J. Denzinger and S. Ennis (Canada)
Intelligent agents, learning/adaptation, cooperative behav ior, Genetic Algorithms
This paper deals with handling the situation when in an ex perienced team of agents one team member is replaced by a new agent with different abilities. We extend earlier work that focused on how the new agent can make use of the strategy of the old agent as seed strategy in an evolutionary learning approach with prototypical situation-action pairs and the nearest-neighbor rule as agent architecture. We now allow also the remaining team members to adapt their behavior using their old strategies as seed strategies in evo lutionary on-line learning. After giving the new agent some time for a basic adaptation, the remaining team members take turns to adapt themselves. Our experimental evaluation shows that this extension allows us to deal with situations where the difference in abilities is too big for the previous approach to succeed. We also can report examples in which additional abilities of the new agent are made use of to get an overall better team performance than with the old agent.
Important Links:
Go Back
