Investigation of Archiving Techniques for Evolutionary Multi-objective Optimizers

Investigation of Archiving Techniques for Evolutionary Multi-objective Optimizers

Abstract: The optimization of multi-objective problems from the Pareto dominance viewpoint can lead to huge sets of incomparable solutions. Many heuristic techniques proposed to these problems have to deal with approximation sets that can be limited or not. Usually, a new solution generated by a heu...

Saved in:
Journal Title: Revista de Informática Teórica e Aplicada
First author: Hudson Geovane de Medeiros
Other Authors: Elizabeth Ferreira Gouvêa Goldbarg;
Marco Cesar Goldbarg
Traslated keyword:
Language: English
Get full text: https://seer.ufrgs.br/rita/article/view/RITA_Vol25_Nr4_11
Resource type: Journal Article
Source: Revista de Informática Teórica e Aplicada; Vol 25, No 4 (Year 2018).
DOI: http://dx.doi.org/10.22456/2175-2745.80478
Publisher: Universidade Federal do Rio Grande do Sul
Usage rights: Reconocimiento - NoComercial - SinObraDerivada (by-nc-nd)
Categories: Physical/Engineering Sciences --> Computer Science, Cybernetics
Physical/Engineering Sciences --> Computer Science, Hardware --AMP-- Architecture
Health Sciences --> Medical Informatics
Abstract: Abstract: The optimization of multi-objective problems from the Pareto dominance viewpoint can lead to huge sets of incomparable solutions. Many heuristic techniques proposed to these problems have to deal with approximation sets that can be limited or not. Usually, a new solution generated by a heuristic is compared with other archived non-dominated solutions generated previously. Many techniques deal with limited size archives, since comparisons within unlimited archives may require significant computational effort. To maintain limited archives, solutions need to be discarded. Several techniques were proposed to deal with the problem of deciding which solutions remain in the archive and which are discarded. Previous investigations showed that those techniques might not prevent deterioration of the archives. In this study, we propose to store discarded solutions in a secondary archive and, periodically, recycle them, bringing them back to the optimization process. Three recycling techniques were investigated for three known methods. The datasets for the experiments consisted of 91 instances of discrete and continuous problems with 2, 3 and 4 objectives. The results showed that the recycling method can benefit the tested optimizers on many problem classes.