Development of a Chess Agent with Minimal Lookahead

Abstract

Currently there are lots of works in board games like Go and Chess, these uses algorithms to take decisions based on exploring the game-tree generated using heuristics, evaluations functions and more recently using complex neural networks. In this work, a chess agent was implemented using Monte Carlo Tree Search using an evaluation function that was implemented using a convolutional neural network trained with supervised learning. The aim of this work is to show that it is not necessary a deep lookahead technique to get good predictions and even it is not necessary complex neural network, so future works can approach better results using the same hardware. The neural network was trained using a chess database and the agent implemented was tested using a position test.

Keywords: Chess Agent, Chess Engine, Convolutional Neural networks (CNN), Evaluation Function, Monte Carlo Tree Search (MCTS)