multi-layered context network with increasing dimensionality that captures larger and more influencing contexts. (Right) Hierarchical context learning framework including geometric and semantic contexts. In this diagram, we only show four semantic neighbors for each image as an example (better to zoom the PDF version).

2012.11253

figure2.jpg

0.360482

f20f4f17dab64289a81af75461920485

The two DW configurations in 1x1x16 supercells used in order to determine the crystallographic plane where the oxygen vacancies are placed in the calculations. In a) BaO centered TT wall with vacancies, in b) TiO_2 centered TT wall with vacancies. The forces of both structures have been relaxed.

2012.11254

batenticen.jpg

0.582117

ac75320f24554a14bb3837cbeaf3dbfd

Properties of the unrelaxed CDWs in a 1×1×16 supercell. Top: schematic atomic structure with exagerated displacements. The blue arrows indicate the direction of polarization; green, blue and red spheres indicate Ba, Ti and O atoms, respectively. The dots indicate positions of Ti atoms in the individual unit cells. Second from top: polarization profile of the supercell. Second from bottom: bound ρ_b and free ρ_f charge density. Bottom: electrostatic potential energy v(z) averaged over the plane orthogonal to z and total charge density ρ obtained from a sliding window average.

2012.11254

fullfig1.jpg

0.393068

855459ad478c4842bdb0a0de144ef1cd

PDOS at the HH wall of the 1x1x16 supercell containing a HH and TT wall. Ti t_2g states, Ti non-t_2g states and the rest are differentiated with different colors. The Fermi level is set as the origin for the energies.

2012.11254

hhti.jpg

0.401585

457d73ba8e7444fb8b15df668463b89d

PDOS at the HH wall of the relaxed 1x1x16 supercell containing a HH and TT wall and an oxygen vacancy at the latter. Ti t_2g states, Ti non-t_2g states and the rest are differentiated with different colors. The Fermi level is set as the origin for the energies.

2012.11254

hhtivac.jpg

0.482519

28dedee96835440cb8c19247d1f5f6c3

Left: projected band structure of the unrelaxed 1×1×16 supercell with no vacancies. Right, the projected density of states resolved in individual unit cells.

2012.11254

novacband.jpg

0.39942

1ba6fdd59e5442778541b4c88cc1e7ef

Density of states of the relaxed 2x2x8 supercell with a HH and TT wall, with two vacancies at the latter, projected into the atomic orbitals of TiO_2 planes. The Fermi level is set as the origin for the energies.

2012.11254

pdostiolayers.jpg

0.502139

a54eb5b8d33f4440915b91f79f0ead8d

At the top, the reference 2×2×8 tetragonal BaTiO_3 supercell with one vacancy at a BaO plane. Below, the energy cost profiles for the i^th vacancy at α position (see text). On the top-right, the section of the supercell corresponding to the central BaO plane where static vacancies are set. Second from the bottom, the polarization profiles of the relaxed supercell for i= 0, 1, 2 and 3. At the bottom, bound and free charge density profiles for the relaxed 2×2×8 supercell with two vacancies at the central BaO plane.

2012.11254

vacprofiles.jpg

0.391881

aa39cb1106ce4a28abc0b4589fb20699

Mean and variance of difference rewards for a set of samples under different noise profiles.

2012.11258

Noise.jpg

0.376894

dcacbcd76262486eb3d46a43bbf3be8a

Mean and variance of difference rewards for a set of samples under different noise profiles.

2012.11258

Noise_Large.jpg

0.421188

9572e52f824b4893b04d5c64ca9b3ee9

Normalized mean prediction error and standard deviation for Dr.ReinforceR reward network R_ψ and COMA critic Q_ω on the on-policy dataset (first row) and the off-policy dataset (second row), for the two environments.

