CDS Lecture: Hyeong Soo Chang, "Reinforcement Learning"
Monday, June 19, 2006
2168 A.V. Williams Building
301 405 6576
Control and Dynamical Systems Invited Lecture Series
Reinforcement Learning with Supervision by Combining Multiple Learnings and Expert Advices
Hyeong Soo Chang
Former ISR Research Assistant
In this talk, we provide a formal coherent learning framework where reinforcement learning is combined with multiple learning and expert advice toward accelerating convergence speed of learning. Our approach is simply to use a nonstationary "potential-based reinforcement function" for shaping the reinforcement signal given to the learning "base-agent." The base-agent employs SARSA(0) or adaptive asynchronous value iteration (VI), and the supervised inputs to the base-agent from the "subagents" involved with other parallel independent reinforcement learnings and if available, from experts are "merged" into the potential-based reinforcement function value and the value is put into the update equation of SARSA(0) for the Q-function estimate or of adaptive asynchronous VI for the optimal value function estimate. The resulting SARSA(0) and adaptive asynchronous VI converge to an optimal policy, respectively.
Hyeong Soo Chang is an Assistant Professor in Computer Science at Sogang University in Seoul, Korea. He received his Ph.D. from Purdue University and was a Research Assistant at the Institute for Systems Research, where he worked with Professors Steven Marcus, Michael Fu and Mark Shayman. His research interests are in systems and controls, reinforcement learning, artificial intelligence and optimal theory controls.
P.S. Krishnaprasad and Michael Fu