arXiv subjects

Xiaoyu Wang

Publications and source records attributed to Xiaoyu Wang.

At least 217 records · Page 12Linked to original sources

arXiv2022

Semi-parametric Makeup Transfer via Semantic-aware Correspondence

The large discrepancy between the source non-makeup image and the reference makeup image is one of the key challenges in makeup transfer. Conventional approaches for makeup transfer either learn disentangled representation or perform pixel-wise correspondence in a parametric way between two images. We argue that non-parametric techniques have a high potential for addressing the pose, expression, and occlusion discrepancies. To this end, this paper proposes a \textbf{S}emi-\textbf{p}arametric \textbf{M}akeup \textbf{T}ransfer (SpMT) method, which combines the reciprocal strengths of non-parametric and parametric mechanisms. The non-parametric component is a novel \textbf{S}emantic-\textbf{a}ware \textbf{C}orrespondence (SaC) module that explicitly reconstructs content representation with makeup representation under the strong constraint of component semantics. The reconstructed representation is desired to preserve the spatial and identity information of the source image while "wearing" the makeup of the reference image. The output image is synthesized via a parametric decoder that draws on the reconstructed representation. Extensive experiments demonstrate the superiority of our method in terms of visual quality, robustness, and flexibility. Code and pre-trained model are available at \url{https://github.com/AnonymScholar/SpMT.

cs.CV↗

arXiv2022

Zeroth-Order Randomized Subspace Newton Methods

Zeroth-order methods have become important tools for solving problems where we have access only to function evaluations. However, the zeroth-order methods only using gradient approximations are $n$ times slower than classical first-order methods for solving n-dimensional problems. To accelerate the convergence rate, this paper proposes the zeroth order randomized subspace Newton (ZO-RSN) method, which estimates projections of the gradient and Hessian by random sketching and finite differences. This allows us to compute the Newton step in a lower dimensional subspace, with small computational costs. We prove that ZO-RSN can attain lower iteration complexity than existing zeroth order methods for strongly convex problems. Our numerical experiments show that ZO-RSN can perform black-box attacks under a more restrictive limit on the number of function queries than the state-of-the-art Hessian-aware zeroth-order method.

math.OC↗

arXiv2022

Narrow bands in magnetic field and strong-coupling Hofstadter spectra

We develop a new, efficient, and general method to determine the Hofstadter spectrum of isolated narrow bands. The method works for topological as well as for trivial narrow bands by projecting the zero $\mathbf{B}$-field hybrid Wannier states -- which are localized in one direction and Bloch extended in another direction -- onto a representation of the magnetic translation group in the Landau gauge. We then apply this method to find, for the first time, the Hofstadter spectrum for the exact single particle charged excitations in the strong coupling limit of the magic angle twisted bilayer graphene at the charge neutrality point and at $|ν|=2$ down to low magnetic fields when the flux through the moiré unit cell is only $\sim 1/25$ of the electronic flux quantum i.e. $\sim 1$T at the first magic angle. The resulting spectra provide a means to investigate Landau quantization of the quasiparticles even if their dispersion is interaction induced.

cond-mat.str-el↗

arXiv2021

Leading and higher twist transverse momentum dependent parton distribution functions in the spectator model

Transverse momentum dependent parton distribution functions, also abbreviated as TMDs, offer a three-dimensional picture of hadrons by taking the intrinsic transverse momentum of the parton into consideration. Hence, they are very important for us to understand the structure of hadrons. In this article, we calculate and summarize all TMDs of quark through the spectator model, from twist-2 to twist-4. Especially, we give complete results of TMDs at twist-4. We adopt a general analytical framework to calculate TMDs, with both scalar and axial-vector spectators being considered. All TMDs are calculated analytically in the light-cone coordinate, and single gluon rescattering is considered to generate T-odd TMDs. T-even TMDs are also calculated to this level, maybe for the first time. Different from the traditional point of view, the twist-4 TMDs can contribute to some physical observables like azimuthal asymmetries. An approximate formula of the Sivers asymmetry, including twist-4 TMDs, is given.

