Pith. sign in

REVIEW 4 major objections 5 minor 63 references

Microstructure-Aware Bayesian Materials Design

T0 review · 4 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read The paper claims that making Bayesian optimization microstructure-aware—feeding microstructural descriptors into the Gaussian process as latent variables compressed by the active subspace method—finds optimal material designs in fewer…

desk verdict Sensible incremental extension of the group's earlier microstructure-aware BO; the convergence gain is real but not yet causally attributed to microstructure content. read the letter →

arxiv 2502.03727 v1 pith:KDB4QWXI submitted 2025-02-06 cond-mat.mtrl-sci

Danial Khatamsaz , Vahid Attari , Raymundo Arroyave This is my paper · ORCID
classification cond-mat.mtrl-sci
keywords microstructure-awaredesignBayesianoptimizationGaussianprocessactivesubspacemethodprocess-structure-property-performancethermoelectricmaterialsthermalconductivitydiscovery
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

Bayesian optimization is a standard way to search for material compositions and processing conditions with as few experiments as possible, but it usually ignores the microstructure that actually mediates between processing and properties. This paper tries to establish that explicitly feeding microstructural descriptors—such as phase fraction, scattering intensity, and entropy—into the optimizer as latent variables makes the search more efficient and accurate. It does so by compressing the descriptors with the active subspace method and conditioning a Gaussian process on both the design variables and the reduced microstructure representation. In a synthetic problem and in a simulated Mg$_2$Sn$_x$Si$_{1-x}$ thermoelectric thermal-conductivity design task, the microstructure-aware optimizer reaches the same optimum as the standard one after fewer iterations. If true, automated materials discovery platforms should treat microstructure characterization as a routine part of the optimization loop rather than as a post-mortem analysis.

What carries the argument

The active subspace method (ASM) carries the argument. At observed points, the framework estimates the covariance matrix of the objective's gradient with respect to the microstructure descriptors and keeps the eigenvectors with the largest eigenvalues; projections onto those eigenvectors define a low-dimensional subspace that captures most of the objective's variability. The Gaussian process is conditioned on the union of the original design variables and this reduced latent representation, and the subspace is dynamically updated as new observations arrive. This turns a high-dimensional, partly unknown microstructure descriptor set into a small set of informative coordinates, which is what lets the optimizer exploit microstructure information without paying the full cost of the extra dimensions.

What would settle it

A head-to-head campaign on a real material where each iteration includes an actual microstructure measurement would settle the claim: if the microstructure-aware optimizer's total cost including characterization time and noise-induced repeat measurements is not lower than a chemistry-only optimizer on the same budget, or if its convergence advantage vanishes when descriptor noise is introduced, the central claim fails.

Watch

Extended reading notes

Core claim

The central claim is that latent-space-aware Bayesian optimization—conditioning the Gaussian process on an active-subspace projection of microstructure descriptors in addition to the controllable chemistry and processing variables—improves both the predictive model and the convergence of the design loop. In both demonstrations, the latent-aware optimizer matches or outperforms the latent-agnostic optimizer after the first few iterations, and all configurations converge to the same optimal region, so the gain is efficiency in reaching a known optimum. The active subspace is recomputed as observations accumulate, and its eigenvalues give activity scores that identify which microstructure features most control the objective; in the thermoelectric case, the radially averaged FFT structure function and Shannon entropy together explain nearly 90% of the total activity score. The paper presents this as a step toward treating microstructure as an explicit, addressable design variable instead of an emergent by-product of processing.

Load-bearing premise

The central assumption is that microstructural descriptors can be obtained exactly and without meaningful cost at every optimization step, because in the demonstrations the descriptors come from the same simulation that produces the property; the paper itself notes that real characterization could add cost, and if measurements are slow, noisy, or expensive, the convergence advantage could shrink or disappear.

