Re-constructing high reliable BP-model neural networks

Reliability or fault-tolerance is one of the most important properties of neural networks. In this paper, a method of re-constructing highly reliable BP-model neural networks and directly training them is submitted. The author used it in a three-layer BP-model formed for the exclusive OR(XOR) problem, the result indicates that not only the reliability of the re-constructed XOR-BP-Model is greatly developed but also its learning speed is increased to some extent, by re-assigning the corresponding weights. Furthermore, the computer aided analysis of the reliability-functional curves shows that this method can be used to construct reliable neural networks using less reliable neurons (or PEs) or components, which is both economic and beneficial.