SCIENCE@home

On the limitations of adaptive resampling in using the student's t-test evolution strategies

J. Kruisselbrink,M. Emmerich,Thomas Bäck

Published 2009 in Annual Conference on Genetic and Evolutionary Computation

ABSTRACT

Evolutionary Algorithms are believed to be relatively robust on noisy objective functions, but generally stagnate in the (later) stages of the evolution process when the population has zoomed in on a particular area of the search space when the noise ratio becomes too large compared to the differences in fitness. The occurrence of stagnation in the search process has been proven for Evolution Strategies using a constant number of repeated samples (resampling size) to evaluate individuals. To prevent stagnation and speed-up convergence, a straightforward and appealing idea is to use the Student's t-test for deciding on the number of individuals to sample in one generation. This paper seeks to study this strategy for (1,lambda)-ES on the noisy sphere model. Besides showing gains achieved with such an adaptive approach in the early stage of runs we also show its limitations: Stagnation cannot be prevented in the long run. Additional studies aim to explain these results.

PUBLICATION RECORD

  • Publication year

    2009

  • Venue

    Annual Conference on Genetic and Evolutionary Computation

  • Publication date

    2009-07-08

  • Fields of study

    Mathematics, Computer Science

  • Identifiers
    DOI 10.1145/1570256.1570376
  • External record

    Open on Semantic Scholar

  • Source metadata

    Semantic Scholar

CITATION MAP

EXTRACTION MAP

CLAIMS

  • No claims are published for this paper.

CONCEPTS

  • No concepts are published for this paper.

REFERENCES

Showing 1-12 of 12 references · Page 1 of 1

CITED BY