融入无监督度量学习的稀疏子空间聚类模型

New Sparse Subspace Clustering Model Integrating with Unsupervised Metric Learning

子空间聚类任务中的无标记数据具有维度高、数据分布分散等特点,传统方法对数据预处理未进行详细地针对化设计且大多使用欧氏距离度量数据间的相似性,使聚类性能提升受限.因此,本文提出融入无监督度量学习的稀疏子空间聚类模型,该算法将距离度量与子空间聚类联合到同一框架,设计由两步组成的聚类过程.该方法对原始数据进行度量学习并重构了稀疏子空间聚类模型,使数据预处理不再是一个单独的步骤,最大限度地将输入的无标记数据之间相似度提高,有效提升了子空间聚类性能、加强了模型泛化能力.我们在真实公开数据集上进行实验测试,实验结果表明该方法优于现有的子空间聚类算法,具有良好的聚类性能.

The unlabeled data are high dimensionality and data distribution are scattered in the subspace clustering,traditional methods have not a detailed customized design for data preprocessing and most of them use Euclidean distance to measure the similarity between data points,which influences the clustering performance.Therefore,this paper proposes a new sparse subspace clustering model integrating with unsupervised metric learning(UML-SSC),which combines distance metric and subspace clustering into the joint framework,and designs a new clustering process consisting of two steps.The proposed method performs metric learning to original data and reconstruct the model,the preprocessing is not a independent step anymore,effectively improve subspace clustering performance and strengthen the generalization ability of the model.We conduct experiments on real-world public datasets,and the experimental results show that this model has good clustering performancewhich is better than existing subspace clustering algorithm.