Editorial extensions

If this is right

  • Design campaigns that measure microstructure descriptors during optimization should reach target compositions and processing conditions in fewer iterations than campaigns that record only chemistry and property.
  • The activity scores give designers a data-driven ranking of which microstructural features to control, so characterization effort can be focused on the features that actually drive the property.
  • Because the active subspace is updated online, the method can start with no prior knowledge of which microstructure feature matters and discover it during the campaign, unlike the earlier approach that required that knowledge in advance.
  • The efficiency gain in the examples is convergence speed to the same optimum, not a better optimum, so the practical payoff is reduced experimental cost per campaign.
  • For self-driving laboratories, the framework implies that real-time microstructure characterization should be built into the optimization loop because the information it provides accelerates subsequent decisions.

Reading between the lines

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

  • Extension: the paper does not price the cost of measuring microstructure, so a natural next step is to compare total campaign cost and let the optimizer decide adaptively whether the expected improvement justifies paying for a characterization step, using activity scores to choose which descriptors to measure.
  • Because the active subspace is estimated from local gradients of the objective, applying the method to real experimental data with noisy or sparse measurements may require gradient surrogates; it is an open quantitative question how much noise the convergence advantage survives.
  • The same latent-space-aware structure could apply to other intermediate variables beyond microstructure, such as stress state or defect density, whenever a measurable intermediate quantity links inputs to the objective.
  • A prospective test would be to run two parallel campaigns on a real alloy or thermoelectric system with equal budgets, one with and one without microstructure feedback, and compare how fast each reaches a target property.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. This manuscript presents a Bayesian optimization (BO) framework that augments the conventional design-variable inputs to a Gaussian process with a low-dimensional active-subspace projection of microstructural descriptors, treated as latent variables. The method is tested on a synthetic problem with six latent variables and on a phase-field simulation of Mg2SnxSi1-x thermoelectric microstructures, where the effective thermal conductivity is minimized. The authors report that the latent-space-aware BO reaches better objective values after a few iterations than a conventional BO that uses only the design variables, and they use activity scores to identify which microstructural descriptors most influence the objective. The paper concludes that microstructure characterization should be integrated into automated materials discovery platforms.

Significance. The proposed framework is a plausible extension of the authors' earlier work and addresses a real gap: bringing process-structure-property linkages into active learning for materials discovery. The synthetic and phase-field demonstrations are clearly motivated, and the dynamic active-subspace update and activity-score diagnostic are useful ideas. The phase-field simulations with CALPHAD-reinforced elasto-chemical modeling and thermal conductivity computation are physically grounded, and the use of a constant seed number for consistency is good practice. If the convergence gains are robust, the work could inform the design of self-driving laboratories. The paper is also honest about some limitations, including the cost trade-off of microstructure characterization, but the evidence as presented does not yet isolate the cause of the improvement, and an internal inconsistency appears in the results discussion.

