Deep learning and computer chess (part 1)

In this paper, a supervised learning approach for training Artificial Neural Networks (ANNs) to evaluate chess positions has been implemented. The architecture used to build the ANN model is based on the architecture mentioned in (Matthew Lai, 2015) paper [1]. The methods that are implemented aim...

Full description

Saved in:
Bibliographic Details
Main Author: Pereddy Vijai Krishna Reddy
Other Authors: He Ying
Format: Final Year Project
Language:English
Published: Nanyang Technological University 2020
Subjects:
Online Access:https://hdl.handle.net/10356/144795
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:In this paper, a supervised learning approach for training Artificial Neural Networks (ANNs) to evaluate chess positions has been implemented. The architecture used to build the ANN model is based on the architecture mentioned in (Matthew Lai, 2015) paper [1]. The methods that are implemented aim to train ANN architecture to understand chess moves and techniques in a manner similar to how chess grandmasters would. We collected over 17,00,000 different chess game positions played by highly skilled chess players. We then used the evaluation function of stockfish, one of the strongest existing chess engines, to label the games. We extracted the positions from the games using Forsyth-Edwards notation and stored them in 2 different files which are later used for training the model after preprocessing the data. The results show how simple Multilayer Perceptrons (MLPs) perform with varying depth of the network.