Several parallel neural network (PNN) architectures are presented in this paper. PNNs can work parallelly and coordinately. The implementation of their training is much easier than that of a single NN. And there are many other attractive characteristics of PNNs such as a modular structure, easy implementation by hardware, high efficiency for their parallel structures (compared with sequential NN architectures), easy implementation of additional learning, etc. PNNs can be used to deal with such problems as data processing, pattern recognition, and classification. The learning and additional learning algorithms for PNNs are presented in this paper. Some simulation results are given to illustrate the advantages of all the PNNs considered.
展开▼