Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Chess and Neural Networks

Author: Gunnar Andersson

Date: 03:28:08 01/28/99

Go up one level in this thread


On January 27, 1999 at 13:30:05, Steve Maughan wrote:

>The problem with using NNs in chess is, as the old quote says, "chess is 99%
>tactics".  This means that a knight on f6 or g6 may mean the difference between
>+Checkmate and -Checkmate due to tactics, but to a NN it will only give a small
>difference.  NN are better at finding positional elements and are therefore
>extremely difficult to use in respect to chess.  If there were coupled to a
>tactical engine they would perform better - but will be much slower at
>evaluating a position, which in turn would slow down the tactical engine.
>
>Othello is a game that may benefit from NN.
>

There have been some attempts to use NNs in Othello. Mark Brockington used NNs
in his program Keyano to calculate the values for rare positions for which the
statistical methods used to estimate the pattern values had too high inaccuracy.
I got the impression that this worked out pretty well in practice, but there are
other, simpler, methods that also solve this problem.

In this year's Princeton II tournament, where all the top programs were present,
I don't think any of the programs used NNs - I know for a fact that the top 10
didn't.

Anyway, NNs might be useful, but as the statistical methods have proven so
successful, I don't think there has been a lot of interest in NNs lately. This
might change, though.

/ Gunnar



>Steve Maughan



This page took 0.01 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.