[Paper Review] Data-Centric Mixed-Variable Bayesian Optimization For Materials Design

Akshay Iyer, Yichi Zhang+6 more|arXiv (Cornell University)|Jul 4, 2019

Machine Learning in Materials Science38 references4 citations

TL;DR

This paper proposes a data-centric mixed-variable Bayesian optimization framework that integrates heterogeneous data from literature, experiments, and simulations to accelerate materials design. By leveraging a Latent Variable Gaussian Process (LVGP) to model qualitative and quantitative variables together, the method efficiently navigates complex, nonlinear design spaces and identifies optimal polymer nanocomposite compositions and morphologies in tens of iterations, demonstrating strong performance on single- and multi-objective optimization tasks.

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.

Motivation & Objective

• To address the challenge of optimizing materials design when both qualitative (e.g., morphology type) and quantitative (e.g., composition) variables exist, leading to disjointed property spaces.
• To overcome data scarcity and high-cost simulations by leveraging existing heterogeneous data from literature, experiments, and simulations.
• To develop a surrogate modeling approach that can quantify uncertainty in regions with limited data, especially for qualitative variables.
• To enable efficient navigation of complex, nonlinear design spaces through a balanced exploration-exploitation strategy.
• To extend the framework to multi-objective optimization for identifying the Pareto frontier with minimal evaluations.

Proposed method

• The framework employs a Latent Variable Gaussian Process (LVGP) to map qualitative variables into a continuous latent space, enabling covariance computation across mixed variable types.
• The LVGP model captures the relationship between design variables (composition, morphology, processing) and material properties, even with sparse or heterogeneous data.
• Expected Improvement (EI) is used as the acquisition function to balance exploration of uncertain regions and exploitation of promising designs.
• The optimization loop sequentially selects new design candidates based on EI, minimizing the number of expensive evaluations.
• For multi-objective optimization, the framework extends EI to handle multiple performance criteria and identifies the Pareto frontier efficiently.
• The method integrates data from diverse sources—literature, experiments, simulations—into a unified surrogate model to maximize knowledge extraction.

Experimental results

Research questions

• RQ1 How can mixed-variable Bayesian optimization be enhanced to handle both qualitative and quantitative design variables in materials design?
• RQ2 Can a data-centric approach effectively integrate heterogeneous data sources to improve surrogate modeling accuracy and uncertainty quantification?
• RQ3 To what extent can the LVGP-based framework reduce the number of required evaluations in materials optimization?
• RQ4 How well does the framework identify optimal designs for insulating polymer nanocomposites across composition and morphology?
• RQ5 Can the framework efficiently identify the Pareto frontier in multi-objective materials design with limited evaluations?

Key findings

• The LVGP-based surrogate model successfully models the joint effect of qualitative and quantitative variables on material properties, even with limited data.
• The framework locates optimal polymer nanocomposite compositions and morphologies in tens of iterations, significantly reducing the number of required experiments or simulations.
• The expected improvement acquisition function effectively balances exploration and exploitation in a complex, nonlinear design space.
• The method demonstrates strong performance on both single- and multi-objective optimization tasks, identifying the Pareto frontier with minimal evaluations.
• Integration of heterogeneous data from literature, experiments, and simulations enhances model generalization and uncertainty quantification.
• The framework enables efficient knowledge discovery from existing data, making it suitable for high-cost materials design problems.