Re: computer-go: FPGA

[Don Dailey <drd@xxxxxxxxxxxxxxxxxxxxx>]

> 
> Exactly,
> 
> for nodes a seconds nothing beats minimax then.
> for search depth nothing beats selectivity. In chess
> the nullmove gets applied bigtime, basically this was the big progress
> in the 90s in computerchess. Last few years progress is realizing
> that knowledge and good preparing like books, a good endgame (that
> would be a good openingsplay in GO) is deciding the game.
> 
> So in contradiction to common believe: the nullmove selectivity sure
> isn't a brute force way of searching. When using nullmove with a 
> Go program you also search a lot of plies extra, as it has the habit
> to cutoff lines bigtime. Especially for got it will kick butt. Note
> there must be a lot of other work to limit the branching factor in
> the openings stage, but it's already a cool form of selectivity which
> with increasing computing power will kick more and more butt.


I was thinking about null move pruning too.  As far as  I can tell, GO
honors  the null move assumption  much stronger  than chess does which
means it  should work even   better than in  chess.   It seems  like a
natural thing  to try and  could  be experimented on  very effectively
with small boards.

Don