A framework for hybrid dynamic evolutionary algorithms : multiple offspring sampling (MOS)

LaTorre de la Fuente, Antonio (2009). A framework for hybrid dynamic evolutionary algorithms : multiple offspring sampling (MOS). Thesis (Doctoral), Facultad de Informática (UPM).

Description

Title: A framework for hybrid dynamic evolutionary algorithms : multiple offspring sampling (MOS)
Author/s:
  • LaTorre de la Fuente, Antonio
Contributor/s:
  • Peña Sánchez, José María
Item Type: Thesis (Doctoral)
Date: 2009
Subjects:
Freetext Keywords: Multiple Offspring Sampling, Hybrid Evolutionary Algorithms, Optimization Problems, Reinforcement Learning
Faculty: Facultad de Informática (UPM)
Department: Arquitectura y Tecnología de Sistemas Informáticos
Creative Commons Licenses: Recognition - No derivative works - Non commercial

Full text

[img]
Preview
PDF - Requires a PDF viewer, such as GSview, Xpdf or Adobe Acrobat Reader
Download (11MB) | Preview

Abstract

Evolutionary Algorithms (EAs) are a set of optimization techniques that have become incredibly popular in the last decades. As they are general purpose algorithms, they have been applied to a wide range of problems, many of them from industrial or scientific disciplines. Several approaches have been proposed, each of them implementing the biological metaphor in their own particular way. This provides each of these evolutionary approaches with different search characteristics, which make them more suitable to different types of problems. This diversity of Evolutionary Algorithms makes possible to face a wider range of optimization problems. However, the selection of a particular Evolutionary Algorithm becomes a crucial decision that can determine the quality of the obtained results. Furthermore, some studies show that synergies among different Evolutionary Algorithms are possible when they are combined appropriately. The use of hybrid algorithms to deal with specific and complex real-world problems is also a fact that proves that hybridization is a powerful tool far beyond the individual algorithms. In this work, the combination of different evolutionary approaches is analyzed thanks to a framework that provides a robust and complete support for the development of Hybrid EAs. This framework is called Multiple Offspring Sampling (MOS) and is based on the key concept of a reproductive technique, which offers an abstraction of the mechanisms used by the different evolutionary approaches to create new individuals, i.e., the particular operators, parameters and encodings of the solutions present in the canonical versions of these algorithms. However, it is now the MOS framework, and not the individual algorithms, the one responsible for creating new individuals by means of the available reproductive techniques. The hybrid algorithms developed with the MOS framework can dynamically evaluate the performance of the different reproductive techniques and adjust their participation accordingly. Several strategies have been proposed for the evaluation of the quality of the techniques and the adjustment of their participation, including some of the more classic alternatives present in the literature to assess the convenience of using the mechanisms offered by MOS. Additionally, the automatic learning of these hybridization strategies by means of Reinforcement Learning mechanisms has also been studied. To conclude, the proposed framework has been tested on a set of well-known problems, from both discrete and continuous domains, obtaining statistically meaningful results confirming that an appropriate combination of different search strategies can lead to an outstading performance compared to the individual algorithms.

More information

Item ID: 2169
DC Identifier: http://oa.upm.es/2169/
OAI Identifier: oai:oa.upm.es:2169
Deposited by: Archivo Digital UPM
Deposited on: 03 Feb 2010 15:00
Last Modified: 20 Apr 2016 11:55
  • Logo InvestigaM (UPM)
  • Logo GEOUP4
  • Logo Open Access
  • Open Access
  • Logo Sherpa/Romeo
    Check whether the anglo-saxon journal in which you have published an article allows you to also publish it under open access.
  • Logo Dulcinea
    Check whether the spanish journal in which you have published an article allows you to also publish it under open access.
  • Logo de Recolecta
  • Logo del Observatorio I+D+i UPM
  • Logo de OpenCourseWare UPM