Computational Core Unit (CCU)

Do you need support in your project work? Then contact us here.

Currently, the CCU supports the Radio Galaxies Classification project of CDL1.
In this project, generative neural networks are developed to generate artificial radio galaxy images to improve the performance of state-of-the-art image classifiers on FIRST[1] radio galaxy images[2]. To gather a decent amount of training data, we set up a clean dataset from various FIRST catalogues.

Further, we created pipelines for testing and Continuous Machine Learning approaches to monitor the performance of the classifier when newly generated galaxy images are added to the training data. As a classifier, we used a pretrained Vision Transformer where the attention-maps provide more insights on how the classifier performed with additional artificial radio galaxy images. The current challenge is to transfer these approaches to the huge number of radio galaxy images from the new LOFAR telescopes [3].

[1] Becker, R. H.; White, R. L.; Helfand, D. J.: The FIRST Survey: Faint Images of the Radio Sky at Twenty Centimeters. The Astrophysical Journal 450, p. 559, Sept. 1995.

[2] Kummer, Janis, Lennart Rustige, Florian Griese, Kerstin Borras, Marcus Brüggen, Patrick L. S. Connor, Frank Gaede, Gregor Kasieczka, and Peter Schleper. “Radio Galaxy Classification with WGAN-Supported Augmentation,” June 30, 2022. https://doi.org/10.48550/arXiv.2206.15131.

[3] van Haarlem, M. P. et al.: LOFAR: The LOw-Frequency ARray. A&A 556,A2, Aug. 2013.