Dynamic compartmental models for algorithm analysis and population size estimation

Hugo Monzón (1, 2) , Hernan Aguirre (1, 2) , Sébastien Verel (3) , Arnaud Liefooghe (4, 5, 6) , Bilel Derbel (5, 6) , Kiyoshi Tanaka (1, 2)

1 Faculty of Engineering [Nagano]
2 Shinshu University [Nagano]
3 LISIC - Laboratoire d'Informatique Signal et Image de la Côte d'Opale
4 LIFL - Laboratoire d'Informatique Fondamentale de Lille
5 BONUS - Optimisation de grande taille et calcul large échelle
6 Université de Lille, Sciences et Technologies

Dynamic Compartmental Models (DCM) can be used to study the population dynamics of Multi- and Many-objective Optimization Evolutionary Algorithms (MOEAs). These models track the composition of the instantaneous population by grouping them in compartments and capture their behavior in a set of values, creating a compact representation for analysis and comparison of algorithms. Furthermore, the use of DCMs is not limited to analysis, by creating models of the same algorithm with different configurations is possible to extract new models by interpolation, and use them to explore fine-grained configurations lying between the ones used as a base. We illustrate the use of the model on some Multi- and Many-objective algorithms, run on enumerable MNK-Landscapes instances with 6 objectives for the analysis, and 5 objectives when used as a tool to do configuration.

Mots clés: Empirical study; Working principles of evolutionary computing; Genetic algorithms; Multi-objective optimization; Compartmental Models; Modeling

Domaines: Optimisation et contrôle [math.OC]

Hugo Monzón, Hernan Aguirre, Sébastien Verel, Arnaud Liefooghe, Bilel Derbel, et al.. Dynamic compartmental models for algorithm analysis and population size estimation. GECCO 2019 - Genetic and Evolutionary Computation Conference, Jul 2019, Prague, Czech Republic. pp.2044-2047, ⟨10.1145/3319619.3326912⟩. ⟨hal-02436226⟩