major comments (4)
  1. [Section 3 (synthetic and phase-field comparisons; Figs. 4 and 6)] The comparison between latent space-agnostic BO and latent space-aware BO changes two factors at once: the GP receives additional input features (the active-subspace projection of f1–f5) and those features are obtained via a gradient-based active subspace method. There is no control in which the agnostic GP is given the same number of extra inputs that are non-informative (e.g., noise variables or arbitrary functions of x), nor a condition in which the latent-aware GP uses the raw latent descriptors without ASM. The convergence advantage in Figs. 4 and 6 could therefore be an artifact of the added input dimensions or of the particular projection, rather than of the microstructural content. Please add control experiments—these are inexpensive in the synthetic setting—to isolate the source of the improvement.
  2. [Section 2.3 and Section 3] The active subspace method is introduced for a function f(x) and requires gradients ∇x f (Eqs. 4–5). In the application, the active subspace is computed in the latent (microstructure) space, so the required quantity is the gradient of the objective with respect to the descriptors f. The manuscript does not state how these gradients are obtained in the phase-field case, where y=κ is the output of a numerical heat-transfer solve and the descriptors include an Otsu threshold, a radially averaged FFT, and Shannon entropy. Please specify the gradient approximation scheme (e.g., analytic, finite-difference, adjoint, or GP-derivative) and discuss its accuracy and cost.
  3. [Section 3, discussion of Fig. 6(c)] The activity-score discussion for the real case is internally inconsistent. The text states that 'the radially averaged FFT structure function and Shannon entropy are the most influential' and then states 'According to Fig. 6(c), latent variables f2 and f5 have the lowest activity scores.' With the latent variables defined as F = [A_f, c_Mg2Sn, c_Mg2Si, PS_max, S_shannon], f5 is the Shannon entropy, so f5 cannot be both the most and the least influential. Please correct the indices or the summary, and make the figure legend and text consistent.
  4. [Section 3 concluding paragraph and Section 4] The recommendation that microstructure characterization 'should be integral to automated—and eventually autonomous—platforms' goes beyond the evidence. In both demonstrations the descriptors are exact, noiseless, deterministic outputs of the same simulation that generates the objective, and the manuscript itself acknowledges that measuring structural features may impose additional costs and defers that trade-off to future work. To support the strong practical recommendation, the paper should either include a demonstration with noisy or partial descriptors, or temper the conclusion to the idealized setting with cost-free descriptors.
minor comments (5)
  1. [Section 3, Eqs. (20)-(21)] The sentence 'we still use Eqs. 20 and 21 to calculate the objective function but the objective function information is extracted from the remaining 5 latent variables' should be reworded; as written it contradicts Eq. (21), which includes f6. The intended meaning is presumably that f6 is hidden from the model and excluded from the ASM, not that y is computed without f6.
  2. [Section 2.5, Eq. (14)] The strain-displacement relation is written as ε_ij = 1/2(∂u_i/∂r_j − ∂u_j/∂r_i), which is the rotation tensor; the plus sign is required for the infinitesimal strain tensor.
  3. [Section 3, Eq. (23)] The expression for ηMax_local contains a malformed parenthesis and should be typeset carefully.
  4. [Keywords] The keyword line in the abstract contains a stray comma after 'Keywords:'.
  5. [Data Availability] The data availability statement is minimal; consider depositing the phase-field and BO data in a public repository or providing a code repository to support reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation found: the GP/ASM/BO pipeline is self-contained, though the comparisons are uncontrolled for added input dimensions.

full rationale

The derivation chain is: design variables x plus microstructure descriptors F form an augmented input; ASM projects F; the GP is conditioned on x and the projected F; BO proposes new x. The target kappa (or synthetic y) is never used to define the input features, nor is any fitted parameter reported as an independent prediction of the same quantity. The synthetic objective (Eqs. 20-21) is deliberately defined as a function of the latent variables that are then supplied to the latent-aware GP, making the demonstration favorable, but this is not a circular reduction: the GP does not receive y itself, and the agnostic baseline is a meaningful though uncontrolled comparator. The real case-study descriptors are intermediate outputs of the same phase-field/heat-transfer simulation that yields kappa, so the comparison tests algorithmic benefit under idealized descriptor availability; the paper explicitly acknowledges that real characterization may impose extra costs and defers that trade-off to future work. Self-citations [29] and [45] motivate the augmented-space and active-subspace ideas, but the current evidence comes from new simulations, and ASM is also supported by external references [40-42,47,48]; no load-bearing claim rests solely on an unverified self-citation. No step in the claimed derivation reduces to its own input by construction.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

No new physical entities are introduced; the latent variables are known microstructure descriptors. The main added complexity is computational, not ontological.

