The simple
operators and representations described above form the backbone of
all GAs, but, because natural genetics is in reality a much more
complex phenomenon than that portrayed so far, it is possible to
conceive of several alternative representations and operators which
have particular advantages for some GA applications, including:
- Introducing the concepts of diploidy and dominance,
whereby solutions are represented by (several) pairs of chromosomes. The
decoding of these (which determines between blue and brown eyes,
say) then depends on whether individual bits are dominant or
recessive. Such a representation allows alternative solutions to be
held in abeyance, and can prove particularly useful for optimization
problems where the solution space is time-varying.
- Introducing
the ideas of niche and speciation
in multimodal problems, whereby
one deliberately tries to maintain diversity (to breed different
species exploiting different niches in the environment) in order to
locate several of the local optima. This can be achieved by
elaborating the selection and recombination rules described above.
- Introducing some sort of intelligent control over the selection of
mating partners, such as the ``inbreeding with intermittent
crossbreeding'' scheme of Hollstien [38]. In this scheme similar
individuals are mated with each other as long as the ``family''
fitness continues to improve. When this ceases new genetic material
is added by crossbreeding with other families.
These and other
advanced operators are discussed in detail in Chapter 5 of the
seminal book by Goldberg [31].
