基于近邻元分析的半监督流形学习算法*

现有的大多数流形学习算法偏重保持流形的几何结构,并未考虑到样本点的标签信息,这在一定程度上限制了流形学习算法在数据分类中的应用.因此文中提出一种基于近邻元分析的半监督流形学习算法,采用近邻元分析学习距离度量矩阵,在距离度量方式下选择样本点的局部邻域点.基于距离度量方式构造样本点和邻域点的局部几何结构,并在样本点的低维嵌入坐标中保持这种局部几何结构不变.3个不同数据集上的分类实验验证了文中算法的有效性.

Semi-supervised Manifold Learning Algorithm Based on Neighbourhood Components Analysis

In most of the existing manifold learning algorithms, the geometry structure of the data instances is preserved, but the label information is ignored. Therefore, the application of manifold learning algorithms in data classification is limited. In this paper, a semi-supervised manifold learning algorithm based on neighborhood components analysis is proposed. A distance metric matrix is learned by using neighbor components analysis and local neighbors of the sample points are selected by using the new distance metric. The local geometric structures of the sample points and their neighbors are constructed under the new distance metric, and the local geometric structures are preserved in the low-dimensional embedding coordinates of the sample points. The classification experiments conducted on three different datasets demonstrate the efficiency of the proposed algorithm.