Qihang Lin - Colloquium Speaker
We consider gradient-based optimization algorithms for solving a class of non-convex non-concave min-max problems, whose objective function is weakly convex (resp. weakly concave) in the minimization (resp. maximization). It has many important applications in machine learning such as robust learning and training generative adversarial networks. Our method is an inexact proximal point method which solves the weakly monotone variational inequality corresponding to the min-max problem. Our algorithm finds a nearly stationary point of the min-max problem with a provable iteration complexity.