Pith. sign in

REVIEW 3 major objections 3 minor 68 references

Data Shift of Object Detection in Autonomous Driving

T0 review · 3 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that an LLM, prompted and iteratively corrected by a human, can design the core modules of a constrained multi-objective evolutionary algorithm, and that the resulting algorithm outperforms 11 published baselines on standa

desk verdict Abstract is about data shift object detection; body is a different paper (LLM4CMO); benchmark superiority partly reflects in-sample module selection on the same suites used for final comparison. read the letter →

arxiv 2508.11868 v1 pith:VOL5XHLG submitted 2025-08-16 cs.RO cs.AIcs.CV

Lida Xu This is my paper
classification cs.ROcs.AIcs.CV
keywords constrainedmulti-objectiveoptimizationlargelanguagemodelsLLM-aidedalgorithmdesignhybridoperatorsepsilonconstrainthandlingdynamicresourceallocationdual-populationevolutionaryUPF-CPFrelationship
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that a large language model, prompted and iteratively corrected by a human, can design the core modules of a constrained multi-objective evolutionary algorithm (CMOEA): the hybrid operator combinations, the epsilon decay schedule, and the dynamic resource allocation between two populations. The resulting algorithm, LLM4CMO, is built on a dual-population, two-stage framework and is reported to beat each of 11 existing algorithms in paired Wilcoxon tests across 61 benchmark functions and 10 real-world constrained problems, on both the hypervolume and inverted generational distance metrics. The practical stake is that CMOEA design is currently manual, fragile, and time-consuming; if LLMs can co-design such modules, algorithm development for constrained multi-objective problems could become substantially faster. The paper's own ablation study attributes most of the improvement to the LLM-designed hybrid operators, with the LLM-designed epsilon decay function also contributing.

What carries the argument

The carrying object is the pair of populations and the stage-1/stage-2 control loop around them. One population follows the constrained Pareto front, the other the unconstrained front; a dynamic distance-based criterion decides when learning ends, and a classification of UPF-CPF relationships (complete overlap, partial overlap, complete separation, unclear) decides which hybrid operator configuration applies in stage 2. Around that skeleton the LLM-human prompt loop designs three modules: the HOps operator and mating-pool table, the $\epsilon$ decay function, and the dynamic resource allocation rule. The design loop—prompt templates, downstream performance feedback, iterative revision—is the

What would settle it

Run the full LLM-human design loop once using only a subset of the six suites for feedback, freeze the modules, then evaluate LLM4CMO against the 11 baselines on the withheld problems. If the HV/IGD advantage disappears on withheld problems, the claimed benefit of LLM-aided design is an artifact of test-set selection; recomputing the Wilcoxon tests after discarding, rather than counting as losses, the NaN baseline runs on DOC and FCP would show how much of the margin depends on that accounting rule.

Watch

Extended reading notes

Core claim

The central discovery on the paper's own terms is that decomposing a complex optimizer into separable design tasks makes those tasks addressable by an LLM. LLM4CMO separates the stage that learns the relationship between the unconstrained Pareto front (UPF) and the constrained Pareto front (CPF) from the stage that exploits that relationship, and within the second stage uses an LLM-designed hybrid operator table, a piecewise $\epsilon$ decay function with phase-specific forms, and a dynamic resource allocation rule. Prompt templates plus observed downstream performance steer the LLM toward better module designs over several interaction rounds. In the final comparison, LLM4CMO wins the multip

Load-bearing premise

The load-bearing premise is that the final modules were selected using downstream performance on the very six benchmark suites that later serve as the comparison test: the LLM-human loop kept the configurations that scored well on those suites, so the reported superiority is partly a description of that selection procedure rather than an independent test of LLM co-design; the comparison also counts baseline runs that returned NaN as losses rather than excluding them as crashe

Editorial extensions

If this is right

  • A prompt-driven division of labour between human and LLM can replace part of the manual design of constrained multi-objective optimizers, at least on the benchmark distribution tested.
  • Because the hybrid-operator module contributes most of the gain, future automated-design efforts should spend their budget on operator and mating-pool selection before tuning decay schedules or resource allocation.
  • The modular decomposition used here can be transferred to other metaheuristics: each module can be designed and tested separately, which lowers the barrier to LLM-aided algorithm development.
  • The algorithm's runtime stays at the same order of magnitude as the fastest baselines, so the reported quality improvement does not rest on a huge computational cost.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A test the paper does not run: hold out a set of CMOPs never shown to the LLM during the prompt-feedback loop, then compare LLM4CMO with the 11 baselines on those held-out problems. If the advantage shrinks, the reported superiority is partly a selection effect rather than a general design capability.
  • The paper itself notes in its limitations section that the interaction paradigm is subjective and depends on designer observations; a robust extension would close the loop with an automatic performance signal that is independent of the final test suites.
  • The same divide-and-design recipe could be tried on other algorithm families, treating each algorithm's core mechanisms as prompt-designable units, with the caveat that feedback must not be drawn from the evaluation set.
  • Since the LLM-chosen HOps tend to overemphasize diversity on some DOC and MW problems, an adaptive operator-selection mechanism active during the run, rather than fixed by UPF-CPF type, is a natural next step.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 3 minor

