computer simulation project

Jeremy John Ahouse ahouse at hydra.rose.brandeis.edu
Fri Feb 26 15:40:15 EST 1993


In article <86842 at ut-emx.uucp> Mikael Behrens, mbehrens at cs.utexas.edu
writes:
>I am still researching, and would be very interested in any comments about  
>this project.  Could someone suggest good sources of information about the  
>Hawk-dove game, and the Prisoners' dilemma?  So far, the three mentioned
>above are the only "games" I've come across.  If anyone knows of others  
>please let me know.  I'm doing this project on a NeXT workstation.

Make sure to have a look at Thomas Ray's Tierra - it is described in
Alife II conference proceedings "An Approach to Synthesis of Life" -
Addison Wesley 1991 pg 371.

There are also various approaches to modelling evolution (variations
on the GA theme) that you could explore.  I just read an article by
Stephen Smith on this topic.  (see abstract below)

Also you might want to see the latest issue of Theoretical Biology
which has an article by Bard Ermentrout and Leah Edelstein-Keshet on
CA use in Biology.  Sorry I don't have the journal date.

        	Jeremy John Ahouse
        	Center for Complex Systems and Biology Dept
        	Brandeis University
        	Waltham, MA 02254-9110

        	(617) 736-4954
        	email: ahouse at hydra.rose.brandeis.edu


AN INVESTIGATION OF META LEVEL EVOLUTIONARY CONTROL - ABSTRACT
  by Stephen J. Smith

Meta Evolution refers to a simulated evolutionary system where the normal
hierarchical operators such as mutation are encoded into the genome.  The
implicit hierarchical control levels of most simulated evolutionary systems
can thus be intermixed with the genotype much the same way they are in real
biological systems.  The seeming disadvantage of meta evolution is that when
these otherwise distinct control operators are encoded in the genome they
dramatically increase the size of the search space (in the experiments
described here there was an increase of 2^320.  The hoped for advantage of
meta evolutionary systems is that they will be less brittle and more likely
to display innovative solutions to given evolutionary optimization problems.

To test these hypotheses three software toolkits were built on the
Connection Machine CM-5 parallel computer.  They included an evolutionary
system simulator called {\em *GA}, a ``scope'' for viewing the evolutionary
process, called {\em Evoscope}, and a description of the search space called
the ``ramp'' function.  These three tools were then used to create
populations of simple ramp organisms.  The evolution of these populations
was carried out under three main experiments:  1.  A standard evolutionary
system of mutation, crossover and selection.  2.  A system where the
mutation rate for all genes of each genome was encoded within the genome.
3.  A system where the mutation rate for each gene was encoded in the
preceding gene.  These three mutation strategies will be called:  {\em
global mutation}, {\em genome specific mutation} and {\em locus specific
mutation} respectively.  The mutation gene values for the genome and locus
specific mutation rates were modified by a global mutation rate though the
phenotypically expressed genes were mutated based only on the values of the
mutation genes themselves.

The experiments showed that all three systems were able to achieve near
optimal performance within the ramp search space within the same number of
generations.  The meta evolutionary systems achieved this by forcing the
genome and locus specific mutation rates to acceptable values soleley
through selective pressure.  The meta evolutionary strategies performed
slightly worse when the global mutation rate was set higher than optimal but
performed better when global mutation rates were set lower than optimal.

  Stephen J. Smith
  Thinking Machines Corporation               smith at think.com
  245 First Street                            617-234-1000
  Cambridge, MA 02142, USA                    617-234-4444 (fax)



More information about the Comp-bio mailing list