Pith. sign in

REVIEW 4 major objections 4 minor 74 references

Putting an LLM agent in charge of a Bayesian optimization loop preserves sample efficiency and lets natural-language descriptions act as priors that improve on standard BO.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-04 00:45 UTC pith:CV6ZAEAF

load-bearing objection A genuinely new agentic-BO architecture with careful, honest experiments, but the natural-language prior claim is only tested with priors that are correct by construction, and the Branin result contradicts the stated parity with SOTA BO. the 4 major comments →

arxiv 2608.00316 v1 pith:CV6ZAEAF submitted 2026-07-31 cs.LG stat.ML

Agentic Bayesian Optimization through Surrogate-Augmented Autoresearch

Paul Brunzema , Louis Tiao , Nhat Le , Kevin De Angeli , Yao Xuan , Djordje Gligorijevic This is my paper
classification cs.LG stat.ML
keywords agentic Bayesian optimizationlarge language model agentsurrogate backendGaussian processacquisition functionnatural-language priorsrun-time reconfigurationautoresearch
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper introduces agentic Bayesian optimization: a paradigm in which a large language model (LLM) agent is the central decision maker in a Bayesian optimization (BO) campaign, while a Bayesian backend supplies uncertainty-aware candidate proposals. The central claim is that this division of labor preserves the sample efficiency of state-of-the-art BO when the agent has no domain knowledge, out-performs LLM-only optimizers that lack a calibrated surrogate, and converts natural-language problem descriptions into useful priors that improve on standard BO. The authors instantiate this in Sara, the agent, and lenz, a modular backend whose raw trial log survives reconfiguration, so the search strategy—bounds, acquisition function, even the objectives and constraints—can be revised mid-run without discarding data. Experiments span synthetic functions, hyperparameter tuning, and reaction-yield optimization; in a scaling-law study the agent re-frames a constrained single-objective problem as a multi-objective Pareto search after the user changes the requirements.

Core claim

Standard BO fixes its whole policy before the first evaluation: surrogate, acquisition function, search region, and the split of outcomes into objectives and constraints. The paper's central claim is that replacing this fixed policy with an LLM agent—which can inspect the surrogate, request and override proposals, and re-edit the configuration mid-run—costs nothing in sample efficiency and adds two capabilities. First, without any domain knowledge, the agent matches a well-tuned classical BO baseline on synthetic problems, while LLM-only baselines that discard the surrogate underperform, sometimes worse than random search. Second, with a natural-language problem description, the agent turns

What carries the argument

The load-bearing object is the metalevel deliberation process defined in Section 4. The agent's policy A maps a state—trial data, the current configuration (surrogate, acquisition, bounds, objective/constraint partition), append-only context, and deliberation history—to either a computational action (probe, reconfigure, propose) or an evaluation action. Three design choices carry the argument: the separation of propose from commit, so the agent can accept, refine, or override the backend's candidate; the append-only context K_t, so later instructions can supersede earlier ones without invalidating data; and the backend (lenz), whose command-line interface exposes commands for creating proble

Load-bearing premise

The demonstrated gains over standard BO assume the natural-language priors given to the agent are accurate and are correctly decoded; with misleading or vague priors the advantage would shrink or reverse, though the no-prior parity result would survive.

What would settle it

Run the reaction-yield benchmarks again, but hand the agent a deliberately wrong context (for example, a cold anhydrous Grignard recipe for the Suzuki–Miyaura task). If the wrong-prior runs still beat uninformed BO, then the claimed prior advantage does not depend on prior correctness; if they fall below the no-prior runs, the 'natural-language priors improve beyond standard BO' claim requires accurate priors.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Without any natural-language context, the agentic system performs on par with a tuned classical BO policy across low- to high-dimensional synthetic problems, showing that the added agent layer does not sacrifice sample efficiency.
  • With a natural-language description, the same system achieves substantially better early convergence than BO that must discover productive regions from scratch, on both hyperparameter-tuning and reaction-yield benchmarks.
  • LLM-only optimizers that propose points from text summaries without a calibrated surrogate underperform, at times worse than random sampling, when the objective has no recognizable structure; the surrogate is what makes the search systematic.
  • Mid-run reconfiguration—promoting a constraint to an objective and switching to a hypervolume-based acquisition function—works without discarding any evaluations, a capability standard BO does not offer.
  • The agent's tool-use pattern is context-dependent: with a prior it moves quickly to local refinement around the incumbent, while without one it relies on generic surrogate proposals longer; persistent acquisitions and bound edits remain rare on the tested tasks.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • This suggests a strict-generalization reading: any fixed BO policy is the special case where the agent always accepts the surrogate's proposal and never reconfigures; a formal regret comparison between the agent's policy and the fixed policy it could have run would clarify when deliberation pays.
  • The paper's own observation that the LLM can identify standard test functions from a few evaluations despite shifted optima implies synthetic-benchmark comparisons of LLM-based optimizers should be treated skeptically; random GP paths and fresh problem families are a more trustworthy evaluation surface.
  • A testable extension is fine-tuning the agent on the meta-MDP objective with token costs; if a small fine-tuned policy can replicate a frontier model's optimization decisions at a fraction of the token budget, the paradigm would become much cheaper to deploy.
  • The most direct practical consequence left implicit: agentic BO is best suited to expensive, evolving campaigns—such as experimental chemistry, hardware design, or ML system tuning—where requirements change mid-study and domain expertise exists mainly as text. A conversational interface that can re-target the objective is itself the product.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. This paper introduces agentic Bayesian optimization, in which an LLM agent is the central decision maker and delegates probabilistic modeling to a modular BoTorch-based backend (lenz). The framework is formalized as a metalevel decision process (Sec. 4) and instantiated as Sara, an LLM agent with a defined system prompt and CLI toolset. The experiments address four questions: no-prior parity with classical BO (Q1), use of natural-language priors (Q2), mid-run reconfiguration (Q3), and model/prompt ablations (Q4). The main empirical results are that Sara matches or slightly outperforms Ax on synthetic functions without a prior, outperforms LLAMBO and Centaur, improves faster with task descriptions on LCBench and synthetic reaction-yield functions, and can reformulate a constrained problem into a multi-objective one mid-run. The paper includes full system prompts and detailed appendices.

