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
0.603154
daa7fadbd9424b09ae7ae9bc66d393f1
Contraction of the descending superoperators in hyMERA are not as computationally costly as in MERA, due in large part to the holographic causal cone structure of hyMERA. More details can be found in <cit.>. Some of the descending superoperators in the {5,4}-version of hyMERA are shown above.
2012.09591
some-descend.jpg
0.50096
ad95921caeb242b4be58de5a45488714
Subfigure b) displays the causal-cone setup that one would use in order to minimize for the groundstate expectation value of a local Hamiltonian H_local; tensors inside of the causal cone can be iteratively optimized by applying the vertex-inflation strategy as described above. However, the same sequence of coupling constants that appear in the local Hamiltonian used for the iterative-optimization procedure is not guaranteed to be found on another subregion of the hyMERA boundary; therefore, one can expect widely varying subsequences of couplings constants (and thus varying local Hamiltonians). As shown in Subfigure c), One may attribute the optimized portion of the network with containing a local Hamiltonian H_local, as in Subfigure b). As such, n-point function calculations may not yield the expected results pertaining to the localized groundstate optimization, since at least one of the observables in the calculation may be located in a subsection of the boundary with a non-matching local Hamiltonian H'_local. This different local Hamiltonian is comprised of a different subsequence of coupling constants, and thus may not exhibit the same groundstate characteristics as the original boundary subregion.
2012.09591
tensor_causal_opt.jpg
0.531353
6442fcde7a744ba89fdb3c6a198b0c84
A vertex inflation of a {5,4} hyperbolic bulk tessellation, which shows how to "grow" a regular tessellation on a hyperbolic manifold in 2D. Blue As represent the first layer of vertex inflation, as shown in the starting pentagon's vertices. Subsequent layers are tiled around the central polygon by applying the necessary inflation rules, as depicted in the green- and magenta-lettered layers of the resulting bulk tessellation. This procedure is described in more detail by <cit.>. We add tensors to the vertices and edges to the entire bullk of the resultant hyMERA network. If we trace the deflation steps from the boundary toward the center, a string sequence of the form {ABABAAB…} is observed to renormalize into effective couplings {B'A'…} as we scale up the network. This effect can be verified by inscribing a causal cone, as the red-dotted lines in the upper-right quadrant show. Subfigures a)-d) show the successive tiling of the Poincaré disk by applying the inflation rules in a {5,4} regular hyperbolic tiling; Subfigure a) begins with the first layering, and subsequent layerings are added as the thermodynamic limit is reached.
2012.09591
vertex_inflation4.jpg
0.461334
2bf03fec8d0a4c0f9c2ae21fd580b9e6
Typical wireless scene for cyber-physical systems. The system suffers from eavesdropping attack and spoofing attack, which need encryption and authentication.
2012.09593
fig1.jpg
0.420921
050a247cc84e40a38966fcf0e9aa6564
The influence of SNR on authentication probability under different MIMO channels. Sparsity level is 2%.
2012.09593
fig10.jpg
0.424261
438e7adb6f3e45718364ab571ca71512
The influence of data loss ratio on authentication probability under different MIMO channels. SNR = 40dB.
2012.09593
fig11.jpg
0.42491
e8c8fa525dc741a88fb4b57f89220257
The influence of data loss ratio on recovery probability under different sparsity level with 4 MIMO channels. SNR = 40dB.
2012.09593
fig12.jpg
0.514651
6665ea21f4804840bf02e08c2a41d2b4
Overall diagram of the proposed concurrent encryption and authentication method.
2012.09593
fig2.jpg
0.471961
b73ba376c1234e7885465c0425bf1ff9
The influence of cyclical shift times on measurement matrix. (a) distribution after shifting L = 18 times; (b) distribution after shifting 6L times; (c) true Gaussian distribution.
2012.09593
fig3.jpg
0.39645
bcce2699cf1744d4887a6569ddb9af3d
The original signal and the recovered signal.
