Author: Stuart Cracraft
Date: 10:35:57 11/04/04
Go up one level in this thread
On November 04, 2004 at 11:31:36, Rémi Coulom wrote:
>On November 04, 2004 at 09:14:19, GuyHaworth wrote:
>
>>
>>Meel Velliste and I used TD-learning (I think) in the ANN-experiment on
>>'Undestanding KPK'.
>>
>>See 'Chess Endgames and Neural Networks'. ICCA_J v21.4 (1998) pp. 211-227.
>>
>>g
>
>The algorithm used in this paper is not TD-learning. It is supervised learning
>of the KPK tablebase. TD-learning is very different. TD-learning learns the
>value of a position from the outcome of playing from this position. It does not
>require a supervisor that directly provides a value.
>
>Rémi
For anyone interested, there is now a group at Yahoo named "chesstune"
which discusses specific methods for tuning computer chess evaluation
functions. I thought it best to separate it from this group to avoid
the usual putdowns on anything that isn't manually done from several
vocal types.
See:
http://groups.yahoo.com/group/chesstune/
Stuart
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.