Proximal Gradient Methods for Nonsmooth Nonconvex Minimax - A Unified Convergence Analysis

Explore a 26-minute conference talk on proximal gradient methods for nonsmooth nonconvex minimax problems, delivered at the "One World Optimization Seminar in Vienna" workshop held at the Erwin Schrödinger International Institute for Mathematics and Physics. Delve into the analysis of parallel and alternating proximal gradient schemes within a unified framework, expanding on general convergence mechanisms for nonconvex nonsmooth optimization. Discover pointwise global convergence results and refined complexity analyses, departing from the common focus on nearly approximate stationary solutions. Learn how this approach broadens the scope of addressable minimax problems through Non Euclidean proximal steps, extending convergence and complexity results to a wider setting. Gain insights from this joint work with Eyal Cohen, which advances beyond typical weakly convex/concave and smooth models in recent literature.