Kalman Filter Neuron Training

Abstract

An attempt of implementing Kalman filter algorithm in the procedure for training the neural network was made and evaluated. The Kalman filter neuron training program (KNT) was coded. The performance of Kalman filter in KNT was compared to commonly used neuron training algorithm. The study revealed that KNT requires much less calculation time to accomplish neuron training than commonly used other algorithms do. KNT also gave much smaller final error than any other algorithms tested in this study.