Conservative data fusion for decentralized networks

The paper investigates a technique for computing conservative data fusion for Gaussian mixture model (GMM) in decentralized networks with any topology. The main advantage of conservative solutions is that they do not deteriorate the performance of a sensor network in presence of any kind of correlations. The paper exploits normalize geometric mean for computing conservative data fusion. It computes normalized geometric mean by Newton generalized binomial theorem and Monte Carlo technique. It is shown that the solution by Newton's generalized binomial theorem exhibits divergence and numerical instability. On the other hand, Monte Carlo technique offers conservative solution. The tradeoffs are that it requires considerable computational time and is expensive as large numbers of samples are required to get statistical accuracy.