Information Technology Reference
In-Depth Information
such as gene transposition. For another, it can play the role of a status quo
agent, maintaining populations steady by permanently working at homog-
enizing their genetic makeup. Indeed, species are stable entities from their
birth to their extinction (Eldredge and Gould 1972) and operators such as
homologous recombination are fundamental to maintaining the status quo.
On the other hand, we also observed that the performance of a system not
only changes dramatically with mutation rate but also that the performance
peak is accessible to mutation alone. Therefore, mutation rates can be easily
tuned so that systems could evolve with maximum efficiency. In fact, muta-
tion rates are themselves tightly controlled and subjected to selection pres-
sures in nature, another indication that mutation, and not recombination, is
the center of the evolutionary storm.
And finally, we also observed that transposition operators display dynam-
ics similar to those of mutation (i.e., non-homogenizing dynamics) and that
populations undergoing transposition evolve significantly better than
populations undergoing recombination alone, further emphasizing the unique,
homogenizing effect of recombination.
12.2 The Founder Effect
We have seen in the previous section that the evolvability of a system will
depend heavily on the kind of genetic operator used to create genetic modi-
fication. And the size and kind of initial populations is closely related to this
question.
In all evolutionary algorithms, an evolutionary epoch or run starts with an
initial population. Initial populations, though, are generated in many differ-
ent ways, and the performance and the costs (in terms of CPU time) of differ-
ent algorithms depend greatly on the characteristics of initial populations.
The simplest and less time consuming initial population is the totally ran-
dom initial population. However, few evolutionary algorithms are able to
use this kind of initial population due not only to structural constraints but
also to the kind of genetic operators available to create genetic modification.
The initial populations of gene expression programming are totally random
and consist of the linear genomes of the individuals of the population.
In artificial evolutionary systems, the question of the initial diversity is
pertinent for two main reasons. First, for some complex problems, the ran-
dom generation of viable individuals (i.e., individuals with positive fitness)
