Title: Non-Convex Min-Max Games: From Defense Against Adversarial Attacks to Fair Inference
Abstract: Recent applications that arise in machine learning have surged significant interest in solving min-max saddle point games. This problem has been extensively studied in the convex-concave regime for which a global equilibrium solution can be computed efficiently. In this talk, we study the problem in the non-convex regime and show that an $\epsilon$--first order Nash equilibrium of the game can be computed in O($\epsilon^{-2.5}$) gradient evaluations when only one of the player’s objective is convex. We then discuss the consequences and the application of the developed theory in defense against adversarial attacks to neural networks, fair inference, and generative adversarial imitation learning.
Bio: Meisam Razaviyayn is an assistant professor of Industrial and Systems Engineering and Computer Science at the University of Southern California. Prior to joining USC, he was a postdoctoral research fellow in the Department of Electrical Engineering at Stanford University. He received his PhD in Electrical Engineering with minor in Computer Science at the University of Minnesota and he obtained his MS degree in Mathematics. Meisam Razaviyayn is the recipient of IEEE Data Science Workshop Best Paper Award in 2019, the Signal Processing Society Young Author Best Paper Award in 2014, and the finalist for Best Paper Prize for Young Researcher in Continuous Optimization in 2013 and 2016. His research interests include the design and analysis of large scale optimization algorithms arise in modern data science era.
