Direct Preference-Based Evolutionary Multi-Objective Optimization with Dueling Bandits

Authors: Tian Huang, Shengbo Wang, Ke Li

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on 48 benchmark test problems, including the RNA inverse design and protein structure prediction, fully demonstrate the effectiveness of our proposed approach.
Researcher Affiliation Academia 1 School of Computer Science and Engineering , University of Electronic Science and Technology of China 2 Department of Computer Science, University of Exeter
Pseudocode Yes Algorithm 1 D-PBEMO structure
Open Source Code Yes The code of our algorithms and peer algorithms are available at https://github.com/COLA-Laboratory/EMOC/.
Open Datasets Yes Our experiments considers 33 synthetic test instances including ZDT1 to ZDT4 and ZDT6 [72] (m = 2), DTLZ1 to DTLZ6 [18] where m = {3, 5, 8, 10}, and WFG1, WFG3, WFG5, and WFG7 [30] (m = 3). ... In addition, we also consider two scientific discovery problems including 10 two-objective RNA inverse design tasks [55, 70] and 5 four-objective protein structure prediction (PSP) task [69].
Dataset Splits No The paper does not provide specific training/validation/test dataset splits (e.g., percentages or sample counts) needed to reproduce the data partitioning.
Hardware Specification No The paper does not provide specific hardware details such as exact GPU/CPU models, memory amounts, or detailed computer specifications used for running its experiments.
Software Dependencies No In our experiment, the MFE is calculate by Vienna RNA3 package. [Footnote 3] https://github.com/Vienna RNA/Vienna RNA. No specific version numbers for software dependencies are provided beyond this mention.
Experiment Setup Yes The probability and distribution of index for SBX: pc = 1.0 and ηc = 20; The mutation probability and distribution of index for polynomial mutation operator: pm = 1 m and ηm = 20; The population size for different problems can be referenced in Table A2; The maximum number of generation G can be referenced in Table A3; For I-MOEA/D-PLVF and I-NSGA2/LTR, the number of incumbent candidate presented to decison maker (DM) for consultation: µ = 10; The budget for dueling bandits algorithm T is set as 100.