Summary. The manuscript body presents LLM4CMO, a dual-population two-stage constrained multi-objective evolutionary algorithm (CMOEA) in which three core modules—hybrid operators (HOps), the epsilon decay function, and dynamic resource allocation (DRA)—are designed through LLM–human interaction. The algorithm is evaluated on six benchmark suites and ten real-world CMOPs against eleven published baselines, with HV and IGD metrics and Wilcoxon tests. Table V reports that LLM4CMO significantly outperforms all baselines on the aggregate 61-function comparison. Ablation studies and sensitivity analyses are provided, and the code is made available. The central claim is that LLMs can serve as effective co-designers for complex CMOEA modules. I reviewed the body text; the supplied front matter, however, identifies a different paper ("Data Shift of Object Detection in Autonomous Driving"), which is a separate editorial issue.

Significance. If the reported result were fully supported, LLM4CMO would provide notable evidence that LLM-assisted, human-in-the-loop design can improve specialized evolutionary algorithms. The paper has genuine strengths: a broad comparison against 11 published algorithms on standard suites and real-world problems, statistical testing, ablation of the three designed modules, and public code. These features make the empirical claim potentially valuable. However, two load-bearing problems weaken the current evidence: (1) baseline runs that returned NaN are counted as losses in the aggregate statistics, and (2) the LLM–human module-search procedure was validated on the same benchmark suites used in the final comparison. Both issues are correctable in revision, but they currently prevent the paper's headline claim from being accepted at face value.

major comments (3)
  1. [Table XVIII / Table V] In Table XVIII (HV) and Table XIX (IGD), FCP1–FCP4 show NaN for all eleven baselines while LLM4CMO reports finite values; DOC2, DOC5, DOC9 and several real-world rows in Table XX contain additional NaNs. The +/-/= counts at the bottom of these tables and the aggregate 61-function Wilcoxon results in Table V appear to count these NaN entries as baseline losses. A crashed or undefined run is not a measured performance loss; it should be excluded or reported as a failure rate. Recompute the summary statistics and Table V without treating NaN as inferior, and state the exclusion rule explicitly. This materially affects the claimed 44/36 HV/IGD win counts against URCMO and similar entries in Table V.
  2. [Sec. III-B4 / Tables III, IV / Table V] The module-search procedure is in-sample. Section III-B4 states that "a diverse set of CMOPs ... from various benchmark test suites" was used to optimize HOps, epsilon decay, and DRA, and Tables III and IV show candidate modules accepted or rejected by HV on CF, DASCMOP, LIRCMOP, and MW problems. The final Table V comparison uses the same six benchmark suites, including all four of those. The reported superiority is therefore partly a description of the selection criterion, not an independent test of whether LLM-aided design generalizes. A held-out evaluation—problems not used during module selection—is needed to support the cross-benchmark claim. The sensitivity analysis in Sec. XI perturbs hyperparameters around fixed HOps choices and does not address this selection loop.
  3. [Sec. IV-B3] The text claims that "LLM4CMO outperformed existing algorithms on problems FCP1–FCP5" (Sec. IV-B3). Since FCP1–FCP4 have NaN results for all baselines, there is no actual comparison for those four problems; only FCP5 provides baseline values. The Opposite-mechanism benefit for FCP1–FCP4 is therefore not demonstrated. This is a specific consequence of the NaN-counting issue and should be fixed by reporting only completable comparisons, or by providing a failure-rate analysis that distinguishes "algorithm crashed" from "algorithm produced a worse solution."
minor comments (3)
  1. [Front matter] The title and abstract supplied with this submission describe "Data Shift of Object Detection in Autonomous Driving," while the full text is the LLM4CMO paper. The metadata should be corrected to match the manuscript content.
  2. [Throughout] There are numerous typographical and formatting issues: duplicate "Limitations" sections (VI and VII), inconsistent numbering of supplementary sections, "Bico use a archieve population" in Sec. IV-A2, and several broken equations (e.g., Eq. (9), Eq. (13) in the raw text). Please copyedit carefully.
  3. [Sec. III-B4 / Sec. XI] The design-process description emphasizes "human guidance" and "observations of designers." The reproducibility of the LLM–human interaction would be improved by releasing the full interaction transcripts, not only the final prompt templates. The current statement that the complete interaction process is only in the code repository is insufficient for a claim of LLM-aided design.

Circularity Check

1 steps flagged · score 6.0 of 10

Core CMOEA modules are selected using HV feedback on the same benchmark suites later used for the headline comparison, making Table V partly an in-sample description of the selection procedure.

  1. fitted input called prediction [Sec. III-B4 (Design Process), Tables III–IV; Sec. IV-B, Table V; Sec. V Conclusion]
    "Our interactive LLM-based design methodology uses comprehensive downstream task performance as the termination criterion. ... We adopted a diverse set of CMOPs with different UPF-CPF relationship types from various benchmark test suites to optimize three core modules."

