Olwethu Sigwela

Radio Transient Detection in MeerKAT data using Unsupervised Machine Learning

Supervisors:

Dr. Dane Brown

Prof. Oleg Smirnov

Abstract

South Africa will be the world center of radio astronomy once it finishes building the Square Kilometer Array (SKA). This telescope will generate five zettabytes of data per year, far more than any existing radio telescope. Such volumes of data cannot be parsed by human astronomers, making software tools necessary. Unsupervised Machine Learning (UML) algorithms have shown promise in detecting radio transients, objects of great astronomical interest. Radio transients are short, anomalous events in radio telescope observations, typically associated with significant energy releases. Historically, human experts found and analysed these transients, a time-consuming process that could only cover a fraction of radio telescope observations. This study aims to use UML to create a system to detect transients in data from the MeerKAT telescope, the first phase of the SKA. MeerKAT has a large backlog of observations which UML can analyse, flagging transients for further analysis. Once the system is complete, the Centre for Radio Astronomy Techniques and Technologies intend to integrate it into their software pipeline.