Erlangen 2026 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz

AKPIK 2: Parallel Talks

AKPIK 2.6: Talk

Monday, March 16, 2026, 17:15–17:30, KS 00.003

Methodological Preparation for Measurements with the DSA-2000: Simulation, Analysis, and Reconstruction Using Neural Networks — •Jonas Grossmann¹, Kevin Schmitz^{1, 2}, Anno Knierim^{1, 2}, Christian Arauner^{1, 2}, and Tom Gross^{1, 2} — ¹TU Dortmund University, Dortmund, Germany — ²Lamarr Institute for Machine Learning and Artificial Intelligence, Dortmund, Germany

A new generation of radio telescopes is set to begin operations later this decade. Among these is the Deep Synoptic Array 2000 (DSA-2000), a radio survey interferometer. It comprises 2000 small antennas, each 5 m in diameter, distributed over an area of 160 square kilometres in the Nevada Desert, USA. This configuration is optimised for rapid, high-resolution surveys of the entire northern sky. The DSA-2000 will achieve at least an order of magnitude improvement in angular resolution, enabling the discovery of billions of previously unknown radio sources. The resulting data volumes are unprecedented, surpassing the capabilities of conventional analysis methods and necessitating new approaches for translating interferometric measurements into cleaned images. The radionets-project group at TU Dortmund University aims to develop neural networks that, could automatically reconstruct dirty radio images. Training these networks requires large, realistic datasets. Given that all 2000 antennas cannot operate flawlessly at all times, the goal is to train the networks to deliver stable and reliable reconstructions even when subsets of antennas are inactive. This talk outlines the framework in radionets-project required to create the datasets necessary for training these neural networks.

Keywords: Radio Astronomy; Radio Interferometry; Machine Learning; DSA-2000