SPS Seminar - Paying Attention to Radio Galaxy Morphologies

When:  Thursday 5^th May at 14.00

Where:  Microsoft Teams - Online

Speaker:  Micah Bowles (Manchester University)

Hosted by: Hugh Dickinson

Abstract:

Exa-scale astronomy is coming and automating as much astronomical data processing as possible will be essential in extracting scientific value from upcoming instruments on reasonable timescales. In the radio domain we are explicitly aware of the challenges being posed by the various surveys that the SKA and its precursors will undertake. Classifying different types of radio galaxies is one of the challenges which such new data sets pose. To address this challenge, one solution is to apply deep learning models. However, historically, deep learning models have been considered to be opaque and have consequently been met with some scepticism. As a start towards addressing this uncertainty, in this talk I will present recent results on the application of self-attention networks to radio morphology classifications. These models not only classify radio galaxies, but also produce images highlighting which regions of an input map are being used by the deep-learning model to decide on a final classification. I will also explain how we can introduce simple astrophysical priors into such models in order to improve their interpretability even further. Finally, I will outline how these models might be applied to SKA precursor surveys, specifically MeerKAT’s MIGHTEE-POL survey.

Bio:

Micah is a PhD student at the University of Manchester funded through the Alan Turing Institute and supervised by Prof. Anna Scaife. He is a member of the MIGHTEE survey’s polarization core working group, and the Radio Galaxy Zoo: EMU project. His research has focussed on big data imaging and enabling deep learning for radio astronomy. He organizes the joint SKAO-JBCA machine learning journal club.

chemistry cycles and the main drivers of atmospheric processes, such as the dust and water cycles, by combining computer models with the latest observations from the ExoMars Trace Gas Orbiter mission alongside multiple other spacecraft missions. He completed his PhD at the OU in 2015 and has been there since on several post-doctoral projects before his current fellowship.