Using deep neural networks for chess position evaluation

• Main Author: Phang, Benito Yan Feng
• Other Authors: He Ying
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence
• Online Access: https://hdl.handle.net/10356/166011
• Institution: Nanyang Technological University

This paper presents a neural network, based on Giraffe by Lai, that evaluates chess positions. It relies on deep neural networks and supervised learning. The paper aims to compare and suggest improvements to the neural network by following the architecture and feature selection of Lai’s Giraffe closely. The goal of the neural network is to evaluate chess positions without manual feature selection and minimal prior knowledge. This is done by training the neural network to solve a 7-class classification problem, each class representing a different positional evaluation. The manual feature selection is subsequently replaced by a Deep Belief Network based on DeepChess to create an end-to-end machine learning based evaluation function.