free parameters (3)
  • information-capturing threshold = 50%, 70%, 90% (synthetic); 50%, 75%, 90%, 95% (materials)
    User-chosen cutoff on normalized active-subspace eigenvalues; determines active subspace dimension and directly controls the amount of latent information fed to the GP.
  • AGB (grain boundary thermal diffusivity coefficient) = not specified in text
    Hand-set simulation parameter in the effective thermal diffusivity model (Section 2.7); it shapes the microstructure-property map being optimized.
  • initial training sample count = 100
    Number of Latin hypercube samples used to initialize BO in both demonstrations; affects the comparison but is not optimized.
assumptions (5)
  • domain assumption Microstructure descriptors can be measured and included at each optimization step without noise or significant cost.
    Both demonstrations supply descriptors from simulation with no observational noise; the paper defers the cost trade-off to future work (Section 3).
  • domain assumption The active-subspace gradient covariance (Eq. 5) is a valid estimator of objective variability from the available samples.
    The paper does not describe how gradients of thermal conductivity with respect to descriptors are computed, yet the whole projection depends on them.
  • domain assumption Phase-field simulations plus Fourier heat conduction faithfully represent the process-structure-property chain for Mg2Sn_x Si_1-x.
    Thermal conductivity is computed on simulated microstructures; no experimental validation is provided.
  • ad hoc to paper The synthetic objective (Eqs. 20-21) is representative of real microstructure-property maps.
    The toy function is constructed to make latent variables causally mediate design-to-property relations; this is a favorable test condition, not a measured system.
  • standard math Gaussian process regression and Bayesian optimization assumptions hold.
    Standard GP/BO background from Rasmussen and Williams, with kernel as in Eq. 3.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Microstructure-Aware Bayesian Materials Design." pith.science (2026). https://pith.science/paper/KDB4QWXI

@misc{pith2026250203727,
  author       = {Pith},
  title        = {Pith review of: Microstructure-Aware Bayesian Materials Design},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KDB4QWXI}},
  note         = {Machine review of arXiv:2502.03727}
}
abstract

In this study, we propose a novel microstructure-sensitive Bayesian optimization (BO) framework designed to enhance the efficiency of materials discovery by explicitly incorporating microstructural information. Traditional materials design approaches often focus exclusively on direct chemistry-process-property relationships, overlooking the critical role of microstructures. To address this limitation, our framework integrates microstructural descriptors as latent variables, enabling the construction of a comprehensive process-structure-property mapping that improves both predictive accuracy and optimization outcomes. By employing the active subspace method for dimensionality reduction, we identify the most influential microstructural features, thereby reducing computational complexity while maintaining high accuracy in the design process. This approach also enhances the probabilistic modeling capabilities of Gaussian processes, accelerating convergence to optimal material configurations with fewer iterations and experimental observations. We demonstrate the efficacy of our framework through synthetic and real-world case studies, including the design of Mg$_2$Sn$_x$Si$_{1-x}$ thermoelectric materials for energy conversion. Our results underscore the critical role of microstructures in linking processing conditions to material properties, highlighting the potential of a microstructure-aware design paradigm to revolutionize materials discovery. Furthermore, this work suggests that since incorporating microstructure awareness improves the efficiency of Bayesian materials discovery, microstructure characterization stages should be integral to automated -- and eventually autonomous -- platforms for materials development.

Figures

Figures reproduced from arXiv: 2502.03727 by the authors.

Figure 1
Figure 1. Throughout history, material innovations have been a driving force behind societal and technological advancements, from the Stone Age to the AI era. As humanity faces modern challenges like climate change and resource scarcity, the need for novel materials and accelerated development approaches is more urgent than ever. AI-driven autonomous labs and robotics will accelerate materials development, creating a ‘Moore’s… view at source ↗
Figure 2
Figure 2. (Left) Autonomous materials decision-making using microstructure-sensitive design methods, and (Right) the main steps of the [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Illustration of the active subspace definition over a 2- dimensional input space. The active subspace determines the sub￾space that represents the largest objective function variability (the most active direction) to build a lower-dimensional projection of the function. In the example above, the active subspace is represented as 1-dimensional subspace that can be traversed by a scalar. fracture. The local mass flux … view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Performance comparison of different optimization scenarios and the exhaustive search averaged over 100 replications of simu￾lations. The latent space-aware BO with different information-capturing thresholds outperforms the latent space-agnostic BO as higher function va…
Figure 5
Figure 5. Figure 5: Thermal conduction across nanostructured microstructure [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: (a) and (b) Optimum thermal conductivity obtained using different optimization scenarios. Microstructure-aware optimization scenarios are run with different variation-capturing thresholds of 50%, 75%, 90%, and 95%. (a) The average thermal conductivity values and (b) th…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

