Hoyle is a FORR-based program that learns to play many different board games well. Its premise is that agreement among varying heuristic viewpoints is a valid decision-making principle. Hoyle minimizes search, focusing instead upon reasonable rationales and multiple learning methods. Hoyle learns during competition, and demonstrates substantial, learned expertise after relatively little training. Work with Hoyle has pioneered the ability to learn heuristic Advisors in FORR. Susan L. Epstein

Current Work

Integration of visual perception with high-level reasoning. This includes learning to associate patterns of playing pieces with outcomes, and learning game-specific Advisors (see FORR) that are based on the best of those patterns.

Key references

Epstein, S. L., Gelfand, J. and Lesniak, J. 1996. Pattern-Based Learning and Spatially-Oriented Concept Formation with a Multi-Agent, Decision-Making Expert. Computational Intelligence, 12 (1): 199-221.

Epstein, S. L. (1995). On Heuristic Reasoning, Reactivity, and Search. In Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence, 454-461. Montreal: Morgan Kaufmann.

Epstein, S. L. 1992. Prior Knowledge Strengthens Learning to Control Search in Weak Theory Domains. International Journal of Intelligent Systems, 7: 547-586.

Edmund Hoyle was an 18th century English chess player who codified the rules for all the popular board games and card games he knew. "According to Hoyle" means "playing by the rules."

[HOME]