nopperl/arxiv-image-text · Datasets at Hugging Face

dedup-isc-ft-v107-score float64 0.32 1	uid stringlengths 32 32	text stringlengths 0 32.5k	paper_id stringlengths 1 14	original_image_filename stringlengths 5 222
0.419592	4fd4f9ca1acb4b93aab054930f933786	The SDC protection scheme.	2012.09292	SDC-scheme.jpg
0.459905	d79511cdb7fb41cbae8d7081b6d9c271	The SSN protection scheme	2012.09292	SNN-scheme.jpg
0.426108	dad4ee123b9641e4889c6670a77b7dbb	The SSN behavior at runtime.	2012.09292	SSN-scheme-runtime.jpg
0.442143	66f40d295987446f9d0c865793217605	A snapshot of the Antimine game.	2012.09292	antimine.jpg
0.483154	091792a35208418aba41093b57ce44c4	The protection workflow.	2012.09292	antirepackaging-flow.jpg
0.45462	a5c0dec28faa4326bc3e656bd79d963c	CPU overhead distribution	2012.09292	cpu-overhead.jpg
0.444344	45de1a8a4ac948cb97ecec1a75ff0ec8	Transformation of a candidate node into a logic bomb.	2012.09292	creation-of-logic-bomb.jpg
0.430811	cc3e1721e9e04611a5f2141b28c85789	Processing time distribution	2012.09292	graph-processing-time.jpg
0.436888	db3d8f56125b4c39ac48f50f069c5e7f	Size overhead distribution	2012.09292	graph-size-overhead.jpg
0.561766	39a97c6608f74285b00b28c4cda5cbb0	Java AT distribution	2012.09292	java-at.jpg
0.520497	932b5f312c88430c92b3fdb441a9be69	Java AT & logic bomb	2012.09292	javaAT-lb.jpg
0.458288	6ff6a46473a64ee9a71b90dca1fbc67d	Logic bombs distribution	2012.09292	lb.jpg
0.457138	68186cc92cb747388dfdbf085281020f	Memory overhead distribution	2012.09292	memory-overhead.jpg
0.419295	c72a05adfb554d55af56379a8a96cdd6	NRP applied to the method of Antimine.	2012.09292	native-bombdroid-evolution-example.jpg
0.417163	0f88311867ab40318619fa2664a28b82	The Native Repackaging Protection scheme.	2012.09292	native-bombdroid-evolution.jpg
0.445749	e475c5dac5bc4c4fb9791442134d4903	Nesting level distribution	2012.09292	nesting.jpg
0.512977	f98ce4d9d3fc4b0a98e88946a62cdc67	Logic bombs design	2012.09292	repack.jpg
0.429539	b05b7c4c8176400eaf037d849ed8fa14	Dynamic Self-Protection and Tamperproofing: runtime behavior	2012.09292	self-protecting-encrypted-dex-runtime.jpg
0.472756	e757f58a00614211baf39033d5eeb0d4	The Dynamic Self-Protection and Tamperproofing scheme.	2012.09292	self-protecting-encrypted-dex.jpg
0.466693	abebea68546b432c87e651e4bc415910	The original (top side) and the tampered (bottom side) library file.	2012.09292	sochanged.jpg
0.392314	b13c4d91477e494494c98f93b16d64f5	App repackaging: Threat model.	2012.09292	threat-model.jpg
0.536286	f9662910ca424a3281da6cca398f8d0e	NN controllers vs. expert data of GTM	2012.09293	GTMNN.jpg
0.469483	1e5e13dd36da4537ab04518904983b2f	ROA inner-approximations and state constraint set X of GTM	2012.09293	GTMROA.jpg
0.437337	f71875e106ed4df39f134d887f1d4100	A ROA inner-approximation and state constraint set X of the inverted pendulum	2012.09293	InvertPen.jpg
0.460515	6c06878c0c764ec69146f5d7e6153850	NN controller vs. expert data of the inverted pendulum	2012.09293	InvertPenNN.jpg
0.431285	c7996b5e073a49fb98f84949e3914d6e	f(N) vs. LQ^-1 after running 16 iterations of the ADMM algorithm of the inverted pendulum	2012.09293	InvertPen_ADMM_sol.jpg
0.506214	3e8e3054234548a9929063348f79b847	Loop transformation. If ϕ is in the sector [α_ϕ, β_ϕ], then ϕ̃ is in the sector [-1_n_ϕ× 1, 1_n_ϕ× 1].	2012.09293	LoopTransform.jpg
0.543187	25ddcb3d4303420a86833c94ea82120c	NN representation to isolate the nonlinearities ϕ.	2012.09293	NNLFT.jpg
0.482195	013c22ba23ce41dd8b610729e2281cab	Feedback system with plant G and NN π	2012.09293	NominalFeedback.jpg
0.479842	8bfca3bb080f468a96a20e117eb79317	Sector constraints on tanh	2012.09293	tanhSector.jpg
0.426803	f36f42bdcecc401ca28b4743f71ae331	The dispersion relation of a Rashba AFM metal with n_0=x̂ in two different directions ϕ_k=0 (top) and ϕ_k=π/2 (bottom). The solid and dashed curves show the conduction (η=+1) and valence (η=-1) bands, respectively, while red and blue colors mark the s=+1 and s=-1 spin chirality of the subbands, respectively.	2012.09297	disp.jpg
0.5247	92b3f7a9b7c241e79c24dd7c0b2a0d7a	The frequency dependence of the laser-induced nonequilibrium interfacial-type DM interaction in a Rashba AFM system as a function of the Fermi energy ε_F. The induced DM interaction changes sign and peaks in magnitude at the insulator-metal transition. The system parameters are the same as in Fig. <ref>.	2012.09297	im3_3.jpg
0.537575	b69a7fdbaadb4a8794855672c2fd8a35	Numerical results for the laser-induced nonequilibrium bulk-type DM interaction in a Rashba AFM system as a function of the Fermi energy ε_F for different laser polarizations. The system parameters are the same as in Fig. <ref>.	2012.09297	im3_b2.jpg
0.467173	7d73c711ee084a9d88ac2bd4ff68f1f9	Numerical results for the laser-induced nonequilibrium interfacial-type DM interaction in a Rashba AFM system as a function of the Fermi energy ε_F, for a laser pulse with linear polarization along x (blue) and y (red) directions, and circular polarization (green). The two components of the DM tensor related to the interfacial DM interaction are plotted by dashed lines δ D_x^i, and solid lines δ D_y^i. The equilibrium direction of the AFM order parameter is n_0=x̂. The inset shows the nonequilibrium DM interaction inside and just above the gap. For the numerical calculations, we use ε_0=4 eV, 𝒥_sd=1 eV, k_0α_R = 0.4 eV, a=0.4 nm, Ω=0.75 eV and T=300 K.	2012.09297	im3_y.jpg
