A Generalized Scalarization Method for Evolutionary Multi-Objective Optimization

Authors: Ruihao Zheng, Zhenkun Wang

AAAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The experimental studies on various MOPs conform to the theoretical analysis.
Researcher Affiliation Academia School of System Design and Intelligent Manufacturing, Southern University of Science and Technology, Shenzhen, China
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes 1https://github.com/Eric Zheng1024/MOEA-D-GGR
Open Datasets Yes We use ZDT1-ZDT4 (Zitzler, Deb, and Thiele 2000), DTLZ1, DTLZ3 and DTLZ5 (Deb et al. 2005), the multi-objective knapsack problem (MOKP) (Zitzler and Thiele 1999), and the multi-objective traveling salesman problem (MOTSP) (Corne and Knowles 2007) to verify the algorithm performance.
Dataset Splits No The paper uses benchmark multi-objective optimization problems as test instances, which are typically not split into distinct training, validation, and testing sets in the same manner as supervised learning datasets. No explicit mention of dataset splits is provided.
Hardware Specification No The paper does not provide specific hardware details such as GPU/CPU models, processor types, or memory amounts used for running its experiments. It only mentions that algorithms are implemented on the PlatEMO platform.
Software Dependencies No All the algorithms are implemented on the Plat EMO platform (Tian et al. 2017). (No version specified for PlatEMO or other software).
Experiment Setup Yes General Algorithm Settings. The population size N: 100 (m = 2) or 190 (m = 3). The maximal number of function evaluations: 25000 for ZDT1-ZDT4, 100000 for DTLZ1, DTLZ3 and DTLZ5, 200000 for 2-objective MOKP and MOTSP, and 400000 for 3-objective MOKP and MOTSP. The number of independent runs: 30 for each instance. The neighborhood size: Tm = 0.1N and Tr = 0.05N. Reproduction operators: For real number coding, the simulated binary crossover (SBX) and polynomial mutation (PM) are used (Purshouse and Fleming 2007). The SBX control parameters pc, ηc and pe are set to 1, 20 and 0, respectively. The PM control parameters pm and ηm are set to 1/n and 20, where n is the number of decision variables. For binary coding, the uniform crossover and bit-flip mutation are used (Syswerda 1989). The crossover rate is 1; the mutation rate is 2/n for a bit. For permutation coding, the order-based crossover and simple inversion mutation are used (Larranaga et al. 1999). The crossover rate is 1 and the mutation rate is 0.1.