Pith. sign in

REVIEW 3 major objections 5 minor 2 cited by

This paper presents Bgolearn, a Python framework that makes Bayesian optimization practical for materials research, and claims it cuts required experiments by 40–60% while finding new alloys and structures.

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

2026-08-03 11:18 UTC pith:NMGCXPXY

load-bearing objection Useful open-source BO toolkit, but the headline 40–60% claim is not supported by the reported benchmarks—grid search and single-objective GA were never run. the 3 major comments →

arxiv 2601.06820 v2 pith:NMGCXPXY submitted 2026-01-11 cond-mat.mtrl-sci

Bgolearn: a Unified Bayesian Optimization Framework for Accelerating Materials Discovery

Bin Cao , Jie Xiong , Jiaxuan Ma , Yuan Tian , Yirui Hu , Mengwei He , Longhan Zhang , Jiayu Wang
show 6 more authors
Jian Hui Li Liu Dezhen Xue Turab Lookman Jun Wang Tong-Yi Zhang
This is my paper
classification cond-mat.mtrl-sci
keywords Bayesian optimizationmaterials discoverymulti-objective optimizationactive learninghigh-entropy alloysmedium-Mn steeltriply periodic minimal surfacesopen-source software
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.

The paper sets out to show that the adoption barrier for Bayesian optimization in materials science is largely an implementation problem, and that a unified open-source package can solve it. Bgolearn bundles data handling, surrogate models, acquisition functions, and uncertainty quantification behind a simple interface, so a materials researcher can run single- or multi-objective optimization with minimal code. The quantitative anchor is a benchmark result: Bgolearn reaches 90-percent optimality on Hartmann-6D in 18±4 iterations and on Ackley-5D in 22±4 iterations, versus 87±15 and 72±18 for random search, which the paper translates into a 40–60% reduction in experimental effort. Its case studies then show the framework recommending an ultra-hard high-entropy alloy (>1000 HV), a TPMS structure with elastic modulus 8,945 MPa, and medium-Mn steel heat treatments that extend the empirical strength–ductility frontier. The reader is being asked to accept that BO's sample-efficiency advantages, long demonstrated in theory, can now be delivered in practice without specialized programming.

Core claim

On its own terms, the paper's central discovery is that a unified, materials-oriented Bayesian-optimization framework can deliver both ease of use and practical discovery gains. It reports that, across standard single- and multi-objective benchmarks, Bgolearn reaches 90-percent optimality in roughly 18–28 iterations where random search needs 72–87, and it achieves higher hypervolume on ZDT1 and DTLZ2 than genetic search. The framework combines five surrogate families, five single-objective and four multi-objective acquisition functions, and bootstrap-based uncertainty quantification so that non-Gaussian-process surrogates can still drive exploration. Applied to three real problems, Bgolearn

What carries the argument

The carrying mechanism is the Bgolearn package itself: a three-layer architecture—data, surrogate, acquisition—that reduces a Bayesian-optimization loop to a few lines of code while allowing expert customization. Within it, the acquisition functions do the decision-making work: expected improvement is the default single-objective driver, expected hypervolume improvement (EHVI), which picks candidates that most expand the dominated volume in objective space, guides multi-objective searches, and bootstrap-resampled ensembles supply the predictive uncertainty that non-GP surrogates lack. The key design claim is that automating normalization, cross-validation, jitter, and convergence diagnostics

Load-bearing premise

The load-bearing assumption is that the 40–60% reduction measured on smooth, noiseless benchmark functions carries over to real materials experiments, which are noisy, expensive, and sometimes involve discrete or categorical variables.

What would settle it

Run Bgolearn and a random-search baseline on a real materials system—for example, the same medium-Mn steel heat-treatment protocol—starting both from the same 16 initial experiments and proposing three candidates per round, with repeated noisy replicates. If Bgolearn does not reach the same strength–ductility frontier in materially fewer evaluations, or does not beat random search at all, the headline reduction claim is falsified.

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

If this is right

  • If the 40–60% benchmark figure transfers, a materials team with a fixed experimental budget can explore roughly twice as wide a design space, or cut the time-to-discovery by about half.
  • Multi-objective optimization is built in, so strength-versus-ductility and similar trade-offs can be managed directly rather than by scalarizing targets.
  • The bootstrap uncertainty estimate extends BO beyond Gaussian processes, allowing large or mixed-variable datasets—common in composition and processing spaces—to drive the search.
  • The graphical interface lets researchers with no programming background run BO and auto-generates equivalent code, changing who is able to use active learning in the lab.
  • The three case studies identify concrete candidate materials and processing schedules that were previously unexplored.

Where Pith is reading between the lines

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

  • Editorial inference: the headline 40–60% savings is measured on noiseless analytic functions; the same advantage on real materials experiments, where noise and discrete variables are unavoidable, is plausible but unproven by the data shown.
  • Editorial inference: each real-world case study validates only a small number of recommended candidates, and there is no random-search control run in the same materials spaces; replication with more candidates and a baseline would sharpen the evidence.
  • Editorial inference: a direct extension would be a closed-loop campaign that alternates Bgolearn recommendations with noisy replicate experiments and records the number of evaluations needed to reach a target property, testing whether the noiseless benchmark ratio survives.
  • Editorial inference: because the framework is modular and open source, its acquisition-function defaults could be stress-tested against alternative policies (for example, pure random exploration or batch random sampling) on the same benchmarks to separate framework gains from BO's intrinsic gains.

