Deep learning and computer chess (part 2)

Abstract

Monte Carlo Tree Search (MCTS) is a probabilistic search algorithm that uses random simulations to build a search tree. It is computationally expensive, and the quality of the results correlate with the effectiveness of the algorithm. This goal of this project was to develop enhancements to improve the effectiveness of MCTS-based chess engines. For that purpose, a chess engine running on the basic MCTS algorithm was built and used as the base engine. After a review of the literature to date, the enhancements early playout termination (EPT), score bonus, MCTS-Solver, biased and corrective simulation were chosen and added to the base engine in stages. Results showed that the enhancements EPT, score bonus, MCTS-Solver and biased simulation successfully improved the performance of the engine, while corrective simulation was ineffective. The greatest improvement was shown by score bonus, which provided an ELO-Rating increase of 191. This demonstrates that with enhancements, MCTS-based chess engines can achieve significant improvements in performance and win games off beginner level engines. The success of these enhancements shows the potential for further development to create stronger MCTS-based chess programs.