home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.ai.genetic
- Path: sparky!uunet!gumby!destroyer!news.iastate.edu!IASTATE.EDU!danwell
- From: danwell@IASTATE.EDU (Daniel A Ashlock)
- Subject: Evolution vs Coevolution
- Message-ID: <1993Jan24.103719@IASTATE.EDU>
- Sender: news@news.iastate.edu (USENET News System)
- Reply-To: danwell@IASTATE.EDU (Daniel A Ashlock)
- Organization: Iowa State University
- Date: Sun, 24 Jan 1993 16:37:19 GMT
- Lines: 31
-
- A function optimizer has a global standard of fitness, the height of the
- function being optimized. A GA playing prisoner's dillemma has a changing
- fitness function based on the particular mix of player's in the algorithm.
- I call these two sorts of genetic algorithms evolutionary and coevolutionary.
-
- What I'd like is to see references to transformations of evolutionary
- problems into coevolutionary problems. If possible summarize the effects.
- For example:
-
- Hillis, Daniel "Co-evolving Parasites Inprove Simulated Evolution as an
- Optimization Procedure" Proceedings of the Second Conference on Artificial
- Life.
-
- Summary:
-
- Hillis observed better results when he modified a GA that located sorting
- networks with relatively few comparitors to have an evolving population of test
- sequences whoes fitness was proportional to the number of sorting networks they
- fooled. The fixed population of test sequences let to a 65 comparitor network
- for a 16-element list. The coevolutionary version found a 61 element network.
- It should be noted that the "best" network contains 60 comparitors and was
- found by a human being.
-
- Thanks in advance,
-
- Dan
- Danwell@IASTAET.EDU
-
-
-
-
-