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
0.485728	55d198b681824ceb99684bca31a3fe23	Left: A Legendrian perturbation of the zero section. The vertical chords denote the two Reeb chords between the zero-section j^10 and the perturbation. Right: The perturbed version of the zero-section after a suitable Legendrian RI-move.	2012.15559	d1.jpg
0.495009	cf5144cc239242bd80a3c8f31e77430b	Λ_t is obtained from Λ_0 by a linear rescaling of the front inside { z ≥ 0} in the universal cover J^1^2 followed by the canonical projection J^1→ J^1S^1. The front of Λ_t undergoes the composite move shown in Figure <ref> consisting of two consecutive RII-moves along with RIII-moves.	2012.15559	d4.jpg
0.489198	65bb6e9357cf4e44bf4f165da49d5bf0	A composite move: the front to the right is obtained by performing two consecutive RII-moves on the front to the left together with an isotopy.	2012.15559	r.jpg
0.467189	75369badc0ff423285611249157e8417	RI: The first Legendrian Reidemeister move in the front projection.	2012.15559	r1.jpg
0.433811	afc166d2809c4226a71cd60d5ce6b8a8	RII: The second Legendrian Reidemeister move in the front projection.	2012.15559	r2.jpg
0.433417	48145ea2aa5a45f39df4b8961b1b6fa8	RIII: The third Legendrian Reidemeister move in the front projection.	2012.15559	r3.jpg
0.516655	ca443961ed0f4005b63c6b47f6d0245b	The left depicts a domain in ^2 with piecewise smooth boundary. After identifying the two horizontal pieces of the boundary, as well as the two vertical pieces, one obtains the Liouville domain shown on the right, with Liouville form described in Section <ref>. The closed exact Lagrangian L is the image of {p=0} and L' is a small Hamiltonian perturbation of L.	2012.15559	t1.jpg
0.569771	0c3a053921f4466db49dba473ec75b66	A Hamiltonian isotopy that wraps the Lagrangian {p=0} around the one-handle with core {q=0}, while fixing a neighbourhood of the latter core. Note that the Hamiltonian function is positive but constant near q=0. Here t'>t.	2012.15559	t2.jpg
0.496787	f2c50805bcfb4f05bc64265f66198370	The standard Legendrian unknot Λ_0 and the Legendrian fibre F. Note that there are precisely two transverse Reeb chords c_0,d_0 between F and Λ_0.	2012.15559	u1.jpg
0.522089	a85a9d5371b6405c96d6325bccba5975	Above: Λ̃_2 has a front which is a linear rescaling of the front of Λ_0 inside J^1. Below: Λ_2 is the projection of Λ̃_2 inside J^1S^1. Except for the mixed chords c_t and d_t that exist for the lift, there are now additional mixed chords.	2012.15559	u2.jpg
0.430169	c89edf6011cb4508ac975a133c274e21	This shows the projection Λ_t of the rescaling Λ̃_t under the universal cover J^1→ J^1S^1.	2012.15559	u3.jpg
0.514158	381e5a24d9d942719e52cd8d5580f979	In this case, the singularity is hidden behind the event horizon and there also occurs cosmological horizon.	2012.15560	Figure_A_New.jpg
0.465958	bcf37645eb5241fcbd2220fb0309c342	There occurs neither singularity nor horizon.	2012.15560	Figure_E_New.jpg
0.46971	8548d44f7cf14cfea8e8a9133d9d4194	A black hole with only event horizon and free of cosmological horizon as well as singularity.	2012.15560	Figure_F_New.jpg
0.444133	ab555add63fd4f75b787a643b0a0cb2a	The source arrangement for three (non-accelerating) black holes.	2012.15561	3blackholes.jpg
0.439905	2182c245f82d4d6d839b7c61186493b5	The source arrangement for the multi-black hole system of section <ref>. In the non-accelerating case, the point z_n, representing the start of the SILM (thick red arrow), and the SILM itself are absent; its neighbouring string (dashed black) instead extends to z→∞.	2012.15561	coordsAcc_v2.jpg
0.387801	97fc1476f2a14f999fff4b750c69ded9	Example tokens of unseen scripts.	2012.15562	example_scripts.jpg
0.469681	d009047d56e84e2dbc7dbe54d57f663e	Number of tokens of each script grouped into the 10 KMeans clusters. We follow <cit.> in splitting the utf-8 tokens into scripts.	2012.15562	kmeans_clusters.jpg
0.510945	f29fe8b56e084e1d8bf25221e87f988c	Sample efficiency. For "MF" we leverage the setting.	2012.15562	sample_efficiency.jpg
0.400779	d85d958482424399a81e94f1b6fbcf6b	mBERT's zero-shot cross-lingual transfer performance with respect to the proportion of s in mBERT's original vocabulary relative to the target language (see Table <ref>).	2012.15562	scatter_mBERT_UNKs.jpg
0.433678	9132572c11af4a559a4d8db2e871b689	Visual comparison of COVID-19 infection segmentation by different methods on 2D COVID-SemiSeg dataset. As can be observed, our method can generate segmentation results with more accurate boundaries and less segmentation mistakes in small infection areas, which is closer to the ground truth.	2012.15564	2dresults.jpg
0.402008	9f887661c51b487b983edcccd4fe9099	The overview of our proposed relation-driven collaborative learning model, where green and blue represent the data flow of general encoder and target encoder for COVID-19 infection segmentation, respectively. Extracted features from these two parallel encoders are concatenated for the input of the shared decoder. To exploit shared knowledge from non-COVID-cases, an additional data flow in orange is adopted. By regularizing feature-level relation consistency of given input, the model is encouraged to explore semantic information from both COVID-19 and non-COVID cases. Since the general encoder is applied to utilize non-COVID data and assist in the learning of COVID-19 segmentation as an auxiliary branch, we only employ the skip connections between the target encoder and shared decoder for the fusion of multi-scale features as shown in the top right corner.	2012.15564	architecture.jpg
0.454226	13e96807353645f492fb0abb709847a9	Examples of COVID-19 infections in CT volumes showing the large variations of shape, size and position of lung infections. The upper row shows the raw images (from axial view) and the lower row shows corresponding annotations of infection areas.	2012.15564	covid.jpg
0.44592	b9671d4b75174f04a92c6a6312a01625	Example of challenging cases for COVID-19 lung infection segmentation with limited labeled data.	2012.15564	hardcase.jpg
0.389573	706bf271846340cbb88034dcf6e31199	Visual comparison of COVID-19 infection segmentation by different methods on 3D COVID-19 segmentation benchmark dataset from axial view.	2012.15564	results.jpg
