1C1-3 他者評価強化学習における合理的集団行動獲得のための評価対象選択(1C1 強化学習)

概要

論文の詳細を見る
This paper deal with observation reduction for acquisition rational collective behavior with mutual reinforcement learning. Mutual Reinforcement Learning (MRL), which applies external evaluation in multiagent systems is based on external evaluation network among interdependent agents. MRL aims at facilitating learning and acquisition of a rational group behavior based on partial observation. However, it is not realistic to acquire the behavior in the case that agents interdepend on many other agents due to explosion of state space. To solve this problem, the method that agents stop observing interdependency which do not effect learning results is applied. Simulation results showed that proposed method decreases observation appropriately.
社団法人日本機械学会の論文
2006-09-25