Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: An algorithm for KBN vs. K

Author: GuyHaworth

Date: 02:55:40 09/04/04

Go up one level in this thread



I like Rob's approach - to look for subgoals and phases.  The method he
recommends is also stated (for one) in Muller/Lamprecht's 'Fundamental Chess
Endings', p18 with the supporting observations:

    - wK stays in opposition to, or a N-move, away from the bK
    - the N stays near the bK, tending to cover flight-square the B doesn't
    - it is essential to know 5k2/5N2/5K2/5B2/8/8/8/8/8 and the W rule.


It would be interesting to use the KBNK EGT to see to what extent these phases
actually occur in DTC- (== DTM-) optimal play.

Also, one can see the distribution of squares the bK is on when mated
(WILHELM?), and test whether White still wins when those mates most remote from
the b-corner squares are removed.  FCE says mate "can only be forced in the
'right corner'" so maybe this is a non-exercise.


I don't know of any Artificial Neural-Network experiments, aiming to capture the
hidden 'higher-level knowledge' of KBNK in the ANN parameters.  Such an
experiment would be interesting.  I wonder if it would capture the phasing and
principles above as well.






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.