0.466274	fed11e0ad3764a398d033501a7f923f6	Comparing approaches for counterfactual generation between gradient descent in feature space (left) and Latent-CF gradient descent in the autoencoder latent space (right)	2012.09301	Architecture.jpg
0.509296	33aa2b74e62648e0a1030b225aa9a1b3	feature gradient descent with clipping	2012.09301	clip.jpg
0.420561	a87198ca5b8b47c9b0a615654984abbf	Histogram of Kernel Density Estimate probabilities per pixel computed on original images and counterfactuals. In this case the counterfactual images generated by baseline methods have more low probability pixels than the original images, signifying out of distribution changes.	2012.09301	hist.jpg
0.434347	c7fcaa5aa7a84d4dbaadc19b2a74ac15	In-distribution probability over all test image pixels	2012.09301	kde.jpg
0.531464	0e24ccd66e734b718f91b0dd042b0e26	Latent-CF	2012.09301	latent.jpg
0.459375	01c2d4e4c68349af85d0f55a62cd9ce3	MAD loss	2012.09301	mad.jpg
0.725533	a99d38559ddc4901b6615d9c8adcd883	Original Image	2012.09301	original.jpg
0.477346	358d94942b354afeb1754a7787461ae9	Prototype-CF	2012.09301	proto.jpg
0.490449	082f2c3fb6fa437ab74a304d41c442d1	feature gradient descent	2012.09301	simple.jpg
0.453704	a12eb82863c44ac4aecdfe429f493007	Fraction of pixels changed by each method	2012.09301	sparsity.jpg
0.445113	95f18aa9d5804ea0af90689f077df8d7	Counterfactual run time for each method	2012.09301	time.jpg
0.445868	02402cfc5dc54742a2c88d9a9b3b49d1	Adjustment Graphs	2012.09306	adjustments.jpg
0.413961	c372dddab7314821b98c7a8a618e0183	Simulated spectra for two fake galaxies with a simulated drift of 0.5 pix/hr added. The resulting stacked spectrum and extracted spectrum are on the right of each plot (with a green line overlaid showing the strength of the emission line when there is no drift). (left) For the brighter galaxy (H_160 < 25 mag), while the shift can been seen in the gaussians in the final stack's spatial profile, the overall final profile shape is hardly changed. (right) For the fainter galaxy (H_160 > 25 mag), the effect from the drift is more noticeable and significant when working with faint emission lines, as can be seen in the increase in flux gained from the extracted 1D spectrum when there is no drift (green line).	2012.09308	bright-vs-faint-drift.jpg
0.406632	f22be385618746738527c8cc78f90ee2	Simulated spectra for two fake galaxies (no drift added). (left) For a reasonably bright source (H_160 < 25 mag), >4 hrs produces detections of both strong emission lines and the source's continuum. Both of these features have been identified in the extracted 1D spectrum and the collapsed spatial profile. (right) For a fainter source (H_160 > 25 mag), >4 hrs may result in a faint detection of just the emission lines (not enough to detect continuum). The faint emission lines have been identified in the extracted 1D spectrum.	2012.09308	bright-vs-faint.jpg
0.467612	8d157a9a33784e87917b0c6d2e97ee71	(left) Optical path of MOSFIRE from Ref. , zoomed in to highlight the location of the guider and the guider optics involved. The “guider optics” tags in the diagram are encircled with purple rectangles. (right) Layout of the guider and science detector FOVs, along with their respective coordinate systems and their separation on sky at PA = 0.	2012.09308	guider-MOSFIRE-coordinates.jpg
0.498995	2a44984dcf114c1e8749aac0392d79ea	A basic schematic showing the two different tracking testing that can be done using a single rising star field (sssec:guider-star-field). As the star field begins to rise in the night (yellow box), the elevation of the field changes rapidly but the rotation of the detector would be minimal. As the star field traverses its highest point in transit (blue box), the elevation of the field would remain relatively constant while the rotation of the detector would be changing more rapidly.	2012.09308	guider-changing-el.jpg
0.439534	992631aef78e4fb299f0c8e04ab2c973	A basic schematic of the testing setup used for sssec:guider-fixed-el (not to scale). Using the same guiding star at a fixed elevation, and rotating the science detector around it, we ran imaging covering a full 360circle. If there was no significant flexure compensation still needed between the optical guider and the NIR science detector, this test would be able to map out a near-perfect great circle on the sky.	2012.09308	guider-great-circle.jpg
0.429885	0151701f8b974b37907cf4aa9ba7a531	Coordinate offsets in the guider FOV for the rotating testing, colored by elevation. (left) Offsets from testing using the current system. (right) The same offsets with additional correction applied, shrinking the spread of points.	2012.09308	guider_correction.jpg
0.466688	ceb6445b0f454e3fa20820059ad7e637	Basic schematic of the Keck I telescope, showing the path of the incoming light and the locations of the primary and secondary mirrors. Additionally, the underside of the primary mirror is shown, highlighting the location of Cassegrain instruments such as the MOSFIRE spectrograph. The curved arrow indicates the rotation of the instrument.	2012.09308	keck_schematic.jpg
0.454978	cbad3f9526fd49239cf6a690c54b364d	FCS model pixel shifts for a range of elevation and physical rotator angle compared to a reference.	2012.09308	model.jpg
