CHAPTER 3. DEFINING DIVERSITY 58
a large area of the search space and not focussed our search onto an unprofitable area.
When we decide our search has continued for long enough, we will typically take the best
performing individual found so far, ignoring the other individuals in the population. The
optimal diversity in this context will be that which optimises the explore-exploit trade off,
such that the best points in the search space are found quickly and efficiently. This is
in contrast to ensemble diversity methods, which create diversity with the intention that
the ‘population’ of ensemble members will be combined. As such, evolutionary diversity
methods do not concern themselves with creating a population that is complementary in
any way, but instead with just ensuring the maximum amount of the hypothesis space is
being explored in order to find the best single individual.
In spite of these differences, some researchers have found interesting parallels. Yao [153]
evolves a population of neural networks, using fitness sharing techniques to encourage diversity, then combines the entire population as an ensemble instead of just picking the best
individual.