Chang, Hyeong Soo (Sogang Univ.)

On Combining Multiple Heuristic Policies in Minimax Control

Scheduled for presentation during the Regular Session "Stochastic systems" (FrP08), Friday, July 28, 2006, 17:00−17:25, Room I

17th International Symposium on Mathematical Theory of Networks and Systems, July 24-28, 2006, Kyoto, Japan

This information is tentative and subject to change. Compiled on April 26, 2024

Abstract

This paper presents formal methods of combining available heuristic policies for solving a minimax control problem formulated as a two-person zero-sum discounted Markov games. We show that a single policy combined by the methods, from a given pair of policy sets for the controller and its opponent, improves the minimax value of the controller with an error that depends on the degree of the "local equilibrium" with respect to the policy sets. Each method can be easily implemented with Monte-Carlo simulation and can be applied in on-line manner.