Information Technology Reference
In-Depth Information
accomplish this includes the special domain Dc for encoding the random nu-
merical constants, which, for the sake of simplicity and efficiency, is only
implemented in the genes encoding the ADFs (one can obviously extend this
organization to the homeotic genes, but I have the feeling that nothing is gained
from that except a considerable increase in computational effort). The struc-
ture of the homeotic genes remains exactly the same and they continue to
control how often each ADF is called upon and how these ADFs interact
with one another.
Consider, for instance, the chromosome below with two homeotic genes
and two conventional genes encoding ADFs with random numerical con-
stants (the Dc's are shown in bold):
0123456789001234567890012345678012345678
**?b?aa
4701
+/Q?ba?
8536
*0Q-10010/Q-+01111
(6.5)
and the respective arrays of random numerical constants:
C
0
= {0.664, 1.703, 1.958, 1.178, 1.258, 2.903, 2.677, 1.761, 0.923, 0.796}
C
1
= {0.588, 2.114, 0.510, 2.359, 1.355, 0.186, 0.070, 2.620, 2.374, 1.710}
The genes encoding the ADFs are expressed exactly as normal genes with a
Dc domain (see section 5.2, Genes with Multiple Domains to Encode RNCs)
and, therefore, the ADFs will, most probably, include random numerical con-
stants (Figure 6.12). Then these ADFs with RNCs are called upon as many
times as necessary from any of the main programs encoded in the homeotic
genes. As you can see in Figure 6.12, in this case, ADF
0
is invoked twice in
Cell
0
and once in Cell
1
, whereas ADF
1
is used just once in Cell
0
and called
three different times in Cell
1
. Let's now see how the system copes with the
evolution of these complex entities by solving a difficult problem requiring
random numerical constants.
6.4.2 Designing Analog Circuits with the ADF-RNC Algorithm
The analog circuit we are going to design in this section is the same of section
5.6.5 and, therefore, whenever possible, we will keep the same settings, namely,
the number of runs, population size, number of generations, function set, training
set, fitness function, number of genes and head sizes and, of course, identical
rates of modification in order to facilitate the comparisons between the different
systems. Although we know that random numerical constants play a crucial
