关于过程神经元网络的理论探讨

Theoretical discussion on the process neural network theory

过程神经元网络是一种适合于处理过程式信号输入的网络,其基本单元是过程神经元——新近出现的神经元模型.过程神经元和传统神经元有着本质的区别,但二者之间也存在着紧密的联系,前者可用后者以任意精度无限逼近.文中给出2个定理及其详细证明,分别论述了过程神经元的2种传统神经元逼近模型:时域特征扩展模型和正交分解特征扩展模型.基于第2个定理,给出了过程神经元网络相关的2个推论.最后,针对过程神经元网络面临的主要问题进行讨论,指出了一些具有前景的研究方向.文中得到的结果对过程神经元模型及其网络的研究具有一定的理论意义.

Process neural networks （PNNs） are networks suitable for processing signal input, whose elementary unit is the process neuron, a newly developed neuron model. The process neuron is different from traditional neurons in nature, but there is an inherent relationship between them. The former can be infi- nitely approached by the latter with arbitrary precision. Two theorems are presented and proved in this paper, giving two models for approaching corresponding process neurons：the time-domain feature expansion model and the orthogonal decomposition feature expansion model. And two corollaries are given based on the second theorem. Finally, some problems with PNNs are discussed and several research topics suggested. The conclusions are significant to theoretical research on process neurons and PNNs.