Scaling deep learning for discovery

Stable crystal discovery using a machine-learning model for oxidation-state balancing in a single crystal structure has been found in Materials Project 16 and OQMD 17 1e. The training sets are sampled uniformly from the materials from the Materials Project and from our structural pipeline. Following Roost (non-representation learning from stoichiometry)58, GNNs are effective at predicting the formation energy of a composition and structure.