Best Practices for Multi-Fidelity Bayesian Optimization in Materials and Molecular Research

2024-10-01

arXiv.org

AI Summary

Multi-fidelity computational resources coexist in materials and molecular discovery, yet multi-fidelity Bayesian optimization (MFBO) lacks systematic, domain-specific guidance for chemical applications. Method: This work establishes the first comprehensive MFBO evaluation framework tailored to chemistry, integrating multi-fidelity surrogate models with expected improvement (EI) and upper confidence bound (UCB)-based acquisition functions. We systematically assess the coupling between model informativeness and evaluation cost on synthetic benchmarks and three real-world tasks: molecular property optimization, catalyst screening, and polymer design. Results: MFBO achieves comparable or superior solution quality while reducing high-fidelity evaluations by 35–60% on average. We introduce the first chemistry-oriented MFBO applicability criteria and configuration guidelines, uncovering fundamental mechanisms by which fidelity trade-offs govern optimization efficacy—providing a reusable, experiment-driven methodology for accelerated discovery.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Reasoning under Uncertainty: Stochastic Optimization
• Machine Learning: Optimization

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Economics, Online Markets and Human Computation: Fairness, privacy, and diversity in economic environments
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Multi-fidelity Bayesian Optimization (MFBO) is a promising framework to speed up materials and molecular discovery as sources of information of different accuracies are at hand at increasing cost. Despite its potential use in chemical tasks, there is a lack of systematic evaluation of the many parameters playing a role in MFBO. In this work, we provide guidelines and recommendations to decide when to use MFBO in experimental settings. We investigate MFBO methods applied to molecules and materials problems. First, we test two different families of acquisition functions in two synthetic problems and study the effect of the informativeness and cost of the approximate function. We use our implementation and guidelines to benchmark three real discovery problems and compare them against their single-fidelity counterparts. Our results may help guide future efforts to implement MFBO as a routine tool in the chemical sciences.

Problem

• Evaluating parameters in Multi-fidelity Bayesian Optimization for chemical tasks
• Providing guidelines for MFBO use in experimental settings
• Benchmarking MFBO against single-fidelity methods in real discovery problems

Innovation

• Multi-fidelity Bayesian Optimization for materials discovery
• Guidelines for MFBO use in experimental settings
• Benchmarking MFBO against single-fidelity methods