Are multiple runs of genetic algorithms better than one?

  • Authors:

  • Erick Cantú-Paz;David E. Goldberg

  • Affiliations:

  • Center for Applied Scientific Computing, Lawrence Livermore National Laboratory, Livermore, CA;Department of General Engineering, University of Illinois at Urbana-Champaign, Urbana, IL

  • Venue:
  • GECCO'03 Proceedings of the 2003 international conference on Genetic and evolutionary computation: PartI
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

There are conflicting reports over whether multiple independent runs of genetic algorithms (GAs) with small populations can reach solutions of higher quality or can find acceptable solutions faster than a single run with a large population. This paper investigates this question analytically using two approaches. First, the analysis assumes that there is a certain fixed amount of computational resources available, and identifies the conditions under which it is advantageous to use multiple small runs. The second approach does not constrain the total cost and examines whether multiple properly-sized independent runs can reach the optimal solution faster than a single run. Although this paper is limited to additively-separable functions, it may be applicable to the larger class of nearly decomposable functions of interest to many GA users. The results suggest that, in most cases under the constant cost constraint, a single run with the largest population possible reaches a better solution than multiple independent runs. Similarly, a single large run reaches the global faster than multiple small runs. The findings are validated with experiments on functions of varying difficulty.