pith. sign in

arxiv: 2602.03901 · v4 · submitted 2026-02-03 · 💻 cs.LG · cs.NE

NeuroPareto: Calibrated Acquisition for Costly Many-Goal Search in Vast Parameter Spaces

Rong Fu , Chunlei Meng , Youjin Wang , Haoyu Zhao , Jiaxuan Lu , Kun Liu , JiaBao Dou , Simon James Fong This is my paper

Pith reviewed 2026-05-16 08:32 UTC · model grok-4.3

classification 💻 cs.LG cs.NE
keywords multi-objective optimizationPareto frontBayesian optimizationsurrogate modelsacquisition functionuncertainty estimationdeep Gaussian processes
0
0 comments X

The pith

NeuroPareto integrates calibrated neural classifiers and deep Gaussian process surrogates to guide costly evaluations toward high-quality Pareto fronts in high-dimensional spaces.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces NeuroPareto as a method for multi-objective optimization when each evaluation is expensive and the parameter space is large. It combines rank-centric filtering to screen candidates, a Bayesian classifier that quantifies uncertainty across domination levels, and deep Gaussian process surrogates that split uncertainty into reducible and irreducible parts. A lightweight acquisition network, updated from past hypervolume gains, then selects the next points to evaluate. Experiments on standard DTLZ and ZDT test suites plus a subsurface energy task show the approach reaches closer Pareto fronts and larger hypervolume than classifier-only or surrogate-only baselines while keeping extra computation low.

Core claim

NeuroPareto is a single architecture that fuses rank-centric filtering, epistemic uncertainty estimation via a calibrated Bayesian classifier over non-domination tiers, and history-conditioned acquisition via a lightweight network trained on hypervolume improvements; deep Gaussian process surrogates supply refined means and risk-aware signals, all maintained through hierarchical screening and amortized updates so that accuracy holds while evaluation cost drops.

What carries the argument

The integrated architecture of rank-centric filtering, uncertainty disentanglement in a calibrated Bayesian classifier, deep Gaussian process surrogates, and an online-trained lightweight acquisition network.

Load-bearing premise

The rank-centric filtering and uncertainty estimates stay reliable enough in high-dimensional spaces that the hierarchical screening and amortized updates continue to deliver accurate guidance without large extra overhead.

What would settle it

If NeuroPareto is run on the same DTLZ and ZDT problems and the measured hypervolume or Pareto proximity falls below that of the classifier-enhanced or surrogate-assisted baselines, the performance claim does not hold.

Figures

Figures reproduced from arXiv: 2602.03901 by Chunlei Meng, Haoyu Zhao, JiaBao Dou, Jiaxuan Lu, Kun Liu, Rong Fu, Simon James Fong, Youjin Wang.

Figure 1
Figure 1. Figure 1: Overview of the NeuroPareto framework for high-dimensional, budget-constrained multi-objective opti￾mization. The pipeline consists of three synergistic modules: The Bayesian Rank Classifier (gθ) screens a massive candidate pool Crank using temperature-calibrated softmax pk˜ and adaptive MC dropout to quantify classifier epis￾temic uncertainty uepclf. The Complexity-Reduced Deep GP pipeline processes the f… view at source ↗
Figure 2
Figure 2. Figure 2: Sensitivity of final hypervolume to inducing point count [PITH_FULL_IMAGE:figures/full_fig_p013_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Convergence curves on Type A problems [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Convergence curves on 100D bi-objective problems. The results indicate faster convergence and higher final [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Geothermal optimization results: (a) approximated Pareto fronts; (b) hypervolume progression over optimiza [PITH_FULL_IMAGE:figures/full_fig_p015_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Evolution of average uncertainty during optimization on a 100D problem. Epistemic uncertainty decreases as [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Computational overhead distribution per iteration. [PITH_FULL_IMAGE:figures/full_fig_p016_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Convergence curves (IGD) over function evaluations. NeuroPareto demonstrates faster convergence and lower [PITH_FULL_IMAGE:figures/full_fig_p017_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Ablation study on 100D DTLZ2: contribution of each component to final IGD. Removing key modules [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: High-dimensional scaling: final IGD as a function of available budget (log-scale). The method remains [PITH_FULL_IMAGE:figures/full_fig_p018_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Reliability diagram comparing temperature scaling and MC-Dropout. The closer the curve to the diagonal, [PITH_FULL_IMAGE:figures/full_fig_p018_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Pareto front approximation (2D): comparison between the true Pareto front and found solutions. [PITH_FULL_IMAGE:figures/full_fig_p019_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Relationship between proxy ranking and the true hypervolume contribution for a large candidate pool. The [PITH_FULL_IMAGE:figures/full_fig_p020_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Validation negative log predictive density (NLPD) across iterations. Selective full-GP evaluation mitigates [PITH_FULL_IMAGE:figures/full_fig_p021_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Predictive uncertainty decomposition on a 1-D slice of 100-D DTLZ2. The epistemic band expands away [PITH_FULL_IMAGE:figures/full_fig_p022_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Median hypervolume achieved as a function of the number of rank categories [PITH_FULL_IMAGE:figures/full_fig_p023_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Predictive correlation matrix among objectives on 100-D DTLZ2, estimated by the LMC Deep GP after 300 [PITH_FULL_IMAGE:figures/full_fig_p024_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Dropout-based uncertainty index (vertical) versus high-fidelity reference information measure (horizontal) [PITH_FULL_IMAGE:figures/full_fig_p026_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Hypervolume progression on DTLZ2-100D under the static weighted rule and the learned acquisition [PITH_FULL_IMAGE:figures/full_fig_p037_19.png] view at source ↗
read the original abstract

