Prospecting MeerKAT Continuum Data for Enigmatic Radio Sources with Unsupervised Vector-Quantised Variational Autoencoders

Abstract

We present a novel application of Vector quantised variational autoencoders (VQ-VAEs) as an unsupervised tool to probe deep 1.28 GHz radio continuum images taken from the MeerKAT Galaxy Cluster Legacy Survey (MGCLS) and compare their performance to other machine learning methods. We examine the effectiveness of VQ-VAEs in identifying radio continuum sources with anomalous structures in the image-plane domain. We compare performance to a supervised training set and other published anomaly detection methods, focusing especially on simple autoencoders (AE), Memory Unit Autoencoders (MUAE) and a human-in-the-loop software Astronomaly based on bootstrap your own latent (BYOL). Our investigations show that VQ-VAE perform better than simple AEs, are not as accurate as MUAE but are much faster and are also faster and comparable to Astronomaly while not requiring labelled data. We observe that they are able to remove a majority of the typical sources in such data, even when trained in an unsupervised manner on unlabelled data. We also provide our testing set of a large sample of manually labelled radio sources, in particular radio galaxies, taken from the MGCLS at 1.28 GHz. We find VQ-VAEs to be potentially useful as part of automated approaches towards searching through high volumes of data which are key in extracting the full scientific potential of the Square Kilometre Array and its pathfinders.