Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!news.mathworks.com!hookup!swrinde!pipex!uknet!festival!ainews!aisb!andrewt
From: andrewt@aisb.ed.ac.uk (Andrew Tuson)
Subject: Re: A quick question...
Message-ID: <D00v8x.6wz@aisb.ed.ac.uk>
Sender: news@aisb.ed.ac.uk (Network News Administrator)
Reply-To: andrewt@aisb.ed.ac.uk (Andrew Tuson)
Organization: Dept of AI, University of Edinburgh, Scotland
References:  <CzzEsB.EKH@freenet.carleton.ca>
Date: Tue, 29 Nov 1994 09:19:44 GMT
Lines: 34

In article <CzzEsB.EKH@freenet.carleton.ca>, av574@FreeNet.Carleton.CA (Tim
 Sallans) writes:

Hi Tim!

# I'm kind of new to this field, and posted a question on comp.alife...
# 
# I thought I might repeat it here (this seems like a more appropriate
# forum <grin>).
# 
# Has there been any work done where mutation, transcription, crossover,
# genome lengthening or shortening (through duplication or deletion),
# and other such reproduction parameters are coded in the parent's
# genome, as opposed to being an external (developer provided) static
# parameter?

Yes, it has been done before. In the area of Evolution Strategies,
especially if I remember correctly. I've a few ideas in this direction
myself - I'd be interested in any references that anyone has!

Another way is to use a `learning rule` that uses information on how
each operator is performing to adjust the operator probabilities -
Lawrence Davis has done work on this - read his book or his paper in a past
ICGA (can't remember which one offhand - sorry!).


# Would this accelerate the evolution of the genome, by optimizing the
# search strategy, while simultaneously seeking solutions to the
# fitness function (assuming a static fitness function for now :) )?

In short, yes. However I suspect that the real story is somewhat more subtle.

Andrew Tuson (andrewt@aisb.ed.ac.uk)
Department of Artificial Intelligence, University of Edinburgh, Scotland, UK.
