Searcharxiv⌕ Search

arXiv subjects

Qingsong Wang

Publications and source records attributed to Qingsong Wang.

51 records · Page 3Linked to original sources

It begins with a boundary: A geometric view on probabilistically robust learning

Although deep neural networks have achieved super-human performance on many classification tasks, they often exhibit a worrying lack of robustness towards adversarially generated examples. Thus, considerable effort has been invested into reformulating standard Risk Minimization (RM) into an adversarially robust framework. Recently, attention has shifted towards approaches which interpolate between the robustness offered by adversarial training and the higher clean accuracy and faster training times of RM. In this paper, we take a fresh and geometric view on one such method -- Probabilistically Robust Learning (PRL). We propose a mathematical framework for understanding PRL, which allows us to identify geometric pathologies in its original formulation and to introduce a family of probabilistic nonlocal perimeter functionals to rectify them. We prove existence of solutions to the original and modified problems using novel relaxation methods and also study properties, as well as local limits, of the introduced perimeters. We also clarify, through a suitable $Γ$-convergence analysis, the way in which the original and modified PRL models interpolate between risk minimization and adversarial training.

cs.LG↗

A Momentum Accelerated Algorithm for ReLU-based Nonlinear Matrix Decomposition

Recently, there has been a growing interest in the exploration of Nonlinear Matrix Decomposition (NMD) due to its close ties with neural networks. NMD aims to find a low-rank matrix from a sparse nonnegative matrix with a per-element nonlinear function. A typical choice is the Rectified Linear Unit (ReLU) activation function. To address over-fitting in the existing ReLU-based NMD model (ReLU-NMD), we propose a Tikhonov regularized ReLU-NMD model, referred to as ReLU-NMD-T. Subsequently, we introduce a momentum accelerated algorithm for handling the ReLU-NMD-T model. A distinctive feature, setting our work apart from most existing studies, is the incorporation of both positive and negative momentum parameters in our algorithm. Our numerical experiments on real-world datasets show the effectiveness of the proposed model and algorithm. Moreover, the code is available at https://github.com/nothing2wang/NMD-TM.

cs.LG↗

Hitting time for one class of Gaussian proccesses

In this paper we propose the method to find the hitting probabilities for Gaussian integrators. Using second quantization we obtain the sseries representation for such probabilities despite the fact that integrators can be non-Markov processes.

math.PR↗

Factor modelling for high-dimensional functional time series

Many economic and scientific problems involve the analysis of high-dimensional functional time series, where the number of functional variables $p$ diverges as the number of serially dependent observations $n$ increases. In this paper, we present a novel functional factor model for high-dimensional functional time series that maintains and makes use of the functional and dynamic structure to achieve great dimension reduction and find the latent factor structure. To estimate the number of functional factors and the factor loadings, we propose a fully functional estimation procedure based on an eigenanalysis for a nonnegative definite and symmetric matrix. Our proposal involves a weight matrix to improve the estimation efficiency and tackle the issue of heterogeneity, the rationale of which is illustrated by formulating the estimation from a novel regression perspective. Asymptotic properties of the proposed method are studied when $p$ diverges at some polynomial rate as $n$ increases. To provide a parsimonious model and enhance interpretability for near-zero factor loadings, we impose sparsity assumptions on the factor loading space and then develop a regularized estimation procedure with theoretical guarantees when $p$ grows exponentially fast relative to $n.$ Finally, we demonstrate that our proposed estimators significantly outperform the competing methods through both simulations and applications to a U.K. temperature data set and a Japanese mortality data set.

stat.ME↗

Inertial Accelerated Stochastic Mirror Descent for Large-Scale Generalized Tensor CP Decomposition

The majority of classic tensor CP decomposition models are designed for squared loss, employing Euclidean distance as a local proximal term. However, the Euclidean distance is unsuitable for the generalized loss function applicable to various types of real-world data, such as integer and binary data. Consequently, algorithms developed under the squared loss are not easily adaptable to handle these generalized losses, partially due to the lack of the gradient Lipschitz continuity. This paper considers the generalized tensor CP decomposition. We use the Bregman distance as the proximal term and propose an inertial accelerated block randomized stochastic mirror descent algorithm (iTableSMD). Within a broader multi-block variance reduction and inertial acceleration framework, we demonstrate the sublinear convergence rate for the subsequential sequence produced by the iTableSMD algorithm. We further show that iTableSMD requires at most O(ε^{-2}) iterations in expectation to attain an ε-stationary point and establish the global convergence of the sequence. Numerical experiments on real datasets demonstrate that our proposed algorithm is efficient and achieve better performance than the existing state-of-the-art methods.

