论文部分内容阅读
现有基于边信息的半监督维数约减算法大都是直接将保留边信息和数据拓扑结构的目标函数相加,因此数据拓扑结构中的错误连接不会因已知的边信息而得到修正.提出通过边信息传播及修正机制将边信息融入到数据拓扑结构图中的方法,从而在保留边信息的同时保留更为真实的数据拓扑结构信息.实验结果表明本文所提出的算法较之其它算法,对数据降维后用于分类时可取得较高的准确率,且算法对创建的KNN图中的参数K最具鲁棒性.
The existing semi-supervised dimension reduction algorithms based on edge information mostly directly add the objective function of preserving edge information and data topology, so the wrong connection in the data topology will not be modified by the known edge information. Proposed a method that the side information is integrated into the topological structure of the data through the side information dissemination and correction mechanism so as to keep the edge information while preserving the more realistic data topology information.The experimental results show that the proposed algorithm compared with other algorithms , The higher accuracy can be obtained when the data is reduced in dimension for classification, and the algorithm is most robust to the parameter K in the created KNN graph.