On Concentration Bounds for Bayesian Identification of Linear Non-Gaussian Systems

We adopt a Bayesian perspective to identify the unknown parameters of linear stochastic systems with possibly non-Gaussian disturbance distributions. The key idea of our algorithm is to alternately execute
$L$
randomly selected linear state-feedback controllers and keep track of a maximum a posteriori estimator. The proposed algorithm asymptotically achieves the concentration of posterior distributions around the true system parameters. We also derive probabilistic bounds for the concentration based on the classical results regarding the asymptotic properties of posterior distributions. An empirical demonstration is provided as well.