Head of Discipline of Statistics, School of Computer Science and Statistics, Trinity College Dublin, Ireland
I.1.09 (I block, 1st floor)
Source separation is one of the initial data analysis tasks for multi-channel image data, particularly in astronomy where satellites such as COBE, WMAP and more recently Planck have obtained multi-channel images at microwave frequencies. In this work, we look at a factor analysis approach to source separation. First I discuss how prior information, available from understanding of the physics of the sources, can be incorporated into the analysis. Priors for the images of each source are modelled as Gaussian Markov random fields. Then I show that an analytical approximation to the posterior is possible which allows for practical separation of large images; Planck data consists of 9 images at different microwave frequencies, each of about 10 million pixels. The work is applied to reconstruct the Cosmic Microwave Background (CMB) signal from satellite observation, by separating it from other sources, using WMAP 7 year data. The performance and limitations of the approximation are also discussed.
About the speaker:
Simon Wilson's research interests are centered around the analysis of large amounts of data with complex structures via the Bayesian approach to statistical inference. He heads the STATICA group at Trinity College Dublin, which aims to develop new methods of statistical analysis for such diverse applications as reliability, telecommunications, astronomy and ecology. Research is focused on the development of efficient algorithms for timely analysis.