0.42851	988cc19127124438b6956368bbdde515	Full optical train for MOSFIRE, from Ref.	2012.09308	optical_path_full_McLean2012.jpg
0.426295	309bf68a00924f53a08e6b01a5824f14	(top left) An example of a raw J-band frame, zoomed in to shows a handful of slits in the data, highlighting how the atmospheric skylines can be arranged depending upon where the slits are located in the mask. (top right & bottom) The same datasets from Figure <ref> showing different examples of the (very small) relative mask drift in these datasets.	2012.09308	slit_drift_map_SPIE_paper.jpg
0.456154	4a22c398871e4bababd4781373defd0a	(top left) An example of the star drift measured in previous MOSFIRE data (these data are also shown in Figure <ref>), with the moment of realignment highlighted; for this test we purposefully did not realign. (top right & bottom) The data resulting from modifying the guider flexure model parameters.	2012.09308	star_drift_guider.jpg
0.517803	19fa4849dad843709db5be3ea16cac09	(top left) An example of a raw J-band frame, zoomed in on the slit holding an alignment star. The blue rectangle marks a region devoid of atmospheric skylines that is used to generate the spatial profile seen in the right of the 2D spectrum. This profile can then be used as the measurement of the star's spatial location in that raw frame. (top right & bottom) Three example datasets from 2017-2019 showing different examples of the relative spatial drifts of an alignment star in these datasets. The scatter points are colored to indicate elevation for each frame, and the scatter point shapes denote the two different nods used in observing. Note that the pixel scale is 018/pix.	2012.09308	star_drift_map_SPIE_paper.jpg
0.378232	344558451186465aae0a7469faade57c	Illustration of <ref>	2012.09310	RP20_illustration_prop1.jpg
0.404597	e0e3135f539f46deb79b418160eb2393	An example of PCL architecture, where ResNet-34 is adopted as backbone. The features are shown as the shape of their tensors, and proper reshaping is performed. ⊗ denotes matrix multiplication, and θ and ϕ are two 1 × 1 × 1 convolutions.	2012.09311	detailed_network.jpg
0.42205	aefc80be0b7542dfb1a9fbad77a4bcde	Visualization of false negative (FN) and false positive (FP) predictions on DFDC dataset.	2012.09311	failure_cases.jpg
0.383989	c07a2270e49f4ec296a608a27a99f39b	Illustration of the workflow of I2G. For each source and target frame pair, a morphed mask is generated by taking a convex hull of the landmarks followed by elastic deformation and Gaussian blur. The masked region of the target image is replaced by that of the source with blending techniques <cit.>.	2012.09311	i2g.jpg
0.444865	2700fa846bda474a8df543b11ceb1b66	Visualization of PCL architecture. The consistency branch focuses on measuring the consistency of image patches according to their source features, and outputs a visualization map that indicates the manipulated region. A classification branch is applied after the learned feature map and predicts the binary score for Deepfake detection.	2012.09311	network.jpg
0.379878	f36ae633a932408db1859ed07cdc907a	Visualization of the consistency heatmaps, which localize the manipulated regions. The predictions are obtained in the cross-dataset experiments, where our model is trained with only real videos of FF++, augmented by I2G. The ground truth masks are generated by DSSIM, as discussed in Section <ref>.	2012.09311	qualitative_results.jpg
0.465907	12c58a836c4a42cdb0158f6ea1f45e2d	Visualization of the consistency maps, which localize the manipulated regions. The predictions are obtained in the cross-dataset experiments, where our model is trained with only real videos of FF++, augmented by I2G. The ground truth masks are generated by DSSIM, as discussed in Section 4.2.	2012.09311	qualitative_results_2.jpg
0.399549	767b603c30fc420d818e5872c1b41187	An overview of our patch-wise consistency learning. An “self-inconsistent” image is generated by stitching the target and source faces, where the mask indicates the region of modification and separates the patches into two groups. For each patch, we compute its source features consistencies with the rest, and their corresponding consistency scores are shown on the right.	2012.09311	teaser.jpg
0.469445	8585168655bd457aa18b82c05d786eee	Schematic of the GoldEnvSim components.	2012.09312	Fig1.jpg
0.482261	e9d8c0826f5d47d4994f3cc5434ff24f	Visualization of the surface pressure on simulation domain at the time 10/03/2019 11:00:00 GMT. The unit is in hPa.	2012.09312	Fig11.jpg
0.40786	d2b4a3fa24b441eca054fcd621b71464	Visualization of the accumulated rainfall over simulation period. The result is sum of convective precipitation and non-convective precipitation.	2012.09312	Fig13.jpg
0.399482	ca6b8047709048faa56a5370840cad28	Interface of the GoldEnvSim software.	2012.09312	Fig2.jpg
0.371538	a77be345c490494fb5675c5a0e6210e1	The IDV was integrated into the GoldEnvSim for WRF output dataset readout and visualization.	2012.09312	Fig4.jpg
0.406305	2e1d80ca4db741d2aef737c85d9b0179	Visualization of the temperature field at 2m-height simulated at 11/03/2019 09:00:00 GMT. The unit is in ^oC.	2012.09312	Fig5.jpg
