基于自表示的双图规格化特征选择聚类

特征选择得到的识别特征可以用于聚类分析,提高聚类分析的质量。受数据自表示特性和双图规则化学习的启发,提出了一种新的特征选择聚类算法。利用数据和特征的自表示特性,不仅保留了数据的流形信息,而且保留了特征空间的流形信息。此外,为了充分发挥双图模型的作用和鉴别局部聚类的效果,加入局部判别特征选择聚类,大大提高了聚类的有效性和鲁棒性。

FEATURE SELECTION CLUSTERING BASED ON SELF-REPRESENTING DOUBLE GRAPH NORMALIZATION

The recognition features obtained through feature selection can be used in cluster analysis to improve the quality. Inspired by the self-representation characteristics of data and regular learning of double graphs, a new feature selection clustering algorithm was proposed. Utilizing the self-representation characteristics of data and features, not only the manifold information of the data but also the manifold information of the feature space was retained. In addition, in order to make the dual graph model work at full capacity, and identify the effects of local clusters, we added the local discriminative features for clustering, which had greatly improved the effectiveness and robustness of clustering.