摘要
提出了一种基于误差的径向基神经网络竞争学习法,它以网络的输出误差为度量,通过竞争调节神经元中心,RLS算法训练网络的权值,并利用IPL算法判断网络神经元的冗余性。仿真结果表明,该算法提高了网络的输出精度,简化了网络结构,其运算速度也较快。
This paper presents a new error-based learning algorithm for radial basis function neural network. Through competitive learning, the algorithm adjusts the center of each network hidden unit firstly. It uses the RLS(regularized least squares)algorithm to train the weight vector of the network secondly. At last the redundance of the network can be reduced by the IPL(incremental projection learning).Through simulation, the algorithm proves to be able to enhance the precision of the network and simplify the structure of network. Speed of the new algorithm is faster than former ones.
出处
《计算机工程》
CAS
CSCD
北大核心
2003年第17期126-127,F003,共3页
Computer Engineering
关键词
径向基神经网络
惩罚竞争学习法
RLS算法
IPL算法
Radial basis function neural network
Rival penalized competitive learning
Regularized least squares
Incremental projection learning