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

3 major / 5 minor

Summary. The paper presents Bgolearn, an open-source Python framework for single- and multi-objective Bayesian optimization tailored to materials discovery. It describes the modular architecture (data, surrogate, and acquisition layers), multiple acquisition functions and surrogate models, bootstrap-based uncertainty quantification, and a GUI ('BgoFace'). The central quantitative claim is that benchmark experiments show a 40–60% reduction in the number of required experiments relative to random search, grid search, and genetic algorithms. The paper also reports three application case studies: TPMS elastic-modulus optimization, ultra-hard high-entropy alloy discovery, and medium-Mn steel heat-treatment design, the latter two with new experimental validation. Extensive appendix material provides benchmark definitions, data tables, and implementation details.

Significance. If the efficiency claim is properly substantiated, Bgolearn would be a practically valuable contribution to materials informatics: it lowers the programming barrier, offers multi-objective acquisition functions, and is backed by real experimental demonstrations (HEA nanoindentation and medium-Mn steel tensile data) rather than synthetic examples alone. The open-source release, large user base, and detailed appendix tables also support reproducibility. However, the headline benchmark claim is not supported by the reported baseline comparisons, and the multi-objective results do not measure experimental-effort reduction. The paper's significance therefore hinges on fixing or reframing this claim.

major comments (3)
  1. [Abstract/Conclusion vs. 'Comparison with Baselines' and Table 1] The statement that Bgolearn 'reduces the number of required experiments by 40–60% compared with random search, grid search, and genetic algorithms' is not derivable from Table 1. The table contains only Random Search, Latin Hypercube Sampling, NSGA-II, and Bgolearn variants; no grid search or single-objective genetic algorithm appears. The text itself says comparisons were made 'with random search and Latin hypercube sampling (LHS), and, for multi-objective problems, with NSGA-II.' Moreover, the reported ratios are not uniformly 40–60%: Hartmann-6D Bgolearn-GP achieves ~79% reduction vs random search and ~69% vs LHS; Ackley-5D shows ~69% and ~54%. Please add the missing baselines or revise the abstract/conclusion to state exactly what was compared.
  2. [Table 1, multi-objective rows (ZDT1, DTLZ2)] The multi-objective rows report normalized hypervolume after 50 fixed iterations. Hypervolume at a fixed budget measures solution quality, not the number of experiments needed to reach a target. Thus the paper's central 'experimental effort reduction' claim cannot be extended to multi-objective problems on the basis of these data. Report an effort-to-target metric (e.g., iterations to reach a fixed hypervolume fraction) for all methods, or explicitly restrict the effort-reduction claim to the single-objective benchmarks.
  3. [Real-World Materials Discovery, TPMS subsection] The TPMS case study improves the best elastic modulus from 8,560 MPa (in 50 initial samples) to 8,945 MPa after four additional evaluations, a ~4.5% gain. Without a control—e.g., random search, LHS, or a genetic algorithm given the same 4-evaluation budget on the same surrogate—this example demonstrates only that Bgolearn found a better point, not that it reduces experimental effort. The HEA and medium-Mn steel examples likewise lack a quantitative baseline comparison, although the experimental validations themselves are valuable.
minor comments (5)
  1. [Introduction, references] The sentence 'as evidenced by a series of publications5, 11–15, which are not listed here individually' is contradictory because references 11–15 are listed. Please rephrase.
  2. [Results, Functional materials] There is a typo: 'combining Gaussian process surrogates with with EI et al. acquisition strategies' should read 'with EI and other acquisition strategies.'
  3. [Results, HEA subsection] 'exceeding the upper bound of the best performance reported in literates' should be 'literature.' Also, the recommended Al46.47Co9.16Cr23.47Cu7.22Fe8.10Ni5.58 has Al slightly above the maximum Al in the training data (46.2 at.%); please clarify whether this is intended and how extrapolation was assessed.
  4. [Comparison with Baselines / Figure 3] The benchmark description says 'three new candidate points per iteration' for all methods, but the single-objective results are reported as iterations to 90% optimality. It would help to state explicitly whether each iteration means one batch of three evaluations or a single evaluation, and how batch size affects the iteration counts.
  5. [BgoFace user study] The user study (15 researchers, 5 without programming experience) is reported without a protocol or statistical detail. If this is kept, please provide the task, metrics, and variability; otherwise, mark it as anecdotal.

