(426c) Latent Variable Machine Learning Framework for Catalysis

Machine learning has been successfully applied in recent years to screen materials for a variety of purposes. However, despite recent advances, most screening-based machine learning approaches are limited in generality and transferability, requiring new models to be created from scratch for each new application. This is particularly apparent in catalysis, where there are many possible intermediates and transition states of interest, in addition to a large number of potential catalytic materials. In this work we developed a new machine learning architecture which is built on chemical principles and allows creation of general, reusable models. Our new architecture uses latent variables to simplify the learning process, thereby making it more data-efficient while promoting transfer learning. We show that this architecture allows the creation of models that can be reused for many different applications, providing improved efficiency and convenience. The integration of latent variables also provides physical interpretability, as predictions can be explained in terms of the learned chemical environment as represented by the latent space. Lastly, we show that our new machine learning architecture is general and robust enough to handle multiple types of systems.