0.415268	5c67a9e5fbf9497d90e4a721a692550c	Visualization of the general feature relation matrices of non-COVID cases (left column) and COVID-19 cases (middle column) and their absolute difference (right column) during the training procedure.	2012.15564	visual1.jpg
0.444966	1cce54dda3e64446b33254ee1ceb6e70	Visualization of the general feature relation matrix (left column) and target feature relation matrix (middle column) of COVID-19 cases and their absolute difference (right column) during the training procedure.	2012.15564	visual2.jpg
0.442316	532a3b58b739451c908c452676baf436	High level look at model architecture	2012.15565	Architecture.jpg
0.413301	4f490c0e1316434c9bca09b90da7b38a	Architecture Overview	2012.15565	Overview.jpg
0.499661	dab8d32bc104479e9e8aec719c2829e6	Caption: four men are playing frisbee near a fenced-off area	2012.15565	coco1.jpg
0.404766	c80a5bf1648d4f73a30465a44e70e2a1	Caption: a zebra grazing in a dirt and grass covered enclosure	2012.15565	coco2.jpg
0.435636	f6f95315358d41f0ab7ac8f9aab4d5e0	Home Page	2012.15565	entry_page.jpg
0.472528	d390f946390046c688e77118689bc185	Sample Image Histogram with Red, Green and Blue Channels	2012.15565	histogram.jpg
0.562238	5d8925ab0feb418db0db41ef0888cd9b	InceptionV3 network architecture	2012.15565	inceptionv3.jpg
0.454522	6508a48041004550b1ffce494f8822b6	Training loss	2012.15565	loss.jpg
0.479901	06c1fa1485d84e1b8804dec0c12c969d	The result of the spoken query	2012.15565	query.jpg
0.410141	78f2c2626cdc46f5b858c0981097995b	Final output from application	2012.15565	result.jpg
0.431301	6bbf8fb68d784dedb836a255f40e80ca	Caption: a man holding is tray a tall on a surfboard	2012.15565	sample_result1.jpg
0.500066	d63d6e41f7b043c8b2a21dca57e37735	Caption: a small elephant with a tower tennis a up of flowers huge bus tennis trees	2012.15565	sample_result2.jpg
0.489164	750d9cf7571c4c04a3aaa95b930b7965	Flow chart describing adaptive asymmetric DAgger (A2D), introduced in this work, which builds on DAgger <cit.> by further refining the expert conditioned on the trainee's policy.	2012.15566	ad_vs_a2d_super_simple.jpg
0.47368	2c3aae065c494cb2bed2adacb4ae1177	Divergence.	2012.15566	lam_sweep_divergence_TigerDoor_True_tests_results_AdaptAsymDagger_cr_logs_lam_sweep_4_.jpg
0.443805	6288bcf17c064ddba61f926d2b05cde6	Reward.	2012.15566	lam_sweep_results_TigerDoor_True_tests_results_AdaptAsymDagger_cr_logs_lam_sweep_4_.jpg
0.410381	93295ef9e81b4ba4a35b02f6932ed4db	Frozen Lake.	2012.15566	sec4_divergence_IceLake_True_cr_logs_LavaGap_LavaGapCompiledRun_.jpg
0.419041	e4ae4d98ff204b1698d00597cd5150b1	Tiger Door.	2012.15566	sec4_divergence_TigerDoor_True_cr_logs_TigerDoor_TigerDoorCompiledRun_.jpg
0.426621	69ad2b2d794c4e3da06768954334a591	Frozen Lake.	2012.15566	sec4_representation_IceLake_True_cr_logs_LavaGap_LavaGapCompiledRun_.jpg
0.459807	e97cc402e90f4442a189db8d92ea5b14	Tiger Door.	2012.15566	sec4_representation_TigerDoor_True_cr_logs_TigerDoor_TigerDoorCompiledRun_.jpg
0.454327	c0f4c74e3f404cdf967585e3e375cf77	Frozen Lake.	2012.15566	sec4_results_IceLake_True_cr_logs_LavaGap_LavaGapCompiledRun_.jpg
0.469332	f7f924f2b9c2452fbac1d52a55525543	Tiger Door.	2012.15566	sec4_results_TigerDoor_True_cr_logs_TigerDoor_TigerDoorCompiledRun_.jpg
0.403372	1708d7c938a543569e53ef0c0eb0631e	Residues with respect to experiment for the energies of 551 spherical and deformed even-even nuclei calculated with the WK Mic-Mac model based on the Gogny D1S force (filled red symbols) and the HFB results calculated with the Gogny D1M force (open blue symbols).	2012.15569	D1S_luis-new-1.jpg
0.416924	04661d39dd784af2baf5dd1295638a74	Residues with respect to the experimental energies of 551 spherical and deformed even-even nuclei predicted by the HFB calculation with the Gogny D1S force (magenta symbols) <cit.> and by the Mic-Mac Gogny-based models fitted to experimental energies (red symbols) and to HFB energies (cyan symbols).	2012.15569	D1S_luis.jpg
0.408578	9f5d9b192c3a4d97b9b920fb5747efb5	Residues with respect to experiment for the energies of 551 spherical and deformed even-even nuclei calculated with the WK Mic-Mac models based on the Gogny D1S force (red symbols) and on the phenomenological Wyss potential (blue symbols).	2012.15569	Dif_be.jpg
0.418666	044a0f4b014a44a9b8c8396a0b398fdb	Residues with respect to experiment for the energies of 551 spherical and deformed even-even nuclei obtained with the Mic-Mac models FRDM <cit.>, LSD <cit.> and our WK model using the Gogny-based and the phenomenological (Wyss) potentials as external mean fields.	2012.15569	Dif_be_all.jpg
0.444527	4002564ba547431095850e60a65802b8	Binned data and the corresponding Gaussian fits for the different Mic-Mac models considered in the present study (see text for more detail).	2012.15569	f-stats.jpg
0.447328	461adcf5d9c3478aafd5d6717bfe10a3	Gauss diagrams and Reidemeister moves	2012.15571	draw.jpg
0.463209	4597a782a1bb46aa88e19ee5230a7eed	The main example and the corresponding word in the group G	2012.15571	figurebelowV.jpg
0.460437	f06a4d71e0994cc4b1e5c4cbea191f3b	A long free knot K and the word bababab b'ab'ab'ab'	2012.15571	informV.jpg
0.486814	7d42422b9d6442b1ab064074b7cef2f4	Transformation of a word in three places	2012.15571	threeplaces.jpg
0.435709	69b5ed67788041349c60780b903137bb	The average, upper and lower bounds of the ratio of biased examples in Quoref_dev and our challenge set for the randomly sampled 10 subsets.	2012.15573	confident_interval.jpg
0.419451	bbb0c4d196784b1b85122987c4e979f6	A sample from the Quoref dataset.	2012.15573	coref-example.jpg
0.346388	90a4c63c8dd54a059a5a8a579ba89a6e	A QA example that relies on simple lexical overlap without requiring coreference reasoning.	2012.15573	lexical-example.jpg
0.427583	efc5b1f63ece4cc48f000fff1e54becb	Examples of gradient-weighted class activation maps of different models generated according to <cit.>. These three retinal images are mis-classified by DenseNet121-RGB <cit.> but correctly classified by our Single-branch SalStructIQA and Dual-branch SalStructIQA. Obviously, with salient structures, the models tend to learn discriminate information from salient regions for image quality assessment.	2012.15575	CAM.jpg
0.534435	3240e06ff00b47999d814aad9580ccd2	Confusion matrices of BRISQUE <cit.>, NBIQA <cit.>, TS-CNN<cit.>, DenseNet121-MCF <cit.>, our Single-branch SalStructIQA and Dual-branch SalStructIQA.	2012.15575	ConfusionMatrixSOTA-v03.jpg
