Testing of several function approximations for a partially observable card game

• Item type: SIG Technical Reports(1)
• 公開日: 2014-05-30
• タイトル: Testing of several function approximations for a partially observable card game
• タイトル / 言語: en
• タイトル: Testing of several function approximations for a partially observable card game
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: 分析
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: Hokkaido University
• 著者所属: Hokkaido University
• 著者所属: Hokkaido University
• 著者所属: Hokkaido University
• 著者所属(英): en
• 著者所属(英): Hokkaido University
• 著者所属(英): en
• 著者所属(英): Hokkaido University
• 著者所属(英): en
• 著者所属(英): Hokkaido University
• 著者所属(英): en
• 著者所属(英): Hokkaido University
• 著者名: JanaCathrinBackhus; Hidetoshi, Nonaka; Takeshi, Yoshikawa; Masanori, Sugimoto
• 著者名(英): Jana, CathrinBackhus; Hidetoshi, Nonaka; Takeshi, Yoshikawa; Masanori, Sugimoto
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Game playing is considered as an important research topic in the field of Artificial Intelligence, because its features resemble those of real world problems. Research aims contain creating strong computer players for the entertainment of humans and creating human-like learning algorithms. Wizard is a partially observable multi-player card game that is structured in rounds with different grades of information imperfectness. A computer player using reinforcement learning algorithm is considered for the game. Function approximation plays an important role in reinforcement learning. In our research, we tested different function approximation approaches for the game Wizard under the aspect of having different grades of information imperfectness in the game rounds. Experimental results show that the success of the function approximations is differing a lot for the game rounds as the game winning rate of the learning agent is differing between 0% and 26.74%.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Game playing is considered as an important research topic in the field of Artificial Intelligence, because its features resemble those of real world problems. Research aims contain creating strong computer players for the entertainment of humans and creating human-like learning algorithms. Wizard is a partially observable multi-player card game that is structured in rounds with different grades of information imperfectness. A computer player using reinforcement learning algorithm is considered for the game. Function approximation plays an important role in reinforcement learning. In our research, we tested different function approximation approaches for the game Wizard under the aspect of having different grades of information imperfectness in the game rounds. Experimental results show that the success of the function approximations is differing a lot for the game rounds as the game winning rate of the learning agent is differing between 0% and 26.74%.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AA12049625
• 書誌情報: 研究報告エンタテインメントコンピューティング（EC）; 巻 2014-EC-32, 号 14, p. 1-6, 発行日 2014-05-30
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会