摘要
本文提出了一种感知器的动态稀疏化（ｄｙｎａｍｉｃｄｉｌｕｔｉｏｎ）概念，同时估计权值和减少神经元间的连接权个数．动态稀疏化有效地克服了传统的静态稀疏化（先确定权值，然后减少连接权个数）的缺陷．计算机实验结果说明了算法的优越性．
In this paper, a dynamic dilution concept for perceptrons is proposed, which estimates the weights and reduces the number of connections at the same time. The dynamic dilution overcomes the weakness of the static dilution. Computer simulations are conducted to show its advantages.
出处
《自动化学报》
EI
CSCD
北大核心
1995年第1期93-98,共6页
Acta Automatica Sinica
关键词
感知器
学习算法
动态稀疏化
神经网络
Perceptron, learning algorithm, dynamic dilution,static dilution.