2012.09593
fig6.jpg
0.371989
9ebbea479aaf4ab5b1f2bb8cf206c8f0
The transmitting message which contain both message information and tag information.
2012.09593
fig7.jpg
0.414017
6c48f7295e1b4303b3f6d101cc279e99
The influence of SNR on recovery probability under different number of MIMO channels. Sparsity level is 2%.
2012.09593
fig8.jpg
0.476894
351ea2fc090e472cbe1e735ee911c57d
The influence of sparsity level on the recovery probability under perfect transmission.
2012.09593
fig9.jpg
0.426065
e30e671d595c4dcd9260cf6bfdc7f790
Differential sensitivity, simplified regression coefficient (SimRC; left) and correlation (center, right) for five uniform prior intervals. Forward responses are computed using the non-linear toy model.
2012.09595
Compre_UniormPrior.jpg
0.422753
48ddd3ad686e4d55a02c2e880ea96e20
Differential sensitivity, simplified regression coefficient (SimRC; left) and correlation (center, right) for four different correlation lengths of a uniform multivariate Gaussian prior distribution. Forward responses are computed using the linear toy model.
2012.09595
LinearModel_priorconstrains.jpg
0.418236
ec7d23ff51284eb78efccb2671480588
Differential sensitivity, simplified regression coefficient (SimRC; left) and correlation (center, right) for three different ensemble sizes. Forward responses are computed using the linear toy model.
2012.09595
LinearModel_samplesize.jpg
0.485827
cf39e1a122f64a69aeca7184f74dfabe
a) Differential sensitivity, simplified regression coefficient (SimRC; left) and correlation (center, right) for five prior standard deviations (STD). Forward responses are computed using the non-linear toy model; b) Zoom-in of a).
2012.09595
NonLinModel_GaussSTD.jpg
0.406936
b460c94258c24340b086f20d96a9f7d1
Differential sensitivity, simplified regression coefficient (SimRC; left) and correlation (center, right) for a uniform Gaussian prior distribution and the different simulated measurement signals for four FDEM receiver coils with different distance to the transmitter coil.
2012.09595
comparefianl_EMmodel.jpg
0.420606
79aee532f53c470fbe5f25292011bdb4
Differential sensitivity, simplified regression coefficient (SimRC; left) and correlation (center, right) for four different correlation lengths of a uniform multivariate Gaussian prior distribution. Forward responses are computed using the non-linear toy model.
2012.09595
fig_corrlengthvariation_Gaussiankernel.jpg
0.401186
9c0299e1e5dc4c4f8e3b60c0dc11e726
Forward response for the FDEM forward model around the Gaussian prior distribution mean.
2012.09595
linearization.jpg
0.449566
8a0496d3cc1b42c0a8f6c468fb52a98a
Shows the Spearman rank correlation between the inferred ranking from each of these models, along with existing methods, and the known true ranking.
2012.09598
fig-5b-spearman-sim-1.jpg
0.473874
c70caac658a74a0097fc9aaba7b55625
An example of a win/loss matrix and the corresponding reordered matrix according to the I&SI method. The entries shaded in red in the matrix are the inconsistencies, where the lower-ranked individual wins more frequently than the higher-ranked individual.
2012.09598
win_loss_matrix.jpg
0.443765
d2c40fedd048464d83f5ea8aa0688e03
2012.09599
S11.jpg
0.469273
083c2cb8578c49269ef7c84f5e3fbeae
2012.09599
S15.jpg
0.531255
4ec916da54a242dea068047a1490ba44
J_1, J_2, J_3 is the green, red and blue circle, respectively.
2012.09599
S16.jpg
0.390774
6984bc8eff704c39bca09f0925f917c7
2012.09599
S17.jpg
0.541729
178bb99f1f8a4770b6bfc229c463395f
In the first two drawings, we have different diagrams of the same link. The last two drawings show the effect of a positive half twist followed by a trivial Dehn filling on the first two diagrams. We see that the results are different.
