Palestra: Selection in Evolutionary Multiobjective Optimization

11/09/2015 - 14:00
Sala 351- IC-3,5

                              TODOS SÃO BEM VINDOS!
                Universidade Estadual de Campinas - UNICAMP
                                   Instituto de Computação - IC
                    PALESTRA DA SÉRIE DE SEMINÁRIOS'2015 
                   Sexta-feira, 11/09/2015, às 14:00 horas
                                           Sala 351- IC-3,5 


                                       Prof. Carlos M. Fonseca
                 CISUC, Department of Informatics Engineering, 
                                 University of Coimbra. Portugal



      The potential of evolutionary algorithms in multiobjective
      optimization was identified early in their history. That
      potential has been realized over the years with the development
      of increasingly elaborate Evolutionary Multiobjective
      Optimization (EMO) algorithms that have found many important
      applications in the real world, and have contributed
      significantly to the growth in popularity of multiobjective
      optimization in general.

      Although EMO has traditionally emphasized the approximation of
      the whole Pareto-optimal front in an a posteriori articulation
      of preferences setting, preference-driven EMO algorithms capable
      of handling interaction with a Decision Maker (DM) were proposed
      early in EMO development. While the identification of a most
      preferred solution is usually seen as the ultimate goal in
      practice, recognizing that the search for diverse sets of
      alternative solutions to be presented to the DM (whether in a
      progressive or an a posteriori articulation of preferences
      scenario) implies some sort of set-oriented preference for
      diversity has been a turning point in EMO algorithm
      development. State-of-the-art algorithms such as IBEA, SMS-EMOA,
      MO-CMA-ES and HyPE, for example, implement multiobjective
      selection based on a notion of set quality that is then used to
      infer the quality of the individuals in the population and to
      introduce bias towards the better ones at the parental and/or
      environmental selection stages. However, how to combine such
      set-oriented preferences with the more traditional search for a
      single most-preferred solution remains largely an open question.

      This talk focuses on the problem of selecting a diverse subset
      of non-dominated solutions from a larger set of candidate
      solutions according to DM preference information. The expression
      of set-oriented preferences by the DM, their incorporation in
      EMO algorithms, and computational aspects of the resulting
      subset selection problems are considered. Existing
      quality-indicator and decomposition approaches are reviewed and
      discussed, and an alternative perspective is introduced where
      set quality is not specified as such by the DM, but is inferred
      from the uncertainty associated with DM solution-oriented
      preferences instead. Recent results obtained by instantiating
      this idea in the form of a portfolio optimization problem are
      presented and discussed, and opportunities for further work are
      outlined at the end.



        Carlos M. Fonseca is an Associate Professor at the Department
        of Informatics Engineering of the University of Coimbra,
        Portugal, and the Head 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. He was a Research Associate with the Department
        of Automatic Control and Systems Engineering of the University
        of Sheffield from 1994 until 1997, and a Lecturer at the
        Department of Electronic Engineering and Informatics, Faculty
        of Science and Technology, University of Algarve, Faro,
        Portugal, from 1997 until October 2010, when he joined the
        University of Coimbra.

        His research has been devoted mainly to evolutionary
        computation and multi-objective optimization. In the 1990's,
        he proposed MOGA, a "first-generation" multi-objective
        evolutionary algorithm with support for progressive preference
        articulation through goals and priorities, and began the
        development of the attainment-function approach to the
        experimental evaluation of stochastic multi-objective
        optimization algorithms. Since then, he has focused on the
        study of evolutionary algorithm dynamics, including
        representation and convergence aspects; further development of
        statistical methodologies for the experimental evaluation of
        multi-objective optimization algorithms, and of efficient
        algorithms to support them; and the development of new,
        computationally efficient, approaches to preference
        articulation in evolutionary multi-objective optimization. He
        has also contributed to several applications in the
        engineering and management domains.

        He was a General co-Chair of the International Conference on
        Evolutionary Multi-Criterion Optimization (EMO) in 2003, 2009
        and 2013, and a Technical co-Chair of the IEEE Congress on
        Evolutionary Computation (CEC) in 2000 and 2005. He is a
        member of the Evolutionary Multi-Criterion Optimization
        Steering Committee and a member of the International Society
        on Multiple Criteria Decision Making (MCDM), the Portuguese
        Operations Research Association (APDIO), and the Portuguese
        Association of Automatic Control.

        Responsável: Profa. Ariadne M. B. R. Carvalho
        Fone: (19) 3521-5864
        Instituto de Computação, Unicamp