Saved in:
Bibliographic Details
Main Authors: Park, Junhyung, Bloebaum, Patrick, Kasiviswanathan, Shiva Prasad
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2410.06191
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913538205810688
author Park, Junhyung
Bloebaum, Patrick
Kasiviswanathan, Shiva Prasad
author_facet Park, Junhyung
Bloebaum, Patrick
Kasiviswanathan, Shiva Prasad
contents We study the least-square regression problem with a two-layer fully-connected neural network, with ReLU activation function, trained by gradient flow. Our first result is a generalization result, that requires no assumptions on the underlying regression function or the noise other than that they are bounded. We operate in the neural tangent kernel regime, and our generalization result is developed via a decomposition of the excess risk into estimation and approximation errors, viewing gradient flow as an implicit regularizer. This decomposition in the context of neural networks is a novel perspective of gradient descent, and helps us avoid uniform convergence traps. In this work, we also establish that under the same setting, the trained network overfits to the data. Together, these results, establishes the first result on benign overfitting for finite-width ReLU networks for arbitrary regression functions.
format Preprint
id arxiv_https___arxiv_org_abs_2410_06191
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Benign Overfitting for Regression with Trained Two-Layer ReLU Networks
Park, Junhyung
Bloebaum, Patrick
Kasiviswanathan, Shiva Prasad
Machine Learning
Artificial Intelligence
We study the least-square regression problem with a two-layer fully-connected neural network, with ReLU activation function, trained by gradient flow. Our first result is a generalization result, that requires no assumptions on the underlying regression function or the noise other than that they are bounded. We operate in the neural tangent kernel regime, and our generalization result is developed via a decomposition of the excess risk into estimation and approximation errors, viewing gradient flow as an implicit regularizer. This decomposition in the context of neural networks is a novel perspective of gradient descent, and helps us avoid uniform convergence traps. In this work, we also establish that under the same setting, the trained network overfits to the data. Together, these results, establishes the first result on benign overfitting for finite-width ReLU networks for arbitrary regression functions.
title Benign Overfitting for Regression with Trained Two-Layer ReLU Networks
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2410.06191