Significance. If the central claim holds, agentic BO is a meaningful advance: it combines an LLM's ability to consume unstructured priors and adapt strategy with calibrated GP-based search, and it demonstrates a dynamic-reconfiguration capability not available in standard BO. The paper is commendable for shipping full prompts, a detailed CLI reference, explicit anti-patterns, and honest discussion of benchmark recognition (Sec. D.1). The empirical support for the strongest headline—improvement beyond standard BO via natural-language priors—is not yet convincing, and the no-prior parity claim is partly confounded by pretraining memory. With additional controls, the contribution could be significant.

major comments (4)
  1. [A.2, Table 5, Fig. 15; Sec. 6.3] The Q2 claim that natural-language priors improve beyond standard BO is tested only with priors that are correct by construction. The reaction-yield contexts in Fig. 15 specify the reaction, catalyst, solvent, and base, and Table 5 places each function's optimum in exactly the regime those cues identify (e.g., Suzuki mu_T=70 C, mu_w=0.45). The ablation in Fig. 11 only contrasts this informative prior with no prior. No experiment supplies a misleading, vague, or partially wrong prior, so the demonstrated gain may reflect prior accuracy rather than agentic use of priors. Please add wrong/perturbed prior conditions and, if the claim is limited to correct priors, state that limitation.
  2. [5.2, Table 2, Sec. 7] Sara's system prompt was explicitly iterated after observing failure modes in early experiments (Table 2 lists directives added because 'we observed' specific behaviors). The paper does not establish that the final benchmark tasks were held out from this prompt-development loop. Since the headline results compare Sara against baselines on these same tasks, prompt overfitting to the test suite is a live threat. Please clarify whether prompt tuning was performed on a separate development set; if not, add a validation split or a comparison with a generic untuned agent prompt.
  3. [6.3, 6.2, 6.5] The 'beyond standard BO' comparison in the prior-informed setting is asymmetric: Sara, LLAMBO, and Centaur receive the natural-language description, while Ax receives no prior at all. This conflates the availability of a domain prior with the agentic architecture. A standard BO method with the same prior injected through a conventional mechanism (e.g., piBO with user beliefs, or a narrow initial search region derived from the description) would separate prior encoding from agentic control. Without such a baseline, the distinctive added value of agentic BO over standard BO is not established.
  4. [6.1, D.1] The no-prior synthetic benchmarks are contaminated by pretraining memory. The paper itself reports that the bash-only agent explicitly identified the benchmark in 4/10 Hartmann, 3/10 constrained Hartmann, and 10/10 Ackley-10 runs despite renaming and shifting (Sec. D.1). This means the Q1 parity claim on these functions is not a clean measure of optimization ability. The GP-sample-path experiments are a good control and show a clear surrogate benefit, but they are multi-objective or high-dimensional and do not fully substitute for a single-objective no-prior parity test. Please either report no-prior parity on fresh single-objective surfaces or soften the claim.
minor comments (4)
  1. [Table 3] The formatting of Table 3 is ambiguous in the text: sub-/superscript q25/q75 values are run together with medians (e.g., '0.00060.0010 0.0005'). Please reformat so medians and quartiles are clearly distinguished and define the order explicitly.
  2. [Sec. 6.1, Table 3] The summary statement 'Sara matches Ax' is slightly stronger than the table for Branin with Opus 4.8 (median 0.0020 vs Ax 0.0006). The difference is small, but please ensure the prose and table are consistent.
  3. [Sec. 6.5, C.1] The finding that reasoning level 'off' outperforms higher reasoning levels on Mizoroki-Heck is interesting, but no statistical test is reported. A Mann-Whitney test or confidence intervals would clarify whether this is a robust effect.
  4. [General] The paper would benefit from an explicit code/data availability statement. The full prompts and CLI reference are valuable, but releasing lenz and the evaluation harness would make the empirical claims much easier to reproduce and extend.

Circularity Check

1 steps flagged

Reaction-yield 'NL prior improves BO' result is partly built into the benchmark construction; no-prior and LCBench evidence remain independent.

specific steps
  1. self definitional [Section A.2 (Table 5 and Figure 15), Section 6.3; abstract claim 'uses natural-language priors to improve beyond standard BO']
    "We instantiate four reactions whose regime parameters reflect their qualitative process requirements (Table 5): Suzuki–Miyaura coupling (warm, aqueous, mildly basic; ymax = 95%) ... The context string given to the agent names the reaction and its catalyst system, solvent, and base—enough chemical information to identify the favorable operating regime, but no explicit parameter values (cf. Figure 15)."