0.471203	7453c1d908e34557ae869733c0df72d6	Hybrid visualization of the 2m-temperature field over typography simulated at 11/03/2019 09:00:00 GMT. This facilitates users a fusionable view on multiple variables.	2012.09312	Fig6.jpg
0.369947	449d0484a29341a089c10bcada5eff41	Representation of the wind bards with the speed (in m/s) is colored at the time 10/03/2019 06:00:00 GMT.	2012.09312	Fig8.jpg
0.446553	3c55b1c41b4e4766aa81530f24264059	The combined decoder and regressor architecture. The decoder maps C space input to the 16x16 intermediary image space I while the regressor maps from I to a single output ∈ℝ.	2012.09313	architecture.jpg
0.469523	71a0e313c9a64a8cb909eceff2a1d7c7	Example of configuration space for simple distance to line estimation. Here d is the distance in meters to the line and θ is the yaw angle.	2012.09313	cspace.jpg
0.39697	c9ba14ee83b4481cb72a8992e854d1bd	Original camera image from the AirSim simulation environment. Crop outlined in red, post processed training image inset and outlined in green. View is from the point in configuration space (d=1.5481, θ=0.0136).	2012.09313	front_center_gray_rect_crop_bw.jpg
0.372351	1b9629fc77f74dffbd36c5c4962187c3	A set of six training images (top), their reconstructions (middle), and reconstructions conditioned on z=0 (bottom) from six different points in configuration space.	2012.09313	reconstructions_comparison.jpg
0.381419	b10cc4f2c59746899423ba1fce80ec9d	Points in C while varying z. The perspective defined by c remains fixed while z varies the line break perturbation.	2012.09313	z_comparison.jpg
0.471656	fdfe668d8b774b6bbee438c9726f4d78	2D materials MnCl3 by element substitution with formation energy -1.45 eV	2012.09314	Fig10_MnCl3.jpg
0.400754	19e487fd6d994127b257a86c4f6e1c33	2D material V2S3 by element substitution with formation energy -0.907 eV	2012.09314	Fig10_V2S3.jpg
0.422555	84fd2e133cbc40d0b56f0f5b90e23686	2D materials WF3 by element substitution with formation energy -2.255 eV	2012.09314	Fig10_WF3.jpg
0.458394	25d56b4b912349ffa6aad7154744abe0	2D material YI2 by element substitution with formation energy -1.26 eV	2012.09314	Fig10_YI2.jpg
0.419854	1681e48695a947e294ac849cb38b365a	Generated materials distribution information.	2012.09314	ICSD.jpg
0.505764	539fd7205a7446839e3bbf498c8970fb	The framework of template based structure prediction. The main parts are: predicting new formula's space group; finding candidate template formulas according to the same element number and the same space group; calculating and sorting ElMD between the new formula and template formulas; selecting top n templates to do element substitution to get the new formula's structure.	2012.09314	TAMP.jpg
0.416629	0029975a912343feace85b128d853558	False-positive curve of the 2D materials RF classifier.	2012.09314	false_pos.jpg
0.439291	a16f9bf42daf40ffa5111deceb5b7259	Formation energy distribution of predicted 2D materials.	2012.09314	fe_hist.jpg
0.476996	2b4b918f33d840a98d3d0c5ccceea051	Framework for generation and prediction of 2D materials. It comprises four components. The green part: a GAN based composition generation module for breeding chemically valid materials. The blue part: a composition based random forest 2D materials classifier. The orange part: a template based element substitution structure predictor, as well as yellow part: DFT validation.	2012.09314	framework.jpg
0.414895	7eedf4148c1d41a4ae31ddbe89418093	Element frequency distribution. The abscissa represents 112 chemical elements arranged according to the atomic number from hydrogen (H) to Copernicium (Cn), vertical axis indicates the frequency of each element. The red curve denotes the distribution information of 2dMaterials dataset. The green curve shows our generated candidate materials. The blue line represents our predicted 2D materials.	2012.09314	frequency.jpg
0.471841	c45ab0eace244d4ea63abd541deaa7d7	Architecture of MatGAN. Generator (G) learns from known material compositions to generate realistic samples while discriminator (D) learns to determine whether a sample is a real one or generated one. Alternative training of D and G will improve the performance of both G and D.	2012.09314	gan.jpg
0.434182	b19f3adf13b044c9bdde793b213372b7	20,000 2D materials in the V2DB	2012.09314	npj_pred2K+2d.jpg
0.415628	6ba961f6fa404d089a52b0f091b4d1f9	6,000 2D materials in MP and V2DB vs. known ones	2012.09314	our+npj+2d_6000.jpg
0.435763	80c58bd517374ff1b25a3ab515c78e77	New 2D materials (Probability > 0.95) vs. known ones	2012.09314	our1485_pred+2d.jpg
0.398358	e224865768354d38930b3a3379421ae4	20,000 2D materials in the whole MP composition space	2012.09314	our20k_pred+2d.jpg
0.410501	3c33a5750e364bfa9cda4108d75e3f4f	One hot representation of material composition PuP2H6CO8. Brown color indicates the atom number of corresponding element in the specific material.	2012.09314	representation.jpg
0.564577	c4d8bb8cdc914a218e84f14bcde26a28	ROC curve of the 2D materials RF classifier	2012.09314	roc_cv.jpg
0.562145	8ea004431e2d43f98a44764b0472a974	𝔼[X^α_t] for different values of α.	2012.09317	example_esp.jpg
0.434194	16c52671f66d45b49cd208d5e6130715	P^α_1,0(t) for different values of α.	2012.09317	example_p10.jpg
0.480043	f1b9f4c5411d4bca84fadf6a33603df5	P^α_1,1(t) for different values of α.	2012.09317	example_p11.jpg
0.49437	e44fbd1ab7164b9881375ae21a4fa041	Var(X^α_t) for different values of α.	2012.09317	example_var.jpg