The pursuit of optimal trade-offs in high-dimensional search spaces under stringent computational constraints poses a fundamental challenge for contemporary multi-objective optimization. We develop NeuroPareto, a cohesive architecture that integrates rank-centric filtering, uncertainty disentanglement, and history-conditioned acquisition strategies to navigate complex objective landscapes. A calibrated Bayesian classifier estimates epistemic uncertainty across non-domination tiers, enabling rapid generation of high-quality candidates with minimal evaluation cost. Deep Gaussian Process surrogates further separate predictive uncertainty into reducible and irreducible components, providing refined predictive means and risk-aware signals for downstream selection. A lightweight acquisition network, trained online from historical hypervolume improvements, guides expensive evaluations toward regions balancing convergence and diversity. With hierarchical screening and amortized surrogate updates, the method maintains accuracy while keeping computational overhead low. Experiments on DTLZ and ZDT suites and a subsurface energy extraction task show that NeuroPareto consistently outperforms classifier-enhanced and surrogate-assisted baselines in Pareto proximity and hypervolume.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper proposes NeuroPareto, an integrated architecture for multi-objective optimization in high-dimensional, computationally expensive spaces. It combines rank-centric filtering, a calibrated Bayesian classifier to estimate epistemic uncertainty across non-domination tiers, Deep Gaussian Process surrogates that disentangle reducible and irreducible uncertainty, and a lightweight acquisition network trained online from historical hypervolume improvements. Hierarchical screening and amortized updates are used to control overhead. Experiments on DTLZ/ZDT suites and a subsurface energy extraction task report consistent gains over classifier-enhanced and surrogate-assisted baselines in Pareto proximity and hypervolume.

Significance. If the performance claims hold under the stated assumptions, the work offers a practical contribution to Bayesian optimization for many-objective problems by explicitly separating uncertainty types and amortizing surrogate updates. The online training of the acquisition network directly from hypervolume improvements is a clear methodological strength that avoids circularity.

major comments (3)
  1. [Experiments] Experiments section: The headline claim of reliable performance in 'vast parameter spaces' rests on the stability of rank-centric filtering and Deep-GP uncertainty estimates, yet the reported DTLZ/ZDT suites are low-to-moderate dimensional and the subsurface task provides no parameter counts, input dimensionality, or scaling ablation; this leaves the central scalability assertion untested.
  2. [Method] Method description (hierarchical screening and Deep-GP component): The assumption that epistemic uncertainty signals remain reliable beyond ~20–30 dimensions without extra regularization is load-bearing for the claimed advantage over baselines, but no stress-test or failure-mode analysis is supplied; degradation here would directly inject noise into the acquisition network and erase the reported gains.
  3. [Results] Results tables/figures: No error bars, statistical significance tests, or ablation isolating the contribution of uncertainty disentanglement versus rank-centric filtering are presented, making it impossible to determine whether the outperformance is robust or sensitive to post-hoc hyperparameter choices.
minor comments (2)
  1. [Abstract] The abstract would benefit from a single sentence stating the typical input dimensionality and number of objectives used in the experiments.
  2. [Method] Notation for the acquisition network input (historical hypervolume improvements) could be formalized with an equation to improve reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below and outline revisions to strengthen the experimental validation and methodological discussion.

read point-by-point responses

  1. Referee: [Experiments] Experiments section: The headline claim of reliable performance in 'vast parameter spaces' rests on the stability of rank-centric filtering and Deep-GP uncertainty estimates, yet the reported DTLZ/ZDT suites are low-to-moderate dimensional and the subsurface task provides no parameter counts, input dimensionality, or scaling ablation; this leaves the central scalability assertion untested.

