pith. sign in

arxiv: 2604.18024 · v1 · submitted 2026-04-20 · 💻 cs.LG

Clusterability-Based Assessment of Potentially Noisy Views for Multi-View Clustering

Mudi Jiang , Jiahui Zhou , Xinying Liu , Zengyou He , Zhikui Chen This is my paper

Pith reviewed 2026-05-10 04:45 UTC · model grok-4.3

classification 💻 cs.LG
keywords multi-view clusteringclusterability scorenoisy view detectionview quality assessmentpre-clustering analysisdata preprocessing
0
0 comments X

The pith

A new Multi-View Clusterability Score detects noisy views by measuring per-view structure, joint space, and cross-view consistency.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a pre-clustering method to evaluate the quality of individual views in multi-view datasets. It defines the Multi-View Clusterability Score through three parts that check structural patterns inside each view, the combined data space, and how neighborhoods align across views. This score is intended to flag low-quality or noisy views that would otherwise impair the clustering outcome. A reader would care because most prior work handles noise only inside the clustering step rather than assessing views in advance. Experiments on real datasets show that noisy views harm results and that the new score identifies them more reliably than single-view clusterability measures.

Core claim

The Multi-View Clusterability Score (MVCS) quantifies the strength of latent cluster-related structures in multi-view data through three complementary components: per-view structural clusterability, joint-space clusterability, and cross-view neighborhood consistency, and it supports noisy-view analysis and detection before clustering is performed.

What carries the argument

The Multi-View Clusterability Score (MVCS) built from per-view structural clusterability, joint-space clusterability, and cross-view neighborhood consistency to measure cluster structure strength across views.

If this is right

  • Noisy views significantly degrade clustering performance in multi-view settings.
  • Removing or down-weighting views identified as noisy by the score improves overall clustering results.
  • The score enables effective pre-clustering analysis and detection of noisy views.
  • The approach outperforms single-view clusterability measures when applied to noisy-view analysis.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The score could be inserted into automated data-cleaning pipelines that filter multi-view inputs before any clustering step.
  • The same components might be adapted to assess view quality for multi-view classification or other supervised tasks.
  • Controlled synthetic experiments with known noise injection levels could directly measure how sensitive each component is to degradation.
  • Different application domains may require different ways to combine or threshold the three components for best results.

Load-bearing premise

The three components together can be combined to accurately identify which views, when removed or down-weighted, will improve the results of downstream clustering.

What would settle it

A real-world multi-view dataset in which removing or down-weighting the views flagged as noisy by MVCS yields clustering performance that is equal to or worse than keeping all views or using existing single-view clusterability measures.

Figures

Figures reproduced from arXiv: 2604.18024 by Jiahui Zhou, Mudi Jiang, Xinying Liu, Zengyou He, Zhikui Chen.

Figure 1
Figure 1. Figure 1: t-SNE visualization with varying numbers of noisy views on the Mfeat dataset. [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: t-SNE visualization with varying numbers of noisy views on the COIL dataset. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Clustering accuracy (ACC) of ten multi-view clustering algorithms when each view is individually replaced by [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of clusterability metrics under increasing numbers of noisy views across datasets. The vertical axis [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Noisy-view detection results of different clusterability scores. Subfigures (a), (b), and (c) correspond to MVCS, [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
read the original abstract

In multi-view clustering, the quality of different views may vary substantially, and low-quality or degraded views can impair overall clustering performance. However, existing studies mainly address this issue within the clustering process through view weighting or noise-robust optimization, while paying limited attention to data-level assessment before clustering. In this paper, we study the problem of pre-clustering noisy-view analysis in multi-view data from a clusterability perspective. To this end, we propose a Multi-View Clusterability Score (MVCS), which quantifies the strength of latent cluster-related structures in multi-view data through three complementary components: per-view structural clusterability, joint-space clusterability, and cross-view neighborhood consistency. To the best of our knowledge, this is the first clusterability score specifically designed for multi-view data. We further use it to perform potentially noisy view analysis and noisy-view detection before clustering. Extensive experiments on real-world datasets demonstrate that noisy views can significantly degrade clustering performance, and that, compared with existing clusterability measures designed for single-view data, the proposed method more effectively supports noisy-view analysis and detection.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper introduces a Multi-View Clusterability Score (MVCS) for pre-clustering assessment of potentially noisy views in multi-view data. MVCS quantifies latent cluster structures via three components—per-view structural clusterability, joint-space clusterability, and cross-view neighborhood consistency—and is applied to noisy-view analysis and detection. The authors claim this is the first multi-view-specific clusterability score, that noisy views degrade clustering performance, and that MVCS outperforms single-view clusterability measures, supported by experiments on real-world datasets.

