8
61

Improved Algorithms for Convex-Concave Minimax Optimization

Abstract

This paper studies minimax optimization problems minxmaxyf(x,y)\min_x \max_y f(x,y), where f(x,y)f(x,y) is mxm_x-strongly convex with respect to xx, mym_y-strongly concave with respect to yy and (Lx,Lxy,Ly)(L_x,L_{xy},L_y)-smooth. Zhang et al. provided the following lower bound of the gradient complexity for any first-order method: Ω(Lxmx+Lxy2mxmy+Lymyln(1/ϵ)).\Omega\Bigl(\sqrt{\frac{L_x}{m_x}+\frac{L_{xy}^2}{m_x m_y}+\frac{L_y}{m_y}}\ln(1/\epsilon)\Bigr). This paper proposes a new algorithm with gradient complexity upper bound O~(Lxmx+LLxymxmy+Lymyln(1/ϵ)),\tilde{O}\Bigl(\sqrt{\frac{L_x}{m_x}+\frac{L\cdot L_{xy}}{m_x m_y}+\frac{L_y}{m_y}}\ln\left(1/\epsilon\right)\Bigr), where L=max{Lx,Lxy,Ly}L=\max\{L_x,L_{xy},L_y\}. This improves over the best known upper bound O~(L2mxmyln3(1/ϵ))\tilde{O}\left(\sqrt{\frac{L^2}{m_x m_y}} \ln^3\left(1/\epsilon\right)\right) by Lin et al. Our bound achieves linear convergence rate and tighter dependency on condition numbers, especially when LxyLL_{xy}\ll L (i.e., when the interaction between xx and yy is weak). Via reduction, our new bound also implies improved bounds for strongly convex-concave and convex-concave minimax optimization problems. When ff is quadratic, we can further improve the upper bound, which matches the lower bound up to a small sub-polynomial factor.

View on arXiv
Comments on this paper