Scalable Training of Diffusion Samplers via Fixed-point Score Matching

This talk explores recent advancements in neural sampling using diffusion models. I will begin by highlighting the shortcomings of prior methods and explaining how recent approaches, inspired by the success of data-driven paradigms, overcome these limitations. I will conclude by showcasing practical applications in materials science and reinforcement learning.

Speaker bio

Denis Blessing is a fourth-year PhD candidate at the Karlsruhe Institute of Technology (KIT) and a 2025 Google PhD Fellowship recipient. His research focuses on scalable algorithms for approximate inference and neural sampling with diffusion and flow models.

Watch Next