    The reaction-yield ground truth is constructed so its optimum (e.g., Suzuki μT=70°C, μw=0.45 in Table 5) is the quantitative image of the same qualitative chemistry ('warm, aqueous, mildly basic'; 'THF/water mixture with a mild base') that is then supplied to Sara as the natural-language prior. A correct LLM mapping from the prior to operating conditions therefore lands near the optimum by construction; the measured 'improvement beyond standard BO' on this family does not test a prior independent of the answer, nor any vague or misleading prior. The abstract's unqualified claim thus reduces, for this benchmark family, to the construction of the test function from the prior content. LCBench and no-prior results supply independent evidence, so the circularity is partial.

full rationale

We found no evidence of self-citation chains, imported uniqueness theorems, or fitted constants whose predictions equal their inputs. The no-prior parity results (Section 6.1), the GP sample-path benchmarks (Sections A.1/D.1), and the LCBench HPO experiments (Section 6.2) are self-contained and externally meaningful; evaluation counts are honest, and the dynamic-reconfiguration demo is a capability illustration rather than a fitted prediction. However, the Q2 claim that 'natural-language priors improve beyond standard BO' is partially circular for the reaction-yield family: the benchmark's optimum regime parameters (Table 5) were chosen to 'reflect their qualitative process requirements,' and the prior text supplied to the agent states exactly those qualitative requirements (Figure 15). Thus the speedup attributed to the prior is, on those four tasks, guaranteed by how the test function was constructed, not by a general property of NL priors; no vague/misleading/wrong prior is tested. The paper's own disclosures—prompt directives tuned after observing failure modes (Table 2) and the reported benchmark recognition by the bash-only agent (Section D.1)—are additional validity threats but do not by themselves constitute a circular derivation. Because independent evidence (LCBench, no-prior parity) remains, the circularity is partial rather than total.

Axiom & Free-Parameter Ledger

3 free parameters · 6 axioms · 0 invented entities

No new physical or mathematical entities are postulated. The system components Sara and lenz are new software artifacts with benchmark evidence, not entities without independent handles. The main ledger items are tunable design choices (prompt directives, benchmark regime placements) that act as de facto free parameters for the empirical claims.

free parameters (3)
  • System-prompt directives (Table 2) = Six directives plus an anti-pattern list; final wording in Appendix F
    Directives were selected after observing agent failure modes in early experiments (Table 2: 'We observed that without explicit ownership, the agent can become lazy...'). The prompt is an iteratively fitted component that shapes all reported results.
  • Reaction-yield benchmark regime placements (Table 5) = e.g., Suzuki μ_T=70 °C, μ_w=0.45; Grignard μ_T=5 °C, μ_w=0.0
    Optima placed by the authors at chemically sensible locations so the author-written natural-language context points at the true regime; this alignment creates the 'prior helps' result.
  • lenz 'suggest --around' default radius = 0.1 (fraction of domain width)
    Hand-chosen default in the lenz CLI (Appendix G) that fixes local-refinement step size; not swept or justified.
axioms (6)
  • standard math GP with Matérn kernel, input normalization, output standardization, and logEI-family acquisition provide calibrated uncertainty for the search spaces used.
    Standard BO machinery (Rasmussen & Williams 2006; Balandat et al. 2020); invoked throughout Section 5.1 and the lenz CLI.
  • domain assumption Pretraining obfuscation (renamed parameters, shifted optima, token-named sandboxes) prevents LLMs from using memorized solutions in the synthetic benchmarks.
    Load-bearing for Figure 5. The paper itself shows the veil is thin: bash-only Sara recognized Ackley-10 in 10/10 runs and Hartmann in 4/10 runs despite the obfuscation (§D.1).
  • domain assumption The natural-language descriptions given to Sara convey true, actionable information about the objective (reaction class maps to operating regime; funnel-MLP description maps to parameter scales).
    Prior-advantage claims (Q2, Figures 7 and 11) assume the LLM-decoded prior is correct; no experiments with misleading or noisy priors are reported.
  • domain assumption Random-Fourier-feature GP sample paths (A.1) are absent from LLM pretraining and are genuinely multi-modal, non-separable landscapes.
    Supports the unconfounded part of the surrogate-value claim (Figure 20). Reasonable but not verifiable from the paper alone.
  • domain assumption Ten seeds of one LLM generation are representative of agentic BO policy behavior.
    Agent behavior is nondeterministic and prompt-sensitive (Section 7); all headline results are medians over 10 seeds with wide IQRs.
  • standard math The meta-MDP formulation (Section E) is a faithful completion of the decision process.
    Used for conceptual framing; the authors state the induced policy is not optimal for it and explicitly leave learning the policy to future work.

pith-pipeline@v1.3.0-alltime-deepseek · 36403 in / 20865 out tokens · 202407 ms · 2026-08-04T00:45:06.054027+00:00 · methodology