hep-ph↗

arXiv2021

YMIR: A Rapid Data-centric Development Platform for Vision Applications

This paper introduces an open source platform to support the rapid development of computer vision applications at scale. The platform puts the efficient data development at the center of the machine learning development process, integrates active learning methods, data and model version control, and uses concepts such as projects to enable fast iterations of multiple task specific datasets in parallel. This platform abstracts the development process into core states and operations, and integrates third party tools via open APIs as implementations of the operations. This open design reduces the development cost and adoption cost for ML teams with existing tools. At the same time, the platform supports recording project development histories, through which successful projects can be shared to further boost model production efficiency on similar tasks. The platform is open source and is already used internally to meet the increasing demand for different real world computer vision applications.

cs.AI↗

arXiv2021

A Little Bit of Carbon Can do a Lot for Superconductivity in H$_3$S

First-principles calculations were carried out to provide a chemical basis for proposed structures associated with the recently reported room-temperature superconductivity in a carbonaceous sulfur hydride material under pressure. Calculations were performed on supercells of H$_3$S doped with 1.85-25\% carbon, corresponding to SH$_3$~$\rightarrow$~CH$_3$ or SH$_3$~$\rightarrow$~CH$_4$ substitutions, primarily at pressures of 270 GPa where the maximum critical temperature, $T_c$, has been reported. In the first type of substitution, the carbon atoms can be six-fold coordinated, stabilizing a CH$_6$ configuration within the cubic H$_3$S framework structure that forms under pressure. In the second, the carbon can be four-fold coordinated as methane intercalated into the H-S lattice, with or without an additional hydrogen in the framework. The results indicate that unusual local bonding configurations with respect to carbon can be stabilized under pressure. The doping breaks degenerate bands, lowering the density of states at the Fermi level ($N_F$), and localizing electrons in C-H bonds. Low levels of CH$_4$ doping do not increase $N_F$ to values as high as those calculated for $Im\bar{3}m$ H$_3$S, but they do result in a larger logarithmic average phonon frequency, and an electron-phonon coupling parameter comparable to that of $R3m$ H$_3$S. The $T_c$s estimated for carbon doping levels ranging from 1.85-5.7\% are compatible with experimental measurements for the C-S-H superconductor.

cond-mat.supr-con↗

arXiv2021

MAAC: Novel Alert Correlation Method To Detect Multi-step Attack

With the continuous improvement of attack methods, there are more and more distributed, complex, targeted attacks in which the attackers use combined attack methods to achieve the purpose. Advanced cyber attacks include multiple stages to achieve the ultimate goal. Traditional intrusion detection systems such as endpoint security management tools, firewalls, and other monitoring tools generate a large number of alerts during the attack. These alerts include attack clues, as well as many false positives unrelated to attacks. Security analysts need to analyze a large number of alerts and find useful clues from them and reconstruct attack scenarios. However, most traditional security monitoring tools cannot correlate alerts from different sources, so many multi-step attacks are still completely unnoticed, requiring manual analysis by security analysts like finding a needle in a haystack. We propose MAAC, a multi-step attack alert correlation system, which reduces repeated alerts and combines multi-step attack paths based on alert semantics and attack stages. The evaluation results of the real-world datasets show that MAAC can effectively reduce the alerts by 90\% and find attack paths from a large number of alerts.

cs.CR↗

arXiv2021

Bandwidth-based Step-Sizes for Non-Convex Stochastic Optimization

Many popular learning-rate schedules for deep neural networks combine a decaying trend with local perturbations that attempt to escape saddle points and bad local minima. We derive convergence guarantees for bandwidth-based step-sizes, a general class of learning rates that are allowed to vary in a banded region. This framework includes many popular cyclic and non-monotonic step-sizes for which no theoretical guarantees were previously known. We provide worst-case guarantees for SGD on smooth non-convex problems under several bandwidth-based step sizes, including stagewise $1/\sqrt{t}$ and the popular step-decay (constant and then drop by a constant), which is also shown to be optimal. Moreover, we show that its momentum variant converges as fast as SGD with the bandwidth-based step-decay step-size. Finally, we propose novel step-size schemes in the bandwidth-based family and verify their efficiency on several deep neural network training tasks.

