Inertial Newton Algorithms Avoiding Strict Saddle Points

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autore principale: Castera, Camille
Natura: Preprint
Pubblicazione: 2021
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866909100755910656
author Castera, Camille
author_facet Castera, Camille
contents We study the asymptotic behavior of second-order algorithms mixing Newton's method and inertial gradient descent in non-convex landscapes. We show that, despite the Newtonian behavior of these methods, they almost always escape strict saddle points. We also evidence the role played by the hyper-parameters of these methods in their qualitative behavior near critical points. The theoretical results are supported by numerical illustrations.
format Preprint
id arxiv_https___arxiv_org_abs_2111_04596
institution arXiv
publishDate 2021
record_format arxiv
spellingShingle Inertial Newton Algorithms Avoiding Strict Saddle Points
Castera, Camille
Optimization and Control
Machine Learning
We study the asymptotic behavior of second-order algorithms mixing Newton's method and inertial gradient descent in non-convex landscapes. We show that, despite the Newtonian behavior of these methods, they almost always escape strict saddle points. We also evidence the role played by the hyper-parameters of these methods in their qualitative behavior near critical points. The theoretical results are supported by numerical illustrations.
title Inertial Newton Algorithms Avoiding Strict Saddle Points
topic Optimization and Control
Machine Learning
url https://arxiv.org/abs/2111.04596