0 comments
read the original abstract

Bayesian optimization (BO) has become the standard tool for sample-efficient optimization and owes its efficiency to uncertainty-aware search driven by generic statistical priors. Richer domain priors can improve BO in principle, but encoding them through tailored kernels or problem structure is difficult and rarely done in practice. LLMs can help sidestep this difficulty by making informal priors from natural language, code, and documentation directly available to the optimizer. However, existing LLM-based BO methods either insert the LLM into a fixed role (surrogate, acquisition proxy, or configuration interface) or hand it broad control, sacrificing the systematic exploration that makes BO reliable. We introduce agentic Bayesian optimization: a paradigm in which an LLM agent is the central decision maker in the BO loop while a Bayesian backend provides the uncertainty-aware optimization substrate. The agent configures the problem, queries the backend, selects and commits evaluations, and can revise the optimization strategy during the run by tightening bounds, switching acquisition functions, proposing targeted evaluations, or even reframing the problem following new instructions or observed evidence. We instantiate this idea in Sara, a surrogate-augmented autoresearch agent, and lenz, a modular BoTorch-based backend that the agent can inspect and modify through a structured interface. Across synthetic and real-world benchmarks, Sara preserves the reliability of state-of-the-art BO without prior knowledge, outperforms LLM-based baselines, and uses natural-language priors to improve beyond standard BO. We further demonstrate the practical value of agentic BO in dynamic settings, where Sara reconfigures the full optimization problem on the fly as requirements change, a capability not previously available in standard BO.

discussion (0)

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

Reference graph

Works this paper leans on

74 extracted references · 19 linked inside Pith

  1. [1]

    Searching for optimal solutions with llms via Bayesian optimization

    Dhruv Agarwal, Manoj Ghuhan Arivazhagan, Rajarshi Das, Sandesh Swamy, Sopan Khosla, and Rashmi Gangadharaiah. Searching for optimal solutions with llms via Bayesian optimization. In International Conference on Learning Representations (ICLR), 2025

    2025

  2. [2]

    Optuna : A next-generation hyperparameter optimization framework

    Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. Optuna : A next-generation hyperparameter optimization framework. In ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), 2019

    2019

  3. [3]

    Unexpected improvements to expected improvement for Bayesian optimization

    Sebastian Ament, Samuel Daulton, David Eriksson, Maximilian Balandat, and Eytan Bakshy. Unexpected improvements to expected improvement for Bayesian optimization. Advances in Neural Information Processing Systems (NeurIPS), 36: 0 20577--20612, 2023

    2023

  4. [4]

    Jiang, Samuel Daulton, Benjamin Letham, Andrew Gordon Wilson, and Eytan Bakshy

    Maximilian Balandat, Brian Karrer, Daniel R. Jiang, Samuel Daulton, Benjamin Letham, Andrew Gordon Wilson, and Eytan Bakshy. BoTorch : A framework for efficient monte-carlo Bayesian optimization. In Advances in Neural Information Processing Systems (NeurIPS), 2020

    2020

  5. [5]

    Random search for hyper-parameter optimization

    James Bergstra and Yoshua Bengio. Random search for hyper-parameter optimization. Journal of Machine Learning Research, 13: 0 281--305, 2012

    2012

  6. [6]

    Algorithms for hyper-parameter optimization

    James Bergstra, R \'e mi Bardenet, Yoshua Bengio, and Bal \'a zs K \'e gl. Algorithms for hyper-parameter optimization. In Advances in Neural Information Processing Systems (NeurIPS), 2011

    2011

  7. [7]

    Boiko, Robert MacKnight, Ben Kline, and Gabe Gomes

    Daniil A. Boiko, Robert MacKnight, Ben Kline, and Gabe Gomes. Autonomous chemical research with large language models. Nature, 624: 0 570--578, 2023

    2023

  8. [8]

    Baye SQP : Bayesian optimization through sequential quadratic programming

    Paul Brunzema and Sebastian Trimpe. Baye SQP : Bayesian optimization through sequential quadratic programming. In Advanced in Neural Information Processing Systems (NeurIPS), 2025

    2025

  9. [9]

    Bayesian optimization via continual variational last layer training

    Paul Brunzema, Mikkel Jordahn, John Willes, Sebastian Trimpe, Jasper Snoek, and James Harrison. Bayesian optimization via continual variational last layer training. In International Conference on Learning Representations (ICLR), volume 2025, pages 56480--56509, 2025

    2025

  10. [10]

    Krueger, Thomas L

    Frederick Callaway, Bas van Opheusden, Sayan Gul, Priyam Das, Paul M. Krueger, Thomas L. Griffiths, and Falk Lieder. Rational use of cognitive resources in human planning. Nature Human Behaviour, 6: 0 1112--1125, 2022

    2022

  11. [11]

    LLINBO : Trustworthy llm-in-the-loop Bayesian optimization

    Chih-Yu Chang, Milad Azvar, Chinedum Okwudire, and Raed Al Kontar. LLINBO : Trustworthy llm-in-the-loop Bayesian optimization. arXiv preprint arXiv:2505.14756, 2025

    arXiv 2025

  12. [12]

    EvoPrompting : Language models for code-level neural architecture search

    Angelica Chen, David Dohan, and David So. EvoPrompting : Language models for code-level neural architecture search. In Advances in Neural Information Processing Systems (NeurIPS), 2023 a

    2023

  13. [13]

