From Denoising Diffusion Models to Transport for Generative Modeling and Inference

Explore the cutting-edge world of generative modeling and inference in this 44-minute lecture by Arnaud Doucet from Oxford University. Delve into the powerful class of denoising diffusion models, understanding their capabilities and limitations. Discover how recent alternative approaches based on transport ideas can overcome these constraints. Focus on the diffusion Schrodinger bridge, an entropy-regularized version of optimal transport, and learn about its numerical approximation methods. Gain insights into the evolving landscape of generative modeling techniques and their applications in the field of artificial intelligence and machine learning.