31 August 2026 to 18 September 2026
Albano Building 3
Europe/Stockholm timezone

BayesLIM: End-to-end, differentiable, Bayesian forward models for joint 21cm signal, foreground, and instrumental parameter inference

9 Sept 2026, 15:20
25m
Albano Building 3

Albano Building 3

Hannes Alfvéns väg 12, 10691 Stockholm, Sweden

Speaker

Nicholas Kern (University of Michigan)

Description

Next-generation 21cm cosmology experiments promise to revolutionize our understanding of the high redshift universe; however, these experiments are overwhelmed by foregrounds many orders of magnitude brighter. This sets a precise specification on instrumental modeling requirements that have to date made direct detection of the 21cm at the EoR elusive. However, next-gen computing and inference paradigms such as auto-diff and GPU acceleration will make joint modeling of the 21cm signal, foregrounds, and systematics possible, even for large-scale interferometric datasets. I'll discuss the BayesLIM approach, the first end-to-end, differentiable Bayesian forward model for low-frequency LIM, and the path forward for cosmological end-to-end modeling at low frequencies. (Arxiv:2504.07090)

Presentation materials

There are no materials yet.