2012.11258

Statistics.jpg

0.45559

0b14aa1ac92a4ca599477754fe4e07f3

Illustration of static and dynamic adaptation of DOS under external perturbation by adding a small amount of electron dN. The effect of perturbing μ by a small amout dμ could be understood in a similar fashion. Note that the dynamic adaptation is closer to reality.

2012.11259

DOS-change-by-dN.jpg

0.414593

1003789162334b0d8d7fd54a4dc4330c

Demonstration of the correlation between the Fermi softness (S_F) and the reactivity of various top sites (tpf, tpc and tpl, as shown in the inset) on Pt nanoparticles as represented by the CO adsorption strength (upper panel). For comparison, correlation with the d-band center is also presented (lower panel). tpl79: top site on Pt atom in Pt nanoparticle made up of 79 Pt atoms.

2012.11259

E_COads_vs_sf_PtNP.jpg

0.379813

39c455e6bb6f449c8e97af37c117a00b

Correlation between the global Fermi softness (S_F) and the surface reactivity as indicated by the oxygen adsorption strength on Pd-derived surfaces (see the supplementary material of reference <cit.> for details). For comparison, the correlation of the d-band center for the same systems is also shown.

2012.11259

Eads-SF-vs-Eads-Ed.jpg

0.420581

2f2aea20202b4a9b88965979013de682

Comparison of correlations between oxygen adsorption energy (Δ E_O) on Pt-related surfaces and S_F of surface Pt atom at two electronic temperatures: 0 eV (A, no spreading in dFDD) and 0.4 eV (B). How well Δ E_O is correlated with S_F is measured by the Spearman's r and larger absolute value indicates better correlation.

2012.11259

Eads-vs-sf-zero-and-finite-Tel.jpg

0.487747

6a28702d3d6a43f78af9b8f6f2ea004b

Plot of A) Fermi-Dirac distribution (FDD for short) function and B) its derivative (dFDD for short) at three representative electronic temperatures: 0.01 eV, 0.4 eV and 1 eV. One could also see from panel A the approximate equivalence of the chemical potential and the energy associated with Fermi level at these diverse T_el's, as indicated by the dashed vertical lines.

2012.11259

FDD-and-dFDD-Tel.jpg

0.407939

117c9cfbfa2242d9951f9d4b06b63e70

Spatially resolved chemical reactivity of a 1D MoS_2 edge. A) The geometrical structure of MoS_2, with three typical local sulfur environments (S#1, S#2, and S#3) and two typical molybdenum atom environments (Mo#1 and Mo#2). B) The projected density of states (g(ε)) of specific sulfur and molybdenum atoms. C) A 3D perspective of the local Fermi softness s_F(r), as represented by some isosurface (blue). D) The s_F(r) is plotted as projection onto two planes normal to the x direction: plane a intersects two S#1 and one Mo#2 atoms, and plane b contains one Mo#1, two S#2, and one S#3 atoms. E) Reaction energy profiles for the reaction 2H_ad→H_2 at two kinds of bridge sites (br-x or br-y, see the inset) on the S#1 dimer edge, obtained from nudged elastic band calculations.

2012.11259

SF-MoS2.jpg

0.476401

50f87e0dbf7b4763bdcd9676d8dd2cbb

Graphical illustration of S_F, defined as a weighted sum of the density of states ∫ g(ε)w(ε)dε, where the weight function w(ε) is chosen as the derivative of the FermiâDirac distribution function at non-zero electronic temperature.

2012.11259

Sf-concept.jpg

0.487786

8475e652d7ac4ae186c0359aca73475c

Schematic illustration of stabilization of total single electron energies upon shifting from ε_d (A) to a higher value (ε_d', B) of d-band center of transitional metal surface when interacting with an adsorbate.