2012.09599
S18.jpg
0.452878
285ca503bc3b42249fdcb9d7bda08f6f
2012.09599
S25.jpg
0.473184
444ac82d3bb643f1a17c755193526300
2012.09599
S3.jpg
0.466201
ae0a42141c3840db913138b83e2da7a5
2012.09599
S5.jpg
0.450867
91822ad39ed14b0b8e8ea761d06d3fc4
2012.09599
S6.jpg
0.473937
f87aaa117ca44f6fa481ace2e619022e
2012.09599
S7.jpg
0.47428
dffb0abda00d47a7bc78e5f07e4fec98
2012.09599
S8.jpg
0.437075
60b83cc27da0420db46e6294957339f3
The sensitivity of DFCN with the variation of λ on six datasets.
2012.09600
f11.jpg
0.406426
c0d96efd7ff345cd919bff72c0d3396d
Illustration of the Structure and Attribute Information Fusion (SAIF) module.
2012.09600
f3.jpg
0.416713
559843ee262b45fcb9cfa603a5e1372f
Clustering results of the graph autoencoder with different reconstruction strategy. GAE-L_w, GAE-L_a, and IGAE correspond to the reconstruction of weighted attribute matrix, adjacency matrix, and both.
2012.09600
f4_2.jpg
0.399938
5baca41fc94947409edf31d95241ae16
Ablation comparisons of cross-modality dynamic fusion mechanism and triplet self-supervised strategy in SAIF. The baseline refers to a naive united framework consisting of AE and IGAE. -C, -S, and -T indicate that the baseline utilizes the cross-modality dynamic fusion mechanism, single or triplet self-supervised strategy, respectively.
2012.09600
f5_1.jpg
0.403739
bd07489d1c504b7096a56e9609ff44e7
2D visualization on six datasets. The first, second, and last row correspond to the distribution of raw data, baseline and DFCN (baseline + SAIF), respectively.
2012.09600
f8_1.jpg
0.53237
9216db304eb348ae9c04d7a8c606f6f9
Network structure comparison. Different from the existing structure and attribute information fusion networks (such as SDCN), our proposed method is enhanced with an information fusion module. With this module, 1) both the decoder of AE and IGAE reconstruct the inputs with a learned consensus latent representation. 2) The target distribution is constructed with sufficient negotiation between AE and IGAE. 3) A self-supervised triplet learning strategy is designed.
2012.09600
f9_4.jpg
0.425613
6b66608a2de04c949bc41e1ed0e25f16
NAP for pedestrian. Left is with classification confidence 95% while right is with 80%
2012.09602
NAP_pedestrian.jpg
0.473005
87d7e65c5a014933a936489a76eb1745
NAP for vehicle. A is with classification confidence 95% while B is with 80%
2012.09602
NAP_vehicle.jpg
0.415365
b141cb9c226148f891fbe73e214593fb
Dispersion curves of t_K as a dashed-dotted line and of ε_b(K) as solid, dashed and dotted lines, associated with various U values, respectively; those data have been obtained with t_K=tcos(Ka/2), where t,a are the one-electron bandwidth and the lattice parameter, respectively.
2012.09611
coop.jpg
0.490062
bdeac3beeab94967ba5a24bc13407c87
Plots of E_F(T_c,c_0),t_K(T_c,c_0),U(T_c,c_0) calculated for T_c=1K and t=3eV; the unit for c_0 is the number of conduction electrons per atomic site; δ f with f=E_F,t_K,U is defined as δ f=|1-f(300K,c_0)/f(1K,c_0)|; the scale is linear for E_F,t_K,U but logarithmic for δ E_F,δ t_K,δ U.
2012.09611
uef.jpg
0.487286
a5a1467b51364f9b943ab4545bc8131d
Estimated MMD^2 values plotted against parameters of the S-V model. The parameters are uniformly sampled 200 times from the prior range one at a time, keeping the others fixed to the true values denoted by the dark green lines. See Tab. <ref> for the prior ranges.