Significance. If the central claims hold after supplying missing details, the work offers a data-level tool for identifying low-quality views prior to clustering, which could improve multi-view clustering pipelines by enabling early view selection or weighting. The paper earns credit for conducting experiments on real-world datasets that illustrate the performance degradation from noisy views.

major comments (2)
  1. [MVCS definition (§3)] The definition of MVCS (abstract and §3) provides no explicit equations, derivation, or functional form for aggregating the three components (per-view structural clusterability, joint-space clusterability, cross-view neighborhood consistency) into a single score, nor any thresholding or weighting scheme. This is load-bearing for the claim that MVCS identifies views whose removal or down-weighting improves downstream clustering.
  2. [Experiments] The experimental section reports that MVCS supports noisy-view detection and outperforms single-view baselines, yet supplies no parameter choices, exact computation protocol for the three components, ablation studies on the aggregation, or validation details showing that flagged views measurably improve clustering when removed. Without these, post-hoc tuning cannot be ruled out and generalizability is unclear.
minor comments (1)
  1. [Abstract] The abstract could briefly indicate the high-level combination rule for the three components to help readers assess the novelty claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the manuscript to improve clarity and completeness.

read point-by-point responses

  1. Referee: [MVCS definition (§3)] The definition of MVCS (abstract and §3) provides no explicit equations, derivation, or functional form for aggregating the three components (per-view structural clusterability, joint-space clusterability, cross-view neighborhood consistency) into a single score, nor any thresholding or weighting scheme. This is load-bearing for the claim that MVCS identifies views whose removal or down-weighting improves downstream clustering.

    Authors: We agree that the aggregation requires explicit specification. In the revised manuscript we will add the full mathematical definitions and derivations for each of the three components, the precise functional form used to combine them into the MVCS score, and any weighting or normalization steps employed. This will make the score reproducible and directly support the downstream claims about noisy-view identification. revision: yes

  2. Referee: [Experiments] The experimental section reports that MVCS supports noisy-view detection and outperforms single-view baselines, yet supplies no parameter choices, exact computation protocol for the three components, ablation studies on the aggregation, or validation details showing that flagged views measurably improve clustering when removed. Without these, post-hoc tuning cannot be ruled out and generalizability is unclear.

    Authors: We acknowledge these details are currently insufficient. The revised version will include all hyper-parameter choices, the exact step-by-step computation protocol for each component, ablation studies on the aggregation function, and additional validation experiments demonstrating measurable clustering improvement after removal of MVCS-flagged views. These additions will address concerns about post-hoc tuning and generalizability. revision: yes

Circularity Check

0 steps flagged

No circularity: MVCS is introduced as a novel composite score without reduction to fitted inputs or self-citations

full rationale

The paper defines MVCS explicitly through three new components (per-view structural clusterability, joint-space clusterability, cross-view neighborhood consistency) and uses it for noisy-view detection. No quoted equation or step shows any component being defined in terms of the final score, a fitted parameter renamed as prediction, or a load-bearing self-citation whose uniqueness theorem is imported from the same authors. The combination is presented as a proposed construct validated by downstream experiments rather than derived by construction from the target clustering improvement. This is the common case of an independent proposal.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no equations or implementation details, so no free parameters, axioms, or invented entities can be identified; the three components of MVCS are named at a conceptual level only.

pith-pipeline@v0.9.0 · 5502 in / 1192 out tokens · 34899 ms · 2026-05-10T04:45:51.071665+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

37 extracted references · 37 canonical work pages

  1. [1]

