Towards a Matrix-free Covariance Matrix Adaptation Evolution Strategy

Jarosław Arabas , Dariusz Jagodziński

Abstract

In this paper, we discuss a method for generating new individuals such that their mean vector and the covariance matrix are defined by formulas analogous to the Covariance Matrix Adaptation Evolution Strategy (CMA-ES). In contrast to CMA-ES, which generates new individuals using multivariate Gaussian distribution with an explicitly defined covariance matrix, the introduced method uses combinations of difference vectors between archived individuals and univariate Gaussian random vectors along directions of past shifts of the population midpoints. We use this method to formulate the Differential Evolution Strategy (DES) – an algorithm that is a crossover between Differential Evolution (DE) and CMA-ES. The numerical results presented in the paper indicate that DES is competitive against CMA-ES in performing both local and global optimization.
Author Jarosław Arabas (FEIT / IN)
Jarosław Arabas,,
- The Institute of Computer Science
, Dariusz Jagodziński (FEIT / IN)
Dariusz Jagodziński,,
- The Institute of Computer Science
Journal seriesIEEE Transactions On Evolutionary Computation, ISSN 1089-778X, (A 50 pkt)
Issue year2019
Pages1-6
Publication size in sheets0.3
Keywords in EnglishCovariance matrices , Sociology , Optimization , History , Gaussian distribution , Indexes
ASJC Classification1703 Computational Theory and Mathematics; 2614 Theoretical Computer Science; 1712 Software
DOIDOI:10.1109/TEVC.2019.2907266
URL http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8673614&isnumber=4358751
projectDevelopment of new algorithms in the areas of software and computer architecture, artificial intelligence and information systems and computer graphics . Project leader: Arabas Jarosław, , Phone: +48 22 234 7432, start date 01-08-2018, planned end date 30-09-2019, II/2018/DS/1, Implemented
WEiTI Działalność statutowa
Languageen angielski
Score (nominal)50
ScoreMinisterial score = 50.0, 22-04-2019, ArticleFromJournal
Ministerial score (2013-2016) = 50.0, 22-04-2019, ArticleFromJournal
Publication indicators Scopus SNIP (Source Normalised Impact per Paper): 2017 = 4.724; WoS Impact Factor: 2017 = 8.124 (2) - 2017=8.481 (5)
Citation count*
Cite
Share Share

Get link to the record


* presented citation count is obtained through Internet information analysis and it is close to the number calculated by the Publish or Perish system.
Back