AIRS in the AIR

• 

  庞旭芳

  AIRS特种机器人中心副研究员

  主持人

  嘉宾介绍

  个人简介
• 

  王捷

  佐治亚理工学院工业工程系博士生

  Entropic Regularization for Reliable Machine Learning

  嘉宾介绍

  Jie Wang is a 4-th year Ph.D. student in Industrial Engineering at the H. Milton Stewart School of Industrial and Systems Engineering. He received BS degree in Pure of Mathematics Major from The Chinese University of Hong Kong, Shenzhen. His main research studies decision-making under uncertainty. His research has been published on several journals and conferences including Operations Research, Information and Inference a Journal of the IMA, NeurIPS, AISTATS, and ISIT. He has received several awards, such as 2022 ISyE Robert Goodell Brown Research Excellence award, Winner in 2022 INFORMS Poster Competition, and Winner for Best Theoretical Paper in 2023 INFORMS Workshop on DMDA.

  个人简介

  Despite the growing prevalence of artificial neural networks in real-world applications, their vulnerability to adversarial attacks remains to be a significant concern, which motivates us to investigate the robustness of machine learning models. While various heuristics aim to optimize the distributionally robust risk using the Wasserstein metric, such a notion of robustness frequently encounters computation intractability. To tackle the computational challenge, we develop a novel approach to adversarial training that integrates entropic regularization into the distributionally robust risk function. This regularization brings a notable improvement in computation compared with the original formulation. We develop stochastic gradient methods with near-optimal sample complexity to solve this problem efficiently. Moreover, we establish the regularization effects and demonstrate this formulation is asymptotic equivalence to a regularized empirical risk minimization (ERM) framework, by considering various scaling regimes of the entropic regularization $\eta$ and robustness level $\rho$. These regimes yield gradient norm regularization, variance regularization, or a smoothed gradient norm regularization that interpolates between these extremes. We numerically validate our proposed method in supervised learning and reinforcement learning applications and showcase its state-of-the-art performance against various adversarial attacks.

视频回顾