0.473837	b9d16de241d046f89ffd102e8649ce3e	The proposed deep CNN architectures incorporating with our proposed large-size salient structures and tiny-size salient structures.	2012.15575	architecture.jpg
0.471748	9682172190484906b0e8d4339046ca64	The cumulative distributions of tiny-size salient structures of each class.	2012.15575	cum.jpg
0.51764	1df2f0ca40254a5c86573d498741062f	Examples for retinal images with different quality grades. From left to right, the quality grade is Good, Usable and Reject, respectively.	2012.15575	fig4intro1.jpg
0.403581	1248b4654bcf42a79e274c5200447d70	The flowchart of our proposed SalStructIQA.	2012.15575	fig4intro2.jpg
0.398647	746be58387c945f4889fd0282937cf8b	Examples for the probability map of large-size salient structures and the masks of detected large-size salient structures.	2012.15575	fig4method_LS.jpg
0.394937	8df055b38ffe4e1396b87e089d2b41b5	Examples for the standardized line response maps and masks of detected tiny-size salient structures.	2012.15575	fig4method_TS.jpg
0.520029	c0ae9348356540d98e5af595149fc567	Plot of the (R_1(θ_1),R_2(θ_1)), θ_1=0.2i, allowed region for the NLSM with N=6, k=1 and various analytic regions b_1=0, b_2=π/2, 0.9π, and b_1=-0.1π, b_2=1.1π. The vertex at the center of the red circle (see rotated and rescaled inset) is the R-matrix in eq.(<ref>) with N=6, k=1. A new vertex appears in the smaller regions, namely the one at the center of the purple circle that corresponds to a solution with R_1=R_2, or N=6, k=0.	2012.15576	NLSM_allowed_region_C_edited.jpg
0.569883	ad0142a6ffcf43f4bc98d3dd298b4cd2	Pictorial description of the 1→ 1 R-matrix describing the amplitude for a particle to bounce from the wall possibly changing its identity a→ b. A double Wick rotation relates this process to pair production from an initial boundary state \|B⟩. In the second case we can define a bulk S-matrix using the usual asymptotic states.	2012.15576	boundary_edited.jpg
0.521379	e3d5c4b0114f4e5681464150a8831872	Allowed values of A(θ_1),B(θ_1) for θ_1=0.9879 i. The solid curve is the analytic result and the points are the numerical data. All points of the curve correspond to the same integrable R-matrix that has a free parameter α that changes along the boundary, see eq.(<ref>). The vertex corresponds to the limit α→∞.	2012.15576	offdiag_bdy_exact_num.jpg
0.479948	2eb2caf7c3f746a8b36c839d4911c55a	Plot of the (R_1(θ_1),R_2(θ_1)) allowed region for the pYB model for θ_1=0.9879i. We consider the case N=6, and k=1,2,3. In this case, the whole boundary is integrable and we can compare the numerics (dots) with the analytical result (solid lines). The curves intersect at R_1(θ)=R_2(θ) common to all k in the ansatz (<ref>). The vertices (encircled) appear when a pole or zero enters that physical region and has to be cancelled by changing the CDD factor.	2012.15576	pYB_N6kall.jpg
0.486597	1c56524c667f49abbe5b048950643cd4	Physical region of the reflection process (ε≥ 0). The positive real axis corresponds to the physical reflection (ε∈ℝ_≥ m) which is the boundary value of an analytic function in the physical region (shaded). Crossing can be imposed on the upper line, this is the minimal region where crossing gives a constraint. Boundary bound states could appear as poles on the imaginary axis (dashed line) but we do not allow them here.	2012.15576	theta_region.jpg
0.453134	f3193f0e41cf481dbdb7cddbb02785ca	Extended region of analyticity. We considered two cases: b_1=0, π/2≤ b_2≤π and b_1=π-b_2, b_2>π. It might seem that in the first case analyticity is guaranteed by crossing but this is not so because of the non-trivial crossing equation (<ref>).	2012.15576	theta_region_2.jpg
0.447843	e598aa8218d54faf9c42d11d20390dd7	Extended region of analyticity. We generically extend the region below the real axis (across the physical cut) to a curve . A well defined dual problem requires a regularization, namely a bound on the analytic function on . An example of such curve is in fig.<ref> in which case the curve is θ=b_1. There we also extended the region to θ = b_2 which can be done in the same way.	2012.15576	theta_region_C.jpg
0.442932	a832b563955f420b82d50eedf607cdde	Numerical simulation of 2D corrugation pattern for (a1) 8 site Hubbard model with 4 filled particles, (b1) 9 site Hubbard model with 6 filled particles, and (c1) 13 site Hubbard model with 2 filled particles. The different sizes of solid circles and different lengths of red edges denote the disorder of the interaction strength U_j and hopping strength t_j of Eq. (<ref>). (a2)-(c2) Plots of ℱ and C_1 as function of disorder strength b. The system parameters are (a) t=1, a=0, U=-0.8t (b) t=1, a=0.2t, U=-0.8t and (c) t=1, a=0.2t, U=-0.5t. Similar with 1D system, the induced fluctuation of C_1 is minor in comparison with disorder free case (b=0 ) indicating that the existing result of 1D can be extended to 2D or higher dimensional system.	2012.15577	fig_2D.jpg
0.575068	32be05c4ced0449daa8e104fb9f02f66	Transition of the ground state driven by the imaginary hopping. (a)-(b) depict the variation of the energy spectrum with respect to hopping strength t, and (c) presents the switch of the correlator C_3. The numerical simulation is performed for the 6 site Hubbard model with 4 filled particles. The other system parameters are t_0=1, and U=-1.5t_0. The hopping strength t is in units of t_0. The red line denotes the η- pairing state. C_3 experiences an evident jump at the critical point indicating a dramatic change of the ground state. Such a first-order transition may associate with quantum phase transition to some extent.	2012.15577	fig_gene.jpg
0.488643	03a167f7c4d741ce8773757b0833a53b	Plots of the overlap ℱ and correlator C_j as a function of the strength of interaction disorder b for (a) t=1, a=0.1t , U=-0.5t (b) t=1, a=0.3t, U=-1.5t (c) t=1, a=0, U=-4t, and (d) t=1, a=0.2t, U=-4t. The numerical simulation is performed for the 6 site 1D Hubbard model at half filling and s^z=0. Here the strength of the hopping disorder a is set to be constant for each subfigure and the correlator C_j is averaged over all sites separated by a distance j. When b=0, no matter what value a takes, as long as U is non-zero, one can always get a perfect η-pairing ground state. The variation of C_j indicates that the increase of b will not result in the significant change of the η-pairing ground state; the presence of the hopping disorder can suppress the fluctuation of C_j compared to the disorder free case, which can be seen from (c)-(d). Therefore, the value of the correlator is the consequence of the interplay between two such disorders, which provides a scheme to prepare η-pairing ground state in the experiment.	2012.15577	fig_ran_r.jpg