cs.LG↗

arXiv2021

Evaluating the origins of the secondary bias based on the correlation of halo properties with the linear density field

Using two sets of large $N$-body simulations, we study the origin of the correlations of halo assembly time ($z_{\rm f}$), concentration ($v_{\rm max}/v_{\rm 200}$) and spin ($λ$) with the large-scale evolved density field at given halo mass, i.e. the secondary bias. We find that the secondary bias is the secondary effect of the correlations of halo properties with the linear density estimated at the same comoving scale. Using the linear density on different scales, we find two types of correlations. The internal correlation, which reflects the correlation of halo properties with the mean linear over-density $δ_{\rm L}$ within the halo Lagrangian radius $R_{\rm L}$, is positive for both $z_{\rm f}$ and $v_{\rm max}/v_{\rm 200}$, and negative for $λ$. The external correlation, which describes the correlation of halo properties with linear overdensity at $R>R_{\rm L}$ for given $δ_{\rm L}$, shows trends opposite to the internal correlation. Both of the external and internal correlations depend only weakly on halo mass, indicating a similar origin for halos of different masses. Our findings offer a transparent perspective on the origin of the secondary bias. The secondary bias can be largely explained by the competition of the external and internal correlations together with the correlation of the linear density field on different scales. These two types of correlations combined can establish the complex halo-mass dependence of the secondary bias observed in the simulations.

astro-ph.CO↗

arXiv2021

Enhancing Social Relation Inference with Concise Interaction Graph and Discriminative Scene Representation

There has been a recent surge of research interest in attacking the problem of social relation inference based on images. Existing works classify social relations mainly by creating complicated graphs of human interactions, or learning the foreground and/or background information of persons and objects, but ignore holistic scene context. The holistic scene refers to the functionality of a place in images, such as dinning room, playground and office. In this paper, by mimicking human understanding on images, we propose an approach of \textbf{PR}actical \textbf{I}nference in \textbf{S}ocial r\textbf{E}lation (PRISE), which concisely learns interactive features of persons and discriminative features of holistic scenes. Technically, we develop a simple and fast relational graph convolutional network to capture interactive features of all persons in one image. To learn the holistic scene feature, we elaborately design a contrastive learning task based on image scene classification. To further boost the performance in social relation inference, we collect and distribute a new large-scale dataset, which consists of about 240 thousand unlabeled images. The extensive experimental results show that our novel learning framework significantly beats the state-of-the-art methods, e.g., PRISE achieves 6.8$\%$ improvement for domain classification in PIPA dataset.

cs.CV↗

arXiv2021

Asymmetric Heavy Tails and Implicit Bias in Gaussian Noise Injections

Gaussian noise injections (GNIs) are a family of simple and widely-used regularisation methods for training neural networks, where one injects additive or multiplicative Gaussian noise to the network activations at every iteration of the optimisation algorithm, which is typically chosen as stochastic gradient descent (SGD). In this paper we focus on the so-called `implicit effect' of GNIs, which is the effect of the injected noise on the dynamics of SGD. We show that this effect induces an asymmetric heavy-tailed noise on SGD gradient updates. In order to model this modified dynamics, we first develop a Langevin-like stochastic differential equation that is driven by a general family of asymmetric heavy-tailed noise. Using this model we then formally prove that GNIs induce an `implicit bias', which varies depending on the heaviness of the tails and the level of asymmetry. Our empirical results confirm that different types of neural networks trained with GNIs are well-modelled by the proposed dynamics and that the implicit effect of these injections induces a bias that degrades the performance of networks.

