Information Technology Reference
In-Depth Information
genome. Therefore, with GEP, a remarkable thing happened: the second evo-
lutionary threshold - the phenotype threshold - was crossed (Dawkins 1995).
And this means that only the genome (slightly modified) is passed on to the
next generation. Consequently, one no longer needs to replicate and mutate
rather cumbersome structures as all the modifications take place in a simple
linear structure which only later will grow into an expression tree.
The pivotal insight of gene expression programming consisted in the in-
vention of chromosomes capable of representing any parse tree. For that
purpose a new language -
Karva
language - was created in order to read and
express the information encoded in the chromosomes.
Furthermore, the chromosome structure was designed to allow the crea-
tion of multiple genes, each coding for a smaller program or sub-expression
tree. It is worth emphasizing that gene expression programming is the only
genetic algorithm with multiple genes. Indeed, the creation of more complex
individuals composed of multiple genes is extremely simplified in truly func-
tional genotype/phenotype systems. In fact, after their inception, these sys-
tems seem to catapult themselves into higher levels of complexity and count-
less new ideas are waiting to be explored. In this topic we will encounter
some of these rather complex entities like, for instance, the uni- and multi-
cellular systems, where different cells put together different combinations of
genes and evolvable genotype/phenotype artificial neural networks and de-
cision trees that not only learn but also adapt their structures to solve a wide
variety of problems.
The basis for all this novelty resides on the simple, yet revolutionary struc-
ture of GEP genes. This structure not only allows the encoding of any con-
ceivable program but also allows an efficient evolution. This versatile struc-
tural organization also allows the implementation of a very powerful set of
genetic operators which can then very efficiently search the solution space.
As in nature, the search operators of gene expression programming always
generate valid structures (be they complex mathematical expressions, or com-
plex artificial neural networks, or sophisticated decision trees) and therefore
are remarkably suited to creating genetic diversity.
In the next chapter we are going to learn all the details about the structural
and functional organization of GEP chromosomes; how the language of the
chromosomes is translated into the language of the expression trees; how the
chromosomes work as genotype and the expression trees as phenotype; and
how an individual program is created, matured, and reproduced, leaving off-
spring with new properties, thus, capable of adaptation.
