Semi-supervised Dimensionality Reduction Method with Side-information Propagation and Revise

The existed side information based semi-supervised dimensionality reduction algorithms add the objective functions that preserve side information and topology information directly,so the error edge in data topology can't be revised.We proposed a method that integrates side information into data structures by transferring and revising mechanism,so it can preserve side information and more really data topology.Experiment results show that data was reduced dimension by this algorithm can gain higher accuracy than other algorithms,and this algorithm is robust to parameter k of KNN graph.