Circularity Check

0 steps flagged

No significant circularity: the central benchmark results are measured on independent test functions and the case-study claims are supported by new validation experiments.

full rationale

The paper's central quantitative claim (40–60% reduction in required experiments) is grounded in benchmark runs on standard external functions (Hartmann-6D, Ackley-5D, ZDT1, DTLZ2) with iteration counts and hypervolumes measured from the runs, not fitted from the claim itself. The case-study findings (TPMS elastic modulus, HEA hardness, medium-Mn steel strength/ductility) are validated by fresh simulations or experiments and are not constructed from the benchmark results. The self-citations to prior Bgolearn applications (refs. 18, 25–31) are used as supporting evidence of practical impact, but they are not the basis of the algorithmic derivation and are independently falsifiable empirical studies. There is no equation-level reduction of a prediction to an input, no fitted parameter renamed as a prediction, and no uniqueness theorem imported from the authors' prior work. The reviewer-flagged concerns—that grid search and single-objective genetic algorithms are named but not benchmarked, that the benchmarks are noiseless analytic functions while real materials are noisy, and that the '40–60%' figure does not exactly match Table 1—are validity and reporting issues, not circularity. Accordingly, no circular steps are identified.

Axiom & Free-Parameter Ledger

9 free parameters · 8 axioms · 0 invented entities

The claims rest less on new theory than on the correctness of the software's surrogate/UQ defaults and on the transferability of noiseless synthetic benchmarks to noisy experimental materials problems. The free parameters are mostly algorithmic defaults or design-space choices; the axioms are domain assumptions that make the case studies stand for real discovery.

free parameters (9)
  • GP kernel hyperparameters (σ_f², ℓ_d, σ_noise²) = MLE per run; not reported
    Optimized by marginal likelihood (Eq. 8) in every GP surrogate; performance of benchmarks and case studies depends on these fits.
  • UCB exploration coefficient β = default β=1.0 (or β_n = sqrt(2 log(D n² π²/(3δ))), δ=0.1 in Eq. 11)
    Controls exploration-exploitation; chosen by default rather than optimized; affects iteration counts and recommendations.
  • Bootstrap replicates B = 8 default; examples use 5, 8, 10
    Arbitrary choice for non-GP uncertainty quantification in 'Bootstrap Uncertainty Quantification'; changes acquisition uncertainty estimates.
  • qNEHVI noise scaling factor α = 0.1
    Equation (32): σ̂_obs = α·mean(std(y_i)); a heuristic that directly sets the assumed observation noise in qNEHVI.
  • Benchmark initial sample size and batch size = 20 initial LHS points; 3 candidates per iteration
    Fixed experimental design in benchmarks; the 40–60% claim is specific to this budget and would change with other budgets.
  • 90% optimality stopping threshold = 90%
    Defines 'iterations to 90% optimality' in Table 1; different thresholds would change the reported reduction percentages.
  • TPMS initial sample size and candidate grid = 50 initial samples; 5,000 candidates; α_i∈[0,1], t_i∈[-0.5,0.5]
    Case-study design choices; the reported result is a 4.5% improvement over the best initial sample.
  • HEA composition step and variable bounds = 0.01 at.% step; bounds from training data
    Discretizes the Al–Co–Cr–Cu–Fe–Ni design space; affects the recommended Al46.47 composition.
  • Medium-Mn parameter grid = AustTemp 700–880 °C (1 °C), AnnTemp 600–750 °C (1 °C), AnnTime 30–120 min (1 min)
    Search-space choices for the steel case; define what schedules were optimized over and thus which recommendations are possible.
axioms (8)
  • domain assumption The analytic benchmark functions are representative of real materials optimization problems.
    Appendix C uses Hartmann-6D, Ackley-5D, ZDT1, and DTLZ2 to support the experimental-effort reduction claim; real materials evaluations are noisy, expensive, and often categorical.
  • domain assumption Bootstrap ensemble variance (Eq. 19) is a calibrated estimate of predictive uncertainty for acquisition functions.
    The Methods assert this; no calibration check is provided, yet RF/GB/MO-UCB case studies rely on it.
  • domain assumption The TPMS numerical framework (ref. [38]) correctly computes elastic modulus for the level-set geometries.
    The TPMS case labels 50 initial and 4 validation configurations with this framework; no experimental validation is presented.
  • domain assumption The 155 literature hardness values for Al–Co–Cr–Cu–Fe–Ni are accurate and all as-cast (Table 4).
    The HEA discovery trains on these values; several rows contain zero at.% for multiple elements yet are treated as HEA as-cast data.
  • domain assumption For fixed Fe–0.3C–8Mn–2Al, the three heat-treatment variables AustTemp, AnnTemp, AnnTime fully determine yield strength and elongation.
    The medium-Mn case uses only these three parameters; cooling rate, prior processing, and atmosphere are not modeled.
  • standard math GP hyperparameter optimization by marginal likelihood avoids severe overfitting on small datasets.
    Standard GP assumption; with 16–155 training points, kernel fits could overfit in the HEA and steel cases.
  • domain assumption Benchmark evaluations are noiseless, so the measured iteration counts transfer to noisy experimental settings.
    The benchmarks are deterministic functions; the paper's 'experimental effort' language implies real noisy experiments.
  • domain assumption Nanoindentation hardness conversion to Vickers using Poisson's ratio 0.33 is valid for the new HEA.
    Appendix C.4 uses this conversion to report 1028.7±47.6 HV; if wrong, the headline HEA hardness shifts.

pith-pipeline@v1.3.0-alltime-deepseek · 24636 in / 18825 out tokens · 196726 ms · 2026-08-03T11:18:09.668560+00:00 · methodology

