学习矢量量化算法的性能分析

本文针对学习矢量量化算法Ⅰ型和Ⅱ型（ＬＶＱ１、ＬＶＱ２）进行了深入的分析研究，取得了以下结果：（１）对应于ＬＶＱ１算法，提出了一种选取学习步长的优化方案；（２）得到了ＬＶＱ１算法关于线性可分模式分类问题的一个重要结论；（３）严格推导出ＬＶＱ２算法实质上是一种使分类错误减小的梯度下降法；（４）得出了ＬＶＱ２算法对于类间混叠的模式分类问题不存在稳定平衡状态这一重要结论；（５）针对ＬＶＱ２算法处理类间混

The Performance Analysis of Learning Vector Quantization Algorithm

Learning Vector Quantization algorithm of type I and type 2(LVQ1,LVQ2) are analyzed thoroughly,and the achievements obtained can be concluded as the following aspects;(1) an optimized scheme of selecting learning step size for LVQ1 is proposed;(2)a significant conclusion is obtained that is concerned with LVQ1 algorithm for the linearly-separable pattern classification problem;(3)LVQ2 algorithm is derived rigorously as a gradient descent method to minimize the classification error;(4)another significant conclusion is achieved that any stable equilibrium state does not exist for LVQ2 algorithm dealing with the overlapping classification problem;(5) an effective modified LVQ2 algorithm is developed to overcome the defect of LVQ2 algorithm for overlapping classification.