2012.09612
MMDvParam_4moments.jpg
0.476705
e079a23897354641a8ef1fbb12629c47
Averaged power delay profiles simulated from the PG model for different SNR levels.
2012.09612
SNR_APDP.jpg
0.438172
2a10acfef7734ae783e823fbcbabffcb
Methodologies for calibration of stochastic radio channel models: (a) State-of-the-art methodology based on multipath extraction and clustering; (b) proposed method based on generic summaries (here exemplified by log-temporal moments) and approximate Bayesian computation (ABC).
2012.09612
introFigure.jpg
0.489581
09caa5f70444483b832147cde8780b05
Given a kernel k, the distributions ℙ and ℚ are mapped to their kernel mean embeddings μ_ℙ and μ_ℚ using Equation <ref>. The MMD is obtained by computing the distance between μ_ℙ and μ_ℚ in the function space ℋ_k, as expressed in Equation <ref>. This figure is inspired by <cit.>.
2012.09612
kernel_embedding.jpg
0.478065
593ff53accc8417990076aaea4d04e06
Local linear regression adjustment of parameter θ inspired from <cit.>. First, the regression model is fitted based on accepted parameter and statistic values. Then, the parameters are adjusted based on the fitted model, which can move them outside the prior range if (s_i - s_obs) is large.
2012.09612
mismatch.jpg
0.412215
bb52b9ea34bb4311a9f949e4caa701fe
Boxplots of the estimated MMD^2 between 𝐗_true (= 1000) and 𝐗' as a function of computed by repeating the experiment 100 times for each value of . 𝐗' is generated from ' = [2× 10^-8, 6× 10^7, 10^8, 2× 10^-8, 10^-9, 5× 10^-10]^⊤ and 𝐗_true from _true = [5× 10^-8, 2× 10^7, 10^9, 10^-8, 2× 10^-9, 10^-9]^⊤. The dashed green line is a corresponds to the value of the MMD^2 being approximated. Since this value is not available in closed-form, it is approximated by using = 10^4.
2012.09612
mmd_nsim.jpg
0.466667
972037dbff8c4ca3abbd48b3e62ecd9c
Mean (left) of 𝐳_0 versus 𝐳_1 simulated from the S-V model, along with the corresponding observed summary computed from the measured data (red). The mean and variance of z_0 is unitless, while that of z_1 is [log s] and [(log s)^2], respectively. The observed summary lies in the point cloud generated by the model. In contrast, the S-V model is not able to replicate the higher moments of the data, as seen from the variance plot (right), indicating model misspecification. Each of the 2000 simulated summaries correspond to one parameter drawn from the prior.
2012.09612
scatter.jpg
0.441186
2e9d415531fc4a1788d9c1feee9d4d26
Violin plots of ABC posterior samples of PG model parameters after T=10 iterations for different SNR levels. The APDP corresponding to each SNR is shown in Fig. <ref>.
2012.09612
snr.jpg
0.458635
44954e64266d483a9538d0295c281fe9
Violin plots of the ABC posterior samples of PG model parameters as a function of PMC iterations for measured data.
2012.09612
violin_meas_pg.jpg
0.421442
ab4aed05939b4e3b8a19605c01514aa6
Violin plots of ABC posterior samples of S-V model parameters as a function of PMC iterations for measured data.
2012.09612
violin_meas_sv.jpg
0.48398
493c3956acf1442b969cb2d787a597f1
Violin plots of ABC posterior samples of PG model parameters as a function of PMC iterations. A violin plot is similar to a box plot with the addition of a rotated kernel density plot on each side. _true = [0.6, 15, 0.5, 10^-9]^⊤ is denoted by the dark green dashed line.
