Abstract
During the actual application of ANN (artificial neural network), the sample functions needed to be learned often present subsection attributes. It is very difficult to simulate such kinds of sample functions only with single structure ANNs. We are inspired by the running principle of human's intelligent activities which can only be accomplished with the support of different brain function subareas. The multiple subareas cooperate with each other, and then the high quality intelligent activities of humans can be performed. In this paper a kind of largescale ANN, with heterogeneous function subnets, is proposed in order to overcome the shortages of traditional ANNs. At last, the capability of the largescale ANN is evaluated with a typical example, and its performances are also contrasted with traditional ANNs

This publication has 11 references indexed in Scilit: