Random Matrix Theory for Stochastic Gradient Descent

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Park, Chanju, Favoni, Matteo, Lucini, Biagio, Aarts, Gert
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912172061229056
author Park, Chanju
Favoni, Matteo
Lucini, Biagio
Aarts, Gert
author_facet Park, Chanju
Favoni, Matteo
Lucini, Biagio
Aarts, Gert
contents Investigating the dynamics of learning in machine learning algorithms is of paramount importance for understanding how and why an approach may be successful. The tools of physics and statistics provide a robust setting for such investigations. Here we apply concepts from random matrix theory to describe stochastic weight matrix dynamics, using the framework of Dyson Brownian motion. We derive the linear scaling rule between the learning rate (step size) and the batch size, and identify universal and non-universal aspects of weight matrix dynamics. We test our findings in the (near-)solvable case of the Gaussian Restricted Boltzmann Machine and in a linear one-hidden-layer neural network.
format Preprint
id arxiv_https___arxiv_org_abs_2412_20496
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Random Matrix Theory for Stochastic Gradient Descent
Park, Chanju
Favoni, Matteo
Lucini, Biagio
Aarts, Gert
High Energy Physics - Lattice
Disordered Systems and Neural Networks
Machine Learning
Investigating the dynamics of learning in machine learning algorithms is of paramount importance for understanding how and why an approach may be successful. The tools of physics and statistics provide a robust setting for such investigations. Here we apply concepts from random matrix theory to describe stochastic weight matrix dynamics, using the framework of Dyson Brownian motion. We derive the linear scaling rule between the learning rate (step size) and the batch size, and identify universal and non-universal aspects of weight matrix dynamics. We test our findings in the (near-)solvable case of the Gaussian Restricted Boltzmann Machine and in a linear one-hidden-layer neural network.
title Random Matrix Theory for Stochastic Gradient Descent
topic High Energy Physics - Lattice
Disordered Systems and Neural Networks
Machine Learning
url https://arxiv.org/abs/2412.20496