Incomplete multi-view clustering based on information fusion with self-supervised learning

Yilong Cai, Qianyu Shu, Zhengchun Zhou, Hua Meng

Abstract: Highlights•The proposed IMCFL framework is based on self-supervised learning and combines contrastive learning and a nearest-neighbor similarity graph.•We introduce the spectral loss function based on the similarity map to make the information of the hidden data class clusters more compact.•We employ the noise-augmentation method to construct positive pairs, which enables the network to handle positive pairs at longer distances and improves the robustness of the model.•In terms of missing data completion, we estimate missing samples via kernel regression, which establishes cross-view connections between samples and has a higher accuracy.