a novel multiple kernel clustering method

ISCAS OpenIR

	a novel multiple kernel clustering method
	Zhang Lujiang; Hu Xiaohui
	2012
会议名称	8th International Conference on Emerging Intelligent Computing Technology and Applications, ICIC 2012
会议录名称	Communications in Computer and Information Science
页码	87-92
会议日期	July 25, 2012 - July 29, 2012
会议地点	Huangshan, China
收录类别	EI
ISSN	1865-0929
ISBN	9783642318368
部门归属	(1) School of Automation Science and Electrical Engineering Beijing University of Aeronautics and Astronautics Beijing China; (2) Institute of Software Chinese Academy of Sciences Beijing China
摘要	Recently Multiple Kernel Learning (MKL) has gained increasing attention in constructing a combinational kernel from a number of basis kernels. In this paper, we proposed a novel approach of multiple kernel learning for clustering based on the kernel k-means algorithm. Rather than using a convex combination of multiple kernels over the whole input space, our method associates to each cluster a localized kernel. We assign to each cluster a weight vector for feature selection and combine it with a Gaussian kernel to form a unique kernel for the corresponding cluster. A locally adaptive strategy is used to localize the kernel for each cluster with the aim of minimizing the within-cluster variance of the corresponding cluster. We experimentally compared our methods to kernel k-means and spectral clustering on several data sets. Empirical results demonstrate the effectiveness of our method. © 2012 Springer-Verlag.; Recently Multiple Kernel Learning (MKL) has gained increasing attention in constructing a combinational kernel from a number of basis kernels. In this paper, we proposed a novel approach of multiple kernel learning for clustering based on the kernel k-means algorithm. Rather than using a convex combination of multiple kernels over the whole input space, our method associates to each cluster a localized kernel. We assign to each cluster a weight vector for feature selection and combine it with a Gaussian kernel to form a unique kernel for the corresponding cluster. A locally adaptive strategy is used to localize the kernel for each cluster with the aim of minimizing the within-cluster variance of the corresponding cluster. We experimentally compared our methods to kernel k-means and spectral clustering on several data sets. Empirical results demonstrate the effectiveness of our method. © 2012 Springer-Verlag.
关键词	Intelligent Computing
主办者	IEEE Computational Intelligence Society; International Neural Network Society; National Science Foundation of China
语种	英语
内容类型	会议论文
URI标识	http://ir.iscas.ac.cn/handle/311060/15801
专题	中国科学院软件研究所
推荐引用方式 GB/T 7714	Zhang Lujiang,Hu Xiaohui. a novel multiple kernel clustering method[C],2012:87-92.