Pith. sign in

REVIEW 4 major objections 6 minor 48 references

How to Guide LLM Generation: Dual-Surrogate Guided Search for Automated Heuristic Design

T0 review · 4 major / 6 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read This paper claims that scoring pre-generation operator-parent actions with two learned surrogates—one predicting the child code's latent distribution, one predicting its utility—lets LLM-based automated heuristic design spend its limited qu

desk verdict A genuinely new dual-surrogate formulation for pre-generation action selection in LLM-based AHD, mathematically coherent and honestly framed, but the empirical case is statistically under-powered and the central controlled margin is within noise. read the letter →

arxiv 2607.13911 v1 pith:NSG6K3BE submitted 2026-07-15 cs.NE

Yuhan Wang , Chaoda Peng , Xingyu Wu , Sheng-Hao Wu , Zhi-Hui Zhan This is my paper
classification cs.NE
keywords automatedheuristicdesignlargelanguagemodelssurrogate-guidedsearchoperator-parentactionselectionuncertainty-awareacquisitionlatentrepresentationlearningsampleefficiencycombinatorialoptimization
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

The paper addresses the pre-generation decision in LLM-based automated heuristic design (AHD): given an archive of evaluated heuristics, which one(s) should be reused as parents, and which generation operator should transform them, before spending the next LLM query and black-box evaluation. The authors claim that this choice is best made by learning from accumulated code-score and transition records rather than by rules such as archive ranking, random sampling, fixed operator schedules, or tree policies. They propose Dual-Surrogate Guided Search (DGS), which trains a transition surrogate to predict the distribution of the child code's latent representation and an instance-conditioned utility surrogate to predict per-instance performance of sampled child latents, then combines predicted utility and both uncertainties in an acquisition rule to select the next action. Using a 200-generation budget across five heuristic-design tasks, DGS reports the best average rank on three tasks, a tie on one, and second on the fifth, with controlled ablations showing that jointly learning operator and parent beats rule-based operator selection and operator-only or parent-only surrogates. The core claim is that the bottleneck in LLM-AHD is not prompt engineering alone but allocating the next generation call to the most promising operator-parent pair.

What carries the argument

The load-bearing machinery is the two-surrogate decomposition of an unobservable outcome: the transition surrogate (Eq. 16) turns an operator-parent action into a distribution τ(x) over the child's learned latent representation z, and the utility surrogate (Eq. 5) turns any latent into per-instance performance estimates via ensemble heads fed with interaction features ξ(z, q_j) = [z, q_j, z⊙q_j, |z−q_j|, cos(z,q_j)]. The acquisition rule (Eq. 19) sums expected utility, ensemble disagreement, and transition uncertainty to rank actions. The paper also introduces a shared task-adaptive latent representation z(a) = norm(W e(a) + ρ R_φ(e(a))) over frozen code embeddings, plus instance embeddings

What would settle it

Take a fixed archive and a fixed pool of candidate actions; train the two surrogates on a random half of the transition records; test whether the acquisition score ranks the held-out actions by actually observed child score better than chance (e.g., Spearman rank correlation). If correlation is at or below zero, the surrogate is not learning a valid action-to-utility map. Separately, rerun the controlled TSP comparison with at least 20 independent seeds: if the distribution of best scores overlaps heavily, the reported 0.043 mean gap is noise.

Watch

Extended reading notes

Core claim

The central claim is that a pre-generation operator-parent action can be given a score before its child code exists, and that this score is a valid guide for allocating the next LLM query. The paper's mechanism: a transition surrogate maps an action x = (o, S) — a generation operator plus a set of archived parents — to a diagonal Gaussian over child latents, using a summary of parent latents and a learned operator embedding. An instance-conditioned utility surrogate, an ensemble of heads, predicts per-instance performance from a heuristic-instance interaction feature, and is trained with both pointwise MSE and within-instance rank supervision. The acquisition rule α(x) = E[μ_u(z)] + β_u E[σ_

Load-bearing premise

The load-bearing assumption is that a diagonal Gaussian over a learned latent, whose parameters come from parent-latent summaries and an operator embedding, can predict the distribution of the child code an LLM will actually generate, and that this model, trained online on data the same search produced, generalizes well enough that the acquisition score ranks actions in the same order as their true expected utility.

