Deep Reinforcement Learning for the Game of Simulated Curling

• Item type: Symposium(1)
• 公開日: 2018-07-26
• タイトル: Deep Reinforcement Learning for the Game of Simulated Curling
• タイトル / 言語: en
• タイトル: Deep Reinforcement Learning for the Game of Simulated Curling
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: Deep reinforcement learning
• キーワード / 主題Scheme: Other
• キーワード / 主題: Machine learning
• キーワード / 主題Scheme: Other
• キーワード / 主題: Monte-Carlo tree search
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_5794
• 資源タイプ: conference paper
• 著者所属: Ulsan National Institute of Science and Technology
• 著者所属: Ulsan National Institute of Science and Technology
• 著者所属: Ulsan National Institute of Science and Technology
• 著者所属: Korea University
• 著者所属(英): en
• 著者所属(英): Ulsan National Institute of Science and Technology
• 著者所属(英): en
• 著者所属(英): Ulsan National Institute of Science and Technology
• 著者所属(英): en
• 著者所属(英): Ulsan National Institute of Science and Technology
• 著者所属(英): en
• 著者所属(英): Korea University
• 著者名: Kyowoon, Lee; Sol-A, Kim; Jaesik, Choi; Seong-Whan, Lee
• 著者名(英): Kyowoon, Lee; Sol-A, Kim; Jaesik, Choi; Seong-Whan, Lee
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Recently, deep reinforcement learning have achieved super-human performance in the deterministic games with discrete action spaces, such as Atari games and Go. However, it is still not clear how to utilize deep neural networks for discrete actions in complex games in which a minute change of an action could alter the outcome of the games dramatically. To solve this issue, we incorporates a deep neural network for learning game strategy with a kernel-based Monte Carlo tree search for finding actions from continuous space especially for the game of curling. Without any hand-crafted feature, we train our network in supervised learning manner and then reinforcement learning. Recently, our framework outperforms existing programs equipped with several hand-crafted features for curling and won in the Game AI Tournaments (GAT-2018).
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Recently, deep reinforcement learning have achieved super-human performance in the deterministic games with discrete action spaces, such as Atari games and Go. However, it is still not clear how to utilize deep neural networks for discrete actions in complex games in which a minute change of an action could alter the outcome of the games dramatically. To solve this issue, we incorporates a deep neural network for learning game strategy with a kernel-based Monte Carlo tree search for finding actions from continuous space especially for the game of curling. Without any hand-crafted feature, we train our network in supervised learning manner and then reinforcement learning. Recently, our framework outperforms existing programs equipped with several hand-crafted features for curling and won in the Game AI Tournaments (GAT-2018).
• 書誌情報: WCI2018論文集; 巻 2018, p. 5-7, 発行日 2018-07-26
• 出版者 / 言語: ja
• 出版者: 情報処理学会