Speaker: Luhuan Wu (CCM) Topic: Practical and Asymptotically Exact Conditional Sampling in Diffusion Models with Applications to Protein Design
Abstract: This work develops an asymptotically exact conditional sampler for diffusion generative models. The main idea relies on the Sequential Monte Carlo technique which simulates a weighted system of particles to approximate the exact conditional distribution. We apply our method to protein motif-scaffolding task and shows that it provides the state-of-the-art performance on a few benchmark problems.