- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper proposes a Prototype-driven Reliability-aware Semantic Consensus Learning (PRSCL) framework for incomplete multi-view clustering (IMVC). Specifically, PRSCL first employs a gating mechanism to adaptively evaluate the quality of available views and perform instance-level weighted fusion, aiming at producing reliable consensus semantic representations and semantic prototypes. Then, PRSCL employs entropy-regularized optimal transport to derive soft assignments between samples and consensus semantic prototypes, and enforce cross-view consistency at the clustering level through bidirectional mutual supervision. Experimental results are provided to show the effectiveness of the proposed method.
Critical strengths
- Various types of experimental results are provided for empirical validation.
- The reliability of missing views is considered by an automatic weighting scheme.
Critical weaknesses
- The overall objective function comprises as many as 5 distinct loss terms. This inevitably raises the concern that the reported performance gains are largely achieved by heavily stacking loss components.
- The cross-view mutual supervision mechanism based on pseudo-labels is prone to be affected by incorrect clustering divisions (inaccurate prototypes and erroneous pseudo-labels) in the early stage of training, which will lead to unstable clustering performance, especially when the data missing rate is high.
- Only 4 datasets are used for experimental validation.
Questions for authors
- The proposed method seems to be a stacking of existing technologies. What is the essential contribution of this paper?
- How to overcome the problem of cumulative errors caused by the inaccuracy of the prototype and the inaccuracy of the pseudo-labels?
- Does the definition of cross-view semantic consistency rely on the assumption that each view is sufficient for clustering?
5 comments
Rebuttal Part 2: Responses to W1–W2
Response to W1: We agree that using multiple loss terms may raise concerns about redundant constraints. However, the five constituent terms in PRSCL serve complementary purposes rather than simply stacking optimization objectives. Specifically, they address representation reconstruction , feature-level cross-view alignment , semantic-level cross-view alignment , and reliability-aware fusion . The latter jointly consists of semantic discrimination and entropy regularization, which coordinate consensus learning and adaptive view weighting. The ablation results in Table 2 of our paper further verify the complementary contributions of these objectives. The performance gains stem from their complementary interplay, rather than merely adding more loss terms.
Table 2. Ablation study on the BDGP dataset under different missing rates.
| r=0.0 ACC | r=0.0 NMI | r=0.0 ARI | r=0.1 ACC | r=0.1 NMI | r=0.1 ARI | r=0.3 ACC | r=0.3 NMI | r=0.3 ARI | r=0.5 ACC | r=0.5 NMI | r=0.5 ARI | |||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ✓ | 33.52 | 9.03 | 4.89 | 39.44 | 17.77 | 10.34 | 34.08 | 13.68 | 7.35 | 35.21 | 9.38 | 6.52 | ||
| ✓ | 94.04 | 90.87 | 91.21 | 93.64 | 90.32 | 91.23 | 92.12 | 84.74 | 86.59 | 91.61 | 83.15 | 85.36 | ||
| ✓ | 92.52 | 84.17 | 83.03 | 94.24 | 84.60 | 86.32 | 64.28 | 52.54 | 46.06 | 62.04 | 43.46 | 36.83 | ||
| ✓ | ✓ | 96.20 | 91.11 | 90.97 | 95.92 | 90.22 | 92.87 | 94.12 | 84.63 | 87.27 | 85.84 | 68.96 | 68.71 | |
| ✓ | ✓ | 96.16 | 91.06 | 90.88 | 95.04 | 90.54 | 92.16 | 94.04 | 84.36 | 87.08 | 86.41 | 70.43 | 69.93 | |
| ✓ | ✓ | 96.84 | 93.16 | 93.71 | 96.28 | 90.96 | 92.37 | 91.16 | 83.35 | 85.68 | 89.72 | 79.37 | 82.29 | |
| ✓ | ✓ | ✓ | 97.96 | 94.36 | 95.00 | 97.16 | 91.50 | 93.07 | 94.76 | 85.50 | 87.47 | 93.28 | 81.92 | 84.05 |
Response to W2: We agree that inaccurate early prototypes may introduce noisy mutual supervision. PRSCL mitigates this issue through several designs. First, encoder–decoder warm-up provides meaningful initial representations before prototype-based training, while and continue to provide learning signals independent of cluster assignments. Second, instead of using hard pseudo-labels, entropy-regularized OT generates balanced soft assignments, avoiding over-sharp supervision and reducing the risk of cluster collapse (Section 2.4). Third, the shared prototypes are periodically re-estimated by K-means on the evolving reliability-weighted consensus representations (Eq. (8)); thus, inaccurate early prototypes are not fixed but can be progressively revised. Meanwhile, the reliability-aware gate excludes missing views from consensus construction without explicit imputation, avoiding additional noise from unreliable completion. Therefore, although early assignment errors cannot be completely eliminated, PRSCL reduces their propagation and enables the prototypes and soft assignments to be refined jointly during training.
