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
0.525179	dd90888dc3b24d36afda3e1907c2af54	(Left) Bode plot – frequency response of modulation system showing modulation amplitude for various modulation frequencies. The system is capable of delivering modulation two times the requirement. (Right) 3000μrad amplitude and 1000 Hz modulation circle recorded on the CCID75 camera using a laser beam reflected on the modulation system.	2012.11668	Modulation_frequncy_response.jpg
0.387479	5efeb0285db541d7b420e0d3603cf15e	(Left) Acquisition camera procured from Basler. (Right) Quantum efficiency more than 65% for wavelengths 0.4-0.8μm.	2012.11668	QE_acq.jpg
0.436859	703c3e6c171d43ad894044c52caf4f5f	The visible pyramid wavefront sensor: (left) double pyramid, (middle) camera triplet lens and (right) Yorick-Yao<cit.> simulated pyramid wavefront sensor pupil images on the detector.	2012.11668	VPWFS.jpg
0.449462	9b567d95365c474aa8d08df738779139	The different cases zoomed at a distance d(t)^1/2 around 𝐱. The configuration ω^(t+1/2) is depicted in black, and 𝐱+ω_,j^(t) in blue.	2012.11672	3arm-bis.jpg
0.522536	07d24280467e462f8d1c3818b102cf5c	On the left, an instance of A^_010(0,R) where the primal path is depicted by a bold path, and the dual ones by dashed paths. In the middle, an instance of A^_010(0,R). On the right, A^_010(0,R)∘ A_1(0,R).	2012.11672	3arm.jpg
0.431181	efeb45a8add34a56ac05ba5dd4219ff1	The initial and final graphs ^(0) and ^(2M+1) of the two processes are the same, but the intermediate graphs (^(t))_0<t≤ 2M (top) and (^(t))_0<t<M (bottom) are different (the figure depicts the diamond graphs). In the top process, horizontal tracks are exchanged successively, by pushing the gray rhombus from right to left; in the bottom process, the rhombus is pushed downwards, effectively exchanging vertical tracks. Throughout the two processes, we record the right-most coordinate of the union of all the clusters intersecting the base.	2012.11672	BBtilde.jpg
0.439036	17b71f98ab9a4eec861281f61dffb106	A picture of the set Ω(ω^(t+1/2),x) (in grey). The dark grey cluster is the unique cluster of the annulus crossing from inside to outside. Everything outside Λ_d(t)^1/2(x) is included in Ω(ω^(t+1/2),x). The red part is the place where we try to couple as best as possible ω^(t+1/2) and 𝐱+ω^(t)_,j(t)-T^(t)(i). In the proof of Proposition <ref>, note that when applied to ω^(t+1/2)∈ E(x) (which means that there exists a unique cluster crossing the annulus), all the conditions on the t-extrema of other nails and homotopy classes of loops are not impacted by what happens inside Ω(ω^(t+1/2),𝐱)^c. As a consequence, there is a “screening” property and the conditioning inside is simply the existence of the red arms, meaning that 𝐱 is equal to the left-most top-most vertex in the cluster that is crossing the annulus.	2012.11672	Omega.jpg
0.433029	7783c8e9663248688bdf6870a105d164	We depicted in black, dotted black, red and blue respectively the primal, dual, diamond and medial lattices. The primal configuration ω is in bold and the dual one ω^* in dashed bold. Finally, the loop configuration ω is in black.	2012.11672	configurations.jpg
0.512411	2d999090103c4bc5917f2cb67de6ae78	The quantities h(t), a(t), b(t) and d(t). One may deduce from the picture that ⌊ t/(2N)⌋=12 as there are 12 tracks of angle α stacked above level -N.	2012.11672	definitions_bdh.jpg
0.45108	a8056020d0e74a14abee779d2daf4d17	A picture of the event E_j(k,m) (except that the minimality of m is not really explicitly depicted). Compared to Figure <ref>, additional conditions are forced on the tracks t_j-m and t_j+m and what happens in between.	2012.11672	eventE.jpg
0.482448	c045d40a5afd439bb18d2bafa370c4fc	A picture of the event E_j(k). Note that we are on a torus, hence t_1-M^-=t_M^+.	2012.11672	eventEintro.jpg
0.462412	b90a0bfae84e4772966c0051c77066ee	We depicted the example of a nail having two top t-extrema.	2012.11672	eventarms.jpg
0.51034	e65f9f0a817c443aad611ee19276119f	On the left, an example of a transformation of the red points that continuously transforms a blue loop with a certain homotopy class into another blue loop with a different homotopy class. It is therefore crucial to encode homotopy classes in a coherent way, which will be done in Section <ref>. Also, on the right, a justification of why we did not choose to keep track of a simpler property of loops, namely, the inclusion between loops. In the picture, the two loops created by opening one of the blue dots, or one of the red dots, have the same inclusion properties with respect to other loops, but have very different large scale connectivity properties (in particular they are far apart in Camia-Newman and Schramm-Smirnov distances). They do however have different homotopy classes with respect to other loops.	2012.11672	example_transformation.jpg
0.48102	3a9d13225a9c4b6a89411d9959b3bd47	On the left, an example of a domain Ω(f). Note that the sets Ω(f+kα) have a nested structure (the red part denotes Ω(f)∖Ω(f+α)). On the right, the impact of changing the boundary is compared with the impact of changing the boundary in a family of subdomains which are images by simple transformations of domains Ω(f) (with potentially different functions f). The existence of such a decomposition is made possible by the fact that the boundary of Ω is C^1.	2012.11672	finitedomainproof.jpg
0.477021	8bec4a6c6fd74064944485fdd99c4c77	In dashed black, the segments corresponding to the oriented edges in _η. Also, the crosses correspond to the points z(ℓ_x) for the loops. Finally, the associated segments are depicted in dashed red.	2012.11672	homotopy.jpg
0.503417	14fce53776364a1aacb2c82007b43fae	An example of a graph (α), where α_i is equal to π2 for i≤ 3, and α above. The diamond graph is drawn in light black lines, the white points refer to the vertices of the dual lattice, and the black points and the stronger black lines refer to the primal lattice. One sees that both the lower and upper parts are portions of a rotated rectangular lattice, and that below this rectangular lattice is simply the square lattice. The vertices of t_0^- are drawn in red, and those of t_0^+ in blue.	2012.11672	isoradial-lattice-unaltered.jpg
0.538505	3702c77aa636426299e74fb89f21448a	An example of a graph _k for k=4. What happens between tracks t_2 and t_5 is a mixture of the isoradial lattice with angles π/2 for i≤ 3 and α for i≥ 4, and ' is obtained by exchanging the tracks 4 and 5. The only diamond that does not belong to or ' is in gray.	2012.11672	isoradial-lattice.jpg
0.450969	c4b93a9c817348d68b9d8de1c28e41fc	When performing a track exchange between t_0 and t_-1, the vertex 0 is modified locally so that the coordinate of the right-most point in t_0^-∪ t_-1^- moves by either -1 (first line) or cosα (second line). The first outcome occurs certainly when both gray rhombi contain primal edges, and with positive probability otherwise; the second outcome may only occur when at least one of the two gray rhombi contains a dual edge. When the second outcome occurs, δ^ IIC = cosα. For the first outcome, δ^ IIC may take values 0, cosα - 1, or -1. The first two values appear if the incipient infinite cluster contains a vertex below 0 with first coordinate 0, or one above 0, with first coordinate cosα - 1, respectively. The same outcomes occur for any environment in t_0 and t_-1 to the left of 0.	2012.11672	isoradial_extrema_right.jpg
0.50787	5aebbff5f8ee45f98ea2af15000acee0	In black, the nails (i.e. the elements indexed by I(ω)) and in bold black the marked nails. The red segments depict the information provided by (^(t),^(t),^(t)). The blue loops are the loops surrounding at least two marked nails, i.e. the loops contributing to [·]_∙,0 and [·]_∙,1. The grey area are the boxes Λ_√(η)N(x), which one should think of potentially much bigger than the minimal size O( N) of nails, but much smaller than the minimal distance η N between vertices of _η(N).	2012.11672	keeping_track.jpg
0.487506	774205f3211d4c369d1ba63d61ed817a	On the left is an isoradial lattice in the left half-plane, almost identical to (α). The three arm events for any point on the vertical axis have the same probability in the left and right graphs.	2012.11672	left_hp.jpg
