Researchers have designed a machine learning method that can predict the structure of new materials with five times the efficiency of the current standard, removing a key roadblock in developing advanced materials for applications such as energy storage and photovoltaics.