Machine learning for atomic and molecular simulations

Michele Ceriotti
École Polytechnique Fédérale de Lausanne (EPFL)

Machine learning is finding applications to more and more tasks, in science as much as in everyday life. In this talk I will focus on how atomic and molecular simulations are being transformed by the use of statistical regression models, that make it possible to approximate accurately and efficiently atomistic properties computed from a few reference electronic-structure calculations.
I will argue about the advantages that are brought about by a physically-motivated framework, and about the insights that can be obtained by a critical application of ML methods. Examples will be given spanning molecular and condensed matter systems, and properties as diverse as magnetic nuclear chemical shieldings and the electron charge density, underscoring the general applicability of the approach.

Presentation (PDF File)

Back to Workshop IV: Using Physical Insights for Machine Learning