Re: computer-go: Mathematical Go

[Martin Mueller <mmueller@xxxxxxxxxxxxxx>]

I disagree with some of the comments about endgames that were made here.
For example, look at the experiments in my paper on "Global and Local Game
Tree Search".
My conclusion was that to solve endgames, combinatorial games are much more
efficient than anything else. And even adding just a bit of local knowledge
to a global minimax search improved things by orders of magnitude.

http://www.cs.ualberta.ca/~mmueller/publications.html


Regarding ko fights, a lot of very good work has been done by Berlekamp,
Spight, Fraser and others in Berkeley. Myself and Bill Fraser have
implemented many of these techniques, but we are still quite a bit away
from being able to build a completely automated endgame solver including ko
and all the other complications. Still, the results of combined
human-computer analysis are very impressive in my humble opinion. Look for
Fraser and Spight on

http://www.msri.org/publications/video/index0.html

	Martin