Training Neural Networks to Play Backgammon Variants Using Reinforcement Learning
Backgammon is a board game that has been studied considerably by computer scientists. Apart from standard backgammon, several yet unexplored variants of the game exist, which use the same board, number of checkers, and dice but may have different rules for moving the checkers, starting positions and movement direction. This paper studies two popular variants in Greece and neighboring countries, named Fevga and Plakoto. Using reinforcement learning and Neural Network function approximation we train agents that learn a game position evaluation function for these games. We show that the resulting agents significantly outperform the open-source program Tavli3D.
KeywordsNeural Network Reinforcement Learning Input Unit Board Game Training Neural Network
Unable to display preview. Download preview PDF.
- 1.BackGammon Variants, http://www.bkgm.com/variants
- 4.Schaeffer, J., Hlynka, M., Vili, J.: Temporal Difference Learning Applied to a High-Performance Game-Playing Program. In: Proceedings IJCAI, pp. 529–534 (2001)Google Scholar
- 5.Sutton, R.S.: Learning to predict by the methods of temporal differences. Machine Learning, 9–44 (1988) Google Scholar
- 6.Sutton, R.S., Barto, A.G.: Reinforcement Learning: An Indroduction. MIT Press, Cambridge (1998)Google Scholar
- 7.Szepesvári, C.: Algorithms for Reinforcement Learning (Electronic Draft Version) (June 2010), http://www.sztaki.hu/~szcsaba/papers/RLAlgsInMDPs-lecture.pdf
- 12.Veness, J., Silver, D., Uther, W., Blair, A.: Bootstrapping from Game Tree Search. Advances in Neural Information Processing Systems 22, 1937–1945 (2009)Google Scholar