The “One-Fifth Rule” with Rollbacks for Self-Adjustment of the Population Size in the (1 + (λ, λ)) Genetic Algorithm

Self-adjustment of parameters can significantly improve the performance of evolutionary algorithms. A notable example is the (1 + (λ, λ)) genetic algorithm, where adaptation of the population size helps to achieve the linear running time on the OneMax problem. However, on problems which interfere wi...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Automatic control and computer sciences Jg. 55; H. 7; S. 885 - 902
Hauptverfasser: Bassin, A. O., Buzdalov, M. V., Shalyto, A. A.
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Moscow Pleiades Publishing 01.12.2021
Springer Nature B.V
Schlagworte:
Computer Science
Control Structures and Microprogramming
Evolutionary algorithms
Genetic algorithms
Linear functions
Performance degradation
Performance enhancement
parameter adaptation
MAX-3SAT
linear functions
1 + (λ
λ)) GA
ISSN:0146-4116, 1558-108X
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Self-adjustment of parameters can significantly improve the performance of evolutionary algorithms. A notable example is the (1 + (λ, λ)) genetic algorithm, where adaptation of the population size helps to achieve the linear running time on the OneMax problem. However, on problems which interfere with the assumptions behind the self-adjustment procedure, its usage can lead to the performance degradation. In particular, this is the case with the “one-fifth rule” on problems with weak fitness-distance correlation. We propose a modification of the “one-fifth rule” in order to have less negative impact on the performance in the cases where the original rule is destructive. Our modification, while still yielding a provable linear runtime on OneMax, shows better results on linear functions with random weights, as well as on random satisfiable MAX-3SAT problems.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0146-4116
1558-108X
DOI:10.3103/S0146411621070208