Publications

Toward Generalized Multistage Clustering: Multiview Self-Distillation

Published in TNNLS, 2024

DistilMVC is a multi-stage deep multi-view clustering framework that leverages multi-view self-distillation—using a teacher-student model to distill dark knowledge of pseudo-label distributions—combined with contrastive learning and mutual information maximization across views to correct overconfident pseudo-labels, yielding state-of-the-art clustering performance on real-world scenario.

Recommended citation: Wang J, Xu Z, Wang X, Li T. Toward Generalized Multistage Clustering: Multiview Self-Distillation. IEEE Transactions on Neural Networks and Learning Systems. 2024 Nov 11.
Download Paper | Download Bibtex | Code

Self‐supervised multi‐view clustering in computer vision: A survey

Published in IET Computer Vision, 2024

This survey examines the rise of self‐supervised multi‐view clustering—outlining its motivations and advantages, categorizing common datasets and data challenges, reviewing representation‐learning and self‐supervised methods with application examples, and highlighting open research problems for future exploration.

Recommended citation: Wang J, Xu Z, Yang X, Li H, Li B, Meng X. Self‐supervised multi‐view clustering in computer vision: A survey. IET Computer Vision. 2024 Sep;18(6):709-34.
Download Paper | Download Bibtex

Self‐supervised image clustering from multiple incomplete views via constrastive complementary generation

Published in IET Computer Vision, 2023

This paper proposes CIMIC‐GAN, a multi‐view image clustering framework that integrates GAN‐based data imputation with dual contrastive learning on both complete and incomplete views to fully exploit complementary and consistent information, significantly enhancing clustering performance under high missing rates.

Recommended citation: Wang J, Xu Z, Yang X, Guo D, Liu L. Self‐supervised image clustering from multiple incomplete views via constrastive complementary generation. IET Computer Vision. 2023 Mar;17(2):189-202
Download Paper | Download Bibtex

Balancing complementarity and consistency via delayed activation in incomplete multi-view clustering

Published in PRCV, 2024

We introduce CoCo-IMC, an incomplete multi-view clustering framework that employs a dual delayed-activation network to balance complementary and consistent information across views, then recovers missing data by minimizing conditional entropy and maximizing mutual information—demonstrating superior performance over 12 state-of-the-art baselines on four public datasets.

Recommended citation: Li B, Xu Z, Yun J, Wang J. Balancing complementarity and consistency via delayed activation in incomplete multi-view clustering. InChinese Conference on Pattern Recognition and Computer Vision (PRCV) 2024 Oct 18 (pp. 531-545). Singapore: Springer Nature Singapore.
Download Paper | Download Bibtex

Hierarchical mutual information analysis: towards multi-view clustering in the wild

Published in IJCNN, 2024

We introduce a hierarchically consistent deep multi‐view clustering framework that first uses dual prediction to recover missing views and achieve instance‐level alignment, then employs contrastive reconstruction for class‐level alignment—significantly outperforming state‐of‐the‐art methods under both view‐missing and unaligned conditions.

Recommended citation: Wang J, Xu Z, Yang X, Guo D, Liu L. Self‐supervised image clustering from multiple incomplete views via constrastive complementary generation. IET Computer Vision. 2023 Mar;17(2):189-202.
Download Paper | Download Slides | Download Bibtex