2012.11259

demo-d-band-model.jpg

0.424796

6fb155c4df0442b1b955f6547ac617d9

Orbital interactions of surface band and three types of adsorbate orbital: half-occupied HOMO (A), doubly occupied HOMO (B) and empty LUMO (C). The five columns respectively correspond to five different states of surface band, being at Fermi level, slightly below ε_F, well below ε_F, slightly above ε_F and well above ε_F. Red bar: magnitude of energy gain after orbital interaction; Green bar: magnitude of energy penalty upon orbital interaction; Solid upward (downward) pointing arrow: electron with spin up (down) occupying some electronic state; Gray arrow: electron occupying some “intermediate” electronic state that is about to fill the unoccupied state at the Fermi level of surface band.

2012.11259

orbital-interactions.jpg

0.392249

85be5f44abaf4970844a23a1033c7946

Charging results of EVs controlled by PPC with tuning parameters β=2× 10^3 (top), 4× 10^3 (mid) and 6× 10^3 (bot) for selected days (with no less than 30 charging sessions, i.e., N≥ 30) from Dec. 2, 2019 to Jan. 1, 2020. Each bar represents a charging session.

2012.11261

battery.jpg

0.551064

2c5b7364fb454381ad72921192670196

Substation charging rates generated by the PPC (orange) in the closed-loop control shown in Algorithm <ref>, together with the MPC generated (blue) and global optimal (dashed black) charging rates.

2012.11261

charging_caltech.jpg

0.401171

f47e5a5b5d284cd5bc906120621dac01

Cost-energy curves for the offline optimization in (<ref>)-(<ref>) (for the example in Section <ref>), MPC (defined in (<ref>)-(<ref>)) and PPC (introduced in Section <ref>).

2012.11261

costs_energy.jpg

0.492933

9eddfa5e51154bb8a6b5a9a37bd0f2fb

Average rewards (defined in (<ref>)) in the training stage with a tuning parameter β=6× 10^3. Shadow region measures the variance.

2012.11261

learning_curve.jpg

0.466523

bcff564a23ab4d31aec83b91f3ffaaa1

Feasible trajectories of power signals and the computed maximum entropy feedback in Example <ref>.

2012.11261

toy_example.jpg

0.464716

5dcbc1069d6b4eb7bc66dd8b6ce1e601

Trade-offs of cost and charging performance. The dashed curve in the left figure corresponds to offline optimal cost. The tested days are selected (with no less than 30 charging sessions, i.e., N≥ 30) from Dec. 2, 2019 to Jan. 1, 2020.

2012.11261

tuning.jpg

0.41819

767cb8a5e7474ecc8d95d7f45e3b2a16

(Bottom) Pulsed peak power dependence of emission spectra for 4 diameter microdisks with Sn contents between 7 % to 10.5 % with corresponding L-L curves at 25 (Top). The continuous curves in the figures on top are calculated curves using the laser rate equations with the indicated thresholds and beta factors. The insets show typical spectra above lasing threshold.

2012.11262

FIG3.jpg

0.412057

edc2b4f786c14e37ba096bd6d3c980a7

a L-L curve obtained at 25 K from a 10 diameter microdisk with 10.5 % of Sn underetched with a CF_4 plasma. The dashed lines are drawn to guide the eyes. The inset shows a SEM image of such microdisks b L-L curve obtained at 25 K from a 8 diameter microdisk fabricated from the same epitaxial layer with 10.5 % of Sn but underetched with a SF_6 plasma.

2012.11262

FIG3bis.jpg

0.383255

bc7a49ef05664b069611e2949801a0dd

a 3D Scanning Electron Microscopy image of a GeSn microdisk b Bright Field (BF) TEM image of a GeSn microdisk. The lamella was milled by Focused Ion Beam through a microdisk with a 10 diameter c Electromagnetic field |E_z|/|E_z-inc| of a TM whispering gallery mode (L=26, N=1) with an energy of 0.5352 eV in the Oxz plane. d Same as c) in the Oxy plane, the dotted circle shows the edges of the Ge pillar. e 2D map of Raman shift measured on a 8 diameter microdisk. The Raman position of TO vibration mode of the as-grown layer is taken as a reference.