    Wenhu Chen, Xueguang Ma, Xinyi Wang, and William W. Cohen. Program of thoughts prompting: Disentangling computation from reasoning for numerical reasoning tasks. Transactions on Machine Learning Research (TMLR), 2023 b

    2023

  14. [14]

    Towards learning universal hyperparameter optimizers with transformers

    Yutian Chen, Xingyou Song, Chansoo Lee, Zi Wang, Richard Zhang, David Dohan, Kazuya Kawakami, Greg Kochanski, Arnaud Doucet, Marc'Aurelio Ranzato, Sagi Perel, and Nando de Freitas. Towards learning universal hyperparameter optimizers with transformers. In Advances in Neural Information Processing Systems (NeurIPS), 2022

    2022

  15. [15]

    Labo: Llm-accelerated Bayesian optimization through broad exploration and selective experimentation

    Zhuo Chen, Xinzhe Yuan, Jianshu Zhang, Jinzong Dong, Ruichen Zhou, Yingchun Niu, Tianhang Zhou, Yu Yang Fredrik Liu, Yuqiang Li, Nanyang Ye, et al. Labo: Llm-accelerated Bayesian optimization through broad exploration and selective experimentation. In International Conference on Machine Learning, 2026

    2026

  16. [16]

    Language-based Bayesian optimization research assistant (bora)

    Abdoulatif Ciss \'e , Xenophon Evangelopoulos, Vladimir V Gusev, and Andrew I Cooper. Language-based Bayesian optimization research assistant (bora). In International Joint Conference on Artificial Intelligence (IJCAI), pages 4967--4975, 2025

    2025

  17. [17]

    Differentiable expected hypervolume improvement for parallel multi-objective Bayesian optimization

    Samuel Daulton, Maximilian Balandat, and Eytan Bakshy. Differentiable expected hypervolume improvement for parallel multi-objective Bayesian optimization. In Advances in Neural Information Processing Systems (NeurIPS), 2020

    2020

  18. [18]

    Parallel Bayesian optimization of multiple noisy objectives with expected hypervolume improvement

    Samuel Daulton, Maximilian Balandat, and Eytan Bakshy. Parallel Bayesian optimization of multiple noisy objectives with expected hypervolume improvement. In Advances in Neural Information Processing Systems (NeurIPS), 2021

    2021

  19. [19]

    Scalable constrained Bayesian optimization

    David Eriksson and Matthias Poloczek. Scalable constrained Bayesian optimization. In International Conference on Artificial Intelligence and Statistics (AISTATS), 2021

    2021

  20. [20]

    Gardner, Ryan D

    David Eriksson, Michael Pearce, Jacob R. Gardner, Ryan D. Turner, and Matthias Poloczek. Scalable global optimization via local Bayesian optimization. In Advances in Neural Information Processing Systems (NeurIPS), 2019

    2019

  21. [21]

    Dynamic priors in Bayesian optimization for hyperparameter optimization

    Lukas Fehring, Marcel Wever, Maximilian Splieth \"o ver, Leona Hennig, Henning Wachsmuth, and Marius Lindauer. Dynamic priors in Bayesian optimization for hyperparameter optimization. arXiv preprint arXiv:2511.02570, 2025

    Pith/arXiv arXiv 2025

  22. [22]

    Can LLMs beat classical hyperparameter optimization algorithms? a study on autoresearch

    Fabio Ferreira, Lucca Wobbe, Arjun Krishnakumar, Frank Hutter, and Arber Zela. Can LLMs beat classical hyperparameter optimization algorithms? a study on autoresearch. arXiv preprint arXiv:2603.24647, 2026

    Pith/arXiv arXiv 2026

  23. [23]

    Peter I. Frazier. A tutorial on Bayesian optimization. arXiv preprint arXiv:1807.02811, 2018

    Pith/arXiv arXiv 2018

  24. [24]

    PAL : Program-aided language models

    Luyu Gao, Aman Madaan, Shuyan Zhou, Uri Alon, Pengfei Liu, Yiming Yang, Jamie Callan, and Graham Neubig. PAL : Program-aided language models. In International Conference on Machine Learning (ICML), 2023

    2023

  25. [25]

    Gpytorch: Blackbox matrix-matrix Gaussian process inference with gpu acceleration

    Jacob Gardner, Geoff Pleiss, Kilian Q Weinberger, David Bindel, and Andrew G Wilson. Gpytorch: Blackbox matrix-matrix Gaussian process inference with gpu acceleration. Advances in Neural Information Processing Systems (NeurIPS), 31, 2018

    2018

  26. [26]

    Bayesian optimization with inequality constraints

    Jacob R Gardner, Matt J Kusner, Zhixiang Eddie Xu, Kilian Q Weinberger, and John P Cunningham. Bayesian optimization with inequality constraints. In International Conference on Machine Learning (ICML), volume 2014, pages 937--945, 2014

    2014

  27. [27]

    Bayesian Optimization

    Roman Garnett. Bayesian Optimization. Cambridge University Press, 2023

    2023

  28. [28]

    Bayesian optimization with unknown constraints

    Michael A Gelbart, Jasper Snoek, and Ryan P Adams. Bayesian optimization with unknown constraints. arXiv preprint arXiv:1403.5607, 2014

    Pith/arXiv arXiv 2014

  29. [29]

