Neural learning of Kalman filtering, Kalman control, and system identification

This paper shows how to implement Kalman estimation (including filtering and prediction) and control, and system identification, within a neural network (NN) whose only input is a stream of noisy measurement data. The operation of the fully-integrated algorithm is illustrated by a numerical example. The resulting network is a multilayer recurrent NN that may be useful for engineering applications. The algorithm is found to impose constraints on the NN circuitry and architecture. It is of interest that the derived circuit bears certain resemblances to the putative dasialocal circuitpsila of mammalian cerebral cortex. These similarities are discussed with reference to speculations on the possible fundamental operations of cerebral cortex.