基于切空间的局部嵌入映射近邻选择

针对欧式距离下局部嵌入映射近邻点选择的缺陷,从其定义出发,引入切空间距离,改进了近邻点的选择方法,从而能够更好的满足LLE对于局部线性的要求。论文用剩余方差测试其性能,通过对S-curve数据和Swiss-roll数据的仿真可以看到,基于切空间距离的方法能够更好的表示数据的输入/输出映射质量。

In this paper,we present a new algorithm for selecting neighbors of Locally linear embedding(LLE). Because the Euclidean distance can not express the local linear characteristic of data accurately,so we introduce tangent space distance,and we use residual variance to test the performance of the algorithm.From the simulation results of S-curve data and Swiss-roll data we can see that the neighbor selecting method based on tangent space distance can express the I/O mapping quality better than classical method...