Factored Particles for Scalable Monitoring

Exact monitoring in dynamic Bayesian net­ works is intractable, so approximate algo­ rithms are necessary. This paper presents a new family of approximate monitoring al­ gorithms that combine the best qualities of the particle filtering and Boyen-Koller meth­ ods. Our algorithms maintain an approxi­ mate representation the belief state in the form of sets of factored particles, that cor­ respond to samples of clusters of state vari­ ables. Empirical results show that our al­ gorithms outperform both ordinary particle filtering and the Boyen-Koller algorithm on large systems.