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
|
|---|---|---|---|---|
0.384332 | 2ac6e64c0a1243d3919a4ebd5367282b | Correlationg accuracy on ANLI with importance alignment (𝒜^H and 𝒜^S) for all wrong examples. The cross, circle, and triangle refer to rounds 1, 2, and 3 of ANLI, respectively. | 2012.13354 | accuracy.jpg |
0.460144 | 331cc61dd34b4d74b24bf9567100ae89 | Accuracy on ANLI is not correlated with importance alignment (𝒜^H and 𝒜^S). The cross, circle, and triangle refer to rounds 1, 2, and 3 of ANLI, respectively. | 2012.13354 | accuracy_subset.jpg |
0.45267 | f4750f192b5d449e8adfd9d18fb739b1 | Correlating accuracy on HANS with importance alignment for all wrong examples. Alignment values are averaged across rounds because HANS is not divided into rounds. | 2012.13354 | hans_accuracy.jpg |
0.381074 | 48d44e6351134101be1361ee8b546b5f | Correlating accuracy on ANLI with importance alignment (𝒜^H and 𝒜^S) for all wrong examples. The cross, circle, and triangle refer to rounds 1, 2, and 3 of ANLI, respectively. | 2012.13354 | hans_accuracy_heuristics.jpg |
0.417282 | 6a4edb67510a49429307fe9cddb8eaf1 | Accuracy on HANS is not correlated with importance alignment. Alignment values are averaged across rounds because HANS is not divided into rounds. | 2012.13354 | hans_accuracy_subset.jpg |
0.472551 | 3ba196b091e34f4e8c8ec0bd6f9293ba | Correlating accuracy on HANS with importance alignment for all wrong examples. Alignment values are averaged across rounds because HANS is not divided into rounds. | 2012.13354 | hans_accuracy_subset_heuristics.jpg |
0.453122 | f0f7a10ab64a4c0b96583a8740c9c59d | An example illustrating different token-level importance values. “Model Importance” is color coded by absolute value integrated gradients attribution for BERT-base. The other three rows show the oracle importance scores estimated by the hard, soft and expert oracles (darker values indicate more important). | 2012.13354 | sampleexample.jpg |
0.519564 | 38a64a635a29416c863c71f8950ed32a | The Gauss map Φ:m → S^(N-2), maps points p defined on the worldsheet to the unit sphere S^(N-2). | 2012.13358 | map_Gauss.jpg |
0.45102 | 78f0fde9c32d407ea3e410408969d3e9 | Plugin generalization architectural views. | 2012.13363 | alg-generalization.jpg |
0.488293 | d2eadc93251d4b4c934ce8d2c5adf5f2 | Algorithms generalization architectural view. | 2012.13363 | algorithms.jpg |
0.549179 | de198713cebf4ada8f6f9ab223a42923 | Algorithms generalization architectural view. | 2012.13363 | algorithms.jpg |
0.475142 | be8b8f5dc7ac4af7b771bce9dc091650 | Authentication process using an identity provider. | 2012.13363 | authenticationworkflow.jpg |
0.490361 | 365bd46effc94a92997facc5248a841f | Client-Server architectural view. | 2012.13363 | clientserver.jpg |
0.407061 | f96fb534276e427e8899f1c413a77e94 | Graph data model architectural view. | 2012.13363 | datamodel.jpg |
0.534474 | e4e853ca06fc4c4f991fa30aaab8493f | Decomposition architectural view. | 2012.13363 | decomposition.jpg |
0.44109 | c64dc21f9a354818bc2a37b69ce5cd17 | Decomposition architectural view. | 2012.13363 | decomposition.jpg |
0.487662 | 5171028f8e5d46a3b6884c725ff83fe6 | Controller generalization architectural view. | 2012.13363 | generalization-controllers.jpg |
0.469733 | 14e42f1fea0a4a92b667c3f91919a6f3 | Formatter generalization architectural view. | 2012.13363 | generalization-formatters.jpg |
0.445989 | 6b7b8c67375c4c7b8e5102e0c78ffc5f | Repository generalization architectural view. | 2012.13363 | generalization-repositories.jpg |
0.548139 | 7df6bf4f349a40d594272026b3d5fd1e | Security generalization architectural view. | 2012.13363 | generalization-security.jpg |
0.438316 | feb950e41de6497d8e1f3e6b192fe8ed | Service generalization architectural view. | 2012.13363 | generalization-services.jpg |
0.464665 | a65b481bf250401eba9d9f9e8ca9869f | Layered architectural view. | 2012.13363 | layered.jpg |
0.473849 | dadb6e6477094a0d9492ab2241b4a057 | Layered architectural view. | 2012.13363 | layered.jpg |
0.494724 | 5c6e553f62224c9c90e3da6ccaf63bde | Example representation of phylogenetic data. | 2012.13363 | phylogenetic.jpg |
0.443456 | 3598b21c616e49cbad76067b2f0fb493 | Example representation of typing data. | 2012.13363 | typing.jpg |
0.453343 | 1aacf2ee92f0471e9113cabde0c989c9 | Use cases. | 2012.13363 | usecases.jpg |
0.412934 | d7332f512137498d879eee39b3989ed2 | Uses architectural view. | 2012.13363 | uses.jpg |
