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
|
|---|---|---|---|---|
0.429867 | 5b4ad4e7aea84f82bb5111591dd85a49 | Relative residuals in X-ray emissivity profile derived for the optimal ellipticity in the region [0.2,1] R_500 (ϵ=0.2), compared to the optimal ellipticity in the region above R_500 (ϵ=0.1). These optimal ellipticity values are obtained using the "uniform interior" data representation (as illustrated by the green curve in Fig. <ref>). For comparison, the relative residuals with respect to the spherical symmetry assumption are also shown. Here x versus y means: |x-y|/x, with x and y, the triaxial profiles derived for a given ellipticity. The vertical lines indicate R_500 and R_200 respectively. | 2012.13413 | Residual_profile_new.jpg |
0.545375 | e9ba05183f324741a91a73dcf4e1af88 | Cluster CL77: Comparison of the potential and dark matter density distributions. Isopotential contours are overlaid on the dark matter density distribution for the three orientations of the projected simulation cube. | 2012.13413 | cl77_contours.jpg |
0.492972 | 50af1c7a17314a6b8a81c895f151b862 | Fit of density isocontours with elliptical model. Illustration of the sensitivity of the fit to the presence of fluctuations. | 2012.13413 | contours2Dfit.jpg |
0.553302 | 9bfa1e0fbdd0473e95a7f2aa004ef5fa | Isocontours comparison. Left: potential versus gas density (original data cube), right: 2-σ criterion versus original density distribution. | 2012.13413 | contours2s_vs_orig.jpg |
0.507595 | 3d1a844cd6c945269dd47d09e299c4ea | Illustration of the gas density isocontours in 2D (3S data sets of CL135 projected along the z axis), overlaid with fitted ellipses. Isocontours are displayed with a constant binning interval of Δ (lnρ) = 0.05, where we show every 6 bins for clarity. | 2012.13413 | fig1.jpg |
0.47287 | 23a5fe455f8f402f892eb0133d2556d0 | Illustration of the three representations of 3D data on which we apply the PCA approach (Eq. <ref>). From left to right: isosurfaces; uniform distribution within isosurfaces; and cumulative isosurfaces. | 2012.13413 | illustration.jpg |
0.477 | cf7c548926724877ba29c0469d8c6903 | Interaction strength parameters (A_1, -A_2, -B_1 and -B_2) plotted as a function of parameter α defined using θ = (1/2) tanh^-1 (sinα). | 2012.13415 | F1.jpg |
0.401958 | 2144a5259a0c43fb9da38882f6d0f2ea | The largest eigenvalue, e^N θ/√(c) of the operator, 𝒟̂_P plotted in the plane of α-log(N). | 2012.13415 | F2.jpg |
0.513271 | 4dd11bd94aae4c3eb0d7693a5f6815dd | Contours of fixed values of overlap of two states |^ 0_ -BB_+^0|^2 is fitted with power law functions in the α- log(N) plane. The data is fitted with the power law function of the form f(N) = A - B N^- γ with γ=0.473. Data points are shown as dots of same colour for contours of fixed |^ 0_ -BB_+^0|^2 corresponding to 0.09 (black) , 0.54 (blue) and 0.90 (red). The parameter B has values of 2.647, 1.491 and 0.524 for overlap 0.90, 0.54 and 0.09 respectively. The parameter n̂ for all the three curves has been parameterized by defining θ_1 = π/2 and ϕ_1 = 2 sin^-1 (√(2^-1/N)) and A_ takes value 1.572. We plot log(A_-f(N)) as a function of log(N) in the inset clearly depicting the emergence of a power law dependence at large N. | 2012.13415 | F4.jpg |
0.451166 | 253f6265f55441f7af41d18c362bf18d | The parameter β plotted as a function of log(N). Dashed horizontal line is the guide to the eye to identify the domain of N for which the functions f_2 reduces to f_3. | 2012.13415 | F5.jpg |
0.460228 | 3be63dcdeee84d5ea958d58a1c9839cc | A schematic of the central spin model where the interaction with the central spin is shown in solid green lines and interaction between bath spins is shown in dashed blue lines. | 2012.13415 | FIG0.jpg |
0.476411 | f91baad4ee3643429db3918514190c31 | One-loop diagrams generating a non-zero EDM for the d quark in the presence of soft breaking of generalized parity through non-hermitian vector-like mass matrices for the SU(2)-singlets. The leading contribution arises from the diagram where h', and the heavy mirror quarks, D_j, propagate inside the loop. Analogous diagrams are present in both the up-quark and lepton sectors. | 2012.13416 | fig_1loopEDM.jpg |
