Mitchell, T. J. (2010) An exploration of evolutionary computation ...

For effective operation of the mutative self-adaptation mechanism, the strategy is
widely regarded to offer superior adaptive properties when compared with the
alternative (Bäck and Schwefel, 1993). This is due to the possibility that a highly-fit
offspring is generated with a step-size parameter that is entirely inappropriate for its new
location. This may arise when a recombinant with a very large mutation step-size
fortuitously jumps to a distant and highly fit region of the search space. If the offspring is
able to pass directly into subsequent generations (elitism), optimization is likely to stagnate
as further progress will be thwarted by the originally useful but now unsuitably large step-
size. This situation could not arise in the strategy, as the anomalous offspring would
expire after transmitting some of its strong genetic material through recombination.
2.3.2.4 Selection
The selection operator in the ES facilitates the drift of the population towards regions of
increasing fitness within the parameter space. Selection works in an opposing yet
complementary manner to the variation operators and identifies the direction in which
search should proceed. As was discussed earlier in this section, selection in the ES is
performed deterministically. In the case of the comma (or extinctive strategy), the
fittest individuals are chosen from the offspring; whereas in the plus (or
preservative strategy), selection is made amongst both the parent and offspring
populations. Schwefel and Rudolph (1995) established the concept of maximal lifespan
with the introduction of the exogenous parameter to indicate the number of generations
for which each individual is permitted to survive. The resulting strategy provides
a generalisation of the deterministic selection scheme, such that when the ES
presents an instance of the extinctive comma strategy; furthermore, when the
resulting ES is equivalent to the preservative plus strategy. The parameter may also be
set to any intermediary value in between these two extremes .
2.4 EA Similarities and Differences
Both the ES and GA derive inspiration from biological evolution; however, the specific
implementation of each EA is quite different. For example, in the theory that relates to the
GA it is assumed that the genes of the optimal solution are scattered throughout the
population; evolution is then the process of recombining these genes to produce the
optimum. ES theory, on the other hand, assumes that the optimum solution will be located
through a processes of organised, but random, mutative leaps through the object space.
27