0.516132	0a5cd15d300c4d62b52cbf1091469aa0	Comparison of the two-particle spectrum within the subspace (0,0) between the non-Hermitian setting and its parent Hermitian system for (a) U=-0.8t, (b) U=-2t, and (c) U=-4.5t, respectively. The upper and lower panels present the spectrum of the Hermitian and non-Hermitian system, respectively. The red circle and gray shading denote the bound pair and scattering state. The parent Hermitian system can be obtained by assuming it→ t. For the Hermitian system, the bound pair with the lowest energy lies in the K=0 subspace while the ground state of two-particle non-Hermitian setting locates on the subspace indexed by K=π. It is shown that the presence of the imaginary hopping not only makes all scattering energy bands imaginary but also reverses the whole bound band. In the condition of small U, there can exist an EP characterized by the divergence of ∂ϵ _K/∂ K. Such non-Hermiticity alters significantly the paring mechanism and hence favors superconductivity.	2012.15577	fig_spectrum.jpg
0.499045	ee8e8debfeaa47ef8aa238bea637c7d7	Eigenvalues of _LE, weighted for the leading edge isentropic Mach number.	2012.15579	LE_eigs.jpg
0.523345	65c48aa8cdbd427782c63dcec01a0ab5	An isometric view of the blade envelope in Figure <ref>.	2012.15579	cad_1_small.jpg
0.495068	56d827cb43fa4105baeb908d344c21ab	Close-up of the leading edge of the blade envelope in Figure <ref>. The demarcated iso-contour lines denote displacements at 5, 10 and 15 mm away from the nominal blade.	2012.15579	cad_2_small.jpg
0.429132	2bb06509c88540f894204f507562fd81	A CAD representation of the controlled loss and mass flow rate blade envelope.	2012.15579	cad_3_small.jpg
0.456392	d9be4f00cc5144ffa8453b1254fab036	The blade envelope from Figure <ref> can be suitably scaled for 3D printing and placed on desktops.	2012.15579	cad_4.jpg
0.425935	b510d0da02f84245a4ed7e951a4958d5	Conditional tuning of the leading edge isentropic Mach number.	2012.15579	inv_design_LE.jpg
0.420202	751b1238ba7e4bf8854571a78322cfde	Conditional tuning of the peak isentropic Mach number.	2012.15579	inv_design_peak2.jpg
0.40119	638de54d1928415da077e76a65b7101a	Loss and mass flow function for random training designs (blue) and invariant designs within the intersection of the inactive subspaces (yellow).	2012.15579	invariant_loss_and_mdot.jpg
0.408503	7570e59fcdec4d8895fb66db979c89b5	The blade envelope for loss and mass flow.	2012.15579	loss_mass_env.jpg
0.56072	4ea0f0aab0544494afe54f77dbe323a0	Eigenvalues of _peak, weighted for the peak isentropic Mach number.	2012.15579	peak_eigs.jpg
0.376286	d1d7122cefad4b659125cb315341fc20	Validation of polynomial model for the mass flow function. This coincides with the sufficient summary plot because the global polynomial approximation is a linear model.	2012.15579	validation_lm.jpg
0.381439	15fe1c79cfb341cba40724096ee538c6	An illustration of reinforcement learning, or multi-armed bandit problems.	2012.15584	bandit-crop.jpg
0.458755	7982e57452e24c399f7a0a24f6a6cd3b	The roadmap of this thesis.	2012.15584	chapters-crop.jpg
0.51847	a4b911d639c9438099461ecd4586e489	The illustration of a confidence ellipsoid, where X=(x_1, ⋯, x_d) is the model matrix. The shape of the ellipsoid depends on the information matrix X^TX.	2012.15584	ellipsoid-crop.jpg
0.45439	107729ae2fa84b1599fadb0d624cffd2	The process of dense subgraph discovery.	2012.15584	graphmining-crop.jpg
0.47967	bfac3a2045ca4dd6936e57c14056e99e	Phase portrait of system (<ref>) with parameters β=0.5, δ=0.2, p=2 and c=5, for different initial conditions not necessarily representing realistic cases. Empty circles represent the initial state, while solid circles represent final states. The red hyperplane corresponds to U(t) ≡ (i.e., the critical value of U, when (t)=1) while the blue hyperplane corresponds to the fast manifold in which I(t) and V(t) are proportional (.i.e, I(t)=c/p V(t)). Note that only the initial states with U_0>=1 corresponds to scenarios with >1.	2012.15585	FhasePortN.jpg
0.416652	6856cca1a6294d4895d052c331f4292a	Evolution of (t) when an antiviral treatment is initiated at a time of approximately 0.75 t̂_V, and different antiviral inhibition effects η_p are considered, smaller and greater than the critical value η_p^c. The black dashed line represents =1. As it can be seen, for values of η_p < η_p^c, (t) crosses 1 at larger times for larger values of η_p as it is stated in Lemma <ref>. This implies that if η_p < η_p^c, larger values of η_p delays the virus peak time, as it is stated in Theorem <ref>.iii. Furthermore, the figure confirms that, for real patient date, t^e is close to t̂_V.	2012.15585	Rt.jpg
0.461044	b5180879ca9041f1af0cf7929040fe89	Uncertainty analysis on viral load evolution for p-parameter 95% confidence interval. Note the uncertainty in viral load peak. Empty dots are data of COVID-19 patients.	2012.15585	Sensivity_Analysis_Case_p.jpg
0.441802	c1de0eaf61774a5aba3793a19e9988d7	Viral load time evolution with treatment initial time given by t_tr=0.7t^e. Values of η_p smaller, approximately equal and greater than η_p^c are simulated to demonstrate the results in Theorem <ref>.The black line denotes the untreated case (η_p=0).	2012.15585	allpatCL_07_te.jpg
0.427573	cd102936b7fd446f94cd72de69a8456d	Free virus behavior when treatment is started at viral load detection level (t_tr=t_DL). Values of η_p smaller (η_p^1,η_p^2), approximately equal (η_p^3,η_p^4) and greater (η_p^5,η_p^6) than η_p^c are simulated to demonstrate the results in Theorem <ref>. The black line denotes the untreated case (η_p=0).	2012.15585	allpatCL_DL.jpg
0.429959	925f2b7b84a140ada54e35d7862793df	Viral load time evolution with treatment initial time given by t_tr=t^e. Values of η_p smaller, approximately equal and greater than η_p^c are simulated to demonstrate the results in Theorem <ref>. The black line denotes the untreated case (η_p=0).	2012.15585	allpatCL_te.jpg

