Learning Diffusion Priors from Observations by Expectation Maximization

François Rozet, Gérôme Andry, François Lanusse +1

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, whi…