Electrical Engineering Seminar

Friday, November 7 at 11am

SEC LL2.224

"Representation-based Reinforcement Learning: Bridging World Model and Policy via Spectral Representations"

Speaker: Bo Dai, Assistant Professor, Georgia Tech, Staff Research Scientist, Google DeepMind

Reinforcement learning often faces a trade-off between world model flexibility and computational tractability. Linear dynamics induces tractable planning and exploration but with huge approximation gap, while flexible models can capture complex dynamics in world model and policy but often introduce nonlinearity, making planning and exploration challenging. In this talk, we explore how representation learning can help overcome this dilemma and bridge world model to policy. We present algorithms that extract flexible representations, which enabling practical and provable planning and exploration from flexible world model. We provide theoretical guarantees our algorithm for RL in MDP and POMDP settings, and empirical results demonstrating the superiority of our approach on various benchmarks.

Bio: Bo Dai is an assistant professor in Georgia Tech and a staff research scientist in Google DeepMind. He obtained his Ph.D. from Georgia Tech. His research interest lies in developing principled and practical algorithms for reinforcement learning and generative models. He regularly serves as area chair or senior program committee member at major AI/ML conferences such as ICML, NeurIPS, AISTATS, and ICLR.