Felix Streichert, Holger Ulmer, and Andreas Zell

Hybrid Representation for Compositional Optimization and Parallelizing MOEAs

Dagstuhl Seminar Proceedings on Practical Approaches to Multi-Objective Optimization


Abstract

In many real-world optimization problems sparse solution vectors are often preferred. Unfortunately, evolutionary algorithms can have problems to eliminate certain components completely, especially in multi-modal or neutral search spaces. A simple extension of the real-valued representation enables evolutionary algorithms to solve these types of optimization problems more efficiently. In case of multi-objective optimization some of these compositional optimization problems show most peculiar structures on the Pareto front. Here, the Pareto front is often non-convex and consists of multiple local segments. This feature invites parallelization based on the 'divide and conquer' principle, since subdivision into multiple local multi-objective optimization problems seems to be feasible. Therefore, we introduce a new parallelization scheme for multi-objective evolutionary algorithms based on clustering.


Download

[pdf]


BibTeX

    
@InProceedings{Streichert05Hybrid,  
	author =      {Felix Streichert and Holger Ulmer and Andreas Zell},  
	title =       {Hybrid Representations for Composition Optimization and Parallelizing MOEAs},  
	booktitle =   {Practical Approaches to Multi-Objective Optimization},  
	year =        {2005},  
	editor =      {J{"u}rgen Branke and Kalyanmoy Deb and Kaisa Miettinen and Ralph E. Steuer},  
	number =      {04461},  
	series =      {Dagstuhl Seminar Proceedings},  
	publisher =   {Internationales Begegnungs- und Forschungszentrum (IBFI), Schloss Dagstuhl, Germany},  
	note =        {$<$http://drops.dagstuhl.de/opus/volltexte/2005/251$>$},  
}