(245b) Going Beyond Accurate Models - What Comes Next in Machine Learning

Machine learning (ML) for molecules and materials has rapidly advanced in the last few years and we're now able to accurately predict many properties. But what comes after creating an accurate model? In this talk, I will discuss progress we're making in areas of explaining predictions, extracting physical insights using accurate ML models, and how to use predictions in practice. Explaining predictions is part of a broader field of explainable artificial intelligence (XAI) and is an important emerging topic when ML predictions affect human decision making, like deciding which compound to test or predicting material properties that impact human health. We're also exploring how to extract principled equations from ML models, and seeing how they can connect to mechanisms. Finally, I'll briefly touch on how we can take ML models from our computers or cloud to be deployed on phones and desktops without installation or computer resources.