Unsupervised deep clustering via contractive feature representation and focal loss. (March 2022)

• Record Type: Journal Article
• Authors: Cai, Jinyu; Wang, Shiping; Xu, Chaoyang; Guo, Wenzhong
• Is Part Of: Pattern recognition. Volume 123(2022)
• Publication Date: 2022-03
• Subjects: Unsupervised learning -- Clustering -- Contractive feature representation -- Focal loss -- Auto-encoder; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2021.108386
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: Propose a novel deep clustering framework with joint optimization. Learn effective embedded features by contractive representation learning. Improve the label assignment mechanism by introducing focal loss. Design a mechanism to adopt focal loss into clustering in an unsupervised manner. Abstract: Deep clustering aims to promote clustering tasks by combining deep learning and clustering together to learn the clustering-oriented representation, and many approaches have shown their validity. However, the feature learning modules in existing methods hardly learn a discriminative representation. In addition, the label assignment mechanism becomes inefficient when dealing with some hard samples. To address these issues, a new joint optimization clustering framework is proposed through introducing the contractive representation in feature learning and utilizing focal loss in the clustering layer. The contractive penalty term added in feature learning would cause the local feature space contraction, resulting in learning more discriminative features. To our certain knowledge, this is also the first work to utilize the focal loss to improve the label assignment in deep clustering method. Moreover, the construction of the joint optimization framework enables the proposed method to learn feature representation and label assignment simultaneously in an end-to-end way. Finally, we comprehensively compare with some state-of-the-art clustering approaches on several clustering tasks to demonstrate the effectiveness of the proposed method.