EUVIMEDEUVIMED entdecken
Crossref

Towards Robust Multi-View Clustering via Early-to-late Manifold Consistency Calibration

Ruimeng Liu, Chang Tang, Bo Wang, Zhenglai Li, Lianbo Guo, Xinwang Liu

IEEE Transactions on Pattern Analysis and Machine Intelligence · 2026 · 1-18

DOI: 10.1109/tpami.2026.3752509 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026. Verfasst von Ruimeng Liu und 5 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

For multi-view data, structural heterogeneity across views is a common property, with cross-view sparsity variation serving as a typical manifestation. To first account for this sparsity-induced heterogeneity, we develop Adaptive View-specific Encoding, which incorporates the sparsity ratio of each view as prior knowledge for view-aware representation learning and constraint modulation. Based thereon, we further tackle the broader issue of view quality imbalance in multi-view fusion, where low-quality views introduce unreliable structural relations and view-specific constraints further amplify latent discrepancies. To this end, we propose a globallocal manifold-guided fusion mechanism named Early-to-late Manifold Consistency Calibration, which leverages the relatively stable global manifold preserved in early-fused encoded features to modulate robust late-stage fusion of adaptively encoded local features. Specifically, the early-stage global features serve as a structural anchor for assessing the manifold consistency of features of each adaptively encoded local view. The resulting consistency scores are then used to adaptively reweight late-stage global fusion, attenuating structurally unreliable views while reinforcing those aligned with the common manifold. In this manner, latent manifold consistency enables early fusion to calibrate late-stage fusion, thereby establishing a principled connection between adaptive view-specific encoding and cross-view distribution alignment. The full framework, named Manifold Consistency Guided Adaptive Sparse Autoencoder for Multiview Clustering (MASA), is validated on multiple datasets and demonstrates superior effectiveness and robustness across diverse experimental settings. On ALOI-100, MASA further enhances clustering stability, reducing the fluctuation standard deviation over the final rounds from 3.02% to 1.28%.

Zitieren

APA

Liu, R., Tang, C., Wang, B., Li, Z., Guo, L., & Liu, X. (2026). Towards Robust Multi-View Clustering via Early-to-late Manifold Consistency Calibration. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1-18. https://doi.org/10.1109/tpami.2026.3752509

Vancouver

Liu R, Tang C, Wang B, Li Z, Guo L, Liu X. Towards Robust Multi-View Clustering via Early-to-late Manifold Consistency Calibration. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2026:1-18. doi:10.1109/tpami.2026.3752509

BibTeX
@article{Liu2026,
  author = {Liu, Ruimeng and Tang, Chang and Wang, Bo and Li, Zhenglai and Guo, Lianbo and Liu, Xinwang},
  title = {Towards Robust Multi-View Clustering via Early-to-late Manifold Consistency Calibration},
  journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year = {2026},
  pages = {1--18},
  doi = {10.1109/tpami.2026.3752509}
}
Kostenlos starten →