Compelling ReLU Networks to Exhibit Exponentially Many Linear Regions at Initialization and During Training

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Milkert, Max, Hyde, David, Laine, Forrest
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917017146097664
author Milkert, Max
Hyde, David
Laine, Forrest
author_facet Milkert, Max
Hyde, David
Laine, Forrest
contents In a neural network with ReLU activations, the number of piecewise linear regions in the output can grow exponentially with depth. However, this is highly unlikely to happen when the initial parameters are sampled randomly, which therefore often leads to the use of networks that are unnecessarily large. To address this problem, we introduce a novel parameterization of the network that restricts its weights so that a depth $d$ network produces exactly $2^d$ linear regions at initialization and maintains those regions throughout training under the parameterization. This approach allows us to learn approximations of convex, one dimensional functions that are several orders of magnitude more accurate than their randomly initialized counterparts. We further demonstrate a preliminary extension of our construction to multidimensional and non-convex functions, allowing the technique to replace traditional dense layers in various architectures.
format Preprint
id arxiv_https___arxiv_org_abs_2311_18022
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Compelling ReLU Networks to Exhibit Exponentially Many Linear Regions at Initialization and During Training
Milkert, Max
Hyde, David
Laine, Forrest
Machine Learning
Artificial Intelligence
In a neural network with ReLU activations, the number of piecewise linear regions in the output can grow exponentially with depth. However, this is highly unlikely to happen when the initial parameters are sampled randomly, which therefore often leads to the use of networks that are unnecessarily large. To address this problem, we introduce a novel parameterization of the network that restricts its weights so that a depth $d$ network produces exactly $2^d$ linear regions at initialization and maintains those regions throughout training under the parameterization. This approach allows us to learn approximations of convex, one dimensional functions that are several orders of magnitude more accurate than their randomly initialized counterparts. We further demonstrate a preliminary extension of our construction to multidimensional and non-convex functions, allowing the technique to replace traditional dense layers in various architectures.
title Compelling ReLU Networks to Exhibit Exponentially Many Linear Regions at Initialization and During Training
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2311.18022