Deep Unsupervised Learning using Nonequilibrium Thermodynamics

TLDR

Modeling complex data sets with highly flexible probability distributions while keeping learning, sampling, inference, and evaluation tractable remains a central challenge in machine learning. We propose an approach that simultaneously delivers both flexibility and tractability. Inspired by non‑equilibrium statistical physics, the method iteratively destroys structure in the data via a forward diffusion process and then learns a reverse diffusion that restores structure, producing a highly flexible yet tractable generative model. This framework enables rapid learning, sampling, and probability evaluation in deep generative models with thousands of layers or time steps, supports conditional and posterior inference, and we provide an open‑source implementation.

Abstract

A central problem in machine learning involves modeling complex data-sets using highly flexible families of probability distributions in which learning, sampling, inference, and evaluation are still analytically or computationally tractable. Here, we develop an approach that simultaneously achieves both flexibility and tractability. The essential idea, inspired by non-equilibrium statistical physics, is to systematically and slowly destroy structure in a data distribution through an iterative forward diffusion process. We then learn a reverse diffusion process that restores structure in data, yielding a highly flexible and tractable generative model of the data. This approach allows us to rapidly learn, sample from, and evaluate probabilities in deep generative models with thousands of layers or time steps, as well as to compute conditional and posterior probabilities under the learned model. We additionally release an open source reference implementation of the algorithm.