0.49154	07bb44497cd8401db0b952625638791b	An example where the configuration on the two diamonds left of the grey diamond ◊ determines the configuration on ◊ after the star-triangle transformation, irrespectively of the configuration on or right of ◊.	2012.11672	mixing_track_exchange.jpg
0.427397	b4844cdc84af4c448a9d4e88ecf885a4	An example with two clusters crossing. Note that the set X_i does not have to be included in the rectangle R. A way to construct a large set of points with the properties of X_i is to choose for each 1≤ j<⌊κ/48η⌋-1, on each line {(x,y)∈^2:y=(2j+1)η}, a vertex of Ω_i which is at a distance smaller than η of Γ_i.	2012.11672	setX.jpg
0.507711	cbd803ab766141ee8e773b04c3bb63c5	The three diamonds together with the drawing, on the left, of the triangle (in which case the dual graph in dashed has a star) and, on the right, of the star (in which case the dual graph has a triangle).	2012.11672	star-triangle.jpg
0.500458	57a9f98b53e84b23bc63a93ce9648440	The edge e and its subtended angle θ_e; bold edges are those of 𝔾, thin ones are those of the diamond graph.	2012.11672	subtended-angle.jpg
0.480002	28e94cecb69845219de1279478d3bf20	A picture of the possible transformations in the star-triangle coupling (the probabilities in the case of multiple outcomes are described in the definition). We also pictured the reverse map.	2012.11672	transformation.jpg
0.44304	74d53d22fff94d3b9a3f8b3d2e057a20	End-to-end architecture for the network that is trained using cross-entropy loss per label for video classification. Frame level features are used in the cluster-and-aggregate pooling layer which outputs a fixed dimensional video representation. This representation is passed through context gating and mixture of experts layers. Blue and green components indicate layers with trainable parameters. Diagram modified from <cit.>	2012.11673	e2eTraining.jpg
0.485533	428de95b82994fff98b9421a04cc284f	Partition function (top) and the Heat Capacity (bottom) of the Thermal State for the one degree of freedom Hyperbolic Hamiltonian H(x) = κ pq as a function of the inverse temperature β̅:= ħκβ.	2012.11674	FigH.jpg
0.455963	72492d99377045908a360b7a861c51e8	Partition function (dashed) and the Heat Capacity (continuous) of the Thermal State for the the two degrees of freedom Loxodromic Hamiltonian H(x) = κ (p_1 q_1 + p_2 q_2) + ω(p_2q_1 - p_1 q_2) as a function of the inverse temperature β. Top: ħω = 1 and ħκ = 4. Middle: ħω = 1 and ħκ = 1. Bottom: ħω = 0.2 and ħκ = 1.	2012.11674	FigL.jpg
0.462574	83108097ddd448d4b520f82e69104a25	The expected phase diagram of the six-vertex model contains four regions: I and II are called ferroelectric, III is antiferroelectric and IV is disordered. The latter two regions correspond to Δ < -1 and Δ∈ [-1,1], respectively. The present paper concerns only regions III and IV. The gray lines represent lines of constant Δ; the black quarter-circle corresponds to Δ = 0.	2012.11675	phase_diagram.jpg
0.455142	7444ba030001400db60e7ff971e2f7c7	Fluorescence spectra of a 250nm CMOS integrated circuit chip, measured with a TES sensor (blue solid line) and a Vortex silicon-drift detector (orange dashed line). Emission lines from W and Cu are visible in both measurements; prominent peaks are labeled with their corresponding element and line name.	2012.11680	250nm_spectrum_oct2020.jpg
0.426064	45b4fc39ca3144a0aaec487add3ed815	Fluorescence spectra of a 350nm CMOS integrated circuit chip, measured with a TES sensor (blue solid line) and a Vortex silicon-drift detector (orange dashed line). Prominent peaks are labeled with their corresponding element and line name.	2012.11680	350nm_spectrum_oct2020.jpg
0.443417	cf188260b8794bda8ec12fed8a1420db	Fluorescence spectra of a 130nm SiGe BiCMOS integrated circuit chip, measured with a TES sensor (blue solid line) and a Vortex silicon-drift detector (orange dashed line). Relative to the other chips, emission lines from Co, Ta, and Hf are also present, but are clearly resolved only in the TES-measured spectrum.	2012.11680	FASPAX_spectrum_oct2020.jpg
0.416095	0c49306121964dcf93c57136e2ff2f19	V and ∇	2012.11825	PRIN1_PRIN2.jpg
0.522651	517f7411c2fb4604a34be153646b9c30	V and ∇	2012.11825	fig1.jpg
0.523853	8dc4d448d7e14eebb658b15ee46c2bf9	V and ∇	2012.11825	fig2.jpg
0.51433	839feafef17e4e7abce88699babea2ab	V and ∇	2012.11825	fig3.jpg
0.4989	62e7b1aa25fb45669904c269d7636334	V and ∇	2012.11825	fig4.jpg
0.498201	76eb79f0d6154b2797dc1ffa5e744a65	V and ∇	2012.11825	fig5.jpg
0.52556	d4d8c2be581242449b76e93ea099677a	V and ∇	2012.11825	fig6.jpg
0.506462	7cef2d59203e43f38b3eb07225033db6	V and ∇	2012.11825	fig7.jpg
0.526218	a5644e73b3a44244ab444e9640318db5	V and ∇	2012.11825	fig8.jpg
0.540723	dfae6ad1e8a24d83aae304c756ac5979	V and ∇	2012.11825	fig9.jpg
0.4615	22b45f415f26410680ffb71a5c1d8ca8	This pictorial representation shows how we update when the iteration goes outside the ball in <cit.> method	2012.11826	SilBall.jpg
0.502508	fbcc64b45db74f1bafbdf8a7ce4b61c1	Case 2: two free-streaming neutrinos plus self-interacting relativistic species. The Figure shows select parameters for five runs with different decoupling widths of 10% (red), 20% (blue), 30% (green), 40% (yellow), and 80% (grey) of . The MCMC runs otherwise have an identical setup and are for the data combination P18 +lens +BAO.	2012.11830	2FSb_dzdec_select_zoom.jpg
0.48804	4040e099e2db4d1cbeea52b703819f98	Case 1: all species interacting. Full parameter space corresponding to Figure <ref>.	2012.11830	P_Pb_C1_R1z6m_reduced_zoom_G0p5_I2_1d2dx5.jpg
0.431382	996aadcd701d4244b4aee77fe21780ff	Case 2: two free-streaming neutrinos plus self-interacting relativistic species. Full parameter space corresponding to Figure <ref>.	2012.11830	P_Pb_Pbr_C2FS_2FSz6m_reduced_zoom_G0p5_I2.jpg
0.420745	b5c8384122ab4a0a8ebd87a05c15166a	Case 4: varying fraction of self-interacting relativistic species. Full parameter space corresponding to Figure <ref>.	2012.11830	P_Pb_Pbr_Cur_urz6i_reduced_zoom_G0p5_I2.jpg
0.55735	6d55d766554e472ba530f4420c999ef1	Case 3: varying fraction of self-interacting relativistic species, including fluid-like, with fixed =3.046. Full parameter space corresponding to Figure <ref>.	2012.11830	Pb_CPSI_PSIz6m_PSIfl_reduced_zoom_G0p5_I2.jpg
0.450102	698d4cc34e6d4d5792d471eb9977ed4a	Case 2: two free-streaming neutrinos plus self-interacting relativistic species, including fluid-like. Full parameter space corresponding to Figure <ref>.	2012.11830	Pb_Pbr_C2FS_2FSz6m_2FSfl_reduced_zoom_G0p5_I2.jpg
0.460739	9c34ee43a7234b3aa9602621287de2d1	Case 4: varying fraction of self-interacting relativistic species, including fluid-like. Full parameter space corresponding to Figure <ref>.	2012.11830	Pb_Pbr_Cur_urz6i_urfl_reduced_zoom_G0p5_I2.jpg
0.45084	a67701e67b764193b0ef081889b4c4eb	Case 1: all species interacting, no high-ℓ polarization. Full parameter space corresponding to Figure <ref>.	2012.11830	Pbnh_Pbnhr_C1_R1z6m_reduced_zoom_G0p5_I2_1d2dx3OL.jpg
0.4077	e27ccc48233d48c2b101043e7793fcd6	Case 1: all species interacting, including H_0 prior. Full parameter space corresponding to Figure <ref>.	2012.11830	Pbr_C1_R1z6n_reduced_zoom_G0p5_I2.jpg
0.406326	d05468585a654f65bde3afe3c07838e9	Experimental constraints for the Majoron coupled to neutrino mass eigenstates in the g_ϕ-m_ϕ plane translated from the constraints on the coupling to neutrino flavor eigenstates in <cit.>. We show these constraints translated under different assumptions about the number of interacting neutrinos. The gray bands show different values of in this parameter space with the upper lines of the bands corresponding to the 1-interacting neutrino case and the lower lines to the 3-interacting neutrino case.	2012.11830	gphiconst.jpg
0.405362	41945d989b344d0994dd2ccc530772d4	Opacity in terms of z for G_ν = 10^-1^-2 (solid) and G_ν = 10^-3^-2 (dashed). Grid lines are z where opacity is 0.5. We also show the transition function 𝒯(z) with corresponding and Δ = 0.4.	2012.11830	opacity.jpg
