Abstract:

This paper proposes an approach to select the dynamical neighborhood for each point while constructing the tangent subspace based on the sampling density and manifold curvature. The parameters of the approach can be automatic determined by computing the residual variances. This paper applies it to ISOMAP, and experimental results validate the optimization of the neighborhood and the accurate of the embedding results.

Key words: manifold learning, tangent space, dynamical neighborhood, sampling density, manifold curvature

摘要：

针对流形学习的邻域优化问题，提出一种动态邻域的算法。基于局部采样密度和流形弯曲度估计切空间，并为所有样本点动态地选择邻域，其参数可通过计算残差自动确定。实验结果表明，将这种算法应用于ISOMAP后，邻域得到进一步优化，嵌入结果也更加准确。

关键词: 流形学, 切空, 动态邻域, 采样密度, 流形弯曲度

CLC Number: 

• TP18