I need to understand exactly in RBF neural network in training phase the RBF centers will determine using k-means clustering. The RBF widths are usually all fixed to the same value which is proportional to the maximum distance between the chosen centers. the question is which center the center of each cluster that adapted by k-means clustering or the neuron center?
.
http://mccormickml.com/2013/08/15/radial-basis-function-network-rbfn-tutorial/
is a nice introduction to RBF NN
Regarding the width :
"If you use k-means clustering to select your prototypes, then one simple method for specifying the beta coefficients is to set sigma equal to the average distance between all points in the cluster and the cluster center."
(so you end up with different widths for different units)
.
Another simpler choice is to set a single width for all units, either from the maximum distance between centers or from their mean distance (centers = centers given by the k-means or by any other initialization process - randomly picking k training patterns for instance) ; see http://www.cs.bham.ac.uk/~jxb/INC/l14.pdf
.
many different approaches to RBF NN training are presented here :
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.109.312&rep=rep1&type=pdf
.
- As stated by Clerot, another approach that has flexability is to combine the RBF as a mapping and use a standard 1- or 2- hidden layer Back Prop (BP). Here, the required nonlinearity comes from the RBF and BP does the minimizing of the decision surface.
- Badges
- Science topic