A dynamic-weight learning in multi-agent system

Abstract

This paper deals with the topic of learning in the reactive multi-agent system. The key question addressed is how several agents learn to coordinate their actions so that they could resolve a given environmental task together. In approaching this question, two constraints will have to be taken into consideration: one is the incompatibility constraint, that is, the fact that different actions may be mutually exclusive; and the other is the local information constraint, that is, the fact that typically each agent knows only a fraction of the environment. The agent is selfish on its own. In order to gain maximal group benefit from the multi-agent system (MAS) learning, this paper attempts to present an improved approach of learning in MAS. This approach, which is based on the organization-structure and dynamic-weight by considering the credit of the agent, is an improvement of the learning method for better outcomes.