Conference

Learning Diffusion Priors from Observations by Expectation Maximization

Bibliographic Details
Title: Learning Diffusion Priors from Observations by Expectation Maximization
Authors: Rozet, François, Andry, Gérôme, Lanusse, François, Louppe, Gilles
Source: Advances in Neural Information Processing Systems, 37 (2024-05-22); The Thirty-Eighth Annual Conference on Neural Information Processing Systems, Vancouver, Canada [CA], December 10-15, 2024
Publisher Information: Curran Associates, 2024.
Publication Year: 2024
Subject Terms: Computer Science - Learning, Statistics - Machine Learning, Engineering, computing & technology, Computer science, Ingénierie, informatique & technologie, Sciences informatiques
Description: Diffusion models recently proved to be remarkable priors for Bayesian inverse problems. However, training these models typically requires access to large amounts of clean data, which could prove difficult in some settings. In this work, we present DiEM, a novel method based on the expectation-maximization algorithm for training diffusion models from incomplete and noisy observations only. Unlike previous works, DiEM leads to proper diffusion models, which is crucial for downstream tasks. As part of our methods, we propose and motivate an improved posterior sampling scheme for unconditional diffusion models. We present empirical evidence supporting the effectiveness of our approach.
Document Type: conference paper
http://purl.org/coar/resource_type/c_5794
conferenceObject
peer reviewed
Language: English
Relation: https://openreview.net/forum?id=7v88Fh6iSM; https://openreview.net/forum?id=7v88Fh6iSM; urn:issn:1049-5258
Access URL: https://orbi.uliege.be/handle/2268/319891
Rights: open access
http://purl.org/coar/access_right/c_abf2
info:eu-repo/semantics/openAccess
Accession Number: edsorb.319891
Database: ORBi
Be the first to leave a comment!
You must be logged in first