Abstract:

The process neural networks (PNNs) are networks that adapt to the process of signal input , whose elementary unit is the process neuron (PN) , an emerging neuron model. Both essential difference and close correlation exist between the process neuron and the traditional neurons , for example , PN can be approximated by traditional neurons with arbitrarγpreclslOn. In this paper , the PN model and some PNNs are introduced. Then , two PN
approximating theorems are presented and proved in detail. Each theorem gives an approximating model to the PN model , i. e. , the time-domain feature expansion model and the orthogonal decomposition feature expansion model. Moreover , a corollarγis given for the real-valued output PNN based on the second theorem. Mterwards , a simulation of analog signals is carried out , showing that the PNN can well suppress the white noises contained in signals. Finally , some problems about PNNs are discussed and further research orientations are suggested.

Key words: artificial neural network, process neuron, simulation, function orthogonal basis, Fourier series, feature expanslOn