Statistics and Biostatistics seminar series
Room: M3 3127
Annealed sequential Monte Carlo method with non-standard applications
In this talk, I will describe an "embarrassingly parallel'' method for Bayesian inference, annealed Sequential Monte Carlo (ASMC) with an adaptive determination of annealing parameters. The ASMC method efficiently provides an approximate posterior distribution and an unbiased estimator for the marginal likelihood. This unbiasedness property can be used to test the correctness of posterior simulation. We have adapted the annealed SMC method to two non-standard applications: 1) Bayesian inference of phylogenetic trees and evolutionary parameters from biological sequence data; 2) Estimation of parameters in nonlinear ordinary differential equations and model selection.
We illustrate our method by comparing it with other methods such as standard Markov chain Monte Carlo algorithms using simulation studies and real data analysis.