In molecular simulations, very large amounts of data are generated. However, the vast majority of these data is not used today, as usually only few mean values are stored. In this research project, we will investigate how machine learning (ML) methods can be used to learn from this data. The aim is on the one side to increase the robustness and efficiency of the molecular simulations and to increase information gain on the other. Appropriate approaches are available and will be implemented, tested and evaluated using molecular simulations in the field of thermodynamics as examples. If successful, this work will result in a new class of ML-based methods for molecular simulation.

[back]