    A compre- hensive survey on multi-view clustering,

    U. Fang, M. Li, J. Li, L. Gao, T. Jia, and Y . Zhang, “A compre- hensive survey on multi-view clustering,”IEEE Transactions on Knowledge and Data Engineering, vol. 35, no. 12, pp. 12 350– 12 368, 2023

    work page 2023

  2. [2]

    Late fusion incomplete multi-view clus- tering,

    X. Liu, X. Zhu, M. Li, L. Wang, C. Tang, J. Yin, D. Shen, H. Wang, and W. Gao, “Late fusion incomplete multi-view clus- tering,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 41, no. 10, pp. 2410–2423, 2018

    work page 2018

  3. [3]

    Incomplete multi-view clustering via multi-level contrastive learning,

    J. Yin, P. Wang, S. Sun, and Z. Zheng, “Incomplete multi-view clustering via multi-level contrastive learning,”IEEE Transac- tions on Knowledge and Data Engineering, 2025

    work page 2025

  4. [4]

    Robust multi-view clustering with noisy correspondence,

    Y . Sun, Y . Qin, Y . Li, D. Peng, X. Peng, and P. Hu, “Robust multi-view clustering with noisy correspondence,”IEEE Trans- actions on Knowledge and Data Engineering, vol. 36, no. 12, pp. 9150–9162, 2024

    work page 2024

  5. [5]

    Cluster-graph convolution networks for robust multi-view clus- tering,

    W. Zheng, X.-Y . Jing, W. Liu, F. Wu, C. Hu, and B. Du, “Cluster-graph convolution networks for robust multi-view clus- tering,”Knowledge-Based Systems, p. 114163, 2025

    work page 2025

  6. [6]

    Fast multi-view cluster- ing via ensembles: Towards scalability, superiority, and simplic- ity,

    D. Huang, C.-D. Wang, and J.-H. Lai, “Fast multi-view cluster- ing via ensembles: Towards scalability, superiority, and simplic- ity,”IEEE Transactions on Knowledge and Data Engineering, vol. 35, no. 11, pp. 11 388–11 402, 2023

    work page 2023

  7. [7]

    Highly-efficient incomplete large-scale multi-view clustering with consensus bipartite graph,

    S. Wang, X. Liu, L. Liu, W. Tu, X. Zhu, J. Liu, S. Zhou, and E. Zhu, “Highly-efficient incomplete large-scale multi-view clustering with consensus bipartite graph,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2022, pp. 9776–9785

    work page 2022

  8. [8]

    The methods for improving large-scale multi-view clustering efficiency: a survey,

    Z. Yang and Y . Tan, “The methods for improving large-scale multi-view clustering efficiency: a survey,”Artificial Intelligence Review, vol. 57, no. 6, p. 153, 2024

    work page 2024

  9. [9]

    Multi-view clustering with noisy views,

    Y . Ye, X. Liu, and J. Yin, “Multi-view clustering with noisy views,” inProceedings of the 2018 2nd International Confer- ence on Computer Science and Artificial Intelligence, 2018, pp. 339–344

    work page 2018

  10. [10]

    Investigating and mitigating the side effects of noisy views for self-supervised clustering algorithms in practical multi-view scenarios,

    J. Xu, Y . Ren, X. Wang, L. Feng, Z. Zhang, G. Niu, and X. Zhu, “Investigating and mitigating the side effects of noisy views for self-supervised clustering algorithms in practical multi-view scenarios,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 22 957– 22 966

    work page 2024

  11. [11]

    Adaptive weighted multi-view clustering,

    S. Liu and L. Lin, “Adaptive weighted multi-view clustering,” inProceedings of the Conference on Health, Inference, and Learning, ser. Proceedings of Machine Learning Research, B. J. Mortazavi, T. Sarker, A. Beam, and J. C. Ho, Eds., vol. 209. PMLR, 2023, pp. 19–36

    work page 2023

  12. [12]

    Deep safe multi-view clustering: Reducing the risk of clustering performance degradation caused by view increase,

    H. Tang and Y . Liu, “Deep safe multi-view clustering: Reducing the risk of clustering performance degradation caused by view increase,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 202–211

    work page 2022

  13. [13]

