Free Search - Comparative Analysis 100

Research output: Contribution to journalArticle

Abstract

Search methods’ abilities for adaptation to various multidimensional tasks where optimisation parameters are hundreds, thousands and more, without retuning of algorithms’ parameters seems to be a great challenge for modern computational intelligence. Many evolutionary, swarm and adaptive methods, which perform well on numerical tests with up to ten dimensions are suffering insuperable stagnation when applied to 100 and more dimensional tests. This article presents a comparison between particle swarm optimisation, differential evolution both with enhanced adaptivity and Free Search applied to 100 multidimensional heterogeneous real-value numerical tests. The aim is to extend the knowledge on how high dimensionality reflects on search space complexity, in particular to identify minimal time and minimal number of objective function evaluations required by used methods for reaching acceptable solution with non-zero probability on tasks with high dimensions’ number. The achieved experimental results are summarised and analysed. Brief discussion on concepts, which support search methods effectiveness, concludes the article.
Original languageEnglish
Pages (from-to)118-132
JournalInternational Journal of Metaheuristics
Volume3
Issue number2
Publication statusPublished - 2014

Fingerprint

Function evaluation
Particle swarm optimization (PSO)
Artificial intelligence

Cite this

@article{a08c600b1db44dc1998e60cb1b3cc724,
title = "Free Search - Comparative Analysis 100",
abstract = "Search methods’ abilities for adaptation to various multidimensional tasks where optimisation parameters are hundreds, thousands and more, without retuning of algorithms’ parameters seems to be a great challenge for modern computational intelligence. Many evolutionary, swarm and adaptive methods, which perform well on numerical tests with up to ten dimensions are suffering insuperable stagnation when applied to 100 and more dimensional tests. This article presents a comparison between particle swarm optimisation, differential evolution both with enhanced adaptivity and Free Search applied to 100 multidimensional heterogeneous real-value numerical tests. The aim is to extend the knowledge on how high dimensionality reflects on search space complexity, in particular to identify minimal time and minimal number of objective function evaluations required by used methods for reaching acceptable solution with non-zero probability on tasks with high dimensions’ number. The achieved experimental results are summarised and analysed. Brief discussion on concepts, which support search methods effectiveness, concludes the article.",
author = "Kalin Penev",
year = "2014",
language = "English",
volume = "3",
pages = "118--132",
journal = "International Journal of Metaheuristics",
number = "2",

}

Free Search - Comparative Analysis 100. / Penev, Kalin.

In: International Journal of Metaheuristics, Vol. 3, No. 2, 2014, p. 118-132.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Free Search - Comparative Analysis 100

AU - Penev, Kalin

PY - 2014

Y1 - 2014

N2 - Search methods’ abilities for adaptation to various multidimensional tasks where optimisation parameters are hundreds, thousands and more, without retuning of algorithms’ parameters seems to be a great challenge for modern computational intelligence. Many evolutionary, swarm and adaptive methods, which perform well on numerical tests with up to ten dimensions are suffering insuperable stagnation when applied to 100 and more dimensional tests. This article presents a comparison between particle swarm optimisation, differential evolution both with enhanced adaptivity and Free Search applied to 100 multidimensional heterogeneous real-value numerical tests. The aim is to extend the knowledge on how high dimensionality reflects on search space complexity, in particular to identify minimal time and minimal number of objective function evaluations required by used methods for reaching acceptable solution with non-zero probability on tasks with high dimensions’ number. The achieved experimental results are summarised and analysed. Brief discussion on concepts, which support search methods effectiveness, concludes the article.

AB - Search methods’ abilities for adaptation to various multidimensional tasks where optimisation parameters are hundreds, thousands and more, without retuning of algorithms’ parameters seems to be a great challenge for modern computational intelligence. Many evolutionary, swarm and adaptive methods, which perform well on numerical tests with up to ten dimensions are suffering insuperable stagnation when applied to 100 and more dimensional tests. This article presents a comparison between particle swarm optimisation, differential evolution both with enhanced adaptivity and Free Search applied to 100 multidimensional heterogeneous real-value numerical tests. The aim is to extend the knowledge on how high dimensionality reflects on search space complexity, in particular to identify minimal time and minimal number of objective function evaluations required by used methods for reaching acceptable solution with non-zero probability on tasks with high dimensions’ number. The achieved experimental results are summarised and analysed. Brief discussion on concepts, which support search methods effectiveness, concludes the article.

M3 - Article

VL - 3

SP - 118

EP - 132

JO - International Journal of Metaheuristics

JF - International Journal of Metaheuristics

IS - 2

ER -