2012.09612
violin_sim_pg.jpg
0.432287
bdf87c8f5e2b412280bec6cf9cac74d8
Violin plots of ABC posterior samples of S-V model parameters as a function of PMC iterations. Note that a violin plot is similar to a box plot with the addition of a rotated kernel density plot on each side. The dark green lines denote the true parameter values _true = [5× 10^-8, 2× 10^7, 10^9, 10^-8, 2× 10^-9, 10^-9]^⊤.
2012.09612
violin_sim_sv.jpg
0.477867
0e5bdef16e4a4aa8be1eea9486e152f0
HTS-RL and SeedRL on GFootball 11 11 easy task.
2012.09849
ablation_step_reward.jpg
0.448007
2c0b52c693fc4d99ac88bc66a8c28a66
Empirical synchronization time.
2012.09849
exp_plot_gamma.jpg
0.493044
c7df4badbc694690a8efce66f9aa5f96
Structure and flow of (a) A3C, (b) GA3C, (c) IMPALA, (d) A2C and (e) our HTS-RL.
2012.09849
flow_all.jpg
0.529058
c6837ada00394929a5ae62b944dd0d37
Processing timeline of (a) A3C , (b) GA3C/IMPALA, (c) A2C and (d) our HTS-RL.
2012.09849
timeline_v3.jpg
0.412946
d83e8409e4ab4d6d847b87140873d438
Graph showing the daily downloaded data per user in March 2020 for those in metro and rural areas, as well as LA county.
2012.09850
daily_downloaded_data_notitle.jpg
0.488088
85b4b65f6b594888ad1377af84f8f574
Graph showing the monthly average downloaded data per user from January 2019 to June 2020, partitioned by test unit county population.
2012.09850
monthly_downloaded_data_notitle.jpg
0.433982
71dc812188434199817954468dbf8898
Graph comparing the 95% quantile packet loss per day for the months January, March 2019 and January, March 2020.
2012.09850
packet_loss_per_day.jpg
0.480971
3c0f92013ea64292ba7a07a553ed4d83
Graph comparing the 95% quantile packet loss per hour for the months January, March 2019 and January, March 2020.
2012.09850
packet_loss_per_hour.jpg
0.437047
72a7b3ff7f204eefacaccad0404262af
Design Space Exploration. A top-k engine with a parallelism of 16 and 196KB key/value buffer is sufficient for our settings.
2012.09852
ab2.jpg
0.420352
77e3c68c8b6c401e8f2aca33601ae98c
SpAtten Architecture Overview. Modules on the critical path (6,7,8,10,11) are fully pipelined to maximize the throughput.
2012.09852
arch_overview_new.jpg
0.465394
1034a25af42546fea5dc60d7529f78cb
NLP model architecture with attention. only contains the summarization stage. contains summarization and generation stages.
2012.09852
background.jpg
0.401092
a7a7718d7d1041dd84953f46b13d9b3f
Co-designed Transformer has slightly more attention but much less FC computations because of 's high efficiency on attention.
2012.09852
barhat.jpg
0.421703
a1efc0ccb0894d509c62caef1570a574
Sample signal shapes of the static (left column), dynamic (middle column) and stochastic (right column) signals from DM subhalos with c = 100, 10^3, 10^4 and the Primordial Black Hole limit c →∞. The top row shows the unsubtracted signal shapes, computed from Eqs. (<ref>) and (<ref>), while the bottom row shows the subtracted signal shapes after the terms degenerate with the timing model have been removed. The first two columns show the phase shift δϕ(t), Eq. (<ref>), from a single transiting subhalo and the last is the signal shape from multiple transiting subhalos, Δϕ, in Eq. (<ref>). The signal shapes are taken from a realization of the Monte Carlo. The mass of the subhalos are 10^-2 M_⊙, 10^-11 M_⊙ and 10^-14 M_⊙ for the static, dynamic and stochastic signals respectively. Only the closest subhalo is kept for the static and dynamic signals, while all subhalos are kept for the stochastic signal.
