A new multilayer feedforward small-world neural network with its performances on function approximation

摘要

In this paper, by the use of the research results from complex network, a new multilayer feedforward small-world neural network is presented. Firstly, based on the construction ideology of Watts-Strogatz network model and community structure, a new multilayer feedforward small-world neural network is built up, which heavily relies on the rewiring probability. Secondly, the network model is briefly described by mathematical method. Finally, in order to investigate the performances of new small-world neural network, function approximation and fault tolerance are used to test the network performances. Simulation results show that the new neural network has the best approximate performance when the rewiring probability is nearby 0.1, and the approximate speed comparison also shows that small-world neural network is superior to regular network and random network at this time.