Learning Equivariant Energy Based Models with Equivariant Stein Variational Gradient Descent

Explore the concept of learning equivariant energy-based models using equivariant Stein variational gradient descent in this 53-minute video presentation. Delve into the efficient sampling and learning of probability densities by incorporating symmetries in probabilistic models. Discover the Equivariant Stein Variational Gradient Descent algorithm, an equivariant sampling method based on Stein's identity. Examine how equivariant SVGD incorporates symmetry information through equivariant kernels, improving sample complexity and quality. Learn about equivariant energy-based models for modeling invariant densities using contrastive divergence. Investigate applications in regression, classification tasks for image datasets, many-body particle systems, and molecular structure generation. Follow along as the presenter covers motivations, equivariant kernels, equivariant EBMs, particle systems, de novo molecular design, and protein folding, concluding with a Q&A session.

- Intro
- Motivations and Overview
- Incorporating Equivariance Using an Equivariant Kernel Equivariant SVGD
- Equivariant EBMs
- Many-Body Particle Systems
- De novo Molecular Design
- Protein Folding
- Q+A