Author: Jay Scott
Date: 15:35:32 01/06/98
Go up one level in this thread
On January 06, 1998 at 17:30:40, Don Dailey wrote: >In >every way (except raw speed) the Deep Blue team is handicapped so you >can not expect them to compete with the highly tuned micro programs. This is not true. The hardware has one big advantage besides outrageous speed: it's parallel at the transistor level. You can get more evaluation function terms at each node by adding transistors, and adding transistors does not slow you down. So a hardware chess machine has a different speed/knowledge tradeoff than a software chess program. Adding knowledge is cheap, at least until you use up your real estate, so it makes sense to add as much knowledge as you can. In software, adding knowledge almost always slows you down, so you have to be more careful. With as many devices as you can get on a chip nowadays, I imagine that its natural for chess hardware to have far more evaluation terms than software can afford. >So does Deep Blue suck? In rating points per node searched, YES. Well, what about rating points per log nodes searched? I don't think we have enough information to judge how well the Deep Blue team did in exploiting their potential advantage in knowledge. Their development time was relatively short, as these things go, so perhaps not well. But on the other hand they had plenty of talent, and they're perfectly capable of coming up with good new ideas. This IBM web page <http://www.av.ibm.com/2-2/CoverStory/> says that "some of Deep Blue's weights" were experimentally tuned by machine learning. DB is not a conventional program. Jay
This page took 0.01 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.