Editorial extensions

If this is right

  • DGS is operator-agnostic: the paper shows it remains effective when the operator set is expanded with a tree-path synthesis operator, so learned action selection can be layered onto any archive-based LLM-AHD loop.
  • Because the utility surrogate conditions on individual instances, the search can favor heuristics that generalize across the training set rather than merely maximizing aggregate score.
  • The utility-trained latent space makes offline heuristic selection more effective: a GP-UCB selector built on these latents reduces simple regret faster than one built on raw code embeddings.
  • The controlled comparison shows that jointly scoring operator and parent is necessary: operator-only or parent-only surrogates are weaker than the joint surrogate, and all surrogate variants except operator-only beat the rule-based controls.

Reading between the lines

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

  • The transition surrogate is trained on self-generated data, raising the risk of a feedback loop where the acquisition rule favors actions the surrogate already predicts well, suppressing genuinely novel exploration; a testable extension is to inject a fraction of random or bandit-chosen actions into the training data and measure whether surrogate validity and final performance improve.
  • The controlled margin between the joint surrogate and the strongest non-surrogate control is 0.043 on a three-run mean, so statistical significance is not established; a reader should ask for repeated seeds and confidence intervals before treating the advantage as robust.
  • The instance-conditioned utility surrogate points toward a broader use: a latent embedding trained to predict per-instance utility could rank pregenerated heuristics on unseen instances, turning the search module into an algorithm-selection tool beyond AHD.
  • The diagonal-Gaussian form of the transition surrogate is a strong simplification; if some operators produce highly multimodal child distributions, the acquisition rule's expectation over τ(x) could systematically underestimate uncertainty, suggesting mixture or normalizing-flow extensions.
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

4 major / 6 minor

Summary. This paper proposes Dual-Surrogate Guided Search (DGS), a module for LLM-based automated heuristic design that selects, before each LLM query, an operator-parent action (o,S) from the current archive. DGS learns a shared latent representation of heuristic code (Eq. 3), trains an instance-conditioned utility surrogate ensemble (Eqs. 4-12) that predicts per-instance performance of a child latent, and trains a transition surrogate (Eqs. 14-17) that predicts a diagonal Gaussian over the child latent induced by an action. These are combined in the acquisition rule of Eq. (19) with exploration bonuses from utility-ensemble disagreement and transition uncertainty. The method is evaluated on TSP, KP, OBP, ASP, and CVRP-ACO against FunSearch, EoH, ReEvo, and MCTS-AHD under a 200-generation budget, with a controlled TSP comparison and ablations. The central claims are that DGS improves sample efficiency and that the joint operator-parent surrogate outperforms rule-based action selection; the main results are reported as average ranks and three-run means.

Significance. If the reported gains hold, the paper makes a useful and original contribution: it treats pre-generation action selection as a learned decision problem rather than a hand-coded rule, and the two-surrogate decomposition is a natural way to score actions before the child code exists. The objective functions are well-formed, Algorithm 1 is explicit, and the paper honestly frames the main result as 'competitive' rather than claiming uniform superiority. The utility rank loss (Eq. 8), the Gaussian NLL (Eq. 17), and the acquisition rule (Eq. 19) are internally coherent. However, the empirical evidence for the central claim is currently not strong enough: the key controlled margin is within run-to-run noise, and no error bars or significance tests are reported. The significance of the work therefore depends on the authors providing statistically grounded comparisons.

