关于前向神经网络的设计问题

前向神经网络的设计问题，一直是神经网络的理论研究中的难点和热点，虽经许多学者的潜心研究，但到目前为止并未得到满意的解决．本文利用Ｍ－Ｐ神经元模型的几何意义，将前向神经网络的设计问题化为“复盖问题”求解，从而为前向神经网络的设计提供了一个比较满意的解决方法．最后给出几个模拟的例子，以资证方法的有效性．

An Principle for Design of Multi-layer Neural Networks

In the paper, a geometrical representation of M-P neural model is presented. From the representation, a clear visual picture and interpretation of the model can be seen. Based on the interpretation a new principle for design a sort of multi-layer weight -sum-and -threshold neural networks is presented. Finilly, The simulation results of three representative hard classification problems are shown. These simulation results are given to illustrate the effectiveness of the new method.