0.402014 | b964d74a81a448549004fd5b85de49d0 | Uses architectural view. | 2012.13363 | uses.jpg |
0.486898 | eac7cdb76caa4d44ad305e63c788e6b4 | Example of a workflow system. | 2012.13363 | workflow.jpg |
0.498438 | 2789efbeaa044146a9803d82aa57c567 | Network architecture overview. Given Cine-MR volumes with the size t × h × w from the training dataset as input, we pass it through the segmentation network to obtain output segmentation masks for the LV cavity and Myo. A segmentation loss is computed based on the ground truth. To make predictions for 11 indices and cardiac phase detection, we utilize a multi-task spatio-temporal network with two parallel branches. Then a multi-task loss is calculated on the target prediction for both regression and classification tasks and is back-propagated to the segmentation network. | 2012.13364 | LVQaun_model.jpg |
0.414912 | 6704a4bedfcd4e9c865c7b38892ff321 | Representation of gradient flow between segmentation, regression and classification tasks in our cardiac quantification framework. | 2012.13364 | LV_gradient.jpg |
0.448689 | ff3af223a1f44ca78d1107356c002d47 | A schematic representation of LV indices for short-axis cardiac Cine-MR image. The LV and Myo cavity areas are shown with green and blue colors in (a) and three LV cavity dimensions with black arrows in (b). (c) shows Six myocardial regional-wall thicknesses, namely anterolateral (AL), inferolateral (IL), inferior (I), inferoseptal (IS), anterior (A), inferoseptal (IS) and anteroseptal (AS). The cardiac phase (systole or diastole) is shown in (d). | 2012.13364 | LV_indcies.jpg |
0.417182 | 93a7376730f140bf824bcfc4f925c716 | LV MR image slices before(first row) and after pre-processing and normalisation (second row). | 2012.13364 | LV_preprocessing.jpg |
0.421533 | ebaa96fbea8f49d3a2e6eecd9e6ecd46 | Qualitative segmentation results of DR-UNet 2D. The red contour shows the ground truth segmentation contour, and the blue color is overlaid as the proposed model prediction output. The row 1-2 shows the first 10 slices with their segmentation error. The row 2-4 illustrate frames 11 to 20 with segmentation error respectively. It can be seen, that segmentation error is quite low for both endocardium and epicardium contours in most of the frames. | 2012.13364 | LV_segOut.jpg |
0.482337 | b56b14a7bf7c4267868b3b0e029e9e79 | The schematic illustration of our left ventricle segmentation network equipped with dilated convolutions(4 dilation rate 1-8) in the bottleneck to capture multi-scale features and residual blocks in the encoder path. | 2012.13364 | UNet_model.jpg |
0.385965 | 37f5236c69d8449ea48c0a43566b1e35 | Multi-objective selection of linear models. For each variable to be explained, we represent the position of all linear models in the objective space (corrected Akaike criterion AICc and adjusted R^2). The color of points gives the number of observations. | 2012.13367 | Fig3.jpg |
0.373761 | 3c45e0126e9e4af1ae54b2153d5cc1cb | The effect of different widths of solvent layers on the SWAXS profile. The chop parameter has minimal effect on the SWAXS profile due to our use of uniform implicit solvent model. Note that in the FM algorithm, we used the largest width of solvent layer corresponding to chop parameter of 0.001. | 2012.13370 | chopplt.jpg |
0.418784 | 48b0397d1cd84e48853cae0f1094decf | (a) The fits from different FM steps to the experimental SWAXS profile of 25bp dsDNA. The profiles are offset by different scale factors for easy visualization. The scattering intensity is shown using a logarithmic scale that magnifies subtle disagreements. (b) The refinement traces of χ^2 (solid) and |∇_d(χ^2)| (transparent) for all the FM steps. Good convergence can be observed by decaying χ^2 and L2-norm of the gradient within a few hundred iterations. (c) The total electrons for all FM steps. (d) The density maps of the refined electron densities from different FM steps. The density maps are presented with the same coloring scheme as the curves. The maps were generated using PyMOL (Schrodinger, New York, NY). Each subsequent density map varies in resolution and recapitulates the experimental data to a specified q_max. We construct a wire plot associated with each density map by extracting coordinates of the voxels that contain more than 80% of the maximum density. These maps are shown with the same color code used in the top-left panel. The first density map is from DENSS using small-angle data. See main text for detail. | 2012.13370 | demo3.jpg |
