因子 K-均值聚类的一致性
统计理论
2014-02-14 v5 统计理论
摘要
因子 K-均值 (FKM) 聚类是一种在低维子空间中对对象进行聚类的方法。该方法的优势在于能够同时获得对象的划分以及反映聚类结构的低维子空间。本文推导了当样本量无限增加时，保证 FKM 聚类估计量几乎必然收敛的条件。该结果在一个包含 FKM 聚类的更一般模型中得到了证明。
引用
@article{arxiv.1301.0676,
title = {Strong Consistency of Factorial K-means Clustering},
author = {Yoshikazu Terada},
journal= {arXiv preprint arXiv:1301.0676},
year = {2014}
}
备注
A revised version of this was accepted in Annals of the Institute of Statistical Mathematics. Please refer to the accepted version of this. In the accepted ver., I describe a new interesting fact that there exists some cases in which reduced k-means clustering becomes equivalent to FKM clustering as n goes to infinity and provide a rough large deviation inequality for FKM clustering