0.419592

4fd4f9ca1acb4b93aab054930f933786

The SDC protection scheme.

2012.09292

SDC-scheme.jpg

0.459905

d79511cdb7fb41cbae8d7081b6d9c271

The SSN protection scheme

2012.09292

SNN-scheme.jpg

0.426108

dad4ee123b9641e4889c6670a77b7dbb

The SSN behavior at runtime.

2012.09292

SSN-scheme-runtime.jpg

0.442143

66f40d295987446f9d0c865793217605

A snapshot of the Antimine game.

2012.09292

antimine.jpg

0.483154

091792a35208418aba41093b57ce44c4

The protection workflow.

2012.09292

antirepackaging-flow.jpg

0.45462

a5c0dec28faa4326bc3e656bd79d963c

CPU overhead distribution

2012.09292

cpu-overhead.jpg

0.444344

45de1a8a4ac948cb97ecec1a75ff0ec8

Transformation of a candidate node into a logic bomb.

2012.09292

creation-of-logic-bomb.jpg

0.430811

cc3e1721e9e04611a5f2141b28c85789

Processing time distribution

2012.09292

graph-processing-time.jpg

0.436888

db3d8f56125b4c39ac48f50f069c5e7f

Size overhead distribution

2012.09292

graph-size-overhead.jpg

0.561766

39a97c6608f74285b00b28c4cda5cbb0

Java AT distribution

2012.09292

java-at.jpg

0.520497

932b5f312c88430c92b3fdb441a9be69

Java AT & logic bomb

2012.09292

javaAT-lb.jpg

0.458288

6ff6a46473a64ee9a71b90dca1fbc67d

Logic bombs distribution

2012.09292

lb.jpg

0.457138

68186cc92cb747388dfdbf085281020f

Memory overhead distribution

2012.09292

memory-overhead.jpg

0.419295

c72a05adfb554d55af56379a8a96cdd6

NRP applied to the method of Antimine.

2012.09292

native-bombdroid-evolution-example.jpg

0.417163

0f88311867ab40318619fa2664a28b82

The Native Repackaging Protection scheme.

2012.09292

native-bombdroid-evolution.jpg

0.445749

e475c5dac5bc4c4fb9791442134d4903

Nesting level distribution

2012.09292

nesting.jpg

0.512977

f98ce4d9d3fc4b0a98e88946a62cdc67

Logic bombs design

2012.09292

repack.jpg

0.429539

b05b7c4c8176400eaf037d849ed8fa14

Dynamic Self-Protection and Tamperproofing: runtime behavior

2012.09292

self-protecting-encrypted-dex-runtime.jpg

0.472756

e757f58a00614211baf39033d5eeb0d4

The Dynamic Self-Protection and Tamperproofing scheme.

2012.09292

self-protecting-encrypted-dex.jpg

0.466693

abebea68546b432c87e651e4bc415910

The original (top side) and the tampered (bottom side) library file.

2012.09292

sochanged.jpg

0.392314

b13c4d91477e494494c98f93b16d64f5

App repackaging: Threat model.

2012.09292

threat-model.jpg

0.536286

f9662910ca424a3281da6cca398f8d0e

NN controllers vs. expert data of GTM

2012.09293

GTMNN.jpg

0.469483

1e5e13dd36da4537ab04518904983b2f

ROA inner-approximations and state constraint set X of GTM

2012.09293

GTMROA.jpg

0.437337

f71875e106ed4df39f134d887f1d4100

A ROA inner-approximation and state constraint set X of the inverted pendulum

2012.09293

InvertPen.jpg

0.460515

6c06878c0c764ec69146f5d7e6153850

NN controller vs. expert data of the inverted pendulum

2012.09293

InvertPenNN.jpg

0.431285

c7996b5e073a49fb98f84949e3914d6e

f(N) vs. LQ^-1 after running 16 iterations of the ADMM algorithm of the inverted pendulum

2012.09293

InvertPen_ADMM_sol.jpg

0.506214

3e8e3054234548a9929063348f79b847

Loop transformation. If ϕ is in the sector [α_ϕ, β_ϕ], then ϕ̃ is in the sector [-1_n_ϕ× 1, 1_n_ϕ× 1].

2012.09293

LoopTransform.jpg

0.543187

25ddcb3d4303420a86833c94ea82120c

NN representation to isolate the nonlinearities ϕ.

2012.09293

NNLFT.jpg

0.482195

013c22ba23ce41dd8b610729e2281cab

Feedback system with plant G and NN π

2012.09293

NominalFeedback.jpg

0.479842

8bfca3bb080f468a96a20e117eb79317

Sector constraints on tanh

2012.09293

tanhSector.jpg

0.426803

f36f42bdcecc401ca28b4743f71ae331

The dispersion relation of a Rashba AFM metal with n_0=x̂ in two different directions ϕ_k=0 (top) and ϕ_k=π/2 (bottom). The solid and dashed curves show the conduction (η=+1) and valence (η=-1) bands, respectively, while red and blue colors mark the s=+1 and s=-1 spin chirality of the subbands, respectively.

2012.09297

disp.jpg

0.5247

92b3f7a9b7c241e79c24dd7c0b2a0d7a

The frequency dependence of the laser-induced nonequilibrium interfacial-type DM interaction in a Rashba AFM system as a function of the Fermi energy ε_F. The induced DM interaction changes sign and peaks in magnitude at the insulator-metal transition. The system parameters are the same as in Fig. <ref>.

2012.09297

im3_3.jpg

0.537575

b69a7fdbaadb4a8794855672c2fd8a35

Numerical results for the laser-induced nonequilibrium bulk-type DM interaction in a Rashba AFM system as a function of the Fermi energy ε_F for different laser polarizations. The system parameters are the same as in Fig. <ref>.

