Author: Graham Laight
Date: 02:26:08 07/26/01
Go up one level in this thread
I believe the following proposal to be both important for chess, and EXTREMELY important for AI in general. Basically, I think that the most interesting thing you can do is to see whether you can use the self-learning AI techniques to learn some things about chess. What I would like to see you do (and I'd love to hear how successful you are!) is to attempt to use AI techniques to learn to play better chess using the following steps: * build some evaluation function "components" (you may look in Crafty's eval.c file for examples of things that chess programs evaluate) * get some sample positions * see if you can use genetic algorithms (GA) to select a good combination of components and weightings to evaluate these positions * then, see whether you can train NNs to "classify" new positions automatically into the correct completed evaluation function * you might also try generating eval functions with GAs fully automatically - using "components" that work with the absolute and relative positions of different pieces. This would be awesome! My only doubt about its suitability as an academic project is that computer professors tend to like very specific things which are well "nailed down". Therefore, if you present this as a project proposal, you'll need to go in with something that looks like a detailed plan for the project. If you take this as your project, I promise I'll buy the 3 of you a bottle of champagne each - such is my opinion of the value this work would have. -g On July 25, 2001 at 16:21:51, Iddo Bentov wrote: >hola.. > >we are 3 students starting a project in AI course at university.. > >we're thinking about ideas for a project in computer chess, we tried >to focus on ideas that involve chess endings.. > >so far we haven't found interesting ideas that haven't been tried >before.. > >our request.. if anyone has any new/interesting ideas in computer >chess that he thought about and are worth trying.. please reply >and tell us? the ideas don't have to be of the kind that actually >improve the current strong chess programs.. just ideas that are >interesting.. > >thanks.. > iddo, itay and yan
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.