CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: Astronomical source detection and background separation via hierarchical Bayesian nonparametric mixtures Authors:  Andrea Sottosanti - University of Padua (Italy) [presenting]
Mauro Bernardi - University of Padua (Italy)
Roberto Trotta - Imperial College London (United Kingdom)
David van Dyk - Imperial College London (United Kingdom)
Abstract: The search of gamma-ray sources in the extra-galactic space is one of the main targets of the Fermi telescope project, which aims to identify and study the nature of high energy phenomena in the universe. This requires to separate their signal from a gamma-ray background component diffuse over the entire area observed by the telescope. From a statistical perspective, we can account for both these phenomena using a mixture of two densities: the first models the spread of photons around the sources, while the second includes the information from the background contamination. We propose a novel approach to the signal extraction of gamma-ray sources using a Dirichlet process mixture, that allows to discover and locate a possible infinite number of clusters in the map, and a new flexible Bayesian nonparametric model based on b-spline basis functions to account for the irregular shape of the background. The resultant is then a mixture of two Dirichlet process mixture models. From the results obtained on a region of the Fermi map we can conclude that the proposed approach both guarantees a posterior estimation of the number of sources and a complete separation of their signals from the background noise.