An Improvement on Extended Kalman Filter for Neural Network Training

Information overload has resulted in difficulties of managing and processing information. Reduction of data using well-defined techniques such as rough set may provide a means to overcome this problem. Extracting useful imformation and knowledge from data is a major concern in information science. Artificial intelligence systems, such as neural network systems, are widely used to extract and infer knowledge from databases. This study explored the training of a neural network inference system using the extended Kalman filter (EKF) learning algorithm. The inference accuracy, inference duration and training performance of this extended Kalman filter neural network were compared with the standard back-propagation algorithm and an improved version of the back-propagation neural network learning algorithm. It was discovered that the extended Kalman filter trained neural network required less

Text FSKTM_2005_5 IR.pdf Download (2MB)

Actions (login required)

View Item