Probabilistic neural network vs MLP with softmax

Hi all,

I was wondering if anybody has any experience with PNN (Probabilistic neural network) http://en.wikipedia.org/wiki/Probabilistic_neural_network to classify vs an MLP with back-propagation and softmax output layer ?

Further more can someone point me to the fist use of using an MLP for classification ? If possible a paper. Was this using a soft-max, if not when was soft-max introduced in MLP ? any papers or review papers that cite the origin of MLP with soft-max.

Thanks