Newsgroups: comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!newsfeed.pitt.edu!scramble.lm.com!news.math.psu.edu!chi-news.cic.net!nntp.coast.net!zombie.ncsc.mil!news.mathworks.com!newsfeed.internetmci.com!info.ucla.edu!ihnp4.ucsd.edu!munnari.OZ.AU!news.mel.connect.com.au!harbinger.cc.monash.edu.au!news.rmit.EDU.AU!matilda.vut.edu.au!rhh
From: rhh@matilda.vut.edu.au (Robert Hinterding)
Subject: Re: Crossover vs Mutation
Message-ID: <DnpwAo.2w7@matilda.vut.edu.au>
Organization: Victoria University of Technology
References: <4gfvv5$d7r@soleil.uvsq.fr> <DnBHuv.4r7@matilda.vut.edu.au> <4h3os4$hkb@soleil.uvsq.fr>
Date: Mon, 4 Mar 1996 00:35:12 GMT
Lines: 55

Olivier Chocron <chocron> writes:

>rhh@matilda.vut.edu.au (Robert Hinterding) wrote:
>>Olivier Chocron <chocron> writes:
>>
>>
>>Argueing about what happens in nature or who said what seems rather pointless
>>as evolutionary computation algorithms are only rather loosely based on nature.
>>What is more important is which of the many operators and their variations
>>are useful for different problem types.  It fairly simple to modify your favorite
>>algorithm so that for example it produces only one child and uses only crossover
>>or mutation to produce the child, not both.  If you leave all the other 
>>parameters constant and vary the probability of using either crossover or
>>mutation and do lots of runs, then you will have the answer for the problems
>>you used.  Not nearly as much fun as argueing about it, but brings out 
>>interesting results.
>Well, of course you can but it is not as usefull as discovering why here or
>there, mutation or crossover will work better or not.Then you will be able to
>choose a method with a better insight that running all possibilities and see
>what is best and it would be a great deal of spared time for later runs on
>different problems.The easiest way might not be the best.

What I was getting at is that Natural Evolution and Evolutionary Computation
are VERY different.  Natural Evolution is constrained by the physics of life
and the current manifestation of life on earth.  In Evolutionary Computation
we are currenlty only looking at problems which are trivial compared to living
organsims, and we are free to choose representations, operators and algorithms
which all have some effect on the problems we are trying to solve.

My own feeling is that Evolutionary Computation algorithms are simple (in terms
of complexity), but that their behaviour is complex. By getting emperical
evidence of which operators and representations are effective for different
problems types, we should be able a gain a better understanding of these 
operators and representations.  This should help us to able to predict which
algorithms, representation and operators to use for different problems and
give us a better understanding of how Evolutionary Computation works.

>If I follow your philosophy, I could go farer and say:
>Optimization is pointless as you can try all possibilities and you will see
>which is better...

I am not at all sure of what you are getting at. 
My point is that if you want to find out if a certain operator is useful,
run tests to check out your hypothesis.  You don't always get the results
you want, but it leads you to thinking about it more fully and can lead to
useful insights.  

Cheers,
Robert

-- 
Robert Hinterding                       Email: rhh@matilda.vut.edu.au 
VICTORIA UNIVERSITY OF TECHNOLOGY       Fax:   +61 3 9688 4050
P.O. Box 14428, Melb Mail Centre        Phone: +61 3 9688 4686              
AUSTRALIA 3000                          Home Page: http://dingo.vut.edu.au/~rhh
