Hybrid Representations for Composition Optimization and Parallelizing MOEAs


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 realvalued 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 of 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.

Extracted Key Phrases

9 Figures and Tables

Cite this paper

@inproceedings{Streichert2005HybridRF, title={Hybrid Representations for Composition Optimization and Parallelizing MOEAs}, author={Felix Streichert and Holger Ulmer and Andreas Zell}, booktitle={Practical Approaches to Multi-Objective Optimization}, year={2005} }