0.425776 | f795743ce1534e348c51bf34169ab691 | The effect of electron density smearing on the SWAXS profile. The smearing parameter, σ, has minimal effects on the SWAXS profile when σ < 1.0. Note that in the FM algorithm, 0.3 < σ < 1.0 always. | 2012.13370 | gaussianplt.jpg |
0.396963 | fbda34f4792b40a2b914302efd654060 | (a) We simulate a normally distributed random variable X and convert it to the linear and customized χ^2 metric shown in pink and blue respectively, using μ=3.39× 10^5 and σ=1.67× 10^5 from one experimental point at q=1.25^-1. The solid line is the distribution of Eq. (<ref>) with r = μ/σ = 2.03. The inset focuses on y > 1 and shows the differences between two distributions. (b) The distribution of Eq. (<ref>) with different signal-to-noise values ranging from 0.75 to 25.0. | 2012.13370 | logchi2.jpg |
0.443741 | 07a0ec8776df43cb93cc096cec38e18d | The application of FM algorithm to different systems: (a) RNA hairpin (hpRNA), (b) RNA triplex (tsRNA), (c) metastasis-associated lung adenocarcinoma transcript 1 (MALAT1) and (d) a short strand of single-stranded RNA (ssRNA). Each panel displays the experimental data and FM fit, with residuals computed from Eq. (<ref>), the FM refined electron density and corresponding wire plot as in Fig. <ref>. The two cross-sections associated with either density map or wire plot represent sections through the midsection (left) or across the top (right) viewing into the central axis of the molecule. See main text for detailed explanation. | 2012.13370 | results.jpg |
0.468783 | f557d53eea3b479688b1c23cb672b313 | The schematics of Algorithm <ref>. At each FM step, the electron density is smeared before being split. After splitting, the algorithm carves the electron density to reveal finer structures on length scales defined by q_max^(i) and proceeds to the next FM step. | 2012.13370 | schematics_green.jpg |
0.47988 | d92eb44ead2941a7b178531e5d150164 | The effect of different voxel splitting sizes on the SWAXS profile. The splitting parameter, z, has minimal effect on the SWAXS profile when z > 0.75. Note that in the FM algorithm, it marches to higher q and z > 1.0 always. | 2012.13370 | zoomplt.jpg |
0.396507 | 2a04c2275a774413a5476f0e3d17924f | Relative gain of NL over Linear processing in downlink for 4x4 and 8x4 MU-MIMO configuration for indoor measurement positions. | 2012.13371 | mobi_gains_dl_indoor_pat_3.jpg |
0.391374 | 9ad8ad648e084645a1adff05e411133c | Relative gain of NL over Linear processing in uplink for 4x4 and 8x4 MU-MIMO configuration for indoor measurement positions. | 2012.13371 | mobi_gains_ul_indoor_pat_3.jpg |
0.450492 | 4291662be95f4da4a3b0b4c81746d1f9 | Relative gain of NL over Linear processing in uplink for 4x4 and 8x4 MU-MIMO configuration for outdoor measurement positions. | 2012.13371 | mobi_gains_ul_outdoor_pat_3.jpg |
0.508957 | 0c4a68d3db064f18884b466844c24b0f | The impact of ue specific dmrs power boosting on system performance. | 2012.13371 | pilot_boost_pat_2.jpg |
0.451132 | 3818aced808341a49582f794489f3254 | Activation of output of the transform function at different stages of GCNet. Best viewed in color. | 2012.13375 | activation_32.jpg |
0.484007 | 5d1a16299b8e45ab8dc380e3306fdc26 | Distributions of class selectivity index of ResNet-50 baseline and ResNet-50+GCNet on different layers. For deeper layers, GCNet exhibits less class selectivity compared to the baseline. Best viewed in color. | 2012.13375 | merged.jpg |
0.461085 | 7d639d2750524968b742b3dedfb8d173 | Visualizations of context modeling attention maps (heatmaps) of GCNet and NLNet (red points denote query positions). Their learnt attention maps are mostly similar. Also, they learn to focus more on hard cases like relatively small size, deformation, occlusion, and blur. Best viewed in color. | 2012.13375 | nl_attn_map.jpg |
0.498388 | 4603449f33284fe58cbe4debbe38a769 | Architecture of the main blocks. The feature maps are shown as feature dimensions, e.g. CxHxW denotes a feature map with channel number C, height H and width W. ⊗ denotes matrix multiplication, ⊕ denotes broadcast element-wise addition, and ⊙ denotes broadcast element-wise multiplication. | 2012.13375 | structure-all.jpg |
