🚀 go-pugleaf

In article <1993Jan22.213851.8114@wuecl.wustl.edu> liu@rodin.wustl.edu (Liu Ann Pheng) writes:

   Greetings!  I have a question for the net readers:

   What is the relationship (as well as main differences, if any) between
   genetic algorithm and evolutionary programming?  

   Thanks,
   Liu, Ann Pheng

Genetic algorithms (GA) and Evolutionary Programming (EP) (ala L. Fogel and D.
Fogel) are similar in that they both use analogies to evolution to do problem
solving search. They differ on about every other aspect.  The most import
difference is the LEVEL of evolution they choose to make the analogy.  In GAs,
the level is that of genetics, i.e. evolution is modeled as a process that
alters a genetic code which is translated into a phenotype that is evaluated.
Evolution in GA occurs via _genetic operators_, i.e. crossover, inversion,
point mutation.  In EP, the manipulation is done directly on the phenotype
rather than on a description that is converted into the phenotype.  EP uses a
mutation operator that takes the population member and alters it so that is
still reasonably similar.  This is a higher level model of evolution and
potentially less flexible but more exploitable.

This can be seen best by an example.  Consider how each would evolve neural
networks. In a standard GA, a fixed-length string would be used with an
interpretation function to convert it to a neural network for evaluation.
Manipulation is done on the string making how the network is represented in the
string, i.e. how the interpretation function converts the string, a real
concern for GAs. In EP, the network itself is the member of the population and
is manipulated directly to create offspring. Mutation would alter the weight
values directly by adding a small random amount to it.  Other manipulations
would alter connections in the network.  I have intentionaly chosen an example
that benefits EP here to show its advantage over GA since it is the lesser
known of the two, but there are many tasks where the flexibility of a GA's
operators makes it the proper choice. 

There are other differences which are mainly due to the traditional support
mechanisms used with these techniques in practice (e.g. a determinsitic
selection method for EP and a probabilistic one for GA), but the essential
difference is their level of analogy to evolution.

-pete angeline

-- 
-------------------------------------------------------------------------------
Peter J. Angeline            ! Laboratory for AI Research (LAIR)
Graduate Research Assistant  ! THE Ohio State University, Columbus, Ohio 43210
ARPA: pja@cis.ohio-state.edu ! "Nature is more ingenious than we are."