Among many techniques proposed to solve optimization problems, metaheuristics are becoming increasingly popular. Different metaheuristics have emerged over the last two decades, most of them being classified as nature or bio-inspired (giving rise to a topic in the Natural Computing area). Despite their rapid development, there are many challenges, and a lot of open problems still remain: large search spaces (regarding decision variables, number of objectives, restrictions), fitness evaluation (costly, noisy, dynamic), complexity and convergence analyses, parameters setting, and so on. Heuristics, auto-adaptation, parallelism, multiple-population, surrogates, and hybridization are among the alternatives being proposed to cope with these problems. Although there is a wide range of successful applications, the large number of available metaheuristics may turn difficult the choice of the most suitable one for a specific problem. In this way, Metalearning for metaheuristic recommendation appears as an interesting alternative. The seminar will present the current main research conducted by the ComPaNat research group, drawing special attention to metalearning in the optimization context.
Myriam Delgado: PhD in Electrical Engineering from the State University of Campinas (2002), and head of the research group Parallel and Natural Computing (ComPaNat) at Federal University of Technology – Paraná - Brazil. On leave (sabbatical year) for a post-doctoral, partially supported by CNPq/Brazil, and Visiting Senior Lecturer at University of Kent – Canterbury since 01/08/2015, under the supervision of Prof. Alex Freitas. Has experience in Computer Science, with an emphasis in Computational Intelligence: Fuzzy Systems, Swarm Intelligence (Ant Colony and Particle Swarm), Evolutionary Algorithms (especially Estimation of Distribution Algorithms and Differential Evolution) with applications in the areas of Computer Networks, Optimization and Logistics, and Bioinformatics.