基于结构正则化方法的半监督降维研究

The Study of Semi-supervised Dimensionality Reduction Based on Structure Regularization Method

提出了一种鲁棒的半监督降维算法——同时降维和学习数据的结构特征,亦称之为结构正则化半监督降维算法.首先,通过交替方向优化算法和聚类发现数据的结构(包括内部结构分布和数据分割),实现数据结构化学习;然后把内部结构和数据分割作为降维的正则化项来学习降维映射;在迭代过程中,降维的结果也影响数据的结构化学习,基于数据结构学习和降维映射之间的相互作用,得到更精确的数据结构,降到最合适的维度.实验结果证明了所给方法的有效性.

A robust semi-supervised dimensionality reduction algorithm is proposed in this paper with the structure feature of both in dimensionality reduction and learning data.This algorithm is also called structure regularization semi-supervised reduction algorithm.Firstly,the data structuring learning is realized through the structure of alternating direction optimization algorithm and cluster discovery data(including the intrinsic structure distribution and the data segment);then,both the intrinsic data structure and data segment are formulated as regularization terms for dimensionality reduction.The results of the dimensionality reduction also affect the structure learning step in the following iterations.The interaction based on data structure learning and dimensionality reduction representation results in more accurate data structure as well as the most appropriate dimensionality reduction.The experimental results turn out that the given method is effective.