Averaging efficiently in the presence of noise

  • Peter Stagge
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 1498)


In this paper the problem of averaging because of noise on the fitness function is addressed. In simulations noise is mostly restricted to the finite precision of the numbers and can often be neglected. However in biology fluctuations are ubiquitous and also in real world applications, where evolutionary methods are used as optimization tools, the presence of noise has to be coped with [1]. This article originated from the second point: Optimizing the structure of Neural Networks their fitness is the result of a learning process. This value depends on the stochastic initialization of the connection strengths and thus represents a noisy fitness value. To reduce noise one can average over several evaluations per individual which is costly. The aim of this work is to introduce a method to reduce the number of evaluations per individual.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. 1.
    S. Rana, L.D. Whitley, and R. Cogswell. Searching in the Presence of Noise. In Parallel Problem Solving from Nature,4, pages 198–207. Springer Verlag, 1996.Google Scholar
  2. 2.
    H. Abarbanel, R. Brown, J. Sidorowich, and L. Tsimring. Analysis of observed chaotic data in physical systems. Rev. Mod. Phys., 65:1331–1392, 1993.MathSciNetCrossRefGoogle Scholar
  3. 3.
    Simon Haykin. Neural Networks. Macmillan College Publishing Company, New York, 1994.Google Scholar
  4. 4.
    Thomas Bäck. Evolutionary Algorithms in Theory and Practice. Oxford University Press, 1996.Google Scholar
  5. 5.
    Ingo Rechenberg. Evolutionsstrategie '94. Friedrich Frommann Holzboog Verlag, Stuttgart, 1994.Google Scholar
  6. 6.
    Hans-Georg Beyer. Toward a Theory of Evolution Strategies: Some Asymptotical Results from (1,+ λ)-theory. Evolutionary Computation, 1(2):165–188, 1993.Google Scholar
  7. 7.
    B. Miller and D. Goldberg. Genetic Algorithms, Selection Schemes, and the Varying Effects of Noise. Evolutionary Computation, 4(2):113–131, 1997.Google Scholar
  8. 8.
    Ulrich Hammel and Thomas Bäck. Evolution Strategies on Noisy Functions How to Improve Convergence Properties. In Y. Davidor, H.P. Schwefel, and R. Manner, editors, Parallel Problem Solving from Nature, 3, pages 159–168. Springer Verlag, 1994.Google Scholar
  9. 9.
    Herbert Büning and Götz Trenkler. Nichtparametrische statistische Methoden. Walter de Gruyter, Berlin, 1994.Google Scholar
  10. 10.
    Bernhard Sendhoff, Martin Kreutz, and Werner von Seelen. A condition for the genotype-phenotype mapping: Causality. In Thomas Bäck, editor, International Conference on Genetic Algorithms. Morgan Kaufman, July 1997.Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 1998

Authors and Affiliations

  • Peter Stagge
    • 1
  1. 1.Institut für NeuroinformatikRuhr-Universität-BochumBochumGermany

Personalised recommendations