0.383446 | 0bb70f89bb5b4e59bded432938cfeeca | Two instantiations of the non-local block: Embedded Gaussian and Gaussian. The feature maps are shown by their dimensions, e.g. CxHxW. ⊗ denotes matrix multiplication, and ⊕ is broadcast element-wise addition. For two matrices with different dimensions, broadcast operations first broadcast features in each dimension to match the dimensions of the two matrices. The feature maps marked in red (e.g. `5 att') are statistically analyzed in Tables <ref>, <ref> and <ref>. | 2012.13375 | structure-nl.jpg |
0.415691 | 581e8df0cb3b4180b3e2b27313e466b4 | Visualization of attention maps (heatmaps) for different query positions (red points) in a non-local block on COCO object detection. The three attention maps are all almost the same. More examples are presented in Figure <ref>. | 2012.13375 | teaser.jpg |
0.438754 | ebf25d6607034686927d14c25ed952c6 | Visualization of attention maps (heatmaps) for different query positions (red points) in a non-local block on COCO object detection. For the same image, the attention maps of different query points are almost the same. Best viewed in color. | 2012.13375 | vis.jpg |
0.511122 | 69d2ee416d504ced9564c4bbae74ffc2 | Performance of 6 physical microscopic car-following models. | 2012.13376 | 6_Physics_models.jpg |
0.406055 | 54015577005741e793a83a2766e96115 | The performance comparison of the prediction-only PIDL-IDM and PUNN over 50 experiments. | 2012.13376 | PIDL-IDM_varia.jpg |
0.417847 | 014c02c70f4742619e70b2e1c7a02d39 | Structure of the DL-based CFM. | 2012.13376 | PUNN.jpg |
0.467641 | e6c9d70a55d742bf8473911cb0edea24 | Relationship between raw trajectory data, observed data, collocation data, and training-validation-test split. | 2012.13376 | data_stuff.jpg |
0.498469 | d106004855ee4f6fb3725434ba304778 | Training and validation error curves for the accelerating regime when (n_O,n_C,α) = (20, 20, 0.1). | 2012.13376 | error_curve.jpg |
0.488561 | 147c6ee6fa9d4a6c9b5c91828c5b1273 | Architecture of PIDL. | 2012.13376 | just_PIDL.jpg |
0.485154 | 321e5c23dff14241ab96cd579e8d2772 | Illustrative flowcharts for estimating x, a vector encapsulating all parameters to be estimated, using controls optimized by (a) GRAPE, (b) RL, and (c) the analytical scheme. In all panels, the blue dashed line indicates that subsequent control should be optimized according to updated estimations of the parameters of the previous step. In (a), GRAPE has to be rerun whenever an updated estimation of the parameters is encountered. In (b), the training of the RL agent is only conducted once and controls for updated parameters can be generated directly from the agent without additional training. In (c), while the initial optimal control has to be found by either GRAPE or RL, subsequent updates can be efficiently done according to Theorem <ref>. | 2012.13377 | fig1.jpg |
0.470639 | 560713626ff84cb49b59e4c67e1c0c7f | Classical CR bound as a function of t in (a) Example 1 and (b) Example 2. The target time T=5. The parameters are (B,ϑ,φ)=(1,π/4,π/4) in Example 1 and (ω_1,ω_2,g)=(1.0,1.2,0.1) in Example 2. The dashed black curves show the results without control. The solid blue curves show the results from RL. The dash-dotted green curves and the dotted orange curves represent results obtained by GRAPE with the steepest gradient ascent (GRAPE-GD) and Adam (GRAPE-Adam) methods, respectively. For GRAPE, the initial guess of the control is zero in Example 1 and the initial guess is random in Example 2. | 2012.13377 | fig10.jpg |
0.451018 | 65b75a5d5d1c4a909cb58b7395f46b01 | Plot of (T)^-1 as a function of the target time T in (a) Example 1 with (B,ϑ,φ)=(1,π/4,π/4) and (b) Example 2 with (ω_1,ω_2,g)=(1.0,1.2,0.1). Diamonds on the blue lines show results obtained by RL. Stars located at each T are GRAPE results from Ref. <cit.>. The dashed black curves show the value without control. The solid black curves represent the optimal precision limit 4T/3 under unitary dynamics. | 2012.13377 | fig11.jpg |
