Machine Learning for Chemical Properties and Materials

Machine learning (ML) is quickly becoming a premier tool for modeling chemical processes and materials. Generally, ML provides a surrogate model trained on the dataset of some reference data. This model establishes a relationship between structure and underlying chemical properties, guiding chemical discovery. Designing high-quality training data sets is crucial to overall model accuracy. To address this problem, I will describe the active learning strategy, in which new data are automatically collected for atomic configurations that produce large ML uncertainties. The locality approximation underpinning favorable computational scaling of the ML models, is another severe limitation that fails to capture long-range effects that may arise from charge transfer, polarization, electrostatic or dispersion interactions. I will also discuss how ML models can overcome nonlocality (via introduction of interaction layers, self-consistent cycles, or charge equilibration schemes) and exemplify their performance for chemical problems with nonlocalities. All these advances are exemplified by applications to molecules and materials. Exciting new method development and explosive growth of user-friendly ML frameworks, designed for chemistry, demonstrate that the field is evolving towards physics-based models augmented by data science.