pith. sign in

arxiv: 1907.02577 · v1 · pith:GO7RNAVSnew · submitted 2019-07-04 · ⚛️ physics.comp-ph · cond-mat.mtrl-sci· cs.LG· stat.ML

Data-Centric Mixed-Variable Bayesian Optimization For Materials Design

Akshay Iyer , Yichi Zhang , Aditya Prasad , Siyu Tao , Yixing Wang , Linda Schadler , L Catherine Brinson , Wei Chen This is my paper

Pith reviewed 2026-05-25 08:51 UTC · model grok-4.3

classification ⚛️ physics.comp-ph cond-mat.mtrl-scics.LGstat.ML
keywords Bayesian optimizationmaterials designmixed-variable optimizationlatent variable Gaussian processpolymer nanocompositesmulti-objective optimizationsurrogate modeling
0
0 comments X

The pith

A Bayesian optimization framework using latent variable Gaussian processes handles both qualitative and quantitative variables to optimize materials designs with limited data.

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

The paper presents a data-centric Bayesian optimization approach for materials design problems involving mixed qualitative variables such as composition types and quantitative variables such as processing conditions. It centers on the latent variable Gaussian process to embed categories into continuous space, allowing a surrogate model to quantify uncertainty from sparse data drawn from literature, experiments, and simulations. Expected improvement then directs the search through nonlinear spaces to identify optimal designs. The method is shown on a polymer nanocomposite insulation case and extended to locate multi-objective Pareto fronts within tens of iterations.

Core claim

The framework pivots around the Latent Variable Gaussian Process (LVGP), a novel Gaussian Process technique which projects qualitative variables on a continuous latent space for covariance formulation, as the surrogate model to quantify lack of data uncertainty. Expected improvement, an acquisition criterion that balances exploration and exploitation, helps navigate a complex, nonlinear design space to locate the optimum design, as demonstrated in concurrent identification of optimal composition and morphology for insulating polymer nanocomposites and in multi-objective Pareto frontier search.

What carries the argument

Latent Variable Gaussian Process (LVGP), which projects qualitative variables onto a continuous latent space to formulate covariances and serve as the surrogate model in mixed-variable Bayesian optimization.

If this is right

  • Locates optimal composition and morphology for insulating polymer nanocomposites within tens of iterations.
  • Extends directly to multiple objectives to identify the Pareto frontier.
  • Integrates data from literature, experiments, and simulations into a single search process.
  • Balances exploration and exploitation via expected improvement in nonlinear mixed-variable spaces.

Where Pith is reading between the lines

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

  • The same embedding approach could lower experimental costs in other materials systems that combine categorical choices with continuous parameters.
  • Testing the learned latent distances against independent similarity metrics from domain data would strengthen transfer to new material classes.
  • The framework suggests a route to automate knowledge discovery by treating published results as additional observations in the same surrogate model.

Load-bearing premise

The latent-space projection of qualitative variables in LVGP preserves the relevant similarity structure among categories for the target material properties without requiring domain-specific validation of the embedding.

What would settle it

In the insulating polymer nanocomposites case study, the LVGP-based optimizer requires substantially more evaluations than a baseline mixed-variable method to reach designs with equivalent or better insulation performance.

read the original abstract

Materials design can be cast as an optimization problem with the goal of achieving desired properties, by varying material composition, microstructure morphology, and processing conditions. Existence of both qualitative and quantitative material design variables leads to disjointed regions in property space, making the search for optimal design challenging. Limited availability of experimental data and the high cost of simulations magnify the challenge. This situation calls for design methodologies that can extract useful information from existing data and guide the search for optimal designs efficiently. To this end, we present a data-centric, mixed-variable Bayesian Optimization framework that integrates data from literature, experiments, and simulations for knowledge discovery and computational materials design. Our framework pivots around the Latent Variable Gaussian Process (LVGP), a novel Gaussian Process technique which projects qualitative variables on a continuous latent space for covariance formulation, as the surrogate model to quantify "lack of data" uncertainty. Expected improvement, an acquisition criterion that balances exploration and exploitation, helps navigate a complex, nonlinear design space to locate the optimum design. The proposed framework is tested through a case study which seeks to concurrently identify the optimal composition and morphology for insulating polymer nanocomposites. We also present an extension of mixed-variable Bayesian Optimization for multiple objectives to identify the Pareto Frontier within tens of iterations. These findings project Bayesian Optimization as a powerful tool for design of engineered material systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript presents a mixed-variable Bayesian optimization framework for materials design that uses Latent Variable Gaussian Process (LVGP) surrogates to embed qualitative design variables into a continuous latent space for covariance computation. Expected improvement is employed as the acquisition function. The approach is demonstrated on a polymer-nanocomposite case study seeking optimal composition and morphology for insulating properties and is extended to multi-objective optimization to recover the Pareto frontier within tens of iterations.