2012.11262

Fig1n.jpg

0.429087

a169cfb5b27a4692b24ea49db9ec50bf

a-bottom, PL spectra from as-grown GeSn layers under 1 cw pump power (0.9 □ power density), except for the x_Sn=7 % layer pumped with a 14 pump power (12 □ power density) and a signal multiplied by a factor of 5. Middle, PL spectra from layers patterned into 4 diameter microdisks, under cw excitation of 1 . Top, same as middle but excitation is turned into a pulsed mode with a 44 peak power, (38 □ power density). Spectra has been offset and divided by a factor of 500. b Experimental values of fundamental band gap obtained from PL spectra under cw pumping of microdisks. Orange lines correspond to modeled direct and indirect band gap energies E_GΓ and E_GL of relaxed GeSn layers with bowing parameters b_L=0.89 eV and b_Γ=2.77 eV. The arrow shows the indirect-direct bandgap crossover. c same as (b) but for as-grown layers assuming 30 % of residual compressive strain due to the lattice parameter mismatch (between GeSn and Ge).

2012.11262

Fig2n.jpg

0.448253

5efd39f0b14242c2bca55db16241d2ee

a Carrier density threshold to reach transparency as a function of the directness parameter Δ E_L-Γ at 20 K in strain-free material. The black circles show the corresponding relaxed GeSn microdisks with the Sn contents probed in this work b Review of low temperature lasing thresholds in GeSn devices as a function of the Sn content in the active media (FP stands for Fabry-Perot). We show only threshold below 25 K. Relaxed GeSn disk cavities are shown as blue circles,<cit.> while results obtained with compressive GeSn ridge waveguides appear as pink diamonds <cit.>, including complex multilayers. More data and details on experimental conditions used for previous works can be found in the comparison table reported in ref.<cit.>. Black, grey and light blue squares are current values. The dashed line is drawn to show the trend.

2012.11262

Fig5n.jpg

0.425196

79c5372b12fb448d9678879a195a0e8a

a L-L curves at temperatures from 25 to 105, for GeSn microdisks with a 4 diameter and Sn contents x_Sn of 7 %, 8.1 % and 10.5 %. No lasing is observed at temperatures above 100. b Emission spectra above lasing threshold for temperatures varying from 25 to 105.c Integrated PL signal from microdisks with different Sn contents as a function of temperature measured under cw excitation below lasing threshold.d Calculated electron population ratio in the Γ valley as a function of temperature for different conduction band barrier energies Δ E_L-Γ=E_L-E_Γ.e Maximal lasing temperature as a function of investigated % of Sn and corresponding directness parameter.

2012.11262

Figure4new.jpg

0.429428

c8fb0a33736d4e018ac667689fb65622

Typical trajectory of x(t) given by Eq. (<ref>) with D_+=10 (blue regions), and D_-=0 (red regions). For this trajectory exponential waiting times with ⟨τ⟩_+ =1 and ⟨τ⟩_- =5 were used.

2012.11265

Fig1.jpg

0.467007

5e170cbbbd674910aa4b4a49e475420c

Alternating process for the diffusivity, starting from the state `+' and N=2k+1. For the case of equilibrium initial conditions exposed in section <ref>, for N=1 τ_1 works as the forward recurrence time with PDF Eq. (<ref>).

2012.11265

Fig2.jpg

0.50121

e15a22b9452f4593a8e9c6d07d79f34f

For a system with, ⟨τ⟩_+=1 and ⟨τ⟩_-=5, P(z,t) in semi-log scale, with z=x/√(t). For short times t=0.1 (red circles) and t=0.5 (blue crosses) P(z,t) is represented by Eq. (<ref>) ( black solid line) with a tent like shape. For large times t=20 (orange circles) and t=30 (green diamonds) P(z,t) converges to the Gaussian statistics Eq. (<ref>) (magenta solid line). In all the cases D_+=10 and D_-=0 were used. Compared with Figure <ref> in this case the Gaussian curve is above the tent curve, contrary to the case with equal mean waiting times. This is because the coefficient of the Gaussian curve Eq. (<ref>) is bigger compared with the weight of the delta peak in Eq. (<ref>). In Figure <ref> we have the opposite, the weight of the corresponding delta function in Eq.(<ref>) is bigger compared with the Gaussian Eq. (<ref>).

