AstroAI Workshop 2026
Dakshesh Kololgi
Learning the Cosmic Web: Inferring Cosmic Web Environments of Galaxies from Surveys
Presenter: Dakshesh Kololgi (University College London)
Title: PLANT: Learning the Cosmic Web: Inferring Cosmic Web Environments of Galaxies from Surveys
Date/Time: Monday, June 15, 4:00 PM - 5:30 PM
VIRTUAL
Abstract: Forthcoming and ongoing large galaxy surveys such as DESI offer unprecedented opportunities to study how the Cosmic Web influences galaxy formation and evolution.
A central challenge is the inference of galaxy environments in observational data, where the underlying dark matter density is not directly accessible. We present a novel graph neural network framework that learns the mapping from graph-encoded local galaxy connectivity to the large-scale tidal field without explicit density reconstruction. Our method is trained on galaxy catalogues from the IllustrisTNG-300 simulation, with targets computed from the dark-matter tidal tensor. In our initial work (Kololgi et al. 2025, RASTI. DOI: 10.1093/rasti/rzag025), we showed that this approach accurately recovers Cosmic Web structures in simulation data. Building on this, we extend the framework to full simulation-based inference via neural posterior estimation (using normalising flows), and forward-model key observational effects, such as survey selection and redshift-space distortions, yielding calibrated per-galaxy posteriors over the tidal eigenvalues. We are additionally exploring domain adaptation (e.g. CORAL) so that our models better generalise from simulations to observational data.
Our work provides a scalable and physically motivated pathway for probabilistic inference of Cosmic Web environments in galaxy surveys, which will facilitate systematic studies of how large-scale structure shapes galaxy formation and evolution.