Recursive Bound-Constrained AdaGrad with Applications to Multilevel and Domain Decomposition Minimization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Gratton, Serge, Kopaničáková, Alena, Toint, Philippe
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918093240926208
author Gratton, Serge
Kopaničáková, Alena
Toint, Philippe
author_facet Gratton, Serge
Kopaničáková, Alena
Toint, Philippe
contents Two OFFO (Objective-Function Free Optimization) noise tolerant algorithms are presented that handle bound constraints, inexact gradients and use second-order information when available.The first is a multi-level method exploiting a hierarchical description of the problem and the second is a domain-decomposition method covering the standard addditive Schwarz decompositions. Both are generalizations of the first-order AdaGrad algorithm for unconstrained optimization. Because these algorithms share a common theoretical framework, a single convergence/complexity theory is provided which covers them both. Its main result is that, with high probability, both methods need at most $O(ε^{-2})$ iterations and noisy gradient evaluations to compute an $ε$-approximate first-order critical point of the bound-constrained problem. Extensive numerical experiments are discussed on applications ranging from PDE-based problems to deep neural network training, illustrating their remarkable computational efficiency.
format Preprint
id arxiv_https___arxiv_org_abs_2507_11513
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Recursive Bound-Constrained AdaGrad with Applications to Multilevel and Domain Decomposition Minimization
Gratton, Serge
Kopaničáková, Alena
Toint, Philippe
Optimization and Control
Artificial Intelligence
Numerical Analysis
49K20, 65M55, 65Y20, 68Q25, 68T05, 90C26, 90C30
F.2.1; G.1.8; I.2.5
Two OFFO (Objective-Function Free Optimization) noise tolerant algorithms are presented that handle bound constraints, inexact gradients and use second-order information when available.The first is a multi-level method exploiting a hierarchical description of the problem and the second is a domain-decomposition method covering the standard addditive Schwarz decompositions. Both are generalizations of the first-order AdaGrad algorithm for unconstrained optimization. Because these algorithms share a common theoretical framework, a single convergence/complexity theory is provided which covers them both. Its main result is that, with high probability, both methods need at most $O(ε^{-2})$ iterations and noisy gradient evaluations to compute an $ε$-approximate first-order critical point of the bound-constrained problem. Extensive numerical experiments are discussed on applications ranging from PDE-based problems to deep neural network training, illustrating their remarkable computational efficiency.
title Recursive Bound-Constrained AdaGrad with Applications to Multilevel and Domain Decomposition Minimization
topic Optimization and Control
Artificial Intelligence
Numerical Analysis
49K20, 65M55, 65Y20, 68Q25, 68T05, 90C26, 90C30
F.2.1; G.1.8; I.2.5
url https://arxiv.org/abs/2507.11513