Astro-Seminar: Searching for new Globular Clusters in the Andromeda Galaxy (M31) with Machine Learning

Globular clusters (GCs) are valuable probes of the history of their host galaxies, tracing out the history of mergers and star formation leading to the build-up of massive galaxies. This project build on my previous work on visually identifying new GCs in the halo of M31 (Huxor et al. 2014), on the Pan-Andromeda Archaeological Survey, which covers >400 square degrees of sky. The problems that arose in such visual inspection (classification) led me to follow the recent work of Wang et al. (2022), employing convolutional neural networks (CNNs) – in several student projects - to automate the procedure. We address the issue of a small training set of known GCs (used by Wang et al.) by supplementing these data with synthetic clusters. These additional data do indeed improve performance of the classifier. If time permits, I will also describe early experiments in using the vision understanding features of large language models (LLMs) to achieve the same goal.