University of California, Los Angeles,
Keywords: glass, material discovery, machine learning
Summary:Glasses play a key role in modern society (e.g., fiber optics, nuclear waste immobilization, or protective cover for smartphones). However, the design of new glasses is often plagued by poorly efficient Edisonian â€œtrial-and-errorâ€ discovery approaches. As an alternative route, new approaches relying on artificial intelligence and machine learning can accelerate the discovery and optimization of novel glasses with enhanced properties. Here, I will review some recent progress in adopting machine learning to predict the properties of glasses as a function of their composition and accelerate the discovery of new glasses with unusual functionalities. I will discuss how "physics-informed" machine learning can overcome the traditional limitations of "blind machine learning," that is, by (i) enhancing the potential for extrapolation far from the training set, (ii) increasing the interpretability of the models, and (iii) reducing the need for large datasets.