基于灰色关联分析的高阶神经网络剪枝算法

在综合分析网络纵向、横向灰色关联分析特点的基础上提出了一种新的基于灰色关联分析的剪枝算法,并将其用于训练高阶神经网络.该算法运用灰色关联分析对比网络各节点输出值序列之间联系的紧密程度,用网络纵向灰色关联分析确定剪枝连接,再用网络横向灰色关联分析确定相应的并枝连接,实现网络结构的动态修剪.训练后的高阶神经网络具有合理的网络拓扑结构和较好的泛化能力.实验验证了该算法的合理性、有效性.

Pruning algorithm training high-order neural network based on grey incidence analysis

A comprehensive analysis of the characteristics of the network lateral and vertical grey incidence analyses is given, and then a new pruning algorithm based on grey incidence analysis for high-order neural network is presented. In this algorithm, grey incidence analysis is used to compare the correlation degree of each output sequence of the network units. The pruned connections and the incorporated connections are determined by the network vertical and lateral grey incidence analyses, respectively. The topological structure of the network is adjusted dynamically. After training, the pruned network has the optimum topological structure and obtains good generalization. The simulation results show the rationality and effectiveness of the proposed approach.