0.485728

55d198b681824ceb99684bca31a3fe23

Left: A Legendrian perturbation of the zero section. The vertical chords denote the two Reeb chords between the zero-section j^10 and the perturbation. Right: The perturbed version of the zero-section after a suitable Legendrian RI-move.

2012.15559

d1.jpg

0.495009

cf5144cc239242bd80a3c8f31e77430b

Λ_t is obtained from Λ_0 by a linear rescaling of the front inside { z ≥ 0} in the universal cover J^1^2 followed by the canonical projection J^1→ J^1S^1. The front of Λ_t undergoes the composite move shown in Figure <ref> consisting of two consecutive RII-moves along with RIII-moves.

2012.15559

d4.jpg

0.489198

65bb6e9357cf4e44bf4f165da49d5bf0

A composite move: the front to the right is obtained by performing two consecutive RII-moves on the front to the left together with an isotopy.

2012.15559

r.jpg

0.467189

75369badc0ff423285611249157e8417

RI: The first Legendrian Reidemeister move in the front projection.

2012.15559

r1.jpg

0.433811

afc166d2809c4226a71cd60d5ce6b8a8

RII: The second Legendrian Reidemeister move in the front projection.

2012.15559

r2.jpg

0.433417

48145ea2aa5a45f39df4b8961b1b6fa8

RIII: The third Legendrian Reidemeister move in the front projection.

2012.15559

r3.jpg

0.516655

ca443961ed0f4005b63c6b47f6d0245b

The left depicts a domain in ^2 with piecewise smooth boundary. After identifying the two horizontal pieces of the boundary, as well as the two vertical pieces, one obtains the Liouville domain shown on the right, with Liouville form described in Section <ref>. The closed exact Lagrangian L is the image of {p=0} and L' is a small Hamiltonian perturbation of L.

2012.15559

t1.jpg

0.569771

0c3a053921f4466db49dba473ec75b66

A Hamiltonian isotopy that wraps the Lagrangian {p=0} around the one-handle with core {q=0}, while fixing a neighbourhood of the latter core. Note that the Hamiltonian function is positive but constant near q=0. Here t'>t.

2012.15559

t2.jpg

0.496787

f2c50805bcfb4f05bc64265f66198370

The standard Legendrian unknot Λ_0 and the Legendrian fibre F. Note that there are precisely two transverse Reeb chords c_0,d_0 between F and Λ_0.

2012.15559

u1.jpg

0.522089

a85a9d5371b6405c96d6325bccba5975

Above: Λ̃_2 has a front which is a linear rescaling of the front of Λ_0 inside J^1. Below: Λ_2 is the projection of Λ̃_2 inside J^1S^1. Except for the mixed chords c_t and d_t that exist for the lift, there are now additional mixed chords.

2012.15559

u2.jpg

0.430169

c89edf6011cb4508ac975a133c274e21

This shows the projection Λ_t of the rescaling Λ̃_t under the universal cover J^1→ J^1S^1.

2012.15559

u3.jpg

0.514158

381e5a24d9d942719e52cd8d5580f979

In this case, the singularity is hidden behind the event horizon and there also occurs cosmological horizon.

2012.15560

Figure_A_New.jpg

0.465958

bcf37645eb5241fcbd2220fb0309c342

There occurs neither singularity nor horizon.

2012.15560

Figure_E_New.jpg

0.46971

8548d44f7cf14cfea8e8a9133d9d4194

A black hole with only event horizon and free of cosmological horizon as well as singularity.

2012.15560

Figure_F_New.jpg

0.444133

ab555add63fd4f75b787a643b0a0cb2a

The source arrangement for three (non-accelerating) black holes.

2012.15561

3blackholes.jpg

0.439905

2182c245f82d4d6d839b7c61186493b5

The source arrangement for the multi-black hole system of section <ref>. In the non-accelerating case, the point z_n, representing the start of the SILM (thick red arrow), and the SILM itself are absent; its neighbouring string (dashed black) instead extends to z→∞.

2012.15561

coordsAcc_v2.jpg

0.387801

97fc1476f2a14f999fff4b750c69ded9

Example tokens of unseen scripts.

2012.15562

example_scripts.jpg

0.469681

d009047d56e84e2dbc7dbe54d57f663e

Number of tokens of each script grouped into the 10 KMeans clusters. We follow <cit.> in splitting the utf-8 tokens into scripts.

2012.15562

kmeans_clusters.jpg

0.510945

f29fe8b56e084e1d8bf25221e87f988c

Sample efficiency. For "MF" we leverage the setting.

2012.15562

sample_efficiency.jpg

0.400779

d85d958482424399a81e94f1b6fbcf6b

mBERT's zero-shot cross-lingual transfer performance with respect to the proportion of s in mBERT's original vocabulary relative to the target language (see Table <ref>).

2012.15562

scatter_mBERT_UNKs.jpg

0.433678

9132572c11af4a559a4d8db2e871b689

Visual comparison of COVID-19 infection segmentation by different methods on 2D COVID-SemiSeg dataset. As can be observed, our method can generate segmentation results with more accurate boundaries and less segmentation mistakes in small infection areas, which is closer to the ground truth.

2012.15564

2dresults.jpg

0.402008

9f887661c51b487b983edcccd4fe9099

The overview of our proposed relation-driven collaborative learning model, where green and blue represent the data flow of general encoder and target encoder for COVID-19 infection segmentation, respectively. Extracted features from these two parallel encoders are concatenated for the input of the shared decoder. To exploit shared knowledge from non-COVID-cases, an additional data flow in orange is adopted. By regularizing feature-level relation consistency of given input, the model is encouraged to explore semantic information from both COVID-19 and non-COVID cases. Since the general encoder is applied to utilize non-COVID data and assist in the learning of COVID-19 segmentation as an auxiliary branch, we only employ the skip connections between the target encoder and shared decoder for the fusion of multi-scale features as shown in the top right corner.

2012.15564

architecture.jpg

0.454226

13e96807353645f492fb0abb709847a9

Examples of COVID-19 infections in CT volumes showing the large variations of shape, size and position of lung infections. The upper row shows the raw images (from axial view) and the lower row shows corresponding annotations of infection areas.

2012.15564

covid.jpg

0.44592

b9671d4b75174f04a92c6a6312a01625

Example of challenging cases for COVID-19 lung infection segmentation with limited labeled data.

2012.15564

hardcase.jpg

0.389573

706bf271846340cbb88034dcf6e31199

Visual comparison of COVID-19 infection segmentation by different methods on 3D COVID-19 segmentation benchmark dataset from axial view.

2012.15564

results.jpg

0.415268

5c67a9e5fbf9497d90e4a721a692550c

Visualization of the general feature relation matrices of non-COVID cases (left column) and COVID-19 cases (middle column) and their absolute difference (right column) during the training procedure.

2012.15564

visual1.jpg

0.444966

1cce54dda3e64446b33254ee1ceb6e70

Visualization of the general feature relation matrix (left column) and target feature relation matrix (middle column) of COVID-19 cases and their absolute difference (right column) during the training procedure.

