Particle Denoising Diffusion Sampler

Denoising diffusion models have become ubiquitous for generative modeling. The core idea is to transport the data distribution to a Gaussian by using a diffusion. Approximate samples from the data distribution are then obtained by estimating the time-reversal of this diffusion using score matching i...

Full description

Saved in:
Bibliographic Details
Published in:arXiv.org 2024-06
Main Authors: Phillips, Angus, Hai-Dang Dau, Hutchinson, Michael John, De Bortoli, Valentin, Deligiannidis, George, Doucet, Arnaud
Format: Article
Language:English
Subjects:
Iterative methods
Matching
Noise reduction
Normal distribution
Samplers
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Denoising diffusion models have become ubiquitous for generative modeling. The core idea is to transport the data distribution to a Gaussian by using a diffusion. Approximate samples from the data distribution are then obtained by estimating the time-reversal of this diffusion using score matching ideas. We follow here a similar strategy to sample from unnormalized probability densities and compute their normalizing constants. However, the time-reversed diffusion is here simulated by using an original iterative particle scheme relying on a novel score matching loss. Contrary to standard denoising diffusion models, the resulting Particle Denoising Diffusion Sampler (PDDS) provides asymptotically consistent estimates under mild assumptions. We demonstrate PDDS on multimodal and high dimensional sampling tasks.
ISSN:2331-8422