Fast computation of particle filters

Miklos Rasonyi; UoE maths dept

9th June, 2010 at 1.00 pm - Sanderson Room

Joint UoE/HWU videoconference seminar

Particle filters are used to effectively compute the expectation of complex functionals of a stochastic process, e.g. when filtering a signal from noisy observations. These algorithms are a priori not parallelizable (though there have been such attempts). We present a modification which could run on a rectangular processor array and, using parallelism and local communication, computes these filters quickly and accurately (as test runs indicate).