Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
FSEO: Few-Shot Evolutionary Optimization via Meta-Learning for Expensive Multi-Objective Optimization
Authors: Xunzhao Yu
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our computational studies can be divided into three parts: 1. Appendix D evaluates our meta-learning model performance on two problems and analyzes model component contributions via ablation comparisons with model variants. 2. Sections 5.1 to 5.2 investigate the performance of our FSEO framework in enhancing sampling efficiency. Extensive ablation studies are conducted to provide guidance for practical applications of our FSEO framework. 3. Section 5.3 and Appendix H demonstrate the performance and broad applicability of our FSEO framework on real-world problems. |
| Researcher Affiliation | Academia | Xunzhao Yu Department of Economics, University of Warwick EMAIL |
| Pseudocode | Yes | Algorithm 1 FSEO Framework. ... Algorithm 2 Meta-learning(Di, Nm, |Dm|, B, α) ... Algorithm 3 Adaptation(γ , S , β) ... Algorithm 4 Update(γ , S , β) |
| Open Source Code | Yes | 1Code is available at https://github.com/Xunzhao Yu/FSEO |
| Open Datasets | Yes | The computational study is conducted on DTLZ test problems [11]. All DTLZ problems have d=10 decision variables and 3 objectives, as the setups that have been widely used in [35]. |
| Dataset Splits | Yes | During the optimization process, an initial dataset S is sampled using Latin Hypercube Sampling (LHS) method [28], then extra evaluations are conducted until the evaluation budget has run out. Note that we aim to use related tasks to save 9d evaluations without a loss of SAEA optimization performance. Hence, the total evaluation budgets for MOEA/D-FS and the comparison algorithms are different. MOEA/D-FSs and comparison algorithms initialize their surrogates with 10, 100 samples, respectively. |
| Hardware Specification | No | The paper does not provide information about compute workers and memory since its experiments do not have specific requirements on memory or other computation resource. |
| Software Dependencies | No | The paper does not explicitly mention software versions for dependencies, only referring to the 'Plat EMO: A MATLAB platform for evolutionary multi-objective optimization [educational forum]' [39]. |
| Experiment Setup | Yes | Optimizaion problems. The computational study is conducted on DTLZ test problems [11]. All DTLZ problems have d=10 decision variables and 3 objectives, as the setups that have been widely used in [35]. ... Optimization setups. The parameter setups for this multi-objective optimization experiment are listed in Table 6. During the optimization process, an initial dataset S is sampled using Latin Hypercube Sampling (LHS) method [28], then extra evaluations are conducted until the evaluation budget has run out. ... Batch size B; Surrogate learning rates α, β; (from Algorithm 1 input) |