** Next:** 2 Does low diversity
** Up:** 1 Hypothesis
** Previous:** 1 Hypothesis
** Contents**

Tackett suggested that the rate of growth is proportionate
to the amount of selection pressure [Tackett, 1994] and
Langdon and Poli (1998a) showed that removing selection
pressure stops code growth. To see the effect of varied selection
pressure on code growth, one can look at different selection schemes and
objective functions used in the literature.
Smith and Harries (1998)
investigated selection schemes and variable tournament sizes to show
the same trend: less selection pressure generally leads to reduced
code growth.
Poli (2003) studied a method for reducing code growth by
periodically worsening the fitness of above average sized individuals.
In the control experiment with no code growth control, the results
(Fig. 1, page 211) show trends of smaller tournament sizes
producing less growth.
Considering the theories of code growth, particularly that individuals
with similar
size and shape are less likely to produce children
with worse fitness than two dissimilar individuals, an increased
rate of growth is expected with higher selection pressure.
In this case, higher selection pressure becomes more likely to pick the same
individual repeatedly.
Given
that the algorithm contains some bias for producing bigger individuals
(as bigger children
are more likely to have a higher chance of survival),
the repeated selection of these individuals will consistently
produce offspring which in turn have a better chance of survival.

** Next:** 2 Does low diversity
** Up:** 1 Hypothesis
** Previous:** 1 Hypothesis
** Contents**
S Gustafson
2004-05-20