Significance. If the LVGP embeddings are shown to recover physically relevant category similarities, the framework could offer a practical route for incorporating literature, experimental, and simulation data into efficient search over mixed-variable spaces common in materials problems. The data-centric emphasis and multi-objective extension are positive features, but the absence of quantitative benchmarks against standard encodings leaves the practical advantage unquantified.

major comments (3)
  1. [case study] Case study section: the manuscript reports that the framework was tested on a polymer-nanocomposite optimization problem and an extension to multiple objectives, yet provides no quantitative performance metrics (e.g., regret curves, success rates), error bars, or direct comparisons to baselines such as one-hot encoding or standard GPs with dummy variables. This absence prevents assessment of whether LVGP delivers measurable improvement.
  2. [LVGP and case study] LVGP description and case-study results: the central claim that LVGP enables effective navigation of mixed-variable spaces rests on the assumption that Euclidean distances in the learned latent space recover the relevant similarity structure among qualitative categories (fillers, morphologies) with respect to dielectric constant and breakdown strength. No post-hoc validation, alignment with independent physical knowledge, or ablation against random embeddings is presented to support this.
  3. [LVGP] LVGP implementation: the latent-space dimension is listed among the free parameters, but the manuscript gives no procedure, cross-validation, or sensitivity analysis for its selection, nor does it report the specific dimension(s) used in the case study.
minor comments (2)
  1. [abstract] The abstract and introduction would benefit from explicit statements of the number of iterations or function evaluations required to reach the reported optima.
  2. [methods] Notation for the latent variables and the form of the covariance kernel should be introduced with a short equation block for clarity.

Simulated Author's Rebuttal

3 responses · 0 unresolved

Thank you for the opportunity to respond to the referee's comments. We address each major comment below.

read point-by-point responses

  1. Referee: [case study] Case study section: the manuscript reports that the framework was tested on a polymer-nanocomposite optimization problem and an extension to multiple objectives, yet provides no quantitative performance metrics (e.g., regret curves, success rates), error bars, or direct comparisons to baselines such as one-hot encoding or standard GPs with dummy variables. This absence prevents assessment of whether LVGP delivers measurable improvement.

    Authors: We agree that the current manuscript lacks regret curves, error bars, success rates, and direct baseline comparisons. In the revised version we will add these quantitative metrics, including averaged regret curves with error bars over repeated runs and comparisons to one-hot and dummy-variable encodings. revision: yes

  2. Referee: [LVGP and case study] LVGP description and case-study results: the central claim that LVGP enables effective navigation of mixed-variable spaces rests on the assumption that Euclidean distances in the learned latent space recover the relevant similarity structure among qualitative categories (fillers, morphologies) with respect to dielectric constant and breakdown strength. No post-hoc validation, alignment with independent physical knowledge, or ablation against random embeddings is presented to support this.

    Authors: The LVGP learns embeddings from data to capture objective-relevant similarities. We acknowledge the value of explicit validation and will add a post-hoc analysis of the learned latent spaces together with an ablation comparing performance against random embeddings. revision: yes

  3. Referee: [LVGP] LVGP implementation: the latent-space dimension is listed among the free parameters, but the manuscript gives no procedure, cross-validation, or sensitivity analysis for its selection, nor does it report the specific dimension(s) used in the case study.

    Authors: We will report the specific latent dimension used in the case study and include a description of its selection along with a sensitivity analysis. revision: yes

Circularity Check

0 steps flagged

No circularity: LVGP surrogate and case-study validation are independent of input data

full rationale

The paper introduces LVGP as a novel surrogate that learns latent embeddings for qualitative variables to enable a standard GP kernel, then applies expected improvement for optimization. The central performance claim is demonstrated via an external case study on polymer nanocomposites rather than any equation or result that reduces to a fitted parameter defined from the same data. No self-definitional steps, fitted-input predictions, or load-bearing self-citations appear in the provided text; the derivation chain remains self-contained against the case-study benchmark.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The framework depends on the LVGP model whose latent dimensions and kernel hyperparameters are fitted to data; standard GP stationarity and smoothness assumptions are invoked without additional justification.

