商品情報にスキップ
1 1

Pruning of Redundant Information to Improve Performance for Agent Control in A Changing Environment

Pruning of Redundant Information to Improve Performance for Agent Control in A Changing Environment

通常価格 ¥770 JPY
通常価格 セール価格 ¥770 JPY
セール 売り切れ
税込

カテゴリ: 論文誌(論文単位)

グループ名: 【C】電子・情報・システム部門

発行日: 2012/11/01

タイトル(英語): Pruning of Redundant Information to Improve Performance for Agent Control in A Changing Environment

著者名: Lutao Wang (Graduate School of Information, Production and Systems, Waseda University), Shingo Mabu (Graduate School of Information, Production and Systems, Waseda University), Wei Xu (Graduate School of Information, Production and Systems, Waseda Univers

著者名(英語): Lutao Wang (Graduate School of Information, Production and Systems, Waseda University), Shingo Mabu (Graduate School of Information, Production and Systems, Waseda University), Wei Xu (Graduate School of Information, Production and Systems, Waseda University), Kotaro Hirasawa (Graduate School of Information, Production and Systems, Waseda University)

キーワード: Evolutionary Computation,Reinforcement Learning,Agent Control,GNP,Tile-world

要約(英語): Genetic Network Programming(GNP) is a new evolutionary computation method which is competent for many agent control problems. However, some redundant nodes exist in the program of GNP, which can easily cause the over-fitting problem and decrease its performance. In order to prune these nodes, a new method named "Credit GNP" is proposed in this paper. The novelties are, firstly, Credit GNP has a unique structure, where each node has an additional "credit branch" which can skip the redundant nodes. Secondly, Credit GNP combines evolution and reinforcement learning, i.e., off-line evolution and on-line learning to prune the redundant nodes. Which node to prune and how many nodes to prune are determined automatically considering different environments. Simulation results on the Tile-world problem show that Credit GNP could generate not only better programs, but also more general rules for agent control. The superiority of the proposed method over the conventional GNP, GP and standard reinforcement learning is proved.

本誌: 電気学会論文誌C(電子・情報・システム部門誌) Vol.132 No.11 (2012) 特集:電気関係学会関西連合大会

本誌掲載ページ: 1829-1839 p

原稿種別: 論文/英語

電子版へのリンク: https://www.jstage.jst.go.jp/article/ieejeiss/132/11/132_1829/_article/-char/ja/

販売タイプ
書籍サイズ
ページ数
詳細を表示する