63 extracted references · 60 canonical work pages

  1. [1]

    Bocquet-Appel, When the world’s population took o ff: the springboard of the neolithic demographic transition, Science 333 (6042) (2011) 560–561

    J.-P. Bocquet-Appel, When the world’s population took o ff: the springboard of the neolithic demographic transition, Science 333 (6042) (2011) 560–561

    work page 2011

  2. [2]

    Mokyr, The lever of riches: Technological creativity and eco- nomic progress, Oxford University Press, 1992

    J. Mokyr, The lever of riches: Technological creativity and eco- nomic progress, Oxford University Press, 1992

    work page 1992

  3. [3]

    Rhodes, The making of the atomic bomb, Simon and Schus- ter, 2012

    R. Rhodes, The making of the atomic bomb, Simon and Schus- ter, 2012

    work page 2012

  4. [4]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, I. Polosukhin, Attention is all you need, Ad- vances in neural information processing systems 30 (2017)

    work page 2017

  5. [5]

    Devlin, M.-W

    J. Devlin, M.-W. Chang, K. Lee, K. Toutanova, Bert: Pre- training of deep bidirectional transformers for language under- standing, arXiv preprint arXiv:1810.04805 (2018)

    arXiv 2018

  6. [6]

    Brynjolfsson, A

    E. Brynjolfsson, A. McAfee, The second machine age: Work, progress, and prosperity in a time of brilliant technologies, WW Norton & Company, 2014

    work page 2014

  7. [7]

    M. I. Ho ffert, K. Caldeira, G. Benford, D. R. Criswell, C. Green, H. Herzog, A. K. Jain, H. S. Kheshgi, K. S. Lackner, J. S. Lewis, et al., Advanced technology paths to global climate stability: energy for a greenhouse planet, science 298 (5595) (2002) 981– 987

    work page 2002

  8. [8]

    A. Jain, Y . Shin, K. A. Persson, Computational predictions of energy materials using density functional theory, Nature Re- views Materials 1 (1) (2016) 1–13

    work page 2016

Show all 63 references
  1. [9]

    K. T. Butler, D. W. Davies, H. Cartwright, O. Isayev, A. Walsh, Machine learning for molecular and materials science, Nature 559 (7715) (2018) 547–555

    2018

  2. [10]

    Himanen, A

    L. Himanen, A. Geurts, A. S. Foster, P. Rinke, Data-driven ma- terials science: status, challenges, and perspectives, Advanced Science 6 (21) (2019) 1900808

    2019

  3. [11]

    D. T. Fullwood, S. R. Niezgoda, B. L. Adams, S. R. Kalidindi, Microstructure sensitive design for performance optimization, Progress in Materials Science 55 (6) (2010) 477–562. 13

    2010

  4. [12]

    M. D. White, A. Tarakanov, P. Withers, C. Race, K. J. Law, Dig- ital fingerprinting of microstructures, Computational Materials Science 218 (2023) 111985

    2023

  5. [13]

    R. C. Reed, The superalloys: fundamentals and applications, Cambridge University Press, 2008

    2008

  6. [14]

    Raabe, C

    D. Raabe, C. C. Tasan, E. A. Olivetti, Strategies for improving the sustainability of structural metals, Nature 575 (7781) (2019) 64–74

    2019

  7. [15]

