Tutorial “Modern sampling methods and their applications”

Abstract

This minicourse covers methods for sampling from distributions whose densities are known up to a normalizing constant. It focuses on sampling using diffusion processes (denoising diffusion samplers) and their connections to stochastic optimal control, reinforcement learning, and generative flow networks. We will discuss foundational papers in the field and several recent results, including work by researchers at the AI and Digital Science Institute FCS HSE University.