2017 年 137 巻 5 号 p. 759-767
The learning automaton (LA) team model has been proposed as one method for modeling multi-agent systems. It is modeled as a non-cooperative game of learning automata. In this model, each LA operates independently from each other, and there exists a Nash equilibrium, i.e. the existance of an optimal mixed strategy in the mixed strategy space of the game has been proven. However, for modelling multi-agent systems more generally, the information exchange among agents and the acquisition of cooperative behaviors such as the formation of autonomous community are required. In this paper, in order to complement the LA team model, we propose a new LA team model with some fully or partially collaborative learning behaviors. In this new model, each automaton performs reinforcement learning process in order to identify random environments exchanging information with its adjacent automata. Several computer simulations indicate the availability of the proposed model.
J-STAGEがリニューアルされました! https://www.jstage.jst.go.jp/browse/-char/ja/