Inductive and non-inductive methods of clustering
Inductive and non-inductive methods of clustering
复制标题
DOI：
10.1109/grc.2012.6468710
复制
发表时间：
2012-08
期刊：
影响因子：
--
通讯作者：
S. Miyamoto
中科院分区：
文献类型：
--
作者：
S. Miyamoto
本文通过引入归纳聚类和非归纳聚类的概念，综述了各种聚类方法。这些概念与半监督分类研究中的归纳学习和转导学习的概念是平行的。当聚类结果自然地在整个感兴趣空间上归纳出分类函数时，这种方法称为归纳聚类。相反，如果一种方法不能诱导出这样的函数，它就被称为非归纳法。归纳类中的典型例子是清晰和模糊的c-均值，而非归纳类中的典型例子是凝聚层次聚类。我们展示了这两类聚类方法在聚类算法的理论考虑上是如何产生差异的，特别是当使用正定核函数时，这两个概念被明显地对比。此外，还考虑了对这两类进行半监督分类。
This paper aims to overview a variety of methods of clustering by introducing the concepts of inductive and non-inductive clustering. These concepts are in parallel with the concepts of inductive and transductive learning in the studies of semi-supervised classification. When the result of clustering naturally induces functions for classification on the whole space of interest, the method is called that of inductive clustering. In contrast, a method is called non-inductive, if it does not induce such a function. Typical examples in the inductive class are crisp and fuzzy c-means, while one of the non-inductive class is agglomerative hierarchical clustering. We show how differences of the two classes of methods of clustering occur in the theoretical consideration of clustering algorithms, in particular two concepts are clearly contrasted when positive-definite kernel functions are employed. Moreover semi-supervised classification is considered for the two classes.