26 Current Trends in Bayesian Methodology with Applications
therefore, allowed the locations as well as the dipo le moments of existing
dipoles to evolve according to random walk dynamics (such approaches date
back at least as far as [6]); we term this model the random walk model. More
recent work has demonstrated that it is possible, with care , to implement par-
ticle filters for a model in which dipole locations are not permitted to move
over time; we term this the static model.
1.5.3.2 Methodology
The approach developed in [64] makes use of a number of techniques to obtain
substantially better performance than a simple particle filter. These changes
are essential to obtain adequate performa nce with the static model. In par-
ticular, it attempts