The True Nature of Multi-Dimensional Gaussian Mutation
The aim of this work is to pay attention of researchers who deal with evolutionary algorithms to the fact, that the most probably location of the mutated points in multi-dimensional Gaussian mutation is not in the nearest neighborhood of the base point, but in a certain distance proportional to the norm of the standard deviation vector, which increases with the landscape dimension. This fact may cause a decrease in the evolutionary algorithm sensitivity to narrow peaks when increasing the landscape dimension. Some new Gaussian-like mutation is proposed in order to overcome this problem.
KeywordsTarget Class Rule Induction Neighborhood System Landscape Dimension Quality Rule
Unable to display preview. Download preview PDF.
- Fogel, L.J., Owens, A.J., Walsh, M.J., “Artificial Intelligence through Simulated Evolution”, Wiley, New York, 1966.Google Scholar
- Michalewicz, Z., “Genetic Algorithms + Data Structures = Evolution Programs”,SpringerVerlag, Berlin Heidelberg, 1996.Google Scholar
- Rechenberg, I., “Cybernetic solution path of an experimental problem”, Roy. Aircr. Establ., libr. Transl. 1122, Farnborough, Hants., UK, 1965.Google Scholar