0.426095 | b9d4e206ed404d0f9b48ebc28445cbed | Additional box diagrams contributing to kaon mixing in the models under consideration include any of the up-type quarks propagating inside the loop, as well as (left) two W bosons, (center) one W and one W', and (right) two W's. Quantitatively, it is diagrams with one W and one W' that are the most relevant. | 2012.13416 | fig_box.jpg |
0.404069 | edef3b36e69246ee92e685315993daed | Constraints on the size of the coefficient of the Planck-suppressed HDO of Eq.(<ref>), as a function of the scale of spontaneous symmetry breaking, v_ϕ. The region v_ϕ≲ 18 TeV (green) is in conflict with direct bounds on the mass of W' and Z' resonances, under the assumption that v_ϕ∼ v', as discussed in section <ref>. Values of ϵ that are too small (blue) do not destabilize the domain wall network early enough to either avoid a domain-wall-dominated era, or to ensure collapse before the onset of BBN, and are therefore ruled out. The region of parameter space in pink is experimentally allowed, but the level of fine-tuning in the electroweak sector worsens as v_ϕ is increased (corresponding to a darker shade). The dashed line at v_ϕ≃ 3.5 · 10^4 TeV corresponds to a fine-tuning of (10^-10) in the electroweak sector. (For illustration, we have set the quartic coupling in the pseudo-scalar potential to be κ_ϕ = 1 in this plot.) | 2012.13416 | fig_epsilonmin.jpg |
0.432243 | 330782c58bef40899dd5191f69791b1a | In blue, region of parameter space where the frequency peak of the stochastic gravitational wave signal, f_*, as well as the corresponding energy density, Ω_ gw h^2 (f_*), can fall in light of the experimental constraints on the various model parameters summarized in figure <ref>. Dotted lines correspond to constant ϵ. The region to the right of the dashed line corresponding to v_ϕ = 3.5 · 10^4 TeV features a level of fine-tuning worse than 1 part in 10^10, and it is therefore less attractive. Sensitivity curves for a variety of gravitational wave experiments are shown, including the pulsar timing arrays EPTA <cit.>, NANOGrav <cit.>, SKA <cit.> (observation time of 5, 10 and 20 years as indicated), as well as the space-based interferometers LISA <cit.>, and DECIGO <cit.>. (For illustration, we have set the quartic coupling in the pseudo-scalar potential to be κ_ϕ = 1 in this plot.) | 2012.13416 | fig_gwsignal.jpg |
0.477953 | 5d656fab95684505bf1f8b7dd5bbb317 | Schematic illustration of the particle spectrum of parity solutions to the strong CP problem in their least tuned version. The lightest exotic particles are W' and Z' resonances, followed by a top partner appearing at a scale of order v'. A mirror Higgs is also part of the low-lying spectrum, with m_h'≃√(2 λ) v'. (For illustration, we have chosen λ = (1), but note that h' could be much lighter if λ≪ 1.) Additional mirror quarks have masses of order the see-saw scale, M ≫ v'. The lepton sector must also be “see-sawed”, with mirror leptons similarly appearing well above v' (although we emphasize that the see-saw scales in the quark and lepton sectors need not coincide). | 2012.13416 | fig_spectrum.jpg |
0.439495 | 56f94a37774143a0b23f590a01221d42 | Variation of Im(β)_max as a function of Re for m/u_0 = 10 and m/u_0 = 100 for the Keplerian flow. The dashed and dotted lines represent Im(β)_max = 0.91 and 2.12 respectively. | 2012.13417 | Im_beta_max_vs_Re.jpg |
0.496608 | 74983cf486224f1eaa0db227b6c1286e | Argand diagram for Re = 10^10, m/u_0 = 10 and for α_1 = 1.0, 5.0 and 10.0 for the Keplerian flow. | 2012.13417 | WKB_kep_argand_re_10_10_v_10.jpg |
0.507225 | 3950a9bc1eeb4875a43ba1527a810f2c | Argand diagram for Re = 10^10, m/u_0 = 10^2 and for α_1 = 1.0, 5.0 and 10.0 for the Keplerian flow. | 2012.13417 | WKB_kep_argand_re_10_10_v_10_2.jpg |
0.447407 | 5b22c21aaf3b4901b39230a456dd4f92 | Argand diagram for Re = 10, 100 and 1000 for m/u_0 = 10 and α_1 = 1.0 for the Keplerian flow. For each Re, α_3 is varied from -2000 to 2000. | 2012.13417 | WKB_kep_argand_re_10_1_2_3_v_10.jpg |
0.461517 | 640ff17ea6bb4ee68deafd1881331206 | Argand diagram for Re = 10^4, m/u_0 = 10 and for α_1 = 1.0, 5.0 and 10.0 for the Keplerian flow. | 2012.13417 | WKB_kep_argand_re_10_4_v_10.jpg |
0.470059 | 51852d857f5840bb9ed57485b3add874 | Argand diagram for Re = 10^4, m/u_0 = 10^2 and for α_1 = 1.0, 5.0 and 10.0 for the Keplerian flow. | 2012.13417 | WKB_kep_argand_re_10_4_v_10_2.jpg |
