论文部分内容阅读
本文给出一种新颖的神经元网络结构学习算法,用其训练分类神经元网络,可以清除冗余的(?)层节点,获得相对于当前分类问题的神经元网络骨架结构,从而降低了计算复杂性,更适用于简单的并行电路的实现,模拟实验表明,用结构学习算法训练的分类神经元网络可以清除初始结构中的冗余节点,又不降低原来网络的分类功能,实验结果是令人满意的。
In this paper, we present a novel neural network structure learning algorithm, which can be used to train neural networks to classify redundant nodes, and to obtain the neural network skeleton structure relative to the current classification, thus reducing the computational cost The complexity is more suitable for the realization of simple parallel circuits. The simulation results show that the neural network trained by structural learning algorithm can eliminate the redundant nodes in the initial structure without reducing the classification function of the original network. The experimental result is that Satisfied.