Recent progress towards a universal machine learning model for reaction energetics in heterogeneous catalysis

Machine learning (ML) promises to increase the efficiency of screening a large number of materials for catalytic reactions. However, most existing ML models can only be applied to a specific reaction; therefore, new models usually have to be built from scratch for a new application. The effort and expense needed to create large datasets is also a major drawback of many ML methods. Hence, developing ML models that can be broadly applied to a wide range of different materials and reactions is crucial to further increase efficiency. In this review, we discuss recently developed ML methods in the field of heterogeneous catalysis that represent progress towards more general models. Notable progress has been made in improving generality which can lead to significant increases in efficiency and convenience.