Flexible block-iterative analysis for the Frank-Wolfe algorithm

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Braun, Gábor, Halbey, Jannis, Pokutta, Sebastian, Woodstock, Zev
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866909963468668928
author Braun, Gábor
Halbey, Jannis
Pokutta, Sebastian
Woodstock, Zev
author_facet Braun, Gábor
Halbey, Jannis
Pokutta, Sebastian
Woodstock, Zev
contents We prove that the block-coordinate Frank-Wolfe (BCFW) algorithm converges with state-of-the-art rates in both convex and nonconvex settings under a very mild "block-iterative" assumption. This appears to be the first result on BCFW addressing the setting of nonconvex objective functions with Lipschitz-continuous gradients and no additional assumptions. This analysis newly allows for (I) progress without activating the most-expensive linear minimization oracle(s), LMO(s), at every iteration, (II) parallelized updates that do not require all LMOs, and therefore (III) deterministic parallel update strategies that take into account the numerical cost of the problem's LMOs. Our results apply for short-step BCFW as well as an adaptive method for convex functions. New relationships between updated coordinates and primal progress are proven, and a favorable speedup is demonstrated using FrankWolfe.jl.
format Preprint
id arxiv_https___arxiv_org_abs_2409_06931
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Flexible block-iterative analysis for the Frank-Wolfe algorithm
Braun, Gábor
Halbey, Jannis
Pokutta, Sebastian
Woodstock, Zev
Optimization and Control
49M27, 49M37, 65K05, 90C26, 90C30
We prove that the block-coordinate Frank-Wolfe (BCFW) algorithm converges with state-of-the-art rates in both convex and nonconvex settings under a very mild "block-iterative" assumption. This appears to be the first result on BCFW addressing the setting of nonconvex objective functions with Lipschitz-continuous gradients and no additional assumptions. This analysis newly allows for (I) progress without activating the most-expensive linear minimization oracle(s), LMO(s), at every iteration, (II) parallelized updates that do not require all LMOs, and therefore (III) deterministic parallel update strategies that take into account the numerical cost of the problem's LMOs. Our results apply for short-step BCFW as well as an adaptive method for convex functions. New relationships between updated coordinates and primal progress are proven, and a favorable speedup is demonstrated using FrankWolfe.jl.
title Flexible block-iterative analysis for the Frank-Wolfe algorithm
topic Optimization and Control
49M27, 49M37, 65K05, 90C26, 90C30
url https://arxiv.org/abs/2409.06931