Browse Teach Curate

Stochastic Extragradient with Flip-Flop Anchoring for Minimax Optimization

A shuffling-based stochastic extragradient algorithm (SEG-FFA) for finite-sum minimax optimization that combines flip-flop component ordering with an anchoring step, and is proven to converge faster than independent (with-replacement) sampling for strongly-monotone/strongly-convex-strongly-concave problems.

Data Text

2501.00511

This paper studies stochastic extragradient (SEG) methods for finite-sum minimax (saddle-point) optimization, where one variable minimizes and another maximizes the same objective. It shows that shuf…