Saved in:
Bibliographic Details
Main Authors: Boonsiriphatthanajaroen, Natthawut, Henderson, Shane G.
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2504.04213
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909567291490304
author Boonsiriphatthanajaroen, Natthawut
Henderson, Shane G.
author_facet Boonsiriphatthanajaroen, Natthawut
Henderson, Shane G.
contents We study the convergence properties of the original and away-step Frank-Wolfe algorithms for linearly constrained stochastic optimization assuming the availability of unbiased objective function gradient estimates. The objective function is not restricted to a finite summation form, like in previous analyses tailored to machine-learning applications. To enable the use of concentration inequalities we assume either a uniform bound on the variance of gradient estimates or uniformly sub-Gaussian tails on gradient estimates. With one of these regularity assumptions along with sufficient sampling, we can ensure sufficiently accurate gradient estimates. We then use a Lyapunov argument to obtain the desired complexity bounds, relying on existing geometrical results for polytopes.
format Preprint
id arxiv_https___arxiv_org_abs_2504_04213
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle A New Convergence Analysis of Two Stochastic Frank-Wolfe Algorithms
Boonsiriphatthanajaroen, Natthawut
Henderson, Shane G.
Optimization and Control
We study the convergence properties of the original and away-step Frank-Wolfe algorithms for linearly constrained stochastic optimization assuming the availability of unbiased objective function gradient estimates. The objective function is not restricted to a finite summation form, like in previous analyses tailored to machine-learning applications. To enable the use of concentration inequalities we assume either a uniform bound on the variance of gradient estimates or uniformly sub-Gaussian tails on gradient estimates. With one of these regularity assumptions along with sufficient sampling, we can ensure sufficiently accurate gradient estimates. We then use a Lyapunov argument to obtain the desired complexity bounds, relying on existing geometrical results for polytopes.
title A New Convergence Analysis of Two Stochastic Frank-Wolfe Algorithms
topic Optimization and Control
url https://arxiv.org/abs/2504.04213