    Khosravani, A

    A. Khosravani, A. Cecen, S. R. Kalidindi, Development of high throughput assays for establishing process-structure-property linkages in multiphase polycrystalline metals: Application to dual-phase steels, Acta Materialia 123 (2017) 55–69

    2017

  8. [16]

    S. R. Kalidindi, M. De Graef, Materials data science: current status and future outlook, Annual Review of Materials Research 45 (1) (2015) 171–193

    2015

  9. [17]

    G. J. Snyder, E. S. Toberer, Complex thermoelectric materials, Nature materials 7 (2) (2008) 105–114

    2008

  10. [18]

    Zebarjadi, K

    M. Zebarjadi, K. Esfarjani, M. Dresselhaus, Z. Ren, G. Chen, Perspectives on thermoelectrics: from fundamentals to device applications, Energy & Environmental Science 5 (1) (2012) 5147–5162

    2012

  11. [19]

    D. Xue, P. V . Balachandran, J. Hogden, J. Theiler, D. Xue, T. Lookman, Accelerated search for materials with targeted properties by adaptive design, Nature communications 7 (1) (2016) 1–9

    2016

  12. [20]

    Lookman, P

    T. Lookman, P. V . Balachandran, D. Xue, R. Yuan, Active learn- ing in materials science with emphasis on adaptive sampling us- ing uncertainties for targeted design, npj Computational Materi- als 5 (1) (2019) 21

    2019

  13. [21]

    C. S. Ha, D. Yao, Z. Xu, C. Liu, H. Liu, D. Elkins, M. Kile, V . Deshpande, Z. Kong, M. Bauchy, et al., Rapid inverse de- sign of metamaterials based on prescribed mechanical behav- ior through machine learning, Nature Communications 14 (1) (2023) 5765

    2023

  14. [22]

    Arr ´oyave, D

    R. Arr ´oyave, D. L. McDowell, Systems approaches to materials design: past, present, and future, Annual Review of Materials Research 49 (2019) 103–126

    2019

  15. [23]

    Morand, T

    L. Morand, T. Iraki, J. Dornheim, S. Sandfeld, N. Link, D. Helm, Machine learning for structure-guided materials and process design, Materials & Design 248 (2024) 113453

    2024

  16. [24]

    Flores Ituarte, S

    I. Flores Ituarte, S. Panicker, H. P. Nagarajan, E. Coatanea, D. W. Rosen, Optimisation-driven design to explore and exploit the process–structure–property–performance linkages in digi- tal manufacturing, Journal of Intelligent Manufacturing 34 (1) (2023) 219–241

    2023

  17. [25]

    X. Liao, M. Liao, C. Wei, Z. Huang, W. Duan, X. Duan, D. Cai, L. Gremillard, Z. Yang, D. Jia, et al., A process-structure- property model via physics-based /data-driven hybrid methods for freeze-cast porous ceramics in si3n4-si2n2o case system, Acta Materialia 269 (2024) 119819

    2024

  18. [26]

    A. Tran, T. Wildey, Solving stochastic inverse problems for property–structure linkages using data-consistent inversion and machine learning, JOM 73 (1) (2021) 72–89

    2021

  19. [27]

    D.-Y . Wu, T. C. Hufnagel, Efficient searching of processing pa- rameter space to enable inverse microstructural design of mate- rials, Acta Materialia 264 (2024) 119562

    2024

  20. [28]

    Rixner, P.-S

    M. Rixner, P.-S. Koutsourelakis, Self-supervised optimization of random material microstructures in the small-data regime, npj Computational Materials 8 (1) (2022) 46

    2022

  21. [29]

    Molkeri, D

    A. Molkeri, D. Khatamsaz, R. Couperthwaite, J. James, R. Arr ´oyave, D. Allaire, A. Srivastava, On the importance of microstructure information in materials design: Psp vs pp, Acta Materialia 223 (2022) 117471

    2022

  22. [30]