0 comments
read the original abstract

Efficient exploration of vast compositional and processing spaces remains a major challenge in accelerated materials discovery. Bayesian optimization (BO) provides a principled approach to identify optimal materials with minimal experimentation, but its adoption has been limited by implementation complexity and a lack of domain-specific tools. Here, we present Bgolearn, a versatile Python framework that brings BO to materials research through intuitive interfaces, robust algorithms, and materials-focused workflows. Bgolearn supports single- and multi-objective optimization, multiple acquisition strategies, diverse surrogate models, and uncertainty quantification, enabling effective navigation of complex design spaces. Benchmark studies show that Bgolearn reduces experimental effort by 40-60\% compared with random search, grid search, and genetic algorithms, while achieving comparable or superior solution quality. Its effectiveness is demonstrated across case studies, including the discovery of maximum-elastic-modulus triply periodic minimal surface structures, ultra-high-hardness high-entropy alloys, and high-strength, high-ductility medium-Mn steels, and is further supported by numerous publications. With a modular architecture that integrates seamlessly into existing materials workflows and a graphical interface (BgoFace) that removes programming barriers, Bgolearn establishes a practical, reliable platform for Bayesian optimization in materials science. The software is openly available at https://github.com/Bin-Cao/Bgolearn.

Figures

Figures reproduced from arXiv: 2601.06820 by Bin Cao, Dezhen Xue, Jian Hui, Jiaxuan Ma, Jiayu Wang, Jie Xiong, Jun Wang, Li Liu, Longhan Zhang, Mengwei He, Tong-Yi Zhang, Turab Lookman, Yirui Hu, Yuan Tian.

Figure 1
Figure 1. Figure 1: The components and workflow of Bgolearn for materials discovery. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the Bgolearn software architecture, data flow, and the integrated BgoFace [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: a, Optimization efficiency comparison across methods, including Bgolearn, random search, Latin hypercube sampling, and NSGA-II (only applicable to multi-target optimization), evaluated on four benchmark functions. b, Optimization traces of each method, where the simple regret represents the difference between the best observed objective value and the global optimum. Real-World Materials Discovery Discovery… view at source ↗
Figure 4
Figure 4. Figure 4: a, Evolution of the elastic modulus across optimization iterations for TPMS structures. The red dotted line indicates the best configuration with the highest elastic modulus identified in the initial training set. b, Experimentally measured yield strength (MPa) and total elongation (%) of medium-Mn steels. The black line denotes the experimental Pareto front, while the red star markers indicate the steels … view at source ↗
Figure 5
Figure 5. Figure 5: Depth–load curves measured from three independent nanoindentation experiments on [PITH_FULL_IMAGE:figures/full_fig_p033_5.png] view at source ↗

discussion (0)

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

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. SVGym (SciVerseGym): An Environment for Reinforcement Learning and Bayesian Optimization in Crystal Discovery

    cs.AI 2026-06 conditional novelty 6.0

    SciVerseGym is a new open Gymnasium environment that frames sequential crystal discovery as an MDP with local/global actions, configurable evaluators, and support for RL, Bayesian optimization, and related workflows.

  2. Efficient and Principled Scientific Discovery through Bayesian Optimization: A Tutorial

    cs.LG 2026-04 accept novelty 2.0

    Bayesian optimization automates the scientific discovery cycle by modeling observations with surrogate models and using acquisition functions to select experiments that balance known information with new exploration.

Reference graph

Works this paper leans on

56 extracted references · 1 linked inside Pith · cited by 2 Pith papers

  1. [1]

    Springer, 2016

    Turab Lookman, Francis J Alexander, and Krishna Rajan.Information science for materials discovery and design, volume 1. Springer, 2016

    2016

  2. [2]

    Active learn- ing in materials science with emphasis on adaptive sampling using uncertainties for targeted design.npj Computational Materials, 5(1):21, 2019

    Turab Lookman, Prasanna V Balachandran, Dezhen Xue, and Ruihao Yuan. Active learn- ing in materials science with emphasis on adaptive sampling using uncertainties for targeted design.npj Computational Materials, 5(1):21, 2019

    2019

  3. [3]

    Julia Ling, Maxwell Hutchinson, Erin Antono, Sean Paradiso, and Bryce Meredig. High- dimensional materials and process optimization using data-driven experimental design with well-calibrated uncertainty estimates.Integrating Materials and Manufacturing Innovation, 6(3):207–217, 2017

    2017

  4. [4]

    Bayesian optimization algorithms for accelerator physics.Physical review accelerators and beams, 27(8):084801, 2024

    Ryan Roussel, Auralee L Edelen, Tobias Boltz, Dylan Kennedy, Zhe Zhang, Fuhao Ji, Xi- aobiao Huang, Daniel Ratner, Andrea Santamaria Garcia, Chenran Xu, et al. Bayesian optimization algorithms for accelerator physics.Physical review accelerators and beams, 27(8):084801, 2024

    2024

  5. [5]

    Sequential closed-loop bayesian optimization as a guide for organic molecular metallophotocatalyst formulation discovery.Nature Chem- istry, 16(8):1286–1294, 2024

