【作者】 杨剑； 李伏欣； 王珏；

【Author】 Yang Jian 1,2,Li Fuxin 1,2,Wang Jue 1 1(Institute of Automation Chinese Academy of Sciences,Beijing 100080,China) 2(Graduate School of the Chinese Academy of Sciences,Beijing 100039,China)

【机构】 中国科学院自动化研究所；

【摘要】 本文提出了一种基于局部切距离的近邻分类算法,它借助流形学习的思想,是在事物本身所形成低维流形中的近邻法。利用流形的局部切空间表示,直接从样本集出发求出测试样本到各局部切空间的距离,在这一度量上用近邻法进行分类。与已有的切空间距离分类器算法相比,它不需要太多的领域知识和基于对数据集构成深刻认识的假设条件。并通过一些实验说明了算法有较强的分类能力。更多还原

【Abstract】 In this paper, Local Tangent Distance K-Nearest Neighbour algorithm was proposed, which performed the K-nearest neighbor algorithm in the low-dimensional manifolds formed by things themselves. By the local tangent space representation theorem, distances between the test point and local tangent spaces were directly computed from sample dataset. And using the distance, the K-nearest neighbor algorithm was implemented to classification. LTDKNN needs fewer field knowledge and hypothesis on the dataset than the Tangent Distance K-Nearest Neighbor classifier. The validity of the algorithm was illustrated by some experiments.更多还原