    Tables III and IV record HV changes on CF, DASCMOP, LIRCMOP, and MW while HOps and epsilon-decay variants are proposed, tested, accepted, or rejected; the final configurations are chosen because they score best on those HV values. Table V then reports LLM4CMO's Wilcoxon wins on 'all 61 functions' from the same six suites, and the conclusion claims 'LLM4CMO consistently outperformed baseline methods.' The headline superiority on those suites is therefore partly a record of the selection procedure—the module design loop used the same HV metric and the same benchmark distribution as the final evaluation—rather than an independent out-of-sample test of the LLM-aided design claim. The real-world problems in Table XX are out-of-sample and provide some independent content, but they are not the ma

full rationale

The paper is not self-citation-heavy and does not import a uniqueness theorem from its own authors. The central circularity is selection-on-target: the LLM-human interaction loop uses downstream HV performance as its termination and acceptance criterion, explicitly on CF, DASCMOP, LIRCMOP, and MW from the six benchmark suites, and the final 'consistently outperformed' claim is reported on those same suites. This makes a substantial part of Table V in-sample: the modules were chosen because they scored well on the same metric and problem distribution that the paper then uses as evidence. The design process is human-in-the-loop and not a formal argmax, so the reduction is not fully deterministic, but the quoted text shows that downstream-task performance on the evaluation suites is the selection signal. The real-world CMOPs provide a partially independent validation, and the comparisons against 11 externally published algorithms are meaningful, which prevents the score from being higher. The NaN baseline handling in Table XVIII is a statistical validity concern but is not itself a circularity finding.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

No new physical entities, mediators, or forces are introduced. The 'unclear' UPF-CPF type and the opposition-offspring mechanism are algorithmic devices within the existing CMOEA framework, not independently evidenced entities. The main unpaid cost is the set of hand-fitted and benchmark-fitted parameters (six groups listed above), plus the domain assumption that the UPF-CPF taxonomy and the chosen benchmarks support the claimed generalization.

free parameters (6)
  • epsilon_0 (initial epsilon for decay) = 0.2
    Eq. (7), Sec. III-B2; set by hand, sensitivity-tested in Tables X-XI.
  • Stage-1 transition thresholds in Eq. (4) = rs<0.001 with g>10; rs<0.02 with g>100; rs<0.05 with g>150; hard switch at g>250
    Hand-set dynamic relaxation conditions; ablation variant LLM4CMOWoRR shows the relaxed thresholds change results on many problems.
  • Epsilon decay function shape (Eq. 13) = a=15; t1=FEswitch+0.2(maxFE-FEswitch); t2=min(t1+0.3(maxFE-FEswitch),maxFE); periods 150 and 200; switch points [0.195,0
    Produced by LLM interaction on the benchmark suites (Sec. XII); switch points sensitivity-tested in Tables XII-XIII, but the functional form itself is a fitted choice.
  • Ns = max(25, (1-fr2)*N) for popAux size = Cofficients 25 and 1; tested over Ns in {30,40,50,60}
    Eq. (5); 'determined in the final step of algorithm design' (Sec. III-A), not derived from any principle.
  • HOps per-type operator and mating-pool assignments (Table II) = Type-specific DE/GA/DE-transfer combinations with tournament or random pools
    Chosen via LLM-human interaction with downstream HV feedback on CF, DASCMOP, MW, LIRCMOP (Tables III-IV); these are fitted to the evaluation suites.
  • DRA decision function coefficients (Eq. 9) = Supplied by LLM prompt output
    Sec. III-B3; obtained through template prompting and selected by downstream performance, with the paper noting minimal interaction was required.
assumptions (4)
  • standard math HV and IGD are valid measures of CMOEA quality and the Wilcoxon rank-sum and multiproblem signed-rank tests at 0.05 correctly establish significance
    Sec. IV-A1; metrics from refs [51],[52] and tests are standard practice in the field.
  • domain assumption The four-way UPF-CPF relationship taxonomy (complete overlap, partial overlap, complete separation, unclear) captures problem structure, and the classification method inherited from URCMO is accurate enough to select operators
    Sec. II-C and Sec. IX; the paper merges URCMO's six types into four and adds an 'unclear' fallback, assuming this coarse classification is reliable.
  • domain assumption The six benchmark suites plus ten real-world instances are a representative spread of CMOP difficulty
    Sec. IV-A1; the suites are standard, but the paper assumes performance here transfers to unseen constrained multi-objective problems.
  • ad hoc to paper Downstream HV on the evaluation suites is a legitimate objective for selecting LLM-designed modules
    Sec. III-B4: the interaction termination criterion is 'comprehensive downstream task performance'; this is the in-sample selection premise that weakens the headline comparison.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Data Shift of Object Detection in Autonomous Driving." pith.science (2026). https://pith.science/paper/VOL5XHLG

@misc{pith2026250811868,
  author       = {Pith},
  title        = {Pith review of: Data Shift of Object Detection in Autonomous Driving},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VOL5XHLG}},
  note         = {Machine review of arXiv:2508.11868}
}
read the original abstract