    Xiaobo Li, Yu Che, Linjiang Chen, Tao Liu, Kewei Wang, Lunjie Liu, Haofan Yang, Ed- ward O Pyzer-Knapp, and Andrew I Cooper. Sequential closed-loop bayesian optimization as a guide for organic molecular metallophotocatalyst formulation discovery.Nature Chem- istry, 16(8):1286–1294, 2024

    2024

  6. [6]

    Best practices for multi- fidelity bayesian optimization in materials and molecular research.Nature Computational Science, pages 1–10, 2025

    V ´ıctor Sabanza-Gil, Riccardo Barbano, Daniel Pacheco Guti ´errez, Jeremy S Luterbacher, Jos´e Miguel Hern´andez-Lobato, Philippe Schwaller, and Lo¨ıc Roch. Best practices for multi- fidelity bayesian optimization in materials and molecular research.Nature Computational Science, pages 1–10, 2025

    2025

  7. [7]

    Benchmarking the perfor- mance of bayesian optimization across multiple experimental materials science domains.npj Computational Materials, 7(1):188, 2021

    Qiaohao Liang, Aldair E Gongora, Zekun Ren, Armi Tiihonen, Zhe Liu, Shijing Sun, James R Deneault, Daniil Bash, Flore Mekki-Berrada, Saif A Khan, et al. Benchmarking the perfor- mance of bayesian optimization across multiple experimental materials science domains.npj Computational Materials, 7(1):188, 2021

    2021

  8. [8]

    Active learning-guided accelerated discovery of ultra-efficient high-entropy thermoelectrics.Advanced Materials, page e15054, 2025

    Hanhwi Jang, Wooseok Lee, Hwa-Jung Kim, Sohyang Cha, Hosun Shin, Won Bo Lee, Min- Wook Oh, Yeon Sik Jung, and YongJoo Kim. Active learning-guided accelerated discovery of ultra-efficient high-entropy thermoelectrics.Advanced Materials, page e15054, 2025

    2025

  9. [9]

    Materials design with target-oriented bayesian optimization.npj Computational Materials, 11(1):209, 2025

    Yuan Tian, Tongtong Li, Jianbo Pang, Yumei Zhou, Dezhen Xue, Xiangdong Ding, and Turab Lookman. Materials design with target-oriented bayesian optimization.npj Computational Materials, 11(1):209, 2025

    2025

  10. [10]

    A survey and benchmark of high-dimensional bayesian optimization of discrete sequences.Advances in Neural Information Processing Systems, 37:140478–140508, 2024

    Miguel Gonz ´alez-Duque, Richard Michael, Simon Bartels, Yevgen Zainchkovskyy, Søren Hauberg, and Wouter Boomsma. A survey and benchmark of high-dimensional bayesian optimization of discrete sequences.Advances in Neural Information Processing Systems, 37:140478–140508, 2024. 18

    2024

  11. [11]

    Bias free multiobjective active learning for materials design and discovery.Nature communications, 12(1):2312, 2021

    Kevin Maik Jablonka, Giriprasad Melpatti Jothiappan, Shefang Wang, Berend Smit, and Brian Yoo. Bias free multiobjective active learning for materials design and discovery.Nature communications, 12(1):2312, 2021

    2021

  12. [12]

    Artificial-intelligence-guided design of ordered gas diffusion layers for high-performing fuel cells via bayesian machine learning.Nature Communications, 16(1):6528, 2025

    Jing Sun, Pengzhu Lin, Lin Zeng, Zixiao Guo, Yuting Jiang, Cailin Xiao, Qinping Jian, Jiayou Ren, Lyuming Pan, Xiaosa Xu, et al. Artificial-intelligence-guided design of ordered gas diffusion layers for high-performing fuel cells via bayesian machine learning.Nature Communications, 16(1):6528, 2025

    2025

  13. [13]

    Bayesian reaction optimization as a tool for chemical synthesis.Nature, 590(7844):89–96, 2021

    Benjamin J Shields, Jason Stevens, Jun Li, Marvin Parasram, Farhan Damani, Jesus I Mar- tinez Alvarado, Jacob M Janey, Ryan P Adams, and Abigail G Doyle. Bayesian reaction optimization as a tool for chemical synthesis.Nature, 590(7844):89–96, 2021

    2021

  14. [14]

    Increasing certainty in systems biology models using bayesian multimodel inference.Nature Communi- cations, 16(1):7416, 2025

    Nathaniel Linden-Santangeli, Jin Zhang, Boris Kramer, and Padmini Rangamani. Increasing certainty in systems biology models using bayesian multimodel inference.Nature Communi- cations, 16(1):7416, 2025

    2025

  15. [15]

    Mlmd: a programming-free ai platform to predict and design materials.npj Computational Materials, 10(1):59, 2024

    Jiaxuan Ma, Bin Cao, Shuya Dong, Yuan Tian, Menghuan Wang, Jie Xiong, and Sheng Sun. Mlmd: a programming-free ai platform to predict and design materials.npj Computational Materials, 10(1):59, 2024

    2024

  16. [16]

    Efficient hyperparameter tuning for predicting student performance with bayesian optimization.Multimedia tools and applications, 83(17):52711–52735, 2024

    Saleh Albahli. Efficient hyperparameter tuning for predicting student performance with bayesian optimization.Multimedia tools and applications, 83(17):52711–52735, 2024

