Homotopy Importance Sampler For Noisy Dynamics

We propose a Bayesian estimation method for moments of a state vector
that obeys stochastic nonlinear dynamics and its observations. The method
uses a homotopy procedure to improve the convergence of an MCMC importance
sampler. Designed to sample non-Gaussian statistics, the method can also
be used to sample very low uncertainty Gaussian statistics dynamics. In this
talk I will describe the method and show comparisons that suggest that
the method is efficient and comparatively accurate on a variety of practical
and challenging problems.