提出一种自适应邻域选择算法,适用于所有基于局部的流形学习算法.该算法能够根据数据集分布的不同密度和曲率选择合适的邻域大小,同时结合局部多维尺度变换(LMDS),在合适的邻域下直接降维并通过全局整合得到数据集的低维坐标.实验表明该算法可较好恢复较复杂数据集的低维几何结构. An adaptive neighborhood selection algorithm is proposed, which is suitable for all local manifold learning algorithms. The algorithm can select the appropriate neighborhood size according to the different density and curvature of the data set distribution. Combined with local multidimensional scaling (LMDS) The dimensionality reduction of the dataset is achieved by global dimensioning under the proper neighborhood.The experiment shows that the algorithm can recover the low-dimensional geometry of the more complex dataset better.