Abstract
We apply the modified extended Kalman filter to develop a learning algorithm for feedforward networks. The resulting algorithm improves the existing extended Kalman filter based scheme in both speed of convergence and accuracy of estimation.