DeepMind successfully developed AlphaGo, a Go program, beating top human players. However, the current Go programs based on AlphaGo still cannot be used in Go Tutoring, e.g., life-and-death problems, block connections, global board evaluations. In this technique, we solve these problems for tutoring based on CGI programs, developed by our group in the past.

Although the current AlphaGo-like programs can surpass top human players, they still cannot exactly prove and solve some common tutoring problems well, like life-and-death (L&D) questions. Here, we propose to combine deep reinforcement learning and relevance zone techniques to solve L&D problems.

此圍棋教學系統，除了可以提供解圍棋棋書中的死活問題，也可用於通常對局中的死活問題判斷，也能夠偵測棋塊間的連接性、局勢的領地計算等，結合我們過去的棋力分級系統，能夠提供完整有效的教學。