基于假设检验理论及动态随机化的藏式围棋博弈研究

Tibetan go is a traditional board game for Tibetan people. With the changes of history, Tibetan go is on the verge of extinction. The chess play record is scare. Unique chess regulations make it more obvious than go with play phase characteristics. Present various tree searching and situation evaluation algorithms, including algorithm based on deep learning Alpha Go use, can not apply to Tibetan Go.. The project proposes tree searching algorithm based on hypothesis test theory and situation evaluation algorithm based on dynamic randomization for Tibetan go. Besides, we should develop high performance Tibetan go system using the proposed algorithms, analyze the algorithms by experiments. The novel searching algorithm based on hypothesis test theory can improve pruning efficiency of search tree, accelerate process to get optimal move sequence. The situation evaluation algorithm based on dynamic randomization can improve evaluation accuracy. . The research has important theoretical significance on computer game tree search algorithm and situation evaluation algorithm in the case of the absence of massive play record data. It also has bright application prospect on the development of high performance Tibetan go and other computer game systems. It also can help protect, spread and develop Tibetan go culture which is on the brink of extinction, promote the prosperity of national traditional culture.

藏式围棋棋谱稀少，独特的棋规使其比围棋具有更明显的阶段性特征。目前各种类型的博弈算法，包括Alpha Go使用的基于深度学习的博弈算法，都无法适用于藏式围棋。.本项目拟提出基于假设检验理论的藏式围棋搜索算法，以及动态随机化局面评估算法，进而使用所提出的算法设计实现高性能的藏式围棋博弈软件，并通过实验分析算法性能。本项目创新性地将假设检验理论及动态随机化应用于藏式围棋博弈，拟提出的搜索算法可提高博弈树剪枝效率，加快最优着法序列生成过程；拟提出的局面评估算法能够提高博弈树节点评估准确性。.该项研究在缺乏海量数据情况下的计算机博弈搜索算法和局面评估算法方面具有重要的理论意义，在藏式围棋等高性能博弈系统开发方面具有良好的应用前景。该项研究还有利于保护、传承及发展藏式围棋这一濒临失传的藏族特色文化，促进民族传统文化的繁荣。

本项目在围棋人工智能等取得战胜人类高手、藏式围棋棋谱稀少且面临灭绝亟待保护的背景下开展研究的。主要研究了基于假设检验理论和动态随机化的藏式围棋博弈研究，将假设检验理论应用于藏式围棋博弈的不同阶段，用于确定其进度阈值，将动态随机化应用于分层局面评估中，提高搜索和评估效果。.本研究提出的基于假设检验理论的搜索方法不仅能够应用于藏式围棋，也能够应用于围棋等一类游戏中，具有通用性。除此之外，还提出了基于强化学习和深度学习的藏棋博弈研究。开发了高水平的藏棋博弈软件。发表了9篇SCI论文,7篇核心期刊或者EI检索论文，包括1篇国际顶尖会议论文。培养了7名硕士研究生。