Waterloo.AI Seminar Series • Exploiting Symmetries for Probabilistic Generative Modelling

Thursday, June 23, 2022 1:00 pm - 2:30 pm EDT (GMT -04:00)

Please note: This seminar will be given in person in DC 1304 and online.

Priyank Jaini, Research Scientist, Brain Team
Google Research

Symmetries play a crucial role in Physics and Mathematics. In this talk, I will explore generative models for efficient sampling and inference by incorporating inductive biases in the form of symmetries. I will begin by introducing Equivariant Stein Variational Gradient Descent (SVGD) algorithm — an equivariant sampling method based on Stein’s identity for sampling from symmetric distributions. Subsequently, I will discuss training equivariant energy based models using Equivariant-SVGD to model invariant probability distributions with applications in many-body particle systems and molecular structure generation.

In the second part, I will briefly talk about sampling transition paths between metastable conformations of molecular systems e.g., folded and unfolded protein, chemical reactions from reactants to products etc. I will formulate the problem as a stochastic optimal control problem and discuss strategies to efficiently sample these transition paths by incorporating inductive biases in the form of both symmetries as well as second order molecular dynamics and demonstrate applications for protein structures like Alanine Dipeptide, Polyproline, and Chignolin.


Bio: Priyank Jaini is a Research Scientist in the Brain team at Google Research in Toronto. Before joining Google, he was post-doctoral fellow with Prof. Max Welling at the University of Amsterdam. Priyank completed his PhD at the University of Waterloo under the supervision of Prof. Pascal Poupart and Prof. Yaoliang Yu working on problems on learning and inference using probabilistic generative models. 

His recent research has mainly focused on developing flexible, expressive, and efficient generative models through works on Normalizing Flows, Energy Based Models, and Diffusion Models. He is particularly interested in incorporating inductive biases in the form of symmetries through equivariances in probabilistic modelling and applying to downstream tasks like molecular generation and modelling many-body particle systems.