2012.15564

visual2.jpg

0.442316

532a3b58b739451c908c452676baf436

High level look at model architecture

2012.15565

Architecture.jpg

0.413301

4f490c0e1316434c9bca09b90da7b38a

Architecture Overview

2012.15565

Overview.jpg

0.499661

dab8d32bc104479e9e8aec719c2829e6

Caption: four men are playing frisbee near a fenced-off area

2012.15565

coco1.jpg

0.404766

c80a5bf1648d4f73a30465a44e70e2a1

Caption: a zebra grazing in a dirt and grass covered enclosure

2012.15565

coco2.jpg

0.435636

f6f95315358d41f0ab7ac8f9aab4d5e0

Home Page

2012.15565

entry_page.jpg

0.472528

d390f946390046c688e77118689bc185

Sample Image Histogram with Red, Green and Blue Channels

2012.15565

histogram.jpg

0.562238

5d8925ab0feb418db0db41ef0888cd9b

InceptionV3 network architecture

2012.15565

inceptionv3.jpg

0.454522

6508a48041004550b1ffce494f8822b6

Training loss

2012.15565

loss.jpg

0.479901

06c1fa1485d84e1b8804dec0c12c969d

The result of the spoken query

2012.15565

query.jpg

0.410141

78f2c2626cdc46f5b858c0981097995b

Final output from application

2012.15565

result.jpg

0.431301

6bbf8fb68d784dedb836a255f40e80ca

Caption: a man holding is tray a tall on a surfboard

2012.15565

sample_result1.jpg

0.500066

d63d6e41f7b043c8b2a21dca57e37735

Caption: a small elephant with a tower tennis a up of flowers huge bus tennis trees

2012.15565

sample_result2.jpg

0.489164

750d9cf7571c4c04a3aaa95b930b7965

Flow chart describing adaptive asymmetric DAgger (A2D), introduced in this work, which builds on DAgger <cit.> by further refining the expert conditioned on the trainee's policy.

2012.15566

ad_vs_a2d_super_simple.jpg

0.47368

2c3aae065c494cb2bed2adacb4ae1177

Divergence.

2012.15566

lam_sweep_divergence_TigerDoor_True_tests_results_AdaptAsymDagger_cr_logs_lam_sweep_4_.jpg

0.443805

6288bcf17c064ddba61f926d2b05cde6

Reward.

2012.15566

lam_sweep_results_TigerDoor_True_tests_results_AdaptAsymDagger_cr_logs_lam_sweep_4_.jpg

0.410381

93295ef9e81b4ba4a35b02f6932ed4db

Frozen Lake.

2012.15566

sec4_divergence_IceLake_True_cr_logs_LavaGap_LavaGapCompiledRun_.jpg

0.419041

e4ae4d98ff204b1698d00597cd5150b1

Tiger Door.

2012.15566

sec4_divergence_TigerDoor_True_cr_logs_TigerDoor_TigerDoorCompiledRun_.jpg

0.426621

69ad2b2d794c4e3da06768954334a591

Frozen Lake.

2012.15566

sec4_representation_IceLake_True_cr_logs_LavaGap_LavaGapCompiledRun_.jpg

0.459807

e97cc402e90f4442a189db8d92ea5b14

Tiger Door.

2012.15566

sec4_representation_TigerDoor_True_cr_logs_TigerDoor_TigerDoorCompiledRun_.jpg

0.454327

c0f4c74e3f404cdf967585e3e375cf77

Frozen Lake.

2012.15566

sec4_results_IceLake_True_cr_logs_LavaGap_LavaGapCompiledRun_.jpg

0.469332

f7f924f2b9c2452fbac1d52a55525543

Tiger Door.

2012.15566

sec4_results_TigerDoor_True_cr_logs_TigerDoor_TigerDoorCompiledRun_.jpg

0.403372

1708d7c938a543569e53ef0c0eb0631e

Residues with respect to experiment for the energies of 551 spherical and deformed even-even nuclei calculated with the WK Mic-Mac model based on the Gogny D1S force (filled red symbols) and the HFB results calculated with the Gogny D1M force (open blue symbols).

2012.15569

D1S_luis-new-1.jpg

0.416924

04661d39dd784af2baf5dd1295638a74

Residues with respect to the experimental energies of 551 spherical and deformed even-even nuclei predicted by the HFB calculation with the Gogny D1S force (magenta symbols) <cit.> and by the Mic-Mac Gogny-based models fitted to experimental energies (red symbols) and to HFB energies (cyan symbols).

2012.15569

D1S_luis.jpg

0.408578

9f5d9b192c3a4d97b9b920fb5747efb5

Residues with respect to experiment for the energies of 551 spherical and deformed even-even nuclei calculated with the WK Mic-Mac models based on the Gogny D1S force (red symbols) and on the phenomenological Wyss potential (blue symbols).

2012.15569

Dif_be.jpg

0.418666

044a0f4b014a44a9b8c8396a0b398fdb

Residues with respect to experiment for the energies of 551 spherical and deformed even-even nuclei obtained with the Mic-Mac models FRDM <cit.>, LSD <cit.> and our WK model using the Gogny-based and the phenomenological (Wyss) potentials as external mean fields.

2012.15569

Dif_be_all.jpg

0.444527

4002564ba547431095850e60a65802b8

Binned data and the corresponding Gaussian fits for the different Mic-Mac models considered in the present study (see text for more detail).

2012.15569

f-stats.jpg

0.447328

461adcf5d9c3478aafd5d6717bfe10a3

Gauss diagrams and Reidemeister moves

2012.15571

draw.jpg

0.463209

4597a782a1bb46aa88e19ee5230a7eed

The main example and the corresponding word in the group G

2012.15571

figurebelowV.jpg

0.460437

f06a4d71e0994cc4b1e5c4cbea191f3b

A long free knot K and the word bababab b'ab'ab'ab'

2012.15571

informV.jpg

0.486814

7d42422b9d6442b1ab064074b7cef2f4

Transformation of a word in three places

2012.15571

threeplaces.jpg

0.435709

69b5ed67788041349c60780b903137bb

The average, upper and lower bounds of the ratio of biased examples in Quoref_dev and our challenge set for the randomly sampled 10 subsets.

2012.15573

confident_interval.jpg

0.419451

bbb0c4d196784b1b85122987c4e979f6

A sample from the Quoref dataset.

2012.15573

coref-example.jpg

0.346388

90a4c63c8dd54a059a5a8a579ba89a6e

A QA example that relies on simple lexical overlap without requiring coreference reasoning.

2012.15573

lexical-example.jpg

0.427583

efc5b1f63ece4cc48f000fff1e54becb

Examples of gradient-weighted class activation maps of different models generated according to <cit.>. These three retinal images are mis-classified by DenseNet121-RGB <cit.> but correctly classified by our Single-branch SalStructIQA and Dual-branch SalStructIQA. Obviously, with salient structures, the models tend to learn discriminate information from salient regions for image quality assessment.