    2024

  17. [17]

    A bayesian active learning platform for scalable combination drug screens.Nature Communications, 16(1):156, 2025

    Christopher Tosh, Mauricio Tec, Jessica B White, Jeffrey F Quinn, Glorymar Ibanez Sanchez, Paul Calder, Andrew L Kung, Filemon S Dela Cruz, and Wesley Tansey. A bayesian active learning platform for scalable combination drug screens.Nature Communications, 16(1):156, 2025

    2025

  18. [18]

    Active learning accelerates the discovery of high strength and high ductility lead-free solder alloys.Materials & Design, 241:112921, 2024

    Bin Cao, Tianhao Su, Shuting Yu, Tianyuan Li, Taolue Zhang, Jincang Zhang, Ziqiang Dong, and Tong-Yi Zhang. Active learning accelerates the discovery of high strength and high ductility lead-free solder alloys.Materials & Design, 241:112921, 2024

    2024

  19. [19]

    Botorch: A framework for efficient monte-carlo bayesian opti- mization.Advances in neural information processing systems, 33:21524–21538, 2020

    Maximilian Balandat, Brian Karrer, Daniel Jiang, Samuel Daulton, Ben Letham, Andrew G Wilson, and Eytan Bakshy. Botorch: A framework for efficient monte-carlo bayesian opti- mization.Advances in neural information processing systems, 33:21524–21538, 2020

    2020

  20. [20]

    Ae: A domain-agnostic platform for adaptive experimentation

    Eytan Bakshy, Lili Dworkin, Brian Karrer, Konstantin Kashin, Benjamin Letham, Ashwin Murthy, and Shaun Singh. Ae: A domain-agnostic platform for adaptive experimentation. In Conference on neural information processing systems, pages 1–8, 2018

    2018

  21. [21]

    GPyOpt: A bayesian optimization framework in python.http: //github.com/SheffieldML/GPyOpt, 2016

    The GPyOpt authors. GPyOpt: A bayesian optimization framework in python.http: //github.com/SheffieldML/GPyOpt, 2016

    2016

  22. [22]

    A tutorial on multiobjective optimization: fun- damentals and evolutionary methods.Natural computing, 17(3):585–609, 2018

    Michael TM Emmerich and Andr ´e H Deutz. A tutorial on multiobjective optimization: fun- damentals and evolutionary methods.Natural computing, 17(3):585–609, 2018. 19

    2018

  23. [23]

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

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

    2021

  24. [24]

    Accessed: 2025-10-30

    bgolearn download statistics (versions 2.4.0, 2.3.9, 2.3.8) — pepy.tech.https: //pepy.tech/projects/bgolearn?timeRange=threeMonths&category= version&includeCIDownloads=true&granularity=daily&viewType= line&versions=2.4.0%2C2.3.9%2C2.3.8, 2025. Accessed: 2025-10-30

    2025

  25. [25]

    Tianliang Li, Lifei Chen, Bin Cao, Siyuan Liu, Lixing Lin, Zeyu Li, Yingying Chen, Zhen- zhen Li, Tong-yi Zhang, and Lingyan Feng. Optimize the quantum yield of g-quartet-based circularly polarized luminescence materials via active learning strategy-bgoface.Materials Genome Engineering Advances, 3(3):e70031, 2025

    2025

  26. [26]

    Machine learning-engineered nanozyme system for synergistic anti-tumor ferroptosis/apoptosis therapy.Small, 21(5):2408750, 2025

    Tianliang Li, Bin Cao, Tianhao Su, Lixing Lin, Dong Wang, Xinting Liu, Haoyu Wan, Haiwei Ji, Zixuan He, Yingying Chen, et al. Machine learning-engineered nanozyme system for synergistic anti-tumor ferroptosis/apoptosis therapy.Small, 21(5):2408750, 2025

    2025

  27. [27]

    Accelerated design of age-hardened mg-ca-zn alloys with enhanced mechanical properties via machine learning.Computational Materials Science, 249:113665, 2025

    Chenhui Zhang, Yuhui Zhang, Benpeng Ren, Yurong Wu, Yanling Hu, Yanfu Chai, Longshan Xu, and Qinghang Wang. Accelerated design of age-hardened mg-ca-zn alloys with enhanced mechanical properties via machine learning.Computational Materials Science, 249:113665, 2025

    2025

  28. [28]

    Bayesian active learning for accelerated design of broadband polarization- insensitive metasurfaces.Intelligent Computing, 4:0135, 2025

    Jiwei Liu, Gangjie Lian, Wenbin You, Ruixuan Zhang, Yifeng Cheng, Chang Zhang, and Renchao Che. Bayesian active learning for accelerated design of broadband polarization- insensitive metasurfaces.Intelligent Computing, 4:0135, 2025

    2025

  29. [29]

    Active learning-based research of foaming agent for epb shield soil conditioning in gravel stratum.Measurement, 239:115509, 2025

    Chiyu Wang, Wen Zhao, Qian Bai, and Xin Wang. Active learning-based research of foaming agent for epb shield soil conditioning in gravel stratum.Measurement, 239:115509, 2025

    2025

  30. [30]

