Information Technology Reference
In-Depth Information
100
90
80
All Op
GR
GR+GT
70
60
50
40
30
20
10
0
0
20
40
60
80
100
120
140
160
180
200
Population size
Figure 12.10.
Dependence of success rate on population size in healthy and
strong populations evolving under a mix of several operators (All Op) and popula-
tions evolving exclusively by recombining genes (GR:
p
gr
= 1.0; and GR+GT:
p
gr
= 1.0 and
p
gt
= 0.5). The success rate was evaluated over 100 identical runs.
12.4 The Role of Neutrality in Evolution
We already know that the automatic evolution of computer programs can
only be done smoothly and efficiently if a genuine genotype/phenotype map-
ping is available. The creation of such a mapping requires some creative
thinking because proteins and computer programs are very different things.
Thankfully, computer programs are much easier to understand than proteins
and it is not necessary to know, for instance, the rules that determine the
three-dimensional structure of proteins in order to create a simple genotype/
phenotype system capable of evolving computer programs. What are, then,
the fundamental properties common to the DNA/protein system and an arti-
ficial system especially designed to evolve computer programs? Obviously,
the first is the creation of the genome/program dyad; and second, no matter
what, the genome must always produce valid programs. And how can that be
accomplished?
Turning to nature for inspiration can help. How does the DNA/protein
system cope with complexity? Is the information somehow fragmented in
the genome? Then perhaps the fragmentation of the genome in genes could
