A0277
Title: Variational inference for a Bayesian nonparametric model with structural breaks
Authors: Yong Song - University of Melbourne (Australia) [presenting]
John Maheu - McMaster University (Canada)
Xuan Vu - University of Melbourne (Austria)
Abstract: A variational inference (VI) algorithm is proposed for a dynamic Bayesian nonparametric framework, in which the standard Markov chain Monte Carlo (MCMC) method is impractical for large data sets due to its high computational cost. Because this dynamic nonparametric framework includes parameter structural breaks, the conditional analytic solution from the exponential family cannot be applied. Instead, a novel structured VI through Rao-Blackwellisation is proposed. Through both the simulation study and the application to the US banking data, the usefulness of the new method is shown.