2012.11265

Fig7.jpg

0.412161

adc9bb9cc90d41f5914107b59bb4a90a

Distribution of diffusion coefficients P(D) obtained via TAMSD analysis of simulated trajectories of a two state system with D_+=10, D_-=0 and exponentially distributed waiting times. From the linear plots of the TAMSD versus the lag time estimates of D were extracted. We show two cases, the first for a system with same mean waiting times ⟨τ⟩_+ = ⟨τ⟩_-= ⟨τ⟩ =1 (red boxes). And the PDF of D for a system with different mean waiting times with ⟨τ⟩_+ =1 and ⟨τ⟩_-=5 is also shown (blue boxes). For the system with same mean waiting times the average diffusivity found in the simulations is ⟨ D ⟩=4.98 and for the case of different mean waiting times we have ⟨ D ⟩ = 1.69. In both cases we used t=1000 and 1000 trajectories.

2012.11265

Fig8.jpg

0.526269

28da972911ac47a9b8a2d05451212e7f

Distribution of displacements P(x,t) obtained by simulations of a two state system with D_+> D_->0 and gamma distributed waiting times τ∼ Gamma(3,1) at D_+ and τ∼ Gamma(6,1) at D_- following Eq. (<ref>). We compare with Eq. (<ref>) (solid lines) with t=0.5, ⟨τ⟩_+=3, ⟨τ⟩_-=6, D_+=10. For D_-=0.1 (red triangles), D_-=5 (cyan squares) and D_-=9 (magenta circles). Exponential like decaying is present at small values for x, when D_+=10>>D_-=0.1 (red solid line). In the cases when D_-⟶ D_- (cyan and magenta solid lines) P(x,t) follows a full Gaussian distribution.

2012.11265

FigA1.jpg

0.504619

8f87379c69634e51bc3a4f27d3dbcf09

Distribution of displacements P(x,t) in semi-log scale, obtained from simulations, of a two state system with uniform and gamma distributed waiting times within the short time limit and displaying the whole span of x. P(x,t) for uniformly distributed waiting times is shown in red triangles. And the case of gamma distributed waiting times is shown in blue squares. We employed the same set of parameters as those used in Fig. <ref> at the left panel. Both cases fit with Eq. (<ref>) (red and blue solid lines).

2012.11265

FigA2.jpg

0.508386

fa1f7b94ab444732aeb429e551113d02

Fragment of Ch(, 𝔾^8(a_0, a_8)) (please, print in colors!)

2012.11269

dav-chase.jpg

0.466694

662f490f86a34d8c81d5b11f323b35db

Poncelet 3-periodic families in the various concentric ellipse pairs studied in the article. Properties and loci of the confocal pair (elliptic billiard) were studied in <cit.>. For each family the particular triangle center which is stationary is indicated.

2012.11270

0100_all_3b.jpg

0.412162

d475d71ebff442aca3ed91732d5cea59

Diagram of transformations that take one 3-periodic family into another. The families are specified in each box while the transformations label the arrows. The second (resp. third) line in each box lists the stationary point(s) (resp. main invariants) in the family.

2012.11270

0160_transf_overview.jpg

0.488299

6982bf4be2ed477e8b92718774b7b774

Defect in Poincaé _3

2012.11274

Wilsonline.jpg

0.422626

a6513e2b7d6c4028a98e059092afefed

Riemann sphere

2012.11274

nonreplica.jpg

0.479954

fb81694a556241a8839ef7b0663d90f2

Source at the boundary.

2012.11280

gamma_boundary.jpg

0.469732

e72c3b2457084d1097446139fb25d691

Interior source.

