Active learning for regression of structure–property mapping: the importance of sampling and representation

Abstract

We develop an active workflow for calibrating microstructure–property relationships when a large dataset of microstructures is available, but the cost associated with evaluating the properties associated is high.