一种改进的神经网络相关性剪枝算法

摘要

神经网络的剪枝有利于网络结构的简化,而作为剪枝算法中的比较重要的相关性剪枝算法,在计算了隐层节点输出的线性相关性和方差后,对于如何根据线性相关值和方差值删除节点并没有给出明确的界限.文章通过研究神经网络的相关性剪枝算法,给出一种以网络的误差传递为思想,根据方差值删除节点的方法,并通过实验证明,该方法不仅能够有效的简化网络结构,保证网络精度,而且计算简单.%Pruning networks helps researcher get simple network structure. As one of the most important pruning methods,rnrelevance pruning networks method has not given a standard to delete the hidden notes by notes' variance and output relations.rnThis paper tries to solve this problem by giving a simple method which deletes hidden notes according to notes' variance byrnerror transfer. The experiment proves that using the method can get a simple network with high precision and it is easy torncalculate.