一种改进的多层前馈神经网络结构研究

Study on an Improved Structure of Multi-Layer Feed-Forward Neural Network

通过将传统多层前馈神经网络结构改造为具有跨层连接的网络,提出了相应的多层前馈神经网络改进算法。通过分析网络误差函数,从理论上证明了:相对于无跨层连接网络,有跨层连接网络能以更加简洁的结构逼近理想状态。最后,用一个隐层神经元解决了传统网络必须用两个隐元才能解决的异或问题。

A transformation for the traditional multi-layer forward feed neural network structure is made to the cross level connection network, proposed the corresponding multi-layered forward feed neural network algorithm improved. The analysis network error function theoretically has proven;in the not cross level connection network, has the cross level connection network to be able to approach the perfect condition by a more succinct structure. Finally, it is solved that the XOR problem with one neuron not two.