AS Seminar: Kalman filtering with unknown covariance matrices

The state-of-the-art algorithms for Kalman filtering in agent networks with information diffusion mostly assume that both the process and measurement noise covariance matrices are known. In the seminar, I'll present an alternative Bayesian message passing procedure removing this requirement. The ignorance of the process noise covariance matrix is overcome by a cheap hypotheses-testing procedure and an intrinsic optimization of a relevant prior distribution. The solution has a very low number of tuning parameters, and the simulation results demonstrate that the estimation performance is close to the diffusion Kalman filter.