摘要
次成分分析神经网络是一种自动迭代求取输入数据自相关矩阵的次成分方法,近十年来在国际上得到广泛深入的研究。本文将次成分学习算法归纳为普通发散、突然发散、动态发散、数值发散和自稳定特性等四种发散现象和一种特性来分析,并指出了该领域存在的问题和下一步发展趋势,为神经网络次成分分析理论奠定了理论基础。
The minor component analysis neural network is a method for adaptively extract the minor component of the autocorrelation matrix of the input data,which has been researched in the last decade. This paper analyzes and summarizes the minor component analysis learning algorithm as general diver- gence, sudden divergence, dynamic divergence, numerical divergence and self-stabilizing property, and points out the existing problems and the future development trend of this fields, and this work lays sound theoretical foundations for the neural MCA theory.
出处
《计算机工程与科学》
CSCD
北大核心
2013年第3期108-114,共7页
Computer Engineering & Science
基金
国家自然科学基金资助项目(61074072)
国家杰出青年基金资助项目(61025014)
关键词
神经网络
次成分分析
自稳定特性
neural network
minor component analysis
self-stabilizing property