With the widespread adoption of machine learning technologies in autonomous driving systems, their role in addressing complex environmental perception challenges has become increasingly crucial. However, existing machine learning models exhibit significant vulnerability, as their performance critically depends on the fundamental assumption that training and testing data satisfy the independent and identically distributed condition, which is difficult to guarantee in real-world applications. Dynamic variations in data distribution caused by seasonal changes, weather fluctuations lead to data shift problems in autonomous driving systems. This study investigates the data shift problem in autonomous driving object detection tasks, systematically analyzing its complexity and diverse manifestations. We conduct a comprehensive review of data shift detection methods and employ shift detection analysis techniques to perform dataset categorization and balancing. Building upon this foundation, we construct an object detection model. To validate our approach, we optimize the model by integrating CycleGAN-based data augmentation techniques with the YOLOv5 framework. Experimental results demonstrate that our method achieves superior performance compared to baseline models on the BDD100K dataset.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

68 extracted references · 56 canonical work pages

  1. [1]

    Vehicle scheduling of an urban bus line via an improved multiobjective genetic algorithm,

    X. Zuo, C. Chen, W. Tan, and M. Zhou, “Vehicle scheduling of an urban bus line via an improved multiobjective genetic algorithm,”IEEE Transactions on Intelligent Transportation Systems, vol. 16, no. 2, pp. 1030–1041, 2014

    work page 2014

  2. [2]

    Butterfly optimizer for placement and sizing of distributed generation for feeder phase balancing,

    S. Mishra, A. Kumar, D. Singh, and R. Kumar Misra, “Butterfly optimizer for placement and sizing of distributed generation for feeder phase balancing,” inComputational Intelligence: Theories, Applications and Future Directions-Volume II: ICCI-2017. Springer, 2019, pp. 519– 530

    work page 2017

  3. [3]

    Evolutionary multi criteria design optimization of robot grippers,

    R. Saravanan, S. Ramabalan, N. G. R. Ebenezer, and C. Dharmaraja, “Evolutionary multi criteria design optimization of robot grippers,” Applied Soft Computing, vol. 9, no. 1, pp. 159–172, 2009

    work page 2009

  4. [4]

    An augmented lagrange multiplier based method for mixed integer discrete continuous optimization and its applications to mechanical design,

    B. Kannan and S. N. Kramer, “An augmented lagrange multiplier based method for mixed integer discrete continuous optimization and its applications to mechanical design,”Journal of Mechanical Design, vol. 116, pp. 405–411, 1994

    work page 1994

  5. [5]

    A coevolutionary framework for constrained multiobjective optimization problems,

    Y . Tian, T. Zhang, J. Xiao, X. Zhang, and Y . Jin, “A coevolutionary framework for constrained multiobjective optimization problems,”IEEE Transactions on Evolutionary Computation, vol. 25, no. 1, pp. 102–116, 2020

    work page 2020

  6. [6]

    A fast and elitist multiobjective genetic algorithm: NSGA-II,

    K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, “A fast and elitist multiobjective genetic algorithm: NSGA-II,”IEEE Transactions on Evolutionary Computation, vol. 6, no. 2, pp. 182–197, 2002

    2002

  7. [7]

    MOEA/D: A multiobjective evolutionary algorithm based on decomposition,

    Q. Zhang and H. Li, “MOEA/D: A multiobjective evolutionary algorithm based on decomposition,”IEEE Transactions on Evolutionary Compu- tation, vol. 11, no. 6, pp. 712–731, 2007

    2007

  8. [8]

    Constrained optimization by�constrained particle swarm optimizer with�-level control,

    T. Takahama and S. Sakai, “Constrained optimization by�constrained particle swarm optimizer with�-level control,” inSoft Computing as Transdisciplinary Science and Technology: Proceedings of the fourth IEEE International Workshop WSTST’05. Springer, 2005, pp. 1019– 1029

    work page 2005

Show all 68 references
  1. [9]

    Handling multiobjective optimization problems with complex constraints: A con- straints grouping-based approach,

    B. Xu, Y . Zheng, W. Li, X. Gao, D. Gong, J. He, and Z. Fan, “Handling multiobjective optimization problems with complex constraints: A con- straints grouping-based approach,”IEEE Transactions on Systems, Man, and Cybernetics: Systems, 2025

    2025

  2. [10]

    A constraint-handling tech- nique for decomposition-based constrained many-objective evolutionary algorithms,

    F. Ming, W. Gong, L. Wang, and L. Gao, “A constraint-handling tech- nique for decomposition-based constrained many-objective evolutionary algorithms,”IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 53, no. 12, pp. 7783–7793, 2023

    2023

  3. [11]

    A competitive and cooperative evolutionary framework for ensemble of constraint handling techniques,

    Y . Li, W. Gong, Z. Hu, and S. Li, “A competitive and cooperative evolutionary framework for ensemble of constraint handling techniques,” IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 54, no. 4, pp. 2440–2451, 2024

    2024

  4. [12]

    A dual-population algorithm based on alternative evolution and degeneration for solving constrained multi-objective optimization problems,

    J. Zou, R. Sun, S. Yang, and J. Zheng, “A dual-population algorithm based on alternative evolution and degeneration for solving constrained multi-objective optimization problems,”Information Sciences, vol. 579, pp. 89–102, 2021

