Institute for Advanced Study/Princeton University Early Universe/Cosmology Lunch Discussion

Abstract 1:The upcoming generation of cosmological surveys such as LSST will aim to map the Universe in great detail and on an unprecedented scale. This of course implies new and outstanding challenges at all levels of the scientific analysis, from pixel level data reduction to cosmological inference. In this talk, I will illustrate how recent advances in deep learning and associated automatic differentiation frameworks, can help us tackle these challenges and rethink our approach to data analysis for cosmological surveys. 
We will see how at the pixel level, combining physical models of the instrument (which account for noise/PSF) with data-driven deep generative models can enable us to solve a number of astronomical inverse problems ranging from deconvolution to deblending galaxy images. I will then illustrate how neural density estimation methods can help us leverage the full information content of numerical simulations for Bayesian inference, either on intermediate fields, like weak-lensing convergence maps, or all the way to cosmological parameters. And finally, I will present our efforts to develop tools for automatically differentiable physical models, from analytic cosmological observables to N-body simulations, opening the door to a range of novel and efficient gradient-based inference techniques, and allowing for fast hybrid physical/ml models.

Abstract 2: I will discuss methods and forecasts for cosmology with cluster weak lensing, focusing on the papers https://arxiv.org/abs/2012.01956 and https://arxiv.org/abs/1906.06499 We advocate an approach that is analogous to galaxy-galaxy lensing, in which the primary observable is the mean tangential shear profile of clusters above a threshold in some mass proxy (richness, SZ decrement, X-ray luminosity, etc.), across the linear and non-linear regimes, together with the mean space density of those clusters. The 1906.06499 paper describes the further gains from incorporating the cluster-galaxy cross-correlation and galaxy auto-correlation as additional observables.
I will briefly advertise https://arxiv.org/abs/1907.06611 which calculates weak lensing covariance matrices, a tricky but soluble problem. These papers forecast achievable statistical errors on sigma8 below 1% for DES and approaching 0.1% for VRO. I will comment on what I see as the most challenging systematic uncertainties that must be controlled to realize this statistical promise. Time permitting, we could discuss the surprising DES Y1 cluster cosmology results https://arxiv.org/abs/2002.11124.