Skip to Main content Skip to Navigation
Theses

Algorithmes Heuristiques et Techniques d'Apprentissage : Applications au Problème de Coloration de Graphe

Abstract : The last couple of decades have seen a surge of interest and sophistication in using heuristics to solve combinatorial optimization problems. However, the theoretical and practical research of these algorithms show there are many important challenges yet to be overcome { e.g. it seems that it is quite difficult to make a heuristic integrate a global vision over its own evolution or over its exploration path. Taking the well-known graph coloring problem as an experimental framework, we develop several new heuristics that integrate certain learning mechanisms so as to render the search process more "self-aware". For instance, we introduce an algorithm that is able to record its trajectory and to interpret its own evolution. A search space analysis showed that the best discovered potential solutions tend to occur relatively close to each other, clustered in spheres of xed radius. Using such learned information, we developed : (i) diversication algorithms that "pay attention" not to visit the same sphere repeatedly, (ii) intensication algorithms assuring an in-depth exploration of a closed perimeter using a breath-rst-search traversal of its spheres, or (iii) evolutionary algorithms that are able to keep the individuals sufficiently distant at all times, while not sacricing population quality. In fact, we present numerous other techniques (e.g. new evaluation functions) that are able to render the heuristic search more "well-informed".
Document type :
Theses
Complete list of metadatas

Cited literature [111 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00476541
Contributor : Anne-Marie Plé <>
Submitted on : Monday, April 26, 2010 - 4:14:49 PM
Last modification on : Friday, October 23, 2020 - 4:41:48 PM
Long-term archiving on: : Tuesday, September 28, 2010 - 1:08:24 PM

Identifiers

  • HAL Id : tel-00476541, version 1

Citation

Daniel Cosmin Porumbel. Algorithmes Heuristiques et Techniques d'Apprentissage : Applications au Problème de Coloration de Graphe. Informatique [cs]. Université d'Angers, 2009. Français. ⟨tel-00476541⟩

Share

Metrics

Record views

410

Files downloads

3813