Deep learning, and in general differentiable programming, allow expressing many scientific problems as end-to-end learning tasks, while retaining some inductive bias derived from physics-based understanding. Common themes in scientific machine learning involve learning surrogate functions of expensive simulators, sampling complex distributions directly or time-propagation of known or unknown differential equation systems efficiently.

In this talk, we will analyse our recent work in applying deep learning surrogates and auto-differentiation in atomistic simulations of materials. In particular, we will explore active learning of machine learning potentials with differentiable uncertainty and their application to uncover the mechanism of ion diffusion in superionic inorganic conductor LGPS. Lastly, we will describe the application of differentiable simulations for learning interaction potentials from experimental data and for reaction-path finding without prior knowledge of collective variables.