Machine learning: Gaming bot could aid working with incomplete data

Assistant Professor Xia “Ben” Hu and his graduate students in the Department of Computer Science and Engineering are collaborating with JJ World, a leading gaming company in China, to develop a gaming bot for one of the most popular Asian card games, Sheng ji.

The new bot will not only provide a better gaming experience for users by matching their skill set, but will also advance the field of machine learning in environments with incomplete information.

Previous research has already been done on games such as Go, an abstract strategy board game. These games have a set number of moves and player strategies, enabling a computer to compute and search for the next steps. The available information is complete.

“Sheng ji is a typical game with incomplete information,” Hu said. “When a player makes a decision, she needs to consider the hidden cards from the other three players. To tackle the information asymmetry we need to implement a model to predict the hidden cards from the other players.”

Hu said the system is expected to be divided into three modules: the memory module, the prediction module and the decision module. The memory module, as the game progresses, will record all the known information that serves as the input of hands prediction in the prediction module. The outputs of the memory module and prediction will be further used for decision making in the decision module.