Parameter meta-optimization of metaheuristic optimization algorithms

  • Authors:

  • Christoph Neumüller;Stefan Wagner;Gabriel Kronberger;Michael Affenzeller

  • Affiliations:

  • Heuristic and Evolutionary Algorithms Laboratory School of Informatics, Communications and Media, Upper Austria University of Applied Sciences, Hagenberg, Austria;Heuristic and Evolutionary Algorithms Laboratory School of Informatics, Communications and Media, Upper Austria University of Applied Sciences, Hagenberg, Austria;Heuristic and Evolutionary Algorithms Laboratory School of Informatics, Communications and Media, Upper Austria University of Applied Sciences, Hagenberg, Austria;Heuristic and Evolutionary Algorithms Laboratory School of Informatics, Communications and Media, Upper Austria University of Applied Sciences, Hagenberg, Austria

  • Venue:
  • EUROCAST'11 Proceedings of the 13th international conference on Computer Aided Systems Theory - Volume Part I
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

The quality of a heuristic optimization algorithm is strongly dependent on its parameter values. Finding the optimal parameter values is a laborious task which requires expertise and knowledge about the algorithm, its parameters and the problem. This paper describes, how the optimization of parameters can be automated by using another optimization algorithm on a meta-level. To demonstrate this, a meta-optimization problem which is algorithm independent and allows any kind of algorithm on the meta- and base-level is implemented for the open source optimization environment HeuristicLab. Experimental results of the optimization of a genetic algorithm for different sets of base-level problems with different complexities are shown.