major comments (4)
  1. [Table 2 and §Effect of Joint Operator-Parent Selection] The load-bearing claim that the joint operator-parent surrogate outperforms the strongest rule-based control rests on H4 (-6.406) vs H3 (-6.449), a 0.043 margin on three-run means. The paper reports no error bars, confidence intervals, or significance tests for any table. With three runs, this margin is not distinguishable from run-to-run noise. Note also that Table 3 reports DGS Original at -6.351 on the same task while H4 is -6.406; the difference between two DGS variants is itself of the same order as the H4-H3 gap. This is the central controlled comparison for RQ2, so it must be supported by additional seeds and a statistical test (or, at minimum, error bars and per-run values).
  2. [Table 1 and §Main Results] The main result summary uses average ranks, but the actual per-task scores are not in the body; they are stated to be in the supplementary material, which was not part of the manuscript as provided. The reader therefore cannot verify the claimed 'best on TSP, KP, and OBP' or the average ranks reported in Table 1. A central empirical claim should be reproducible from the submitted manuscript. Include a per-task table with means and standard deviations in the main text, or otherwise supply enough per-task evidence to check the average ranks.
  3. [Eqs. (14)-(17), §Transition Surrogate] The acquisition rule in Eq. (19) treats τ(x) as the distribution of the child latent for action x and uses sτ(x) = ||σ_τ(x)|| as an exploration bonus. The manuscript provides no diagnostic that this diagonal-Gaussian transition model is a reasonable approximation of actual LLM-generated child latents, nor that its predicted uncertainty is calibrated. The ablation 'w/o λτ' in Table 3 shows sensitivity to this term, but sensitivity does not establish validity. Please report held-out transition NLL or calibration/coverage checks on held-out transitions, or otherwise justify the Gaussian/parent-summary parameterization.
  4. [Table 3, §Ablation Study] Table 3 is used to conclude that each component is necessary, but all values are three-run means without variance or significance tests. On OBP, many variants are within about 0.03 of each other (e.g., -2.166 vs -2.174); on TSP, the range is about 0.55. Without error bars, the ordering of configurations may not be stable. Report per-run values, standard deviations, or paired tests over the same seeds or initial archives.
minor comments (6)
  1. [Eq. (7)] The text says 'if a batch contains no valid within-instance pair, the rank term is set to zero.' Make this explicit in the equation or immediately after, since the current notation R+ could be empty.
  2. [§Method Overview] The five EoH operators are only named; a one-sentence description in the main text would make the method self-contained without forcing the reader to consult the supplement.
  3. [Table 2 caption] The caption says 'Mean best' without specifying training vs test scores; the text clarifies later, but the caption should state 'training score' explicitly.
  4. [Figure 2] With only three runs, min-max bands are not a robust uncertainty display. Consider plotting means with standard errors or per-run curves.
  5. [§Experimental Setup] The paper does not state how β_u, λ_τ, K, and M_MC were chosen. Add a short paragraph on hyperparameter selection and sensitivity.
  6. [Reproducibility] No code-release or seed-management statement is included. For an empirical paper of this kind, please state whether code will be released and how seeds are handled.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: surrogate targets are observed code-score and child-latent outcomes, and main claims are checked against external baselines.

full rationale

The derivation chain is a supervised surrogate loop, not a tautological reduction. The utility surrogate's targets in Eq. 6 are recorded black-box evaluation scores y_ij; the transition surrogate's targets in Eq. 17 are the discovered child latents z(a_c). Neither target is defined in terms of the acquisition value alpha(x) from Eq. 19. The archive used for training is produced by the search itself (Algorithm 1, lines 8-19), which is an online/self-generated training distribution, but this is not an equation-level circularity: the surrogate's predictions are still checked against the next LLM sample and evaluator call. The central empirical claims are evaluated against external baselines (FunSearch, EoH, ReEvo, MCTS-AHD) on held-out test settings (Table 1), and the controlled study in Table 2 varies only the action-selection rule, so the comparison does not reduce to the surrogate's own fitted values. Self-citations (e.g., Wu et al. 2024b for interaction features, Wu et al. 2023 for variation operators, and the EoH-style operator set from Liu et al. 2024a) are used as design inspiration or external components, not as an unverified uniqueness/justification argument. The small controlled margin between H4 and H3 (0.043 on three-run means, without error bars) is a statistical-validity concern, but it is not circularity.

