Denoising Diffusion Probabilistic Models with Prakhar Srivastava (UC Irvine)

Abstract : Denoising Diffusion Probabilistic Models (DDPM) are a new class of powerful generative models that have been shown to generate high fidelity spatio-temporal data. They are a point of fascination not only for researchers, but have become a society-wide phenomenon. In this talk, I will explore the principles behind probabilistic diffusion models and their underlying mathematical foundations. Finally, I'll present current advances in autoregressive diffusion models for videos.

Bio : I'm a second year PhD student in Computer Science at University of California, Irvine. My current research focus is on deep generative models. Before this, I was a research engineer at NEC Heidelberg (Germany) working on open information extraction and knowledge graphs. Prior to this, I completed my Master's at EPFL (Switzerland) where I got an opportunity to work with Schlumberger's petrophysical research group on Bayesian Approximation and uncertainty estimation. I have obtainedmy Bachelor's from Indian Institute of Technology, Jodhpur in Electrical Engineering.