    Authors: We appreciate the referee's point on explicit validation. The DTLZ/ZDT suites are standard benchmarks that support configurable dimensionality, and the subsurface task involves a practically high-dimensional parameter space. In revision we will explicitly state the input dimensions for every task (including the subsurface case), add a scaling ablation varying dimension and objective count on a subset of problems, and clarify how hierarchical screening is intended to support larger spaces. This will directly address the untested aspect of the scalability claim. revision: yes

  2. Referee: [Method] Method description (hierarchical screening and Deep-GP component): The assumption that epistemic uncertainty signals remain reliable beyond ~20–30 dimensions without extra regularization is load-bearing for the claimed advantage over baselines, but no stress-test or failure-mode analysis is supplied; degradation here would directly inject noise into the acquisition network and erase the reported gains.

    Authors: The Deep GP architecture is selected precisely because its layered structure improves uncertainty calibration in higher dimensions relative to shallow GPs, while rank-centric filtering reduces sensitivity to noisy uncertainty estimates. We did not provide dedicated stress tests above 30 dimensions because the primary contribution is the integrated pipeline rather than a standalone high-dimensional GP study. In revision we will add a limitations paragraph discussing expected degradation modes and the stabilizing role of the online acquisition network, but we cannot retroactively run new high-dimensional stress tests within the current experimental budget. revision: partial

  3. Referee: [Results] Results tables/figures: No error bars, statistical significance tests, or ablation isolating the contribution of uncertainty disentanglement versus rank-centric filtering are presented, making it impossible to determine whether the outperformance is robust or sensitive to post-hoc hyperparameter choices.

    Authors: We fully agree that error bars, significance testing, and component ablations are necessary. In the revised manuscript we will report mean and standard deviation over 10 independent runs with error bars on all tables and figures, apply Wilcoxon signed-rank tests with p-values to compare NeuroPareto against each baseline, and insert a new ablation table that isolates the Deep-GP uncertainty disentanglement from the rank-centric filtering step. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation relies on standard online training and experimental validation

full rationale

The paper describes a NeuroPareto architecture integrating rank-centric filtering, uncertainty disentanglement via Bayesian classifiers and Deep GPs, and a lightweight acquisition network trained online from historical hypervolume improvements. This training approach is a standard, non-circular practice in Bayesian optimization and does not reduce the central performance claims (Pareto proximity and hypervolume gains on DTLZ/ZDT and subsurface tasks) to fitted inputs by construction. No self-definitional equations, uniqueness theorems imported from self-citations, or ansatz smuggling appear in the provided derivation chain. The claims rest on empirical outperformance against baselines rather than tautological reductions. The reader's noted assumption about uncertainty reliability in high dimensions is a validity concern, not a circularity issue.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

Abstract-only review; the method implicitly relies on standard assumptions of Gaussian process regression and Bayesian classification plus several fitted components whose exact count cannot be determined without the full text.

free parameters (2)
  • acquisition network weights
    Trained online from historical hypervolume improvements; treated as learned parameters.
  • uncertainty disentanglement hyperparameters
    Deep GP surrogate parameters that separate reducible and irreducible uncertainty.
axioms (1)
  • domain assumption Non-domination ranking remains informative in high-dimensional objective spaces
    Central to the rank-centric filtering step described in the abstract.

pith-pipeline@v0.9.0 · 5482 in / 1322 out tokens · 22054 ms · 2026-05-16T08:32:56.805640+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

35 extracted references · 35 canonical work pages

  1. [1]

    An expensive many-objective optimization algorithm based on efficient expected hypervolume improvement.IEEE Transactions on Evolutionary Computation, 27(6):1822–1836, 2022

    Yong Pang, Yitang Wang, Shuai Zhang, Xiaonan Lai, Wei Sun, and Xueguan Song. An expensive many-objective optimization algorithm based on efficient expected hypervolume improvement.IEEE Transactions on Evolutionary Computation, 27(6):1822–1836, 2022

    work page 2022

  2. [2]

