Resource management in a multi-agent system by means of reinforcement learning and supervised rule learning

Abstract

In this paper two learning methods are presented: reinforcement learning and supervised rule learning. The former is a classical approach to a learning problem in multi-agent systems. The latter is a novel approach, according to the author's knowledge, which has several advantages. Both methods are used for resource management in a multi-agent system. The environment is a Fish Bank game, where agents manage fishing companies. Both learning methods are applied to generate ship allocation strategy. In this article the system architecture and learning processes are described and experimental results comparing the performance of implemented types of agents are presented. © Springer-Verlag Berlin Heidelberg 2007.