    2021

  5. [13]

    A dual- population-based evolutionary algorithm for constrained multiobjective optimization,

    M. Ming, A. Trivedi, R. Wang, D. Srinivasan, and T. Zhang, “A dual- population-based evolutionary algorithm for constrained multiobjective optimization,”IEEE Transactions on Evolutionary Computation, vol. 25, no. 4, pp. 739–753, 2021

    2021

  6. [14]

    Utilizing the relationship between unconstrained and constrained pareto fronts for constrained multiobjective optimization,

    J. Liang, K. Qiao, K. Yu, B. Qu, C. Yue, W. Guo, and L. Wang, “Utilizing the relationship between unconstrained and constrained pareto fronts for constrained multiobjective optimization,”IEEE Transactions on Cybernetics, vol. 53, no. 6, pp. 3873–3886, 2022

    2022

  7. [15]

    A coevolutionary algorithm with detection and supervision strategies for constrained multiobjective optimization,

    S. Liu, J. Feng, S. Yang, J. Zheng, and Q. Xiao, “A coevolutionary algorithm with detection and supervision strategies for constrained multiobjective optimization,”IEEE Transactions on Evolutionary Com- putation, 2024

    2024

  8. [16]

    A survey on evolutionary constrained multiobjective optimization,

    J. Liang, X. Ban, K. Yu, B. Qu, K. Qiao, C. Yue, K. Chen, and K. C. Tan, “A survey on evolutionary constrained multiobjective optimization,” IEEE Transactions on Evolutionary Computation, vol. 27, no. 2, pp. 201–221, 2022

    2022

  9. [17]

    A survey of large language models,

    W. X. Zhao, K. Zhou, J. Li, T. Tang, X. Wang, Y . Hou, Y . Min, B. Zhang, J. Zhang, Z. Donget al., “A survey of large language models,”arXiv preprint arXiv:2303.18223, vol. 1, no. 2, 2023

    2023 arXiv

  10. [18]

    A systematic survey on large language models for algorithm design,

    F. Liu, Y . Yao, P. Guo, Z. Yang, Z. Zhao, X. Lin, X. Tong, M. Yuan, Z. Lu, Z. Wanget al., “A systematic survey on large language models for algorithm design,”arXiv preprint arXiv:2410.14716, 2024

    2024

  11. [19]

    LLM4SR: A sur- vey on large language models for scientific research,

    Z. Luo, Z. Yang, Z. Xu, W. Yang, and X. Du, “LLM4SR: A sur- vey on large language models for scientific research,”arXiv preprint arXiv:2501.04306, 2025

    2025 arXiv

  12. [20]

    Chain-of-thought prompting elicits reasoning in large language models,

    J. Wei, X. Wang, D. Schuurmans, M. Bosma, F. Xia, E. Chi, Q. V . Le, D. Zhouet al., “Chain-of-thought prompting elicits reasoning in large language models,”Advances in Neural Information Processing Systems, vol. 35, pp. 24 824–24 837, 2022

    2022

  13. [21]

    Mathematical discoveries from program search with large language models,

    B. Romera-Paredes, M. Barekatain, A. Novikov, M. Balog, M. P. Kumar, E. Dupont, F. J. Ruiz, J. S. Ellenberg, P. Wang, O. Fawziet al., “Mathematical discoveries from program search with large language models,”Nature, vol. 625, no. 7995, pp. 468–475, 2024. 13

    2024

  14. [22]

    Evolution of heuristics: towards efficient automatic algorithm design using large language model,

    F. Liu, X. Tong, M. Yuan, X. Lin, F. Luo, Z. Wang, Z. Lu, and Q. Zhang, “Evolution of heuristics: towards efficient automatic algorithm design using large language model,” inProceedings of the 41st International Conference on Machine Learning, ser. ICML’24. JMLR.org, 2024

    2024

  15. [23]

    Large language model for multiobjective evolutionary optimization,

    F. Liu, X. Lin, S. Yao, Z. Wang, X. Tong, M. Yuan, and Q. Zhang, “Large language model for multiobjective evolutionary optimization,” in International Conference on Evolutionary Multi-Criterion Optimization. Springer, 2025, pp. 178–191

    2025

  16. [24]

    Large language model- aided evolutionary search for constrained multiobjective optimization,

    Z. Wang, S. Liu, J. Chen, and K. C. Tan, “Large language model- aided evolutionary search for constrained multiobjective optimization,” inInternational Conference on Intelligent Computing. Springer, 2024, pp. 218–230

    2024

  17. [25]

    Jain and K

    H. Jain and K. Deb, “An evolutionary many-objective optimization algorithm using reference-point based nondominated sorting approach, part II: Handling constraints and extending to an adaptive approach,” IEEE Transactions on Evolutionary Computation, vol. 18, no. 4, pp. 602–622, 2013

    2013

  18. [26]

    Push and pull search for solving constrained multi-objective optimiza- tion problems,

