Boundary Decomposition for Nadir Objective Vector Estimation

Authors: Ruihao Zheng, Zhenkun Wang

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We implement BDNE for black-box MOPs and use 28 black-box problems to validate its performance. The results indicate that BDNE remarkably outperforms the existing methods.
Researcher Affiliation Academia Ruihao Zheng Zhenkun Wang School of System Design and Intelligent Manufacturing, Southern University of Science and Technology 12132686@mail.sustech.edu.cn, wangzhenkun90@gmail.com
Pseudocode Yes Algorithm 1 BDNE
Open Source Code Yes The source code is available at https://github.com/Eric Zheng1024/BDNE.
Open Datasets Yes Moreover, we select problems with different shapes of feasible objective regions, including 6 existing test problems [46, 47, 48, 49] (DTLZ3, m DTLZ3, Ma F2, DTLZ5, IMOP4, and IMOP6) and 4 real-world problems [50, 51, 52] (MP-DMP, ML-DMP, RE3-4-7, and RE5-3-1).
Dataset Splits No The paper does not explicitly provide percentages or counts for training, validation, and test dataset splits. The datasets used are multi-objective optimization problems, which are typically evaluated as a whole rather than with predefined splits for training, validation, and testing like in supervised learning.
Hardware Specification Yes Experiments are executed on a computer equipped with two 3.00-GHz Intel Xeon Gold 6248R CPUs (48 cores in total), an NVIDIA T400 GPU, and 128GB of RAM.
Software Dependencies No Our codes rely on the MATLAB-based platform called Plat EMO [59], which is freely available for research purposes (see https://github.com/BIMK/Plat EMO).
Experiment Setup Yes For BDNE, we set µ = 100 and τl = 200. Then τu = 14 according to the setting of τl and FEmax. The general algorithm settings are summarized in Table 2, where FEmax means maximum number of function evaluations. Table 2: General algorithm settings. Symbol Description τu(τl) Maximum number of iterations for each CMA-ES procedure (the MOEA solving LLOPs). N N m Z; number of generated LLOPs in each iteration of m ULOPs. P |P| = N; population of the MOEA for solving N LLOPs. A |A| = N; elite archive preserved the current best solutions for N LLOPs. wi,j The j-th boundary weight vector of the i-th objective. m N FEmax Operator 3 60 180,000 SBX+PM [53] 5 100 300,000 8 160 480,000