0.463911 | da4857ddc9f745e2a2ac89b9673dc0bc | (a) Generalizability regarding the strength of the magnetic field in Example 1. Here evaluated at the target time T=5 is shown as a function of B' in [B-2π/T,B+2π/T] where B=1. The RL agent is trained at (B,ϑ,φ)=(1,π/4,π/4) and the dephasing rate γ=0.2, which is subsequently used to generate control pulses for B' under dephasing rates γ=0.1 (upper panel) and γ=0.4 (lower panel). (b) Generalizability regarding individual magnetic fields in Example 2. The CR bounds evaluated at the target time T=5 are shown as a function of ω_1' with fixed ω_2=1.2 and coupling strength g=0.1. The RL agent is trained at (ω_1,ω_2,g)=(1.0,1.2,0.1) and dephasing rates γ_1,2=0.1, which are subsequently used to generate control pulses for a range of ω_1' under dephasing rates γ_1,2=0.05 (upper panel) and γ_1,2=0.2 (lower panel). The dashed gray curves show the CR bound without control for comparison. | 2012.13377 | fig12.jpg |
0.49796 | f1495e4f5455417b917c9e49d4a1da81 | Schematics of the reinforcement learning framework. At the jth time step, the RL agent observes the current state as the density matrix ρ_j≡ρ(jΔ t) and takes an action as the control on the system. As a consequence, the state evolves to a new one at the (j+1)th time step while the agent receives an immediate reward. After completing one episode, the state is reset to the probe state. | 2012.13377 | fig2.jpg |
0.439289 | f9cd30a6e43a44cab5bfde6915d6a6e9 | (a) Generalizability for the strength of the magnetic field in Example 1. For RL, we show at the target time T=5 as a function of B' in [B-2π/T,B+2π/T] where B=1. We use the neural networks that are trained at (B,ϑ,φ)=(1,π/4,π/4) to generate the pulse sequences for B'. The blue crosses show the results of RL. The red squares represent the results using new pulse sequences optimized by GRAPE. The solid dark blue (red) curve show the analytical results using the pulse sequences found by RL (GRAPE) at (B,ϑ,φ)=(1,π/4,π/4). The dashed gray curve shows the CR bound without control. (b) Representative pulse profiles obtained with RL. The cyan lines show the results of B'=B. The purple lines show the results of B'=B+2π/T. (c) Representative pulse profiles obtained with GRAPE. The red lines show the results of B'=B. The brown lines show the results of B'=B+2π/T. | 2012.13377 | fig3.jpg |
0.425338 | 871dbc4569304ad3a2be0c578df4b7ed | Generalizability for the direction of the magnetic field in Example 1. The strength of the magnetic field is fixed as B=1. (a) Pseudo-color-plots showing the CR bound at the target time T=5 as a function of the parameters (ϑ',φ') in the plane [0,π]×[0,2π]. Note that the color bar has a logarithmic scale. Shown on the left is the CR bound using the analytical scheme. Shown on the right is the CR bound under controls generated by RL using the policy network trained at (B,ϑ,φ)=(1,π/4,π/4), the location of which is denoted by a blue circle. Two cuts (shown as blue dashed lines) correspond to the results in (b) and (c). (b) The CR bound as a function of φ' at fixed ϑ'=0.25π. (c) The CR bound as a function of ϑ' at fixed φ'=0.25π. | 2012.13377 | fig4.jpg |
0.397157 | 809eebe24e6a4519bf007316aff04394 | Generalizability for the individual magnetic fields in Example 2. The ZZ-coupling strength is fixed at g=0.1. (a) The CR bound as a function of ω_2' at fixed ω_1'=1.0. (b) The CR bound as a function of ω_1' at fixed ω_2'=1.2. | 2012.13377 | fig5.jpg |
0.449926 | 6e985892c67940d9a5d1057b1557a1fe | (a) Generalizability for the magnetic field using only the local controls σ_x^(1), σ_y^(1), σ_x^(2) and σ_y^(2). We show at the target time T=5 as a function of ω_1' in [1.0-π/T,1.0+π/T]. The other parameters are fixed at ω_2'=1.2 and g=0.1. (b) Lack of generalizability for the ZZ-coupling strength. For RL, we show at the target time T=5 as a function of g' in [0.1-π/T,0.1+π/T]. The magnetic fields ω_1'=1.0 and ω_2'=1.2. | 2012.13377 | fig6.jpg |
0.419698 | 96d3f1cbe6ac475b8072ddd7864677c3 | Performance of RL using different ranges of action space. We show the value of as a function of the range u_m where |u_a^(i)|⩽ u_m. The other parameters are T=5 and Δ t=0.1. The same random seed is used. The black dashed and red dotted horizontals are the CR bound without control in Examples 1 and 2, respectively. | 2012.13377 | fig7.jpg |
0.398933 | 3aa8e83197274c27b1381620077e7eae | Training processes of RL in (a) Example 1 and (b) Example 2. Here T=5 and Δ t=0.1. Blue marks represent results from a total of 2000 RL episodes with replay memory size 10 000, green marks correspond to a training involving 10 000 RL episodes with replay memory size 20 000, and red marks represent 20 000 RL episodes with replay memory size 100 000. | 2012.13377 | fig8.jpg |
