To what extent does imitating biological evolution benefit genetic programming?
Genetic algorithms are a set of search algorithms that have been inspired by biological evolution [holland]. They have been used in applications from *** to creative logo design[blprnt], by using variations of biological mutation, fitness-based selection, and populations.
Over the years, the benefits provided by copying biology have been debated. For example, the inclusion of sexual recombination, which is a part of nearly all plant and animal reproduction, drastically degrades the performance of genetic algorithms [nordin] something which evolutionary biologists still don’t have a computationally sound explanation for in biology (R. Redfield, personal communication, February 4, 2010). On the other hand, biological evolution has produced solutions to many different environmental conditions – from the darkest sea-floors, to the driest deserts – which could be used to inform computer science.
There is already a selection of literature on genetic algorithms which would allow a review to explore which aspects of biological evolution are worth emulating, and which have been unhelpful to the field of computer science.
By summarizing the state of the art of genetic algorithms, and comparing that with an introductory understanding of biology, I hope to describe several mechanisms of biological evolution, how they transfer to genetic algorithms, and present a base comparison of whether they are useful in computer science. Due to the situational nature of the search problems to which genetic algorithms are applied, I would expect categorical classification of benefit of these mechanisms to be difficult.
[holland]
Holland, John H. Adaptation in Natural and
Artificial Systems. Ann Arbor, MI: University
of Michigan Press 1975.
[nordin]
Peter Nordin, Frank Francone, and Wolfgang Banzhaf, 1996, Explicitly defined introns and destructive
crossover in genetic programming, Advances in Genetic Programming 2, chapter 6, pp. 111–134, MIT
Press, Cambridge,MA, USA.
[blprnt] www.blprnt.com/variance/