In this talk I focus on just two contributions from my thesis: the WoLF principle and the GraWoLF algorithm. I show that the WoLF variable learning rate causes learning to converge to optimal responses in settings of simultaneous learning. I demonstrate this converging effect both theoretically in a subclass of single-state games and empirically in a variety of multiple-state domains. I then describe GraWoLF, a combination of policy gradient techniques and the WoLF principle. I show compelling results of applying this algorithm to a card game with an intractably large state space as well as an adversarial robot task. These results demonstrate that WoLF-based algorithms can effectively learn in the presence of other learning agents, and do so even in complex tasks with limited agents.

Speaker Bio:

Michael Bowling is an assistant professor at the University of Alberta in Edmonton, Canada. He recently completed his Ph.D. in Computer Science at Carnegie Mellon University in the area of artificial intelligence. He has been actively involved in the emerging field of multiagent learning. In addition to his thesis contributions, he has given several invited talks on multiagent learning and co-presented a tutorial at last year's International Joint Conference on Artificial Intelligence. He has also participated extensively in the RoboCup initiative, an international, autonomous, robot soccer competition. In 1998, he was the student leader of the Carnegie Mellon small-size robot team, which won the RoboCup world championship. His research interests include multiagent learning, multiagent planning, game theory, reinforcement learning, mobile robots, and computer game environments.

<< Back