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

DGS is a method paper: the central claim is empirical, so the ledger records the modeling and tuning choices the empirical result depends on. The main burden sits on the transition surrogate's diagonal-Gaussian/parent-summary assumption (Eqs. 14-16) and on the online training loop in which surrogates are fit to data produced by actions the surrogates themselves selected (Algorithm 1). The free parameters β_u, λ_τ, K, M_MC and the imputation floor 4s_j enter the acquisition and training objectives directly without reported values, leaving open the possibility of benchmark-specific tuning. No new physical or mathematical entities are postulated, so the invented-entities list is empty.

free parameters (6)
  • β_u (exploration strength) = not stated in main text
    Weights the two uncertainty bonuses in the acquisition rule Eq. 19; if tuned per benchmark rather than fixed a priori, the comparisons partly fit the method to the test tasks.
  • λ_τ (transition-uncertainty weight) = not stated in main text
    Relative contribution of transition uncertainty in Eq. 19. The ablation pattern is sensitive to this constant: removing only λ_τ (-6.899) hurts more than removing both uncertainty terms via β_u (-6.464), an interaction the paper does not explain.
  • K (top-K parent pool size) = not stated in main text
    Defines the candidate action set X_t in Eq. 18; inherited from EoH-style practice but the concrete value is not given, and it caps which parents DGS may ever select.
  • M_MC (Monte Carlo child-latent samples) = not stated in main text
    Number of sampled child latents used to estimate the expectations in Eq. 19 (Algorithm 1, line 14); controls estimator noise of the acquisition score.
  • invalid-score imputation floor (4 × s_j) = 4
    Invalid per-instance scores are imputed as ȳ_j − 4s_j when building the utility table T_u; the factor 4 is chosen by hand and directly shapes utility targets for programs that fail on some instances.
  • representation hyperparameters (ρ, λ_r, λ_d, d_z, H, n_0, n_w) = not stated in main text
    Residual scale, rank/drift weights, latent dimension, ensemble size, and initial/warmup budgets enter Eqs. 3, 10-12 and Algorithm 1 without stated values; these silently affect representation quality and acquisition behavior.
assumptions (6)
  • domain assumption The ModernBERT-derived latent z(a) captures performance-relevant structure of heuristic code.
    The entire utility surrogate rests on Eq. 3 embeddings being predictive of scores; the paper provides no evidence in the main text that a frozen text encoder is adequate for these code-style objects.
  • ad hoc to paper Child latents induced by an action x follow a diagonal Gaussian parameterized by a parent-latent summary and operator embedding.
    Eq. 16 postulates a diagonal-Gaussian child-latent distribution; no goodness-of-fit check or calibration evidence is offered for this modeling convenience.
  • domain assumption LLM generation operators map parent latents to child latents in a learnable, reasonably stationary way.
    The transition surrogate (Eqs. 13-17) assumes that the stochastic LLM generator behaves like a stable conditional distribution over latents; proprietary API models are non-stationary and prompt-sensitive.
  • domain assumption The top-K archive by aggregate score is a sufficient parent pool.
    Eq. 18 restricts all candidate actions to B_t; DGS cannot select parents outside the top-K, so the parent-pool construction, inherited from EoH/MCTS-AHD, caps the search and is a fixed input rather than a learned one.
  • domain assumption Surrogates trained on at most 200 transition records generalize well enough to rank untried actions.
    Algorithm 1 fits the transition and utility surrogates on at most t records early in search; with latent dimension d_z and ensemble heads this is a small-sample regime, and the paper provides no learning-curve diagnostics.
  • standard math A UCB-style sum of predicted mean, ensemble variance, and transition variance is a valid acquisition for action selection.
    Standard Bayesian-optimization practice (cf. Srinivas et al. 2009), applied here with estimated rather than calibrated uncertainties; reasonable but unproven in this pre-generation setting.

