Research projects within the AstroAI-Lab

AstroAI-Lab is a rather new research group and we are just getting started. Therefore the list of research projects is rather small.

Bachelor and master student projects

You are a student and looking for some interesting thesis projects in astrophysics and/or machine learning? Our group has plenty open projects in the broad field of galaxy formation and physics-informed ML. Do not hesitate to contact us if you like to know more about the projects or the group.

People and their projects

Eva: Star formation models in cosmological simulations

State-of-the-art cosmological hydrodynamical simulations of galaxy formation have reached the point at which their outcomes result in galaxies with ever more realism. Still, the employed sub-grid models are relatively simple empirical models that include several free parameters. One of the challenges is the robust identification and characterization of star forming regions and various methods have been proposed. However, the results of these methods are often inconsistent and the underlying assumptions are not always physically well motivated. In this project, we investigate how we can robustly identify star forming regions in cosmological simulations and explore novel prescriptions of calculating the star formation efficiency of these regions.

Rebekka: Galactic chemical enrichment in MW dwarf galaxies

With the advent of large spectroscopic surveys the amount of high quality chemodynamical data in the Milky Way and its satellite galaxies increased tremendously. Accurately and correctly capturing and explaining the detailed features in the high-quality observational data is notoriously difficult for state-of-the-art numerical models. Here we investigate a set of new cosmological galaxy formation models that include improved prescriptions for chemical enrichment and compare our findings to the observational data.

Ufuk: Galaxy morphology models with AI

We investigate the use of machine learning to create galaxy morphology models and encode the information contained in modern state-of-the-art galaxy simulations. Simulation data from the IllustrisTNG project is used to calculate the Eigengalaxies as the basis vectors of the transformed image space using Principal Component Analysis in two and three dimensions and it is demonstrated, how eigengalaxies encode specific morphological information. In the final step the NVIDA StyleGAN2-ada is trained on the 2D galaxy images. We find that a simple PCA model with 49 eigengalaxies can recover the general shape of 9763 two dimensional galaxy images and even if 99% of all images have a reconstruction error less than 9%, a more advanced model like a GAN is needed to describe highly nonlinear structures (e.g spiral arms) and small scale structures. All results with additional animations and interactive dashboards can be accessed through the GitHub-Repository of Ufuk's thesis.