0.41082 | 0df8e56ac3664a56b5fab7c9d65aefce | Time complexity of different methods. The x axis shows the system size, i.e., the number of time slices N. The y axis shows the wall-clock time cost in seconds. Diamonds on the blue line show the time cost in one RL training episode. Squares on the red line and circles on the black line are the times required by one iteration of GRAPE with and without storing the superoperators, respectively. Notice that the base-10 logarithmic scale is used for both axes. | 2012.13377 | fig9.jpg |
0.480477 | baf64704995949d88d6a8eba868aab8e | 2012.13380 | graph_confidence_deterministic.jpg |
|
0.587129 | 06edadb3ee5d4825944f2e904f7cab2b | 2012.13380 | graph_confidence_fair.jpg |
|
0.438081 | 46c2715900a144c0a02d5e8682db9d6d | 2012.13380 | graph_ratio.jpg |
|
0.411692 | 18fcf637184f46d2b457fbb1dc9ed636 | Phase diagram of MSK model: the left and right one are expected for and ^2 respectively. The green shaded region corresponds to RS regime proved in Theorem <ref> and <ref>. The expected RSB phase is labelled in both cases. In both pictures, the red curve and blue curve correspond to ^2 ρ(^2)=1 and ^2max_∈spec('^2)()=1 respectively. The right picture for ^2 is based on the evidence from some examples, see Example <ref>. | 2012.13381 | combine_phase_v1-3.jpg |
0.481746 | 543cf442d25641aa89af4e685c6ea99b | An example episode of high CAPE and substantial CAPE error: 5 days from May 24th to 28th, 2012 over Topeka, Kansas, color coded as usual. Reanalyses and model are shown at 6-hourly; IGRA soundings are every 12 hours (an additional records at 27th 18 UTC). Vertical lines mark the two examples discussed in text. CAPE biases arise from underestimated specfic humidity: in ERA renalyses on May 26 because T is low despite accurate RH, and in WRF on both May 26 and 27 because conditions are too dry even though temperature is high. | 2012.13383 | Figure10episode.jpg |
0.400548 | a9067a0463a3441c8d43288fc656a93b | The diurnal cycle in T–H space in all datasets. Color code follows the convention throughout this work. (Left) The example of Figure <ref>. Thick lines connect 00 UTC (right end) and 12 UTC (left end) values on the day of high CAPE error (May 26th 2012), with time of maximum error marked with 'x'. Faint lines connect 6-hourly values in reanalyses and model over the full 5-day period. (Right) Mean summertime diurnal cycles. Lines connect mean of 00 UTC and 12 UTC points for each dataset, for all profiles (dot-dashed), and for high-CAPE (solid) and low-CAPE (dotted) subsets, defined as 00 UTC SBCAPE values above 90th / below 10th percentile in each dataset, and values 12 hours later. In all cases the IGRA diurnal cycle involves flat or slightly increasing H; the WRF cycle decreasing H; and the ERA cycles too low humidity and too small a daytime T rise in high CAPE conditions. | 2012.13383 | Figure11slopecolumn.jpg |
0.424748 | 7dab7cd36bcf4650b9d54a4b05d582c1 | Comparison of SBCAPE and MUCAPE for all datasets, using all soundings considered. Data is binned by SBCAPE value, and we exclude values under 200 J/kg. (Left) Mean ratio of MUCAPE over SBCAPE, and (right) mean of ratio of the most unstable pressure level over surface pressure. Note that y axes are log scale. For both CAPE and pressure level, the ratio approaches 1.0 as CAPE increases: in higher CAPE conditions, the most unstable level is closer to the surface. | 2012.13383 | Figure1ratiomusb.jpg |
0.444194 | df34dfdbb1ad444db869f864bb634051 | Probability density functions (top row) and quantile ratio plots (bottom row) of CAPE from reanalysis (ERAI and ERA5), high-resolution model output (WRF), and radiosonde observations (IGRA) for MJJA 2001-2012, with times and locations matched to IGRA observations. Points with zero CAPE are excluded (36-40% of datasets, see Table <ref>). Left column shows full distribution and right column the high tail (90th percentile and above). For IGRA, the 90th percentile is ∼2800 J/kg, the 95th ∼3200 J/kg, the 97.5th ∼4000 J/kg. In PDFs (top), plots are cut off at 6000 J/kg on the x-axis, omitting less than 0.1% of all points. The most extreme values are: IGRA (6834 J/kg), ERAI (5198), ERA5 (6291), WRF (7294). In quantile ratio plots (bottom), a slope downward to the right indicates a narrower distribution. Model and reanalyses consistently underpredict CAPE values in this high tail. | 2012.13383 | Figure2distsbcape.jpg |
