Training
Radial Basis Neural Networks with the Extended Kalman Filter
Dan
Simon
332 Stilwell Hall
Department of Electrical Engineering
Cleveland State University
1960 East 24th Street
Cleveland, OH 44115
Radial Basis Function (RBF) neural networks are three-layer neural networks. Several ways have been proposed for training RBF networks. Recently, Professor Simon has proposed the use of Kalman filters for training RBF networks [1]. This web page makes available the classical Iris data that can be used to test RBF networks, along with various m-files that can be run in the MATLAB environment. M-files are written in a very high-level language that can be easily read, almost like pseudo code. The data files and m-files are contained in the following zip file.
RBFKalman.zip - 16 kilobytes
If you download RBFKalman.zip to your hard drive by clicking on the above link, then unzip the file (using, for example, WinZip), you can run RBF experiments and reproduce the results in reference [1]. These include results on gradient descent training, Kalman filter training, and decoupled Kalman filter training. When you unzip the file on your hard drive, look at the readme.txt file for more detailed information. If you don't have software to unzip the file, you can download a free evaluation version of WinZip from www.winzip.com.
References
Department of Electrical and Computer Engineering
Last Revised: