Author: Walter Faxon
Date: 02:58:08 11/22/05
Go up one level in this thread
On November 22, 2005 at 04:00:02, Andrew Wagner wrote: >For your information, comment, amusement, and criticism, I have decided to >announce a possible new project I may be launching. > >Over the next 6 weeks or so, I will be doing some research and preliminary >planning for a project to "evolve" a chess engine. If I decide to actually >launch the project, it would look something like this: >1.) Using OO design principles, and a significant amount of conditional >compiling, create a framework for an engine where all features can be specified >by a DNA-type string. I.E., one character acts as a gene specifying the board >representation, one specifies whether and how it will use SEE, and so on. >2.) Design an environment within which the engine can actually evolve. My basic >concept for this is to have a few "predator" engines to play against, and a few >instances of my engine with randomly-specified DNA. Each of my engines will play >a specified number of games, randomly choosing a predator to play each time. My >engines' fitness factor will be determined by its record in those games. >3.) Once the engine framework and evolutionary environment are in place, I'll >simply help the engine to evolve by introducing new genes which will allow it to >naturally select what features it will add. > >Thoughts and suggestions are of course welcome. Hi. You might be interested to know that using a genetic algorithm to tune a neural net for a game evaluation function is a patented technology of Natural Selection, Inc. NSI's CEO Dr. David B. Fogel wrote about using it for a checkers program in his book "Blondie24". See: http://www.natural-selection.com/Press/2001/pr11262001.htm Though according to at least one book reviewer the results weren't all that Dr. Fogel claimed them to be: http://www.jimloy.com/checkers/blondie.htm Surprisingly, I wasn't able to find the patent with a quick look at the USPTO site (http://patents.uspto.gov/). I clearly remember reading about it. Maybe it was just applied for, as the first link states. But looking at what I did find is discouraging. So much has been patented! Patents on obvious applications. Very annoying. -Walter P.S. Instead of "predator" engines, co-evolution should be preferred. An engine that knows any chess will wipe the floor with your creatures, who won't learn anything from the experience.
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.