2012.09857
large_signal_plot.jpg
0.44251
3bcfc08b2642414c903cd0949ba7a85c
Comparison of the PBH constraints between this work and <cit.>. The left panel compares the pulsar term results while the right panel compares the Earth term results. The meanings of the labels are described in the main text. The pulsar parameters used here are N_p=200, Δ t=2, t_rms=50 and T=20. The SNR thresholds are set to SNR=4 for the pulsar term and SNR=2 for the earth term for consistency with <cit.>.
2012.09857
may_new_comparison_cropped.jpg
0.392618
a5731da1890f4b12ac3664589663bea2
Limits from PTAs on the dark matter mass fraction f = Ω/Ω_DM in subhalos of mass M. The left panel compares different subhalo concentration parameters, c = 100, 10^4, 10^8, and the PBH limit, c→∞, as well as the different search types, “Earth" and “Pulsar". The right panel shows how the constraints change when different assumptions are used when computing the SNR for the Earth and the Pulsar terms discussed in Sec. <ref>. For the pulsar term we compare the constraints from keeping only the closest subhalo around each pulsar, labelled “Closest only", to the constraint obtained from keeping all of the subhalo contributions, “All". For the Earth term we compare the optimal filter to the one generated by the MC to understand the mis-filtering effect. The PTA parameters used were N_P=200, Δ t=2, t_rms=50 and T=20.
2012.09857
reach_plot.jpg
0.440522
1021e94ac538430d953d78085078e2a6
Surrogate optimality condition satisfaction graph during the SLR training of ResNet-18 on CIFAR-10 for 50 epochs (1: satisfied, 0: not satisfied). The condition is satisfied periodically.
2012.10079
SOC_0.01.jpg
0.438874
938f8b6c2de44eea9f28fdb04ff46a42
Hardpruning accuracy on TuSimple benchmark with ADMM vs. SLR training for several compression rates. SLR has a greater advantage over ADMM as compression rate increases.
2012.10079
lane_seg_comp.jpg
0.502027
433fcc40eb9e49888b259d6757933af5
Cost convergence
2012.10083
character.jpg
0.45196
2e1df669246b412dbab353b7a8bc417e
Database of projector's SPDs
2012.10083
projector_light_.jpg
0.45514
e89235260c3348339873757289648d01
Camera sensitivity functions
2012.10083
sensitivity.jpg
0.427432
b9ea908441084a8b8c3f890ec856887f
The schematic diagram of our proposal. (a) Experimental setup; (b) Level diagram of atom and light configuration.
2012.10098
Fig1.jpg
0.433789
cb65637e31a94404b289dd5ece574a53
The energy spectrum for different COM momentum. Different color lines correspond to different λ. (a) δ>0, λ^2<ω E_r/4 for the red line, λ^2=ω E_r/4 for the blue line, λ^2=ω/4(E_r+δ) for the green line and λ^2>ω/4(E_r+δ) for the magenta line. The red dot is the energy minimum for the first three cases, which is at p_min=q_r/2, the magenta dot is the energy minimum for the last case, which is at p_min<q_r/2; (b) δ=0.λ^2<ω E_r/4 for the red line, λ^2=ω E_r/4 for the blue line, λ^2>ω E_r/4 for the green and magenta lines. The red dot is the energy minimum for the first two cases, which is at p_min=q_r/2, the green and magenta dots are the energy minimum for the last two cases, which is at p_min=0. The inset shows the expectation value of cavity photon number ⟨ n_c⟩/N for the eigenstates with different p where δ=0 and λ=√(ω E_r)/2.
2012.10098
Fig2.jpg
0.415392
d868dde8044e43adadfa421614891c98
Phase diagram. (a). The expectation value of cavity photon number ⟨ n_c⟩. (b). The expectation value of ⟨ S_z⟩. (c). The expectation value of cavity photon number for different δ. The red, blue, green, magenta line correspond to δ=0, δ=0.1, δ=0.2 and δ=0.3 respectively. (c). The expectation value of ⟨ S_z⟩ for different δ. The red, blue, green, magenta line correspond to δ=0, δ=0.05, δ=0.1 and δ=0.15 respectively.