2012.15575

CAM.jpg

0.534435

3240e06ff00b47999d814aad9580ccd2

Confusion matrices of BRISQUE <cit.>, NBIQA <cit.>, TS-CNN<cit.>, DenseNet121-MCF <cit.>, our Single-branch SalStructIQA and Dual-branch SalStructIQA.

2012.15575

ConfusionMatrixSOTA-v03.jpg

0.473837

b9d16de241d046f89ffd102e8649ce3e

The proposed deep CNN architectures incorporating with our proposed large-size salient structures and tiny-size salient structures.

2012.15575

architecture.jpg

0.471748

9682172190484906b0e8d4339046ca64

The cumulative distributions of tiny-size salient structures of each class.

2012.15575

cum.jpg

0.51764

1df2f0ca40254a5c86573d498741062f

Examples for retinal images with different quality grades. From left to right, the quality grade is Good, Usable and Reject, respectively.

2012.15575

fig4intro1.jpg

0.403581

1248b4654bcf42a79e274c5200447d70

The flowchart of our proposed SalStructIQA.

2012.15575

fig4intro2.jpg

0.398647

746be58387c945f4889fd0282937cf8b

Examples for the probability map of large-size salient structures and the masks of detected large-size salient structures.

2012.15575

fig4method_LS.jpg

0.394937

8df055b38ffe4e1396b87e089d2b41b5

Examples for the standardized line response maps and masks of detected tiny-size salient structures.

2012.15575

fig4method_TS.jpg

0.520029

c0ae9348356540d98e5af595149fc567

Plot of the (R_1(θ_1),R_2(θ_1)), θ_1=0.2i, allowed region for the NLSM with N=6, k=1 and various analytic regions b_1=0, b_2=π/2, 0.9π, and b_1=-0.1π, b_2=1.1π. The vertex at the center of the red circle (see rotated and rescaled inset) is the R-matrix in eq.(<ref>) with N=6, k=1. A new vertex appears in the smaller regions, namely the one at the center of the purple circle that corresponds to a solution with R_1=R_2, or N=6, k=0.

2012.15576

NLSM_allowed_region_C_edited.jpg

0.569883

ad0142a6ffcf43f4bc98d3dd298b4cd2

Pictorial description of the 1→ 1 R-matrix describing the amplitude for a particle to bounce from the wall possibly changing its identity a→ b. A double Wick rotation relates this process to pair production from an initial boundary state |B⟩. In the second case we can define a bulk S-matrix using the usual asymptotic states.

2012.15576

boundary_edited.jpg

0.521379

e3d5c4b0114f4e5681464150a8831872

Allowed values of A(θ_1),B(θ_1) for θ_1=0.9879 i. The solid curve is the analytic result and the points are the numerical data. All points of the curve correspond to the same integrable R-matrix that has a free parameter α that changes along the boundary, see eq.(<ref>). The vertex corresponds to the limit α→∞.

2012.15576

offdiag_bdy_exact_num.jpg

0.479948

2eb2caf7c3f746a8b36c839d4911c55a

Plot of the (R_1(θ_1),R_2(θ_1)) allowed region for the pYB model for θ_1=0.9879i. We consider the case N=6, and k=1,2,3. In this case, the whole boundary is integrable and we can compare the numerics (dots) with the analytical result (solid lines). The curves intersect at R_1(θ)=R_2(θ) common to all k in the ansatz (<ref>). The vertices (encircled) appear when a pole or zero enters that physical region and has to be cancelled by changing the CDD factor.

2012.15576

pYB_N6kall.jpg

0.486597

1c56524c667f49abbe5b048950643cd4

Physical region of the reflection process (ε≥ 0). The positive real axis corresponds to the physical reflection (ε∈ℝ_≥ m) which is the boundary value of an analytic function in the physical region (shaded). Crossing can be imposed on the upper line, this is the minimal region where crossing gives a constraint. Boundary bound states could appear as poles on the imaginary axis (dashed line) but we do not allow them here.

2012.15576

theta_region.jpg

0.453134

f3193f0e41cf481dbdb7cddbb02785ca

Extended region of analyticity. We considered two cases: b_1=0, π/2≤ b_2≤π and b_1=π-b_2, b_2>π. It might seem that in the first case analyticity is guaranteed by crossing but this is not so because of the non-trivial crossing equation (<ref>).

2012.15576

theta_region_2.jpg

0.447843

e598aa8218d54faf9c42d11d20390dd7

Extended region of analyticity. We generically extend the region below the real axis (across the physical cut) to a curve . A well defined dual problem requires a regularization, namely a bound on the analytic function on . An example of such curve is in fig.<ref> in which case the curve is θ=b_1. There we also extended the region to θ = b_2 which can be done in the same way.

2012.15576

theta_region_C.jpg

0.442932

a832b563955f420b82d50eedf607cdde

Numerical simulation of 2D corrugation pattern for (a1) 8 site Hubbard model with 4 filled particles, (b1) 9 site Hubbard model with 6 filled particles, and (c1) 13 site Hubbard model with 2 filled particles. The different sizes of solid circles and different lengths of red edges denote the disorder of the interaction strength U_j and hopping strength t_j of Eq. (<ref>). (a2)-(c2) Plots of ℱ and C_1 as function of disorder strength b. The system parameters are (a) t=1, a=0, U=-0.8t (b) t=1, a=0.2t, U=-0.8t and (c) t=1, a=0.2t, U=-0.5t. Similar with 1D system, the induced fluctuation of C_1 is minor in comparison with disorder free case (b=0 ) indicating that the existing result of 1D can be extended to 2D or higher dimensional system.

2012.15577

fig_2D.jpg

0.575068

32be05c4ced0449daa8e104fb9f02f66

Transition of the ground state driven by the imaginary hopping. (a)-(b) depict the variation of the energy spectrum with respect to hopping strength t, and (c) presents the switch of the correlator C_3. The numerical simulation is performed for the 6 site Hubbard model with 4 filled particles. The other system parameters are t_0=1, and U=-1.5t_0. The hopping strength t is in units of t_0. The red line denotes the η- pairing state. C_3 experiences an evident jump at the critical point indicating a dramatic change of the ground state. Such a first-order transition may associate with quantum phase transition to some extent.

2012.15577

fig_gene.jpg

0.488643

03a167f7c4d741ce8773757b0833a53b

Plots of the overlap ℱ and correlator C_j as a function of the strength of interaction disorder b for (a) t=1, a=0.1t , U=-0.5t (b) t=1, a=0.3t, U=-1.5t (c) t=1, a=0, U=-4t, and (d) t=1, a=0.2t, U=-4t. The numerical simulation is performed for the 6 site 1D Hubbard model at half filling and s^z=0. Here the strength of the hopping disorder a is set to be constant for each subfigure and the correlator C_j is averaged over all sites separated by a distance j. When b=0, no matter what value a takes, as long as U is non-zero, one can always get a perfect η-pairing ground state. The variation of C_j indicates that the increase of b will not result in the significant change of the η-pairing ground state; the presence of the hopping disorder can suppress the fluctuation of C_j compared to the disorder free case, which can be seen from (c)-(d). Therefore, the value of the correlator is the consequence of the interplay between two such disorders, which provides a scheme to prepare η-pairing ground state in the experiment.

