対戦相手に合わせた手加減によって練習相手としての活用を目指すライバル将棋AI

書誌事項

タイトル別名
  • Rival Shogi AI as a Practice Partner Adjusted to Match the Player's Strength
公開日
2023-11-10
資源種別
conference paper
公開者
情報処理学会

説明

モンテカルロ木探索を用いる強い将棋 AI を練習相手として活用させるために提案された手加減手法の追加検証を中級者以上の AI と人間を対象に行う.対局実験の結果,手加減はできており不自然な指し手も少ないことが確認できた.しかし,対戦相手と同程度の強さへの手加減はあまりできていなかった.そこで,人間との対局の棋譜を分析しその理由を考察することで改善点を挙げた.

Additional experiments were conducted on the proposed method for controlling the strength of a strong Shogi AI using Monte Carlo tree search as a training partner for intermediate and higher Shogi AIs and human players. The experimental results showed that the proposed method can control the moves adequately and that there are only a few unnatural moves. However, it did not do much to weaken them to the same level of strength as their opponents. Therefore, we analyze the game records of the games played with human players and discuss the reasons and areas for improvement.

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