Adapting mutation and recombination operators to range-aware relations in real-world application data

Publikationstyp
Conference Paper
Date Issued
2022-07
Sprache
English
Author(s)
Plump, Christina  
Bernhard Johannes Berger  orcid-logo
Drechsler, Rolf  
Institut
Eingebettete Systeme E-13  
TORE-URI
http://hdl.handle.net/11420/13536
Start Page
755
End Page
758
Citation
Genetic and Evolutionary Computation Conference (GECCO 2022)
Contribution to Conference
Genetic and Evolutionary Computation Conference, GECCO 2022  
Publisher DOI
10.1145/3520304.3529066
Scopus ID
2-s2.0-85136335473
Optimisation problems with higher-dimensional search spaces do usually not only come with equality or inequality constraints, but also with dependencies between the different variables. In real-world applications, especially in experimental data from material sciences, these relations as well as the constraints may not be true for the entire search space, but only for certain areas. Other constraints or relations may hold then for different areas of the search space. We build on correlation-aware mutation and recombination operators that are used in genetic algorithms and adjust them to be able to deal with area-specific constraints and relations. This can be configured by a domain expert using a domain-specific language. Our approach is evaluated with well-known benchmark functions, carefully designed distributions, and data description files and shows a better capability of generating feasible solutions in the population.
Subjects
dependencies
domain knowledge
evolutionary algorithm