Xiaocheng Luan, Yun Peng, and Timothy Finin, University of Maryland Baltimore County
One of the common ways to achieve interoperability among the autonomous agents is to use a broker agent (or a facilitator). Simple broker agents provide match-making services based on the capability information volunteered by individual agents and the (recommendation) request. The problem is, even with a very good agent capability description language and a powerful match-making mechanism, if the actual capability information volunteered by each individual agent is not accurate, it won’t be of much help. Given that the autonomous agents might be written by different people, at different time, and for different purpose, this is likely to occur. This work is an attempt to solve such problems by incorporating learning into broker agents so that the broker agents can capture more accurate information about the capabilities of individual agents.