    Z. Fan, W. Li, X. Cai, H. Li, C. Wei, Q. Zhang, K. Deb, and E. Goodman, “Push and pull search for solving constrained multi-objective optimiza- tion problems,”Swarm and Evolutionary Computation, vol. 44, pp. 665– 679, 2019

    2019

  19. [27]

    An improved epsilon constraint-handling method in MOEA/D for CMOPs with large infeasible regions,

    Z. Fan, W. Li, X. Cai, H. Huang, Y . Fang, Y . You, J. Mo, C. Wei, and E. Goodman, “An improved epsilon constraint-handling method in MOEA/D for CMOPs with large infeasible regions,”Soft Computing, vol. 23, pp. 12 491–12 510, 2019

    2019

  20. [28]

    Constraint-pareto domi- nance and diversity enhancement strategy based evolutionary algorithm for solving constrained multiobjective optimization problems,

    Z. Liu, F. Han, Q. Ling, H. Han, and J. Jiang, “Constraint-pareto domi- nance and diversity enhancement strategy based evolutionary algorithm for solving constrained multiobjective optimization problems,”IEEE Transactions on Evolutionary Computation, 2025

    2025

  21. [29]

    Angle-based constrained dominance principle in MOEA/D for constrained multi- objective optimization problems,

    Z. Fan, W. Li, X. Cai, K. Hu, H. Lin, and H. Li, “Angle-based constrained dominance principle in MOEA/D for constrained multi- objective optimization problems,” in2016 IEEE Congress on Evolu- tionary Computation (CEC). IEEE, 2016, pp. 460–467

    2016

  22. [30]

    MOEA/D with angle-based constrained dominance principle for constrained multi- objective optimization problems,

    Z. Fan, Y . Fang, W. Li, X. Cai, C. Wei, and E. Goodman, “MOEA/D with angle-based constrained dominance principle for constrained multi- objective optimization problems,”Applied Soft Computing, vol. 74, pp. 621–633, 2019

    2019

  23. [31]

    A cooperative multistep mutation strategy for multiobjective optimization problems with deceptive constraints,

    K. Qiao, K. Yu, C. Yue, B. Qu, M. Liu, and J. Liang, “A cooperative multistep mutation strategy for multiobjective optimization problems with deceptive constraints,”IEEE Transactions on Systems, Man, and Cybernetics: Systems, vol. 54, no. 11, pp. 6670–6682, 2024

    2024

  24. [32]

    A novel dual-stage dual-population evolutionary algorithm for constrained multiobjective optimization,

    M. Ming, R. Wang, H. Ishibuchi, and T. Zhang, “A novel dual-stage dual-population evolutionary algorithm for constrained multiobjective optimization,”IEEE Transactions on Evolutionary Computation, vol. 26, no. 5, pp. 1129–1143, 2021

    2021

  25. [33]

    Handling constrained multi- objective optimization problems via bidirectional coevolution,

    Z.-Z. Liu, B.-C. Wang, and K. Tang, “Handling constrained multi- objective optimization problems via bidirectional coevolution,”IEEE Transactions on Cybernetics, vol. 52, no. 10, pp. 10 163–10 176, 2021

    2021

  26. [34]

    A multi-stage evolutionary algorithm for multi-objective optimization with complex constraints,

    H. Ma, H. Wei, Y . Tian, R. Cheng, and X. Zhang, “A multi-stage evolutionary algorithm for multi-objective optimization with complex constraints,”Information Sciences, vol. 560, pp. 68–91, 2021

    2021

  27. [35]

    Balancing objective optimization and constraint satisfaction in constrained evolu- tionary multiobjective optimization,

    Y . Tian, Y . Zhang, Y . Su, X. Zhang, K. C. Tan, and Y . Jin, “Balancing objective optimization and constraint satisfaction in constrained evolu- tionary multiobjective optimization,”IEEE Transactions on Cybernetics, vol. 52, no. 9, pp. 9559–9572, 2021

    2021

  28. [36]

    Evolution- ary multitasking with global and local auxiliary tasks for constrained multi-objective optimization,

    K. Qiao, J. Liang, Z. Liu, K. Yu, C. Yue, and B. Qu, “Evolution- ary multitasking with global and local auxiliary tasks for constrained multi-objective optimization,”IEEE/CAA Journal of Automatica Sinica, vol. 10, no. 10, pp. 1951–1964, 2023

    1951

  29. [37]

    Constrained multiobjective optimization via multitasking and knowledge transfer,

    F. Ming, W. Gong, L. Wang, and L. Gao, “Constrained multiobjective optimization via multitasking and knowledge transfer,”IEEE Transac- tions on Evolutionary Computation, vol. 28, no. 1, pp. 77–89, 2022

    2022

  30. [38]

    A multistage algorithm for solving multiobjective optimization problems with multiconstraints,

    R. Sun, J. Zou, Y . Liu, S. Yang, and J. Zheng, “A multistage algorithm for solving multiobjective optimization problems with multiconstraints,” IEEE Transactions on Evolutionary Computation, vol. 27, no. 5, pp. 1207–1219, 2022

    2022

  31. [39]

