Semi-supervised partial multi-label classification via consistency learning. (November 2022)

• Record Type: Journal Article
• Authors: Tan, Anhui; Liang, Jiye; Wu, Wei-Zhi; Zhang, Jia
• Is Part Of: Pattern recognition. Volume 131(2022)
• Publication Date: 2022-11
• Subjects: Semi-supervised partial multi-label learning -- Label correlation -- HSIC; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2022.108839
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We solve the inconsistency of the distributions in features and labels and acquire the label level instance correlation via HSIC for partial multi-label datasets in semi-supervised scenarios. We propose a semi-supervised partial multi-label method inability of training the feature mapping, recovering ground-truth labels and alleviating noisy labels, and we also derive a nonlinear version of the proposed method. We confirm empirically the effectiveness of the proposed methods, and we further verify the importance of the HISC technique for label-level instance correlation estimation on unseen instances. Abstract: Partial multi-label learning refers to the problem that each instance is associated with a candidate label set involving both relevant and noisy labels. Existing solutions mainly focus on label disambiguation, while ignoring the negative effect of the inconsistency between feature information and label information. Specifically, the existence of completely unlabeled instances makes the estimation of label co-occurrence difficult. To tackle these problems, we propose a novel framework for partial multi-label learning in semi-supervised scenarios by solving the inconsistency between features and labels. In the first stage, the label-level correlation matrix on both labeled and unlabeled instances is derived via Hilbert-Schmidt Independence Criterion (HSIC). The correlation matrix can characterize the label correlation of labeled instances and can propagate the label correlation of unlabeled instances. In the second stage, the proposed framework achieves the training of feature mapping, the recovery of ground-truth labels, and the alleviation of noisy labels in a mutually beneficial manner, and develops an alternative optimization procedure to optimize them. In addition, a nonlinear version is extended by using kernel trick. Experimental studies demonstrate that the proposed methods can achieve competitive superiority against existing well-established methods.