0.576287 | 3a19361c8f0d40ddb68ddd3276299a0d | Variation of (ℛe(u))^2 as a function of time, for ℛe(β)_max and Im(β)_max from FIG. <ref> corresponding to α_1 = 1.0, 5.0 and 10.0. | 2012.13417 | f_t_squre.jpg |
0.472406 | 11a4f4d8962d4cb4a3a0a76ea5f73645 | Variation of Im(β) in three dimensions as a function of α_1 and α_3 for Re = 10 and m/u_0 = 10 for the Keplerian flow. Im(β)_max = 0.33. | 2012.13417 | kep_wkb_3d_plot_v_10_re_10.jpg |
0.467585 | 2fa5abf7933b4accb72c7d18f980f1ee | Variation of Im(β) in three dimensions as a function of α_1 and α_3 for Re = 10^2 and m/u_0 = 10 for the Keplerian flow. Im(β)_max = 0.822. | 2012.13417 | kep_wkb_3d_plot_v_10_re_10_2.jpg |
0.421772 | 5380908d5a254fcf9eb79f798a496a90 | Variation of Im(β) in three dimensions as a function of α_1 and α_3 for Re = 10^3 and m/u_0 = 10 for the Keplerian flow. Im(β)_max = 0.896. | 2012.13417 | kep_wkb_3d_plot_v_10_re_10_3.jpg |
0.430253 | 1bbc93401a194b7fb28ca338b64bb203 | Variation of Im(β) in three dimensions as a function of α_1 and α_3 for Re = 10^4 and m/u_0 = 10 for the Keplerian flow. Im(β)_max = 0.91. | 2012.13417 | kep_wkb_3d_plot_v_10_re_10_4.jpg |
0.422371 | 14d8a0b5b2f54d809faf468920e465ed | Variation of Im(β) as a function of α_1 and α_3 for Re = 10^4 and m/u_0 = 10^2 for the Keplerian flow. | 2012.13417 | kep_wkb_v_10_2_re_10_4.jpg |
0.4686 | fdf4cb19a74d4944a22727835d891df8 | Variation of Im(β) as a function of α_1 and α_3 for Re = 10^4 and m/u_0 = 10 for the Keplerian flow. | 2012.13417 | kep_wkb_v_10_re_10_4.jpg |
0.439704 | 32c82f4ca2044706989ad2b4d00f7a6e | Variation of Im(β) as a function of α_1 and α_3 for Re = 10^10 and m/u_0 = 0 for the Keplerian flow. | 2012.13417 | keplarian_wkb_v_0_re_10_10.jpg |
0.499433 | 07306dc0c1824c65bb0fe0ddb1394655 | Variation of Im(β) as a function of α_1 and α_3 for Re = 10^10 and m/u_0 = 10 for the Keplerian flow. | 2012.13417 | keplarian_wkb_v_10_re_10_10.jpg |
0.49708 | 1a05a9b80bf240b48e0a3beea8a2181d | Variation of Im(β) as a function of α_1 and α_3 for Re = 10^2 and m/u_0 = 10 for the Keplerian flow. | 2012.13417 | keplarian_wkb_v_10_re_10_2.jpg |
0.526005 | b555d67c9e1b42929236d8b3d5babf55 | Variation of Im(β) in three dimensions as a function of α_1 and α_3 for Re = 10^10 and m/u_0 = -10^-2 for plane Couette flow. | 2012.13417 | plane_cou_wkb_3d_plot_v_-10_-2_re_10_10_alpha_100.jpg |
0.575747 | c1b3a059f1064cce851ce21a5aed3dd3 | Variation of Im(β) in three dimensions as a function of α_1 and α_3 for Re = 10^2 and m/u_0 = -10^-2 for plane Couette flow. | 2012.13417 | plane_cou_wkb_3d_plot_v_-10_-2_re_10_2.jpg |
0.455834 | cd4e949550ca4a7386ee1afa28756e2f | Variation of Im(β) as a function of α_1 and α_3 for Re = 10^10 and m/u_0 = -10^-2 for plane Couette flow. | 2012.13417 | plane_cou_wkb_v_-10_-2_re_10_10.jpg |
0.356317 | f5ac2ab14b7d4614b63ead24b011e0f9 | Variation of Im(β) as a function of α_1 and α_3 for Re = 10^2 and m/u_0 = -10^-2 for plane Couette flow. At α_1 = α_3 = 0, Im(β)→∞. It is indicated by white point at the center of the plot. | 2012.13417 | plane_cou_wkb_v_-10_-2_re_10_2.jpg |
0.45835 | 7011e8040c064863b955d7fd85cb0d13 | Variation of Im(β) as a function of α_1 and α_3 for Re = 10^10 and m/u_0 = -10 for plane Couette flow. | 2012.13417 | plane_cou_wkb_v_-10_re_10_10.jpg |
0.486348 | 146e8a0af4c04eb79a87996c0d2f1b51 | Variation of Im(β) as a function of α_1 and α_3 for Re = 10^2 and m/u_0 = -10 for plane Couette flow. | 2012.13417 | plane_cou_wkb_v_-10_re_10_2.jpg |
0.370642 | 21800b37c0184593b5030dad6a5a585f | Variation of Im(β) as a function of α_1 and α_3 for Re = 10^10 and m/u_0 = 0 for plane Couette flow. | 2012.13417 | plane_cou_wkb_v_0_re_10_10.jpg |
0.479928 | 944e4f93c3c348a391955f2c254acb7d | Variation of Im(β)_max as a function of m/u_0, for α_1 = 1, 5, 10 and Re = 10^4 for the Keplerian flow. | 2012.13417 | v_vs_growth_rates.jpg |
