Titre :

A fractal-based decomposition framework for continuous optimization

Auteur(s) :

Firmin, Thomas [Auteur correspondant]
Centre de Recherche en Informatique, Signal et Automatique de Lille - UMR 9189 [CRIStAL]
Talbi, El-Ghazali [Auteur]
Optimisation de grande taille et calcul large échelle [BONUS]
Université de Lille
Centre de Recherche en Informatique, Signal et Automatique de Lille - UMR 9189 [CRIStAL]
Inria Lille - Nord Europe

Mot(s)-clé(s) en anglais :

Continuous optimization
Metaheuristic
High-dimensional optimization
Decomposition
Fractal
Tree search

Discipline(s) HAL :

Informatique [cs]/Autre [cs.OH]

Résumé en anglais : [en]

In this paper, we propose a generic algorithmic framework which defines a unified view of fractal decomposition algorithms for continuous optimization. Fractals allow building a hierarchical decomposition of the decision ...
Lire la suite >In this paper, we propose a generic algorithmic framework which defines a unified view of fractal decomposition algorithms for continuous optimization. Fractals allow building a hierarchical decomposition of the decision space by using a self-similar geometrical object. The proposed generic framework is made of five distinct and independent search components: fractal geometrical object, tree search, scoring, exploration and exploitation. The genericity of the framework allowed the instantiation of popular algorithms from the optimization, machine learning and computational intelligence communities. Moreover, new optimization algorithms can be designed using various strategies of the search components. This shows the modularity of the proposed algorithmic framework. The computational experiments emphasize the behaviors of fractal-based approaches in terms of scalability, robustness, and the balance between exploitation and exploration in the search space. The obtained results show the significance of each search component of the fractal framework, and the necessity to build harder and well-defined benchmarks which can assess the performance of deterministic, axis-aligned and symmetrical decomposition-based algorithms.Lire moins >

Langue :

Anglais

Projet ANR :

Programme de formation doctorale en IA à Lille

Collections :

• Centre de Recherche en Informatique, Signal et Automatique de Lille (CRIStAL) - UMR 9189

Source :

Harvested from HAL