On Convergence of Gradient Descent Ascent: A Tight Local Analysis

TL;DR

This paper provides a detailed local convergence analysis of Gradient Descent Ascent (GDA) for nonconvex-nonconcave minimax problems, revealing optimal stepsize ratios and convergence rates that align with practical observations.

Contribution

It establishes the necessary and sufficient stepsize ratio for local convergence of GDA to a Stackelberg Equilibrium in nonconvex-nonconcave settings, extending theoretical understanding.

Findings

A stepsize ratio of Θ(κ) is necessary and sufficient for local convergence.

The paper proves a nearly tight convergence rate with a matching lower bound.

Numerical experiments support the theoretical convergence guarantees.

Abstract

Gradient Descent Ascent (GDA) methods are the mainstream algorithms for minimax optimization in generative adversarial networks (GANs). Convergence properties of GDA have drawn significant interest in the recent literature. Specifically, for $\min_{\mathbf{x}} \max_{\mathbf{y}} f(\mathbf{x};\mathbf{y})$ where $f$ is strongly-concave in $\mathbf{y}$ and possibly nonconvex in $\mathbf{x}$, (Lin et al., 2020) proved the convergence of GDA with a stepsize ratio $\eta_{\mathbf{y}}/\eta_{\mathbf{x}}=\Theta(\kappa^2)$ where $\eta_{\mathbf{x}}$ and $\eta_{\mathbf{y}}$ are the stepsizes for $\mathbf{x}$ and $\mathbf{y}$ and $\kappa$ is the condition number for $\mathbf{y}$. While this stepsize ratio suggests a slow training of the min player, practical GAN algorithms typically adopt similar stepsizes for both variables, indicating a wide gap between theoretical and empirical results. In this paper, we aim to bridge this gap by analyzing the \emph{local convergence} of general \emph{nonconvex-nonconcave} minimax problems. We demonstrate that a stepsize ratio of $\Theta(\kappa)$ is necessary and sufficient for local convergence of GDA to a Stackelberg Equilibrium, where $\kappa$ is the local condition number for $\mathbf{y}$. We prove a nearly tight convergence rate with a matching lower bound. We further extend the convergence guarantees to stochastic GDA and extra-gradient methods (EG). Finally, we conduct several numerical experiments to support our theoretical findings.