Data-driven approaches in radio astronomy

Presentation slides for my invited talk at the EAS 2025 Symposium S15f: The changing macrocosm of astroinformatics : big data, catalogues, machine learning, challenges, and more. Herein I discuss the tools and techniques that I and others are developing and using to tackle the scale and complexity of data coming from the SKA and its precursor instruments. After laying out the problem of immense data rates and rare, interesting sources we wish to find, I demonstrate how citizen scientists are well suited to the task of finding transients in our datastreams using data from the ThunderKAT survey on MeerKAT. I then show how these volunteer classifications can provide a test bed against which to compare other techniques including anomaly detection models and Gaussian Processes. Anomaly detection algorithms show huge promise at tackling the large data rates we expect from the SKA, whilst still allowing astronomers to select for their (subjective) anomalies of interest e.g. transients. I then show how applying these now verified methods to the ASKAP VAST survey yields a wide range of novel and interesting transients whilst reducing our search volume by orders of magnitude. Finally, I showcase some work on Gaussian processes, which show promise as alternative features with which to describe our data and prove to be more robust than traditional metrics.