    Rank-based learning and local model based evolutionary algorithm for high-dimensional expensive multi-objective problems.arXiv preprint arXiv:2304.09444, 2023

    Guodong Chen, Jiu Jimmy Jiao, Xiaoming Xue, and Zhongzheng Wang. Rank-based learning and local model based evolutionary algorithm for high-dimensional expensive multi-objective problems.arXiv preprint arXiv:2304.09444, 2023

    work page arXiv 2023

  3. [3]

    Evolutionary optimiza- tion methods for high-dimensional expensive problems: A survey.IEEE/CAA Journal of Automatica Sinica, 11 (5):1092–1105, 2024

    MengChu Zhou, Meiji Cui, Dian Xu, Shuwei Zhu, Ziyan Zhao, and Abdullah Abusorrah. Evolutionary optimiza- tion methods for high-dimensional expensive problems: A survey.IEEE/CAA Journal of Automatica Sinica, 11 (5):1092–1105, 2024

    work page 2024

  4. [4]

    Expensive multiobjective evolutionary optimization assisted by dominance prediction.IEEE Transactions on Evolutionary Computation, 26(1):159–173, 2021

    Yuan Yuan and Wolfgang Banzhaf. Expensive multiobjective evolutionary optimization assisted by dominance prediction.IEEE Transactions on Evolutionary Computation, 26(1):159–173, 2021. 9 NeuroPareto

    work page 2021

  5. [5]

    Masksembles for uncertainty estimation

    Nikita Durasov, Timur Bagautdinov, Pierre Baque, and Pascal Fua. Masksembles for uncertainty estimation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 13539–13548, 2021

    work page 2021

  6. [6]

    A surrogate-assisted multi-objective evolutionary algorithm guided by hybrid reference points

    Shuxian Li, Yong Zhang, Qing Wang, Linchun He, Huijun Li, and Bin Ye. A surrogate-assisted multi-objective evolutionary algorithm guided by hybrid reference points. InInternational Conference on Swarm Intelligence, pages 442–450. Springer, 2024

    work page 2024

  7. [7]

    Rtdk-bo: High dimensional bayesian optimization with reinforced transformer deep kernels

    Alexander Shmakov, Avisek Naug, Vineet Gundecha, Sahand Ghorbanpour, Ricardo Luna Gutierrez, Ash- win Ramesh Babu, Antonio Guillen, and Soumyendu Sarkar. Rtdk-bo: High dimensional bayesian optimization with reinforced transformer deep kernels. In2023 IEEE 19th International Conference on Automation Science and Engineering (CASE), pages 1–8. IEEE, 2023

    work page 2023

  8. [8]

    Expensive multi-objective bayesian optimization based on diffusion models

    Bingdong Li, Zixiang Di, Yongfan Lu, Hong Qian, Feng Wang, Peng Yang, Ke Tang, and Aimin Zhou. Expensive multi-objective bayesian optimization based on diffusion models. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 27063–27071, 2025

    work page 2025

  9. [9]

    Yuanchao Liu, Jinliang Ding, Qian Li, Fei Li, and Jianchang Liu. A two-level model management-based surrogate- assisted evolutionary algorithm for medium-scale expensive multiobjective optimization.IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2025

    work page 2025

  10. [10]

    Parallel bayesian optimization of multiple noisy objectives with expected hypervolume improvement.Advances in neural information processing systems, 34: 2187–2200, 2021

    Samuel Daulton, Maximilian Balandat, and Eytan Bakshy. Parallel bayesian optimization of multiple noisy objectives with expected hypervolume improvement.Advances in neural information processing systems, 34: 2187–2200, 2021

    work page 2021

  11. [11]

    Expected hypervolume improvement is a particular hypervolume improvement

    Jingda Deng, Jianyong Sun, Qingfu Zhang, and Hui Li. Expected hypervolume improvement is a particular hypervolume improvement. InProceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 16217–16225, 2025

    work page 2025

  12. [12]

    Inverse distance weighting and radial basis function based surrogate model for high-dimensional expensive multi-objective optimization.Applied Soft Computing, 152:111194, 2024

    Fei Li, Zhengkun Shang, Yuanchao Liu, Hao Shen, and Yaochu Jin. Inverse distance weighting and radial basis function based surrogate model for high-dimensional expensive multi-objective optimization.Applied Soft Computing, 152:111194, 2024

    work page 2024

  13. [13]