0.469327	0d11447374584502ba3a86b53c6bcd8a	Fractional differences between CMB temperature (top left) and polarization (top right) auto-correlation and cross-correlation (bottom left, showing difference rather than fractional difference) angular power spectra, and the CMB lensing power spectrum (bottom right). In all panels, the Planck 2018 binned data <cit.> are shown in black, for comparison. Fractional differences are computed with respect to a best-fit free-streaming neutrinos comparison case. The orange curves show power spectra in a model with both free-streaming and self-interacting neutrinos (Case 2), computed at the best fit cosmological parameters for that model with a decoupling redshift of =12,200. In red the same cosmological parameters are assumed as for the orange curve, but the the power spectra are computed assuming only free-streaming neutrinos. In blue, the power spectra are computed in a cosmology with interacting neutrinos assuming = 12,200 using the best-fit cosmological parameters for the free-streaming reference case. See Sec. (<ref>) for a description of the reference and interacting neutrino models and Table <ref> (P18 +lens +BAO) for the precise parameter choices.	2012.11830	spectra_Pb_2FSb_cut_mode_1_both_ref_v4_combo_diff.jpg
0.408188	3304ec1ab5654a55bca1a294b24ccbfb	The fractional differences between CMB power spectra computed in a scenario with a varying fraction of interacting neutrinos that decouple from self-interactions at = 10,000, as compared to those computed in a cosmology with the same total comprised of all free-streaming neutrinos (all other parameters are also held fixed). Shown are the CMB temperature (top left) and polarization (top right) auto-correlation and cross-correlation (bottom left, showing difference rather than fractional difference) angular power spectra, and the CMB lensing power spectrum (bottom right).	2012.11830	spectra_Pb_2FSb_ref_varying_Neff_combo_diff.jpg
0.432694	ff990fc312934a4bafbdb786bc6ba04f	Fractional differences of the CMB temperature (top left) and polarisation (top right) auto-correlation and cross-correlation (bottom left, showing difference rather than fractional difference) angular power spectra, and the CMB lensing power spectrum (bottom right). All cases are compared to a free-streaming neutrinos comparison case with varying width of decoupling Δ of 10% (more red) to 80% (more yellow) of the decoupling redshift of =100,000.	2012.11830	spectra_Pb_2FSb_ref_varying_dzdec2_combo_diff.jpg
0.418176	11c80b5be7954de58259318308bb57b5	Fractional differences of the CMB temperature (top left) and polarisation (top right) auto-correlation and cross-correlation (bottom left, showing difference rather than fractional difference) angular power spectra, and the CMB lensing power spectrum (bottom right). All cases are compared to a free-streaming neutrinos comparison case with varying width of decoupling Δ of 10% (more red) to 80% (more yellow) of the decoupling redshift of =10,000.	2012.11830	spectra_Pb_2FSb_ref_varying_dzdec_combo_diff.jpg
0.438061	2f0fa8c76c934fc5bf8e7b6e124036d7	Fractional differences between the CMB power spectra computed in a model with self-interacting and free-streaming neutrinos (Case 2, see Sec. <ref>), as compared to a model with only free-streaming neutrinos. Each curve shows a different assumed value of , with all other parameters held fixed. Shown are CMB temperature (top left) and polarization (top right) auto-correlation and cross-correlation (bottom left, showing difference rather than fractional difference) angular power spectra, and the CMB lensing power spectrum (bottom right). In all panels, the Planck 2018 binned data <cit.> are shown in black, for comparison.	2012.11830	spectra_Pb_2FSb_ref_varying_zdec_combo_diff.jpg
0.446805	499cb0c406f840f9811a8d85abb9a357	The relationship between the redshift of neutrino decoupling, z_dec, and the effective neutrino self-interaction strength G_ν. Plotted is the relationship for different numbers of interacting neutrino species. The red, orange, and green lines correspond to 1, 2, and 3 self-interacting neutrino species, respectively. The gray line is result from dimensional analysis, z_dec∼ (G_ν^2 T_0^3 m_pl)^-1/3.	2012.11830	zdecvsGnu.jpg
0.432418	3bd9336316d14b9492f60a08ad8f2657	The results of evolutionary learning (genome bottom and phenome top) instructed to produce (a) 4-mers, (b) 5-mers and 5-gons, (c) 6-gons, and (d) 12-gons.	2012.11832	collection.jpg
0.478878	187c2636397a4b87be092e8ca5e4e979	Evolutionary learning directed to maximize the number of 12-gons, convex pores of 12 sides (shown light blue in images). The main panel shows the yield ϕ, as a function of generation n, produced by the most successful genome (potential and protocol). The particle symbols are the data points, which also show the form of the learned potential. Below the data points are plots of the learned protocol (red and blue lines) and resulting yield (green lines) as a function of time t. The format of those plots is shown bottom right. Large positive values of ϵ indicate strong particle attractions, and large positive and negative values of μ promote dense and sparse substrates, respectively. The snapshots at the top show portions of a simulation box from three different generations, indicated by the blue lines. Parameters: ϕ_0=90 12-gons, ϵ_0=μ_0=20, n_0=27 generations, t_0=10^9 Monte Carlo steps.	2012.11832	fig_12_gons.jpg
0.428927	d6f7993994ec430f83b096afd6af712f	Self-assembly dynamics produced by the most successful genome (potential and protocol, shown top left) from generation 27 of fig_12_gons. N_3 and N_12 are the number of 3-gons (red) and 12-gons (light blue). The snapshots are taken from the times indicated by the blue lines. 12-gons form relatively late in the trajectory, emphasizing that self-assembly is a “sparse reward” problem: the objective ϕ can be zero for most of the simulation. Evolutionary methods that require information from only the final time point of the simulation are natural ways to tackle such problems. Parameters: t_0=10^9 Monte Carlo steps.	2012.11832	fig_12_gons_dynamics.jpg
0.45409	57a31c093a4a45d38381fbb7a112f097	Evolutionary learning with ϕ=C_4, the number of 4-mers, interacting clusters of 4 particles (shown darker in images). The format of the figure is the same as fig_12_gons. Parameters: ϕ_0=350 4-mers, ϵ_0=μ_0=20, n_0=17 generations, t_0=10^9 Monte Carlo steps.	2012.11832	fig_4_mers.jpg
0.447571	722976bf70044373a01e21017c647124	Evolutionary learning with ϕ=min(C_5,N_5), the number of 5-mers (shown darker in images) and 5-gons (shown light blue). The format of the figure is the same as fig_12_gons, with the exception of the top-left image: that shows the outcome of learning after 17 generations with ϕ=C_5. Evolution of pentagonal clusters requires the dual goal of 5-mers and 5-gons. Parameters: ϕ_0=125 5-mers & 5-gons, ϵ_0=μ_0=20, n_0=17 generations, t_0=10^9 Monte Carlo steps.	2012.11832	fig_5_mers_gons.jpg
0.443591	cbfa4ab8df114fdc85f37edc1a423fc3	Evolutionary learning with ϕ=N_6, the number of 6-gons, convex pores of 6 interacting particles (shown light blue: see boxed image). The format of the figure is the same as fig_12_gons. Parameters: ϕ_0=700 6-gons, ϵ_0=μ_0=20, n_0=17 generations, t_0=10^9 Monte Carlo steps.	2012.11832	fig_6_gons.jpg
0.49572	d6b462be85094971becf06726cf7d6d2	Geometry for the interaction potential. The white arrow indicates the line from the center of the particle to the point on its circumference specified by θ=0.	2012.11832	fig_potential.jpg
0.405748	0a9218ca0b044e6cbcd2d97d65bd67c2	The results of evolutionary learning instructed to produce novelty in the space of 3-gons and 4-gons. The scatterplot shows the coverage obtained in this space after 10 generations of learning; the images show examples of the structures produced (the particles underlying the polygons are not shown).	2012.11832	novelty_search.jpg
0.445454	5387deb08f0d4fe9aba5e1e79ec7a0a3	One of the materials identified by the novelty search of fig_novelty shows the σ, H, and Z binding motifs characteristic of dodecagonal quasicrystalsv̧a̧ņ2̧0̧1̧2̧f̧o̧ŗm̧a̧ţi̧o̧ņ. To produce this figure we used the particle- and protocol design identified by the learning algorithm of fig_novelty, and ran a trajectory of length 20 t_0. The particles underlying the pattern of 3-gons and 4-gons are not shown.	2012.11832	novelty_search2.jpg
