(V) Domain and Uncertain in Machine Learning Models for Materials Properties

Machine learning provides a powerful tool to predict materials properties. This is generallly done by fitting a machine learning model to a target propery as a function of readily available features, e.g., composition or atomic structure information. Such approaches are increasingly being used to predict properties from formation energies to band gaps to electromigration effective charge. However, relatively little attention has been paid to the critical issue of assessing the domain of the machine learning model and the accuracy of the predictions within that domain. In this talk we explore the effectiveness of some model error estimation methods, including ensemble and Bayesian methods, and consider how these might be used to obtain accurate error estimates within with the domain of the model. We apply the results on a realistic problem of modeling dilute impurity diffusion coefficients in a host, demonstrating that the model can predict accurate values for new systems but that the domain and errors are essential to consider for effective use of the model.