how reviews work

0 comments
Cite this review

Pith. "Pith review of How to Guide LLM Generation: Dual-Surrogate Guided Search for Automated Heuristic Design." pith.science (2026). https://pith.science/paper/NSG6K3BE

@misc{pith2026260713911,
  author       = {Pith},
  title        = {Pith review of: How to Guide LLM Generation: Dual-Surrogate Guided Search for Automated Heuristic Design},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NSG6K3BE}},
  note         = {Machine review of arXiv:2607.13911}
}
read the original abstract

Large language models (LLMs) have made automated heuristic design (AHD) increasingly practical by generating executable heuristic code from task descriptions and evaluator feedback. Yet under a limited query and evaluation budget, search efficiency depends critically on a pre-generation decision. Before each LLM query and black-box evaluation, the system must choose which archived heuristics to reuse as parents and which generation operator should transform them. Existing methods typically choose such actions with predefined rules, leaving the expected outcome of each concrete operator-parent action only indirectly modeled. Therefore, we propose \emph{\fullmethod{}} (\method{}), a surrogate-guided action-selection module for operator-parent selection in LLM-based AHD. \method{} guides the LLM code-generation process by scoring pre-generation actions with two complementary surrogates. Specifically, a transition surrogate is proposed to predict the latent distribution of the child representation induced by an operator-parent action, while an instance-conditioned utility surrogate is proposed to estimate the expected performance of sampled child latents. Moreover, we propose an uncertainty-aware acquisition rule that combines predicted utility, utility uncertainty, and transition uncertainty to select the next LLM generation action. Across a diverse heuristic-design suite, \method{} is competitive with strong LLM-AHD baselines, and ablation and action-selection analyses suggest that its behavior goes beyond simple archive ranking or fixed operator preferences.

Figures

Figures reproduced from arXiv: 2607.13911 by the authors.

Figure 1
Figure 1. Overview of existing methods and proposed Dual-Surrogate Guided Search (DGS). A generic archive-based loop [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Evolution curves on four tasks. Curves show mean best-so-far scores; shaded bands show the min–max range. [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

48 extracted references · 4 linked inside Pith

  1. [1]

    IEEE Transactions on Cybernetics , volume=

    Evolving scheduling heuristics via genetic programming with feature selection in dynamic flexible job-shop scheduling , author=. IEEE Transactions on Cybernetics , volume=. 2020 , publisher=

    2020

  2. [2]

    Handbook of Metaheuristics , pages=

    A classification of hyper-heuristic approaches: revisited , author=. Handbook of Metaheuristics , pages=. 2019 , publisher=

    2019

  3. [3]

    ACM Computing Surveys , volume=

    A systematic survey on large language models for algorithm design , author=. ACM Computing Surveys , volume=. 2026 , publisher=

    2026

  4. [4]

    IEEE Transactions on Evolutionary Computation , volume=

    van Stein, Niki and B. IEEE Transactions on Evolutionary Computation , volume=. 2024 , publisher=

    2024

  5. [5]

    Proceedings of the International Conference on Machine Learning , year=

    Evolution of heuristics: Towards efficient automatic algorithm design using large language model , author=. Proceedings of the International Conference on Machine Learning , year=

  6. [6]

    Liu, Fei and Zhang, Rui and Xie, Zhuoliang and Sun, Rui and Li, Kai and Lin, Xi and Wang, Zhenkun and Lu, Zhichao and Zhang, Qingfu , eprint=

  7. [7]

    Ye, Haoran and Wang, Jiarui and Cao, Zhiguang and Berto, Federico and Hua, Chuanbo and Kim, Haeyeon and Park, Jinkyoo and Song, Guojie , booktitle=

  8. [8]