2012.11280

gamma_interior.jpg

0.495247

9cd1e165e1774670b2deb8a192b4219b

Inverse solution, α = 0.3α̅.

2012.11280

inv_03.jpg

0.52248

abb5fca6453542ab9914e0c25c505237

Inverse solution, α = 0.3α̅.

2012.11280

inv_03.jpg

0.566466

ebd9a46e3bee43b6bee42ef2c1884733

Inverse solution, α = 0.3α̅.

2012.11280

inv_03.jpg

0.55768

aa5ec9c01faa4c09a2a4e4d92979e561

Inverse solution, α = 3α̅.

2012.11280

inv_3.jpg

0.572433

5ee0b6bb2f6c49d3b8839cddbb197ae7

Inverse solution, α = 3α̅.

2012.11280

inv_3.jpg

0.577355

8f4c48f363654f4b88f87df68a8f8574

Inverse solution, α = 3α̅.

2012.11280

inv_3.jpg

0.499945

936da36af5df4794b40470e955d70394

Re-scaling of the solution depicted in Figure <ref>a).

2012.11280

inv_3_scaled.jpg

0.540498

b2fe06951e4c4f9eaaf39608e804a7de

Re-scaling of the solution depicted in Figure <ref>a).

2012.11280

inv_3_scaled.jpg

0.56029

e160d475daf34269b0ad43d19ddc77bd

Re-scaling of the solution depicted in Figure <ref>a).

2012.11280

inv_3_scaled.jpg

0.417103

544d48117bc54b9fad571fa388764ba8

Inverse solution.

2012.11280

inverse.jpg

0.61733

cab8b52ce24745a696b9296bd8c2af15

Inverse solution.

2012.11280

inverse_boundary.jpg

0.427865

112165e160cf4f3dafb959686b0fc3d8

Inverse solution.

2012.11280

inverse_eightsource.jpg

0.481774

1022c4b4d9074c118ef5c496b7fbdd8a

Inverse solution.

2012.11280

inverse_foursource.jpg

0.463099

343e144b4b004b7586721672f38290ab

Inverse solution.

2012.11280

inverse_interior.jpg

0.467379

b80ebbe8b39d4edd92b876658114bb6a

Inverse solution.

2012.11280

inverse_source.jpg

0.462634

1c83461df62b41ac855474daee7e2fc8

Tikhonov. α = 10^-6.

2012.11280

inverse_sourceAlpha0000001.jpg

0.434832

57e75e055cb341bb85ff51389f0345da

Truncated SVD. α = 10^-6.

2012.11280

inverse_sourceAlpha0000001k.jpg

0.42099

f610bfbeac5c47908ba36921c7f191d1

Tikhonov. α = 10^-4.

2012.11280

inverse_sourceAlpha00001.jpg

0.479344

87bb3176e1d14722a7261424804ad215

Truncated SVD. α = 10^-4.

2012.11280

inverse_sourceAlpha00001k.jpg

0.502324

f34ffe7f5c8d46cd9805cc45e7a1c24f

Tikhonov. α = 10^-2.

2012.11280

inverse_sourceAlpha001.jpg

0.496589

9d555e536ca24c85823b45829cb7668b

Truncated SVD. α = 10^-2.

2012.11280

inverse_sourceAlpha001k.jpg

0.426661

fabde517c85c43aaafd2c41f1654154c

Inverse solution.

2012.11280

inverse_twosource.jpg

0.487916

a7fed92e71744262bb3e9a6332e239c8

k=15, α = 10^-2.

2012.11280

k15a100.jpg

0.512656

d83048ce6abd4dbb8c66cc0ca85713fe

k=15, α = 10^-3.

2012.11280

k15a1000.jpg

0.550797

40cb9f0124a047ea811be5d8f08b9b5c

k=15, α = 10^-4.

