Information Technology Reference
In-Depth Information
a.
0123456789
DUDTUDcdabdcabacbad
012345678
901234567890123456
789010
429984097914824092
675841
b.
Dw
Dt
D
4
0
8
6
7
U
D
2
9
2
9
9
7
4
5
T
U
D
9
0
8
c
a
c
8
1
9
4
d
b
d
4
4
2
1
Figure 2.10.
Translation of chromosomes with two extra domains for handling the
weights and thresholds of neural networks.
a)
A multi-domain chromosome
composed of a conventional head/tail domain encoding the neural network
architecture, and two extra domains - one encoding the weights (Dw) of the neural
network encoded in the head/tail domain and another the thresholds (Dt). Dw and
Dt are shown in different shades.
b)
The sub-ETs codified by each domain. The
one-element sub-ETs encoded in Dw and Dt are placed apart together. “U”, “D”,
and “T” represent, respectively, neurons or functions with connectivity one, two,
and three. How all these sub-ETs interact will be shown in chapter 10.
expression. Indeed, the language of gene expression programming - Karva
language - is a versatile representation that can be used to evolve relatively
complex programs as simple, extremely compact, symbolic strings. In fact,
there is already commercially available software such as Automatic Problem
Solver by Gepsoft that automatically converts K-expressions and GEP chro-
mosomes into several programming languages, such as C, C++, C#, Visual
Basic, VB.NET, Java, Fortran, VHDL, Verilog, and others.
Another advantage of Karva notation is that it can be used to evolve highly
sophisticated programs using any programming language. Indeed, the origi-
nal GEP implementation was written in C++, but it can be done in virtually
any programming language, as it does not rely on any quirks of a particular
programming language. As a comparison, it is worth pointing out that early
GP implementations relied greatly on LISP because the sub-tree swapping
that occurs during reproduction in that system is very simple to implement in
that programming language.
