权益分类	功能权益	普通用户	{{item.name}}会员
{{category.name}}	{{benefitItem.name}}

名人に勝つコンピュータ将棋の開発

开发打败高手的电脑将棋

基本信息

批准号：
18800017
负责人：
橋本剛
金额：
$ 1.54万
依托单位：
Japan Advanced Institute of Science and Technology
依托单位国家：
日本
项目类别：
Grant-in-Aid for Young Scientists (Start-up)
财政年份：
2006
资助国家：
日本
起止时间：
2006 至 2007
项目状态：
已结题

项目摘要

名人に勝つコンピュータ将棋の実現に向けて,動的なマージンを用いるFutility Pruningの提案,部分局面n-gram法の改良,詰め将棋探索で問題となる二重カウント問題の有効な対処法としてWeak Proof Number Searchの開発などを行った。マージンを用いるFutility Pruningの研究について,これまでfutility pruningは,チェスでは有効であるが将棋では探索空間の広大さと選択的探索が主流な事もあり,有効ではないと考えられていた枝刈り技術であった。しかし近年futility pruningを将棋に用いたBonanzaが好成績を収めて以降将棋でもfutility pruningが注目されることになった。futility pruningは大雑把な評価値にマージンを持たせαβのWINDOWから十分外れている場合に枝刈りをする技術である。これまでfutility pruningで用いるマージンには定数が用いられていたが,局面状況によって適切なマージンが変化すべきである場合に対応できないという問題があった。そこで,まずfutility pruningのマージンを定数で扱っている事による問題点を解析し,その改良案として,探索毎に動的なマージンを決定するアルゴリズムを提案,実装し評価を行った。その結果,動的なマージンを用いたfutility pruningが従来のものに比べて175勝138敗1分で統計的に有意に勝ち越し,その優秀性を証明することができた。部分局面n-gram法の改良では,プロ棋士の棋譜から頻度の高い指し手列を抽出し低ノイズで探索に組み込むことに成功した。また,指し手列の出現頻度をカウントし実現確率探索の遷移確率として扱うことでより人間らしい手筋を効率的に読ませることで探索の強化に成功した。詰み探索では従来のdf-pnに代わる二重カウント問題に影響されない新たな探索法Weak Proof Number Searchを提案し,ベンチマークテストでは611手詰みの「寿」をわずか数秒で解くなど従来手法に比べて圧倒的な性能を示した。

Celebrities and celebrities have tried to make progress in the game, and in some cases, the n-gram method has been improved, and in some cases, the n-gram method has been improved, and there is a problem in the exploration of chess. There is a problem in the exploration of chess, and there is a question of how to solve the problem. There are many ways to solve the problem, and there are many ways to improve the performance of the game. In part of the situation, the n-gram method has been improved. You can use the Futility Pruning to study the computer, futility pruning the machine, and explore the mainstream of the space exploration system, as well as the technology and technology. In recent years, futility pruning has used chess Bonanza to make chess games into chess games in order to keep an eye on the chess game. Futility pruning has paid attention to chess games. In the futility pruning plant, there is a great deal of support for the production of α-β-WINDOW. When you use the futility pruning to determine the number of problems, you need to know that the situation is correct, and the situation is different. This is the first time that you have made a decision on how to make a decision on how to make a decision on a proposal and install it on the basis of the resolution of the problem, the resolution of the reform bill, the discussion of the reform bill, and the decision of the reform and improvement bill. According to the results of the test, the results showed that the results were compared with the statistics of 175-138 by using the futility pruning test, and the results showed that the performance was not satisfactory. In part of the situation, the n-gram method has been improved, and the chess player has a high degree of accuracy, which means that the hand column has been drawn out, and the exploration group has been successfully selected. This refers to the fact that you have detected the accuracy rate, the transfer rate, the transfer rate, the rate of failure, and the rate of success in the exploration of human-to-human health. The purpose of this paper is to explore the df-pn on behalf of the two-tier problem. The new method of exploration, Weak Proof Number Search, is proposed, and the performance of the device is better than that of the performance.