2012.15577

fig_ran_r.jpg

0.516132

0a5cd15d300c4d62b52cbf1091469aa0

Comparison of the two-particle spectrum within the subspace (0,0) between the non-Hermitian setting and its parent Hermitian system for (a) U=-0.8t, (b) U=-2t, and (c) U=-4.5t, respectively. The upper and lower panels present the spectrum of the Hermitian and non-Hermitian system, respectively. The red circle and gray shading denote the bound pair and scattering state. The parent Hermitian system can be obtained by assuming it→ t. For the Hermitian system, the bound pair with the lowest energy lies in the K=0 subspace while the ground state of two-particle non-Hermitian setting locates on the subspace indexed by K=π. It is shown that the presence of the imaginary hopping not only makes all scattering energy bands imaginary but also reverses the whole bound band. In the condition of small U, there can exist an EP characterized by the divergence of ∂ϵ _K/∂ K. Such non-Hermiticity alters significantly the paring mechanism and hence favors superconductivity.

2012.15577

fig_spectrum.jpg

0.499045

ee8e8debfeaa47ef8aa238bea637c7d7

Eigenvalues of _LE, weighted for the leading edge isentropic Mach number.

2012.15579

LE_eigs.jpg

0.523345

65c48aa8cdbd427782c63dcec01a0ab5

An isometric view of the blade envelope in Figure <ref>.

2012.15579

cad_1_small.jpg

0.495068

56d827cb43fa4105baeb908d344c21ab

Close-up of the leading edge of the blade envelope in Figure <ref>. The demarcated iso-contour lines denote displacements at 5, 10 and 15 mm away from the nominal blade.

2012.15579

cad_2_small.jpg

0.429132

2bb06509c88540f894204f507562fd81

A CAD representation of the controlled loss and mass flow rate blade envelope.

2012.15579

cad_3_small.jpg

0.456392

d9be4f00cc5144ffa8453b1254fab036

The blade envelope from Figure <ref> can be suitably scaled for 3D printing and placed on desktops.

2012.15579

cad_4.jpg

0.425935

b510d0da02f84245a4ed7e951a4958d5

Conditional tuning of the leading edge isentropic Mach number.

2012.15579

inv_design_LE.jpg

0.420202

751b1238ba7e4bf8854571a78322cfde

Conditional tuning of the peak isentropic Mach number.

2012.15579

inv_design_peak2.jpg

0.40119

638de54d1928415da077e76a65b7101a

Loss and mass flow function for random training designs (blue) and invariant designs within the intersection of the inactive subspaces (yellow).

2012.15579

invariant_loss_and_mdot.jpg

0.408503

7570e59fcdec4d8895fb66db979c89b5

The blade envelope for loss and mass flow.

2012.15579

loss_mass_env.jpg

0.56072

4ea0f0aab0544494afe54f77dbe323a0

Eigenvalues of _peak, weighted for the peak isentropic Mach number.

2012.15579

peak_eigs.jpg

0.376286

d1d7122cefad4b659125cb315341fc20

Validation of polynomial model for the mass flow function. This coincides with the sufficient summary plot because the global polynomial approximation is a linear model.

2012.15579

validation_lm.jpg

0.381439

15fe1c79cfb341cba40724096ee538c6

An illustration of reinforcement learning, or multi-armed bandit problems.

2012.15584

bandit-crop.jpg

0.458755

7982e57452e24c399f7a0a24f6a6cd3b

The roadmap of this thesis.

2012.15584

chapters-crop.jpg

0.51847

a4b911d639c9438099461ecd4586e489

The illustration of a confidence ellipsoid, where X=(x_1, ⋯, x_d) is the model matrix. The shape of the ellipsoid depends on the information matrix X^TX.

2012.15584

ellipsoid-crop.jpg

0.45439

107729ae2fa84b1599fadb0d624cffd2

The process of dense subgraph discovery.

2012.15584

graphmining-crop.jpg

0.47967

bfac3a2045ca4dd6936e57c14056e99e

Phase portrait of system (<ref>) with parameters β=0.5, δ=0.2, p=2 and c=5, for different initial conditions not necessarily representing realistic cases. Empty circles represent the initial state, while solid circles represent final states. The red hyperplane corresponds to U(t) ≡ (i.e., the critical value of U, when (t)=1) while the blue hyperplane corresponds to the fast manifold in which I(t) and V(t) are proportional (.i.e, I(t)=c/p V(t)). Note that only the initial states with U_0>=1 corresponds to scenarios with >1.

2012.15585

FhasePortN.jpg

0.416652

6856cca1a6294d4895d052c331f4292a

Evolution of (t) when an antiviral treatment is initiated at a time of approximately 0.75 t̂_V, and different antiviral inhibition effects η_p are considered, smaller and greater than the critical value η_p^c. The black dashed line represents =1. As it can be seen, for values of η_p < η_p^c, (t) crosses 1 at larger times for larger values of η_p as it is stated in Lemma <ref>. This implies that if η_p < η_p^c, larger values of η_p delays the virus peak time, as it is stated in Theorem <ref>.iii. Furthermore, the figure confirms that, for real patient date, t^e is close to t̂_V.

2012.15585

Rt.jpg

0.461044

b5180879ca9041f1af0cf7929040fe89

Uncertainty analysis on viral load evolution for p-parameter 95% confidence interval. Note the uncertainty in viral load peak. Empty dots are data of COVID-19 patients.

2012.15585

Sensivity_Analysis_Case_p.jpg

0.441802

c1de0eaf61774a5aba3793a19e9988d7

Viral load time evolution with treatment initial time given by t_tr=0.7t^e. Values of η_p smaller, approximately equal and greater than η_p^c are simulated to demonstrate the results in Theorem <ref>.The black line denotes the untreated case (η_p=0).

2012.15585

allpatCL_07_te.jpg

0.427573

cd102936b7fd446f94cd72de69a8456d

Free virus behavior when treatment is started at viral load detection level (t_tr=t_DL). Values of η_p smaller (η_p^1,η_p^2), approximately equal (η_p^3,η_p^4) and greater (η_p^5,η_p^6) than η_p^c are simulated to demonstrate the results in Theorem <ref>. The black line denotes the untreated case (η_p=0).

2012.15585

allpatCL_DL.jpg

0.429959

925f2b7b84a140ada54e35d7862793df

Viral load time evolution with treatment initial time given by t_tr=t^e. Values of η_p smaller, approximately equal and greater than η_p^c are simulated to demonstrate the results in Theorem <ref>. The black line denotes the untreated case (η_p=0).

2012.15585

allpatCL_te.jpg