Nonparametric Bayesian Modelling of Event Data
Gordon Ross (University of Edinburgh)
Friday 7th June 15:00-16:00 Maths 311B
Many physical processes produce event data which can be modelled using a point process framework. However, this data often contains features that can be tricky to model. We are concerned with data which has the following two features: 1) a high level of seasonality where the occurrence of events varies regularly over the course of a day, and 2) a substantial amount of burstiness/clustering where events tend to occur in groups. We develop a Bayesian formulation of the self-exciting Poisson process to capture such data, using a nonparametric Dirichlet Process to capture features such as seasonality and clustering in a data-driven manner.