2012.09297

im3_b2.jpg

0.467173

7d73c711ee084a9d88ac2bd4ff68f1f9

Numerical results for the laser-induced nonequilibrium interfacial-type DM interaction in a Rashba AFM system as a function of the Fermi energy ε_F, for a laser pulse with linear polarization along x (blue) and y (red) directions, and circular polarization (green). The two components of the DM tensor related to the interfacial DM interaction are plotted by dashed lines δ D_x^i, and solid lines δ D_y^i. The equilibrium direction of the AFM order parameter is n_0=x̂. The inset shows the nonequilibrium DM interaction inside and just above the gap. For the numerical calculations, we use ε_0=4 eV, 𝒥_sd=1 eV, k_0α_R = 0.4 eV, a=0.4 nm, Ω=0.75 eV and T=300 K.

2012.09297

im3_y.jpg

0.466274

fed11e0ad3764a398d033501a7f923f6

Comparing approaches for counterfactual generation between gradient descent in feature space (left) and Latent-CF gradient descent in the autoencoder latent space (right)

2012.09301

Architecture.jpg

0.509296

33aa2b74e62648e0a1030b225aa9a1b3

feature gradient descent with clipping

2012.09301

clip.jpg

0.420561

a87198ca5b8b47c9b0a615654984abbf

Histogram of Kernel Density Estimate probabilities per pixel computed on original images and counterfactuals. In this case the counterfactual images generated by baseline methods have more low probability pixels than the original images, signifying out of distribution changes.

2012.09301

hist.jpg

0.434347

c7fcaa5aa7a84d4dbaadc19b2a74ac15

In-distribution probability over all test image pixels

2012.09301

kde.jpg

0.531464

0e24ccd66e734b718f91b0dd042b0e26

Latent-CF

2012.09301

latent.jpg

0.459375

01c2d4e4c68349af85d0f55a62cd9ce3

MAD loss

2012.09301

mad.jpg

0.725533

a99d38559ddc4901b6615d9c8adcd883

Original Image

2012.09301

original.jpg

0.477346

358d94942b354afeb1754a7787461ae9

Prototype-CF

2012.09301

proto.jpg

0.490449

082f2c3fb6fa437ab74a304d41c442d1

feature gradient descent

2012.09301

simple.jpg

0.453704

a12eb82863c44ac4aecdfe429f493007

Fraction of pixels changed by each method

2012.09301

sparsity.jpg

0.445113

95f18aa9d5804ea0af90689f077df8d7

Counterfactual run time for each method

2012.09301

time.jpg

0.445868

02402cfc5dc54742a2c88d9a9b3b49d1

Adjustment Graphs

2012.09306

adjustments.jpg

0.413961

c372dddab7314821b98c7a8a618e0183

Simulated spectra for two fake galaxies with a simulated drift of 0.5 pix/hr added. The resulting stacked spectrum and extracted spectrum are on the right of each plot (with a green line overlaid showing the strength of the emission line when there is no drift). (left) For the brighter galaxy (H_160 < 25 mag), while the shift can been seen in the gaussians in the final stack's spatial profile, the overall final profile shape is hardly changed. (right) For the fainter galaxy (H_160 > 25 mag), the effect from the drift is more noticeable and significant when working with faint emission lines, as can be seen in the increase in flux gained from the extracted 1D spectrum when there is no drift (green line).

2012.09308

bright-vs-faint-drift.jpg

0.406632

f22be385618746738527c8cc78f90ee2

Simulated spectra for two fake galaxies (no drift added). (left) For a reasonably bright source (H_160 < 25 mag), >4 hrs produces detections of both strong emission lines and the source's continuum. Both of these features have been identified in the extracted 1D spectrum and the collapsed spatial profile. (right) For a fainter source (H_160 > 25 mag), >4 hrs may result in a faint detection of just the emission lines (not enough to detect continuum). The faint emission lines have been identified in the extracted 1D spectrum.

2012.09308

bright-vs-faint.jpg

0.467612

8d157a9a33784e87917b0c6d2e97ee71

(left) Optical path of MOSFIRE from Ref. , zoomed in to highlight the location of the guider and the guider optics involved. The “guider optics” tags in the diagram are encircled with purple rectangles. (right) Layout of the guider and science detector FOVs, along with their respective coordinate systems and their separation on sky at PA = 0.

2012.09308

guider-MOSFIRE-coordinates.jpg

0.498995

2a44984dcf114c1e8749aac0392d79ea

A basic schematic showing the two different tracking testing that can be done using a single rising star field (sssec:guider-star-field). As the star field begins to rise in the night (yellow box), the elevation of the field changes rapidly but the rotation of the detector would be minimal. As the star field traverses its highest point in transit (blue box), the elevation of the field would remain relatively constant while the rotation of the detector would be changing more rapidly.

2012.09308

guider-changing-el.jpg

0.439534

992631aef78e4fb299f0c8e04ab2c973

A basic schematic of the testing setup used for sssec:guider-fixed-el (not to scale). Using the same guiding star at a fixed elevation, and rotating the science detector around it, we ran imaging covering a full 360circle. If there was no significant flexure compensation still needed between the optical guider and the NIR science detector, this test would be able to map out a near-perfect great circle on the sky.

2012.09308

guider-great-circle.jpg

0.429885

0151701f8b974b37907cf4aa9ba7a531

Coordinate offsets in the guider FOV for the rotating testing, colored by elevation. (left) Offsets from testing using the current system. (right) The same offsets with additional correction applied, shrinking the spread of points.

