Applications where Radial Basis and Probabilistic Neural Networks are successful respectively?

Can someone please explain the application areas of Radial Basis and Probabilistic Neural Networks? I mean How to identify where a particular network is successful? I am calculating feature vectors through different techniques. Some are giving results with RBFs while others with PNNs. I am not able to identify reasons for the same.

Thank you

[Geetika]

I am applying these networks for pattern recognition problems.

Answer 1

[Greg Heath]

Use RBFs. Like MLPs, under some conditions, they are universal approximators.

I consider PNNs to be a special case of an RBF.

MATLAB's version of an RBF has two nagging defaults.

 1. You cannot specify a starting configuration of hidden nodes.

 2. All of the hidden layer transfer function are spherical with the same specified radius.
    Some generalizations that could be incorporated using the proximity to other classes
    a. Different radii 
    b. Different coordinate aligned ellipses

Hope this helps.

Thank you for formally accepting my answer

Greg

[Geetika]

Thank you very much for your reply. But how can one determine under what situations RBFs or PNNs will be useful? I mean like I am applying these on feature vectors. So, What should be the characteristics of feature vectors to determine this?

[Greg Heath]

As I said above,

 1. RBFs are universal approximators. 
 2. I consider PNNs as a special case of RBFs.
 3.I have no use for PNNs.

HTH

Greg