Fractal Dimension and Perturbation Strength: A Local Optima Networks View - Archive ouverte HAL
Conference Papers Year : 2022

Fractal Dimension and Perturbation Strength: A Local Optima Networks View

Abstract

We study the effect of varying perturbation strength on the fractal dimensions of Quadratic Assignment Problem (QAP) fitness landscapes induced by iterated local search (ILS). Fitness landscapes are represented as Local Optima Networks (LONs), which are graphs mapping algorithm search connectivity in a landscape. LONs are constructed for QAP instances and fractal dimension measurements taken from the networks. Thereafter, the interplay between perturbation strength, LON fractal dimension, and algorithm difficulty on the underlying combinatorial problems is analysed. The results show that higher-perturbation LONs also have higher fractal dimensions. ILS algorithm performance prediction using fractal dimension features may benefit more from LONs formed using a high perturbation strength; this model configuration enjoyed excellent performance. Around half of variance in Robust Taboo Search performance on the data-set used could be explained with the aid of fractal dimension features.
Fichier principal
Vignette du fichier
PPSN_2022__Fractal_Dimension_and_Perturbation_Strength.pdf (3.63 Mo) Télécharger le fichier
Origin Files produced by the author(s)

Dates and versions

hal-03824159 , version 1 (21-10-2022)

Identifiers

Cite

Sarah Thomson, Gabriela Ochoa, Sébastien Verel. Fractal Dimension and Perturbation Strength: A Local Optima Networks View. 17th International Conference on Parallel Problem Solving from Nature (PPSN XVII), Sep 2022, Dortmund, Germany. pp.562-574, ⟨10.1007/978-3-031-14714-2_39⟩. ⟨hal-03824159⟩
28 View
49 Download

Altmetric

Share

More