    Spatial-adaptive active learning identifies ultra-durable and highly active catalysts for acidic oxygen evolution reaction.Science Bulletin, 2025

    Bin Cao, Yin Qin, Yan Luo, Zhehan Ying, Zilin Yan, Lu-Tao Weng, Kaikai Li, and Tong-Yi Zhang. Spatial-adaptive active learning identifies ultra-durable and highly active catalysts for acidic oxygen evolution reaction.Science Bulletin, 2025

    2025

  31. [31]

    Self-driving laboratory for accelerated on-surface synthesis under ultrahigh vacuum.Nano Letters, 25(30):11609–11617, 2025

    Yizhang Li, Qi Huang, Tairan Yang, Zhiwen Zhu, Shaoxuan Yuan, Quan Yang, Xinyi Zhang, and Qiang Sun. Self-driving laboratory for accelerated on-surface synthesis under ultrahigh vacuum.Nano Letters, 25(30):11609–11617, 2025

    2025

  32. [32]

    Scikit-learn: Machine learning in python.the Journal of machine Learning research, 12:2825–2830, 2011

    Fabian Pedregosa, Ga ¨el Varoquaux, Alexandre Gramfort, Vincent Michel, Bertrand Thirion, Olivier Grisel, Mathieu Blondel, Peter Prettenhofer, Ron Weiss, Vincent Dubourg, et al. Scikit-learn: Machine learning in python.the Journal of machine Learning research, 12:2825–2830, 2011

    2011

  33. [33]

    The global optimization problem: an introduction.Towards Global Optimiation 2, pages 1–15, 1978

    Laurence Charles Ward Dixon. The global optimization problem: an introduction.Towards Global Optimiation 2, pages 1–15, 1978

    1978

  34. [34]

    Springer science & business media, 2012

    David Ackley.A connectionist machine for genetic hillclimbing, volume 28. Springer science & business media, 2012. 20

    2012

  35. [35]

    Comparison of multiobjective evolution- ary algorithms: Empirical results.Evolutionary computation, 8(2):173–195, 2000

    Eckart Zitzler, Kalyanmoy Deb, and Lothar Thiele. Comparison of multiobjective evolution- ary algorithms: Empirical results.Evolutionary computation, 8(2):173–195, 2000

    2000

  36. [36]

    Scalable test problems for evolutionary multiobjective optimization

    Kalyanmoy Deb, Lothar Thiele, Marco Laumanns, and Eckart Zitzler. Scalable test problems for evolutionary multiobjective optimization. InEvolutionary multiobjective optimization: theoretical advances and applications, pages 105–145. Springer, 2005

    2005

  37. [37]

    A fast elitist non- dominated sorting genetic algorithm for multi-objective optimization: Nsga-ii

    Kalyanmoy Deb, Samir Agrawal, Amrit Pratap, and Tanaka Meyarivan. A fast elitist non- dominated sorting genetic algorithm for multi-objective optimization: Nsga-ii. InInterna- tional conference on parallel problem solving from nature, pages 849–858. Springer, 2000

    2000

  38. [38]

    Jiaxuan Ma, Sheng Sun, Bin Cao, Bo Qian, Jie Xiong, and Qingcheng Yang. Constraint- aware active learning with conditional generative modeling for bi-objective design of cellular structures based on triply periodic minimal surfaces.Available at SSRN 5758813

  39. [39]

    How atomic bonding plays the hardness behavior in the al–co–cr–cu–fe–ni high entropy family.Small Science, 4(2):2300225, 2024

    Andrea Fantin, Giovanni O Lepore, Michael Widom, Sergey Kasatikov, and Anna M Man- zoni. How atomic bonding plays the hardness behavior in the al–co–cr–cu–fe–ni high entropy family.Small Science, 4(2):2300225, 2024

    2024

  40. [40]

    Exploring high entropy alloys: a review on thermodynamic design and computational modeling strategies for advanced materials ap- plications.Heliyon, 10(22), 2024

    Peter Ifeolu Odetola, Bukola Joseph Babalola, Ayodeji Ebenezer Afolabi, Ufoma Silas Anamu, Emmanuel Olorundaisi, Mutombo Christian Umba, Thabang Phahlane, Olu- soji Oluremi Ayodele, and Peter Apata Olubambi. Exploring high entropy alloys: a review on thermodynamic design and computational modeling strategies for advanced materials ap- plications.Heliyon, ...

    2024

  41. [41]

    Review on wear resistance of laser cladding high- entropy alloy coatings.Journal of Materials Research and Technology, 28:911–934, 2024

    Dingding Xiang, Yusheng Liu, Tianbiao Yu, Di Wang, Xiaoxin Leng, Kaiming Wang, Lin Liu, Jie Pan, Sun Yao, and Zibin Chen. Review on wear resistance of laser cladding high- entropy alloy coatings.Journal of Materials Research and Technology, 28:911–934, 2024

    2024

  42. [42]

    Hot-rolled al-added medium mn steel (fe-8mn-2.85 al-1si-0.2 c): Microstructural evolution and tensile behavior.Materialia, 29:101790, 2023

