Welcome to the research group Machine Learning for Materials Science
The focus of the group is on developing and integrating machine learning techniques into computational methods for materials science. In particular, we target electronic structure methods, aiming to improve both their accuracy and computational efficiency with the help of deep learning.
The basic operational paradigm of electronic structure methods (quantum Monte Carlo, density functional theory, coupled clusters) has been to use as much physics as possible to derive effective models, and leave only little room for parametric freedom, because fitting many parameters to data has been traditionally hard. But modern deep learning techniques excel in fitting huge parametric models to data in a robust way. We seek to revisit this fundamental paradigm and develop methods that take full advantage of deep neural networks for electronic structure theory.