USING GEODESIC DISTANCE TO BUILD RELATIVE TRANSFORMATION FOR ISOMETRIC EMBEDDING

摘要：

Isometric embedding approaches are topologically unstable when confronted with sparse or noisy data, as where the neighbourhood structure is critically distorted. Inspired by the cognitive law, a relative transformation on the original space is proposed to construct the relative space. In the relative space, the distinction between points is improved and the impact of noise is diminished. Furthermore, in order to deal with the highly twisted and folded noisy manifold, a relative transformation based on geodesic distance is proposed, which is then applied to improve an isometric embedding approach by constructing better neighbourhood graph. The improved approach was validated by conducted experiments on both synthetic and real data.

关键词： manifold learning isometric embedding relative transformation geodesic distance relative manifold.

作者: Guihua Wen Xiangbin Liu Shengping Feng Chunshun Lin

作者单位: School of Computer Science and Engineering,South China University of Technology,Guangzhou 510641,Chi Survey Research Center on Situation in Guangdong Province,Guangzhou510620,China

会议类型: 国际会议

会议名称: 2008高等智能国际会议(2008 International Conference on Advanced Intelligence)

会议地点: 北京

会议语种:英文

在线出版日期: 2008-10-18（万方平台首次上网日期，不代表论文的发表时间）