Original title: Morphology of Radio Sources in Representation Space
Nicolas Baron Perez, Marcus Brüggen, Luisa Lucie-Smith · September 2026
Radio sources in space come in complicated shapes, called morphologies, which are difficult to classify. Astronomers wanted to find rare, unknown shapes in a massive telescope survey called LoTSS-DR3, which contains > 13 million sources. Previously, they had recognized 12 shape classes. To find new ones, they ran an ensemble of classifiers on the new data. This tool created mathematical profiles, called representations, and assigned class probabilities to the sources. For sources where the tool was unsure—meaning low class probabilities—the researchers searched for new clusters grouped together in representation spaceAn abstract map where similar data points are grouped together for comparison.. They found that 88% of the sources matched the original clusters. However, looking at the rest revealed six new morphological classes, including rare winged and FR-ambiguous shapes. The resulting 18 classes are heavily skewed: four dominant classes make up 69% of the sample. This shows that representation learning can discover entirely new classes in unexplored data.
You're organizing a closet full of shirts by type, like t-shirts, polos, and dress shirts. Most shirts fit neatly, but some have unusual cuts or fabrics that don't match any category. In this paper, the researchers use a method to group radio sources by their visual patterns, and for the ones that don't fit the known types, it finds new categories based on how similar they are.
This helps astronomers map the universe more precisely by finding rare radio shapes that were previously missed. It changes how future large surveys are analyzed, ensuring we don't overlook important cosmic phenomena.
The abstract does not specify how the method performs on other types of data beyond radio sources from LOFAR surveys. It does not address the computational cost or scalability to even larger datasets. The approach identifies new classes but does not provide detailed physical interpretations for them.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Morphology of Radio Sources in Representation Space — Nicolas Baron Perez, Marcus Brüggen, Luisa Lucie-Smith
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.