Timezone: »

Dual Policy Iteration
Wen Sun · Geoffrey Gordon · Byron Boots · J. Bagnell

Wed Dec 05 02:00 PM -- 04:00 PM (PST) @ Room 517 AB #124

Recently, a novel class of Approximate Policy Iteration (API) algorithms have demonstrated impressive practical performance (e.g., ExIt from [1], AlphaGo-Zero from [2]). This new family of algorithms maintains, and alternately optimizes, two policies: a fast, reactive policy (e.g., a deep neural network) deployed at test time, and a slow, non-reactive policy (e.g., Tree Search), that can plan multiple steps ahead. The reactive policy is updated under supervision from the non-reactive policy, while the non-reactive policy is improved with guidance from the reactive policy. In this work we study this Dual Policy Iteration (DPI) strategy in an alternating optimization framework and provide a convergence analysis that extends existing API theory. We also develop a special instance of this framework which reduces the update of non-reactive policies to model-based optimal control using learned local models, and provides a theoretically sound way of unifying model-free and model-based RL approaches with unknown dynamics. We demonstrate the efficacy of our approach on various continuous control Markov Decision Processes.

Author Information

Wen Sun (Carnegie Mellon University)
Geoffrey Gordon (MSR Montréal & CMU)

Dr. Gordon is an Associate Research Professor in the Department of Machine Learning at Carnegie Mellon University, and co-director of the Department's Ph. D. program. He works on multi-robot systems, statistical machine learning, game theory, and planning in probabilistic, adversarial, and general-sum domains. His previous appointments include Visiting Professor at the Stanford Computer Science Department and Principal Scientist at Burning Glass Technologies in San Diego. Dr. Gordon received his B.A. in Computer Science from Cornell University in 1991, and his Ph.D. in Computer Science from Carnegie Mellon University in 1999.

Byron Boots (Georgia Tech / Google Brain)
J. Bagnell (Carnegie Mellon University)

More from the Same Authors