    Consensus graph learning for multi-view clustering,

    Z. Li, C. Tang, X. Liu, X. Zheng, W. Zhang, and E. Zhu, “Consensus graph learning for multi-view clustering,”IEEE Transactions on Multimedia, vol. 24, pp. 2461–2472, 2021

    work page 2021

  14. [14]

    Consistency meets inconsistency: A unified graph learning framework for multi- view clustering,

    Y . Liang, D. Huang, and C.-D. Wang, “Consistency meets inconsistency: A unified graph learning framework for multi- view clustering,” inProceedings of the 2019 IEEE International Conference on Data Mining (ICDM). IEEE, 2019, pp. 1204– 1209

    work page 2019

  15. [15]

    Flexible multi- view representation learning for subspace clustering

    R. Li, C. Zhang, Q. Hu, P. Zhu, and Z. Wang, “Flexible multi- view representation learning for subspace clustering.” inPro- ceedings of the Twenty-Eighth International Joint Conference on Artificial Intelligence, 2019, pp. 2916–2922

    work page 2019

  16. [16]

    Uniform distribu- tion non-negative matrix factorization for multiview clustering,

    Z. Yang, N. Liang, W. Yan, Z. Li, and S. Xie, “Uniform distribu- tion non-negative matrix factorization for multiview clustering,” IEEE Transactions on Cybernetics, vol. 51, no. 6, pp. 3249– 3262, 2020

    work page 2020

  17. [17]

    Multiview spectral cluster- ing via structured low-rank matrix factorization,

    Y . Wang, L. Wu, X. Lin, and J. Gao, “Multiview spectral cluster- ing via structured low-rank matrix factorization,”IEEE Transac- tions on Neural Networks and Learning Systems, vol. 29, no. 10, pp. 4833–4843, 2018

    work page 2018

  18. [18]

    Contrastive multi- view kernel learning,

    J. Liu, X. Liu, Y . Yang, Q. Liao, and Y . Xia, “Contrastive multi- view kernel learning,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 8, pp. 9552–9566, 2023

    work page 2023

  19. [19]

    Deep adversarial multi-view clustering network,

    Z. Li, Q. Wang, Z. Tao, Q. Gao, and Z. Yang, “Deep adversarial multi-view clustering network,” inProceedings of the 28th International Joint Conference on Artificial Intelligence, 2019, p. 2952–2958

    work page 2019

  20. [20]

    Cross-modal subspace clustering via deep canonical correlation analysis,

    Q. Gao, H. Lian, Q. Wang, and G. Sun, “Cross-modal subspace clustering via deep canonical correlation analysis,” inProceed- ings of the AAAI Conference on Artificial Intelligence, vol. 34, no. 04, 2020, pp. 3938–3945

    work page 2020

  21. [21]

    Joint deep multi-view learning for image cluster- ing,

    Y . Xie, B. Lin, Y . Qu, C. Li, W. Zhang, L. Ma, Y . Wen, and D. Tao, “Joint deep multi-view learning for image cluster- ing,”IEEE Transactions on Knowledge and Data Engineering, vol. 33, no. 11, pp. 3594–3606, 2020

    work page 2020

  22. [22]

    Self-supervised discriminative feature learning for deep multi-view clustering,

    J. Xu, Y . Ren, H. Tang, Z. Yang, L. Pan, Y . Yang, X. Pu, S. Y . Philip, and L. He, “Self-supervised discriminative feature learning for deep multi-view clustering,”IEEE Transactions on Knowledge and Data Engineering, vol. 35, no. 7, pp. 7470– 7482, 2023

    work page 2023

  23. [23]

    Calibrating the excess mass and dip tests of modality,

    M.-Y . Cheng and P. Hall, “Calibrating the excess mass and dip tests of modality,”Journal of the Royal Statistical Society: Series B (Statistical Methodology), vol. 60, no. 3, pp. 579–589, 1998

    work page 1998

  24. [24]

    Using kernel density estimates to investigate multimodality,

    B. W. Silverman, “Using kernel density estimates to investigate multimodality,”Journal of the Royal Statistical Society: Series B (Methodological), vol. 43, no. 1, pp. 97–99, 1981

    work page 1981

  25. [25]

