Leveraging Legacy Data to Accelerate Materials Design via Preference Learning
This work addresses data scarcity for materials scientists by enabling efficient use of legacy data without complex calibration, though it is incremental as it builds on existing Bayesian optimization and preference learning methods.
The paper tackles the shortage of experimental data in materials science by introducing a calibration-free strategy using preference learning to integrate legacy data, demonstrating significant enhancement in Bayesian optimization for organic molecules and inorganic solid-state materials.
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.