子空间聚类算法的研究新进展

高维数据聚类是聚类技术的难点和重点,子空间聚类是实现高维数据集聚类的有效途径,它是在高维数据空间中对传统聚类算法的一种扩展,其思想是将搜索局部化在相关维中进行.该文从不同的搜索策略即自顶向下策略和自底向上策略两个方面对子空间聚类算法的思想进行了介绍,对近几年提出的子空间聚类算法作了综述,从算法所需参数、算法对参数的敏感度、算法的可伸缩性以及算法发现聚类的形状等多个方面对典型的子空间聚类算法进行了比较分析,对子空间聚类算法面临的挑战和未来的发展趋势进行了讨论.

Research and Advances of Subspace Clustering

The clustering of high dimensional data is a key problem in clustering methods. Subspace clustering is an effective approach to realize clustering in high dimensional data. It is an extension of traditional clustering that seeks to find clusters in different subspaces within a high dimensional dataset and it localizes the search for relevant dimensions. In Ibis paper, the ideas of subspace clustering are introduced from two different search strategies such as top -down approach and bottom - up approach. And the recent subspace clustering algorithms are reviewed. The comparison among the typical subspace clustering algorithms is made from some aspects which are the parameters of the algorithm, the sensitivity of parameters to algorithms, the scalability of the algorithm and the shape of the clustering. At last, the paper proposes the challenges of subspace clustering and discusses the trends in the future.