Assist. Prof. Dr. Julija Zavadlav (TUM): Accurate and Efficient Machine Learning-based Molecular Modeling

Molecular simulations have become a cornerstone of many disciplines ranging from material science to medicine. However, the quality of predictions critically depends on the employed model that defines particle interactions. A class of models with tremendous success in recent years are neural network (NN) potentials due to their flexibility and capacity to learn many-body interactions. In this talk, I will present the current state-of-the-art in deep molecular modeling. I will discuss the ongoing challenge of sufficiently large and broad training datasets and our approaches to alleviate this issue, including novel training strategies, combining different data sources, Bayesian uncertainty quantification, and active learning. I will showcase the effectiveness of these approaches for various test case solid and soft matter systems. I will also present our software chemtrain that enables NN potential learning through customizable training routines and advanced training algorithms.
Cookies, tea