Author: Michael Yee
Date: 19:49:37 05/09/05
I came across this interesting/strange paper today: http://www.natural-selection.com/publications_2004.html http://www.natural-selection.com/Library/2004/92jproc12-fogel.pdf Some remarks: (1) started with known good values for material and PSTs (so did in fact incorporate human knowledge) (2) doesn't say how they encoded the inputs for the 3 neural networks (whose total contribution to the eval was at most 3*50 [or -3*50]) (3) had artificial rule that games lasting 50 moves were draws (although I can't tell if this was just for during training) (4) fixed search depth of 4 ply (or 6 ply when extending for quiescence), but maybe only during training? (5) gauged strength with matches vs CM8000 personalities and Pocket Fritz (6) cites Kasparov communication as a reference... (7) badmouths td learning but doesn't even reference knightcap paper (8) doesn't cite what open source programs they "borrowed" from I'm curious why they didn't just play automated matches against a winboard or uci compatible program. I'm also curious how well a simple program on their athlon 2400 would do against pocket fritz. Mostly, I'm skeptical because I can't deduce all of the conditions of the experiments. Michael
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.