CPSC 7430: Topics in flow-based generative modeling and optimal transport

Course Description: This course studies the mathematical underpinnings of modern problems in data science. For example, how can we use the expressive power of neural networks to generate new data from a finite set of training data? We will also investigate algorithms used to efficiently sample approximations of posterior distribution (e.g., variational inference), sampling under tilts, and the case of temporal data, or data lying on a Riemannian manifold. Many if not all of these learning problems can be reduced to an algorithm that leverages ideas from flow-based generative modeling and optimal transport. Specifically, in dissecting how probability distributions "move", how maps and flows can be efficiently learned given constraints on the data, and how optimality principles shape practical algorithms.

Instructor: Aram-Alexandre Pooladian