On Wed, Jun 29, 2011 at 1:30 AM, Peter Drake <[email protected]> wrote: > It doesn't beat RAVE, but it's an interesting result. Our paper will appear > at the International Conference on Artificial Intelligence (ICAI) in Las > Vegas: > > https://webdisk.lclark.edu/drake/publications/sylvester-icai-2011.pdf > > Peter Drake > http://www.lclark.edu/~drake/ > > > > _______________________________________________ > Computer-go mailing list > [email protected] > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go >
Wow, what a title, next someone will write that Chess and Go are no-brainers... Oh wait, that already happened :-) I hope you are aware that some strong MCTS programs use (at least) a factor hundred less playouts to break even with gnugo. In fact, to get to 50% they don't even need a tree at all... (so UCT is perhaps not really that relevant at these levels) Anyway, the paper shows that it is an interesting idea to try to generalize over multiple states. However, this shouldn't really be a big surprise; replacing a lookup table by a general function approximator is know to work well in many domains (and I wouldn't be surprised if many programs use this idea). Also it doesn't exclude you from using UCT/MCTS, e.g., you can just use the approximations to set priors in the tree. Erik BTW Long before UCT, Levente Kocsis showed that linear prediction also works quite well in Chess. _______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