    Llms for Bayesian optimization in scientific domains: Are we there yet? arXiv preprint arXiv:2509.21403, 2025

    Rushil Gupta, Jason Hartford, and Bang Liu. Llms for Bayesian optimization in scientific domains: Are we there yet? arXiv preprint arXiv:2509.21403, 2025

    arXiv 2025

  30. [30]

    The CMA evolution strategy: A tutorial

    Nikolaus Hansen. The CMA evolution strategy: A tutorial. arXiv preprint arXiv:1604.00772, 2016

    Pith/arXiv arXiv 2016

  31. [31]

    Selecting computations: Theory and applications

    Nicholas Hay, Stuart Russell, David Tolpin, and Solomon Eyal Shimony. Selecting computations: Theory and applications. In Uncertainty in Artificial Intelligence (UAI), 2012

    2012

  32. [32]

    Training compute-optimal large language models

    Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, Elena Buchatskaya, Trevor Cai, Eliza Rutherford, DDL Casas, Lisa Anne Hendricks, Johannes Welbl, Aidan Clark, et al. Training compute-optimal large language models. arXiv preprint arXiv:2203.15556, 10, 2022

    Pith/arXiv arXiv 2022

  33. [33]

    Large language models for automated data science: Introducing CAAFE for context-aware automated feature engineering

    Noah Hollmann, Samuel M \"u ller, and Frank Hutter. Large language models for automated data science: Introducing CAAFE for context-aware automated feature engineering. In Advances in Neural Information Processing Systems (NeurIPS), 2023

    2023

  34. [34]

    MetaGPT : Meta programming for a multi-agent collaborative framework

    Sirui Hong, Mingchen Zhuge, Jonathan Chen, Xiawu Zheng, Yuheng Cheng, Ceyao Zhang, Jinlin Wang, Zili Wang, Steven Ka Shing Yau, Zijuan Lin, Liyang Zhou, Chenyu Ran, Lingfeng Xiao, Chenglin Wu, and J \"u rgen Schmidhuber. MetaGPT : Meta programming for a multi-agent collaborative framework. In International Conference on Learning Representations (ICLR), 2024

    2024

  35. [35]

    Benchmarking large language models as ai research agents

    Qian Huang, Jian Vora, Percy Liang, and Jure Leskovec. Benchmarking large language models as ai research agents. arXiv preprint arXiv:2310.03302, 2023

    Pith/arXiv arXiv 2023

  36. [36]

    BO : Augmenting acquisition functions with user beliefs for Bayesian optimization

    Carl Hvarfner, Danny Stoll, Artur Souza, Marius Lindauer, Frank Hutter, and Luigi Nardi. BO : Augmenting acquisition functions with user beliefs for Bayesian optimization. In International Conference on Learning Representations (ICLR), 2022

    2022

  37. [37]

    Jones, Matthias Schonlau, and William J

    Donald R. Jones, Matthias Schonlau, and William J. Welch. Efficient global optimization of expensive black-box functions. Journal of Global Optimization, 13 0 (4): 0 455--492, 1998

    1998

  38. [38]

    Language models (mostly) know what they know

    Saurav Kadavath, Tom Conerly, Amanda Askell, Tom Henighan, Dawn Drain, Ethan Perez, Nicholas Schiefer, Zac Hatfield-Dodds, Nova DasSarma, Eli Tran-Johnson, et al. Language models (mostly) know what they know. arXiv preprint arXiv:2207.05221, 2022

    Pith/arXiv arXiv 2022

  39. [39]

    LILO : Bayesian optimization with natural language feedback

    Katarzyna Kobalczyk, Jerry Lin, Benjamin Letham, Zhuokai Zhao, Maximilian Balandat, and Eytan Bakshy. LILO : Bayesian optimization with natural language feedback. In International Conference on Machine Learning (ICML), 2026. arXiv:2510.17671

    Pith/arXiv arXiv 2026

  40. [40]

    Agustinus Kristiadi, Felix Strieth-Kalthoff, Marta Skreta, Pascal Poupart, Alan Aspuru-Guzik, and Geoff Pleiss. A sober look at LLM s for material discovery: Are they actually good for B ayesian optimization over molecules? In International Conference on Machine Learning (ICML), pages 25603--25622, 2024

    2024

  41. [41]

    Hyperband : A novel bandit-based approach to hyperparameter optimization

    Lisha Li, Kevin Jamieson, Giulia DeSalvo, Afshin Rostamizadeh, and Ameet Talwalkar. Hyperband : A novel bandit-based approach to hyperparameter optimization. Journal of Machine Learning Research, 18 0 (185): 0 1--52, 2018

    2018

  42. [42]

    A study of Bayesian neural network surrogates for Bayesian optimization

    Yucen Li, Tim GJ Rudner, and Andrew Gordon Wilson. A study of Bayesian neural network surrogates for Bayesian optimization. In International Conference on Learning Representations (ICLR), volume 2024, pages 47003--47041, 2024

    2024

  43. [43]

    Large language model agent for hyper-parameter optimization

    Siyi Liu, Chen Gao, and Yong Li. Large language model agent for hyper-parameter optimization. arXiv preprint arXiv:2402.01881, 2024 a

    Pith/arXiv arXiv 2024

  44. [44]

