From Mercury to the Galaxy (and beyond): Unlocking Hidden Structures in High-Dimensional Astronomical Data

Modern astrophysics is increasingly flooded with high-dimensional, complex, and noisy datasets. However, traditional machine learning blindly treats data points as isolated entities, unknowingly discarding a wealth of relational information that could otherwise unravel complex patterns. In an intuitive and jargon-free talk, I will introduce Graph Attention Autoencoders — a new technique that processes datasets as networks of interconnected points. By leveraging these connections, we can enrich the data with a crucial relational context, bringing to light hidden patterns that would otherwise remain submerged in noise. I will also demonstrate how this single, accessible approach provides a unified solution across vastly different domains in astronomy: from automated morpho-spectral geological mapping on planetary surfaces to retrieving dispersed stellar groups in the Milky Way.