Asymmetric Game Tree using a Dynamic Bayesian Network

Student thesis: Master thesis (including HD thesis)

  • Esben Skov Pedersen
4. term, Computer Science, Master (Master Programme)
This project we improve mini max search by rediciting which move the opponent is most likely to take. This prediction is performed by using a dynamic bayesian network. This network has nodes that represents a number of abstract features extracted from the game. The network is trained sing EM-learning based on data from a number of comptuter players playing against each other. These computer players are defined as the weighted sum of the features mentioned before. These weights are found using a genetic approach. The system is able to improve performance of the worst strategies without increasing the search space.
Publication dateJul 2008
ID: 61073133