EUVIMEDEUVIMED entdecken
Crossref

Rethinking Prototype Alignment for Incomplete Multi-View Clustering

Liang Zhao, Yuan Yukun, Ma Shubin, Wang Ziyue

IEEE Transactions on Consumer Electronics · 2026-05 · 72 (2) · 3306-3317

DOI: 10.1109/tce.2026.3679542 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Consumer Electronics, 2026-05. Verfasst von Liang Zhao und 3 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 28 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Science and Technology Project of Liaoning Province, Science and Technology Project of Dalian City.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Multi-view clustering leverages complementary sources to improve grouping, yet real-world applications are often hindered by incomplete data where instances miss certain views. Prior methods typically rely on learning view-to-view correspondences from the subset of fully observed data, a process that is often data-hungry and computationally intensive. To address this, we propose a prototype-oriented framework motivated by a theoretical analysis of gradient consistency. Specifically, our analysis suggests that the optimization direction derived from a representative prototype aligns consistently with that of the underlying cluster centroid. Based on this insight, we introduce a “sample–score–select” strategy that identifies and aligns a single high-quality prototype per mini-batch. Although only one prototype is selected per iteration, the cumulative alignment across the training trajectory ensures the capture of global cluster structure and semantic diversity. This strategy (i) maintains computational efficiency comparable to random sampling, (ii) suppresses noise from incomplete views by filtering out low-quality candidates, and (iii) mitigates the mismatch between the number of prototypes and true clusters. Augmented with a lightweight, pluggable robust imputation module, the proposed framework achieves state-of-the-art accuracy while significantly reducing training time and reliance on paired data. Extensive experiments on multiple benchmarks demonstrate the effectiveness and robustness of this paradigm.

Förderung

  • National Natural Science Foundation of China
  • Science and Technology Project of Liaoning Province
  • Science and Technology Project of Dalian City
  • Fundamental Research Funds for the Central Universities

Zitieren

APA

Zhao, L., Yukun, Y., Shubin, M., & Ziyue, W. (2026). Rethinking Prototype Alignment for Incomplete Multi-View Clustering. IEEE Transactions on Consumer Electronics, 72(2), 3306-3317. https://doi.org/10.1109/tce.2026.3679542

Vancouver

Zhao L, Yukun Y, Shubin M, Ziyue W. Rethinking Prototype Alignment for Incomplete Multi-View Clustering. IEEE Transactions on Consumer Electronics. 2026;72(2):3306-3317. doi:10.1109/tce.2026.3679542

BibTeX
@article{Zhao2026,
  author = {Zhao, Liang and Yukun, Yuan and Shubin, Ma and Ziyue, Wang},
  title = {Rethinking Prototype Alignment for Incomplete Multi-View Clustering},
  journal = {IEEE Transactions on Consumer Electronics},
  year = {2026},
  volume = {72},
  number = {2},
  pages = {3306--3317},
  doi = {10.1109/tce.2026.3679542}
}
Kostenlos starten →