This is a Virtual Event
Location: Zoom from campus In Campbell Hall, Room 131
TITLE: Galaxy Clustering with Emulation and Equivariant Machine Learning
ABSTRACT:
There is significant untapped cosmological information in the clustering of galaxies, particularly at small scales. I will discuss two projects aimed at this. In the Aemulus project we take an emulation approach, populating dark matter only cosmological simulations with a simple galaxy–halo connection model and training machine learning (ML) emulators to predict clustering statistics. I will show that incorporating beyond-standard statistics sensitive to the local environment aids in constraining galaxy bias parameters and increases the precision on recovered cosmological parameters.
To account for uncertainties in galaxy formation, we require improved models of the galaxy–halo connection. I will present a new equivariant ML approach to learning the relationship between dark matter halo and galaxy properties in cosmological simulations. Our approach explicitly respects physical symmetries by describing halos in terms of a large set of invariant dimensionless scalars. I will show that this results in precise predictions of galaxy properties. These frameworks will be critical for the analysis of upcoming spectroscopic surveys.