Saved in:
Bibliographic Details
Main Authors: Beyler, Eliot, Bach, Francis
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2503.12966
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918392434262016
author Beyler, Eliot
Bach, Francis
author_facet Beyler, Eliot
Bach, Francis
contents Score-based generative models achieve state-of-the-art sampling performance by denoising a distribution perturbed by Gaussian noise. In this paper, we focus on a single deterministic denoising step, and compare the optimal denoiser for the quadratic loss, we name ''full-denoising'', to the alternative ''half-denoising'' introduced by Hyv{ä}rinen (2025). We show that looking at the performance in terms of distance between distributions tells a more nuanced story, with different assumptions on the data leading to very different conclusions. We prove that half-denoising is better than full-denoising for regular enough densities, while full-denoising is better for singular densities such as mixtures of Dirac measures or densities supported on a low-dimensional subspace. In the latter case, we prove that full-denoising can alleviate the curse of dimensionality under a linear manifold hypothesis.
format Preprint
id arxiv_https___arxiv_org_abs_2503_12966
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Optimal Denoising in Score-Based Generative Models: The Role of Data Regularity
Beyler, Eliot
Bach, Francis
Machine Learning
Score-based generative models achieve state-of-the-art sampling performance by denoising a distribution perturbed by Gaussian noise. In this paper, we focus on a single deterministic denoising step, and compare the optimal denoiser for the quadratic loss, we name ''full-denoising'', to the alternative ''half-denoising'' introduced by Hyv{ä}rinen (2025). We show that looking at the performance in terms of distance between distributions tells a more nuanced story, with different assumptions on the data leading to very different conclusions. We prove that half-denoising is better than full-denoising for regular enough densities, while full-denoising is better for singular densities such as mixtures of Dirac measures or densities supported on a low-dimensional subspace. In the latter case, we prove that full-denoising can alleviate the curse of dimensionality under a linear manifold hypothesis.
title Optimal Denoising in Score-Based Generative Models: The Role of Data Regularity
topic Machine Learning
url https://arxiv.org/abs/2503.12966