Definition
TD-Gammon is a world-champion strength backgammon program developed by Gerald Tesauro. Its development relied heavily on machine learning techniques, in particular on Temporal-Difference Learning. Contrary to successful game programs in domains such as chess, which can easily out-search their human opponents but still trail these ability of estimating the positional merits of the current board configuration, TD-Gammon was able to excel in backgammon for the same reasons that humans play well: its grasp of the positional strengths and weaknesses was excellent. In 1998, it lost a 100-game competition against the world champion with only 8 points. Its sometimes unconventional but very solid evaluation of certain opening strategies had a strong impact on the backgammon community and was soon adapted by professional players.
Description of the Learning System
TD-Gammonis a conventional game-playing program that uses very shallow search (the first versions only searched one ply)...
This is a preview of subscription content, log in via an institution to check access.
References
Tesauro, G. (1989). Connectionist learning of expert preferences by comparison training. In D. Touretzky (Ed.), Proceedings of the advances in neural information processing systems 1 (NIPS-88) (pp. 99–106). San Francisco: Morgan Kaufmann.
Tesauro, G. (1992). Practical issues in temporal difference learning. Machine Learning, 8, 257–278. http://mlis.www.wkap.nl/mach/abstracts/absv8p257.htm.
Tesauro, G. (1995). Temporal difference learning and TD-Gammon. Communications of the ACM, 38(3), 58–68. http://www.research.ibm.com/massdist/tdl.html.
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2011 Springer Science+Business Media, LLC
About this entry
Cite this entry
(2011). TD-Gammon. In: Sammut, C., Webb, G.I. (eds) Encyclopedia of Machine Learning. Springer, Boston, MA. https://doi.org/10.1007/978-0-387-30164-8_813
Download citation
DOI: https://doi.org/10.1007/978-0-387-30164-8_813
Publisher Name: Springer, Boston, MA
Print ISBN: 978-0-387-30768-8
Online ISBN: 978-0-387-30164-8
eBook Packages: Computer ScienceReference Module Computer Science and Engineering