摘要
通过分析稀疏数据或噪声数据,导出局部线性嵌入(LLE)算法出现失效的原因,由此提出了一种基于小世界邻域优化的局部线性嵌入(SLLE)算法.将复杂网络算法引入到流形学习中,利用小世界算法对LLE算法进行数据优化,并以最短路径和局部集群系数作为局部优化参数,解决了数据点不规则时以欧氏空间作为邻域判别标准在构建局部超平面造成嵌入结果扭曲的难题.通过3组标准测试数据集合比较了SLLE、LLE算法,结果表明SLLE算法的计算效果、鲁棒性、非理想数据的降维结果均优于LLE算法,且计算正确率至少提高10%.
By analyzing the invalidity reason of the local linear embedding （LLE） algorithm in ease of the sparse data or the high noise data, small world neighborhood optimization LLE algorithm （SLLE） is proposed based on the complex networks theory. The data in LLE are optimized using the small world algorithm, and the shortest path and the local neighbor set clustering coefficients are used as the local parameters. As a result, the problem of the embedding distortion using only local linear patch of the manifold to define neighborhood in Euclidean space is effectively solved. Three groups of standard data sets are selected to test and to compare the efficiency and robustness of SLLE arid LLE. The experimental results show that the calculation results, robustness and dimension reduction of SLLE are all better than those of LLE, and accuracy rate of SLLE is;at least 10 percent higher than that of LLE.
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2008年第12期1486-1489,共4页
Journal of Xi'an Jiaotong University
基金
国家自然科学基金青年科学基金资助项目(507050723)
关键词
局部线性嵌入
降维
小世界邻域
local linear embedding
dimension reduction
small world neighborhood
