Leveraging Legacy Data to Accelerate Materials Design via Preference Learning

TL;DR

This paper introduces a calibration-free preference learning approach that leverages legacy experimental data to significantly improve Bayesian optimization in materials design, avoiding complex data calibration.

Contribution

A novel calibration-free strategy that uses pairwise comparisons within legacy data to enhance Bayesian optimization in materials science.

Findings

Bayesian optimization performance improved with legacy data.

The method works for organic molecules and inorganic materials.

Calibration-free approach simplifies data integration.

Abstract

Machine learning applications in materials science are often hampered by shortage of experimental data. Integration with legacy data from past experiments is a viable way to solve the problem, but complex calibration is often necessary to use the data obtained under different conditions. In this paper, we present a novel calibration-free strategy to enhance the performance of Bayesian optimization with preference learning. The entire learning process is solely based on pairwise comparison of quantities (i.e., higher or lower) in the same dataset, and experimental design can be done without comparing quantities in different datasets. We demonstrate that Bayesian optimization is significantly enhanced via addition of legacy data for organic molecules and inorganic solid-state materials.