Abstract

The particle filter is a recursive, importance sampling-based scheme which yields weighted sample approximations to a sequence of posterior probability distributions on spaces of increasing dimension. Sequential Monte Carlo Samplers are a very general class of algorithms which enable the same task to be performed for sequences of distributions which need not be defined on spaces of increasing dimension.

This talk will summarise importance sampling, the particle filter and go on to describe the Sequential Monte Carlo Samplers framework, pointing out relationships to other existing inference alogrithms. A sketch will be given of an application to an audio signal processing problem involving trans-dimensional state spaces.

“Sequential Monte Carlo Samplers”, (with P. Del Moral & A. Jasra), /J. Royal Statist. Soc. /B, vol. 68, no. 3, pp. 411-436, 2006.

http://www.cs.ubc.ca/~arnaud/delmoral_doucet_jasra_sequentialmontecarlosamplersJRSSB.pdf