    Monte Carlo tree search for comprehensive exploration in

    Zheng, Zhi and Xie, Zhuoliang and Wang, Zhenkun and Hooi, Bryan , booktitle=. Monte Carlo tree search for comprehensive exploration in

Show all 48 references
  1. [9]

    Nature , volume=

    Mathematical discoveries from program search with large language models , author=. Nature , volume=. 2024 , publisher=

    2024

  2. [10]

    2605.17137 , archivePrefix=

    Latent heuristic search: Continuous optimization for automated algorithm design , author=. 2605.17137 , archivePrefix=

    arXiv

  3. [11]

    Proceedings of the AAAI Conference on Artificial Intelligence , year=

    Multi-objective evolution of heuristic using large language model , author=. Proceedings of the AAAI Conference on Artificial Intelligence , year=

  4. [12]

    2026 , url=

    Huang, Ziyao and Wu, Weiwei and Wu, Kui and Wang, Jianping and Lee, Wei-Bin , booktitle=. 2026 , url=

    2026

  5. [13]

    0912.3995 , archivePrefix=

    Gaussian process optimization in the bandit setting: No regret and experimental design , author=. 0912.3995 , archivePrefix=

    arXiv

  6. [14]

    2412.03250 , archivePrefix=

    Controlling the mutation in large language models for the efficient evolution of algorithms , author=. 2412.03250 , archivePrefix=

    arXiv

  7. [15]

    and Kotthoff, Lars and B

    van Stein, Niki and Kononova, Anna V. and Kotthoff, Lars and B. 2601.21511 , archivePrefix=

  8. [16]

    Hifo-prompt: Prompting with hindsight and foresight for

    Chen, Chentong and Zhong, Mengyuan and Sun, Jianyong and Fan, Ye and Shi, Jialong , eprint=. Hifo-prompt: Prompting with hindsight and foresight for

  9. [17]

    IEEE Transactions on Cybernetics , volume=

    Transferable adaptive differential evolution for many-task optimization , author=. IEEE Transactions on Cybernetics , volume=. 2023 , publisher=

    2023

  10. [18]

    Dat, Pham Vu Tuan and Doan, Long and Binh, Huynh Thi Thanh , booktitle=

  11. [19]

    Optimizing generative

    Yuksekgonul, Mert and Bianchi, Federico and Boen, Joseph and Liu, Sheng and Lu, Pan and Huang, Zhi and Guestrin, Carlos and Zou, James , journal=. Optimizing generative. 2025 , publisher=

    2025

  12. [20]

    2510.07073 , archivePrefix=

    Hottung, Andr. 2510.07073 , archivePrefix=

  13. [21]

    Generalizable heuristic generation through

    Shi, Yiding and Zhou, Jianan and Song, Wen and Bi, Jieyi and Wu, Yaoxin and Cao, Zhiguang and Zhang, Jie , eprint=. Generalizable heuristic generation through

  14. [22]

    Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V

    Co-evolution of large language models and configuration strategies to enhance surrogate-assisted evolutionary algorithm , author=. Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2 , pages=

  15. [23]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Co-bench: Benchmarking language model agents in algorithm search for combinatorial optimization , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  16. [24]

    IEEE Transactions on Evolutionary Computation , year=

    Automated scheduling heuristic generation and evaluation via large language model , author=. IEEE Transactions on Evolutionary Computation , year=

  17. [25]

    IEEE Transactions on Evolutionary Computation , year=

    Autonomous multi-objective optimization using large language model , author=. IEEE Transactions on Evolutionary Computation , year=

  18. [26]

    Journal of the ACM (JACM) , volume=

    On the online bin packing problem , author=. Journal of the ACM (JACM) , volume=. 2002 , publisher=

    2002

  19. [27]

    An extension of the Lin-Kernighan-Helsgaun

    Helsgaun, Keld , journal=. An extension of the Lin-Kernighan-Helsgaun

  20. [28]

    Ye, Haoran and Wang, Jiarui and Cao, Zhiguang and Liang, Helan and Li, Yong , booktitle=

