计算机工程 ›› 2012, Vol. 38 ›› Issue (06): 181-183.doi: 10.3969/j.issn.1000-3428.2012.06.059

• 人工智能及识别技术 • 上一篇    下一篇

半监督的局部保留投影降维方法

谈 锐 a,b,陈秀宏 a   

  1. (江南大学 a. 数字媒体学院;b. 物联网工程学院,江苏 无锡 214122)
  • 收稿日期:2011-08-01 出版日期:2012-03-20 发布日期:2012-03-20
  • 作者简介:谈 锐(1987-),男,硕士研究生,主研方向:人工智能,模式识别;陈秀宏,教授

Semi-supervised Locality Preserving Projection Dimensionality Reduction Method

TAN Rui a,b, CHEN Xiu-hong a   

  1. (a. School of Digital Media; b. School of Internet of Things Engineering, Jiangnan University, Wuxi 214122, China)
  • Received:2011-08-01 Online:2012-03-20 Published:2012-03-20

摘要: 针对现有数据降维算法不能同时利用标记样本和无标记样本的问题,提出一种半监督局部保留投影降维方法。定义类间相似度和类内相似度,同时最大化类间分离度、最小化类内分离度,保持样本总体结构和局部结构,从而提高数据降维的效果。在人工数据集、UCI数据库和Olivetti人脸库中的测试结果表明,该方法的识别率较高。

关键词: 数据降维, 半监督, 局部结构, 全局结构, 相似度, 分离度

Abstract: Existing algorithms can not effectively use rich labeled and unlabeled sample contains valuable information, which is useful for dimensionality reduction. Aiming at this problem, this paper proposes a novel method called Semi-supervised Locality Preserving Projection (SSLPP). It redefines the between-class similarity and within-class similarity, which is used to maximize the between-class separability and minimizes the within-class separability. In addition, the proposed method preserves the global and locality structure of unlabeled samples. Experimental results in artificial data sets, UCI databases and Olivetti face databases show the usefulness of SSLPP.

Key words: data dimensionality reduction, semi-supervised, local structure, global structure, similarity, separability

中图分类号: