摘要
本文首先提出了一种实现连续映射的三层感知器的构造方法,该方法对陷入局部极小点的多层感知器,可以补偿到全局最小点。然后,给出了一种可对上述三层感知器进行压缩的隐节点的压缩方法,它可以推广到三层以上的感知器和节点激发函数不同的情形。
This paper gives a simple and practical construction method of realizing continuous mapping by a three-layer perceptron, which leads to compensate for the incompletely trained multilayer perceptrons to the global minima. Then, a method is presented to compress the number of hidden nodes, which can be extended to more than three layer perceptrons and to the case of using different activation functions.
出处
《系统工程理论方法应用》
1993年第4期3-8,共6页
Systems Engineering Theory·Methodology·Applications
基金
中国科学院管理
决策与信息系统开放研究实验室
关键词
连续映射
多层感知器
构造
Continuous Mapping, Multilayer Perceptrons, Construction, Hidden Nodes
