Taking a Big Step: Large Learning Rates in Denoising Score Matching Prevent Memorization

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Wu, Yu-Han, Marion, Pierre, Biau, Gérard, Boyer, Claire
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866908351182405632
author Wu, Yu-Han
Marion, Pierre
Biau, Gérard
Boyer, Claire
author_facet Wu, Yu-Han
Marion, Pierre
Biau, Gérard
Boyer, Claire
contents Denoising score matching plays a pivotal role in the performance of diffusion-based generative models. However, the empirical optimal score--the exact solution to the denoising score matching--leads to memorization, where generated samples replicate the training data. Yet, in practice, only a moderate degree of memorization is observed, even without explicit regularization. In this paper, we investigate this phenomenon by uncovering an implicit regularization mechanism driven by large learning rates. Specifically, we show that in the small-noise regime, the empirical optimal score exhibits high irregularity. We then prove that, when trained by stochastic gradient descent with a large enough learning rate, neural networks cannot stably converge to a local minimum with arbitrarily small excess risk. Consequently, the learned score cannot be arbitrarily close to the empirical optimal score, thereby mitigating memorization. To make the analysis tractable, we consider one-dimensional data and two-layer neural networks. Experiments validate the crucial role of the learning rate in preventing memorization, even beyond the one-dimensional setting.
format Preprint
id arxiv_https___arxiv_org_abs_2502_03435
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Taking a Big Step: Large Learning Rates in Denoising Score Matching Prevent Memorization
Wu, Yu-Han
Marion, Pierre
Biau, Gérard
Boyer, Claire
Machine Learning
Denoising score matching plays a pivotal role in the performance of diffusion-based generative models. However, the empirical optimal score--the exact solution to the denoising score matching--leads to memorization, where generated samples replicate the training data. Yet, in practice, only a moderate degree of memorization is observed, even without explicit regularization. In this paper, we investigate this phenomenon by uncovering an implicit regularization mechanism driven by large learning rates. Specifically, we show that in the small-noise regime, the empirical optimal score exhibits high irregularity. We then prove that, when trained by stochastic gradient descent with a large enough learning rate, neural networks cannot stably converge to a local minimum with arbitrarily small excess risk. Consequently, the learned score cannot be arbitrarily close to the empirical optimal score, thereby mitigating memorization. To make the analysis tractable, we consider one-dimensional data and two-layer neural networks. Experiments validate the crucial role of the learning rate in preventing memorization, even beyond the one-dimensional setting.
title Taking a Big Step: Large Learning Rates in Denoising Score Matching Prevent Memorization
topic Machine Learning
url https://arxiv.org/abs/2502.03435