From fitness landscape analysis to designing evolutionary algorithms: the case study in automatic generation of function block applications

Vladimir Mironovich, Maxim Buzdalov, Valeriy Vyatkin

Research output: Chapter in Book/Report/Conference proceedingConference Proceeding (Non-Journal item)

Abstract

Search-based software engineering, a discipline that often requires finding optimal solutions, can be a viable source for problems that bridge theory and practice of evolutionary computation. In this research we consider one such problem: generation of data connections in a distributed control application designed according to the IEC 61499 industry standard.

We perform the analysis of the fitness landscape of this problem and find why exactly the simplistic (1 + 1) evolutionary algorithm is slower than expected when finding an optimal solution to this problem. To counteract, we develop a population-based algorithm that explicitly maximises diversity among the individuals in the population. We show that this measure indeed helps to improve the running times.
Original languageEnglish
Title of host publicationGECCO '18
Subtitle of host publicationProceedings of the Genetic and Evolutionary Computation Conference Companion
EditorsHernan Aguirre
PublisherAssociation for Computing Machinery, Inc
Pages1902-1905
Number of pages4
ISBN (Print)978-1-4503-5764-7
DOIs
Publication statusPublished - 06 Jul 2018
Externally publishedYes
EventGECCO 2018: The Genetic and Evolutionary Computation Conference - Kyoto, Japan
Duration: 15 Jul 201819 Jul 2018
http://gecco-2018.sigevo.org

Conference

ConferenceGECCO 2018: The Genetic and Evolutionary Computation Conference
Country/TerritoryJapan
CityKyoto
Period15 Jul 201819 Jul 2018
Internet address

Keywords

  • evolutionary computation
  • population diversity
  • program synthesis
  • search-based software engineering

Fingerprint

Dive into the research topics of 'From fitness landscape analysis to designing evolutionary algorithms: the case study in automatic generation of function block applications'. Together they form a unique fingerprint.

Cite this