    Bayesian optimization of function networks with partial evaluations.arXiv preprint arXiv:2311.02146, 2023

    Poompol Buathong, Jiayue Wan, Raul Astudillo, Samuel Daulton, Maximilian Balandat, and Peter I Frazier. Bayesian optimization of function networks with partial evaluations.arXiv preprint arXiv:2311.02146, 2023

    work page arXiv 2023

  14. [14]

    Monte carlo tree search based variable selection for high dimensional bayesian optimization.Advances in Neural Information Processing Systems, 35:28488–28501, 2022

    Lei Song, Ke Xue, Xiaobin Huang, and Chao Qian. Monte carlo tree search based variable selection for high dimensional bayesian optimization.Advances in Neural Information Processing Systems, 35:28488–28501, 2022

    work page 2022

  15. [15]

    Dropout as a bayesian approximation: representing model uncertainty in deep learning

    Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: representing model uncertainty in deep learning. InProceedings of the 33rd International Conference on Machine Learning (ICML), pages 1050–1059, 2016

    work page 2016

  16. [16]

    Zaid Abulawi, Rui Hu, Prasanna Balaprakash, and Yang Liu. Bayesian optimized deep ensemble for uncertainty quantification of deep neural networks: a system safety case study on sodium fast reactor thermal stratification modeling.Reliability Engineering & System Safety, page 111353, 2025

    work page 2025

  17. [17]

    Enhancing monte carlo dropout performance for uncertainty quantification.arXiv preprint arXiv:2505.15671, 2025

    Hamzeh Asgharnezhad, Afshar Shamsi, Roohallah Alizadehsani, Arash Mohammadi, and Hamid Alinejad-Rokny. Enhancing monte carlo dropout performance for uncertainty quantification.arXiv preprint arXiv:2505.15671, 2025

    work page arXiv 2025

  18. [18]

    Handling uncertainty with parametric surrogate- assisted optimization for dynamic multi-objective problems.Neural Computing and Applications, pages 1–30, 2025

    Arne De Temmerman, Matthias De Ryck, and Mathias Verbeke. Handling uncertainty with parametric surrogate- assisted optimization for dynamic multi-objective problems.Neural Computing and Applications, pages 1–30, 2025

    work page 2025

  19. [19]

    Amortized variational inference for deep gaussian processes.arXiv preprint arXiv:2409.12301, 2024

    Qiuxian Meng and Yongyou Zhang. Amortized variational inference for deep gaussian processes.arXiv preprint arXiv:2409.12301, 2024

    work page arXiv 2024

  20. [20]

    Sparse gaussian neural processes.arXiv preprint arXiv:2504.01650, 2025

    Tommy Rochussen and Vincent Fortuin. Sparse gaussian neural processes.arXiv preprint arXiv:2504.01650, 2025

    work page arXiv 2025

  21. [21]

    A classifier-ensemble-based surrogate-assisted evolutionary algorithm for distributed data-driven optimization.IEEE Transactions on Evolutionary Computation, 2024

    Xiao-Qi Guo, Feng-Feng Wei, Jun Zhang, and Wei-Neng Chen. A classifier-ensemble-based surrogate-assisted evolutionary algorithm for distributed data-driven optimization.IEEE Transactions on Evolutionary Computation, 2024. 10 NeuroPareto

    work page 2024

  22. [22]

    Neurolgp-sm: A surrogate-assisted neuroevolution approach using linear genetic programming

    Fergal Stapleton, Brendan Cody-Kenny, and Edgar Galván. Neurolgp-sm: A surrogate-assisted neuroevolution approach using linear genetic programming. InInternational Conference on Optimization and Learning, pages 67–81. Springer, 2024

    work page 2024

  23. [23]

    Permutation-based multi-objective evolutionary feature selection for high-dimensional data.arXiv preprint arXiv:2501.14310, 2025

    Raquel Espinosa, Gracia Sánchez, José Palma, and Fernando Jiménez. Permutation-based multi-objective evolutionary feature selection for high-dimensional data.arXiv preprint arXiv:2501.14310, 2025

    work page arXiv 2025

  24. [24]

    Xiaoxu Jiang, Qingda Chen, Jinliang Ding, and Xingyi Zhang. Dual-population evolution based dynamic constrained multiobjective optimization with discontinuous and irregular feasible regions.IEEE Transactions on Emerging Topics in Computational Intelligence, 2025

    work page 2025

  25. [25]

