多重最小成分的神经网络算法及其在曲线拟合中的应用  

Unsupervised Learning Algorithm for Space-Curve Fitting Based on Multiple Minor Components Analysis

作　　者：董天信[1] 姜玉英[1] 徐宗本[1] 

机构地区：[1]西安交通大学理学院,西安710049

出　　处：《西安交通大学学报》2003年第2期205-207,共3页Journal of Xi'an Jiaotong University

摘　　要：讨论了空间曲线的拟合问题与多重最小成分抽取问题的等价性,提出了多重最小成分抽取的无导师、自适应神经网络学习算法,该算法为Oja算法的一种改进,没有任何附加条件.数学分析的结果表明,该算法是一种高效、可靠的多重最小成分的抽取方法,特别可用于空间曲线的拟合.The problem of spacecurve fitting is shown to be equivalent to the extraction of multiple minor components, and can be solved efficiently by a neural network with an unsupervised adaptive learning procedure. The algorithm can be considered as an improvment of the Oja's one without any additional conditions.The suggested learning rule is mathematically analyzed. The algorithm provides a very reliable and efficient method for minor component analysis especially for spacecurve fitting. 

关 键 词：神经网络 最小成分分析 空间曲线拟合 自适应学习算法 最小成分抽取 Oja算法 

分 类 号：O212.4[理学—概率论与数理统计] TP183[理学—数学]