Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Oh, Jong Kwon, Lyu, Hanbaek, Son, Hwijae
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2509.19773
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866914053801115648
author Oh, Jong Kwon
Lyu, Hanbaek
Son, Hwijae
author_facet Oh, Jong Kwon
Lyu, Hanbaek
Son, Hwijae
contents Sobolev training, which integrates target derivatives into the loss functions, has been shown to accelerate convergence and improve generalization compared to conventional $L^2$ training. However, the underlying mechanisms of this training method remain only partially understood. In this work, we present the first rigorous theoretical framework proving that Sobolev training accelerates the convergence of Rectified Linear Unit (ReLU) networks. Under a student-teacher framework with Gaussian inputs and shallow architectures, we derive exact formulas for population gradients and Hessians, and quantify the improvements in conditioning of the loss landscape and gradient-flow convergence rates. Extensive numerical experiments validate our theoretical findings and show that the benefits of Sobolev training extend to modern deep learning tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2509_19773
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Sobolev acceleration for neural networks
Oh, Jong Kwon
Lyu, Hanbaek
Son, Hwijae
Machine Learning
Artificial Intelligence
Sobolev training, which integrates target derivatives into the loss functions, has been shown to accelerate convergence and improve generalization compared to conventional $L^2$ training. However, the underlying mechanisms of this training method remain only partially understood. In this work, we present the first rigorous theoretical framework proving that Sobolev training accelerates the convergence of Rectified Linear Unit (ReLU) networks. Under a student-teacher framework with Gaussian inputs and shallow architectures, we derive exact formulas for population gradients and Hessians, and quantify the improvements in conditioning of the loss landscape and gradient-flow convergence rates. Extensive numerical experiments validate our theoretical findings and show that the benefits of Sobolev training extend to modern deep learning tasks.
title Sobolev acceleration for neural networks
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2509.19773