An efficient ant colony optimization strategy for the resolution of multi-class queries
| Autor: | Michael E. Houle, Kamil Krynicki, Javier Jaen |
|---|---|
| Rok vydání: | 2016 |
| Předmět: |
Information Systems and Management
Theoretical computer science Computer science 02 engineering and technology Resource queries ComputingMethodologies_ARTIFICIALINTELLIGENCE Management Information Systems Ant colony optimization Resource (project management) Artificial Intelligence 0202 electrical engineering electronic engineering information engineering Pruning (decision trees) Multi-class queries Multipheromone Flexibility (engineering) business.industry Ant colony optimization algorithms 020206 networking & telecommunications Ant colony Pheromone 020201 artificial intelligence & image processing Artificial intelligence Computational problem business LENGUAJES Y SISTEMAS INFORMATICOS Software |
| Zdroj: | RiuNet. Repositorio Institucional de la Universitat Politécnica de Valéncia instname |
| ISSN: | 0950-7051 |
| DOI: | 10.1016/j.knosys.2016.05.009 |
| Popis: | Ant Colony Optimization is a bio-inspired computational technique for establishing optimal paths in graphs. It has been successfully adapted to solve many classical computational problems, with considerable results. Nevertheless, the attempts to apply ACO to the question of multidimensional problems and multi-class resource querying have been somewhat limited. They suffer from either severely decreased efficiency or low scalability, and are usually static, custom-made solutions with only one particular use. In this paper we employ Angry Ant Framework, a multipheromone variant of Ant Colony System that surpasses its predecessor in terms of convergence quality, to the question of multi-class resource queries. To the best of the authors knowledge it is the only algorithm capable of dynamically creating and pruning pheromone levels, which we refer to as dynamic pheromone stratification. In a series of experiments we verify that, due to this pheromone level flexibility, Angry Ant Framework, as well as our improvement of it called Entropic Angry Ant Framework, have significantly more potential for handling multi-class resource queries than their single pheromone counterpart. Most notably, the tight coupling between pheromone and resource classes enables convergence that is both better in quality and more stable, while maintaining a sublinear cost. © 2016 Elsevier B.V. All rights reserved. Kamil Krynicki is a FPI fellow of Universitat Politecnica de Valencia, number 3117. This work received support from Spanish Ministry of Economy and Competitiveness and European Development Regional Fund (EDRF-FEDER) with the project SUPEREMOS TIN2014-60077-R and the National Institute of Informatics, Tokyo, Japan. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full Text from ScienceDirect