Sequential Monte Carlo for Complex Models

Abstract:

Sequential Monte Carlo (SMC) methods are applicable across a wide range of non-linear, non-Gaussian state-space models. They are particularly useful for complex models where the prior can be simulated, but not evaluated pointwise. Such situations usually preclude the use of other methods such as Markov chain Monte Carlo (MCMC) or Hamiltonian Monte Carlo (HMC). This talk will review a few techniques for improving the performance of SMC methods for complex models, such as the disturbance state-space model formulation, and the bridge particle filter. These will be demonstrated with some examples drawn from finance, epidemiology, and marine biogeochemistry.