Speakers - Carlos M. Fonseca
Article Index
Page 2 of 7
| An Integrated View of Selection in Evolutionary Algorithms | |
 |
Carlos M. Fonseca University of Coimbra Talk Abstract: Selection plays a double role in evolutionary algorithms. The selection of solutions from the current solution set (or population) to produce new candidate (offspring) solutions through variation is known as parental selection, whereas the selection of solutions to discard in order to make room for new solutions in the population is usually called environmental selection. Intimately related to selection is the concept of solution fitness, which is typically related to the objective function(s) of the problem at hand. In practice, both types of selection must be implemented, although different evolutionary algorithms often emphasize either one or the other. In particular, it is common for only one type of selection to depend on fitness. Selection also has the double purpose of steering the search towards more promising regions of the search space by favouring the best solutions available (exploitation) while maintaining a sufficient level of diversity in order to be able to escape local optima and/or find multiple good solutions (exploration). Over the years, many different approaches to selection in evolutionary algorithms have been proposed in the literature, with the balance between exploration and exploitation gaining heightened importance in the context of multiobjective optimization. However, parental and environmental selection have continued to be seen as different operators, and to be implemented separately. In this talk, selection methods and fitness assignment strategies are reviewed and discussed from the unifying perspective of portfolio optimization, where the fitness of a solution is interpreted as an investment in that solution, and solution diversity emerges naturally from the need to balance return and risk in the portfolio. In addition, parental and environmental selection can be seamlessly integrated in the portfolio optimization formulation. Application examples illustrate the main aspects of the approach. Bio: Carlos M. Fonseca is an Associate Professor at the Department of Informatics Engineering of the University of Coimbra, Portugal, and a member of the Evolutionary and Complex Systems (ECOS) group of the Centre for Informatics and Systems of the University of Coimbra (CISUC). He graduated in Electronic and Telecommunications Engineering from the University of Aveiro, Portugal, in 1991, and obtained a Ph.D. in Automatic Control and Systems Engineering from the University of Sheffield, U.K., in 1996. His research has been devoted mainly to evolutionary computation and multi-objective optimization, with a focus on computationally efficient approaches to preference articulation and experimental performance evaluation in evolutionary multi-objective optimization. He is the Scientific Representative of the Grant Holder of COST Action CA15140 – Improving Applicability of Nature-Inspired Optimisation by Joining Theory and Practice (ImAppNIO), and the leader of a Working Group on Software in that Action. He has served as General, Technical or Track co-Chair of several major international conferences on evolutionary computation, and is a member of the Evolutionary Multi-Criterion Optimization and of the Parallel Problem Solving from Nature Steering Committees. More information about his research contribution can be found from https://eden.dei.uc.pt/~cmfonsec/#bio. |