基于单位分解的神经网络构造

Structure of Neural Network Based on Partition of Unity

针对三层BP(Back Propagation,反向转播)神经网络,提出了隐层神经元的底层"状态域"的概念,通过利用"状态域"和单位分解方法构造了一种新的人工神经网络结构。当网络有输入时,根据输入处于状态域中的位置来激活对应的神经元,整个网络的输出只与所激活的那部分神经元有关,故在进行训练时就不必更新所有的连接权值,因此能大大降低更新权值的维度和学习复杂度,提高学习质量。

In this paper,the concept of the state domain of the underlying layer of the hidden layer neuron is presented for the three-layer BP(Back Propagation)neural network.This paper proposes a new artificial neural network structure by using the concept of the state domain and the partition of unity method.when the input values are given into network,the corresponding neurons are activated according to the position of the input in the state domain,so that the output of the entire network is only related to the activated neurons.