IMPLEMENTING PROBABILISTIC NEURAL NETWORKS

A modified PNN training algorithm is proposed. The standard PNN, thoug h requiring a very short training time, when implemented in hardware e xhibits the drawbacks of being costly in terms of classification time and of requiring an unlimited number of units. The proposed modificati on overcomes the latter drawback by introducing an elimination criteri on to avoid the storage of unnecessary patterns. The distortion in the density estimation introduced by this criterion is compensated for by a cross-validation procedure to adapt the network parameters. The pre sent paper deals with a specific real-world application, i.e. handwrit ten character classification. The proposed algorithm makes is possible to realise the PNN in hardware and, at the same time, compensates for some inadequacies arising from the theoretical basis of the PNN, whic h does not perform well with small training sets.