Ensemble Learning Online Filtering in Wireless Sensor Networks

In many applications, the observed system is assumed to evolve according to a probabilistic state space model. The data likelihood function is, in general, non linear or/and non Gaussian leading to analytically intractable inference. Particle filter is a popular approximate Monte Carlo solution based on a particle representation of the filtering distribution. However, power constraints in sensor networks require an additional approximation (compression) when communicating the particle based representation. In this contribution, we propose an alternative ensemble learning (variational) approximation suitable to the communication constraints of sensor networks. The efficiency of the variational approximation relies on the fact that the online update of the filtering distribution and its compression are simultaneously performed. In addition, the variational approach has the nice property to be parameterization-independent ensuring the robustness of the data processing. The selection of the leader node is based on a trade-off between communication constraints and information content relevance of measured data