Large Language Models for Explainable and Scalable Radio Galaxy Classification

Supervisors: Rob Lyon (LJMU), Andreea Font (LJMU), Hongming Tang (Tsinghua University)
Institution: Liverpool John Moores University

Modern telescopes are capturing vast amounts of detailed image data from galaxies across the universe, revealing intricate structures such as jets, lobes, and diffuse emissions. These features offer critical insights into galaxy evolution, the behaviour of supermassive black holes, and the distribution of dark matter. However, interpreting and labelling such complex features manually is time-consuming and inconsistent, especially as upcoming surveys like Euclid and the Vera Rubin Observatory’s Legacy Survey of Space and Time (LSST) will produce massive datasets containing billions of galaxies. 

In this interdisciplinary project, you will develop state-of-the-art AI tools to automatically analyse and describe galaxy structures in images, generating both standardised tags and plain-language summaries. You will apply advanced machine learning techniques, including Large Language Models (LLMs), to automate the interpretation and annotation of astronomical data, focusing primarily on images from the ASKAP EMU survey—which aims to detect over 40 million galaxies. Additionally, you will build upon and utilise existing human annotations from Radio Galaxy Zoo: EMU, actively contributing to this important Zooniverse project. You will have the opportunity to develop AI methods that integrate crucial model interpretability techniques, creating a system that not only accurately labels galaxy features, but also provides clear explanations for its reasoning. This will offer astronomers a reliable tool capable of uncovering new insights and accelerating discovery. 

As a student, you will work closely with researchers in the Astrophysics Research Institute (ARI), participate in the Radio Galaxy Zoo: EMU project, and benefit from a collaborative visit to China, where you will explore the latest advances in AI explainability. Your work will provide astronomers with powerful tools for handling the immense data volumes generated by new telescopes. The skills you acquire in explainable AI, large-scale data processing, and cross-domain applications will be highly transferable, enhancing your employability in fields such as medical imaging, remote sensing, and computer vision. 

Throughout the project, you will have access to the ARI’s postgraduate training program and data science training provided by the Centre for Doctoral Training LIV.INNO. You’ll also gain specialised skills through Nvidia DLI training, which will help you make the most of high-performance computing resources at Liverpool John Moores University. Additionally, a six-month industry placement will give you the opportunity to broaden your research experience and career skills, making this an exceptional opportunity to contribute to both astrophysics and the broader field of machine learning. 

This project will be carried out over 48 months and is fully funded (tuition fees + stipend set by UKRI guidelines + a research/training budget), inclusive of the 6-month industry placement. We encourage applications from underrepresented groups to help foster diversity, inclusivity, and equity within our research community. Applications are welcomed from those with a either a physics, astrophysics or computer science background. 

The delivery of this project is subject to funding approval. Information on how to apply will follow shortly.