Acquisition of Best Bet for 10,000 Yen Game Using Machine Learning

Ryota Komori, Yasuaki Ito, Koji Nakano


The 10,000 yen game is a two-player game that was originally designed for a programming competition. The game has a simple rule and is classified as a two person zero sum finite definite incomplete information game in game theory. In this paper, we propose the best strategy for the 10,000 yen game. The strategy was found from results acquired by machine learning approach for 10,000 game AI. The machine learning approach uses Deep Q-Network and the learning has been performed by playing games against itself. If a player use the strategy as perfect play, the player can win the game with a probability of at least 50% against any strategies.


reinforcement learning; self learning; incomplete information game; game theory; backward induction

Full Text:



  • There are currently no refbacks.