0.429676 | 5c5bac383e454d33863b0ae254741b45 | (a) Schematic representation of holographic characterization of colloidal spheres and clusters in a microfluidic channel. Particles, including clusters, passing through the illuminating laser beam scatter light that interferes with the rest of the laser beam to create a hologram. Fitting a hologram to a generative model based on the Lorenz-Mie theory of light scattering yields measurements of the particle's effective diameter, d_p^∗, and refractive index, n_p^∗. (b) Scatter plot of single-particle properties measured for a nominally monodisperse sample of micrometer-diameter silica spheres that have been functionalized for a medical diagnostic test. Each plotted point represents the measured properties of one particle and is colored by the relative probability density of observations, ρ(d_p^∗, n_p^∗). Clusters of points are likely to represent distinct types of particles, with at least three clusters being evident here. | 2012.13419 | dimer_schematic04.jpg |
0.488664 | 07303555c92542d499ec3d46b4bee972 | Dependence of the effective-sphere diameter, d_p^∗(θ), and refractive index, n_p^∗(θ), on dimer orientation, θ. Discrete points are the data for simulated dimers plotted in Fig. <ref>(a). Solid (red) curves are fits to Eq. (<ref>). | 2012.13419 | dimerfits04.jpg |
0.448487 | 52898c890a924cddbe95cac6914a171b | Geometry of symmetric dimers used for computing cluster holograms. A pair of spheres initially is aligned with the x̂ axis and then is rotated through two Euler angles, ϕ and θ. | 2012.13419 | euler05.jpg |
0.407642 | e0eaf25a70d244688d4a1d12dced2a19 | (a) The holographically measured flow profile, v_p(z_p), for particles identified as dimers agrees with the flow profile for individual spheres co-dispersed in the same system. This agreement validates the effective-sphere estimate of z_p for the dimers. The dashed curve is a fit to the parabolic profile expected for Poiseuille flow, which also yields estimates for the axial positions of the upper and lower walls, indicated by dashed lines. The projected distribution of axial positions, ρ(z_p), shows that spheres and dimers are both concentrated near the midplane of the channel. (b) Angular orientation distribution, P(θ), for dimers (orange) estimated by projecting measured effective-sphere characterization data onto the computed curves for d_p^∗(θ) and n_p^∗(θ). Discrete (green) points reproduce experimental results on Brownian rods in a Poiseuille flow from <cit.>. The dashed curve is the prediction of Eq. (<ref>) for Γ_max = 50. | 2012.13419 | jeffery04.jpg |
0.422732 | 4409b00ebb9347969c23b0ff25583f45 | Effective-sphere interpretation of a (a) simulated and (b) experimental holograms of colloidal monomers and dimers. Simulations are performed for symmetric dimers composed of silica spheres (n_0 = 1.435) with diameter d_0 = 1.054 at varying polar angle θ. Individual points are colored by θ. The values for d_0 and n_0 are selected to mimic the experimental system. The dimers' effective diameters, d_p^*(θ), and refractive indexes, n_p^*(θ), are fitted to the phenomenological model from Eq. (<ref>) that is plotted as the dashed black curve in (b). Holographic characterization data are segmented into monomers and dimers. Data points identified with dimers are colored by the angle θ of closest approach to the predicted dimer curve. | 2012.13419 | orientation04.jpg |
0.519989 | 6a32957dd3544b85839d8864f6dfaf8d | Error in the effective-sphere axial position, z_p^∗, for the simulated symmetric dimers from Fig. <ref>(a) as a function of true axial position, z_p. Each point is colored by the dimer's orientation, θ. Effective-sphere analysis tends to underestimate z_p, with the errors tending to be greater for dimers aligned at θ = 45 from the imaging plane. | 2012.13419 | z_error04.jpg |
0.496375 | 9a0c71c9bb0b4524bf38afd10775b6a3 | A typical two-loop diagram involving gauge bosons that induce right-handed Majorana neutrino mass in the model. | 2012.13420 | 2loop_gauge.jpg |
0.520688 | 8636b6116b4b4468832a2ad147e61433 | The observed 95% C.L. limit of the production cross section, σ (p p →η^+ η^-), as a function of mass of η^+ scalar obtained from four leptons searches at √(s) = 13 TeV pp collisions by the ATLAS experiment <cit.>. Dotted line corresponds to the future sensitivity limit at the high luminosity LHC with 3 ab^-1 data. | 2012.13420 | 4lep.jpg |
0.51405 | 42f8820b96954a6da0ed78b9e3ff92c5 | Integral function in Eq. (<ref>) as a function of mass ratios, r_1 and r_2. | 2012.13420 | Integral_plot.jpg |
