Correntropy based semi-supervised concept factorization with adaptive neighbors for clustering. (October 2022)

• Record Type: Journal Article
• Authors: Peng, Siyuan; Yang, Zhijing; Nie, Feiping; Chen, Badong; Lin, Zhiping
• Is Part Of: Neural networks. Volume 154(2022)
• Page Start: 203
• Page End: 217
• Publication Date: 2022-10
• Subjects: Concept factorization -- Correntropy -- Semi-supervised learning -- Adaptive neighbors -- Clustering; Neural computers -- Periodicals; Neural networks (Computer science) -- Periodicals; Neural networks (Neurobiology) -- Periodicals; Nervous System -- Periodicals; Ordinateurs neuronaux -- Périodiques; Réseaux neuronaux (Informatique) -- Périodiques; Réseaux neuronaux (Neurobiologie) -- Périodiques; Neural computers; Neural networks (Computer science); Neural networks (Neurobiology); Periodicals; 006.32
• DOI: 10.1016/j.neunet.2022.07.021
• Languages: English
• ISSNs: 0893-6080

Abstract

Abstract: Concept factorization (CF) has shown the effectiveness in the field of data clustering. In this paper, a novel and robust semi-supervised CF method, called correntropy based semi-supervised concept factorization with adaptive neighbors (CSCF), is proposed with improved performance in clustering applications. Specifically, on the one hand, the CSCF method adopts correntropy as the cost function to increase the robustness for non-Gaussian noise and outliers, and combines two different types of supervised information simultaneously for obtaining a compact low-dimensional representation of the original data. On the other hand, CSCF assigns the adaptive neighbors for each data point to construct a good data similarity matrix for reducing the sensitiveness of data. Moreover, a generalized version of CSCF is derived for enlarging the clustering application ranges. Analysis is also presented for the relationship of CSCF with several typical CF methods. Experimental results have shown that CSCF has better clustering performance than several state-of-the-art CF methods.