2012.09308

guider_correction.jpg

0.466688

ceb6445b0f454e3fa20820059ad7e637

Basic schematic of the Keck I telescope, showing the path of the incoming light and the locations of the primary and secondary mirrors. Additionally, the underside of the primary mirror is shown, highlighting the location of Cassegrain instruments such as the MOSFIRE spectrograph. The curved arrow indicates the rotation of the instrument.

2012.09308

keck_schematic.jpg

0.454978

cbad3f9526fd49239cf6a690c54b364d

FCS model pixel shifts for a range of elevation and physical rotator angle compared to a reference.

2012.09308

model.jpg

0.42851

988cc19127124438b6956368bbdde515

Full optical train for MOSFIRE, from Ref.

2012.09308

optical_path_full_McLean2012.jpg

0.426295

309bf68a00924f53a08e6b01a5824f14

(top left) An example of a raw J-band frame, zoomed in to shows a handful of slits in the data, highlighting how the atmospheric skylines can be arranged depending upon where the slits are located in the mask. (top right & bottom) The same datasets from Figure <ref> showing different examples of the (very small) relative mask drift in these datasets.

2012.09308

slit_drift_map_SPIE_paper.jpg

0.456154

4a22c398871e4bababd4781373defd0a

(top left) An example of the star drift measured in previous MOSFIRE data (these data are also shown in Figure <ref>), with the moment of realignment highlighted; for this test we purposefully did not realign. (top right & bottom) The data resulting from modifying the guider flexure model parameters.

2012.09308

star_drift_guider.jpg

0.517803

19fa4849dad843709db5be3ea16cac09

(top left) An example of a raw J-band frame, zoomed in on the slit holding an alignment star. The blue rectangle marks a region devoid of atmospheric skylines that is used to generate the spatial profile seen in the right of the 2D spectrum. This profile can then be used as the measurement of the star's spatial location in that raw frame. (top right & bottom) Three example datasets from 2017-2019 showing different examples of the relative spatial drifts of an alignment star in these datasets. The scatter points are colored to indicate elevation for each frame, and the scatter point shapes denote the two different nods used in observing. Note that the pixel scale is 018/pix.

2012.09308

star_drift_map_SPIE_paper.jpg

0.378232

344558451186465aae0a7469faade57c

Illustration of <ref>

2012.09310

RP20_illustration_prop1.jpg

0.404597

e0e3135f539f46deb79b418160eb2393

An example of PCL architecture, where ResNet-34 is adopted as backbone. The features are shown as the shape of their tensors, and proper reshaping is performed. ⊗ denotes matrix multiplication, and θ and ϕ are two 1 × 1 × 1 convolutions.

2012.09311

detailed_network.jpg

0.42205

aefc80be0b7542dfb1a9fbad77a4bcde

Visualization of false negative (FN) and false positive (FP) predictions on DFDC dataset.

2012.09311

failure_cases.jpg

0.383989

c07a2270e49f4ec296a608a27a99f39b

Illustration of the workflow of I2G. For each source and target frame pair, a morphed mask is generated by taking a convex hull of the landmarks followed by elastic deformation and Gaussian blur. The masked region of the target image is replaced by that of the source with blending techniques <cit.>.

2012.09311

i2g.jpg

0.444865

2700fa846bda474a8df543b11ceb1b66

Visualization of PCL architecture. The consistency branch focuses on measuring the consistency of image patches according to their source features, and outputs a visualization map that indicates the manipulated region. A classification branch is applied after the learned feature map and predicts the binary score for Deepfake detection.

2012.09311

network.jpg

0.379878

f36ae633a932408db1859ed07cdc907a

Visualization of the consistency heatmaps, which localize the manipulated regions. The predictions are obtained in the cross-dataset experiments, where our model is trained with only real videos of FF++, augmented by I2G. The ground truth masks are generated by DSSIM, as discussed in Section <ref>.

2012.09311

qualitative_results.jpg

0.465907

12c58a836c4a42cdb0158f6ea1f45e2d

Visualization of the consistency maps, which localize the manipulated regions. The predictions are obtained in the cross-dataset experiments, where our model is trained with only real videos of FF++, augmented by I2G. The ground truth masks are generated by DSSIM, as discussed in Section 4.2.

2012.09311

qualitative_results_2.jpg

0.399549

767b603c30fc420d818e5872c1b41187

An overview of our patch-wise consistency learning. An “self-inconsistent” image is generated by stitching the target and source faces, where the mask indicates the region of modification and separates the patches into two groups. For each patch, we compute its source features consistencies with the rest, and their corresponding consistency scores are shown on the right.

2012.09311

teaser.jpg

0.469445

8585168655bd457aa18b82c05d786eee

Schematic of the GoldEnvSim components.

2012.09312

Fig1.jpg

0.482261

e9d8c0826f5d47d4994f3cc5434ff24f

Visualization of the surface pressure on simulation domain at the time 10/03/2019 11:00:00 GMT. The unit is in hPa.

2012.09312

Fig11.jpg

0.40786

d2b4a3fa24b441eca054fcd621b71464

Visualization of the accumulated rainfall over simulation period. The result is sum of convective precipitation and non-convective precipitation.

2012.09312

Fig13.jpg

0.399482

ca6b8047709048faa56a5370840cad28

Interface of the GoldEnvSim software.

2012.09312

Fig2.jpg

0.371538

a77be345c490494fb5675c5a0e6210e1

The IDV was integrated into the GoldEnvSim for WRF output dataset readout and visualization.

2012.09312

Fig4.jpg

0.406305

2e1d80ca4db741d2aef737c85d9b0179

Visualization of the temperature field at 2m-height simulated at 11/03/2019 09:00:00 GMT. The unit is in ^oC.

2012.09312

Fig5.jpg

0.471203

7453c1d908e34557ae869733c0df72d6

Hybrid visualization of the 2m-temperature field over typography simulated at 11/03/2019 09:00:00 GMT. This facilitates users a fusionable view on multiple variables.

