電腦圍棋近年來棋力突飛猛進，九路圍棋程式已經可以擊敗人類，十年內十九路圍棋可以挑戰職業棋士。目前設計方法主流為蒙地卡羅樹狀搜尋演算法(MCTS)，此方法非常倚賴事前知識，而棋型為事前知識中最重要的部分。棋型是指棋子在棋盤上的分佈狀況，乃圍棋知識的濃縮，在對局雙方進行攻防之際，棋型可以幫助棋手迅速排除無用的著手，為決定棋力高低的關鍵因素。在實際應用上，必須發展更有效率的分析比對方法與自大量棋譜中擷取棋型的方法，這也是本計畫的主要研究課題。本計畫成果將對圍棋程式棋力的提升有相當大的幫助，發展出的技術將可幫助圍棋程式的進展。同時也將開發出一些樣式比對演算法及圖形理論，預計將有兩篇以上的相關論文發表。
Recently, computer Go has advanced by leaps and bounds. It can beat human top player on 9*9 Go and challenge the Go pro of 19*19 Go in ten years. At present, the most commonly design method is the Carlo Tree Search MCT, which is a method deeply depending on prior knowledge, and the board situation is the most important part of prior knowledge. Go pattern is the special distribution of stones on the chessboard. These patterns contain the critical knowledge of Go and could help the player to filter the useless moves. We need to develop a more efficient method of analysis and comparison in addition to a way choosing board situations in the amount of Go knowledge patterns, which is one of the main studying subjects in this program. We believe the result of this project could help the national gaming industry because it could elevate the on-line game intelligence. The constructed system could be combined with the existed computer Go program and the hybrid system is more powerful. During the development of the system, we will complete more than two papers for describing the proposed pattern matching algorithms and the related graph theory.