0.442276 | fe35c63952d544af8de09eeb6e4d1d3c | As Figure <ref>, but for MUCAPE instead of SBCAPE. Points with zero CAPE are excluded from the analysis (23-35% of the datasets, see Table <ref>). We match the time and locations of model output to IGRA observations. PDF x-axes are cut off at 6000 J/kg, as less than 0.4% of all points lie above the limit. For IGRA, the 90th percentile is about 3360 J/kg, the 95th percentile ∼4000 J/kg, and the 97.5th percentile ∼4540 J/kg. | 2012.13383 | Figure3distmucape.jpg |
0.459233 | f2dd5ebdcbf44fe388b1154292d935e6 | Snapshot of WRF simulation output at 00 UTC, July 21st, 2012. Panel colors show SBCAPE, 2 m temperature, and specific humidity. Ocean values are masked out. Circles show IGRA stations, with circle area showing the magnitude of bias in each variable and color indicating its sign (red = high, green = low). Note the low CAPE bias in the Central U.S. associated with too hot and too dry model conditions. Sounding marked “X” may also be affected by errors in the location of the warm front. | 2012.13383 | Figure4snapshot2012072100.jpg |
0.43014 | 87b6e227883e49c9b750ddeb058616d2 | Comparison of SBCAPE in WRF and radiosonde observations, for all points during summer (MJJA) 2001-2012 when observations are available, inclusive of zeroes. Color bar shows log density (midpoint color is 1% of all observations), and both axes are also log scale. (Top left) Raw data, showing wide scatter. Other panels: Recalculated WRF CAPE using (top right) observed surface temperature, (bottom left) observed surface humidity, and (bottom right) all surface values from observations. All recalculated CAPE values also involve a pressure correction whose effects are small. For analogous figures for ERAI and ERA5, see Supplementary Figures S4–S5. | 2012.13383 | Figure5calibrationdensity.jpg |
0.456945 | b61dcba13c1c4ac0b5fd5004239d21f0 | Mean radiosonde observed SBCAPE in surface temperature and specific humidity parameter space, for the entire dataset: summer (MJJA) 2001–2012 over the contiguous U.S., inclusive of all launch times and of zero values. Colors denote mean CAPE values averaged in bins of 3 K and 1.35 g/kg. Solid and dashed lines mark conditions of 100% and 50% relative humidity. Symbols '+' and 'x' mark two cases ('warm' and 'hot') used in Figure <ref>. Contours show approximate limits for 2000 and 4000 J/kg SBCAPE for all datasets with no surface corrections applied: IGRA (black), ERAI (blue), ERA5 (green), and WRF (red). Similarity of contours means that all datasets show similar bivariate distributions. See Supplementary Figure S6 for full density distributions for all datasets. | 2012.13383 | Figure6heatmapmeancape.jpg |
0.466274 | 6a4a7d2bdbe54da89a6db29b03a13654 | Comparison of SBCAPE in all datasets for specified T–H grid cell: the `Warm' example is centered at 298.5 K and 12.825 g/kg (63.4% RH), and has mean SBCAPE 810 J/kg; the `Hot' example is at 307.5 K and 16.875 g/kg (48.7% RH) with mean SBCAPE 3270 J/kg. Each bin is 3 K in width, and 1.35 g/kg in height. Top row shows uncorrected SBCAPE from reanalyses and model, and bottom corrected with IGRA surface values. Note that since the correction involves adjusting surface T and H, the profiles sampled in top and bottom rows are different. The `warm' bin has 390 profiles in the uncorrected data and 480 in the corrected, while `hot' has 2300 and 2111, respectively. Tickmarks at panel top show the mean of each distribution. Distributions are very similar; correcting surface values only slightly adjusts means (from a maximum bias of -5% in uncorrected data to -2% after correction). | 2012.13383 | Figure7binsdistribution.jpg |
0.470274 | fa77ba1e23594be3b48b4aa147c2d7bd | (Top left) Density of observed surface conditions in temperature – specific humidity parameter space at 12 UTC (early morning in the contiguous U.S.), again for 2001-2012 MJJA radiosonde observations. Contours are repeated from Figure <ref> to mark conditions associated with 2000 and 4000 J/kg SBCAPE. Darkest blue color shown is 5.6%–6.4% of distribution; lightest is 0–0.8%. Grids with no more than three samples are defined as outliers and removed (only 0.03% of all model or reanalysis samples). Nighttime and early morning conditions are tightly distributed in relative humidity (RH ∼80%) and tend to be relatively cool (T < 300 K), with almost no conditions sampled that would tend to produce SBCAPE >2000 J/kg. (Other panels) – heatmaps of density differences between models and observations for (clockwise from top left) ERA5, ERAI, and WRF. Color scale shows fractional difference after normalizing each bin by IGRA raw density. Orange = undersampling and purple = oversampling. Reanalysis and model all underestimate relative humidities (orange near the RH=100% contour) and WRF shows a strong warm dry bias (dark purple in lower right). | 2012.13383 | Figure8heatmap12utc.jpg |
