English

DIPPA: An improved Method for Bilinear Saddle Point Problems

Machine Learning 2021-03-16 v1 Optimization and Control

Abstract

This paper studies bilinear saddle point problems minxmaxyg(x)+xAyh(y)\min_{\bf{x}} \max_{\bf{y}} g(\bf{x}) + \bf{x}^{\top} \bf{A} \bf{y} - h(\bf{y}), where the functions g,hg, h are smooth and strongly-convex. When the gradient and proximal oracle related to gg and hh are accessible, optimal algorithms have already been developed in the literature \cite{chambolle2011first, palaniappan2016stochastic}. However, the proximal operator is not always easy to compute, especially in constraint zero-sum matrix games \cite{zhang2020sparsified}. This work proposes a new algorithm which only requires the access to the gradients of g,hg, h. Our algorithm achieves a complexity upper bound O~(A2μxμy+κxκy(κx+κy)4)\tilde{\mathcal{O}}\left( \frac{\|\bf{A}\|_2}{\sqrt{\mu_x \mu_y}} + \sqrt[4]{\kappa_x \kappa_y (\kappa_x + \kappa_y)} \right) which has optimal dependency on the coupling condition number A2μxμy\frac{\|\bf{A}\|_2}{\sqrt{\mu_x \mu_y}} up to logarithmic factors.

Keywords

Cite

@article{arxiv.2103.08270,
  title  = {DIPPA: An improved Method for Bilinear Saddle Point Problems},
  author = {Guangzeng Xie and Yuze Han and Zhihua Zhang},
  journal= {arXiv preprint arXiv:2103.08270},
  year   = {2021}
}
R2 v1 2026-06-24T00:09:41.680Z