Kevin Schmitz: Deep Learning for Radio Interferometric Imaging

Tuesday 13 Oct 2026, 13:15 → 14:15 Europe/Stockholm

FB54 (AlbaNova Main Building)

Radio astronomy is entering a new era. The Square Kilometre Array (SKA) and its precursors such as MeerKAT will map the radio sky with unprecedented sensitivity, addressing questions ranging from the epoch of reionization to cosmic magnetism and the growth of galaxy clusters. Yet radio interferometers do not take pictures directly. They sample the Fourier transform of the sky at a limited set of points, and reconstructing an image from these incomplete measurements is a difficult inverse problem. The classical algorithms used for decades are struggling with the data volumes of these new instruments and with faint, extended emission.

Machine learning is emerging as a powerful alternative. Neural networks trained on physically realistic simulations, an approach familiar from particle physics, can learn to recover the missing information quickly. Their adoption in radio astronomy, however, depends on key open issues: the reliability of network-based reconstructions, their transfer from simulated to real observations, and the quantification of uncertainties in the resulting images.

In this talk, I will give an overview of deep learning-based imaging in radio interferometry and its recent progress. Using the radionets framework as an example, I will show how a cascade of specialized networks reconstructs both compact sources and diffuse emission, and how it performs on real MeerKAT observations of a galaxy cluster compared with classical methods. I will close with the challenges that remain before these methods can be used routinely in the SKA era.

About the speaker: Kevin Schmitz is a postdoctoral researcher at the Hamburger Sternwarte, part of the Department of Physics at Universität Hamburg. He received his PhD from TU Dortmund in 2023, where he developed machine learning-based analysis strategies for physics in close collaboration with computer scientists. His current research centers on deep learning methods for radio interferometric imaging, with applications to MeerKAT and the upcoming SKA.