抄録
Computer Shogi made rapid progress due to the establishment of a method for learning evaluation functions by self-play. In this paper, we applied this method to Mini-Shogi and verified the effect. Specifically, we developed a Mini-Shogi program based on the computer Shogi engine, YaneuraOu, and learned a non-linear evaluation function using a neural network. Our program won all competitions in which we participated in 2020. Moreover, the experimental results suggest a second-move advantage in Mini-Shogi.