    Aspuru-Guzik, K

    A. Aspuru-Guzik, K. Persson, Materials acceleration platform: Accelerating advanced energy materials discovery by integrat- ing high-throughput methods and artificial intelligence., Mis- sion Innovation (2018)

    2018

  23. [31]

    fourth paradigm

    A. Agrawal, A. Choudhary, Perspective: Materials informatics and big data: Realization of the “fourth paradigm” of science in materials science, Apl Materials 4 (5) (2016)

    2016

  24. [32]

    Raccuglia, K

    P. Raccuglia, K. C. Elbert, P. D. Adler, C. Falk, M. B. Wenny, A. Mollo, M. Zeller, S. A. Friedler, J. Schrier, A. J. Norquist, Machine-learning-assisted materials discovery using failed ex- periments, Nature 533 (7601) (2016) 73–76

    2016

  25. [33]

    Pilania, C

    G. Pilania, C. Wang, X. Jiang, S. Rajasekaran, R. Ram- prasad, Accelerating materials property predictions using ma- chine learning, Scientific reports 3 (1) (2013) 2810

    2013

  26. [34]

    Miracle, D

    D. Miracle, D. Thoma, Autonomous research and develop- ment of structural materials–an introduction and vision, Current Opinion in Solid State and Materials Science 33 (2024) 101188

    2024

  27. [35]

    Y . Liu, T. Zhao, W. Ju, S. Shi, Materials discovery and design using machine learning, Journal of Materiomics 3 (3) (2017) 159–177

    2017

  28. [36]

    C. E. Rasmussen, C. K. I. Williams, Gaussian Processes for Ma- chine Learning (Adaptive Computation and Machine Learning), The MIT Press, 2005

    2005

  29. [37]

    F. S. Costabal, P. Perdikaris, E. Kuhl, D. E. Hurtado, Multi- fidelity classification using gaussian processes: accelerating the prediction of large-scale computational models, Comput. Meth- ods Appl. Mech. Eng. 357 (2019) 112602

    2019

  30. [38]

    Hastie, R

    T. Hastie, R. Tibshirani, J. H. Friedman, J. H. Friedman, The elements of statistical learning: data mining, inference, and pre- diction, V ol. 2, Springer, 2009

    2009

  31. [39]

    D. R. Jones, M. Schonlau, W. J. Welch, E fficient global op- timization of expensive black-box functions, J. Global Optim. 13 (4) (1998) 455–492. doi:10.1023/A:1008306431147

    1998 doi

  32. [40]

    P. G. Constantine, E. Dow, Q. Wang, Active subspace methods in theory and practice: applications to kriging surfaces, SIAM Journal on Scientific Computing 36 (4) (2014) A1500–A1524

    2014

  33. [41]

    Constantine, D

    P. Constantine, D. Gleich, Computing active subspaces with monte carlo, arXiv preprint arXiv:1408.0545 (2014)

    2014 arXiv

  34. [42]

    P. G. Constantine, Active subspaces: Emerging ideas for dimen- sion reduction in parameter studies, SIAM, 2015

    2015

  35. [43]

    Abolhasani, E

    M. Abolhasani, E. Kumacheva, The rise of self-driving labs in chemical and materials sciences, Nature Synthesis 2 (6) (2023) 483–492

    2023

  36. [44]

    Seifrid, R

    M. Seifrid, R. Pollice, A. Aguilar-Granda, Z. Morgan Chan, K. Hotta, C. T. Ser, J. Vestfrid, T. C. Wu, A. Aspuru-Guzik, Autonomous chemical experiments: Challenges and perspec- tives on establishing a self-driving lab, Accounts of Chemical Research 55 (17) (2022) 2454–2466

    2022

  37. [45]

    Khatamsaz, et al., Adaptive active subspace-based e fficient multifidelity materials design, Mater

    D. Khatamsaz, et al., Adaptive active subspace-based e fficient multifidelity materials design, Mater. Des. 209 (2021) 110001

    2021

  38. [46]