free parameters (2)
  • latent space dimension
    Dimensionality of the continuous embedding for each qualitative variable must be chosen or optimized during model fitting.
  • GP kernel hyperparameters
    Length scales and variance parameters of the covariance function are fitted to the mixed data.
axioms (1)
  • domain assumption The latent embedding induces a valid positive-definite covariance function over the mixed-variable domain.
    Required for the Gaussian process surrogate to be well-defined.

pith-pipeline@v0.9.0 · 5798 in / 1200 out tokens · 30641 ms · 2026-05-25T08:51:46.847346+00:00 · methodology

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

41 extracted references · 41 canonical work pages · 3 internal anchors

  1. [1]

    Materials genome initiative for global competitiveness,

    J. P. Holdren, "Materials genome initiative for global competitiveness," National Science and technology council OSTP . Washington, USA, 2011

    work page 2011

  2. [2]

    Computational design of hierarchically structured materials,

    G. B. Olson, "Computational design of hierarchically structured materials," Science, vol. 277, no. 5330, pp. 1237-1242, 1997

    work page 1997

  3. [3]

    Commentary: The Materials Project: A materials genome approach to accelerating materials innovation,

    A. Jain et al. , "Commentary: The Materials Project: A materials genome approach to accelerating materials innovation," Apl Materials, vol. 1, no. 1, p. 011002, 2013

    work page 2013

  4. [4]

    Materials design and discovery with high -throughput density functional theory: the open quantum materials database (OQMD),

    J. E. Saal, S. Kirklin, M. Aykol, B. Meredig, and C. Wolverton, "Materials design and discovery with high -throughput density functional theory: the open quantum materials database (OQMD)," Jom, vol. 65, no. 11, pp. 1501-1509, 2013

    work page 2013

  5. [5]

    Perspective: NanoMine: A material genome approach for polymer nanocomposites analysis and design,

    H. Zhao, X. Li, Y . Zhang, L. S. Schadler, W. Chen, and L. C. Brinson, "Perspective: NanoMine: A material genome approach for polymer nanocomposites analysis and design," APL Materials, vol. 4, no. 5, p. 053204, 2016

    work page 2016

  6. [6]

    NanoMine schema: An extensible data representation for polymer nanocomposites,

    H. Zhao et al., "NanoMine schema: An extensible data representation for polymer nanocomposites," APL Materials, vol. 6, no. 11, p. 111108, 2018

    work page 2018

  7. [7]

    Descriptor - based methodology for statistical characterization and 3D reconstruction of microstructural materials,

    H. Xu, D. A. Dikin, C. Burkhart, and W. Chen, "Descriptor - based methodology for statistical characterization and 3D reconstruction of microstructural materials," Computational Materials Science, vol. 85, pp. 206-216, 2014

    work page 2014

  8. [8]

    A descriptor -based design methodology for developing heterogeneous microstructural materials system,

    H. Xu, Y . Li, C. Brinson, and W. Chen, "A descriptor -based design methodology for developing heterogeneous microstructural materials system," Journal of Mechanical Design, vol. 136, no. 5, p. 051007, 2014

    work page 2014

  9. [9]

    Characterization and Design of Functional Quasi-Random Nanostructured Materials Using Spectral Density Function,

    S. C. Yu et al., "Characterization and Design of Functional Quasi-Random Nanostructured Materials Using Spectral Density Function," Journal of Mechanical Design, 139(7), 071401. 12 https://doi.org/10.1115/1.4036582, vol. 139, no. July, pp. 135 -145, 2016

    work page doi:10.1115/1.4036582 2016

  10. [10]

    A Spectral Density Function Approach for Active Layer Design of Organic Photovoltaic Cells,

    U. Farooq Ghumman et al. , "A Spectral Density Function Approach for Active Layer Design of Organic Photovoltaic Cells," Journal of Mechanical Design, vol. 140, no. 11, pp. 111408-111408-14, 2018

    work page 2018

  11. [11]

    Microstructural Materials Design Via Deep Adversarial Learning Methodology,

    Z. Yang, X. Li, L. Catherine Brinson, A. N. Choudh ary, W. Chen, and A. Agrawal, "Microstructural Materials Design Via Deep Adversarial Learning Methodology," Journal of Mechanical Design, vol. 140, no. 11, pp. 111416-111416-10, 2018

    work page 2018

  12. [12]

    Taking the human out of the loop: A review of bayesian optimization,

    B. Shahriari, K. Swersky, Z. Wang, R. P. Adams, and N. De Freitas, "Taking the human out of the loop: A review of bayesian optimization," Proceedings of the IEEE, vol. 104, no. 1, pp. 148 -175, 2016

    work page 2016

  13. [13]

