P-ISOMAP:一种新的对邻域大小不甚敏感的数据可视化算法

ISOMAP算法对邻域大小敏感,而邻域大小却难以有效选取.本文根据二阶最小生成树不含有"短路"边的特性提出了能有效删除邻域图中的"短路"边因而对邻域大小不甚敏感的P-ISOMAP算法.由于避免了邻域大小难以有效选取的问题,该算法能更容易地对数据进行可视化,也获得了一定程度的拓扑稳定性和鲁棒性.实验结果很好地验证了该算法的有效性.

P-ISOMAP:A New ISOMAP-Based Data Visualization Algorithm with Less Sensitivity to the Neighborhood Size

The success of ISOMAP depends greatly on choosing a suitable neighborhood size,however,it is still an open problem how to do this effectively.Based on characteristics of the SOMST(Second-Order Minimal Spanning Tree) in which shortcut edges can be avoided,this paper presented a variant of ISOMAP,i.e.PISOMAP(Pruned-ISOMAP).P-ISOMAP can prune effectively shortcut edges existed possibly in the neighborhood graph according to their costs over the SOMST,and thus is much less sensitive to the neighborhood size than ISOMAP.Consequently,PISOMAP can be applied to data visualization more easily than ISOMAP for the open problem described above can be avoided to a certain extent;in addition,P-ISOMAP can also be more topologically stable and robust than ISOMAP.Finally,the feasibility and effectivity of P-ISOMAP can be verified by experimental results very well.