    A classification and pareto domination based multiobjective evolutionary algorithm

    Jinyuan Zhang, Aimin Zhou, and Guixu Zhang. A classification and pareto domination based multiobjective evolutionary algorithm. In2015 IEEE congress on evolutionary computation (CEC), pages 2883–2890. IEEE, 2015

    work page 2015

  26. [26]

    Tinkle Chugh, Yaochu Jin, Kaisa Miettinen, Jussi Hakanen, and Karthik Sindhya. A surrogate-assisted refer- ence vector guided evolutionary algorithm for computationally expensive many-objective optimization.IEEE Transactions on Evolutionary Computation, 22(1):129–142, 2016

    work page 2016

  27. [27]

    A classification-based surrogate-assisted evolutionary algorithm for expensive many-objective optimization.IEEE Transactions on Evolutionary Computation, 23(1):74–88, 2018

    Linqiang Pan, Cheng He, Ye Tian, Handing Wang, Xingyi Zhang, and Yaochu Jin. A classification-based surrogate-assisted evolutionary algorithm for expensive many-objective optimization.IEEE Transactions on Evolutionary Computation, 23(1):74–88, 2018

    work page 2018

  28. [28]

    An ensemble surrogate-based framework for expensive multiobjective evolutionary optimization.IEEE Transactions on Evolutionary Computation, 26(4):631–645, 2021

    Qiuzhen Lin, Xunfeng Wu, Lijia Ma, Jianqiang Li, Maoguo Gong, and Carlos A Coello Coello. An ensemble surrogate-based framework for expensive multiobjective evolutionary optimization.IEEE Transactions on Evolutionary Computation, 26(4):631–645, 2021

    work page 2021

  29. [29]

    Takumi Sonoda and Masaya Nakata. Multiple classifiers-assisted evolutionary algorithm based on decomposition for high-dimensional multiobjective problems.IEEE Transactions on Evolutionary Computation, 26(6):1581– 1595, 2022

    work page 2022

  30. [30]

    Efficient global optimization of expensive black-box functions.Journal of Global optimization, 13(4):455–492, 1998

    Donald R Jones, Matthias Schonlau, and William J Welch. Efficient global optimization of expensive black-box functions.Journal of Global optimization, 13(4):455–492, 1998

    work page 1998

  31. [31]

    Sequential model-based optimization for general algorithm configuration

    Frank Hutter, Holger H Hoos, and Kevin Leyton-Brown. Sequential model-based optimization for general algorithm configuration. InInternational conference on learning and intelligent optimization, pages 507–523. Springer, 2011

    work page 2011

  32. [32]

    Differentiable expected hypervolume improvement for parallel multi-objective bayesian optimization.Advances in neural information processing systems, 33:9851–9864, 2020

    Samuel Daulton, Maximilian Balandat, and Eytan Bakshy. Differentiable expected hypervolume improvement for parallel multi-objective bayesian optimization.Advances in neural information processing systems, 33:9851–9864, 2020

    work page 2020

  33. [33]

    Theoretical analyses of multi-objective evolutionary algorithms on multi-modal objectives

    Benjamin Doerr and Weijie Zheng. Theoretical analyses of multi-objective evolutionary algorithms on multi-modal objectives. InProceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 12293–12301, 2021

    work page 2021

  34. [34]

    Multi-objective bayesian optimization over high-dimensional search spaces

    Samuel Daulton, David Eriksson, Maximilian Balandat, and Eytan Bakshy. Multi-objective bayesian optimization over high-dimensional search spaces. InUncertainty in Artificial Intelligence, pages 507–517. PMLR, 2022

    work page 2022

  35. [35]

    Evolution-guided bayesian optimization for constrained multi-objective optimization in self-driving labs.npj Computational Materials, 10(1):104, 2024

    Andre KY Low, Flore Mekki-Berrada, Abhishek Gupta, Aleksandr Ostudin, Jiaxun Xie, Eleonore Vissol-Gaudin, Yee-Fun Lim, Qianxiao Li, Yew Soon Ong, Saif A Khan, et al. Evolution-guided bayesian optimization for constrained multi-objective optimization in self-driving labs.npj Computational Materials, 10(1):104, 2024. 6 Integrated complexity-reduced algorith...

    work page arXiv 2024