Policy Optimization of Finite-Horizon Kalman Filter with Unknown Noise Covariance

This paper is on learning the Kalman gain by policy optimization method. Firstly, we reformulate the finite-horizon Kalman filter as a policy optimization problem of the dual system. Secondly, we obtain the global linear convergence of exact gradient descent method in the setting of known parameters. Thirdly, the gradient estimation and stochastic gradient descent method are proposed to solve the policy optimization problem, and further the global linear convergence and sample complexity of stochastic gradient descent are provided for the setting of unknown noise covariance matrices and known model parameters.