Rapid Learning in Constrained Minimax Games with Negative Momentum
This work addresses constrained minimax optimization problems, which are common in game theory and machine learning, though it appears incremental as it builds on existing negative momentum methods.
The paper tackles constrained minimax games by extending negative momentum techniques to constrained settings with a novel momentum buffer framework, achieving significant performance improvements over original versions and SOTA baselines in experiments on Normal Form Games and Extensive Form Games.
In this paper, we delve into the utilization of the negative momentum technique in constrained minimax games. From an intuitive mechanical standpoint, we introduce a novel framework for momentum buffer updating, which extends the findings of negative momentum from the unconstrained setting to the constrained setting and provides a universal enhancement to the classic game-solver algorithms. Additionally, we provide theoretical guarantee of convergence for our momentum-augmented algorithms with entropy regularizer. We then extend these algorithms to their extensive-form counterparts. Experimental results on both Normal Form Games (NFGs) and Extensive Form Games (EFGs) demonstrate that our momentum techniques can significantly improve algorithm performance, surpassing both their original versions and the SOTA baselines by a large margin.