Beyond Scores: Proximal Diffusion Models

Sam Buchanan (UC Berkeley) · Zhenghan Fang (Johns Hopkins University) · Mateo Diaz (Johns Hopkins University) · Jeremias Sulam (Johns Hopkins University)
backward discretizationconvergence ratesdiffusion modelsforward discretizationgenerative modelshigh-dimensional datakl divergenceproximal diffusion modelsproximal mapsproximal matchingproximal operatorssampling efficiencyscore estimationscore-matching methodsstochastic differential equation

Diffusion models have quickly become some of the most popular and powerful generative models for high-dimensional data. The key insight that enabled their development was the realization that access to the score