Contact Us | Language: čeština English
Title: | Steady success clusters in Differential Evolution |
Author: | Viktorin, Adam; Šenkeřík, Roman; Pluháček, Michal; Zamuda, Aleš |
Document type: | Peer-reviewed article (English) |
Source document: | 2016 IEEE Symposium Series on Computational Intelligence, SSCI 2016. 2016 |
ISBN: | 978-1-5090-4240-1 |
DOI: | https://doi.org/10.1109/SSCI.2016.7850252 |
Abstract: | This paper presents insights into the proportions between the k-means clusters of successful Differential Evolution (DE), donor generating vectors. This is demonstrated by the high certainty that these proportions are similar - and thereby, that these cluster size proportions regularly appear. A characteristic of these proportions is that they are observed at the same specific values in different test functions. It is also shown that, when varying the number of dimensions for a fitness function, the proportions are constant. However, some of the possible dynamics of these proportions are reported later on, in the situation where the optimization algorithm is changed - for instance, control parameters like the population size parameter. This parameter significantly changes, the proportions of the most frequently successful and most unsuccessful vectors. Insights like this are useful for an understanding of the inter-generational complexity that appears within evolutionary algorithms and would thus benefit future algorithm design; for example, plausible metrics for on-line control. © 2016 IEEE. |
Full text: | http://ieeexplore.ieee.org/document/7850252/ |
Show full item record |