Machine learning materials properties for small datasets

In order to make accurate predictions of material properties, current machine-learning approaches generally require large amounts of data, which are often not available in practice. In this work, a novel all-round framework is presented which relies on a feedforward neural network and the selection of physically-meaningful features. Next to being faster in terms of training time, this approach is shown to outperform current graph-network models on small datasets. In particular, the vibrational entropy at 305K of crystals is predicted with a mean absolute test error of 0.009 meV/K/atom (four times lower than previous studies). Furthermore, the proposed framework enables the prediction of multiple properties, such as temperature functions, by using joint-transfer learning. Finally, the selection algorithm highlights the most important features and thus helps understanding the underlying physics.