Papers by Prahalad Rajkumar
The way computer programs play strategy games is quite different from the way humans play. In per... more The way computer programs play strategy games is quite different from the way humans play. In perfect-information games like chess and checkers, a game-tree search is the core technique in a computer program’s arsenal, augmented by good evaluation functions and clever secondary strategies. In other perfect-information games such as go and clobber, there is very little intuition as to how good a position is, and consequently constructing a good evaluation function is not easy. Furthermore, go has a high branching factor. It turns out that Monte-Carlo simulations, i.e. producing repeated random samples and considering their average in making a decision, work surprisingly well in these games. In imperfect-information games such as bridge and scrabble (the latter game has inherent randomness associated with it as well), Monte-Carlo simulations once again turn out to be useful. This paper examines the use of Monte-Carlo simulations in bridge, scrabble, go, clobber, and backgammon, and re...
Uploads
Papers by Prahalad Rajkumar