Lifted Marginal Filtering

Bayesian Filtering in discrete state spaces resulting from symbolic models (e.g. models of human behavior) is a challenging task because the state space space can grow very large due to redundancies and symmetries. For instance, in multiple person tracking by use of anonymous sensors or assisted manufacturing, the identity of different tools and parts is uncertain, but cannot be discarded completely. Lifted Marginal Filtering is a Bayesian Filtering algorithm that exploits these symmetries by representing symmetrical states by a single, parametric state representation. Intuitively, this representation groups together similar entities whose properties follow an exchangeable joint distribution. We introduced a filtering algorithm that works directly on lifted states, without resorting to the original, much larger ground representation.

We showed empirically that this lifted representation can lead to a factorial reduction in the representational complexity of the distribution, and in the approximate cases can lead to a lower variance of the estimate and a lower estimation error compared to the original, ground representation.

Contact

Stefan Lüdtke

stefan.luedtke2@uni-rostock.de