Abstract
In recent years, not only ruggedness but also neutrality has been recognized as an important feature of a fitness landscape for artificial genetic search. In this paper, we propose the use of the Nei's standard genetic distance, which was originally proposed in population genetics, for estimating the degree of neutrality in fitness landscapes. The characteristics of the Nei's standard genetic distance are shown by applying the standard genetic distance to artificial evolution for tunably neutral NK fitness landscapes. Further investigations are conducted in a complex evolutionary robotics fitness landscape to validate the proposed method. The results show that neural network controllers changing the number of hidden neurons have different levels of neutrality as well as ruggedness in their landscapes. This suggests to us that the Nei's standard genetic distance in natural evolution can be successfully applied to estimating the degree of neutrality in artificial evolution after minor modifications.
