Training Artificial Intelligence Agents to Play a Family of Combinatorial Games

This project explores the application of various artificial intelligence techniques in developing strategy for combinatorial games. A family of deterministic 2-player games is played on m-by-n grids, potentially with some cells removed. Players take turns placing pieces on the board until the board is filled, then sequences of pieces are scored based on length. AI agents are trained to play the game using methods including tabular reinforcement learning, evaluation of N-tuples of grid squares, and development of genetic training methods with artificial neural networks. These agents are trained against a variety of non-learning agents, then evaluated against both non-learning agents and one another to assess the quality of decision-making. Additionally, this family of combinatorial games is studied using the theoretical foundations of game theory and strategy.