Powered by Indico
v3.2.8
Galaxy properties couple in opaque ways, as they depend on each other, large-scale environment, and the history of the galaxy. Unfortunately, widely used statistical models of galaxies ignore all of these due to the difficulties of incorporating them. I will introduce graph neural networks as the tools necessary to accurately introduce environmental and historical dependencies into a statistical framework, and prove a surprising connection between the merger histories and environments of galaxies. I will then show how galaxy component coupling is implicitly treated in common inference codes, how this implicit treatment impacts the inference, and how we can let data lead us to fix it.