M. Harré and T. Bossomaier (Australia)
Multi-Agent Systems, Reinforcement Learning, Artificial Economies
Recently there has been an increased interest in the use of reinforcement learning methods for studying adaptive behaviour in multi-agent systems (MAS). This interest is based, in part at least, on the versatility of these systems in which the microscopic behaviour of the system, the be haviour of individual agents, changes according to feed back provided by the environment in which the agents are playing [1]. In this paper we compare two different types of adaptive MAS behaviour, one using a Boltzmann type rein forcement algorithm where an agent is selected from a dis tribution of agents based on past performance and the other uses an artificial economy where agents bid in an auction for the rights to make a move in the game. In both cases they are learning to play an implementation of the game of Dots and Boxes. We show that learning is more difficult in the artificial economy model than in the Boltzmann model, but with an adjustment to the economic model it is able to perform on a par with the Boltzmann method
Important Links:
Go Back