math.OC↗

A Bregman Proximal Stochastic Gradient Method with Extrapolation for Nonconvex Nonsmooth Problems

In this paper, we explore a specific optimization problem that involves the combination of a differentiable nonconvex function and a nondifferentiable function. The differentiable component lacks a global Lipschitz continuous gradient, posing challenges for optimization. To address this issue and accelerate the convergence, we propose a Bregman proximal stochastic gradient method with extrapolation (BPSGE), which only requires smooth adaptivity of the differentiable part. Under the variance reduction framework, we not only analyze the subsequential and global convergence of the proposed algorithm under certain conditions, but also analyze the sublinear convergence rate of the subsequence, and the complexity of the algorithm, revealing that the BPSGE algorithm requires at most O(epsilon\^\,(-2)) iterations in expectation to attain an epsilon-stationary point. To validate the effectiveness of our proposed algorithm, we conduct numerical experiments on three real-world applications: graph regularized nonnegative matrix factorization (NMF), matrix factorization with weakly-convex regularization, and NMF with nonconvex sparsity constraints. These experiments demonstrate that BPSGE is faster than the baselines without extrapolation.

math.OC↗

Metric Graph Approximations of Geodesic Spaces

We study the question of approximating a compact geodesic metric space by metric graphs satisfying a uniform upper bound on their first Betti number. We prove that, up to a suitable multiplicative constant, Reeb graphs of distance functions to a point provide optimal approximation in the Gromov-Hausdsorff sense.

math.MG↗

An inexact proximal majorization-minimization Algorithm for remote sensing image stripe noise removal

The stripe noise existing in remote sensing images badly degrades the visual quality and restricts the precision of data analysis. Therefore, many destriping models have been proposed in recent years. In contrast to these existing models, in this paper, we propose a nonconvex model with a DC function (i.e., the difference of convex functions) structure to remove the strip noise. To solve this model, we make use of the DC structure and apply an inexact proximal majorization-minimization algorithm with each inner subproblem solved by the alternating direction method of multipliers. It deserves mentioning that we design an implementable stopping criterion for the inner subproblem, while the convergence can still be guaranteed. Numerical experiments demonstrate the superiority of the proposed model and algorithm.

math.NA↗

The Persistent Topology of Optimal Transport Based Metric Thickenings

A metric thickening of a given metric space $X$ is any metric space admitting an isometric embedding of $X$. Thickenings have found use in applications of topology to data analysis, where one may approximate the shape of a dataset via the persistent homology of an increasing sequence of spaces. We introduce two new families of metric thickenings, the $p$-Vietoris-Rips and $p$-Čech metric thickenings for all $1\le p\le \infty$, which include all measures on $X$ whose $p$-diameter or $p$-radius is bounded from above, equipped with an optimal transport metric. The $p$-diameter (resp. $p$-radius) of a measure is a certain $\ell_p$ relaxation of the usual notion of diameter (resp. radius) of a subset of a metric space. These families recover the previously studied Vietoris-Rips and Čech metric thickenings when $p=\infty$. As our main contribution, we prove a stability theorem for the persistent homology of $p$-Vietoris-Rips and $p$-Čech metric thickenings, which is novel even in the case $p=\infty$. In the specific case $p=2$, we prove a Hausmann-type theorem for thickenings of manifolds, and we derive the complete list of homotopy types of the $2$-Vietoris-Rips thickenings of the $n$-sphere as the scale increases.

math.MG↗

Some results about the Tight Span of spheres

The smallest hyperconvex metric space containing a given metric space X is called the tight span of X. It is known that tight spans have many nice geometric and topological properties, and they are gradually becoming a target of research of both the metric geometry community and the topological/geometric data analysis community. In this paper, we study the tight span of n-spheres (with either geodesic metric or l infinity-metric).

math.MG↗

Complete crystallographic, spin-electronic and magnetic structure of (Co0.2Cr0.2Fe0.2Mn0.2Ni0.2)3O4: Unraveling the suppression of entropy in high entropy oxides