2012.11280

k15a10000.jpg

0.464048

d2f3d8a6c11841578619d3b94a7c369a

k=3, α = 10^-2.

2012.11280

k3a100.jpg

0.465318

e68447f91fb04ada88962bafc118d54b

k=3, α = 10^-3.

2012.11280

k3a1000.jpg

0.541553

c932822ba0f34baf8914774b805e12e3

k=3, α = 10^-4.

2012.11280

k3a10000.jpg

0.535079

c356fd1866394e5c946ea91147e241e8

k=5, α = 10^-2.

2012.11280

k5a100.jpg

0.426469

995dd03b615f483aa17dce15ad6aa30e

k=5, α = 10^-3.

2012.11280

k5a1000.jpg

0.439425

98b1d400810f4946bdd03ac10f3a93d4

Solution of (<ref>) with ζ = 10^-4.

2012.11280

tikh.jpg

0.57456

09fee4a8104443508f6273938171b8c1

Example 2, true source.

2012.11280

true.jpg

0.464358

0624f2e5c5f04892afc768b733ecd52e

Example 2, true source.

2012.11280

true.jpg

0.540997

46c4789a204c48ed9b6c70f5c0260321

True source.

2012.11280

true_boundary.jpg

0.532658

9cbcf0f9d059460cbe61bb379918aed2

True sources.

2012.11280

true_eightsource.jpg

0.554223

a233adb16b564ecaa7a8befdd0680baf

True sources.

2012.11280

true_foursource.jpg

0.485527

854a011e282f45e290676ac041c82a5d

True source.

2012.11280

true_interior.jpg

0.414676

feaf10781e134c3a8b608f719c2d0763

True source.

2012.11280

true_source.jpg

0.553396

2fa1f25b367741b6bb42e3383115bb2c

True source.

2012.11280

true_sourceR.jpg

0.58299

0a75de8587684b789c65adf03fab0c81

True sources.

2012.11280

true_twosource.jpg

0.492258

9e0a644756fc4035b10e1ba777beffd8

k = 7.

2012.11280

weightsK7.jpg

0.45418

7887363a930c4f37af866a026c7b8baf

k = 70.

2012.11280

weightsK70.jpg

0.522236

4e4eec17b68545abbd3cc53554c98ce5

Bruhat interval W/W_P_1 for E_6

2012.11290

e6.jpg

0.483109

9942ea6965a043d5b6c02c410c79d0c2

Bruhat interval W/W_P_7 for E_7

2012.11290

e7.jpg

0.444363

ad15fffee26d4450b801da0427bfe1e1

Best molecule generated on the different tasks. Tasks from top to bottom: QED, penalized logP, Similarity. Left: Best molecule generated by the curious agent. Right: Best molecule generated by a non curious agent. Note that no chemical stability or synthesizabitly filters were employed on this work, so the structures may look a bit strange to organic chemists.

2012.11293

6mols.jpg

0.426469

00866619b35c43649f67e3d6f760d04d

An illustration of curiosity: The agent generates molecules and gives them to the property prediction network. Initially, the predictions are wrong everywhere in the chemical space, but over time the network learns to predict the properties of molecules it has already encountered well. By using the prediction error as an intrinsic reward, the agent is incentivised to move to regions, it has not yet explored, because the prediction network will make more errors there. The combination of the intrinsic and extrinsic rewards is then used as feedback to train the agent.

2012.11293

drawing-1_with_arrows.jpg

0.418756

f1db71e5a45a4fd6ba0d9eed44f55231

An example case showing the impact of spatial interpolation. (a) is the source image and (d) is the corresponding output feature map of ResNet-18's stage. (b) is obtained by directly 4 × bilinear up-sampling and then 4 × bilinear down-sampling on (d), while (e) is down-sampled first and then up-sampled. (c,f) are the difference maps of (b,d) and (e,d), respectively. It can be seen that even the simplest spatial interpolation operations can cause obvious differences.

2012.11294

interpolation.jpg