Practical Issues in Temporal Difference Learning

[Volker Pittlik]

Geschrieben von:/Posted by: Volker Pittlik at 18 September 2004 14:02:55:

Just stumbled on this:
http://www.research.ibm.com/massive/pitdl.pdf
Possibly interesting...
Volker

[Andrew Fan]

Geschrieben von:/Posted by: Andrew Fan at 18 September 2004 15:18:07:
Als Antwort auf:/In reply to: Practical Issues in Temporal Difference Learning geschrieben von:/posted by: Volker Pittlik at 18 September 2004 14:02:55:


So, this neural net "engine" has been learning to play backgammon since 1992 or even earlier? I wonder how good it is now in 2004.
Anyone care to try it on chess? I am no neural net fan (pun intended).

Andrew

Just stumbled on this:
http://www.research.ibm.com/massive/pitdl.pdf
Possibly interesting...
Volker

[Dann Corbit]

Geschrieben von:/Posted by: Dann Corbit at 19 September 2004 02:27:50:
Als Antwort auf:/In reply to: Re: Practical Issues in Temporal Difference Learning geschrieben von:/posted by: Andrew Fan at 18 September 2004 15:18:07:

So, this neural net "engine" has been learning to play backgammon since 1992 or even earlier? I wonder how good it is now in 2004.
Anyone care to try it on chess? I am no neural net fan (pun intended).

Implementations:
http://www.turm-lage.de/TeX/nessi/
http://satirist.org/learn-game/systems/neurochess.html
http://www.soft-ware.net/spiele/brett/p03284.asp
http://satirist.org/learn-game/systems/sal.html
http://satirist.org/learn-game/projects/morph.html
Papers:
http://digilander.libero.it/gargamellac ... uators.zip
The most successful learning approach in a chess game so far is TD-Leaf (not a neural net), I think. And the best implementation of that is KnightCap.



my ftp site {remove http:// unless you like error messages}