Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | Department of Applied Mathematics | en_US |
| dc.contributor.advisor | Chen, Xiaojun (AMA) | en_US |
| dc.creator | Qiu, Zicheng | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/13780 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | en_US |
| dc.rights | All rights reserved | en_US |
| dc.title | A quasi-newton subspace trust region algorithm for nonmonotone variational inequalities and applications in adversarial learning | en_US |
| dcterms.abstract | Nonmonotone Variational Inequalities (VIs) are widely applied in the fields of data sciences and machine learning. However, the design of algorithms towards nonmonotone VIs is still a challenge. State-of-the-art algorithms can only solve nonmonotone VIs under some strong assumptions, such as pseudomonotonicity or Minty's condition. The main purpose of this thesis is to study a class of nonmonotone VIs with box constraints, which is equivalent to a system of nonsmooth equations. The thesis is primarily divided into two parts. | en_US |
| dcterms.abstract | In the first part of the thesis, we propose a smoothing Quasi-Newton Subspace Trust Region (QNSTR) algorithm for the least squares problems defined by the smoothing approximation of nonsmooth equations. Based on the structure of the nonmonotone VI, we use an adaptive quasi-Newton formula to approximate the Hessian matrix and solve a low-dimensional strongly convex quadratic program with ellipse constraints in a subspace at each step of the QNSTR algorithm efficiently. Moreover, we study the relationship between solutions of the VI and first order stationary points of the least squares problem, and prove the global convergence of the QNSTR algorithm to a solution of the VI under some mild conditions. We also propose a strategy to update the smoothing parameter and establish its complexity. | en_US |
| dcterms.abstract | In the second part of the thesis, we implement the QNSTR algorithm to solve a box constrained nonconvex-nonconcave minimax optimization problem with application to practical problems. Since the objective function of the optimization problem has expectation, we apply the Sample Average Approximation (SAA) method to solve the optimization problem. We prove that any accumulation points of the global minimax point, first order stationary point, second order stationary point of the SAA problem is a global minimax point, first order stationary point, second order stationary point of the original problem respectively, as the sample size N tends to be infinity. We formulate the first order optimality condition of the box constrained SAA minimax problem as a nonmonotone VI, and apply the QNSTR to find a first order stationary point of the SAA problem via the nonmonotone VI. We also present numerical results based on the QNSTR algorithm with different subspaces for generative adversarial networks on several practical adversarial learning problems using real data on eyes. The numerical results show that the QNSTR algorithm is efficient and effective for solving large scale minimax optimization problems. | en_US |
| dcterms.extent | xvii, 98 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2025 | en_US |
| dcterms.educationalLevel | Ph.D. | en_US |
| dcterms.educationalLevel | All Doctorate | en_US |
| dcterms.accessRights | open access | en_US |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13780