Author: Graham Laight
Date: 15:01:35 11/10/00
Go up one level in this thread
On November 10, 2000 at 17:48:18, Bob Durrett wrote: >On November 10, 2000 at 15:09:46, Graham Laight wrote: > ><snip> > >>Another problem here - I'm not 100% convinced that you've addressed one of the >>greatest complaints that programmers' have when you talk to them - when you >>tinker with the tuning, you often find that doing something to make your program >>play better in one type of situation, at the same time you make it worse in >>other types of situation. >> >>Maybe you can modify your thinking to overcome the above weaknesses? >> > >Off the top of my head, it immediately occurs to me that the "learning >algorithms" used so far might be much simpler than those which would be required >here. Development of really good learning algorithms [or even just "ideas"] >might win someone a prize of some sort. [Nobel?] My opinion is - don't INVENT a learning algorithm - FIND one. The man who invented the photocopier (sorry - I've forgotton his name) didn't go back to 1st principles - he studied what was known about photoelectronics. Look to ai research! That's been the inspiration for most of my suggestions. Some stuff (e.g. neural networks or genetic algorithms), doesn't even need to be written - you can buy it ready made off the shelf. >I'll give this more thought and post any "brilliancies" I may come up with >later. However, I suspect that "learning" is a branch of Computer Science that >will produce many PhD theses in the future. Not trivial. I look forward to reading it! -g
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.