Javascript must be enabled for the correct page display

Applying Reinforcement Learning with Monte Carlo Tree Search to The Game of Draughts

Meng, Li (2019) Applying Reinforcement Learning with Monte Carlo Tree Search to The Game of Draughts. Master's Thesis / Essay, Artificial Intelligence.


Download (693kB) | Preview
[img] Text
Restricted to Registered users only

Download (97kB)


Draughts is a popular game among many regions in the world. Applying the Monte Carlo Tree Search (MCTS) algorithm on international draughts and analyzing the playing strength are interesting research objectives. Besides the baseline MCTS algorithm similar to AlphaZero, three different variations of the MCTS algorithm are compared in our experiment. Two of them use multiple neural networks inspired by domain-specific heuristics of draughts or the multiple search tree MCTS. The hybrid algorithm is a combination of both heuristics and multiple search trees. The results of our experiments show that MCTS is indeed capable of improving its playing skills of draughts. All MCTS algorithms are capable of beating a random player, but no algorithms can stably best a player using the Alpha-Beta algorithm with depth 2. The most dominant parameters behind the bad performances are the size of the neural networks and the number of MCTS simulations. The number of input channels and the amount of training examples are also considered crucial. The results of the method with multiple search trees are the best among all MCTS algorithms, which proves that the coordination of policies and values between different search trees can improve the performance of the MCTS algorithm. On the other hand, the usage of domain-specific heuristics is considered insufficient to offset the deficit caused by decreasing the size of neural networks.

Item Type: Thesis (Master's Thesis / Essay)
Supervisor name: Wiering, M.A.
Degree programme: Artificial Intelligence
Thesis type: Master's Thesis / Essay
Language: English
Date Deposited: 29 Jan 2020 12:25
Last Modified: 29 Jan 2020 12:25

Actions (login required)

View Item View Item