Particle filtering for high-dimensional systems

Particle filtering methods aim at tracking probability distributions sequentially in time. One of the main challenges of these methods is their accuracy in high-dimensional state spaces. Namely, it can be shown that if the dimensions of these spaces are sufficiently high, the obtained results by particle filtering are practically useless. In this paper, we propose an approach for addressing this problem. It is based on breaking the high-dimensional distribution of the complete state into smaller dimensional (marginalized) distributions and attempting to track these distributions in a novel way as accurately as possible. We demonstrate the proposed approach with computer simulations.