Deep learning and computer chess

• Main Author: Low, Benedict Yu
• Other Authors: He Ying
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/153244
• Institution: Nanyang Technological University

This report presents a chess evaluation function trained using neural networks, without a priori knowledge of chess. The neural network undergoes two phases. In the first phase, it is trained using unsupervised learning to perform feature extraction. Subsequently in the second phase it undergoes supervised learning to compare two chess positions and select the more favourable one. The entire network is trained using only positions of a chess game and the game’s outcome, and no other information on chess. Although the neural network utilizes a relatively shallow network architecture by modern standards, it is capable of achieving very high accuracies and has shown great promise in its ability to identify key features that results in a favourable game. This project closely follows the implementation and concepts of DeepChess.