Deep Unsupervised Learning using Nonequilibrium Thermodynamics

Jascha Sohl-Dickstein, Eric A. Weiss, Niru Maheswaranathan, and Surya Ganguli submitted 'Deep Unsupervised Learning using Nonequilibrium Thermodynamics' on March 12, 2015. The paper describes an iterative forward diffusion process that destroys structure in a data distribution and a learned reverse diffusion process that restores structure.