Advertisement

Performance and Scalability of Genetic Algorithms on NK-Landscapes

  • Hernán Aguirre
  • Kiyoshi Tanaka
Part of the Studies in Computational Intelligence book series (SCI, volume 153)

Summary

This work studies the working principles, performance, and scalability of genetic algorithms on NK-landscapes varying the degree of epistasis interactions. Previous works that have focused mostly on recombination have shown that simple genetic algorithms, and some improved ones, perform worse than random bit climbers and not better than random search on landscapes of increased epistasis. In our work, in addition to recombination, we also study the effects on performance of selection, mutation, and drift. We show that an appropriate selection pressure and postponing drift make genetic algorithms quite robust on NK-landscapes, outperforming random bit climber on a broad range of classes of problems. We also show that the interaction of parallel varying-mutation with crossover improves further the reliability of the genetic algorithm.

Keywords

Genetic Algorithms NK-Landscapes Epistasis Nonlinear Fitness Functions Selection Drift Mutation Recombination 

Preview

Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.

Copyright information

© Springer-Verlag Berlin Heidelberg 2008

Authors and Affiliations

  • Hernán Aguirre
    • 1
  • Kiyoshi Tanaka
    • 1
  1. 1.Faculty of EngineeringShinshu UniversityNaganoJapan

Personalised recommendations