0.382893 | 703581d5704f41839d61e9bd15b5e506 | Maximum values of the one-loop induced and two-loop induced M_ν_R as a function of M_W_R. The dashed blue line corresponds to the two-loop contribution, while the dashed green and the solid red lines respectively represent one-loop contributions to the ν_R mass with and without Parity symmetry. | 2012.13420 | mR_WR.jpg |
0.545558 | 7abc7517df8b4ba19e6911f73dcc6904 | Feynman diagram for the production of 4-lepton + E_T through Drell-Yan process with pair production of η^±. ν_R decays to lepton pair and neutrino via virtual η. | 2012.13420 | pp_4l.jpg |
0.468114 | 4e858b478feb415ea4b730270d868b8b | Total cross section for ν_R production via e + p →ν_R + n at supernova energies as a function of the CM energy of the electron. | 2012.13420 | sigma_Ee.jpg |
0.469223 | a0d95c757c534d9796e619b9bc54d95e | (Color online) Values of |a_J^F|^2 averaged over initial multipole moments Q_20,Q_30 from the even J^F-momenta are displayed with filled symbols, while the contributions arising from odd J^F-momenta are displayed with empty symbols. The “error bars” characterize the range of the variation due to the spread of initial multipole moments Q_20 and Q_30 and energies of the fissioning nucleus. The |a_J^F|^2 for the SeaLL1 <cit.> and SkM^* <cit.> (displaced by Δ J^F = 0.36 for better visualization) NEDFs are displayed with filled and empty symbols for the even and odd values of J respectively. The average (standard deviation) for ^240Pu are [A^L, Z^L]= [103.6(0.7), 41.0(0.3)] and ^236U [102.4(2.0), 40.4(0.7)] in case of SeaLL1 and [104.3(1.5), 41.4(0.5)] and [97.9(1.2), 38.9(0.4)] in case of SkM^* respectively. The evaluated FF intrinsic spins S^H,L at different FFs separations are shown in the inset for ^236U. Typical behavior of the overlaps ⟨Φ | R̂_x^F (β) | Φ⟩, for one TDDFT trajectory is shown in the inset for ^240Pu. The overlaps' widths narrow with increasing β and the average S^F increases. | 2012.13422 | U1andP1.jpg |
0.471123 | 367da7d3cc944e0bb4a6a01600ddb0fa | (Color online) The average intrinsic spins S^L,H versus the initial FF equivalent incident neutron energy E_n'= E^*-S_n (E^* and S_n are the excitation energy and S_n the neutron separation energy) for the reaction ^239Pu(n,f) with SkM^* NEDF. The solid lines are linear fits over the data, S^L=0.0168 E_n'+9.197 and S^H=0.0732 E_n'+4.933 respectively, as a function of equivalent neutron energy E_n' along with their linear fits. In the inset we display the FF excitation energies and their linear fits E^L_int=0.4505 E_n'+ 13.25 and E^H_int=0.5676 E_n'+13.40. Using E^F_int≈ A^F(T^F)^2/10 <cit.> it follows that on average T^L>T^H. | 2012.13422 | pu240_spin_exen.jpg |
0.471457 | dab4e90650fb4ca9b88946b903ca89a3 | Proposed framework for a performance-based anomaly detection system | 2012.13423 | bigpictureICAC.jpg |
0.440256 | 08e22ad89ac0421397b65edb776500bd | The considered anomaly detection mechanism aims to detect anomalies solely based on job response times. When is it feasible to identify if the system is under attack solely based on response times? Jobs of types a, b and c correspond to short jobs, short jobs impaired by attacks and long attack jobs, respectively. Depending on the system setting, the waiting time of jobs in the waiting queue can be used to distinguish jobs of type a and type b. Under which system configurations does the waiting time contain enough information to allow the automatic distinction between jobs of types a and b, implying the feasibility of locally detecting anomalies solely based on response times? | 2012.13423 | mechanism_detection.jpg |
0.464183 | 4ab2389691e9411aa442ecb5091c8f14 | Anomaly detection framework, accounting for infrastructure configuration actions at system deployment time. | 2012.13423 | securityscheme1.jpg |
0.594965 | 206016144c0b44b7880ebcf9b3349327 | M/G/K model: the proposed analytical model is used to assess the impact of the number of servers K on standard deviation and mean response times. | 2012.13423 | systemmgk1.jpg |
