鲁棒谱多流形聚类算法  

Robust Spectral Multi-manifold Clustering

作　　者：邹鹏[1] 李凡长[2] ZOU Peng;LI Fan-zhang(School of Computer Science and Technology, Soochow University, Suzhou 215006, China;Joint International Research Laboratory of Machine Learning and Neuromorphic Computing, Soochow University, Suzhou 215006, China)

机构地区：[1]苏州大学计算机科学与技术学院,江苏苏州215006 [2]苏州大学机器习与类脑计算国际合作联实验室,江苏苏州215006

出　　处：《小型微型计算机系统》2018年第6期1130-1134,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61672364;61672365;61402310;61373093)资助

摘　　要：提出一种鲁棒谱多流形聚类算法(Robust Spectral Multi-Manifold Clustering,简称RSMMC).现实生活中许多数据都是带有噪声的,先前许多聚类算法在直接处理带噪声的数据,聚类效果受了很大影响.为了减少数据噪声,在谱多流形聚类(Spectral M ulti-M anifold Clustering,简称SM M C)的基础上引入一个噪声消除项,能够在迭代优化的过程中输出一个降噪稀疏投影,该投影进而可用于提取"干净"数据进行训练.实验结果表明,本文算法对复杂非线性数据聚类结果优于相关对比算法,而且对噪音具有较强的鲁棒性.A robust spectral multi-manifold clustering（ RSMMC） algorithm is proposed in this paper. Since the real world data often includes a lot of noise and most of the previous clustering methods directly use the noisy data,the performance of these methods is limited. To reduce the noise of inputting data,a noise elimination term is introduced on the basis of Spectral Multi-Manifold Clustering（ SMMC）,which can be used to produce a noise reduction sparse projection in the process of iterative optimization. As a result,the＂clean＂data can be extracted for training. The experimental results show that the proposed algorithm performs better than the correlation algorithms in the clustering of complex nonlinear data,and it has strong robustness to noise.

关 键 词：鲁棒 谱聚类 多流形 降噪 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]