Newsgroups: comp.ai.neural-nets,comp.ai.genetic
Path: cantaloupe.srv.cs.cmu.edu!bb3.andrew.cmu.edu!newsfeed.pitt.edu!gatech!newsfeed.internetmci.com!in1.uu.net!DIALix!sydney.DIALix.oz.au!quasar!telford
From: telford@threetek.dialix.oz.au (Telford Tendys)
Subject: Re: Experiments with GA and neural nets
In-Reply-To: inmanh@cogs.susx.ac.uk's message of 8 Jan 1996 10:26:03 GMT
Message-ID: <1996Jan11.000324.5442@threetek.dialix.oz.au>
Followup-To: comp.ai.neural-nets,comp.ai.genetic
Organization: 3Tek Systems Pty Ltd., N.S.W., Australia
Date: Thu, 11 Jan 1996 00:03:24 GMT
Lines: 42
Xref: glinda.oz.cs.cmu.edu comp.ai.neural-nets:29185 comp.ai.genetic:7659

> From: inmanh@cogs.susx.ac.uk (Inman Harvey)
> 
> Archmage (Archmage@mageton.demon.co.uk) wrote:
> : ...
> : What I am thinking of would be start with one cell, which has a chunk
> : of DNA (given to it by the GA program?). This DNA is obeyed, causing
> : the cell to grow, move, divide, react to surrounding cells etc.
> : The DNA will contain instructions for cells in some circumstances to
> : become neurons, and react to other neurons.
> 
> : Has anyone else thought of something like this?
> 
> It's a wide-open research area, and I don't think anyone has come
> close to cracking the problem yet -- do not underestimate the
> ramifications!
> 
OK, this may have some relationship to the problem.
Check out New Scientist 1 July 1995 p34
(which I just happened to have on the shelf).
The article outlines embrionic development of neurons
and the way localised chemical signals are used to
``signpost'' the route for each neural link.

I guess this could be worked into a cellular automaton model
by giving the neural net some spacial structure
(presumably a 2 dimensional grid) and overlaying it with
the spacial structure of the cellular automaton.
Then some sort of rules would relate the growth of neural
links to the state of the local automaton cells.

I suspect that any classic cellular automaton (like game
of life, say) tends to have cells all the same size.
This fits with the grid structure nicely so programmers
like it. However, while this may be a good approximation
for bacteria, it doesn't cover neurons very well since
nerve cells are a few orders of magnitude larger (longer?)
than the surrounding cells. Possibly there is scope for
a new style of cellular automata where the cells can grow
longer under the right circumstance (although I can't
immediately visualise programming this).

	- Tel