0.422192 | 1d846fb5dff848f6871c135c514c74fe | States and regions in the US show closely conforming growth in Covid-19 cases after rescaling and translating, and these case data can be fit by SIR models. Top left: a heatmap showing the clustering of cumulative case curves for all US states along with the District of Columbia and Puerto Rico. Clustering was performed with the WPGMA algorithm <cit.> using pairwise mean squared distances between case curves after vertical scaling and horizontal shifting. Distances are shown from low (blue) to high (orange) in the heatmap. Top right: a legend associating each US state and territory code to the heat map rows and columns. Bottom left, bottom right: examples of SIR curves fitting two of the state clusters. In each image, the solid black line is the SIR model fit, and colored lines are actual Covid case data from <cit.> after scaling and translation. | 2012.13424 | ClusterFigure.jpg |
0.429495 | 90d349d312db45ef8afafa67e1891a54 | Networked subpopulations still allows for fitting by an SIR model. Erdős-Réyni networks were simulated for with three different mean degree parameterizations. 100 replica simulations were run with β = 0.2, γ = 0.1, N_i = 1000, and K = 101. Shown: Close to identical estimates are obtained for β̃ using Eqn. <ref> (orange, solid) or least squares fitting (orange, dashed). Mean Relative Squared error is shown for Eqn. <ref> (blue, right axes). | 2012.13424 | Graph_Dynamics.jpg |
0.466092 | 66298bd4d27245eb907394c5202d2501 | A single well-mixed population model can represent aggregate spread of disease through a linked set of subpopulations, even with relatively weak inter-subpopulation mixing. Top: Cumulative cases from the model in Eq. <ref> with β_i j given by Eq. <ref>. Blue curves: subpopulation model with b increasing from b = 10^-5 (lightest) to b = 1 (darkest) with log-spaced values. Red curves: optimal SIR fits, with b=10^-5 (lightest), b ≈ 10^-2, and b ≈ 10^-1 (darkest) cases. Bottom panels β̃ given by Eq. <ref> (left, orange) and by linear regression (right, orange) as a function of b (orange), and the relative mean squared error of the true model relative to a simple SIR model using β̃ and N_tot as a function of b (blue) [The least-squares fitting algorithm that generated these results treats γ and N_tot as known.] (blue). The plots included in this figure are from numerical simulations done with the parameters β = 0.2, γ = 0.1, N_i = 1000, and K = 5, and initial data I_1(0) = 1, I_i(0) = 0 ∀ i > 1 and R_i(0) = 0 ∀ i. Both the β̃ and the regression calculations are done over an entire pandemic period (250 days). | 2012.13424 | Model_Transition.jpg |
0.43994 | a7bef76733734465acf993b96a540259 | R_0 estimates from fitting entire case data curve optimally to well-mixed model (blue) agree with linearized analysis by next-generation matrix method (yellow) <cit.>, but not to empirical fits to the data assuming exponential growth (red). | 2012.13424 | R0_Compare.jpg |
0.432533 | 608134e11d6d44ba999b704a595d9652 | Different connectivity graphs: 𝒢_1 (left) and 𝒢_2 (right) | 2012.13425 | adjacencies.jpg |
0.427284 | c2d3ab69607f4bd28d1fff7c17699311 | Field layout and treatment allocation (numbers) for the motivating agricultural experiment at Rothamsted Research (year 2016). Blue/green shading indicates the superrows and light/dark shading indicates supercolumns considered in the development of alternative models and designs | 2012.13425 | figure1.jpg |
0.480853 | d3a9dfa2d52f4c96a03e148ca5ed1106 | Hasse diagram of the unit structure of the design. Each node has two numbers: the number of levels of the corresponding blocking factor and the corresponding degrees of freedom (in brackets), obtained by subtracting the degrees of freedom for higher factors from the number of levels of the factor under consideration. R, C, r and c represent supperrows, suppercolumns, rows and columns respectively. | 2012.13425 | hassercd.jpg |
0.374124 | ebdb8309ade84869abf35c29c285a9f7 | Lightcurves showing radio observations (scaled to 4.8 GHz) and X-ray observations from MAXI (2-10 keV). The stars on the plot show the observations discussed in this work. | 2012.13426 | radioxrayLC_bw.jpg |
0.378686 | b316238799fb4d2ab2e7ca39efee8f0e | Mixed-Linear model has comparable accuracy to standard DNN. Plot of the test errors for different datasets using different models. We take a ResNet-50 pretrained on ImageNet and fine-tune it using different procedures, (DNN) We fine-tune the whole network on various datasets, (Mixed-Linear) We fine-tune the linearized ResNet-50 (<ref>) in a mixed private setting, Last-Layer Features: We simply fine-tune the final fully connected (FC) layer of the ResNet-50. We show that fine-tuning a linearized DNN using the mixed-privacy framework performs comparable to fine-tuning a DNN and outperforms simply fine-tuning the last FC layer. | 2012.13431 | Error_Comparison.jpg |