2012.09312

Fig6.jpg

0.369947

449d0484a29341a089c10bcada5eff41

Representation of the wind bards with the speed (in m/s) is colored at the time 10/03/2019 06:00:00 GMT.

2012.09312

Fig8.jpg

0.446553

3c55b1c41b4e4766aa81530f24264059

The combined decoder and regressor architecture. The decoder maps C space input to the 16x16 intermediary image space I while the regressor maps from I to a single output ∈ℝ.

2012.09313

architecture.jpg

0.469523

71a0e313c9a64a8cb909eceff2a1d7c7

Example of configuration space for simple distance to line estimation. Here d is the distance in meters to the line and θ is the yaw angle.

2012.09313

cspace.jpg

0.39697

c9ba14ee83b4481cb72a8992e854d1bd

Original camera image from the AirSim simulation environment. Crop outlined in red, post processed training image inset and outlined in green. View is from the point in configuration space (d=1.5481, θ=0.0136).

2012.09313

front_center_gray_rect_crop_bw.jpg

0.372351

1b9629fc77f74dffbd36c5c4962187c3

A set of six training images (top), their reconstructions (middle), and reconstructions conditioned on z=0 (bottom) from six different points in configuration space.

2012.09313

reconstructions_comparison.jpg

0.381419

b10cc4f2c59746899423ba1fce80ec9d

Points in C while varying z. The perspective defined by c remains fixed while z varies the line break perturbation.

2012.09313

z_comparison.jpg

0.471656

fdfe668d8b774b6bbee438c9726f4d78

2D materials MnCl3 by element substitution with formation energy -1.45 eV

2012.09314

Fig10_MnCl3.jpg

0.400754

19e487fd6d994127b257a86c4f6e1c33

2D material V2S3 by element substitution with formation energy -0.907 eV

2012.09314

Fig10_V2S3.jpg

0.422555

84fd2e133cbc40d0b56f0f5b90e23686

2D materials WF3 by element substitution with formation energy -2.255 eV

2012.09314

Fig10_WF3.jpg

0.458394

25d56b4b912349ffa6aad7154744abe0

2D material YI2 by element substitution with formation energy -1.26 eV

2012.09314

Fig10_YI2.jpg

0.419854

1681e48695a947e294ac849cb38b365a

Generated materials distribution information.

2012.09314

ICSD.jpg

0.505764

539fd7205a7446839e3bbf498c8970fb

The framework of template based structure prediction. The main parts are: predicting new formula's space group; finding candidate template formulas according to the same element number and the same space group; calculating and sorting ElMD between the new formula and template formulas; selecting top n templates to do element substitution to get the new formula's structure.

2012.09314

TAMP.jpg

0.416629

0029975a912343feace85b128d853558

False-positive curve of the 2D materials RF classifier.

2012.09314

false_pos.jpg

0.439291

a16f9bf42daf40ffa5111deceb5b7259

Formation energy distribution of predicted 2D materials.

2012.09314

fe_hist.jpg

0.476996

2b4b918f33d840a98d3d0c5ccceea051

Framework for generation and prediction of 2D materials. It comprises four components. The green part: a GAN based composition generation module for breeding chemically valid materials. The blue part: a composition based random forest 2D materials classifier. The orange part: a template based element substitution structure predictor, as well as yellow part: DFT validation.

2012.09314

framework.jpg

0.414895

7eedf4148c1d41a4ae31ddbe89418093

Element frequency distribution. The abscissa represents 112 chemical elements arranged according to the atomic number from hydrogen (H) to Copernicium (Cn), vertical axis indicates the frequency of each element. The red curve denotes the distribution information of 2dMaterials dataset. The green curve shows our generated candidate materials. The blue line represents our predicted 2D materials.

2012.09314

frequency.jpg

0.471841

c45ab0eace244d4ea63abd541deaa7d7

Architecture of MatGAN. Generator (G) learns from known material compositions to generate realistic samples while discriminator (D) learns to determine whether a sample is a real one or generated one. Alternative training of D and G will improve the performance of both G and D.

2012.09314

gan.jpg

0.434182

b19f3adf13b044c9bdde793b213372b7

20,000 2D materials in the V2DB

2012.09314

npj_pred2K+2d.jpg

0.415628

6ba961f6fa404d089a52b0f091b4d1f9

6,000 2D materials in MP and V2DB vs. known ones

2012.09314

our+npj+2d_6000.jpg

0.435763

80c58bd517374ff1b25a3ab515c78e77

New 2D materials (Probability > 0.95) vs. known ones

2012.09314

our1485_pred+2d.jpg

0.398358

e224865768354d38930b3a3379421ae4

20,000 2D materials in the whole MP composition space

2012.09314

our20k_pred+2d.jpg

0.410501

3c33a5750e364bfa9cda4108d75e3f4f

One hot representation of material composition PuP2H6CO8. Brown color indicates the atom number of corresponding element in the specific material.

2012.09314

representation.jpg

0.564577

c4d8bb8cdc914a218e84f14bcde26a28

ROC curve of the 2D materials RF classifier

2012.09314

roc_cv.jpg

0.562145

8ea004431e2d43f98a44764b0472a974

𝔼[X^α_t] for different values of α.

2012.09317

example_esp.jpg

0.434194

16c52671f66d45b49cd208d5e6130715

P^α_1,0(t) for different values of α.

2012.09317

example_p10.jpg

0.480043

f1b9f4c5411d4bca84fadf6a33603df5

P^α_1,1(t) for different values of α.

2012.09317

example_p11.jpg

0.49437

e44fbd1ab7164b9881375ae21a4fa041

Var(X^α_t) for different values of α.

2012.09317

example_var.jpg