A Multi-agent Design of a Computer Player for Nine Men's Morris Board Game using Deep Reinforcement Learning

Deep Reinforcement Learning (DRL) has been recently deployed in many artificial intelligence applications, and game players are not an exception. Nine Men's Morris is a board game that has been addressed and implemented using different AI techniques. In this paper, a multi-agent design of a computer player is introduced that represents the placing, moving, and capturing phases of the Nine Men's Morris. This design is a self-play one that knows nothing about the game other than the rules. Monte Carlo Tree Search (MCTS) is combined with Convolutional Neural Network (CNN) in each agent to provide the DNN with the training data. This combination allows the DNN to play against itself and tune its weights to predict actions. This computer player design ensures a proper training of NN without any human dataset and can compete with expert humans in the board games.