[Paper Review] Bgolearn: a Unified Bayesian Optimization Framework for Accelerating Materials Discovery

Bin Cao, Jie Xiong+11 more|arXiv (Cornell University)|Jan 11, 2026

Machine Learning in Materials Science0 citations

TL;DR

Bgolearn provides a production-ready Bayesian optimization toolkit for materials discovery, enabling single- and multi-objective optimization with diverse surrogates and acquisition functions, plus a GUI, reducing experimental requirements by 40–60%.

ABSTRACT

Efficient exploration of vast compositional and processing spaces is essential for accelerated materials discovery. Bayesian optimization (BO) provides a principled strategy for identifying optimal materials with minimal experiments, yet its adoption in materials science is hindered by implementation complexity and limited domain-specific tools. Here, we present Bgolearn, a comprehensive Python framework that makes BO accessible and practical for materials research through an intuitive interface, robust algorithms, and materials-oriented workflows. Bgolearn supports both single-objective and multi-objective Bayesian optimization with multiple acquisition functions (e.g., expected improvement, upper confidence bound, probability of improvement, and expected hypervolume improvement etc.), diverse surrogate models (including Gaussian processes, random forests, and gradient boosting etc.), and bootstrap-based uncertainty quantification. Benchmark studies show that Bgolearn reduces the number of required experiments by 40-60% compared with random search, grid search, and genetic algorithms, while maintaining comparable or superior solution quality. Its effectiveness is demonstrated not only through the studies presented in this paper, such as the identification of maximum-elastic-modulus triply periodic minimal surface structures, ultra-high-hardness high-entropy alloys, and high-strength, high-ductility medium-Mn steels, but also by numerous publications that have proven its impact in material discovery. With a modular architecture that integrates seamlessly into existing materials workflows and a graphical user interface (BgoFace) that removes programming barriers, Bgolearn establishes a practical and reliable platform for Bayesian optimization in materials science, and is openly available at https://github.com/Bin-Cao/Bgolearn.

Motivation & Objective

• Lower the number of experiments needed to discover high-performance materials via Bayesian optimization.
• Provide a materials-friendly interface and workflows that reduce implementation barriers for researchers.
• Enable both single-objective and multi-objective optimization to handle trade-offs between material properties.
• Offer scalable surrogate models and uncertainty quantification suitable for high-dimensional design spaces.
• Deliver a graphical user interface (BgoFace) and open-source tooling to democratize BO in materials science.

Proposed method

• Introduce a unified, materials-oriented API that simplifies BO to 3–5 lines of code without sacrificing customization.
• Support multi-objective Bayesian optimization with acquisition functions such as EHVI, qNEHVI, MO-PI, and MO-UCB.
• Provide flexible surrogate models including Gaussian processes, random forests, gradient boosting, SVR, and neural networks with cross-validation for model selection.
• Incorporate bootstrap-based uncertainty quantification for non-GP surrogates to enable scalable MOBO.
• Include BgoFace, a GUI that generates equivalent Python code for reproducibility and lowers programming barriers.

Experimental results

Research questions

• RQ1 Can Bgolearn achieve data-efficient single-objective optimization across high-dimensional problems compared to baselines like random search and LHS?
• RQ2 How does Bgolearn perform on multi-objective optimization in terms of Pareto front quality (e.g., hypervolume) relative to NSGA-II and random baselines?
• RQ3 Do diverse surrogate models and acquisition functions in Bgolearn provide robust performance across different material design problems?
• RQ4 Can Bgolearn’s bootstrap-based uncertainty quantification enable scalable MOBO with non-GP surrogates while maintaining solution quality?
• RQ5 What is the practical impact of the GUI (BgoFace) on adoption speed and user experience for materials researchers with limited programming experience?

Key findings

• Bgolearn reduces the number of required experiments by approximately 40–60% compared with random search, grid search, and genetic algorithms while maintaining comparable or superior solution quality.
• In single-objective benchmarks, Bgolearn-GP with EI reaches 90% of the optimum with significantly fewer iterations than baselines (e.g., Hartmann-6D and Ackley-5D).
• For multi-objective problems, Bgolearn with EHVI achieves higher hypervolume than random search and NSGA-II on ZDT1 and DTLZ2 benchmarks.
• Bgolearn-RF achieves competitive performance with substantially lower compute time, making it suitable for high-throughput campaigns.
• Real-world applications demonstrate Bgolearn identifying high-performance TPMS structures, ultra-high-hardness HEAs, and strong/ductile medium-Mn steels, illustrating practical impact across metals, functional materials, and engineering systems.
• The BgoFace GUI enables rapid task completion (10–15 minutes for first-time users) and automatically generates reproducible Python code.