A TOP-DOWN SEARCH GRID BASED ALGORITHM FOR FAST SUBSPACE CLUSTERING
In this paper, a top-down search grid based algorithm is proposed to search all subspace that may contain clusters. Diffcrcnl from bottom-up search grid algorithms, the new approach starts from high-level subspace to low-level subspace, avoiding a lot of useless computation. Active spaces and grids are introduced to prune the search space, which reduces searching candidates dramatically. A new filter method based on active axis numbers is adopted to filter noise objects, since the noise is more serious in high-dimensional space. The advantages of the new approach are: it can discover clusters both on entire space and subspace; the computation complexity is proximate linear with objects number, space dimension, and clusters dimension respectively; it is not sensitive to noise; it can find both disjoint clusters or overlap clusters; it can find clusters of arbitrary shape; it is also able to find any number of clusters in any number of dimensions and the number is not predetermined by a parameter.
Subspace clustering High-dimensional top-down active space active grid
QIANG ZHANG XI CHEN WEI-GONG CHANG JIE ZHANG
Computer Science and Information Engineering College of Tianjin University of Science & Technology, School of Electronic Information Engineering, Tianjin University, Tianjin 300072, China
国际会议
2008 International Conference on Machine Learning and Cybernetics(2008机器学习与控制论国际会议)
昆明
英文
180-183
2008-07-12（万方平台首次上网日期，不代表论文的发表时间）