Adaptive Step Sizes for Preconditioned Stochastic Gradient Descent

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Köhne, Frederik, Kreis, Leonie, Schiela, Anton, Herzog, Roland
Format: Preprint
Veröffentlicht: 2023
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866916398271299584
author Köhne, Frederik
Kreis, Leonie
Schiela, Anton
Herzog, Roland
author_facet Köhne, Frederik
Kreis, Leonie
Schiela, Anton
Herzog, Roland
contents This paper proposes a novel approach to adaptive step sizes in stochastic gradient descent (SGD) by utilizing quantities that we have identified as numerically traceable -- the Lipschitz constant for gradients and a concept of the local variance in search directions. Our findings yield a nearly hyperparameter-free algorithm for stochastic optimization, which has provable convergence properties and exhibits truly problem adaptive behavior on classical image classification tasks. Our framework is set in a general Hilbert space and thus enables the potential inclusion of a preconditioner through the choice of the inner product.
format Preprint
id arxiv_https___arxiv_org_abs_2311_16956
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Adaptive Step Sizes for Preconditioned Stochastic Gradient Descent
Köhne, Frederik
Kreis, Leonie
Schiela, Anton
Herzog, Roland
Optimization and Control
Machine Learning
This paper proposes a novel approach to adaptive step sizes in stochastic gradient descent (SGD) by utilizing quantities that we have identified as numerically traceable -- the Lipschitz constant for gradients and a concept of the local variance in search directions. Our findings yield a nearly hyperparameter-free algorithm for stochastic optimization, which has provable convergence properties and exhibits truly problem adaptive behavior on classical image classification tasks. Our framework is set in a general Hilbert space and thus enables the potential inclusion of a preconditioner through the choice of the inner product.
title Adaptive Step Sizes for Preconditioned Stochastic Gradient Descent
topic Optimization and Control
Machine Learning
url https://arxiv.org/abs/2311.16956