We meet on Wednesdays at 1pm, in the 10th floor conference room of the Statistics Department, 1255 Amsterdam Ave, New York, NY.
Thursday, October 21, 2010
Micky Vidne: October 27th. s(MC)^2 or Hesitant Particle Filter.
In my talk I will describe a recent extension of the Sequential Monte Carlo (SMC) method. SMCs (particle filters) are a commonly used method to estimate a latent dynamical process from sequential noise-contaminated observations. SMCs are extremely powerful but suffer from sample impoverishment, a situation in which very few diﬀerent particles represent the distribution of interest. I will describe our attempt to circumvent this fundamental problem by adding an extra MCMC step in the SMC algorithm. I will illustrate the usefulness of this algorithm by considering a toy neuroscience example.
by michael at 10:16 PM
Labels: ergodicity, forward backward, group meeting, Kalman filter, latent gaussian, markov, MCMC, non-normal, nonlinear filtering, numeric methods, resampling