From Manifold Learning to Foundation Models: AI Representations for Galaxies and AGNs

Abstract: Modern astronomical surveys are generating data volumes that increasingly require machine learning methods capable of discovering structure beyond traditional analysis techniques. In this talk, I will present my recent work on the application of manifold learning and transformer-based foundation models to galaxy and AGN studies. I will discuss how unsupervised representation learning can reveal physically distinct galaxy populations and selection effects, and how modern foundation models encode information about galaxy morphology, stellar populations, and redshift within their latent representations.