    A test for spatial homogeneity in cluster analysis,

    R. C. Dubes and G. Zeng, “A test for spatial homogeneity in cluster analysis,”Journal of classification, vol. 4, no. 1, pp. 33–56, 1987

    work page 1987

  26. [26]

    Deciphering clusters with a de- terministic measure of clustering tendency,

    A. F. Diallo and P. Patras, “Deciphering clusters with a de- terministic measure of clustering tendency,”IEEE Transactions on Knowledge and Data Engineering, vol. 36, no. 4, pp. 1489– 1501, 2023

    work page 2023

  27. [27]

    The faiss XXX, VOL. XX, NO. XX, APRIL 2026 10 library,

    M. Douze, A. Guzhva, C. Deng, J. Johnson, G. Szilvasy, P.- E. Mazar ´e, M. Lomeli, L. Hosseini, and H. J ´egou, “The faiss XXX, VOL. XX, NO. XX, APRIL 2026 10 library,”IEEE Transactions on Big Data, vol. 12, no. 2, pp. 346–361, 2026

    work page 2026

  28. [28]

    Multi- level feature learning for contrastive multi-view clustering,

    J. Xu, H. Tang, Y . Ren, L. Peng, X. Zhu, and L. He, “Multi- level feature learning for contrastive multi-view clustering,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 16 051–16 060

    work page 2022

  29. [29]

    Dynamic auto-weighted multi-view co-clustering,

    S. Hu, X. Yan, and Y . Ye, “Dynamic auto-weighted multi-view co-clustering,”Pattern Recognition, vol. 99, p. 107101, 2020

    work page 2020

  30. [30]

    Consider high-order consistency for multi-view clustering,

    X. You, H. Li, J. You, and Z. Ren, “Consider high-order consistency for multi-view clustering,”Neural Computing and Applications, vol. 36, no. 2, pp. 717–729, 2024

    work page 2024

  31. [31]

    Multi-view subspace clustering via tensor nuclear norm factorization,

    T. Yan, Q. Guo, J.-X. Mi, and W. Li, “Multi-view subspace clustering via tensor nuclear norm factorization,”Pattern Recog- nition, vol. 174, p. 112969, 2026

    work page 2026

  32. [32]

    Multi- view clustering via pseudo-label guide learning and latent graph structure recovery,

    R. Cai, H. Chen, Y . Mi, C. Luo, S.-J. Horng, and T. Li, “Multi- view clustering via pseudo-label guide learning and latent graph structure recovery,”Pattern Recognition, vol. 151, p. 110420, 2024

    work page 2024

  33. [33]

    Tensorized diversity and consistency with laplacian manifold for multi-view clustering,

    T. Wu and G.-F. Lu, “Tensorized diversity and consistency with laplacian manifold for multi-view clustering,”Information Sciences, vol. 690, p. 121575, 2025

    work page 2025

  34. [34]

    Multi-view clustering via multi-stage fusion,

    Y . Gan, Y . You, J. Huang, S. Xiang, C. Tang, W. Hu, and S. An, “Multi-view clustering via multi-stage fusion,”IEEE Transactions on Multimedia, vol. 27, pp. 4571–4583, 2025

    work page 2025

  35. [35]

    Multi- layer multi-level comprehensive learning for deep multi-view clustering,

    Z. Chen, X.-J. Wu, T. Xu, H. Li, and J. Kittler, “Multi- layer multi-level comprehensive learning for deep multi-view clustering,”Information Fusion, vol. 116, p. 102785, 2025

    work page 2025

  36. [36]

    Dcmvc: Dual contrastive multi-view clustering,

    P. Li, D. Chang, Z. Kong, Y . Wang, and Y . Zhao, “Dcmvc: Dual contrastive multi-view clustering,”Neurocomputing, vol. 635, p. 129889, 2025

    work page 2025

  37. [37]

    Self-supervised semantic soft label learning network for deep multi-view clustering,

    W. Yan, T. Yang, and C. Tang, “Self-supervised semantic soft label learning network for deep multi-view clustering,”IEEE Transactions on Multimedia, vol. 27, pp. 4971–4983, 2025. Mudi Jiangreceived the MS degree in software engineering from Dalian University of Technology, China, in 2023. He is currently working toward the PhD degree in the School of Soft...

    work page 2025