From October 2019 to April 2021 I worked on my Bachelor thesis for Informatics in the Physics-based Simulation group of Professor Nils Thuerey at the Technical University of Munich. The thesis was titled “Learning from Differentiable Physics to Simulate Liquids with Graph Networks” and had the goal to extend the simulation toolkit PhiFlow with a FLIP simulation and then train graph networks to simulate liquids based on generated data from the FLIP simulation. Parts of my thesis have been published as a technical report on arXiv.
During this project I gained experience in writing performant and differentiable fluid simulations, extended my knowledge of PyTorch and Tensorflow and added graph networks to my personal repertoire of deep learning tools.