    H. Abdi, L. J. Williams, Principal component analysis, Wiley Interdiscip. Rev. Comput. Stat. 2 (4) (2010) 433–459

    2010

  39. [47]

    S. F. Ghoreishi, S. Friedman, D. L. Allaire, Adaptive dimen- sionality reduction for fast sequential optimization with gaus- sian processes, J. Mech. Design 141 (7) (2019)

    2019

  40. [48]

    T. M. Russi, Uncertainty quantification with experimental data and complex system models, Ph.D. thesis, UC Berkeley (2010)

    2010

  41. [49]

    L.-Q. Chen, Y . Zhao, From classical thermodynamics to phase- field method, Progress in Materials Science 124 (2022) 100868

    2022

  42. [50]

    L. Q. Chen, J. Shen, Applications of semi-implicit fourier- spectral method to phase field equations, Computer Physics Communications 108 (2-3) (1998) 147–158

    1998

  43. [51]

    Attari, A

    V . Attari, A. Cruzado, R. Arroyave, Exploration of the mi- crostructure space in tialzrn ultra-hard nanostructured coatings, Acta Materialia 174 (2019) 459–476

    2019

  44. [52]

    Otsu, A threshold selection method from gray-level his- tograms, IEEE transactions on systems, man, and cybernetics 14 9 (1) (1979) 62–66

    N. Otsu, A threshold selection method from gray-level his- tograms, IEEE transactions on systems, man, and cybernetics 14 9 (1) (1979) 62–66

    1979

  45. [53]

    Honarmandi, V

    P. Honarmandi, V . Attari, R. Arroyave, Accelerated materials design using batch bayesian optimization: A case study for solv- ing the inverse problem from materials microstructure to pro- cess specification, Computational Materials Science 210 (2022) 111417

    2022

  46. [54]

    E. S. Toberer, L. L. Baranowski, C. Dames, Advances in thermal conductivity, Annual Review of Materials Research 42 (2012) 179–209

    2012

  47. [55]

    D. G. Cahill, H. E. Fischer, T. Klitsner, E. Swartz, R. Pohl, Ther- mal conductivity of thin films: Measurements and understand- ing, Journal of Vacuum Science & Technology A: Vacuum, Sur- faces, and Films 7 (3) (1989) 1259–1266

    1989

  48. [56]

    X. L ¨u, W. Shen, J. Chu, Size effect on the thermal conductivity of nanowires, Journal of applied physics 91 (3) (2002) 1542– 1552

    2002

  49. [57]

    Vaqueiro, A

    P. Vaqueiro, A. V . Powell, Recent developments in nanostruc- tured materials for high-performance thermoelectrics, Journal of Materials Chemistry 20 (43) (2010) 9577–9584

    2010

  50. [58]

    Shakouri, Recent developments in semiconductor thermo- electric physics and materials, Annual review of materials re- search 41 (2011) 399–431

    A. Shakouri, Recent developments in semiconductor thermo- electric physics and materials, Annual review of materials re- search 41 (2011) 399–431

    2011

  51. [59]

    Z. Bian, A. Shakouri, Beating the maximum cooling limit with graded thermoelectric materials, Applied Physics Letters 89 (21) (2006) 212101

    2006

  52. [60]

    J. He, T. M. Tritt, Advances in thermoelectric materials research: Looking back and moving forward, Science 357 (6358) (2017) eaak9997

    2017

  53. [61]

    D. M. Rowe, Thermoelectrics handbook: macro to nano, CRC press, 2018

    2018

  54. [62]

    G. J. Snyder, T. S. Ursell, Thermoelectric e fficiency and com- patibility, Physical review letters 91 (14) (2003) 148301

    2003

  55. [63]

    Torquato, H

    S. Torquato, H. Haslach Jr, Random heterogeneous materials: microstructure and macroscopic properties, Appl. Mech. Rev. 55 (4) (2002) B62–B63. 15

    2002

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.