“School of Cognitive Sciences”
Back to Papers HomeBack to Papers of School of Cognitive Sciences
| Paper IPM / Cognitive Sciences / 7514 |
|
||||||
| Abstract: | |||||||
|
Heterogeneity is more the rule than an exception when it comes to cooperative learning in multi-agent systems (MAS). This fact, accounts for an emerging trend among MAS researchers. If heterogeneity is not handled properly, cooperative learning can be inefficient, even impossible. In this paper, two heterogeneous Q-learning agents that cooperate to learn are considered. The heterogeneity is assumed in their actions, which can be caused by different action orders. The problem becomes more complicated when one agent has an extra action. The extra action affects the values of Q-cells in a Q-table. For an agent without the extra action, we propose two cooperation methods which adapt values of Q-cells, and make them meaningful and usable. Simulation results in different case studies justify the superior performance of both proposed cooperation methods, as compared to that of individual learning.
Download TeX format |
|||||||
| back to top | |||||||
