Seminar series
Date
Tue, 13 May 2014
13:00 -
Wed, 14 May 2014
14:00
Location
C4
Speaker
Yves-Lauren Kom Samo
Organisation
University of Oxford
Cox processes arise as a natural extension of inhomogeneous Poisson Processes, when the intensity function itself is taken to be stochastic. In multiple applications one is often concerned with characterizing the posterior distribution over the intensity process (given some observed data). Markov Chain Monte Carlo methods have historically been successful at such tasks. However, direct methods are doubly intractable, especially when the intensity process takes values in a space of continuous functions.
In this talk I'll be presenting a method to overcome this intractability that is based on the idea of "thinning" and that does not resort to approximations.