Fitness Landscape and Evolutionary Algorithm Characterization - GII/JEAF GitHub Wiki

One of the research lines of the GII group is devoted to the analysis and characterization of fitness landscapes and evolutionary algorithms. In this section, the main practical results obtained in this field will be explained.

The main objective of this research line is the development of an usable and reliable characterization methodology of the different Evolutionary Algorithms (EAs) so that users are able to choose the most suitable one for a given problem. This methodology will allow minimizing the typical trial and error stage in which researchers apply different EAs, versions of the algorithms, or sets of parameters to the functions they want to optimize without really taking into account their suitability to the particular features of those functions or problems. This trial and error procedure is usually due to a lack of objective characterizations of the algorithms in the literature in terms of the types of functions they are well suited to handle and, more importantly, which types they are not appropriate for.

Research in this field results on several works published in journals and conferences. Here, these publications are listed:

• Pilar Caamaño, Francisco Bellas, Jose A. Becerra, Richard J. Duro, Evolutionary algorithm characterization in real parameter optimization problems, Applied Soft Computing, Volume 13, Issue 4, April 2013, Pages 1902-1921, ISSN 1568-4946 ( URL )
• Caamano, P.; Bellas, F.; Becerra, J.A.; Diaz, V.; Duro, R.J., "Experimental analysis of the relevance of fitness landscape topographical characterization," Evolutionary Computation (CEC), 2012 IEEE Congress on , vol., no., pp.1,8, 10-15 June 2012 ( URL )
• Pilar Caamaño, Jose A. Becerra, Francisco Bellas, and Richard J. Duro. 2011. Are evolutionary algorithm competitions characterizing landscapes appropriately. In Proceedings of the 13th annual conference companion on Genetic and evolutionary computation (GECCO '11), Natalio Krasnogor (Ed.). ACM, New York, NY, USA, 695-702. ( URL )
• P. Caamaño, A. Prieto, J. A. Becerra, F. Bellas, and R. J. Duro. 2010. Real-valued multimodal fitness landscape characterization for evolution. In Proceedings of the 17th international conference on Neural information processing: theory and algorithms - Volume Part I (ICONIP'10), Kok Wai Wong, B. Sumudu U. Mendis, and Abdesselam Bouzerdoum (Eds.), Vol. Part I. Springer-Verlag, Berlin, Heidelberg, 567-574. ( URL )
• P. Caamano, F. Bellas, J. A. Becerra, and R. J. Duro. 2008. Application domain study of evolutionary algorithms in optimization problems. In Proceedings of the 10th annual conference on Genetic and evolutionary computation (GECCO '08), Maarten Keijzer (Ed.). ACM, New York, NY, USA, 495-502. ( URL )

The work developed provides insights on some relevant features of the functions used for the characterization of the algorithms and also defines some performance and error measures that may be helpful for the user and the researcher that is characterizing an EA. This way a fairer and more focused and detailed comparison of the different alternatives for a given problem may be established rather than trying to determine which algorithm is better overall, as done in so much of the current literature.

To illustrate the whole procedure, an exhaustive and formal comparison test has been carried out using some representative EAs and a real-parameter continuous benchmark function set, where the functions are categorized in terms of two basic features: separability and modality. This comparison is used to establish the basic principles of a procedure for the formal characterization of EAs and the definition of their application domains.