Comparative Analysis of Statistical and Information Measures for Benchmark Fitness Landscapes

Proceedings of the UkSIM Fourth European Modelling Symposium on Computer Modelling and Simulation 2010
Gaļina Merkurjeva, Vitālijs Boļšakovs

Various techniques of fitness landscape analysis for the determination of optimisation problem hardness for evolutionary algorithms are proposed in the literature. However, a few implementations of these techniques and their application in practice are described nowadays. In this paper fitness landscapes of benchmark fitness functions are analysed. Statistical and information measures of fitness landscapes are briefly described. Measures of both types are calculated and interpreted for this benchmark landscapes. Detailed calculations of performed experiments are provided and analysed in this paper. The obtained results will allow estimating hardness of different potential optimisation problems for application of the evolutionary algorithms. Conclusions about the influence of conditions and parameters of the fitness landscape analysis techniques on the results that could be achieved also are made.

Atslēgas vārdi
Fitness landscape; fitness landscape analysis; information measures; statistical measures; genetic algorithm
DOI
10.1109/EMS.2010.27
Hipersaite
http://www.computer.org/csdl/proceedings/ems/2010/4308/00/4308a096-abs.html

Merkurjeva, G., Boļšakovs, V. Comparative Analysis of Statistical and Information Measures for Benchmark Fitness Landscapes. No: Proceedings of the UkSIM Fourth European Modelling Symposium on Computer Modelling and Simulation, Itālija, Pīza, 17.-19. novembris, 2010. Los Alamitos: IEEE Computer Society Conference Publishing Services, 2010, 96.-101.lpp. ISBN 978-0-7695-4308-6. Pieejams: doi:10.1109/EMS.2010.27

Publikācijas valoda
English (en)