Speaker
Description
The rapid growth of astronomical survey data presents both an opportunity and a challenge: while large datasets increase the chances of discovering rare or unexpected sources, manual inspection quickly becomes impractical at scale. This work presents the development and application of anomaly detection pipelines using Astronomaly, an active learning-based framework, applied to two distinct astronomical datasets spanning optical and radio wavelengths.
In the optical domain, approximately 4 million galaxy images from the Dark Energy Camera Legacy Survey was explored. This work involved careful data curation, selection criteria optimisation, and evaluation of multiple active learning strategies. The pipeline identified 1635 anomalies including gravitational lens candidates, galaxy merger candidates, and 18 previously uncatalogued sources with highly unusual morphologies, all from only a few hours of human labelling. In the radio domain, a targeted pipeline combining self-supervised feature extraction with active learning was developed to detect diffuse radio emission in galaxy clusters from the MeerKAT Galaxy Cluster Legacy Survey. Significant effort went into source extraction, resolution-dependent feature comparison, and the development of a novel data cut to preferentially retain extended emission. Of the top 100 ranked sources, 99% exhibited diffuse emission characteristics, with 55% confirmed as cluster-related.
Across both domains, active learning proves essential: unsupervised anomaly detection alone consistently prioritises imaging artefacts and uninteresting outliers over scientifically valuable sources. With minimal human input, active learning rapidly reorients the search toward sources of genuine interest, offering a scalable path to discovery in the era of next-generation facilities such as the Square Kilometre Array and the Vera C. Rubin Observatory.
| Consent on use of personal information: Abstract Submission | Yes, I ACCEPT |
|---|