0.519876	3ee2d3cc74ad40cba9908756d0ea154d	(a) We express the angular interaction potential and the time-dependent self-assembly protocol for a set of model molecules in the form of two neural networks, which together comprise the “genome” for making a material. In images, attractive portions of particles are shown green. (b) Upon specifying a design goal, neuroevolutionary learning can produce a genome whose “phenome” – the result of molecular simulations carried out using the potential and protocol specified by the genome – is a material satisfying that goal.	2012.11832	schematic.jpg
0.482825	e7b5cc26a4bc4fd5b415a4ee80ac31bf	(a) Yield of the self-assembly process using the best generation-27 genome (particle and potential) of fig_12_gons (blue line) compared to the yield obtained using a human-designed particlew̧ḩi̧ţȩļa̧m̧2̧0̧1̧6̧m̧i̧ņi̧m̧a̧ļ combined with evolutionary learning of the protocol (light blue). Also shown are three simulations (red) using a human-designed particle and protocols, the latter consisting of slow cooling at three rates at fixed chemical potential. (b) Yield of the self-assembly process using the best generation-18 genome from fig_collection(c) and fig_6_gons (blue line), compared to that using a particle with perfect three-fold rotational symmetry combined with evolutionary learning of the protocol (light blue).	2012.11832	yields.jpg
0.427807	d69afd447c72415ab522a6546fc2f9d0	The mean number of photons difference in percentile at the exit of the cell between α_a = 0.1 and α_a = 0 versus the FWM gain G(t=1) for (a) probe mode, (⟨â^†â⟩_t=1,α = 0 - ⟨â^†â⟩_t=1,α = 0.1)/⟨â^†â⟩_t=1,α = 0, and (b) conjugate mode, (⟨b̂^†b̂⟩_t=1,α = 0 - ⟨b̂^†b̂⟩_t=1,α = 0.1)/⟨b̂^†b̂⟩_t=1,α = 0. (c) The expectation value difference in percentile for the quantum correlation \|⟨âb̂⟩\| between α_a = 0.1 and α_a = 0, i.e., (\|⟨âb̂⟩\|_t=1,α = 0 - \|⟨âb̂⟩\|_t=1,α = 0.1)/\|⟨âb̂⟩\|_t=1,α = 0, versus the FWM gain.	2012.11839	Response.jpg
0.552872	3fd8cf934d1541e2a08cd3b178821185	TMSV beam (light cone) power as a function of pump power. Dashed line is the theoretical fit according to Eq. (<ref>) with r ≡α P, where P is the pump power.	2012.11839	Ring_vs_Pump1.jpg
0.432621	393cccd116a4460da8e147dfc1d7e3c0	SHG signals induced by 80 μW partially-decoherent TMSV beams of light (blue dots, left-hand side y-axis) and 80 μW coherent beams of light (red dots, right-hand side y-axis) as a function of transmission of ND filters. Dashed blue line is the theoretical fit according to Eq. (<ref>).	2012.11839	SHG_Decoh1.jpg
0.426057	d742a027610b45b6a3d63179546c1378	SHG signal as a function of the power of the light cone (shown in Fig. <ref>(c) as the `ring'). Dashed line is a polynomial fit according to Eq. (<ref>) with ND = 0.	2012.11839	SHG_vs_Ring1.jpg
0.452936	83d771e1c48b4aa5a263674245cf7857	(a) Experimental setup in which a CW laser-pumped ^85Rb vapor cell produces a TMSV state via the FWM process. The TMSV beam (i.e. the `light cone') is separated from the pump beam by a ∼ 2×10^5 :1 polarizing beam splitter after the cell. The SHG signal from the BBO crystal is collected by an EMCCD camera. Two low-pass filters are mounted in front of the camera to eliminate undesired excitation photons. The BBO crystal and the EMCCD camera are enclosed in a light-proofing box to block ambient light. PBS: polarizing beam splitter, ND: neutral density filter, LP: low-pass filter. (b) Level structure of the D1 transition of ^85Rb atom. The optical transitions are arranged in a double-Λ configuration, where ν_p, ν_c and ν_1 stand for probe, conjugate and pump frequencies, respectively, fulfilling ν_p + ν_c = 2ν_1 and ν_c - ν_p = 2ν_HF. The width of the excited state in the level diagram represents the Doppler broadened line. Δ is the one-photon detuning. ν_HF is the hyperfine splitting in the electronic ground state of ^85Rb. (c) Image of the cross-section of the TMSV state, i.e., the `light cone', where the bright central spot is the residual pump beam. The angular separation between the center of the ring and the center of the residual pump beam is ∼ 0.3^∘, and the width of the ring is measured to be ∼ 0.05^∘.	2012.11839	Setup_Combo1.jpg
0.446683	3a04e2a545ee418fb39c9ccfaf26bdcd	Theoretical decoherence characterization of the TMSV state via its SHG signal. To isolate the effect of decoherence on the quantum correlation ⟨âb̂⟩, experiment is performed by holding power p constant, i.e., p≡ p_0.	2012.11839	TheoryFig.jpg
0.451007	b796395fa0bb415a978d829796fc85a5	Overview of our approach. We formulate selective forgetting by four datasets; _r, _f, _f, clean, and _r, extra. Depending on which dataset the DNN should perform well, three patterns of forgetting are introduced. In Pattern A, the DNN forgets samples such as outliers. As illustrated in the figure's left side, images of mice that are input devices are outliers when the task is classifying animals, for example. In Pattern B, the DNN forgets a part of each training sample. The model is supposed to forget to use the sunglasses as a feature for classification in the figure. Pattern C is similar to Pattern B; in the figure, the model is supposed to forget to use the logo or the emblem of a carmaker for classifying types of cars. Pattern B and C are different in their evaluation. Example images are from WebVision <cit.>, CelebA <cit.>, and Cars Dataset <cit.>.	2012.11849	fig_main.jpg
0.390342	6c2a28b304c344dca33a3b25a30402b2	Tile, FashionMNIST: Backdoor with a RLD factor 1.	2012.11849	rld.jpg
0.461641	31a3c4c59c1649bd981df472e5bd5374	Distribution of softmax(f_θ(_f)). We use Fashion-MNIST and the forgotten class is 0.	2012.11849	softmax.jpg
0.408781	dfe2653ada044e28b08994e50c7c5b94	Case D: the solution path of the MTIP.	2012.11855	CaseA1.jpg
0.473503	e024ad037e114b89829b2046a7ffa3e3	Case D: The profiles of F[E(t)], L^-[E(t)], and L^+[E(t)] with respect to time t.	2012.11855	CaseA2_new.jpg
0.478084	62a4d39d259140e4ba71fe8552de6999		2012.11855	CaseA3.jpg
0.469146	361afe98f1674c2e95110a6011d1b8fe	Case A: The profile of F[E(t)] with respect to time.	2012.11855	CaseAA1_new.jpg
0.43617	ff93a16528c64b1682869c3d694b1bd9	Case A: The solution path of the MTIP.	2012.11855	CaseAA2.jpg
0.416942	342898d8f6154be0a96ea5f0b67202ed		2012.11855	CaseAA3.jpg
0.445203	127ed7f5e7fc4e3fbf928f379f4b7259	Case B: The profile of F[E(t)] with respect to time.	2012.11855	CaseBB1.jpg
0.477351	e414ebddf53e4bb1867a795bd066fc54	Case B: The solution path of the MTIP.	2012.11855	CaseBB2.jpg
0.528094	caa21fbb3065480cb62ecadf79c2b83d		2012.11855	CaseBB3.jpg
0.432309	f0bcf51e0ca448199ca4df0505232ac7	Case C: The solution path of the MTIP.	2012.11855	CaseC2.jpg
0.484183	76da0b60e18640f599796d5d738fb952		2012.11855	CaseC3.jpg
0.45327	8b99c0b3177f4ff1b01bbd071ee250b4	The geometry for the path of RDP with terminal points E(t̅+ε), E(t̅), and E(t̅-ε).	2012.11855	Intersection.jpg
0.507986	606d4dbd0ab2409aab9bb6c9b5c3137b	The geometries of the feasible Dubins paths related to F, L^-, and L^+.	2012.11855	L1L2.jpg
0.502135	ce3637f3bc9a441eb91bfa3a4d3a4f48	The geometry of the solution path of RDP with a type of LR.	2012.11855	Path_LR.jpg
0.484622	85a45fba0763460a940a99f389d572a6	The geometry of the solution path of RDP with a type of RS.	2012.11855	Path_RS.jpg
0.501143	345ee61253a3448388f32b768974f17f	The geometry of the circles 𝒞_r and 𝒞_l and the subregions ℛ_1, ℛ_2, and ℛ_3 in the 2-dimensional plane.	2012.11855	Subregions.jpg
0.516921	3044a60f28454ace87ce7d21a436d03f	The regions for which the path of RDP terminates by different types. <cit.>	2012.11855	Types_Region.jpg
0.447366	6a3b54c50c9f40b6b420983fed14789a	Figure showing the performance of a multi-channel LoRaDRL and LoRaSim scheme under frequency jamming attack. A small drop of performance is seen in the case of LoRaDRL as it shifts to the other available frequency. While in the case of LoRaSim, a sudden drop of performance can be seen because of the absence of a feedback loop.	2012.11867	freq_blocking.jpg
0.999988	6b93f3684b2b40faa22218f748ca9d4a	Mapping of our problem setup on DRL. The gateway is considered the agent and the LoRa network represents the environment.	2012.11867	loradrl.jpg
0.467528	bd77cc25d1f941889c0b63b22ec54eb4	Architecture of LoRaWAN consisting of LoRa EDs, LoRa gateway, network server, and end-user. The EDs serve different purposes and transmit the data to the gateway based on the application requirements. The received data packets are forwarded to the network server, which in turn forwards them to the end-user.	2012.11867	lorawan.jpg

