摘要
用ＲＢＦ网络为非线性映射建模，其学习算法对应于求解线性问题，因而学习速度快。然而在样本数据含有加性噪声的情况下，拟合函数会出现迅速振荡，使推广能力受到限制。过去曾提出过一些平滑算法。本文提出了一个新的平滑算法。实验结果表明，网络的推广能力有进一步的提高。
RBF neural networks provide a powerful technique to model nonlinearmapping.The learning algorithm for RBF networks corresponds to the solution of alinear problem, therefore fast. However, if the data are contaminated by additivenoise,the approximation function will oscillate rapidly. As a result,the generalizationproperties are restricted. Some smoothing method had been advanced in earner works.A new smoothing method is presental in this paper. Experimental results show thatthe generalization properties are improved,
出处
《华南理工大学学报（自然科学版）》
EI
CAS
CSCD
1995年第5期31-36,共6页
Journal of South China University of Technology(Natural Science Edition)
基金
国家自然科学基金
攀登计划资助
关键词
神经网络
RBF网络
推广能力
平滑算法
neural networks, non-linear, smoothing technique / RBF networks,generalization properties
