摘要
泛函网络是神经网络的一般化推广,至今还没有统一的系统设计方法能够对给定问题设计出近似最优的结构。为了获得良好的网络结构,本文利用熵聚类的思想,提出一基于熵聚类思想的设计泛函网络的方法,对网络每一神经元的共存且相互影响的基函数和泛函参数进行最优搜索,实现泛函网络结构和泛函参数的共同学习。对一非线性函数进行逼近比较仿真实验,结果表明,逼近效果较好,且收敛速度较快,并表明所设计的泛函网络有效地提高了泛函网络的收敛精度,还可获得更为合理的网络结构。
Functional network introduced recently is an extension of neural networks.Up to the present,there is no general system designing method for designing approximation functional networks structure.In this paper,based on an entropy clustering idea,a novel entropy clustering method for designing functional network is proposed.This method can get the base function and its parameters with optimal searching,achieving the learning between functional network structure and the functional parameters.For a nonlinear function,a comparing simulation experiment is designed,the effect of approximation is better,and convergent speed is also faster.The simulation results show that the proposed method in this paper can produce very rational structure and functional networks convergent precision is improved greatly.
出处
《计算机仿真》
CSCD
北大核心
2011年第2期200-203,共4页
Computer Simulation
关键词
泛函网络
熵聚类
神经元函数
Functional network
Entropy clustering
Neuron function
