Learning to Play the Game of Go
James R. Foulds · 2006
The problem of creating a successful artificial intelligence game playing program for the game of Go represents an important milestone in the history of computer science, and provides an interesting domain for the development of both new and existing problem-solving methods. In particular, the problem of Go can be used as a benchmark for machine learning techniques. Most commercial Go playing programs use rule-based expert systems, relying heavily on manually entered domain knowledge. Due to the complexity of strategy possible in the game, these programs can only play at an amateur level of skill. A more recent approach is to apply machine learning to the problem. Machine learning-based Go playing systems are currently weaker than the rule-based programs, but this is still an active area of research. This project compares the performance of an extensive set of supervised machine learning algorithms in the context of learning from a set of features generated from the common fate graph – a graph representation of a Go playing