Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz
AKPIK 2: Parallel Talks
AKPIK 2.7: Vortrag
Montag, 16. März 2026, 17:30–17:45, KS 00.003
Resource Awareness in Deep Learning-based Imaging in Radio Interferometry — •Anno Knierim1,2, Kevin Schmitz1,2, Christian Arauner1,2, Raphael Fischer1,2, Dominik Elsässer1,2, and Wolfgang Rhode1,2 — 1TU Dortmund University, Dortmund, Germany — 2Lamarr Institute for Machine Learning and Artificial Intelligence, Dortmund, Germany
Recent approaches in radio interferometry aim to improve image cleaning of measurements using machine learning techniques. Reconstructing sources using these novel techniques has the advantage of being agnostic to the initial parameters used in traditional cleaning algorithms.
The radionets-project is a multi-software environment developed at TU Dortmund University. The main deep-learning framework, radionets, reconstructs calibrated data from radio observations using convolutional neural networks (CNNs). The framework aims to achieve a high dynamic range and produce high-resolution source images.
Due to their high complexity, applying deep learning models sustainably requires balancing predictive performance with resource consumption. The environmental impacts extend well beyond training, as model complexity also affects the efficiency during data simulation and inference. This talk presents ongoing efforts to track and estimate the environmental impact of our models using CodeCarbon, PyTorch Lightning, and MLFlow.
Keywords: Radio astronomy; Radio interferometry; Machine learning; Simulations; Sustainability