Accelerated Algorithms for Smooth Convex-Concave Minimax Problems with $\mathcal{O}(1/k^2)$ Rate on Squared Gradient Norm
Optimization and Control
2021-06-11 v2
Abstract
In this work, we study the computational complexity of reducing the squared gradient magnitude for smooth minimax optimization problems. First, we present algorithms with accelerated last-iterate rates, faster than the existing or slower rates for extragradient, Popov, and gradient descent with anchoring. The acceleration mechanism combines extragradient steps with anchoring and is distinct from Nesterov's acceleration. We then establish optimality of the rate through a matching lower bound.
Cite
@article{arxiv.2102.07922,
title = {Accelerated Algorithms for Smooth Convex-Concave Minimax Problems with $\mathcal{O}(1/k^2)$ Rate on Squared Gradient Norm},
author = {TaeHo Yoon and Ernest K. Ryu},
journal= {arXiv preprint arXiv:2102.07922},
year = {2021}
}
Comments
Published at ICML 2021 as a long talk