Gradient Methods with Online Scaling

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Gao, Wenzhi, Chu, Ya-Chi, Ye, Yinyu, Udell, Madeleine
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916469983412224
author Gao, Wenzhi
Chu, Ya-Chi
Ye, Yinyu
Udell, Madeleine
author_facet Gao, Wenzhi
Chu, Ya-Chi
Ye, Yinyu
Udell, Madeleine
contents We introduce a framework to accelerate the convergence of gradient-based methods with online learning. The framework learns to scale the gradient at each iteration through an online learning algorithm and provably accelerates gradient-based methods asymptotically. In contrast with previous literature, where convergence is established based on worst-case analysis, our framework provides a strong convergence guarantee with respect to the optimal scaling matrix for the iteration trajectory. For smooth strongly convex optimization, our results provide an $O(κ^\star \log(1/\varepsilon)$) complexity result, where $κ^\star$ is the condition number achievable by the optimal preconditioner, improving on the previous $O(\sqrt{n}κ^\star \log(1/\varepsilon))$ result. In particular, a variant of our method achieves superlinear convergence on convex quadratics. For smooth convex optimization, we show for the first time that the widely-used hypergradient descent heuristic improves on the convergence of gradient descent.
format Preprint
id arxiv_https___arxiv_org_abs_2411_01803
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Gradient Methods with Online Scaling
Gao, Wenzhi
Chu, Ya-Chi
Ye, Yinyu
Udell, Madeleine
Optimization and Control
Machine Learning
We introduce a framework to accelerate the convergence of gradient-based methods with online learning. The framework learns to scale the gradient at each iteration through an online learning algorithm and provably accelerates gradient-based methods asymptotically. In contrast with previous literature, where convergence is established based on worst-case analysis, our framework provides a strong convergence guarantee with respect to the optimal scaling matrix for the iteration trajectory. For smooth strongly convex optimization, our results provide an $O(κ^\star \log(1/\varepsilon)$) complexity result, where $κ^\star$ is the condition number achievable by the optimal preconditioner, improving on the previous $O(\sqrt{n}κ^\star \log(1/\varepsilon))$ result. In particular, a variant of our method achieves superlinear convergence on convex quadratics. For smooth convex optimization, we show for the first time that the widely-used hypergradient descent heuristic improves on the convergence of gradient descent.
title Gradient Methods with Online Scaling
topic Optimization and Control
Machine Learning
url https://arxiv.org/abs/2411.01803