High entropy oxides (HEOs) are a rapidly emerging class of functional materials consisting of multiple principal cations. The original paradigm of HEOs assumes cationic occupations with the highest possible configurational entropy allowed by the composition and crystallographic structure. However, the fundamental question remains on the actual degree of configurational disorder in HEOs, especially, in systems with a low enthalpy barriers for cation anti-site mixing. Considering the experimental limitations due to the presence of multiple principal cations in HEOs, here we utilize a robust and cross-referenced characterization approach using soft X-ray magnetic circular dichroism, hard X-ray absorption spectroscopy, Mössbauer spectroscopy, neutron powder diffraction and SQUID magnetometry to study the competition between crystal field stabilization energy and configurational entropy governing the cation occupation in a spinel HEO (S-HEO), (Co$_{0.2}$Cr$_{0.2}$Fe$_{0.2}$Mn$_{0.2}$Ni$_{0.2}$)$_3$O$_4$. In contrast to the previous studies, the derived complete structural and spin-electronic model, (Co$_{0.6}$Fe$_{0.4}$)(Cr$_{0.3}$Fe$_{0.1}$Mn$_{0.3}$Ni$_{0.3}$)$_2$O$_4$, highlights a significant deviation from the hitherto assumed paradigm of entropy-driven non-preferential distribution of cations in HEOs. An immediate correlation of this result can be drawn with bulk as well as the local element specific magnetic properties, which are intrinsically dictated by cationic occupations in spinels. The real local lattice picture presented here provides an alternate viewpoint on ionic arrangement in HEOs, which is of fundamental interest for predicting and designing their structure-dependent functionalities.

cond-mat.mtrl-sci↗

A sparse semismooth Newton based augmented Lagrangian method for large-scale support vector machines

Support vector machines (SVMs) are successful modeling and prediction tools with a variety of applications. Previous work has demonstrated the superiority of the SVMs in dealing with the high dimensional, low sample size problems. However, the numerical difficulties of the SVMs will become severe with the increase of the sample size. Although there exist many solvers for the SVMs, only few of them are designed by exploiting the special structures of the SVMs. In this paper, we propose a highly efficient sparse semismooth Newton based augmented Lagrangian method for solving a large-scale convex quadratic programming problem with a linear equality constraint and a simple box constraint, which is generated from the dual problems of the SVMs. By leveraging the primal-dual error bound result, the fast local convergence rate of the augmented Lagrangian method can be guaranteed. Furthermore, by exploiting the second-order sparsity of the problem when using the semismooth Newton method,the algorithm can efficiently solve the aforementioned difficult problems. Finally, numerical comparisons demonstrate that the proposed algorithm outperforms the current state-of-the-art solvers for the large-scale SVMs.

math.OC↗

A Dual Alternating Direction Method of Multipliers for Image Decomposition and Restoration

In this paper, we develop a dual alternating direction method of multipliers (ADMM) for an image decomposition model. In this model, an image is divided into two meaningful components, i.e., a cartoon part and a texture part. The optimization algorithm that we develop not only gives the cartoon part and the texture part of an image but also gives the restored image (cartoon part + texture part). We also present the global convergence and the local linear convergence rate for the algorithm under some mild conditions. Numerical experiments demonstrate the efficiency and robustness of the dual ADMM (dADMM). Furthermore, we can obtain relatively higher signalto-noise ratio (SNR) comparing to other algorithms. It shows that the choice of the algorithm is also important even for the same model.

eess.IV↗

Manifolds with positive orthogonal Ricci curvature

In this paper we study the class of compact Kähler manifolds with positive orthogonal Ricci curvature: $Ric^\perp>0$. First we illustrate examples of Kähler manifolds with $Ric^\perp>0$ on Kähler C-spaces, and construct ones on certain projectivized vector bundles. These examples show the abundance of Kähler manifolds which admit metrics of $Ric^\perp>0$. Secondly we prove some (algebraic) geometric consequences of the condition $Ric^\perp>0$ to illustrate that the condition is also quite restrictive. Finally this last point is made evident with a classification result in dimension three and a partial classification in dimension four.

math.DG↗

On Bismut Flat Manifolds

In this paper, we give a classification of all compact Hermitian manifolds with flat Bismut connection. We show that the torsion tensor of such a manifold must be parallel, thus the universal cover of such a manifold is a Lie group equipped with a bi-invariant metric and a compatible left invariant complex structure. In particular, isosceles Hopf surfaces are the only Bismut flat compact non-Kähler surfaces, while central Calabi-Eckmann threefolds are the only simply-connected compact Bismut flat threefolds.

math.DG↗