摘要
提出了一种新的训练前馈神经网络的局部线性化最小二乘算法。新算法将整个网络的权值训练看作非线性系统的参数识别,并将非线性系统的全局参数识别转化为一系列局部子系统的参数识别,最终将局部子系统的参数识别转化为线性化的最小二乘问题。
A new local linearized least squares algorithm for training feedforward neural networks(FNN)is presented in this paper. The new algorithm takes the weight training of FNN as parametric identification of nonlinear systems and converts the global parametric identification into parametric identification of a series of local subsystems and ultimately converts parametric identification of local subsystems into local linearized least squares problems.
出处
《计算机工程》
CAS
CSCD
北大核心
2002年第3期172-174,共3页
Computer Engineering
基金
国家教育部科学技术重点项目资助(2000.16b)