Decision Tree Learning in General Game Playing

Xinxin Sheng and David Thuente


general game playing, agent, sub-goal, decision tree


This paper introduces a General Game Playing agent that automatically adapts to play different kinds of games without human intervention in a multi-agent system. By using machine learning techniques in heuristic search, especially decision tree learning, the agent is able to identify common winning features in a given game and evaluate a game state with many features as an entirety. We provide four distinctly different game examples to show how the agent’s performance has been improved by introducing decision tree learning. These four games are the strategy game Mini-Chess, the territory taking game Connect Four, the crossing Chinese Checker, and the three player simultaneous game Farmer.

Important Links:

Go Back