CSAIL Event Calendar: Previous Series
|
R-max: A General Polynomial Time Algorithm for Near-Optimal Reinforcement Learning Speaker: Moshe Tennenholtz , Technion and Stanford University In this talk we will present R-max. R-max is an extremely simple model-based reinforcement learning algorithm which can attain near-optimal average reward in polynomial time. In R-max, the agent always maintains a complete, but possibly inaccurate model of its environment and acts based on the optimal policy derived from this model. The model is initialized in an optimistic fashion: all actions in all states return the maximal possible reward (hence the name). During execution, it is updated based on the agent's observations. R-max improves upon several previous algorithms: (1) It is simpler and more general than Kearns and Singh's $E^3$ algorithm, covering many classes of stochastic games. (2) It has a built-in mechanism for resolving the exploration vs. exploitation dilemma. (3) It formally justifies the ``optimism under uncertainty'' bias used in many RL algorithms. (4) It is simpler, more general, and more efficient than Brafman and Tennenholtz's LSG algorithm for learning in single controller stochastic games. (5) It generalizes the algorithm by Monderer and Tennenholtz for learning in repeated games. (6) It is the most efficient algorithm for learning in repeated games, to date, considerably improving and simplifying previous algorithms by Banos and by Megiddo.
See other events that are part of AI Colloquium Series Fall 2001 |
