Amortizing intractable inference in diffusion models for vision, language, and control Moksh Jain

Diffusion models have emerged as effective distribution estimators in vision, language, and reinforcement learning, but their use as priors in downstream tasks poses an intractable posterior inference problem.