Particle physics experiments rely heavily on simulations. Any possible hypothesis of nature can be simulated and subsequently compared to experimental data. Among the different steps in the data processing chain, the simulation of the detector response is typically the most time-consuming one. Recent developments in deep generative models showed impressive results in emulating the expensive full simulation. Especially normalizing flows and diffusion models have shown great performance in these tasks.

The "Machine Learning for Particle Physics" group offers projects for Master Theses that explore the application of such deep generative models to new datasets and detector layouts. For more details, please contact Claudius.Krause(a)oeaw.ac.at.

The codes are based on python and we will be using git for version control. Existing experience in these will be helpful, but motivation to learn them is sufficient.