0.421251 | c616533c033d4419a9ec75db78de2c77 | Plot of the amount of remaining information and test error vs the L_2 regularization coefficient. We forget 10% of the training data sequentiall through 10 forgetting request. | 2012.13431 | information_error_noise_caltech_stanforddogs_wd.jpg |
0.415152 | 27127c3fb6cb4cdab31f402a0114960f | Same experiment as <Ref> for MIT67. | 2012.13431 | information_error_noise_mit67.jpg |
0.444523 | 55b1ad9c009147098c2d030541d2774b | Comparison of complete and stochastic residual gradient estimation for forgetting. uses the complete residual gradient for the forgetting step, exploiting the fact that the loss function is quadratic. However, one can also estimate it stochastically, which is equivalent to fine-tuning on the remaining data to forget. Here we show that indeed both methods work but – when using the same number of steps – complete estimate gives a better solution due to smaller variance and faster convergence (lower test error and information leakage). | 2012.13431 | information_v_error_ft_compare_caltech_stanforddogs.jpg |
0.469262 | 74e6ac16b7f94a48a2f8a08aa9182cc0 | Effect of using a core data close to the task. Plot of the remaining information and test error on Aircrafts using (a) generic ImageNet core data, and (b) ImageNet pre-training + 30% of the Aircrafts. When the core data contain information close to the user task that the network does not need to forget, can exploit this to create better core-weights and a correspondingly better linearized model. This improves both the accuracy of the model and makes forgetting easier, as seen from the accuracy-forgetting curves in the plot. | 2012.13431 | information_v_error_wwo_mp_aircrafts.jpg |
0.459236 | 88341ed78dfd451f9660efd96d8c64af | 100 realizations of R(t), for t∈[0,15], with second order approximation. | 2012.13432 | itofig.jpg |
0.448036 | 27328179a545432da86b6abfe64c9561 | 100 realizations of R(t), for t=15 (second order approximation). | 2012.13432 | itoosc.jpg |
0.471874 | a29b2dc4f41c448795676eacd10ab6d8 | 100 realizations of R(t), for t=15 (first order approximation). | 2012.13432 | oneorderoccil.jpg |
0.408374 | 7ffcc2398bf7431aaa98bae115c5d843 | 100 realizations of R(t), for t∈[0,15], with first order approximation. | 2012.13432 | oneorderstoc.jpg |
0.360683 | b1956075156547e3902c198a9f30fd35 | Phases of the Parsing Pipeline | 2012.13436 | parsing-pipeline.jpg |
0.444858 | e5f558577af34c7e8037cb7ea03bef47 | A trajectory of (<ref>) emanating from x(0)=[ -1 1 1 ]^T for the dissipation parameter in (<ref>). | 2012.13441 | RAZ_CHAOS.jpg |
0.492945 | 8a8a451c3bd74f86900e698677a1686b | Several trajectories of the closed-loop system for the dissipation parameter in (<ref>). The circles denote the initial conditions of the trajectories. | 2012.13441 | RAZ_CLOSED_LOOP.jpg |
0.421325 | 3c46668528584a51af9156b1fab4f918 | (Color Online). Detailed network architecture of the ADL-MVDR module. The real and imaginary parts of the covariance matrices are concatenated before fed into the GRU-Nets. The estimated MVDR coefficients are reshaped back to their original forms before computing the frame-level MVDR weights. | 2012.13442 | ADL_MVDR_module.jpg |
0.477554 | 76643476bff54ed7b8d4928587b86f46 | (Color Online). Illustration of one TCN block (left) that consists of 8 dilated 1-D convolution blocks (right). Each dilated 1-D convolution block consists of 1×1 (i.e., pointwise) convolution layers and a depth-wise separable convolution layer (D convolution) <cit.>. PReLU <cit.> activation and normalization are applied between the convolution layers. Skip connection is also used. | 2012.13442 | AudioEncodingBlock.jpg |
0.424554 | f181320e5e5d442591da6007815b1868 | (Color Online). Network architecture of our proposed ADL-MVDR framework. It consists of a complex filter estimator (i.e., highlighted in blue dashed box) based on temporal convolution network (TCN) blocks for components estimation (depending on the case), and an ADL-MVDR filter (i.e., highlighted in red dashed box) that consists of two GRU-Nets for frame-wise MVDR coefficients estimation. ⊗ and ⊙ denote the operations expressed in Eq. (<ref>) and (<ref>), (<ref>) or (<ref>), respectively, depending on the situation. | 2012.13442 | TASLP_ADL_MVDR.jpg |