  21. [29]

    European Journal of Operational Research , volume=

    New benchmark instances for the capacitated vehicle routing problem , author=. European Journal of Operational Research , volume=. 2017 , publisher=

    2017

  22. [30]

    Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence , pages=

    Large language model-enhanced algorithm selection: Towards comprehensive algorithm representation , author=. Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence , pages=

  23. [31]

    Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Smarter, better, faster, longer: A modern bidirectional encoder for fast, memory efficient, and long context finetuning and inference , author=. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  24. [32]

    International Conference on Learning Representations , volume=

    Offline model-based optimization by learning to rank , author=. International Conference on Learning Representations , volume=

  25. [33]

    Advances in Neural Information Processing Systems , volume=

    Neural combinatorial optimization with heavy decoder: Toward large scale generalization , author=. Advances in Neural Information Processing Systems , volume=

  26. [34]

    IEEE Transactions on Evolutionary Computation , volume=

    On the performance of multiple-objective genetic local search on the 0/1 knapsack problem-a comparative experiment , author=. IEEE Transactions on Evolutionary Computation , volume=. 2002 , publisher=

    2002

  27. [35]

    International Conference on Parallel Problem Solving from Nature , pages=

    Understanding the importance of evolutionary search in automated heuristic design with large language models , author=. International Conference on Parallel Problem Solving from Nature , pages=. 2024 , organization=

    2024

  28. [36]

    IEEE Transactions on Evolutionary Computation , volume=

    Evolutionary computation in the era of large language model: Survey and roadmap , author=. IEEE Transactions on Evolutionary Computation , volume=. 2024 , publisher=

    2024

  29. [37]

    Artificial Intelligence , volume=

    Ensembling neural networks: many could be better than all , author=. Artificial Intelligence , volume=. 2002 , publisher=

    2002

  30. [38]

    Advances in Neural Information Processing Systems , volume=

    Local latent space Bayesian optimization over structured inputs , author=. Advances in Neural Information Processing Systems , volume=

  31. [39]

    Advances in Neural Information Processing Systems , volume=

    Learning to iteratively solve routing problems with dual-aspect collaborative transformer , author=. Advances in Neural Information Processing Systems , volume=

  32. [40]

    ACM Computing Surveys , volume=

    Recent advances in Bayesian optimization , author=. ACM Computing Surveys , volume=. 2023 , publisher=

    2023

  33. [41]

    Zhao, Zhikai and Hua, Chuanbo and Berto, Federico and Lee, Kanghoon and Ma, Zihan and Li, Jiachen and Park, Jinkyoo , booktitle=

  34. [42]

    IEEE Computational Intelligence Magazine , volume=

    How good is neural combinatorial optimization? A systematic evaluation on the traveling salesman problem , author=. IEEE Computational Intelligence Magazine , volume=. 2023 , publisher=

    2023

  35. [43]

    2506.13131 , archivePrefix=

    Novikov, Alexander and V. 2506.13131 , archivePrefix=

    arXiv

  36. [44]

    Handbook of Evolutionary Machine Learning , pages=

    Evolution through large models , author=. Handbook of Evolutionary Machine Learning , pages=. 2023 , publisher=

    2023

  37. [45]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Motif: Multi-strategy optimization via turn-based interactive framework , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  38. [46]

    Machine Learning , volume=

    Finite-time analysis of the multiarmed bandit problem , author=. Machine Learning , volume=. 2002 , publisher=

    2002

  39. [47]

    Proceedings of the Genetic and Evolutionary Computation Conference , pages=

    Surrogate learning in meta-black-box optimization: A preliminary study , author=. Proceedings of the Genetic and Evolutionary Computation Conference , pages=

  40. [48]

    IEEE Transactions on Evolutionary Computation , year=

    Toward automated algorithm design: A survey and practical guide to meta-black-box optimization , author=. IEEE Transactions on Evolutionary Computation , year=

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.