On Wed, 28 May 1997, Alan Brain wrote:
> John Erickson wrote:
> > -----> Isn't what you are describing a genetic algorithm? I saw them applied to circuits
> > in a magnetic field. You start out with a bunch of children, mix them take the best results
> > and so on. I do remember you had to introduce a random mutation factor along the way.
> Random mutation will get you there. Eventually. After some years of numbercrunching.
> But sexual reproduction (and funnily enough, 2 sexes is the most efficient form) is
> needed for rapid evolution. You also end up with a number of populations sharing
> similarities, which is interesting (and in the problem domain in question meant that
> depending on the threat, you had to select a tactic) but that's about all I can say.
> > It was a pretty interesting subject that I got to read up on when I was in school but like
> > many other things I've lost the time to do now. If you have more cool things rolling around
> > in your head spout them out I like reading this stuff.
> Ta. Trouble is, most are to do with Ada (see http://www.adahome.com/) and suchlike,
> plus underwater acoustics, em propagation and so on. Alas, I have had
Alan, I think what John meant was that every few generations you
randomly mutate a single gene (maybe more, usually you define that
parameter). Has anyone else read Koza's 'Genetic Algorithms', this is
pretty well THE book on the topic for the last 10yrs or so.
I've seen some novel applications of GA's/GP/EP too. One involved
backing up a semi-trailer into a loading bay, the program figured out all
the relevant things like angles and speed to use to get it right. Another
figured out probabilities of horse races and suggested what sort of bets
to place to maximise your winnings. Worked pretty well too IIRC.
I did a fair bit of reading on this stuff last year as part of an
AI subject and loved it. Alan you might be interested in one of Koza's
books, it had actual code for setting up problems, I think it was actually
C based. He seemed to be of the opinion that a very wide varietyof
problems could be solved using GA's so long as you defined the problem in
the right way.
I have a funny feeling that you probably new most of this Alan,
but couldn't say ;) But I guess you can't say if you can't say eh?
Although I can't imagine why, all this theory is prettyt public domain
stuff. But military secrets are just that, secret, and there can be a
fine line between general theory and specific aplications at times.
Tony VanDerLinden - | Department of Computer Science
E-mail: | James Cook University of North Queensland
Antony.VanDerLinden_at_... | Australia
Received on Thu Jan 01 1970 - 00:00:00 UTC