AstroAI Lunch Talks - September 29, 2025 - Christian Jespersen
29 Sep 2025 - Joshua Wing
The video can be found here: https://www.youtube.com/watch?v=_ZT0f13lOmk
Speaker: Christian Jespersen (Princeton/CCA)
Title: Graph Neural Networks for Galaxies
Abstract: Galaxies are the medium through which we study the structure of the universe. Yet the statistical models most commonly applied to galaxies are generally oversimplified and cannot capture dependencies on environment or formation history. To address this, I will introduce Graph Neural Networks (GNNs), motivating their use and outlining their key features. GNNs are a general and powerful tool for physical modeling, able to naturally incorporate geometrical constraints, symmetries, and inductive biases (prior knowledge). This leads to both better and more robust models and enables us to ask detailed questions about the roles of environment and formation history in cosmological galaxy modeling. Finally, I will present a surprising equivalence between these two aspects of galaxy formation.