0.525179

dd90888dc3b24d36afda3e1907c2af54

(Left) Bode plot – frequency response of modulation system showing modulation amplitude for various modulation frequencies. The system is capable of delivering modulation two times the requirement. (Right) 3000μrad amplitude and 1000 Hz modulation circle recorded on the CCID75 camera using a laser beam reflected on the modulation system.

2012.11668

Modulation_frequncy_response.jpg

0.387479

5efeb0285db541d7b420e0d3603cf15e

(Left) Acquisition camera procured from Basler. (Right) Quantum efficiency more than 65% for wavelengths 0.4-0.8μm.

2012.11668

QE_acq.jpg

0.436859

703c3e6c171d43ad894044c52caf4f5f

The visible pyramid wavefront sensor: (left) double pyramid, (middle) camera triplet lens and (right) Yorick-Yao<cit.> simulated pyramid wavefront sensor pupil images on the detector.

2012.11668

VPWFS.jpg

0.449462

9b567d95365c474aa8d08df738779139

The different cases zoomed at a distance d(t)^1/2 around 𝐱. The configuration ω^(t+1/2) is depicted in black, and 𝐱+ω_,j^(t) in blue.

2012.11672

3arm-bis.jpg

0.522536

07d24280467e462f8d1c3818b102cf5c

On the left, an instance of A^_010(0,R) where the primal path is depicted by a bold path, and the dual ones by dashed paths. In the middle, an instance of A^_010(0,R). On the right, A^_010(0,R)∘ A_1(0,R).

2012.11672

3arm.jpg

0.431181

efeb45a8add34a56ac05ba5dd4219ff1

The initial and final graphs ^(0) and ^(2M+1) of the two processes are the same, but the intermediate graphs (^(t))_0<t≤ 2M (top) and (^(t))_0<t<M (bottom) are different (the figure depicts the diamond graphs). In the top process, horizontal tracks are exchanged successively, by pushing the gray rhombus from right to left; in the bottom process, the rhombus is pushed downwards, effectively exchanging vertical tracks. Throughout the two processes, we record the right-most coordinate of the union of all the clusters intersecting the base.

2012.11672

BBtilde.jpg

0.439036

17b71f98ab9a4eec861281f61dffb106

A picture of the set Ω(ω^(t+1/2),x) (in grey). The dark grey cluster is the unique cluster of the annulus crossing from inside to outside. Everything outside Λ_d(t)^1/2(x) is included in Ω(ω^(t+1/2),x). The red part is the place where we try to couple as best as possible ω^(t+1/2) and 𝐱+ω^(t)_,j(t)-T^(t)(i). In the proof of Proposition <ref>, note that when applied to ω^(t+1/2)∈ E(x) (which means that there exists a unique cluster crossing the annulus), all the conditions on the t-extrema of other nails and homotopy classes of loops are not impacted by what happens inside Ω(ω^(t+1/2),𝐱)^c. As a consequence, there is a “screening” property and the conditioning inside is simply the existence of the red arms, meaning that 𝐱 is equal to the left-most top-most vertex in the cluster that is crossing the annulus.

2012.11672

Omega.jpg

0.433029

7783c8e9663248688bdf6870a105d164

We depicted in black, dotted black, red and blue respectively the primal, dual, diamond and medial lattices. The primal configuration ω is in bold and the dual one ω^* in dashed bold. Finally, the loop configuration ω is in black.

2012.11672

configurations.jpg

0.512411

2d999090103c4bc5917f2cb67de6ae78

The quantities h(t), a(t), b(t) and d(t). One may deduce from the picture that ⌊ t/(2N)⌋=12 as there are 12 tracks of angle α stacked above level -N.

2012.11672

definitions_bdh.jpg

0.45108

a8056020d0e74a14abee779d2daf4d17

A picture of the event E_j(k,m) (except that the minimality of m is not really explicitly depicted). Compared to Figure <ref>, additional conditions are forced on the tracks t_j-m and t_j+m and what happens in between.

2012.11672

eventE.jpg

0.482448

c045d40a5afd439bb18d2bafa370c4fc

A picture of the event E_j(k). Note that we are on a torus, hence t_1-M^-=t_M^+.

2012.11672

eventEintro.jpg

0.462412

b90a0bfae84e4772966c0051c77066ee

We depicted the example of a nail having two top t-extrema.

2012.11672

eventarms.jpg

0.51034

e65f9f0a817c443aad611ee19276119f

On the left, an example of a transformation of the red points that continuously transforms a blue loop with a certain homotopy class into another blue loop with a different homotopy class. It is therefore crucial to encode homotopy classes in a coherent way, which will be done in Section <ref>. Also, on the right, a justification of why we did not choose to keep track of a simpler property of loops, namely, the inclusion between loops. In the picture, the two loops created by opening one of the blue dots, or one of the red dots, have the same inclusion properties with respect to other loops, but have very different large scale connectivity properties (in particular they are far apart in Camia-Newman and Schramm-Smirnov distances). They do however have different homotopy classes with respect to other loops.

2012.11672

example_transformation.jpg

0.48102

3a9d13225a9c4b6a89411d9959b3bd47

On the left, an example of a domain Ω(f). Note that the sets Ω(f+kα) have a nested structure (the red part denotes Ω(f)∖Ω(f+α)). On the right, the impact of changing the boundary is compared with the impact of changing the boundary in a family of subdomains which are images by simple transformations of domains Ω(f) (with potentially different functions f). The existence of such a decomposition is made possible by the fact that the boundary of Ω is C^1.

2012.11672

finitedomainproof.jpg

0.477021

8bec4a6c6fd74064944485fdd99c4c77

In dashed black, the segments corresponding to the oriented edges in _η. Also, the crosses correspond to the points z(ℓ_x) for the loops. Finally, the associated segments are depicted in dashed red.

2012.11672

homotopy.jpg

0.503417

14fce53776364a1aacb2c82007b43fae

An example of a graph (α), where α_i is equal to π2 for i≤ 3, and α above. The diamond graph is drawn in light black lines, the white points refer to the vertices of the dual lattice, and the black points and the stronger black lines refer to the primal lattice. One sees that both the lower and upper parts are portions of a rotated rectangular lattice, and that below this rectangular lattice is simply the square lattice. The vertices of t_0^- are drawn in red, and those of t_0^+ in blue.

2012.11672

isoradial-lattice-unaltered.jpg

0.538505

3702c77aa636426299e74fb89f21448a

An example of a graph _k for k=4. What happens between tracks t_2 and t_5 is a mixture of the isoradial lattice with angles π/2 for i≤ 3 and α for i≥ 4, and ' is obtained by exchanging the tracks 4 and 5. The only diamond that does not belong to or ' is in gray.

2012.11672

isoradial-lattice.jpg

0.450969

c4b93a9c817348d68b9d8de1c28e41fc

When performing a track exchange between t_0 and t_-1, the vertex 0 is modified locally so that the coordinate of the right-most point in t_0^-∪ t_-1^- moves by either -1 (first line) or cosα (second line). The first outcome occurs certainly when both gray rhombi contain primal edges, and with positive probability otherwise; the second outcome may only occur when at least one of the two gray rhombi contains a dual edge. When the second outcome occurs, δ^ IIC = cosα. For the first outcome, δ^ IIC may take values 0, cosα - 1, or -1. The first two values appear if the incipient infinite cluster contains a vertex below 0 with first coordinate 0, or one above 0, with first coordinate cosα - 1, respectively. The same outcomes occur for any environment in t_0 and t_-1 to the left of 0.

2012.11672

isoradial_extrema_right.jpg

0.50787

5aebbff5f8ee45f98ea2af15000acee0

In black, the nails (i.e. the elements indexed by I(ω)) and in bold black the marked nails. The red segments depict the information provided by (^(t),^(t),^(t)). The blue loops are the loops surrounding at least two marked nails, i.e. the loops contributing to [·]_∙,0 and [·]_∙,1. The grey area are the boxes Λ_√(η)N(x), which one should think of potentially much bigger than the minimal size O( N) of nails, but much smaller than the minimal distance η N between vertices of _η(N).

2012.11672

keeping_track.jpg

0.487506

774205f3211d4c369d1ba63d61ed817a

On the left is an isoradial lattice in the left half-plane, almost identical to (α). The three arm events for any point on the vertical axis have the same probability in the left and right graphs.

2012.11672

left_hp.jpg

0.49154

07bb44497cd8401db0b952625638791b

An example where the configuration on the two diamonds left of the grey diamond ◊ determines the configuration on ◊ after the star-triangle transformation, irrespectively of the configuration on or right of ◊.

2012.11672

mixing_track_exchange.jpg

0.427397

b4844cdc84af4c448a9d4e88ecf885a4

An example with two clusters crossing. Note that the set X_i does not have to be included in the rectangle R. A way to construct a large set of points with the properties of X_i is to choose for each 1≤ j<⌊κ/48η⌋-1, on each line {(x,y)∈^2:y=(2j+1)η}, a vertex of Ω_i which is at a distance smaller than η of Γ_i.

