Daniel Brown is a postdoc at UC Berkeley working with Anca Dragan and Ken Goldberg. He received his PhD in 2020 from the CS department at UT Austin where he was advised by Scott Niekum. He is interested in safe reward inference and inverse reinforcement learning. In particular, Daniel has worked on methods that give a robot or other autonomous agent the ability to provide high-confidence bounds on performance when learning a policy from a limited number of demonstrations, ask risk-aware queries to better resolve ambiguities and perform robust policy optimization from demonstrations, learn more efficiently from informative demonstrations, learn from ranked suboptimal demonstrations, even when rankings are unavailable, and perform fast Bayesian reward inference for visual control tasks.
Inverse Reinforcement Learning, AI Safety, Imitation Learning, and Emergent Behaviors.
