Information Technology Reference
In-Depth Information
a.
01234560123456
AOAbabac ca ba-[m] = 7
AOAbabac ca ba-[d] = 7
ba
Nb
b.
Sub-ET
1
Sub-ET
2
c
A
O
A
b
a
b
a
Figure 3.14.
Illustration of neutral mutations.
a)
The mother and daughter chromo-
somes with the mutation points shown in bold.
b)
The sub-ETs encoded by both
chromosomes (they are exactly the same and, therefore, are shown just once). Note
that the point mutations shown in (a) have no expression in sub-ET
2
as they
occurred downstream of the termination point.
to create new ones. These building blocks are very different from the ones a
mathematician would have chosen but, nonetheless, they work much more
efficiently.
And finally, it is worth emphasizing that in gene expression programming
there are no constraints both in the type of mutation and the number of muta-
tions in a chromosome because, in all cases, the newly created individuals
will always be syntactically correct programs. This important feature distin-
guishes gene expression programming from all GP-style systems, where
straight point mutations would have resulted most of the times in invalid
programs. And an evolutionary algorithm unable to use freely this powerful
operator is severely restricted, since mutation is the most important agent of
genetic diversification (Ferreira 2002a; see also the discussion of the Ge-
netic Operators and Their Power in chapter 12).
3.3.3 Inversion
We know already that the modifications bound to make a big impact occur
usually in the heads of genes. Therefore, the inversion operator was restricted
to these regions. Here any sequence might be randomly selected and inverted.
It is worth pointing out that, since the inversion operator was restricted to the
heads of genes, there is no danger of a function ending up in the tails and,