    Even search in a promising region for constrained multi-objective optimization,

    F. Ming, W. Gong, and Y . Jin, “Even search in a promising region for constrained multi-objective optimization,”IEEE/CAA Journal of Automatica Sinica, vol. 11, no. 2, pp. 474–486, 2024

    2024

  32. [40]

    Evolutionary constrained multiobjective opti- mization: Test suite construction and performance comparisons,

    Z. Ma and Y . Wang, “Evolutionary constrained multiobjective opti- mization: Test suite construction and performance comparisons,”IEEE Transactions on Evolutionary Computation, vol. 23, no. 6, pp. 972–986, 2019

    2019

  33. [41]

    Genetic algorithms,

    J. H. Holland, “Genetic algorithms,”Scientific american, vol. 267, no. 1, pp. 66–73, 1992

    1992

  34. [42]

    Differential evolution–a simple and efficient heuristic for global optimization over continuous spaces,

    R. Storn and K. Price, “Differential evolution–a simple and efficient heuristic for global optimization over continuous spaces,”Journal of Global Optimization, vol. 11, pp. 341–359, 1997

    1997

  35. [43]

    JADE: adaptive differential evolution with optional external archive,

    J. Zhang and A. C. Sanderson, “JADE: adaptive differential evolution with optional external archive,”IEEE Transactions on Evolutionary Computation, vol. 13, no. 5, pp. 945–958, 2009

    2009

  36. [44]

    Robust and adaptive optimization under a large language model lens,

    D. Bertsimas and G. Margaritis, “Robust and adaptive optimization under a large language model lens,”arXiv preprint arXiv:2501.00568, 2024

    2024 arXiv

  37. [45]

    Large language models as optimizers,

    C. Yang, X. Wang, Y . Lu, H. Liu, Q. V . Le, D. Zhou, and X. Chen, “Large language models as optimizers,”arXiv preprint arXiv:2309.03409, 2023

    2023 arXiv

  38. [46]

    Approximation algorithms for bin packing problems: A survey,

    M. R. Garey and D. S. Johnson, “Approximation algorithms for bin packing problems: A survey,” inAnalysis and design of algorithms in combinatorial optimization. Springer, 1981, pp. 147–172

    1981

  39. [47]

    Fast heuristics for large geometric traveling salesman problems,

    G. Reinelt, “Fast heuristics for large geometric traveling salesman problems,”ORSA Journal on Computing, vol. 4, no. 2, pp. 206–217, 1992

    1992

  40. [48]

    Llamea: A large language model evolutionary algorithm for automatically generating metaheuristics,

    N. van Stein and T. B ¨ack, “Llamea: A large language model evolutionary algorithm for automatically generating metaheuristics,”IEEE Transac- tions on Evolutionary Computation, 2024

    2024

  41. [49]

    No free lunch theorems for optimization,

    D. H. Wolpert and W. G. Macready, “No free lunch theorems for optimization,”IEEE Transactions on Evolutionary Computation, vol. 1, no. 1, pp. 67–82, 1997

    1997

  42. [50]

    Deepseek-r1: Incentivizing reasoning capability in LLMs via reinforcement learning,

    D. Guo, D. Yang, H. Zhang, J. Song, R. Zhang, R. Xu, Q. Zhu, S. Ma, P. Wang, X. Biet al., “Deepseek-r1: Incentivizing reasoning capability in LLMs via reinforcement learning,”arXiv preprint arXiv:2501.12948, 2025

    2025 arXiv

  43. [51]

    Multiobjective evolutionary algorithms: a com- parative case study and the strength pareto approach,

    E. Zitzler and L. Thiele, “Multiobjective evolutionary algorithms: a com- parative case study and the strength pareto approach,”IEEE Transactions on Evolutionary Computation, vol. 3, no. 4, pp. 257–271, 2002

    2002

  44. [52]

    The balance between proximity and diversity in multiobjective evolutionary algorithms,

    P. A. Bosman and D. Thierens, “The balance between proximity and diversity in multiobjective evolutionary algorithms,”IEEE Transactions on Evolutionary Computation, vol. 7, no. 2, pp. 174–188, 2003

    2003

  45. [53]

    Multiobjective evolutionary algorithm based on multimethod with dynamic resources allocation,

    W. K. Mashwani and A. Salhi, “Multiobjective evolutionary algorithm based on multimethod with dynamic resources allocation,”Applied Soft Computing, vol. 39, pp. 292–309, 2016

    2016

  46. [54]

    Difficulty adjustable and scalable constrained multiobjective test problem toolkit,

    Z. Fan, W. Li, X. Cai, H. Li, C. Wei, Q. Zhang, K. Deb, and E. Good- man, “Difficulty adjustable and scalable constrained multiobjective test problem toolkit,”Evolutionary Computation, vol. 28, no. 3, pp. 339–378, 2020

    2020

  47. [55]

    Handling constrained multiobjective optimiza- tion problems with constraints in both the decision and objective spaces,

    Z.-Z. Liu and Y . Wang, “Handling constrained multiobjective optimiza- tion problems with constraints in both the decision and objective spaces,” IEEE Transactions on Evolutionary Computation, vol. 23, no. 5, pp. 870–884, 2019