2012.11672

setX.jpg

0.507711

cbd803ab766141ee8e773b04c3bb63c5

The three diamonds together with the drawing, on the left, of the triangle (in which case the dual graph in dashed has a star) and, on the right, of the star (in which case the dual graph has a triangle).

2012.11672

star-triangle.jpg

0.500458

57a9f98b53e84b23bc63a93ce9648440

The edge e and its subtended angle θ_e; bold edges are those of 𝔾, thin ones are those of the diamond graph.

2012.11672

subtended-angle.jpg

0.480002

28e94cecb69845219de1279478d3bf20

A picture of the possible transformations in the star-triangle coupling (the probabilities in the case of multiple outcomes are described in the definition). We also pictured the reverse map.

2012.11672

transformation.jpg

0.44304

74d53d22fff94d3b9a3f8b3d2e057a20

End-to-end architecture for the network that is trained using cross-entropy loss per label for video classification. Frame level features are used in the cluster-and-aggregate pooling layer which outputs a fixed dimensional video representation. This representation is passed through context gating and mixture of experts layers. Blue and green components indicate layers with trainable parameters. Diagram modified from <cit.>

2012.11673

e2eTraining.jpg

0.485533

428de95b82994fff98b9421a04cc284f

Partition function (top) and the Heat Capacity (bottom) of the Thermal State for the one degree of freedom Hyperbolic Hamiltonian H(x) = κ pq as a function of the inverse temperature β̅:= ħκβ.

2012.11674

FigH.jpg

0.455963

72492d99377045908a360b7a861c51e8

Partition function (dashed) and the Heat Capacity (continuous) of the Thermal State for the the two degrees of freedom Loxodromic Hamiltonian H(x) = κ (p_1 q_1 + p_2 q_2) + ω(p_2q_1 - p_1 q_2) as a function of the inverse temperature β. Top: ħω = 1 and ħκ = 4. Middle: ħω = 1 and ħκ = 1. Bottom: ħω = 0.2 and ħκ = 1.

2012.11674

FigL.jpg

0.462574

83108097ddd448d4b520f82e69104a25

The expected phase diagram of the six-vertex model contains four regions: I and II are called ferroelectric, III is antiferroelectric and IV is disordered. The latter two regions correspond to Δ < -1 and Δ∈ [-1,1], respectively. The present paper concerns only regions III and IV. The gray lines represent lines of constant Δ; the black quarter-circle corresponds to Δ = 0.

2012.11675

phase_diagram.jpg

0.455142

7444ba030001400db60e7ff971e2f7c7

Fluorescence spectra of a 250nm CMOS integrated circuit chip, measured with a TES sensor (blue solid line) and a Vortex silicon-drift detector (orange dashed line). Emission lines from W and Cu are visible in both measurements; prominent peaks are labeled with their corresponding element and line name.

2012.11680

250nm_spectrum_oct2020.jpg

0.426064

45b4fc39ca3144a0aaec487add3ed815

Fluorescence spectra of a 350nm CMOS integrated circuit chip, measured with a TES sensor (blue solid line) and a Vortex silicon-drift detector (orange dashed line). Prominent peaks are labeled with their corresponding element and line name.

2012.11680

350nm_spectrum_oct2020.jpg

0.443417

cf188260b8794bda8ec12fed8a1420db

Fluorescence spectra of a 130nm SiGe BiCMOS integrated circuit chip, measured with a TES sensor (blue solid line) and a Vortex silicon-drift detector (orange dashed line). Relative to the other chips, emission lines from Co, Ta, and Hf are also present, but are clearly resolved only in the TES-measured spectrum.

2012.11680

FASPAX_spectrum_oct2020.jpg

0.416095

0c49306121964dcf93c57136e2ff2f19

V and ∇

2012.11825

PRIN1_PRIN2.jpg

0.522651

517f7411c2fb4604a34be153646b9c30

V and ∇

2012.11825

fig1.jpg

0.523853

8dc4d448d7e14eebb658b15ee46c2bf9

V and ∇

2012.11825

fig2.jpg

0.51433

839feafef17e4e7abce88699babea2ab

V and ∇

2012.11825

fig3.jpg

0.4989

62e7b1aa25fb45669904c269d7636334

V and ∇

2012.11825

fig4.jpg

0.498201

76eb79f0d6154b2797dc1ffa5e744a65

V and ∇

2012.11825

fig5.jpg

0.52556

d4d8c2be581242449b76e93ea099677a

V and ∇

2012.11825

fig6.jpg

0.506462

7cef2d59203e43f38b3eb07225033db6

V and ∇

2012.11825

fig7.jpg

0.526218

a5644e73b3a44244ab444e9640318db5

V and ∇

2012.11825

fig8.jpg

0.540723

dfae6ad1e8a24d83aae304c756ac5979

V and ∇

2012.11825

fig9.jpg

0.4615

22b45f415f26410680ffb71a5c1d8ca8

This pictorial representation shows how we update when the iteration goes outside the ball in <cit.> method

2012.11826

SilBall.jpg

0.502508

fbcc64b45db74f1bafbdf8a7ce4b61c1

Case 2: two free-streaming neutrinos plus self-interacting relativistic species. The Figure shows select parameters for five runs with different decoupling widths of 10% (red), 20% (blue), 30% (green), 40% (yellow), and 80% (grey) of . The MCMC runs otherwise have an identical setup and are for the data combination P18 +lens +BAO.

2012.11830

2FSb_dzdec_select_zoom.jpg

0.48804

4040e099e2db4d1cbeea52b703819f98

Case 1: all species interacting. Full parameter space corresponding to Figure <ref>.

2012.11830

P_Pb_C1_R1z6m_reduced_zoom_G0p5_I2_1d2dx5.jpg

0.431382

996aadcd701d4244b4aee77fe21780ff

Case 2: two free-streaming neutrinos plus self-interacting relativistic species. Full parameter space corresponding to Figure <ref>.

2012.11830

P_Pb_Pbr_C2FS_2FSz6m_reduced_zoom_G0p5_I2.jpg

0.420745

b5c8384122ab4a0a8ebd87a05c15166a

Case 4: varying fraction of self-interacting relativistic species. Full parameter space corresponding to Figure <ref>.

2012.11830

P_Pb_Pbr_Cur_urz6i_reduced_zoom_G0p5_I2.jpg

0.55735

6d55d766554e472ba530f4420c999ef1

Case 3: varying fraction of self-interacting relativistic species, including fluid-like, with fixed =3.046. Full parameter space corresponding to Figure <ref>.

2012.11830

Pb_CPSI_PSIz6m_PSIfl_reduced_zoom_G0p5_I2.jpg

0.450102

698d4cc34e6d4d5792d471eb9977ed4a

Case 2: two free-streaming neutrinos plus self-interacting relativistic species, including fluid-like. Full parameter space corresponding to Figure <ref>.

2012.11830

Pb_Pbr_C2FS_2FSz6m_2FSfl_reduced_zoom_G0p5_I2.jpg

0.460739

9c34ee43a7234b3aa9602621287de2d1

Case 4: varying fraction of self-interacting relativistic species, including fluid-like. Full parameter space corresponding to Figure <ref>.

2012.11830

Pb_Pbr_Cur_urz6i_urfl_reduced_zoom_G0p5_I2.jpg

0.45084

a67701e67b764193b0ef081889b4c4eb

Case 1: all species interacting, no high-ℓ polarization. Full parameter space corresponding to Figure <ref>.

2012.11830

Pbnh_Pbnhr_C1_R1z6m_reduced_zoom_G0p5_I2_1d2dx3OL.jpg

0.4077

e27ccc48233d48c2b101043e7793fcd6

Case 1: all species interacting, including H_0 prior. Full parameter space corresponding to Figure <ref>.

2012.11830

Pbr_C1_R1z6n_reduced_zoom_G0p5_I2.jpg

0.406326

d05468585a654f65bde3afe3c07838e9

Experimental constraints for the Majoron coupled to neutrino mass eigenstates in the g_ϕ-m_ϕ plane translated from the constraints on the coupling to neutrino flavor eigenstates in <cit.>. We show these constraints translated under different assumptions about the number of interacting neutrinos. The gray bands show different values of in this parameter space with the upper lines of the bands corresponding to the 1-interacting neutrino case and the lower lines to the 3-interacting neutrino case.

2012.11830

gphiconst.jpg

0.405362

41945d989b344d0994dd2ccc530772d4

Opacity in terms of z for G_ν = 10^-1^-2 (solid) and G_ν = 10^-3^-2 (dashed). Grid lines are z where opacity is 0.5. We also show the transition function 𝒯(z) with corresponding and Δ = 0.4.