2012.10098
Fig3.jpg
0.420563
db92c12942144ee4868e0d1c93c20461
The excitation energies as the function of λ. The parameters are all scaled to ω=m=q_r=1. The brown solid line and the blue dotted line correspond to ϵ_- and ϵ_+ respectively. (a) δ=0, the excitation energy is discontinuous. (b). δ=0.6, the excitation energies is similar with standard Dicke model.
2012.10098
Fig4.jpg
0.471475
78f791b8ccff45c088c27a81e0ef0fa5
The expectation value of cavity photon number ⟨ n_c⟩/N for different δ and λ in the system with (a). small atom decay rate γ=0.1 and (b). lager atom decay rate γ=2. (c). The expectation value of photon number α^2/N for different atomic emission rate γ. γ=0 for blue line, γ=0.1 for red line, γ=1 for green line. (d). Phase boundry λ_c for different parameters. κ=0.1 and γ=0 for blue line, κ=0.5 and γ=0.1 for red soild line, κ=1 and γ=0.1 for green soild line. All parameters are scaled to ω=q_r=m=1.0.
2012.10098
Fig5.jpg
0.388499
b17a88944c1748a5976d7743a8c15ff4
Phase diagram with the influence of nonlinear coupling U. (a). The energy spectrum with different COM momentum for δ=0. the blue line corresponds to U=0 and the red line corresponds to U>0; (b). The expectation value of cavity photon number ⟨ n_c⟩ when U>0, the red line represents the phase boundary, the white line represents phase boundary with U=0. All parameters are scaled to ω=q_r=m=1.0.
2012.10098
Fig6.jpg
0.401065
fb653354e4bf471482e0eb9d805fdd47
Illustration of our frequency density comparator. FDC has two encoder modules with shared weights. The encoder encodes the two input patches into latent space and outputs the embedding of frequency density. FDC then returns the substruction value of the two embeddings.
2012.10102
FDC.jpg
0.375558
14be26d7640447b088f3b77e03d29bad
Illustration of the degradation's importance in SR. Different blur kernels might get different degraded LR images, resulting in different SR models. For a degradation-unknown image 𝐱, only SR model (f_B) with a consistent degradation kernel generates natural HR result (Middle row), while those (f_A and f_C) with inconsistent kernels generate over-smoothing (Top row) or over-sharpening (Bottom row) images.
2012.10102
degradation.jpg
0.437878
447214101f6649e7acaf096e529f630e
Framework of the proposed frequency consistent adaptation. The adaptation generator takes the unknown degraded image as input, and generates LR image G(x), while the frequency density comparator and the wavelet discriminator provide frequency consistency losses to guide the adaptation generator.
2012.10102
estimator.jpg
0.425642
aadc45b00b8347bea38d20608a7d9255
(a) Degradation with different blur kernels show difference in frequency density. The frequency is increasing from middle of horizon axis. Domain A/C represents images degraded with small/big variance blur kernel, and it has a strong/weak density in frequency domain. Domain B is a consistent degradation, thus is close to source domain. (b) Impacts of down- and up-sampling on frequency density. Down-sampling on source domain with scale factor 0.5 might increase frequency density, and up-sampling with 2.0 impacts the opposite way.
2012.10102
freq.jpg
0.403368
eebe0d7088f748349e78338ec11e3d4e
Visual results with 4× compared with state-of-the-art SR methods on real images.
2012.10102
real.jpg
0.427648
8dd5566386e64c9b818e1822bfd92e71
Frequency density distributions of degraded LR and SR results by different methods on real-world images. The frequency distribution distance indicates that ESRGAN and KernelGAN generate inconsistent degraded images. Our FCA estimates proper kernels, thus obtaining LRs whose domain is closer to source domain.
