Dan Simpson • 2/8/2023

Diffusion models; or Yet another way to sample from an arbitrary distribution

This blog post humorously introduces diffusion models as a technique for sampling from complex probability distributions, framing it within the broader context of measure transport problems. It contrasts diffusion models with other methods like Markov chain Monte Carlo (MCMC) and normalizing flows, beginning with the simpler case of constructing transport maps for 1D continuous distributions.

0 comments

#Machine Learning #sampling #Diffusion Models