2012.11830

opacity.jpg

0.469327

0d11447374584502ba3a86b53c6bcd8a

Fractional differences between CMB temperature (top left) and polarization (top right) auto-correlation and cross-correlation (bottom left, showing difference rather than fractional difference) angular power spectra, and the CMB lensing power spectrum (bottom right). In all panels, the Planck 2018 binned data <cit.> are shown in black, for comparison. Fractional differences are computed with respect to a best-fit free-streaming neutrinos comparison case. The orange curves show power spectra in a model with both free-streaming and self-interacting neutrinos (Case 2), computed at the best fit cosmological parameters for that model with a decoupling redshift of =12,200. In red the same cosmological parameters are assumed as for the orange curve, but the the power spectra are computed assuming only free-streaming neutrinos. In blue, the power spectra are computed in a cosmology with interacting neutrinos assuming = 12,200 using the best-fit cosmological parameters for the free-streaming reference case. See Sec. (<ref>) for a description of the reference and interacting neutrino models and Table <ref> (P18 +lens +BAO) for the precise parameter choices.

2012.11830

spectra_Pb_2FSb_cut_mode_1_both_ref_v4_combo_diff.jpg

0.408188

3304ec1ab5654a55bca1a294b24ccbfb

The fractional differences between CMB power spectra computed in a scenario with a varying fraction of interacting neutrinos that decouple from self-interactions at = 10,000, as compared to those computed in a cosmology with the same total comprised of all free-streaming neutrinos (all other parameters are also held fixed). Shown are the CMB temperature (top left) and polarization (top right) auto-correlation and cross-correlation (bottom left, showing difference rather than fractional difference) angular power spectra, and the CMB lensing power spectrum (bottom right).

2012.11830

spectra_Pb_2FSb_ref_varying_Neff_combo_diff.jpg

0.432694

ff990fc312934a4bafbdb786bc6ba04f

Fractional differences of the CMB temperature (top left) and polarisation (top right) auto-correlation and cross-correlation (bottom left, showing difference rather than fractional difference) angular power spectra, and the CMB lensing power spectrum (bottom right). All cases are compared to a free-streaming neutrinos comparison case with varying width of decoupling Δ of 10% (more red) to 80% (more yellow) of the decoupling redshift of =100,000.

2012.11830

spectra_Pb_2FSb_ref_varying_dzdec2_combo_diff.jpg

0.418176

11c80b5be7954de58259318308bb57b5

Fractional differences of the CMB temperature (top left) and polarisation (top right) auto-correlation and cross-correlation (bottom left, showing difference rather than fractional difference) angular power spectra, and the CMB lensing power spectrum (bottom right). All cases are compared to a free-streaming neutrinos comparison case with varying width of decoupling Δ of 10% (more red) to 80% (more yellow) of the decoupling redshift of =10,000.

2012.11830

spectra_Pb_2FSb_ref_varying_dzdec_combo_diff.jpg

0.438061

2f0fa8c76c934fc5bf8e7b6e124036d7

Fractional differences between the CMB power spectra computed in a model with self-interacting and free-streaming neutrinos (Case 2, see Sec. <ref>), as compared to a model with only free-streaming neutrinos. Each curve shows a different assumed value of , with all other parameters held fixed. Shown are CMB temperature (top left) and polarization (top right) auto-correlation and cross-correlation (bottom left, showing difference rather than fractional difference) angular power spectra, and the CMB lensing power spectrum (bottom right). In all panels, the Planck 2018 binned data <cit.> are shown in black, for comparison.

2012.11830

spectra_Pb_2FSb_ref_varying_zdec_combo_diff.jpg

0.446805

499cb0c406f840f9811a8d85abb9a357

The relationship between the redshift of neutrino decoupling, z_dec, and the effective neutrino self-interaction strength G_ν. Plotted is the relationship for different numbers of interacting neutrino species. The red, orange, and green lines correspond to 1, 2, and 3 self-interacting neutrino species, respectively. The gray line is result from dimensional analysis, z_dec∼ (G_ν^2 T_0^3 m_pl)^-1/3.

2012.11830

zdecvsGnu.jpg

0.432418

3bd9336316d14b9492f60a08ad8f2657

The results of evolutionary learning (genome bottom and phenome top) instructed to produce (a) 4-mers, (b) 5-mers and 5-gons, (c) 6-gons, and (d) 12-gons.

2012.11832

collection.jpg

0.478878

187c2636397a4b87be092e8ca5e4e979

Evolutionary learning directed to maximize the number of 12-gons, convex pores of 12 sides (shown light blue in images). The main panel shows the yield ϕ, as a function of generation n, produced by the most successful genome (potential and protocol). The particle symbols are the data points, which also show the form of the learned potential. Below the data points are plots of the learned protocol (red and blue lines) and resulting yield (green lines) as a function of time t. The format of those plots is shown bottom right. Large positive values of ϵ indicate strong particle attractions, and large positive and negative values of μ promote dense and sparse substrates, respectively. The snapshots at the top show portions of a simulation box from three different generations, indicated by the blue lines. Parameters: ϕ_0=90 12-gons, ϵ_0=μ_0=20, n_0=27 generations, t_0=10^9 Monte Carlo steps.

2012.11832

fig_12_gons.jpg

0.428927

d6f7993994ec430f83b096afd6af712f

Self-assembly dynamics produced by the most successful genome (potential and protocol, shown top left) from generation 27 of fig_12_gons. N_3 and N_12 are the number of 3-gons (red) and 12-gons (light blue). The snapshots are taken from the times indicated by the blue lines. 12-gons form relatively late in the trajectory, emphasizing that self-assembly is a “sparse reward” problem: the objective ϕ can be zero for most of the simulation. Evolutionary methods that require information from only the final time point of the simulation are natural ways to tackle such problems. Parameters: t_0=10^9 Monte Carlo steps.

2012.11832

fig_12_gons_dynamics.jpg

0.45409

57a31c093a4a45d38381fbb7a112f097

Evolutionary learning with ϕ=C_4, the number of 4-mers, interacting clusters of 4 particles (shown darker in images). The format of the figure is the same as fig_12_gons. Parameters: ϕ_0=350 4-mers, ϵ_0=μ_0=20, n_0=17 generations, t_0=10^9 Monte Carlo steps.

2012.11832

fig_4_mers.jpg

0.447571

722976bf70044373a01e21017c647124

Evolutionary learning with ϕ=min(C_5,N_5), the number of 5-mers (shown darker in images) and 5-gons (shown light blue). The format of the figure is the same as fig_12_gons, with the exception of the top-left image: that shows the outcome of learning after 17 generations with ϕ=C_5. Evolution of pentagonal clusters requires the dual goal of 5-mers and 5-gons. Parameters: ϕ_0=125 5-mers & 5-gons, ϵ_0=μ_0=20, n_0=17 generations, t_0=10^9 Monte Carlo steps.

2012.11832

fig_5_mers_gons.jpg

0.443591

cbfa4ab8df114fdc85f37edc1a423fc3

Evolutionary learning with ϕ=N_6, the number of 6-gons, convex pores of 6 interacting particles (shown light blue: see boxed image). The format of the figure is the same as fig_12_gons. Parameters: ϕ_0=700 6-gons, ϵ_0=μ_0=20, n_0=17 generations, t_0=10^9 Monte Carlo steps.

2012.11832

fig_6_gons.jpg

0.49572

d6b462be85094971becf06726cf7d6d2

Geometry for the interaction potential. The white arrow indicates the line from the center of the particle to the point on its circumference specified by θ=0.

2012.11832

fig_potential.jpg

0.405748

0a9218ca0b044e6cbcd2d97d65bd67c2

The results of evolutionary learning instructed to produce novelty in the space of 3-gons and 4-gons. The scatterplot shows the coverage obtained in this space after 10 generations of learning; the images show examples of the structures produced (the particles underlying the polygons are not shown).

2012.11832

novelty_search.jpg

0.445454

5387deb08f0d4fe9aba5e1e79ec7a0a3

One of the materials identified by the novelty search of fig_novelty shows the σ, H, and Z binding motifs characteristic of dodecagonal quasicrystalsv̧a̧ņ2̧0̧1̧2̧f̧o̧ŗm̧a̧ţi̧o̧ņ. To produce this figure we used the particle- and protocol design identified by the learning algorithm of fig_novelty, and ran a trajectory of length 20 t_0. The particles underlying the pattern of 3-gons and 4-gons are not shown.

2012.11832

novelty_search2.jpg

0.519876

3ee2d3cc74ad40cba9908756d0ea154d

(a) We express the angular interaction potential and the time-dependent self-assembly protocol for a set of model molecules in the form of two neural networks, which together comprise the “genome” for making a material. In images, attractive portions of particles are shown green. (b) Upon specifying a design goal, neuroevolutionary learning can produce a genome whose “phenome” – the result of molecular simulations carried out using the potential and protocol specified by the genome – is a material satisfying that goal.

