Fast Bayesian inference for poisson processes with changing rate parameters

Datum

11.09.2012

Zeit

13:30 - 14:00

Sprecher

Florian Stimberg

Zugehörigkeit

TU Berlin

Sprache

en

Hauptthema

Physik

Andere Themen

Physik

Host

SIMPLE12 seminar

Beschreibung

Poisson processes can be used to model event-time data e.g. occurrences of sequence motifs in DNA or neural spike trains. Most interesting are models where the rate of the Poisson process can change. A widely used variant are Markov modulated Poisson processes where the rate can jump between a fixed number of states. We propose an efficient MCMC sampler for Bayesian inference in this model and an extension where the number of distinct rate values is unknown. Our sampler's performance depends only minimally on the number of data points and shows a vast improvement over current algorithms for data sets where the number of events is high compared to the number of jumps in the rate function.

Letztmalig verändert: 11.09.2012, 09:36:30