    Efficient global optimization of expensive black -box functions,

    D. R. Jones, M. Schonlau, and W. J. Welch, "Efficient global optimization of expensive black -box functions," Journal of Global optimization, vol. 13, no. 4, pp. 455-492, 1998

    work page 1998

  14. [14]

    Adaptive strategies for materials design using uncertainties,

    P. V . Balachandran, D. Xue, J. Theiler, J. Hogden, and T. Lookman, "Adaptive strategies for materials design using uncertainties," Scientific reports, vol. 6, p. 19660, 2016

    work page 2016

  15. [15]

    Rapid Bayesian optimisation for synthesis of short polymer fiber materials,

    C. Li et al. , "Rapid Bayesian optimisation for synthesis of short polymer fiber materials," Scientific Reports, vol. 7, no. 1, p. 5683, 2017/07/18 2017

    work page 2017

  16. [16]

    Multi -objective Optimization for Materials Discovery via Adaptive Design,

    A. M. Gopakumar, P. V . Balachandran, D. Xue, J. E. Gubernatis, and T. Lookman, "Multi -objective Optimization for Materials Discovery via Adaptive Design," Scientific reports, vol. 8, no. 1, p. 3738, 2018

    work page 2018

  17. [17]

    Accelerated search for materials with targeted properties by adaptive design,

    D. Xue, P. V . Balachandran, J. Hogden, J. Theiler, D. Xue, and T. Lookman, "Accelerated search for materials with targeted properties by adaptive design," Nature communications, vol. 7, p. 11241, 2016

    work page 2016

  18. [18]

    Identifying interphase properties in polymer nanocomposites using adaptive optimization,

    Y . Wang et al., "Identifying interphase properties in polymer nanocomposites using adaptive optimization," Composites Science and Technology, vol. 162, pp. 146-155, 2018

    work page 2018

  19. [19]

    Enhanced Gaussian Process Metamodeling and Collaborative Optimization for Vehicle Suspension Design Optimization,

    S. Tao et al., "Enhanced Gaussian Process Metamodeling and Collaborative Optimization for Vehicle Suspension Design Optimization," presented at the International Design Engineering Technical Conferences and Computers and Information in Engineering Conference, 2017. Available: http://dx.doi.org/10.1115/DETC2017- 67976

    work page doi:10.1115/detc2017- 2017

  20. [20]

    Leveraging the nugget parameter for efficient Gaussian process modeling,

    R. Bostanabad, T. Kearney, S. Tao, D. W. Apley, and W. Chen, "Leveraging the nugget parameter for efficient Gaussian process modeling," International Journal for N umerical Methods in Engineering, vol. 114, no. 5, pp. 501-516, 2018

    work page 2018

  21. [21]

    The application of Bayesian methods for seeking the extremum,

    J. Mockus, V . Tiesis, and A. Zilinskas, "The application of Bayesian methods for seeking the extremum," Towards global optimization, vol. 2, no. 117-129, p. 2, 1978

    work page 1978

  22. [22]

    A new method of locating the maximum point of an arbitrary multipeak curve in the presence of noise,

    H. J. Kushner, "A new method of locating the maximum point of an arbitrary multipeak curve in the presence of noise," Journal of Basic Engineering, vol. 86, no. 1, pp. 97-106, 1964

    work page 1964

  23. [23]

    Gaussian Process Optimization in the Bandit Setting: No Regret and Experimental Design

    N. Srinivas, A. Krause, S. M. Kakade, and M. Seeger, "Gaussian process optimizat ion in the bandit setting: No regret and experimental design," arXiv preprint arXiv:0912.3995, 2009

    work page internal anchor Pith review Pith/arXiv arXiv 2009

  24. [24]

    The correlated knowledge gradient for simulation optimization of continuous parameters using gaussian process re gression,

    W. Scott, P. Frazier, and W. Powell, "The correlated knowledge gradient for simulation optimization of continuous parameters using gaussian process re gression," SIAM Journal on Optimization, vol. 21, no. 3, pp. 996-1026, 2011

    work page 2011

  25. [25]

    A Latent Variable Approach to Gaussian Process Modeling with Qualitative and Quantitative Factors

    Y . Zhang, S. Tao, W. Chen, and D. W. Apley, "A Latent Variable Approach to Gaussian Process Modeling with Qualitative and Quantitative Factors," arXiv preprint arXiv:1806.07504, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  26. [26]

