Using Disruptive Selection to Maintain Diversity in GeneticAlgorithms

  • Authors:

  • Ting Kuo;Shu-Yuen Hwang

  • Affiliations:

  • Department of International Trade, Takming Junior College of Commerce, Institute of Computer Science and Information Engineering, National Chiao Tung University;Department of International Trade, Takming Junior College of Commerce, Institute of Computer Science and Information Engineering, National Chiao Tung University

  • Venue:
  • Applied Intelligence
  • Year:
  • 1997

Quantified Score

Hi-index 0.00

Visualization

Abstract

Genetic algorithms are a class of adaptive search techniques based onthe principles of population genetics. The metaphor underlyinggenetic algorithms is that of natural evolution. With their greatrobustness, genetic algorithms have proven to be a promisingtechnique for many optimization, design, control, and machinelearning applications. A novel selection method, disruptive selection, has been proposed.This method adopts a nonmonotonic fitnessfunction that is quite different fromconventional monotonic fitness functions. Unlikeconventional selection methods, this method favors both superior and inferiorindividuals. Since genetic algorithms allocate exponentiallyincreasing numbers of trials to the observed better parts of thesearch space, it is difficult to maintain diversity in geneticalgorithms. We show that Disruptive Genetic Algorithms (DGAs)effectively alleviate this problem by first demonstrating that DGAscan be used to solve a nonstationary search problem, where the goalis to track time-varying optima. Conventional Genetic Algorithms(CGAs) using proportional selection fare poorly on nonstationarysearch problems because of their lack of population diversity afterconvergence. Experimental results show that DGAs immediately trackthe optimum after the change of environment. We then describe aspike function that causes CGAs to miss the optimum. Experimentalresults show that DGAs outperform CGAs in resolving a spike function.