1C1-6 マルチエージェント強化学習における割引率の社会適応的調節 : 生物に見られる社会的階層構造の自己組織化と神経修飾物質(機械学習)

概要

論文の詳細を見る
In the field of reinforcement learning, researchers have focused a role and dynamics of animal's neuromodulatory system, which is thought to correspond to meta-parameter of the learning system. Recently, the hypothesis is suggested and beginning to be verified that neuromodulator serotonin regulates the temporal discount factor. It is known that dominance hierarchy affects the amount of serotonin in animal society. It is also known that social hierarchy is generated from competitive behavior in a self-organizing manner. In this study, we propose self-organizing multi-agent reinforcement system which grows various temporal discount factors robust and fault-tolerant system. We show that this system has the capability adapting to environmental variability and robustness for vacancy or increasing of agents.
一般社団法人日本機械学会の論文
2011-09-01