Computer Igo (Go) - NOMITAN

We create a computer Igo (Go) program NOMITAN, which is based on the UCT algorithm improved with Go knowledge. We develop new procedures for improving the efficiency of NOMITAN including new search operators, and multi-processor parallelization. We intend for NOMITAN to be the world number champion computer Go player.

Combinatorial Game Theory

Games are often used as training tools for intelligence in real-world conflicts in the political, social or economical domains. We extend Combinatorial Game Theory to games with more than two players. Here it is essential to analyze possible coalitions.

Monte Carlo Proof Number Search

Proving the game theoretical outcome of a game is fundamental to games research. Proof Number Search is a popular technique that builds a proof-tree in an iterative best-first manner, by always selecting the node with the fewest child expansions. Monte Carlo techniques extend this idea by using simulation results to approximate the probability that a given node will prove the tree. This adds heuristic selection to the standard best-first pns algorithm.

Dynamics of Interacting Intelligence

We explore the dynamics of intelligence in games as a form of tension. Tension increases when the strength of players is higher, and also when the difference between player strength is smaller. Additionally, tension will be greater when the degree of fairness of a game is higher. We build models of these dynamics between intelligences.

Tutorial Search Engine Design

We collaborate with industrial partner and engineer a game playing tutor for the game of Shogi. Novice players (under 7 kyu) might easily become frustrated when playing against skillful players. Therefore, the tutor plays a more gentle game so that novice players can be entertained while they learn.