Description
SKA pathfinders can sample wide swathes of the radio sky with unprecedented sensitivity and cadence. As a result, we are now discovering novel radio transients across an immense range of astrophysical regimes - from flare stars to FRBs. I will discuss recent, serendipitous discoveries being made with the MeerKAT and ASKAP radio telescopes. I will show how unsupervised machine learning techniques accelerate the search for interesting and anomalous sources in large datasets such as those expected from the SKA. These anomaly detection models can, with the use of active learning strategies, be customised to find not only anomalies, but those that are verified as interesting systems for a particular science case. The tested models show great success in recovering transients in our large dataset, reducing the volume of sources for vetting by an order of magnitude. Following these successes, I will show how the application of these anomaly detection techniques to ASKAP data has uncovered a huge range of transients including long period transients, further stellar flares and new radio detections of X-ray binary outbursts. I will demonstrate how the huge data rates and superb sensitivity of the SKA necessitates fast techniques for searching for anomalies and show that these techniques provide a way to detect transients at scale and on-the-fly, revolutionising our understanding of the dynamic radio sky.