    Large language models to enhance Bayesian optimization

    Tennison Liu, Raul Astudillo, Vasan Bhavana, and Mihaela van der Schaar. Large language models to enhance Bayesian optimization. In International Conference on Learning Representations (ICLR), 2024 b . arXiv:2402.03921

    Pith/arXiv arXiv 2024

  45. [45]

    The AI scientist: Towards fully automated open-ended scientific discovery

    Chris Lu, Cong Lu, Robert Tjarko Lange, Jakob Foerster, Jeff Clune, and David Ha. The AI scientist: Towards fully automated open-ended scientific discovery. arXiv preprint arXiv:2408.06292, 2024

    Pith/arXiv arXiv 2024

  46. [46]

    Bran, Sam Cox, Oliver Schilter, Carlo Baldassari, Andrew D

    Andres M. Bran, Sam Cox, Oliver Schilter, Carlo Baldassari, Andrew D. White, and Philippe Schwaller. ChemCrow : Augmenting large-language models with chemistry tools. arXiv preprint arXiv:2304.05376, 2023

    Pith/arXiv arXiv 2023

  47. [47]

    OptiMindTune : A multi-agent framework for intelligent hyperparameter optimization

    Meher Bhaskar Madiraju and Meher Sai Preetam Madiraju. OptiMindTune : A multi-agent framework for intelligent hyperparameter optimization. arXiv preprint arXiv:2505.19205, 2025

    Pith/arXiv arXiv 2025

  48. [48]

    Sequential large language model-based hyper-parameter optimization

    Kanan Mahammadli and Seyda Ertekin. Sequential large language model-based hyper-parameter optimization. arXiv preprint arXiv:2410.20302, 2024

    Pith/arXiv arXiv 2024

  49. [49]

    On the design of LQR kernels for efficient controller learning

    Alonso Marco, Philipp Hennig, Stefan Schaal, and Sebastian Trimpe. On the design of LQR kernels for efficient controller learning. In 2017 IEEE 56th Annual Conference on Decision and Control (CDC), pages 5193--5200. IEEE, 2017

    2017

  50. [50]

    Adaptive acquisition selection for Bayesian optimization with large language models

    Giang Ngo, Dat Phan Trong, Dang Nguyen, Sunil Gupta, and Svetha Venkatesh. Adaptive acquisition selection for Bayesian optimization with large language models. In International Conference on Learning Representations (ICLR), 2026

    2026

  51. [51]

    Ax: A platform for adaptive experimentation

    Miles Olson, Elizabeth Santorella, Louis C Tiao, Sait Cakmak, Mia Garrard, Samuel Daulton, Zhiyuan Jerry Lin, Sebastian Ament, Bernard Beckerman, Eric Onofrey, et al. Ax: A platform for adaptive experimentation. In AutoML 2025 ABCD Track, 2025

    2025

  52. [52]

    Random features for large-scale kernel machines

    Ali Rahimi and Benjamin Recht. Random features for large-scale kernel machines. Advances in Neural Information Processing Systems (NeurIPS), 20, 2007

    2007

  53. [53]

    Large language models as uncertainty-calibrated optimizers for experimental discovery

    Bojana Rankovi \'c , Ryan-Rhys Griffiths, and Philippe Schwaller. Large language models as uncertainty-calibrated optimizers for experimental discovery. arXiv preprint arXiv:2504.06265, 2025

    arXiv 2025

  54. [54]

    Carl Edward Rasmussen and Christopher K. I. Williams. Gaussian Processes for Machine Learning. MIT Press, 2006

    2006

  55. [55]

    Turner, and David Duvenaud

    James Requeima, John Bronskill, Dami Choi, Richard E. Turner, and David Duvenaud. LLM processes: Numerical predictive distributions conditioned on natural language. In International Conference on Machine Learning (ICML), 2024

    2024

  56. [56]

    Pawan Kumar, Emilien Dupont, Francisco J

    Bernardino Romera-Paredes, Mohammadamin Barekatain, Alexander Novikov, Matej Balog, M. Pawan Kumar, Emilien Dupont, Francisco J. R. Ruiz, Jordan S. Ellenberg, Pengming Wang, Omar Fawzi, Pushmeet Kohli, and Alhussein Fawzi. Mathematical discoveries from program search with large language models. Nature, 625: 0 468--475, 2024

    2024

  57. [57]

    Toolformer : Language models can teach themselves to use tools

    Timo Schick, Jane Dwivedi-Yu, Roberto Dess \`i , Roberta Raileanu, Maria Lomeli, Luke Zettlemoyer, Nicola Cancedda, and Thomas Scialom. Toolformer : Language models can teach themselves to use tools. In Advances in Neural Information Processing Systems (NeurIPS), 2023

    2023

  58. [58]

    Improving llm-based global optimization with search space partitioning

    Andrej Schwanke, Lyubomir Ivanov, David Salinas, Fabio Ferreira, Aaron Klein, Frank Hutter, and Arber Zela. Improving llm-based global optimization with search space partitioning. arXiv preprint arXiv:2505.21372, 2025

    arXiv 2025

  59. [59]

    Hyperparameter optimization via interacting with probabilistic circuits

    Jonas Seng, Fabrizio Ventola, Zhongjie Yu, and Kristian Kersting. Hyperparameter optimization via interacting with probabilistic circuits. In International Conference on Automated Machine Learning (AutoML), pages 11--1. PMLR, 2025

