Erlangen 2026 – scientific programme
Parts | Days | Selection | Search | Updates | Downloads | Help
AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz
AKPIK 2: Parallel Talks
AKPIK 2.3: Talk
Monday, March 16, 2026, 16:30–16:45, KS 00.003
Simulating Protoplanetary Disks as Training Data for Neural Network based Reconstruction of Radio Interferometer Measurements — •Tom Gross1,2, Kevin Schmitz1,2, Christian Arauner1,2, and Anno Knierim1,2 — 1TU Dortmund University — 2Lamarr Institute for Machine Learning and Artificial Intelligence
Understanding the evolution of planetary systems is an important research field in modern astronomy. Planets form through accretion processes of dust and gas in protoplanetary disks around young stars.
The thermal radiation of the disks is measurable in the radio spectrum using radio interferometers. These enable the observation of small structures but only measure samples of an incomplete Fourier space. This intrinsic limitation results in noisy images, unsuitable for physical analyses.
The radionets-project uses a modern deep-learning framework to reconstruct the missing information. To enhance the scientific interpretation of the reconstructions, the training data needs to represent realistic source distributions. The software FARGO3D simulates the distribution of dust and gas in a protoplanetary disk using a hydrodynamical model. From these distributions, the thermal radiation can be simulated using Monte Carlo methods. This presentation shows the results of the simulations and how they are integrated into the data pipeline for the deep-learning based reconstruction.
Keywords: Radio astronomy; Radio interferometry; Protoplanetary Disks; Simulations; Machine Learning