    2019

  48. [56]

    Indicator-based evolutionary algorithm for solving constrained multiobjective optimization problems,

    J. Yuan, H.-L. Liu, Y .-S. Ong, and Z. He, “Indicator-based evolutionary algorithm for solving constrained multiobjective optimization problems,” IEEE Transactions on Evolutionary Computation, vol. 26, no. 2, pp. 379–391, 2021

    2021

  49. [57]

    A benchmark-suite of real-world constrained multi- objective optimization problems and some baseline results,

    A. Kumar, G. Wu, M. Z. Ali, Q. Luo, R. Mallipeddi, P. N. Sugan- than, and S. Das, “A benchmark-suite of real-world constrained multi- objective optimization problems and some baseline results,”Swarm and Evolutionary Computation, vol. 67, p. 100961, 2021

    2021

  50. [58]

    PlatEMO: A MATLAB plat- form for evolutionary multi-objective optimization [educational forum],

    Y . Tian, R. Cheng, X. Zhang, and Y . Jin, “PlatEMO: A MATLAB plat- form for evolutionary multi-objective optimization [educational forum],” IEEE Computational Intelligence Magazine, vol. 12, no. 4, pp. 73–87, 2017

    2017

  51. [59]

    On improving multiobjective genetic algorithms for design optimization,

    S. Narayanan and S. Azarm, “On improving multiobjective genetic algorithms for design optimization,”Structural Optimization, vol. 18, pp. 146–155, 1999

    1999

  52. [60]

    Comparison of multi-objective optimization methodologies for engineering applica- tions,

    G. Chiandussi, M. Codegone, S. Ferrero, and F. E. Varesio, “Comparison of multi-objective optimization methodologies for engineering applica- tions,”Computers & Mathematics with Applications, vol. 63, no. 5, pp. 912–942, 2012

    2012

  53. [61]

    Synchronous optimal pulsewidth modulation for low-switching-frequency control of medium-voltage mul- tilevel inverters,

    A. K. Rathore, J. Holtz, and T. Boller, “Synchronous optimal pulsewidth modulation for low-switching-frequency control of medium-voltage mul- tilevel inverters,”IEEE Transactions on Industrial Electronics, vol. 57, no. 7, pp. 2374–2381, 2010

    2010

  54. [62]

    Generalized optimal pulsewidth modulation of multilevel inverters for low-switching- frequency control of medium-voltage high-power industrial AC drives,

    Rathore, Akshay K and Holtz, Joachim and Boller, Till, “Generalized optimal pulsewidth modulation of multilevel inverters for low-switching- frequency control of medium-voltage high-power industrial AC drives,” IEEE Transactions on Industrial Electronics, vol. 60, no. 10, pp. ...

    2012

  55. [63]

    Optimal pulsewidth modulation of cascade nine-level (9L) inverter for medium voltage high power industrial AC drives,

    A. Edpuganti, A. Dwivedi, A. K. Rathore, and R. K. Srivastava, “Optimal pulsewidth modulation of cascade nine-level (9L) inverter for medium voltage high power industrial AC drives,” inIECON 2015-41st 14 Annual Conference of the IEEE Industrial Electronics Society. IEEE, 2015,...

    2015

  56. [64]

    Fundamental switching frequency optimal pulsewidth modulation of medium-voltage cascaded seven-level inverter,

    A. Edpuganti and A. K. Rathore, “Fundamental switching frequency optimal pulsewidth modulation of medium-voltage cascaded seven-level inverter,”IEEE Transactions on Industry Applications, vol. 51, no. 4, pp. 3485–3492, 2015

    2015

  57. [65]

    Optimal pulsewidth modulation for common-mode voltage elimination scheme of medium-voltage modular multilevel converter-fed open-end stator winding induction motor drives,

    Edpuganti, Amarendra and Rathore, Akshay Kumar, “Optimal pulsewidth modulation for common-mode voltage elimination scheme of medium-voltage modular multilevel converter-fed open-end stator winding induction motor drives,”IEEE Transactions on Industrial Electronics, vol. 64, no...

    2017

  58. [66]

    The results show that, for most problems, the inclusion of the opposite mechanism has no significant impact; however, a slight performance advantage was observed when it is enabled

    Results of the Base Framework:We conducted ablation studies focusing on the opposite mechanism and the staged optimization phases. The results show that, for most problems, the inclusion of the opposite mechanism has no significant impact; however, a slight performance advanta...

  59. [67]

    The results indicate that while modifications to the DRA module had a relatively minor effect on overall performance, they still contributed positively

    Results of Core Modules:We also performed before- and-after comparisons on the three core modules specifically designed through the LLM. The results indicate that while modifications to the DRA module had a relatively minor effect on overall performance, they still contributed...

  60. [68]

    5 and Fig

    Convergence:We plotted the convergence curves of IGD values for LLM4CMO and 11 baseline algorithms on the MW and LIRCMOP test suites, as presented in Fig. 5 and Fig. 4. Although LLM4CMO may not consistently achieve rapid early-stage convergence compared to some baseline algori...

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.