Adapting the AlphaZero Algorithm to Pawn Dama: Implementation, Training, and Performance Evaluation

Authors : Mehmet Kadir Baran, Erdem Pehlivanlar, Cem Güleç, Alperen Gönül, Muhammet Şeramet

Pages : 27-35

Doi:10.7240/jeps.1620319

View : 147 | Download : 72

Publication Date : 2025-03-25

Article Type : Research Paper

Abstract :This research uses deep reinforcement learning techniques, notably the AlphaZero algorithm, to construct an artificial intelligence system that can play Pawn Dama at a level that surpasses human players. Pawn dama, a simplified variant of Dama, is a perfect platform to explore AI\\\'s ability to think strategically and make decisions. The primary goal is to develop an AI that can use self-play to develop sophisticated strategies and comprehend the game\\\'s dynamics and regulations. The project incorporates MCTS to improve decision-making during games and uses a Convolutional Neural Network (CNN) to enhance the AI\\\'s learning capabilities. Creating an intuitive graphical user interface, putting the reinforcement learning algorithm into practice, and testing the system against real players are steps in the development process. The accomplishment of this project will contribute to the field of strategic game AI research by providing insights that may be applied to other domains and spurring further advancements in AI-driven game strategies.
Keywords : Deep Reinforcement Learning, Deep Learning, AlphaZero Algorithm, Pawn Dama, Monte Carlo Tree Search (MCTS), Convolutional Neural Network (CNN)