0.403896 | d5bbe52595974c4a9e041aeb42d49906 | (Color Online). Schematic diagrams for complex ratio masking and complex ratio filtering. The cRM is a one-to-one mapping, whereas the cRF is a many-to-one mapping. In this example, the cRF has a size of 3×3, where each T-F bin is estimated using nine T-F bins in its neighboring filter range. The center mask is marked with a red fill. | 2012.13442 | cRF_plot.jpg |
0.422432 | 002ef5e6cd144fffac27a38f7ab57234 | (Color Online). Schematic diagrams for the 15-channel linear microphone array and the DOA estimation process with a wide view camera. | 2012.13442 | mic_array.jpg |
0.402764 | 4e17be2d69c5412d97e46cfa28953ee2 | (Color Online). Beam pattern examples for conventional neural mask-based MVDR system and our proposed MC ADL-MVDR system. RT60=0.31 s. Target source at 63^∘, interfering source at 131^∘. | 2012.13442 | sample_beampattern.jpg |
0.440501 | e65530ed47944053ae0e017e7e493957 | (Color Online). Spectrograms of some evaluated systems in Table <ref>, the nonlinear distortion and residual noise are highlighted by the dashed boxes. | 2012.13442 | sample_spec.jpg |
0.40834 | 062a6839fadd42d888f114fc23a82a5e | Illustration of the 𝒫𝒯-symmetric crystal lattice for one, two and three pairs of particles. | 2012.13444 | fig1.jpg |
0.441289 | e6b8d3d8281b496fad9fe69bdc320664 | Spectral density S^s(rŝ,ω) normalized by factor 4S^i(ω)B^2σ^3/r^2, for a pair of particles with gain and loss described by (<ref>), versus polar and azimuthal scattering angles, ϕ and θ, respectively. (a) p = 0, (b) p = 1, (c) p = 1/2 and (d) α = 0 (passive case). Parameters used: ŝ_0 = (1/√(3))(1,1,1), r_0 = (6,0,0), α = 1 (a-c), kσ = 0.1. | 2012.13444 | fig2.jpg |
0.482759 | ce5607f857824733b006729746a0d7a0 | (a) Results of the Monte Carlo simulations. Experimental results for EuTe (squares), EuSe (triangles) and EuS (circles) are also shown, from references [prb17EuTe], [prl2018], and [apl2020], respectively. (b) Results of the Weiss field model. | 2012.13445 | Fig_MC_WF.jpg |
0.418943 | 38dc0ab9593b49aab2fd74a6a87bf85b | Visualization of VGG BA SMALL architecture, showing only parts that contain trainable parameters. Omitted layers are described: (1) each convolution layer is followed by 2D Batch Normalization and then ReLU non-linearity. (2) Immediately before the dense layer, there is also a 2D average pooling layer. Our code is publicly available at <cit.>. | 2012.13447 | VGG_BA_SMALL.jpg |
0.523467 | 8e8efe2ca94f4fb4adb4e5ddb63e2eed | Selected samples from CK+ and FER2013 dataset | 2012.13447 | ckfer.jpg |
0.483244 | 244193d149b44bf6b98403f1945be902 | Confusion matrix of VGG BA SMALL final model on test set. | 2012.13447 | confusion.jpg |
0.443154 | f4528547e7ef4cc3966b2f523a6e6324 | Emoji marked with key points. Coordinates calculated based on human face ratios as used in drawing. See discussion in Section <ref> for reference. This emoji is chosen to correspond to the “neutral” label in this work. | 2012.13447 | emoji_with_marks.jpg |
0.451927 | 72bb1306b1c44bf2a24417fb1a0fc163 | Human face example marked with key points. Coordinates estimated by Dlib HoG + SVM 68 facial landmark model. Marking based on portrait of Kim Jisoo <cit.>. | 2012.13447 | face_with_marks.jpg |
0.394842 | e39c7973a9cf4618acba18a29b114bfc | Examples of good performance and limitations | 2012.13447 | good_bad_eg.jpg |
0.40672 | 00a0158d6f814225b5e1b4abe7bf7807 | SqueezeNet performance on one of our own faces | 2012.13447 | ourfacesqueeze.jpg |
0.562639 | eb5bc5be1f6a4ceb83dc336490855aa8 | Processing pipeline | 2012.13447 | pipeline.jpg |
0.400614 | 8eb1da12d2774fc9ab2e5c64183b8079 | Architecture overview of the system. | 2012.13448 | Flow_chart_woSpeed.jpg |
0.443551 | afa3f4108b8848ae8f1ecee70bc8b434 | The road simulated in the ray-tracing simulator is converted to SUMO network from OpenStreetMap (OSM). | 2012.13448 | OSM.jpg |
0.453407 | 128bb4cf83cf47208268e96a959c08c5 | 2012.13448 | box_plots.jpg |
|
0.4877 | cab436df9fc5465da3a54ba9e7da9207 | The actual and estimated number of vehicles for 100 packets. | 2012.13448 | estimation.jpg |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.