Astrostatistics Meeting

Bayesian source separation at scale, applied to reconstructing CMB

​I look at implementing a fully Bayesian source separation algorithm with application to separating the cosmic microwave background. By use of a Gaussian approximation and a partitioning of the images, we achieve a reconstruction of the posterior mean and variance of the sources in a reasonable amount of computing time.  The method also gives posterior distributions of system hyperparameters such as source spectral indices.  I’ll discuss the outstanding issues in the work.