Home // ADVCOMP 2016, The Tenth International Conference on Advanced Engineering Computing and Applications in Sciences // View article

Learning Method by Sharing Activity Logs in Multiagent Environment

Authors:
Keinosuke Matsumoto
Takuya Gohara
Naoki Mori

Keywords: machine learning; Q-learning; sharing of activity history; agents; hunter game

Abstract:
Applications of multiagent systems are expected from the point of view of the parallel and distributed processing. Reinforcement learning is used as an implementation method for learning agents’ actions. However, the problem is that, the higher the number of agents to deal with, the slower the speed of learning becomes. To solve this problem, this paper proposes a new reinforcement learning method that can learn quickly by using past actions of its own and of other agents. Agents can learn good actions in the early stage of learning by this method. However, if agents keep learning, learning efficiency will deteriorate. The method controls to reduce effects of other agents’ actions in the later stage of learning. In experiments, agents learned good actions in various environments. Thus, the success of the proposed method was verified.

Pages: 71 to 76

Copyright: Copyright (c) IARIA, 2016

Publication date: October 9, 2016

Published in: conference

ISSN: 2308-4499

ISBN: 978-1-61208-506-7

Location: Venice, Italy

Dates: from October 9, 2016 to October 13, 2016