The long term goal is to accelerate molecular simulations using generative models. To do that, several different approaches might be used. In the near future, we plan to be working on recovering the missing degrees of freedom of a coarse grained system using a generative model that is able to generate samples from a conditional distribution. After having obtained a proof of concept of this idea on a small model system, we now want to scale to a larger system where it is unfeasible to run training and inference on a personal computer. In the future, we will be targeting the following points as well, possibly combined with the aforementioned coarse graining approach: 1. Build a long-time-step surrogate for the simulation process to speed up the characterization of molecular dynamics. 2. Extend the surrogate to generalize across the space of molecules by leveraging physical symmetries of molecules, to establish meaningful parameter sharing schemes.