Hong-Kun Xu, Extra Anchored Gradient Method for Convex-Concave Minimax Problems

Abstract: Algorithmic approaches to minimax problems have recently been paid much attention, due to their important applications in machine learning, in particular, in generative adversarial nets (GANs). In this talk we will recall and discuss some traditional and recent extra gradient and extra anchored gradient methods for convex-concave minimax problems.
A theoretical ingredient is Halpern's anchored iterative method for nonexpansive mappings, which has recently been proved to play an important role in solving variational inequalities and minimax problems.