0.444493 | c66b9f7ba0a64a55972c46cdbe95b730 | As in Figure <ref>, but for 00 UTC (late afternoon / early evening in the contiguous U.S.) (Top left) Density of observed surface conditions in T–H diagram. At this time period the density distribution peaks in conditions associated with 2000–4000 J/kg CAPE. Darkest blue color shown is 2.1%–2.4% of distribution; lightest is 0–0.3%. (Other panels) Density differences between renalyses / model and radiosondes. ERA5 and ERAI undersample both the highest relative humidities and the highest temperatures (orange near the RH=100% contour and on the right side), while WRF shows a warm dry bias (purple in lower right). Reanalyses and model all severely undersample the conditions associated with extreme CAPE (orange in upper right). | 2012.13383 | Figure9heatmap0utc.jpg |
0.425516 | a9193fc175094dd5a18b68429729fcea | Hierarchical model view for Gaussian process | 2012.13384 | hieGP.jpg |
0.450878 | 933f00e194104eee80fef3339b4bb1e7 | Hierarchical model view for spatial sequence with corrupted class labels y | 2012.13384 | hieSTSequence.jpg |
0.405939 | 49d762d19cd940a980ef05dceac515e4 | Illustration of geographically weighted model based on kernel function (source of image:<cit.>) | 2012.13384 | kernels.jpg |
0.506175 | 8d8f553c5a624a6587c88251c71f3d23 | Spatial operations on raster data (map algebra) | 2012.13384 | mapalgebra.jpg |
0.402682 | 3cfeef5119ed4194965ae9610c0c977a | A taxonomy of spatial prediction methods | 2012.13384 | taxonomy.jpg |
0.435136 | d83ad104787c47f4a093332b46effdb2 | 2012.13385 | fig_te.jpg |
|
0.453933 | b6ba7de0db0c43079740fcaf85782bd1 | 3D-rendition (a) and sectional view (b) of the modified single-cell 650-MHz superconducting resonator. | 2012.13388 | Figure1OverViewSetup.jpg |
0.456417 | 55cff6c7b1b6474b8fe2e0021971638f | Relative shift in the resonant frequency (a) and maximum axial electric field Ê_̂ẑ as a function of the stem length and two radii ϱ (b). | 2012.13388 | Figure2FreqShift.jpg |
0.532288 | 9dde6e43087d44e8a89c6fa7a68ba63e | Geometry and associated peak-normalized electric (r>0) and magnetic (r<0) fields associated with the Full-cell (a), half-cell (b) and modified field-enhanced resonator (c). The corresponding on-axis axial electric field E_z(r=0,z) (blue) and surface magnetic field B(θ=0,z) (red) are respectively displayed in (d),(e) and (f). In (f) the thin (resp. thick) red trace corresponds to the surface magnetic field on the cavity-wall (resp. rod) surface. | 2012.13388 | Figure3CompareModels.jpg |
0.431193 | 0958131241fa4fe5bc9c8e8025c50671 | Evolution of the maximum temperature at the cathode-rod extremity during the iterative simulations (a) and temperature dependence of the surface resistance R_s (b) used in the thermal simulation. The red and blue traces correspond respectively to the case of Nb (red) and Nb_3Sn. | 2012.13388 | Figure4Temp_final.jpg |
0.471433 | d9df4f31806b46c183a20d9c828cdc34 | Steady-state temperature map (a) and associated peak-normalized electric- (upper plane r>0) and magnetic- (lower plane r<0) field amplitudes (b) for the FE configuration with optimized conical rod. | 2012.13388 | Figure5ThermalModel.jpg |
0.495287 | 65b00071e49b49328ac263ccfc554798 | On-axis axial electric field E_z(r=0) (solid traces) and kinetic energy K (dashed lines) in the FE (red) and HC (blue) configurations (a) and radial component of the electric field E_r (z=) (right). The dashed black line represent the radius of the rod (5 mm). The electromagnetic fields are normalized to the same dissipated power for comparison. | 2012.13388 | Figure6Energygain.jpg |
0.434243 | eeb1791d3a884ed5bc707547210b3473 | Pareto fronts between transverse emittance and bunch duration associated with the multi-objective optimizations for the HC and FE configurations. The colors represent beam brightness B. The inset (b) displays a zoomed portion of the FE plot. | 2012.13388 | Figure7OptResults.jpg |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.