TY - GEN
T1 - Distribution of computational effort in parallel MOEA/D
AU - Durillo, Juan J.
AU - Zhang, Qingfu
AU - Nebro, Antonio J.
AU - Alba, Enrique
PY - 2011
Y1 - 2011
N2 - MOEA/D is a multi-objective optimization algorithm based on decomposition, which consists in dividing a multi-objective problem into a number of single-objective sub-problems. This work presents two variants, called pMOEA/Dv1 and pMOEA/Dv2, of a new parallel model of MOEA/D that have been developed under the observation that different sub-problems may require different computational effort, and thus, demand different number of evaluations. Our interest in this paper is to analyze whether the proposed models are able of outperforming the MOEA/D in terms of the quality of the computed fronts. To cope with this issue, our proposals have been evaluated using a benchmark composed of eight problems and the obtained results have been compared against MOEA/D-DE, an extension of the original MOEA/D where new individuals are generated by an operator taken from differential evolution. Our experiments show that some configurations of pMOEA/Dv1 and pMOEA/Dv2 have been able to compute fronts of higher quality than MOEA/D-DE in many of the evaluated problems, giving room for further research in this line. © Springer-Verlag Berlin Heidelberg 2011.
AB - MOEA/D is a multi-objective optimization algorithm based on decomposition, which consists in dividing a multi-objective problem into a number of single-objective sub-problems. This work presents two variants, called pMOEA/Dv1 and pMOEA/Dv2, of a new parallel model of MOEA/D that have been developed under the observation that different sub-problems may require different computational effort, and thus, demand different number of evaluations. Our interest in this paper is to analyze whether the proposed models are able of outperforming the MOEA/D in terms of the quality of the computed fronts. To cope with this issue, our proposals have been evaluated using a benchmark composed of eight problems and the obtained results have been compared against MOEA/D-DE, an extension of the original MOEA/D where new individuals are generated by an operator taken from differential evolution. Our experiments show that some configurations of pMOEA/Dv1 and pMOEA/Dv2 have been able to compute fronts of higher quality than MOEA/D-DE in many of the evaluated problems, giving room for further research in this line. © Springer-Verlag Berlin Heidelberg 2011.
UR - https://www.scopus.com/pages/publications/84868560197
UR - https://www.scopus.com/record/pubmetrics.uri?eid=2-s2.0-84868560197&origin=recordpage
U2 - 10.1007/978-3-642-25566-3_38
DO - 10.1007/978-3-642-25566-3_38
M3 - RGC 32 - Refereed conference paper (with host publication)
SN - 9783642255656
VL - 6683 LNCS
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 488
EP - 502
BT - Learning and Intelligent Optimization
PB - Springer Verlag
T2 - 5th International Conference on Learning and Intelligent Optimization, LION 2011
Y2 - 17 January 2011 through 21 January 2011
ER -