Proportional infinite-width infinite-depth limit for deep linear neural networks

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Bassetti, Federico, Ladelli, Lucia, Rotondo, Pietro
Natura: Preprint
Pubblicazione: 2024
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866929601922465792
author Bassetti, Federico
Ladelli, Lucia
Rotondo, Pietro
author_facet Bassetti, Federico
Ladelli, Lucia
Rotondo, Pietro
contents We study the distributional properties of linear neural networks with random parameters in the context of large networks, where the number of layers diverges in proportion to the number of neurons per layer. Prior works have shown that in the infinite-width regime, where the number of neurons per layer grows to infinity while the depth remains fixed, neural networks converge to a Gaussian process, known as the Neural Network Gaussian Process. However, this Gaussian limit sacrifices descriptive power, as it lacks the ability to learn dependent features and produce output correlations that reflect observed labels. Motivated by these limitations, we explore the joint proportional limit in which both depth and width diverge but maintain a constant ratio, yielding a non-Gaussian distribution that retains correlations between outputs. Our contribution extends previous works by rigorously characterizing, for linear activation functions, the limiting distribution as a nontrivial mixture of Gaussians.
format Preprint
id arxiv_https___arxiv_org_abs_2411_15267
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Proportional infinite-width infinite-depth limit for deep linear neural networks
Bassetti, Federico
Ladelli, Lucia
Rotondo, Pietro
Machine Learning
Disordered Systems and Neural Networks
Probability
60F05, 60H05, 62E2
We study the distributional properties of linear neural networks with random parameters in the context of large networks, where the number of layers diverges in proportion to the number of neurons per layer. Prior works have shown that in the infinite-width regime, where the number of neurons per layer grows to infinity while the depth remains fixed, neural networks converge to a Gaussian process, known as the Neural Network Gaussian Process. However, this Gaussian limit sacrifices descriptive power, as it lacks the ability to learn dependent features and produce output correlations that reflect observed labels. Motivated by these limitations, we explore the joint proportional limit in which both depth and width diverge but maintain a constant ratio, yielding a non-Gaussian distribution that retains correlations between outputs. Our contribution extends previous works by rigorously characterizing, for linear activation functions, the limiting distribution as a nontrivial mixture of Gaussians.
title Proportional infinite-width infinite-depth limit for deep linear neural networks
topic Machine Learning
Disordered Systems and Neural Networks
Probability
60F05, 60H05, 62E2
url https://arxiv.org/abs/2411.15267