2012.10102
real_freq.jpg
0.357266
019e7a5470ba4e248b966e41b4ebc567
Training and testing pipeline of real-world super-resolution. When applying a SR model to real scene, a typical pipeline includes: 1. Constructing LR-HR pair by performing degradation with a blur kernel. 2. Training a SR model f to reconstruct HR image 𝐈_HR for each 𝐈_LR. 3. Testing SR model on real-world image.
2012.10102
rwsr.jpg
0.442649
9bb31ce5cc024d4dbd248bdcbeea7587
4× SR results of different methods on a real image taken by smartphone. The red rectangle area of each result is cropped and enlarged for visual comparison. Compared with other results, our method generates natural HR image without blur or artifacts.
2012.10102
shortshow.jpg
0.454684
0dfb4a4b85484388bbe3619671f52bd1
Visual results with 4× compared with state-of-the-art SR methods on blur images with ISO.1.
2012.10102
synthetic.jpg
0.482024
a5ba30a61dbc42c6ac9d7ecdcc68b0a4
Diagrams of the ^3P_0 model that contributes to the decay of a meson into two mesons (a) and a baryon into a baryon and a meson (b).
2012.10105
3P0.jpg
0.430222
845295e3cfc24979baea35c0da822b14
Probabilities of different components in the physical X(3872) state obtained in the unquenched chiral quark model as a function of the mass. The mass is varied using the ^3P_0 strength parameter. Lines are DD̅^*, charged channel in black and neutral channel in red, χ_c1(2P) in dashed magenta and χ_c1(1P) in thin black.
2012.10105
EbindvsProbColor.jpg
0.429052
9a481faf6d3945dd91132f2501c13cbc
Bottomonium spectrum in the energy region of 2P and 3P states. Blue boxes shows the states in the Particle Data Group <cit.>. The states in red are the pure QQ̅ states predicted by the model.
2012.10105
bottomonium.jpg
0.470141
0ee820bb273d441eb3f2b5856bdbd1da
Charmonium spectrum in the energy region of 1P and 2P states. Blue boxes shows the states in the Particle Data Group <cit.>. The X(3915) has been included in the 0^++ and in the 2^++ channels since the J quantum number is not known. The quantum numbers of the X(3940) are also not known but it has been included in the 1^++ channel since it has been seen in DD̅^* but not in DD. The states in red are naive QQ̅ states predicted by the model.
2012.10105
charmonium.jpg
0.445569
a812d41a44484d08979a53305c89f03d
Euclidean dS_3 at a fixed time. The two asymptotic AdS_3 regions are depicted in purple and green, respectively. The UV brane is the middle slice and corresponds to the maximum dS_2 static patch (and we exemplarily show two additional static patches).
2012.10109
Dsds.jpg
0.438622
0580e714ace4464eb05bf9ca8e89c69f
Demonstration of nonlinearities. Left: γ(t)=sin(t-1) over σ(t)=sin(t). Right: γ(t)=sin(t-1) over σ(t)=sin(t-0.9)+sin(t-0.9)^3+2 sin(t-0.9)^5+sin(t-0.9)^7
2012.10109
LinearvsNonlinear.jpg
0.42881
c0573fa8a84a4f32b7111b13bb723a8f
Double logarithmic plot of the dimensionless magneto-vortical susceptibility M_5/T divided by the dimensionless chemical potential μ̃ in dependence of B̃. The dashed lines are quadratic fits.
2012.10109
M5g0LogLog.jpg
0.416057
9b74289f2ccc4fc0be508667e1ef82d4
The effective radius R_eff=R cos(β_ϵ) (blue) corresponds to the radius, where the points of the interval are antipodal by the basic definition of the cosine.
2012.10109
circle.jpg
0.400487
518fda2dfa484a5d922fe36dffc79beb
Different rates of convergence.
2012.10109
convergence_asgs.jpg
0.420659
06c06e2587544b1eb48500cae04bf9ea
Right: The dissipated energy E over the first 5 cycles (red to blue). Inset: Average over the first 5 cycles; low amplitude scaling ∼γ_0^2.
2012.10109
energy2N.jpg