作成者 |
|
|
本文言語 |
|
出版者 |
|
|
発行日 |
|
収録物名 |
|
|
巻 |
|
開始ページ |
|
終了ページ |
|
会議情報 |
|
出版タイプ |
|
アクセス権 |
|
権利関係 |
|
|
関連DOI |
|
概要 |
In recent years, game AI has been remarkably evolved. Some game AIs have outperformed top human players for complete information games such as Chess, Shogi, and Go. Compared with AI for perfect inform...ation games, game AI is not so strong for imperfect information games such as Poker and Mahjong. We research and development Ma h jong AI using machine learning method. In this paper, we divide the internal Mahjong AI function into two parts: supervised learning and reinforcement learning. In the supervised learning part, the AI learns the choices of tiles using the top ranked human players in Tenho's game records. In the second part, the AI is reinforced sel ection function during many Mahjong games against three AIs created in the first step supervised learning続きを見る
|