RE: computer-go: Learning from existing games

[Piotr Kaminski <piotr@xxxxxxxxxxxx>]

On the subject of learning by "judging" moves, I recommend two articles:

Bernd Brugmann:  Monte Carlo Go (my reformatted version is at
http://www.ideanest.com/research/documents/Monte%20Carlo%20Go.pdf)

Nicol Schraudolph et al.:  Learning to Evaluate Go Positions Via Temporal
Difference Methods (http://www.inf.ethz.ch/~schraudo/pubs/gochap.pdf)

I've tried my hand at the Monte Carlo approach and have met with some
success, but haven't even been able to fully reproduce Bernd's results, much
less produce a Go player of reasonable caliber.  I haven't published my
code/report, but please ask me in private if you're interested.  Thanks.

		-- P.

 --
  Piotr Kaminski, piotr@xxxxxxxxxxxxxxxxx, http://www.ideanest.com/
  It's the heart afraid of breaking that never learns to dance.