6533b81ffe1ef96bd1277a18

RESEARCH PRODUCT

On the Locality of Standard Search Operators in Grammatical Evolution

Ann ThorhauerFranz Rothlauf

subject

Binary treeTheoretical computer sciencebusiness.industryPerspective (graphical)LocalityEvolutionary algorithmGenetic programmingcomputer.software_genreRandom walkGrammatical evolutionArtificial intelligencebusinesscomputerNatural language processingMathematics

description

Offspring should be similar to their parents and inherit their relevant properties. This general design principle of search operators in evolutionary algorithms is either known as locality or geometry of search operators, respectively. It takes a geometric perspective on search operators and suggests that the distance between an offspring and its parents should be less than or equal to the distance between both parents. This paper examines the locality of standard search operators used in grammatical evolution (GE) and genetic programming (GP) for binary tree problems. Both standard GE and GP search operators suffer from low locality since a substantial number of search steps result in an offspring whose distance to one of its parents is greater than the distance between both of its parents. Furthermore, the locality of standard GE search operators is higher than that of standard GP search operators, which allows more focused search in GE.

https://doi.org/10.1007/978-3-319-10762-2_46