Negative Stepsizes Make Gradient-Descent-Ascent Converge

TL;DR

This paper demonstrates that gradient-descent-ascent (GDA) can converge on min-max problems by using unconventional, time-varying, asymmetric, and negative stepsize schedules, challenging previous beliefs about GDA's failure.

Contribution

The authors introduce slingshot stepsize schedules, a novel approach with negative stepsizes, enabling GDA to converge without additional modifications on classical counterexamples.

Findings

GDA converges with negative, asymmetric, and time-varying stepsizes.

Slingshot stepsize schedules are necessary for convergence.

The method applies to the last iterate, aligning with practical use cases.

Abstract

Efficient computation of min-max problems is a central question in optimization, learning, games, and controls. Arguably the most natural algorithm is gradient-descent-ascent (GDA). However, since the 1970s, conventional wisdom has argued that GDA fails to converge even on simple problems. This failure spurred an extensive literature on modifying GDA with additional building blocks such as extragradients, optimism, momentum, anchoring, etc. In contrast, we show that GDA converges in its original form by simply using a judicious choice of stepsizes. The key innovation is the proposal of unconventional stepsize schedules (dubbed slingshot stepsize schedules) that are time-varying, asymmetric, and periodically negative. We show that all three properties are necessary for convergence, and that altogether this enables GDA to converge on the classical counterexamples (e.g., unconstrained convex-concave problems). All of our results apply to the last iterate of GDA, as is typically desired in practice. The core algorithmic intuition is that although negative stepsizes make backward progress, they de-synchronize the min and max variables (overcoming the cycling issue of GDA), and lead to a slingshot phenomenon in which the forward progress in the other iterations is overwhelmingly larger. This results in fast overall convergence. Geometrically, the slingshot dynamics leverage the non-reversibility of gradient flow: positive/negative steps cancel to first order, yielding a second-order net movement in a new direction that leads to convergence and is otherwise impossible for GDA to move in. We interpret this as a second-order finite-differencing algorithm and show that, intriguingly, it approximately implements consensus optimization, an empirically popular algorithm for min-max problems involving deep neural networks (e.g., training GANs).