    Pareto optimality in multiobjective problems,

    Y . Censor, "Pareto optimality in multiobjective problems," Applied Mathematics and Optimization, journal article vol. 4, no. 1, pp. 41-59, March 01 1977

    work page 1977

  27. [27]

    A sequential design for approximating the pareto front using the expected pareto improvement function,

    D. C. T. Bautista, "A sequential design for approximating the pareto front using the expected pareto improvement function," The Ohio State University, 2009

    work page 2009

  28. [28]

    Nanotechnology in high voltage insulation systems for turbine generators-First results,

    J. R. Weidner, F. Pohlmann, P. Gröppel, and T. Hildinger, "Nanotechnology in high voltage insulation systems for turbine generators-First results," 17th ISH, Hannover, Germany, 2011

    work page 2011

  29. [29]

    Trends in the ultimate breakdown strength of high dielectric-constant materials,

    J. W. McPherson, J. Kim, A. Shanware, H. Mogul, and J. Rodriguez, "Trends in the ultimate breakdown strength of high dielectric-constant materials," IEEE transactions on electron devices, vol. 50, no. 8, pp. 1771-1778, 2003

    work page 2003

  30. [30]

    Materials Informatics and Data System for Polymer Nanocomposites Analysis and Design,

    Wei Chen et al., "Materials Informatics and Data System for Polymer Nanocomposites Analysis and Design," in Big, Deep, and Smart Data in the Physical Sciences, 2018

    work page 2018

  31. [31]

    Niblack, An Introduction to Image Processing

    W. Niblack, An Introduction to Image Processing. Englewood Cliffs, NJ: Prentice-Hall, 1986, pp. 115-116

    work page 1986

  32. [32]

    Morphologically dependent alternating ‐ current and direct ‐current breakdown strength in silica –polypropylene nanocomposites,

    T. Krentz et al. , "Morphologically dependent alternating ‐ current and direct ‐current breakdown strength in silica –polypropylene nanocomposites," Journal of Applied Polymer Science, vol. 134, no. 1, 2017

    work page 2017

  33. [33]

    Phase separation by s pinodal decomposition in isotropic systems,

    J. W. Cahn, "Phase separation by s pinodal decomposition in isotropic systems," The Journal of Chemical Physics, vol. 42, no. 1, pp. 93-99, 1965

    work page 1965

  34. [34]

    Predicting the breakdown strength and lifetime of nanocomposites using a multi -scale modeling approach,

    Y . Huang et al. , "Predicting the breakdown strength and lifetime of nanocomposites using a multi -scale modeling approach," Journal of Applied Physics, vol. 122, no. 6, p. 065101, 2017

    work page 2017

  35. [35]

    Rethinking Interphase Representations for Modeling Viscoelastic Properties for Polymer Nanocomposites

    X. Li et al. , "Rethinking Interphase Representations for Modeling Viscoelastic Properties for Polymer Nanocomposites," arXiv preprint arXiv:1811.06238, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  36. [36]

    Dielectric spectroscopy an alysis using viscoelasticity-inspired relaxation theory with finite element modeling,

    H. Zhao et al. , "Dielectric spectroscopy an alysis using viscoelasticity-inspired relaxation theory with finite element modeling," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 24, no. 6, pp. 3776-3785, 2017

    work page 2017

  37. [37]

    Validation of a novel dielectric constant simulation model and the determination of its physical parameters,

    M. G. Todd and F. G. Shi, "Validation of a novel dielectric constant simulation model and the determination of its physical parameters," Microelectronics journal, vol. 33, no. 8, pp. 627 -632, 2002

    work page 2002

  38. [38]

    On the size and dielectric properties of the interphase in epoxy -alumina nanocomposite,

    P. Maity, N. Gupta, V . Parameswaran, and S. Basu, "On the size and dielectric properties of the interphase in epoxy -alumina nanocomposite," IEEE Transactions on Dielectrics and Electrical Insulation, vol. 17, no. 6, 2010

    work page 2010

  39. [39]

    Random forests,

    L. Breiman, "Random forests," Machine learning, vol. 45, no. 1, pp. 5-32, 2001

    work page 2001

  40. [40]

    Toward the development of a quantitative tool for predicting dispersion of nanocomposites under non-equilibrium processing conditions,

    I. Hassinger et al., "Toward the development of a quantitative tool for predicting dispersion of nanocomposites under non-equilibrium processing conditions," Journal of Materials Science, vol. 51, no. 9, pp. 4238-4249, May 2016

    work page 2016

  41. [41]

    D. E. Goldberg, Genetic algorithms. Pearson Education India, 2006

    work page 2006