pc蛋蛋

【学术报告】Distributionally Robust Universal Classification

发布者:科研办发布时间:2026-11-18浏览次数:202

【主讲人简介】:Dr. Weijun Xie is a Professor in the Department of Data Science at City University of Hong Kong. His research interests are theory and applications of stochastic, discrete, and convex optimization. His work has received multiple awards, including the 2026 INFORMS Gaver Early Career Award, the 2025 INFORMS Computing Prize (Honorable Mention), the 2022 New Investigator Award from the Virginia Space Grant Consortium (NASA), the 2021 NSF CAREER Award, and the 2020 INFORMS Young Researchers Paper Prize. He currently serves as Associate Editor of Operations Research, Mathematical Programming, Manufacturing & Service Operations Management, Journal of Global Optimization, and Naval Research Logistics.


【内容简介】:The Universal Classification (UC) problem aims to find an optimal classifier from a universal policy space that minimizes the expected 0-1 loss, also known as the misclassification risk. However, conventional empirical risk minimization often results in overfitting and poor out-of-sample performance. To address this limitation, we propose the Distributionally Robust Universal Classification (DRUC) formulation, which enhances generalization by incorporating distributional robustness through a Wasserstein distance-based ambiguity set centered at the empirical distribution. To manage the infinite-dimensional nature of the DRUC policy space, we develop its in-sample DRUC counterpart, which allows for a more tractable reformulation while preserving robustness properties. We prove that, asymptotically, the in-sample DRUC formulation converges to the original UC formulation and is equivalent to the DRUC formulation. Under mild conditions, we provide non-asymptotic finite-sample performance guarantees. Furthermore, we derive a mixed-integer linear programming (MILP) reformulation to obtain the optimal in-sample DRUC policy and propose an efficient 2-approximation algorithm. Our numerical experiments show the efficiency of the proposed approximation algorithm and demonstrate the superior out-of-sample performance of the in-sample DRUC formulation.



【讲座时间】:2026年11月19日(星期四)下午15:00


【讲座地点】:人文社科科研楼1801会议室

CopyRight © 2025 All Rights Reserved by pc蛋蛋(中国)开奖记录查询站.pceggs28.net