Saved in:
Bibliographic Details
Main Authors: Oldewage, Elre T., Clarke, Ross M., Hernández-Lobato, José Miguel
Format: Preprint
Published: 2023
Subjects:
Online Access:https://arxiv.org/abs/2310.14901
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909120733380608
author Oldewage, Elre T.
Clarke, Ross M.
Hernández-Lobato, José Miguel
author_facet Oldewage, Elre T.
Clarke, Ross M.
Hernández-Lobato, José Miguel
contents Despite their popularity in the field of continuous optimisation, second-order quasi-Newton methods are challenging to apply in machine learning, as the Hessian matrix is intractably large. This computational burden is exacerbated by the need to address non-convexity, for instance by modifying the Hessian's eigenvalues as in Saddle-Free Newton methods. We propose an optimisation algorithm which addresses both of these concerns - to our knowledge, the first efficiently-scalable optimisation algorithm to asymptotically use the exact inverse Hessian with absolute-value eigenvalues. Our method frames the problem as a series which principally square-roots and inverts the squared Hessian, then uses it to precondition a gradient vector, all without explicitly computing or eigendecomposing the Hessian. A truncation of this infinite series provides a new optimisation algorithm which is scalable and comparable to other first- and second-order optimisation methods in both runtime and optimisation performance. We demonstrate this in a variety of settings, including a ResNet-18 trained on CIFAR-10.
format Preprint
id arxiv_https___arxiv_org_abs_2310_14901
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Series of Hessian-Vector Products for Tractable Saddle-Free Newton Optimisation of Neural Networks
Oldewage, Elre T.
Clarke, Ross M.
Hernández-Lobato, José Miguel
Machine Learning
Despite their popularity in the field of continuous optimisation, second-order quasi-Newton methods are challenging to apply in machine learning, as the Hessian matrix is intractably large. This computational burden is exacerbated by the need to address non-convexity, for instance by modifying the Hessian's eigenvalues as in Saddle-Free Newton methods. We propose an optimisation algorithm which addresses both of these concerns - to our knowledge, the first efficiently-scalable optimisation algorithm to asymptotically use the exact inverse Hessian with absolute-value eigenvalues. Our method frames the problem as a series which principally square-roots and inverts the squared Hessian, then uses it to precondition a gradient vector, all without explicitly computing or eigendecomposing the Hessian. A truncation of this infinite series provides a new optimisation algorithm which is scalable and comparable to other first- and second-order optimisation methods in both runtime and optimisation performance. We demonstrate this in a variety of settings, including a ResNet-18 trained on CIFAR-10.
title Series of Hessian-Vector Products for Tractable Saddle-Free Newton Optimisation of Neural Networks
topic Machine Learning
url https://arxiv.org/abs/2310.14901