1: \begin{abstract}

2: 	In this paper, we focus on solving a class of non-convex non-concave min-max problem using stochastic first-order oracle. Under assumptions of Lipschitz-continuity and the pseudomonotonicity of the first-order oracle, we develop a variant of optimistic mirror descent and show its \textit{non-asymptotic} first-order convergence. Our algorithm is computationally cheaper than the traditional stochastic extragradient method~\citep{iusem2017extragradient} while maintaining the state-of-the-art iteration complexity established in~\citep{iusem2017extragradient} simultaneously.

3: 	% Our analysis for optimistic mirror descent is not restricted to bilinear case as in~\citep{daskalakis2017training}. 

4: 	%We conduct an synthetic experiment to verify our theory and also 

5: 	We conduct an experiment on GAN training to demonstrate the effectiveness of the proposed algorithm.

6: 	%by experimental results on GAN training.

7: 	%  We also provide centralized and decentralized versions of the proposed algorithms.

8: \end{abstract}

9: