Particle filters for the magnetoencephalography inverse problem: increasing the efficiency through a semi-analytic approach (Rao–Blackwellization)
Alberto Sorrentino, Annalisa Pascarella, Cristina Campi, Michele Piana · Journal of Physics Conference Series · 2008
We consider the problem of dynamically estimating the parameters of point-like neural sources from magnetoencephalography data.Since the problem is non-linear, we apply the sequential Monte Carlo algorithms known as particle filters for solving the Bayesian filtering problem.We suggest that the linear dependence of the data on a subset of the parameters allows the analytic computation of the posterior density for these parameters, i.e.Rao-Blackwellization; this considerably improves the accuracy of the method and its statistical efficiency.