2012.11832

schematic.jpg

0.482825

e7b5cc26a4bc4fd5b415a4ee80ac31bf

(a) Yield of the self-assembly process using the best generation-27 genome (particle and potential) of fig_12_gons (blue line) compared to the yield obtained using a human-designed particlew̧ḩi̧ţȩļa̧m̧2̧0̧1̧6̧m̧i̧ņi̧m̧a̧ļ combined with evolutionary learning of the protocol (light blue). Also shown are three simulations (red) using a human-designed particle and protocols, the latter consisting of slow cooling at three rates at fixed chemical potential. (b) Yield of the self-assembly process using the best generation-18 genome from fig_collection(c) and fig_6_gons (blue line), compared to that using a particle with perfect three-fold rotational symmetry combined with evolutionary learning of the protocol (light blue).

2012.11832

yields.jpg

0.427807

d69afd447c72415ab522a6546fc2f9d0

The mean number of photons difference in percentile at the exit of the cell between α_a = 0.1 and α_a = 0 versus the FWM gain G(t=1) for (a) probe mode, (⟨â^†â⟩_t=1,α = 0 - ⟨â^†â⟩_t=1,α = 0.1)/⟨â^†â⟩_t=1,α = 0, and (b) conjugate mode, (⟨b̂^†b̂⟩_t=1,α = 0 - ⟨b̂^†b̂⟩_t=1,α = 0.1)/⟨b̂^†b̂⟩_t=1,α = 0. (c) The expectation value difference in percentile for the quantum correlation |⟨âb̂⟩| between α_a = 0.1 and α_a = 0, i.e., (|⟨âb̂⟩|_t=1,α = 0 - |⟨âb̂⟩|_t=1,α = 0.1)/|⟨âb̂⟩|_t=1,α = 0, versus the FWM gain.

2012.11839

Response.jpg

0.552872

3fd8cf934d1541e2a08cd3b178821185

TMSV beam (light cone) power as a function of pump power. Dashed line is the theoretical fit according to Eq. (<ref>) with r ≡α P, where P is the pump power.

2012.11839

Ring_vs_Pump1.jpg

0.432621

393cccd116a4460da8e147dfc1d7e3c0

SHG signals induced by 80 μW partially-decoherent TMSV beams of light (blue dots, left-hand side y-axis) and 80 μW coherent beams of light (red dots, right-hand side y-axis) as a function of transmission of ND filters. Dashed blue line is the theoretical fit according to Eq. (<ref>).

2012.11839

SHG_Decoh1.jpg

0.426057

d742a027610b45b6a3d63179546c1378

SHG signal as a function of the power of the light cone (shown in Fig. <ref>(c) as the `ring'). Dashed line is a polynomial fit according to Eq. (<ref>) with ND = 0.

2012.11839

SHG_vs_Ring1.jpg

0.452936

83d771e1c48b4aa5a263674245cf7857

(a) Experimental setup in which a CW laser-pumped ^85Rb vapor cell produces a TMSV state via the FWM process. The TMSV beam (i.e. the `light cone') is separated from the pump beam by a ∼ 2×10^5 :1 polarizing beam splitter after the cell. The SHG signal from the BBO crystal is collected by an EMCCD camera. Two low-pass filters are mounted in front of the camera to eliminate undesired excitation photons. The BBO crystal and the EMCCD camera are enclosed in a light-proofing box to block ambient light. PBS: polarizing beam splitter, ND: neutral density filter, LP: low-pass filter. (b) Level structure of the D1 transition of ^85Rb atom. The optical transitions are arranged in a double-Λ configuration, where ν_p, ν_c and ν_1 stand for probe, conjugate and pump frequencies, respectively, fulfilling ν_p + ν_c = 2ν_1 and ν_c - ν_p = 2ν_HF. The width of the excited state in the level diagram represents the Doppler broadened line. Δ is the one-photon detuning. ν_HF is the hyperfine splitting in the electronic ground state of ^85Rb. (c) Image of the cross-section of the TMSV state, i.e., the `light cone', where the bright central spot is the residual pump beam. The angular separation between the center of the ring and the center of the residual pump beam is ∼ 0.3^∘, and the width of the ring is measured to be ∼ 0.05^∘.

2012.11839

Setup_Combo1.jpg

0.446683

3a04e2a545ee418fb39c9ccfaf26bdcd

Theoretical decoherence characterization of the TMSV state via its SHG signal. To isolate the effect of decoherence on the quantum correlation ⟨âb̂⟩, experiment is performed by holding power p constant, i.e., p≡ p_0.

2012.11839

TheoryFig.jpg

0.451007

b796395fa0bb415a978d829796fc85a5

Overview of our approach. We formulate selective forgetting by four datasets; _r, _f, _f, clean, and _r, extra. Depending on which dataset the DNN should perform well, three patterns of forgetting are introduced. In Pattern A, the DNN forgets samples such as outliers. As illustrated in the figure's left side, images of mice that are input devices are outliers when the task is classifying animals, for example. In Pattern B, the DNN forgets a part of each training sample. The model is supposed to forget to use the sunglasses as a feature for classification in the figure. Pattern C is similar to Pattern B; in the figure, the model is supposed to forget to use the logo or the emblem of a carmaker for classifying types of cars. Pattern B and C are different in their evaluation. Example images are from WebVision <cit.>, CelebA <cit.>, and Cars Dataset <cit.>.

2012.11849

fig_main.jpg

0.390342

6c2a28b304c344dca33a3b25a30402b2

Tile, FashionMNIST: Backdoor with a RLD factor 1.

2012.11849

rld.jpg

0.461641

31a3c4c59c1649bd981df472e5bd5374

Distribution of softmax(f_θ(_f)). We use Fashion-MNIST and the forgotten class is 0.

2012.11849

softmax.jpg

0.408781

dfe2653ada044e28b08994e50c7c5b94

Case D: the solution path of the MTIP.

2012.11855

CaseA1.jpg

0.473503

e024ad037e114b89829b2046a7ffa3e3

Case D: The profiles of F[E(t)], L^-[E(t)], and L^+[E(t)] with respect to time t.

2012.11855

CaseA2_new.jpg

0.478084

62a4d39d259140e4ba71fe8552de6999

2012.11855

CaseA3.jpg

0.469146

361afe98f1674c2e95110a6011d1b8fe

Case A: The profile of F[E(t)] with respect to time.

2012.11855

CaseAA1_new.jpg

0.43617

ff93a16528c64b1682869c3d694b1bd9

Case A: The solution path of the MTIP.

2012.11855

CaseAA2.jpg

0.416942

342898d8f6154be0a96ea5f0b67202ed

2012.11855

CaseAA3.jpg

0.445203

127ed7f5e7fc4e3fbf928f379f4b7259

Case B: The profile of F[E(t)] with respect to time.

2012.11855

CaseBB1.jpg

0.477351

e414ebddf53e4bb1867a795bd066fc54

Case B: The solution path of the MTIP.

2012.11855

CaseBB2.jpg

0.528094

caa21fbb3065480cb62ecadf79c2b83d

2012.11855

CaseBB3.jpg

0.432309

f0bcf51e0ca448199ca4df0505232ac7

Case C: The solution path of the MTIP.

2012.11855

CaseC2.jpg

0.484183

76da0b60e18640f599796d5d738fb952

2012.11855

CaseC3.jpg

0.45327

8b99c0b3177f4ff1b01bbd071ee250b4

The geometry for the path of RDP with terminal points E(t̅+ε), E(t̅), and E(t̅-ε).

2012.11855

Intersection.jpg

0.507986

606d4dbd0ab2409aab9bb6c9b5c3137b

The geometries of the feasible Dubins paths related to F, L^-, and L^+.

2012.11855

L1L2.jpg

0.502135

ce3637f3bc9a441eb91bfa3a4d3a4f48

The geometry of the solution path of RDP with a type of LR.

2012.11855

Path_LR.jpg

0.484622

85a45fba0763460a940a99f389d572a6

The geometry of the solution path of RDP with a type of RS.

2012.11855

Path_RS.jpg

0.501143

345ee61253a3448388f32b768974f17f

The geometry of the circles 𝒞_r and 𝒞_l and the subregions ℛ_1, ℛ_2, and ℛ_3 in the 2-dimensional plane.

2012.11855

Subregions.jpg

0.516921

3044a60f28454ace87ce7d21a436d03f

The regions for which the path of RDP terminates by different types. <cit.>

2012.11855

Types_Region.jpg

0.447366

6a3b54c50c9f40b6b420983fed14789a

Figure showing the performance of a multi-channel LoRaDRL and LoRaSim scheme under frequency jamming attack. A small drop of performance is seen in the case of LoRaDRL as it shifts to the other available frequency. While in the case of LoRaSim, a sudden drop of performance can be seen because of the absence of a feedback loop.

2012.11867

freq_blocking.jpg

0.999988

6b93f3684b2b40faa22218f748ca9d4a

Mapping of our problem setup on DRL. The gateway is considered the agent and the LoRa network represents the environment.

2012.11867

loradrl.jpg

0.467528

bd77cc25d1f941889c0b63b22ec54eb4

Architecture of LoRaWAN consisting of LoRa EDs, LoRa gateway, network server, and end-user. The EDs serve different purposes and transmit the data to the gateway based on the application requirements. The received data packets are forwarded to the network server, which in turn forwards them to the end-user.

2012.11867

lorawan.jpg