Hill-climbing strategies on various landscapes

TitreHill-climbing strategies on various landscapes
Type de publicationCommunication
TypeCommunication avec actes dans un congrès
Date du colloque06-10/07/2013
Titre du colloqueProceeding of the fifteenth annual conferenceProceeding of the fifteenth annual conference on Genetic and evolutionary computation conference - GECCO '13
AuteurBasseur, Matthieu , Goëffon, Adrien
EditeurACM Press
VilleAmsterdam, The NetherlandsNew York, New York, USA
Résumé en anglais

Climbers constitute a central component of modern heuristics, including metaheuristics, hybrid metaheuristics and hyperheuristics. Several important questions arise while designing a climber, and choices are often arbitrary, intuitive or experimentally decided. The paper provides guidelines to design climbers considering a landscape shape under study. In particular, we aim at competing best improvement and first improvement strategies, as well as evaluating the behavior of different neutral move policies. Some conclusions are assessed by an empirical analysis on a large variety of landscapes. This leads us to use the NK-landscapes family, which allows to define landscapes of different size, rugosity and neutrality levels. Experiments show the ability of first improvement to explore rugged landscapes, as well as the interest of accepting neutral moves at each step of the search. Moreover, we point out that reducing the precision of a fitness function could help to optimize problems.

URL de la noticehttp://okina.univ-angers.fr/publications/ua7667
Lien vers le document en ligne