    2025

  60. [60]

    Adams, and Nando de Freitas

    Bobak Shahriari, Kevin Swersky, Ziyu Wang, Ryan P. Adams, and Nando de Freitas. Taking the human out of the loop: A review of Bayesian optimization. Proceedings of the IEEE, 104 0 (1): 0 148--175, 2016

    2016

  61. [61]

    Jasper Snoek, Hugo Larochelle, and Ryan P. Adams. Practical Bayesian optimization of machine learning algorithms. In Advances in Neural Information Processing Systems (NeurIPS), 2012

    2012

  62. [62]

    Position: leverage foundational models for black-box optimization

    Xingyou Song, Yingtao Tian, Robert Tjarko Lange, Chansoo Lee, Yujin Tang, and Yutian Chen. Position: leverage foundational models for black-box optimization. In International Conference on Machine Learning (ICML), 2024

    2024

  63. [63]

    Kakade, and Matthias Seeger

    Niranjan Srinivas, Andreas Krause, Sham M. Kakade, and Matthias Seeger. Gaussian process optimization in the bandit setting: No regret and experimental design. In International Conference on Machine Learning (ICML), 2010

    2010

  64. [64]

    Adaptive kernel design for Bayesian optimization is a piece of cake with llms

    Richard Suwandi, Feng Yin, Juntao Wang, Renjie Li, Tsung-Hui Chang, and Sergios Theodoridis. Adaptive kernel design for Bayesian optimization is a piece of cake with llms. Advances in Neural Information Processing Systems (NeurIPS), 38, 2026

    2026

  65. [65]

    Manning, and Chelsea Finn

    Katherine Tian, Eric Mitchell, Huaxiu Yao, Christopher D. Manning, and Chelsea Finn. Just ask for calibration: Strategies for eliciting calibrated confidence scores from language models fine-tuned with human feedback. In Conference on Empirical Methods in Natural Language Processing (EMNLP), 2023

    2023

  66. [66]

    AutoML in the age of large language models: Current challenges, future opportunities and risks

    Alexander Tornede, Difan Deng, Theresa Eimer, Joseph Giovanelli, Aditya Mohan, Tim Ruhkopf, Sarah Segel, Daphne Theodorakopoulos, Tanja Tornede, Henning Wachsmuth, and Marius Lindauer. AutoML in the age of large language models: Current challenges, future opportunities and risks. Transactions on Machine Learning Research (TMLR), 2024

    2024

  67. [67]

    Voyager : An open-ended embodied agent with large language models

    Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, and Anima Anandkumar. Voyager : An open-ended embodied agent with large language models. Transactions on Machine Learning Research, 2023

    2023

  68. [68]

    Can LLMs express their uncertainty? an empirical evaluation of confidence elicitation in LLMs

    Miao Xiong, Zhiyuan Hu, Xinyang Lu, Yifei Li, Jie Fu, Junxian He, and Bryan Hooi. Can LLMs express their uncertainty? an empirical evaluation of confidence elicitation in LLMs . arXiv preprint arXiv:2306.13063, 2024

    Pith/arXiv arXiv 2024

  69. [69]

    Le, Denny Zhou, and Xinyun Chen

    Chengrun Yang, Xuezhi Wang, Yifeng Lu, Hanxiao Liu, Quoc V. Le, Denny Zhou, and Xinyun Chen. Large language models as optimizers. In International Conference on Learning Representations (ICLR), 2024. arXiv:2309.03409

    Pith/arXiv arXiv 2024

  70. [70]

    Reasoning BO : Enhancing Bayesian optimization with the long-context reasoning power of LLM s

    Zhuo Yang, Daolang Wang, Lingli Ge, Beilun Wang, Tianfan Fu, and Yuqiang Li. Reasoning BO : Enhancing Bayesian optimization with the long-context reasoning power of LLM s. arXiv preprint arXiv:2505.12833, 2025

    arXiv 2025

  71. [71]

    ReAct : Synergizing reasoning and acting in language models

    Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, and Yuan Cao. ReAct : Synergizing reasoning and acting in language models. In International Conference on Learning Representations (ICLR), 2023

    2023

  72. [72]

    Zhang, Nishkrit Desai, Juhan Bae, Jonathan Lorraine, and Jimmy Ba

    Michael R. Zhang, Nishkrit Desai, Juhan Bae, Jonathan Lorraine, and Jimmy Ba. Using large language models for hyperparameter optimization. arXiv preprint arXiv:2312.04528, 2023 a

    Pith/arXiv arXiv 2023

  73. [73]

    AutoML-GPT : Automatic machine learning with GPT

    Shujian Zhang, Chengyue Gong, Lemeng Wu, Xingchao Liu, and Mingyuan Zhou. AutoML-GPT : Automatic machine learning with GPT . arXiv preprint arXiv:2305.02499, 2023 b

    Pith/arXiv arXiv 2023

  74. [74]

    Auto- PyTorch : Multi-fidelity metalearning for efficient and robust AutoDL

    Lucas Zimmer, Marius Lindauer, and Frank Hutter. Auto- PyTorch : Multi-fidelity metalearning for efficient and robust AutoDL . IEEE Transactions on Pattern Analysis and Machine Intelligence, 43 0 (9): 0 3079--3090, 2021

    2021