2B1-4 強化学習を用いたコンピュータ将棋における状態表現に関する考察(OS7:エージェントの学習・進化)

書誌事項

タイトル別名
  • 2B1-4 State Representation of Reinforcement Learning for Shogi

この論文をさがす

説明

Recently, evaluation functions for Shogi by using computer has attracted much attention due to Bonanza based on machine learning. The Bonanza has achieved one of the strongest computer players for Shogi, which often defeat human players. In order to learn the evaluation functions, Bonanza utilizes a considerable number of game records. Meanwhile, reinforcement learning can learn evaluation values based on experiences. The reinforcement learning, however, has not succeeded in learning with a large number of fine-grained feature values. In this paper, we investigate the effects of the state representations in the evaluation functions for learning results, where the state representations are derived from the ones of 'Bonanza'.

収録刊行物

詳細情報 詳細情報について

問題の指摘

ページトップへ