Learning Variational Inequalities from Data: Fast Generalization Rates under Strong Monotonicity

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhao, Eric, Chavdarova, Tatjana, Jordan, Michael
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929719847419904
author Zhao, Eric
Chavdarova, Tatjana
Jordan, Michael
author_facet Zhao, Eric
Chavdarova, Tatjana
Jordan, Michael
contents Variational inequalities (VIs) are a broad class of optimization problems encompassing machine learning problems ranging from standard convex minimization to more complex scenarios like min-max optimization and computing the equilibria of multi-player games. In convex optimization, strong convexity allows for fast statistical learning rates requiring only $Θ(1/ε)$ stochastic first-order oracle calls to find an $ε$-optimal solution, rather than the standard $Θ(1/ε^2)$ calls. This note provides a simple overview of how one can similarly obtain fast $Θ(1/ε)$ rates for learning VIs that satisfy strong monotonicity, a generalization of strong convexity. Specifically, we demonstrate that standard stability-based generalization arguments for convex minimization extend directly to VIs when the domain admits a small covering, or when the operator is integrable and suboptimality is measured by potential functions; such as when finding equilibria in multi-player games.
format Preprint
id arxiv_https___arxiv_org_abs_2410_20649
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Learning Variational Inequalities from Data: Fast Generalization Rates under Strong Monotonicity
Zhao, Eric
Chavdarova, Tatjana
Jordan, Michael
Machine Learning
Optimization and Control
Variational inequalities (VIs) are a broad class of optimization problems encompassing machine learning problems ranging from standard convex minimization to more complex scenarios like min-max optimization and computing the equilibria of multi-player games. In convex optimization, strong convexity allows for fast statistical learning rates requiring only $Θ(1/ε)$ stochastic first-order oracle calls to find an $ε$-optimal solution, rather than the standard $Θ(1/ε^2)$ calls. This note provides a simple overview of how one can similarly obtain fast $Θ(1/ε)$ rates for learning VIs that satisfy strong monotonicity, a generalization of strong convexity. Specifically, we demonstrate that standard stability-based generalization arguments for convex minimization extend directly to VIs when the domain admits a small covering, or when the operator is integrable and suboptimality is measured by potential functions; such as when finding equilibria in multi-player games.
title Learning Variational Inequalities from Data: Fast Generalization Rates under Strong Monotonicity
topic Machine Learning
Optimization and Control
url https://arxiv.org/abs/2410.20649