Matlab Functions for Radial Basis Function Networks (1999)
(Version 2.2) Implements a variety of methods based on subset selection and ridge regression to control
model complexity and regression trees to generate RBF centres and radii. Includes a
comprehensive manual, demos of the methods and examples of how to run them.
Matlab Routines for Linear Neural Networks (1996)
Older version of the software package. Implements methods based on subset selection
and ridge regression (local and global versions). No longer supported.
Assessing RBF Networks Using DELVE
Submitted to IJNS, July 1999. Accepted, February 2000.
Combining Regression Trees and RBFs
Submitted to IJNS, July 1999. Accepted, September 2000.
Optimising the Widths of RBFs
Fifth Brazilian Symposium on Neural Networks,
Belo Horizonte, Brazil, 1998
An EM Algorithm for Regularised RBF Networks
International Conference on Neural Networks and Brain,
Beijing, China, 1998
Regularisation in the Selection of RBF Centres
Neural Computation, 7(3):606-623, 1995
Local Smoothing of RBF Networks
International Symposium on Artificial Neural Networks,
Hsinchu, Taiwan, 1995
Recent Advances in Radial Basis Function Networks (1999)
Covers developments from 1996 to 1999: maximum marginal likelihood, optimisation of RBF size
and regression trees to generate RBFs centres and radii.
Introduction to Radial Basis Function Networks (1996)
Describes the theory behind forward selection, ridge regression and model selection
criteria in the context of radial basis function networks for nonparametric regression.
Business Time in the Foreign Exchange Markets
An RBF network is used to model "business time" in various currency markets,
as world wide activity varies in response to such events as lunchtime in Tokyo.
Extrapolating Uncertain Bond Yield Predictions
An RBF network plays a role in extrapolating short and medium term predictions
of return into an entire yield curve for several types of government bonds.
The refmix Data Set
A small data set provided by a chemical engineering company is modelled
by an RBF network. The results are compared to polynomial models.
The Matlab software is available in two formats:
.zip for Windows and .tar.gz for Unix.
If you're on Windows and have WinZip
then it doesn't matter which you download - it can handle either,
while if you're on Unix gunzip
can decompress .gz and
tar will do the rest.
All text files (primarily Matlab m-files) have Unix style newlines which
Matlab under Windows is quite happy with (but some text editors are not).
If you don't already have one, you can obtain a free
for the software manual, technical reports and papers.
All PostScript files are available raw (.ps)
or compressed with gzip
Here is a complete list of titles, file names and approximate sizes in KB.