stat.ML↗

arXiv2021

Hosts and triggers of AGNs in the Local Universe

Based on the spectroscopic and shear catalogs for SDSS galaxies in the local Universe, we compare optically-selected active galactic nuclei (AGNs) with control star-forming and quiescent galaxies on galactic, inter-halo and larger scales. We find that AGNs are preferentially found in two specific stages of galaxy evolution: star-burst and `green valley' phases, and that the stellar population of their host galaxies is quite independent of stellar mass, different from normal galaxies. Combining galaxy-galaxy lensing and galaxy clustering on large scales, we measure the mass of AGN host halos. The typical halo mass is about $10^{12}h^{-1}\rm M_{\odot}$, similar to the characteristic mass in the stellar mass-halo mass relation (SHMR). For given stellar mass, AGN host galaxies and star-forming galaxies share the same SHMR, while quiescent galaxies have more massive halos. Clustering analysis on halo scales reveals that AGNs are surrounded by a larger number of satellites (with stellar mass down to 1/1000 of the mass of the central galaxy) than star-forming galaxies, and that galaxies with larger stellar velocity dispersion have more satellites. The number of satellites also increase with halo mass, reaching unity around $10^{12}h^{-1}\rm M_{\odot}$. Our results suggest a scenario, in which the interaction of the central galaxy with the satellites triggers an early episode of star burst and AGN activities, followed by multiple AGN cycles driven by the non-axisymmetric structure produced by the interaction. The feedback from the starburst and AGN reduces the amount of cold gas for fueling the central black hole, producing a characteristic halo mass scale, $\sim 10^{12}h^{-1}\rm M_{\odot}$, where the AGN fraction peaks.

astro-ph.GA↗

arXiv2021

$\sin (ϕ_Λ-ϕ_S)$ azimuthal asymmetry in transversely polarized $Λ$ production in SIDIS within TMD factorization at EIC

We investigate the $\sin (ϕ_Λ-ϕ_S)$ single-spin asymmetry in the transversely polarized $Λ$ production in semi-inclusive deeply inelastic scattering process within the framework of the transverse momentum dependent~(TMD) factorization. The asymmetry is contributed by the convolution of the polarizing TMD fragmentation function $D_{1T}^\perp$ of the $Λ$ hyperon and the unpolarized TMD distribution function $f_1$ of the target nucleon. We apply two sets of $D_{1T}^{\perp,Λ^\uparrow/q}$, the spectator diquark model result and available parametrization, to numerically estimate the $\sin (ϕ_Λ-ϕ_S)$ asymmetry at the kinematical region of electron ion collider (EIC). We take into account the TMD evolution effect in order to include the scale dependence of the TMD distribution functions and fragmentation functions. It is found that different sets of $D_{1T}^{\perp,Λ^\uparrow/q}$ lead to different ${\sin (ϕ_Λ-ϕ_S)}$ asymmetry, particularly in sign. Future measurements on the $\sin (ϕ_Λ-ϕ_S)$ asymmetry with high precision at EIC can provide important cross check on the available $Λ$ polarizing fragmentation functions as well as constrain them more stringently.

hep-ph↗

arXiv2021

The transverse polarization of $Λ$ hyperons in $e^+e^-\toΛ^\uparrow h X$ processes within TMD factorization

We investigate the transverse polarization of the $Λ$ hyperon in the processes $e^+e^-\toΛ^\uparrow π^\pm X$ and $e^+e^-\toΛ^\uparrow K^\pm X$ within the framework of the transverse momentum dependent~(TMD) factorization. The transverse polarization is contributed by the convolution of the transversely polarizing fragmentation function~(PFF) $D_{1T}^\perp$ of the lambda hyperon and the unpolarized fragmentation function $D_1$ of pion/kaon. We adopt the spectator diquark model result for $D_{1T}^{\perp}$ to numerically estimate the transverse polarization in $e^+e^-\toΛ^\uparrow h X$ process at the kinematical region of Belle Collaboration. To implement the TMD evolution formalism of the fragmentation functions, we apply two different parametrizations on the nonperturbative Sudakov form factors associated with the fragmentation functions of the $Λ$, pion and kaon. It is found that our prediction on the polarization in the $Λπ^+$ production and $\barΛ π^-$ is consistent with the recent Belle measurement in size and sign, while the model predictions on the polarizations in $Λπ^-$ and $ΛK^\pm$ productions show strong disagreement with the Belle data. The reason for the discrepancies is discussed and possible approaches to improve the calculation in the future are also discussed.

hep-ph↗

arXiv2021

Removing Adversarial Noise in Class Activation Feature Space

Deep neural networks (DNNs) are vulnerable to adversarial noise. Preprocessing based defenses could largely remove adversarial noise by processing inputs. However, they are typically affected by the error amplification effect, especially in the front of continuously evolving attacks. To solve this problem, in this paper, we propose to remove adversarial noise by implementing a self-supervised adversarial training mechanism in a class activation feature space. To be specific, we first maximize the disruptions to class activation features of natural examples to craft adversarial examples. Then, we train a denoising model to minimize the distances between the adversarial examples and the natural examples in the class activation feature space. Empirical evaluations demonstrate that our method could significantly enhance adversarial robustness in comparison to previous state-of-the-art approaches, especially against unseen adversarial attacks and adaptive attacks.

cs.LG↗

arXiv2021

Weighted $\sin\left(3ϕ_h -ϕ_S\right)$ asymmetry from pretzelosity in SIDIS at electron ion colliders

We investigate the transverse momentum weighted azimuthal asymmetry with a $\sin\left(3ϕ_h -ϕ_S\right)$ modulation in electroproduction of charged pion production off the transversely polarized proton. The asymmetry is contributed by the product of the second transverse-moment of the pretzelosity distribution function and the first transverse-moment of the Collins function. Using the extracted result for the proton pretzelosity and the available parametrization for the pion Collins function, we numerically predict the $P_{hT}^3$-weighted $\sin\left(3ϕ_h -ϕ_S\right)$ asymmetry for charged pion production at electron ion colliders (EICs). It is found that the weighted asymmetry is around 2 percent in the kinematics region of EICs. Our results show that the planned high luminosity electron ion colliders could be feasible to access pretzelosity in the near future.

hep-ph↗

arXiv2021

Predicting the $\sinϕ_S$ Single-Spin-Asymmetry of $Λ$ Production off transversely polarized nucleon in SIDIS

We investigate the transverse target spin asymmetry $A^{\sinϕ_{S}}_{UT}$ for the unpolarized $Λ$ production in semi-inclusive deep inelastic scattering with the transverse momentum of the final-state lambda hyperon being integrated out. The asymmetry is contributed by the product of the transversity distribution function $h_1(x)$ of the nucleon and the collinear twist-3 fragmentation function $\tilde{H}(z)$ of the $Λ$ hyperon. The later one originates from the quark-gluon-quark correlation and is a naive time-reversal-odd function. We calculate $\tilde{H}$ of the $Λ$ hyperon by adopt a diquark spectator model. Using the numerical result of $\tilde{H}(z)$ and the available parametrization of $h_1(x)$ from SIDIS data, we predict the $\sinϕ_{S}$ asymmetry in the electroproduction of the $Λ$ hyperon in the kinematical region of EIC, EicC and COMPASS. In the phenomenological analysis we include the evolution effect of the distribution functions and the fragmentation functions. The results show that the asymmetries for the $Λ$ production SIDIS process is around 0.1 and may be accessible at EIC, EicC and COMPASS. We also find that the evolution of fragmentation function can affect the size of asymmetry.

hep-ph↗

arXiv2021

JAMIP: an artificial-intelligence aided data-driven infrastructure for computational materials informatics

Materials informatics has emerged as a promisingly new paradigm for accelerating materials discovery and design. It exploits the intelligent power of machine learning methods in massive materials data from experiments or simulations to seek for new materials, functionality, principles, etc. Developing specialized facility to generate, collect, manage, learn and mine large-scale materials data is crucial to materials informatics. We herein developed an artificial-intelligence-aided data-driven infrastructure named Jilin Artificial-intelligence aided Materials-design Integrated Package (JAMIP), which is an open-source Python framework to meet the research requirements of computational materials informatics. It is integrated by materials production factory, high-throughput first-principles calculations engine, automatic tasks submission and monitoring progress, data extraction, management and storage system, and artificial intelligence machine learning based data mining functions. We have integrated specific features such as inorganic crystal structure prototype database to facilitate high-throughput calculations and essential modules associated with machine learning studies of functional materials. We demonstrated how our developed code is useful in exploring materials informatics of optoelectronic semiconductors by taking halide perovskites as typical case. By obeying the principles of automation, extensibility, reliability and intelligence, the JAMIP code is a promisingly powerful tool contributing to the fast-growing field of computational materials informatics.

cond-mat.mtrl-sci↗