    Mukesh Kumar Yadav, Deepak Kumar, Navanit Kumar, and Tapas Kumar Bandyopadhyay. Hot-rolled al-added medium mn steel (fe-8mn-2.85 al-1si-0.2 c): Microstructural evolution and tensile behavior.Materialia, 29:101790, 2023

    2023

  43. [43]

    Stabilizing austenite via intercritical mn partitioning in a medium mn steel.Scripta Materialia, 225:115162, 2023

    Baojia Hu, Qinyuan Zheng, Yi Lu, Chunni Jia, Tian Liang, Chengwu Zheng, and Dianzhong Li. Stabilizing austenite via intercritical mn partitioning in a medium mn steel.Scripta Materialia, 225:115162, 2023

    2023

  44. [44]

    High dislocation density–induced large ductility in deformed and partitioned steels.Science, 357(6355):1029– 1032, 2017

    BB He, Bo Hu, HW Yen, GJ Cheng, ZK Wang, HW Luo, and MX Huang. High dislocation density–induced large ductility in deformed and partitioned steels.Science, 357(6355):1029– 1032, 2017

    2017

  45. [45]

    Target- driven design of high strength yet corrosion resistant medium mn steel via interpretable machine-learning.Materials & Design, page 115217, 2025

    Jiayu Wang, Yao Lu, Xiaoya Wang, Siyan Liang, Jie Xiong, Liang Zhen, and Li Liu. Target- driven design of high strength yet corrosion resistant medium mn steel via interpretable machine-learning.Materials & Design, page 115217, 2025

    2025

  46. [46]

    Materials informatics: Emergence to au- tonomous discovery in the age of ai.arXiv preprint arXiv:2601.00742, 2026

    Turab Lookman, YuJie Liu, and Zhibin Gao. Materials informatics: Emergence to au- tonomous discovery in the age of ai.arXiv preprint arXiv:2601.00742, 2026. 21

    arXiv 2026

  47. [47]

    Gaussian pro- cess optimization in the bandit setting: No regret and experimental design.arXiv preprint arXiv:0912.3995, 2009

    Niranjan Srinivas, Andreas Krause, Sham M Kakade, and Matthias Seeger. Gaussian pro- cess optimization in the bandit setting: No regret and experimental design.arXiv preprint arXiv:0912.3995, 2009. 22 Appendix A Detailed Algorithm Descriptions The code booklet provides a tutorial on applying Bgolearn in various situations with version control. Please visi...

    Pith/arXiv arXiv 2009

  48. [48]

    Partition objective space into cells based on current Pareto front

  49. [49]

    For each cell, compute probability that new point falls in cell and dominates

  50. [50]

    Form >2, we use Monte Carlo approxima- tion: EHVI(x)≈ 1 S SX s=1 max(0,HV(F ∪ {f(s)(x)})−HV(F))(28) wheref (s)(x)are samples from the GP posterior

    Sum weighted contributions from all cells For 2D problems, exact computation is feasible. Form >2, we use Monte Carlo approxima- tion: EHVI(x)≈ 1 S SX s=1 max(0,HV(F ∪ {f(s)(x)})−HV(F))(28) wheref (s)(x)are samples from the GP posterior. fromM u l t i B g o l e a r nimportbgo 25 # EHVI f o r 3− o b j e c t i v e o p t i m i z a t i o n VS rec , improvemen...

  51. [51]

    Initialize batchX q =∅

  52. [52]

    Fori= 1toq: • Evaluate qNEHVI for all remaining candidates conditioned onX q • Selectx ∗ i = arg maxx qNEHVI(Xq ∪ {x}) • Update batch:X q ←X q ∪ {x∗ i }

  53. [53]

    fromM u l t i B g o l e a r nimportbgo # qNEHVI w i t h s i n g l e p o i n t s e l e c t i o n VS rec , improvements , i d x = bgo

    ReturnX q Automatic Noise Estimation:When observation noise is unknown, it can be estimated from training data: ˆσobs =α· 1 m mX i=1 std(yi)(32) whereα= 0.1is a conservative scaling factor andmis the number of objectives. fromM u l t i B g o l e a r nimportbgo # qNEHVI w i t h s i n g l e p o i n t s e l e c t i o n VS rec , improvements , i d x = bgo . f...

  54. [54]

    Squared Exponential: k(x,x ′) =σ 2 f exp − 1 2 DX d=1 (xd −x ′ d)2 ℓ2 d ! (37)

  55. [55]

    Mat´ern 5/2: k(x,x ′) =σ 2 f 1 + √ 5r ℓ + 5r2 3ℓ2 ! exp − √ 5r ℓ ! (38) wherer=∥x−x ′∥2

  56. [56]

    Mat´ern 3/2: k(x,x ′) =σ 2 f 1 + √ 3r ℓ ! exp − √ 3r ℓ ! (39) B.1.2 Parameter Optimization GP parametersθ={σ 2 f , ℓ1, . . . , ℓD, σ2 n}are optimized by maximizing the marginal log-likelihood: logp(y|X,θ) =− 1 2 yT K−1 y y− 1 2 log|K y| −n 2 log(2π)(40) whereK y =K+σ 2 nIis the covariance matrix with noise. Bgolearn uses L-BFGS-B optimization with multipl...

    arXiv 2000