Gibbs sampling and Gaussian constrained realisations

by Dr Phil Bull

Wednesday 5 Dec 2018, 16:30 → 17:30 GB

Universe

Cosmological observations often involve making maps of faint, inherently random fields that have been filtered through complex instrumental responses. This naturally leads to very high dimensional inference problems. I discuss the Gibbs sampling method, which makes it possible to sample from high-dimensional posteriors by breaking them up into chains of more tractable conditional distributions, and show how it can be applied to solve missing data problems that would otherwise severely limit the dynamic range of cosmological observations.