Node decoupled extended Kalman filter based learning algorithm for neural networks

摘要

The use of extended Kalman filter (EKF) is common in estimation of nonlinear system parameters. It has also found application in training of feedforward neural networks. A heuristic modification of the EKF algorithm known as the node decoupled EKF (NDEKF) algorithm, which improves upon the EKF algorithm by significantly reducing computation time and memory requirements, appears very promising. The purpose of this paper is to present the NDEKF algorithm in a form suitable for coding readily into a computer program. Matlab implementation of the algorithm with simulation examples is included.