论文部分内容阅读
标签传播算法(LP)是一种基于图的半监督学习算法,通过保持数据间的某些特殊结构,将部分有标签数据的标签信息迭代传递给无标签数据,直至获得全局的稳定状态.结合标签传播算法和线性鉴别分析提出一种流形结构保持的传播半监督降维算法(SDRMPP),采用流行结构上的重构权重并结合已知的部分标签信息进行标签传播,利用传播后获得的全体软标签信息构造离散度矩阵实现鉴别分析,通过求解目标函数的最优值获得特征抽取空间,从而对测试样本进行分类.在Yale和Feret两个标准人脸库上实验验证了该算法的有效性,尤其在只存有少量有标签样本的情况下,该算法仍能保持良好的分类性能.
The label propagation algorithm (LP) is a graph-based semi-supervised learning algorithm that iteratively passes part of the tagged tag information to unlabeled data by maintaining some special structure between the data until a globally stable state is obtained. Label Propagation Algorithm and Linear Discriminant Analysis This paper proposes a propagation-based semi-supervised dimensionality reduction (SDRMPP) algorithm that preserves the manifold structure. It adopts the structural reconstruction weight and combined with known partial label information for label propagation. The whole soft label information constructs the dispersion matrix to realize the discriminant analysis, obtains the feature extraction space by solving the optimal value of the objective function, and classifies the test sample.Experimental results show that the proposed algorithm is validated by two standard face libraries, Yale and Feret The algorithm can still maintain good classification performance, especially when there are only a few labeled samples.