Event Details

Sequential Monte Carlo as an alternative to reversible jump

Sequential Monte Carlo as an alternative to reversible jump

MCMC remains a popular approach for performing Bayesian model comparison. However, it is well known that the acceptance rate of dimension changing moves is often low, particularly in high dimensional models. In this talk we suggest an alternative approach based on sequential Monte Carlo that may be used in cases where reversible jump fails. Applications to mixture models and the coalescent are presented.

[This is joint work with Richard Culliford, Felipe Medina Aguayo and Daniel Wilson and a draft is available at arxiv.org/abs/1612.06468]