mkamer@cs.columbia.edu (Matthew Kamerman) (12/19/89)
I am working on a Master's Thesis involving parsing Range Maps by growing
an "ecosystem" of competing simulated organisms which use the range map as
resources. Of fundamental importance in this endeavor is avoidance of the
use of arbitrary weightings. Hence, the creatures need to evolve their
own suitable characteristics.
The genetic system which generates the characteristics of the organisms
should be:
1: Minimal - All traits should be producible from as small a set of
genes as possible. This includes gene regulating "homeobox" genes.
2: Closed - All gene descriptors should be parsable, if nothing else,
as "Do Nothing".
3: Extendable - While the vocabulary of primitive genes "codons" is
minimal and closed, the genotype of an organism can be an
arbitrarily large data structure filled with these codons.
4: Practical - While a Turing Strength genetic system can "do anything"
it might take a *very* long time to do it! Weaker, but still
expressive systems might be worth while.
Consideration of biological ecosystems has led me to believe that the
cross-over and dominant/recessive phenomena involved in sexually
reproducing organisms are more important than simple mutation in
producing a diverse but structured mix of organisms. Genetic Systems
capable of supporting sexual reproduction are especially desired.
Pointers, keywords, etc. are welcome. I'm willing to do quite a bit
of sifting for one nugget of information, so please lay it on. My
undergraduate work was in Biology and my graduate in Computer Science,
so by all means get technical!
Thank you all very, very much,
Matthew Kamerman
Requests for summaries, etc, will be saved and honored ~January '90..