Presentation Information
[N-1-35]Human-AI Classification of Shogi Players using 2DCNN and LSTM
◎Sakuya Watanabe1, Shingo Ando2 (1. Graduate School of Shonan Institute of Technology, 2. Shonan Institute of Technology)
Keywords:
Cheating Detection,Deep Learning,Convolutional Neural Network,Long Short-Term Memory,Shogi
Cheating by illicitly consulting AI during shogi games to gain an unfair advantage has become a serious issue. Previous studies on chess have classified whether a player is cheating by applying a 3D Convolutional Neural Network (3DCNN) to game records. While CNNs are effective at extracting local temporal features, they have difficulty capturing long-term dependencies. Therefore, with the aim of future application to cheating detection, this study proposes a method for classifying whether a player is an AI or a human from shogi game records using a combination of a 2D Convolutional Neural Network (2DCNN) and a Long Short-Term Memory (LSTM) network. Numerical experiments demonstrate the effectiveness of the proposed method by comparing the loss and classification accuracy of each approach.