H.B. Keller Colloquium
Yisong Yue is a professor of Computing and Mathematical Sciences at the California Institute of Technology. He was previously a research scientist at Disney Research. Before that, he was a postdoctoral researcher in the Machine Learning Department and the iLab at Carnegie Mellon University. He received a Ph.D. from Cornell University and a B.S. from the University of Illinois at Urbana-Champaign.
This talk explores how to leverage programmatic domain knowledge to improve policy learning (which includes reinforcement & imitation learning). I will consider two aspects. First, how can we express policy classes using domain specific programming languages to yield interesting inductive biases that lead to sample-efficient learning while preserving flexibility and improving interpretability? Second, building upon the data programming paradigm in supervised learning, how can we use expert-written programs as a form of auxiliary supervision to improve the reliability of policy learning? I will present problem framings, algorithms, and experiments for two settings: efficient learning of programmatically interpretable policies, and controllable generation of behaviors.