Rebuttal Part 3: Responses to W3
Response to W3: Evaluations over a broader spectrum of datasets are essential to validate the generalizability of incomplete multi-view clustering algorithms. Following your suggestion, we have supplemented two additional benchmark datasets, i.e., CUB (600 instances, 2 views) and YouTubeFace10 (38654 instances, 4 views), which cover fine-grained bird recognition and face identification tasks with distinct data distributions. The full quantitative results under four missing rates () are reported in Table 3, including ACC, NMI and ARI. As shown, our PRSCL achieves competitive or state-of-the-art performance across most settings, even under severe view missing. These results further demonstrate the effectiveness and robustness of our method.
Table 3. Performance comparison on CUB and YouTubeFace10. The best result of each metric under each missing rate is highlighted in bold.
| Dataset | Method | r=0.0 ACC | r=0.0 NMI | r=0.0 ARI | r=0.1 ACC | r=0.1 NMI | r=0.1 ARI | r=0.3 ACC | r=0.3 NMI | r=0.3 ARI | r=0.5 ACC | r=0.5 NMI | r=0.5 ARI |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CUB | DCP | 53.20 | 63.90 | 40.05 | 51.60 | 59.21 | 32.94 | 37.47 | 50.60 | 22.62 | 21.17 | 21.12 | 20.05 |
| CPSPAN | 63.50 | 69.01 | 51.23 | 63.83 | 70.78 | 53.73 | 61.67 | 64.57 | 48.16 | 64.67 | 66.79 | 53.01 | |
| ProImp | 81.60 | 74.33 | 65.87 | 78.93 | 72.93 | 63.65 | 74.80 | 68.52 | 57.45 | 70.90 | 64.92 | 52.75 | |
| RPCIC | 74.33 | 73.24 | 61.78 | 69.17 | 69.41 | 54.93 | 59.50 | 63.38 | 47.06 | 63.33 | 63.68 | 48.24 | |
| PMIMC | 57.83 | 65.80 | 49.85 | 63.50 | 69.21 | 53.16 | 75.83 | 72.28 | 60.28 | 64.83 | 66.50 | 53.80 | |
| FreeCSL | 75.01 | 68.95 | 56.77 | 66.17 | 67.51 | 54.13 | 67.33 | 68.56 | 56.27 | 56.66 | 58.35 | 41.97 | |
| HSACC | 62.07 | 69.51 | 49.76 | 60.07 | 68.10 | 48.83 | 61.23 | 66.65 | 45.19 | 52.90 | 55.82 | 38.45 | |
| DCG | 77.17 | 71.97 | 60.50 | 75.83 | 72.36 | 60.58 | 74.01 | 67.18 | 54.96 | 64.17 | 65.97 | 51.42 | |
| CAMERA | 79.67 | 73.81 | 63.90 | 79.94 | 73.51 | 64.23 | 78.50 | 71.58 | 61.01 | 71.11 | 65.82 | 54.13 | |
| GLGC | 58.83 | 56.42 | 41.78 | 54.43 | 52.10 | 34.94 | 54.17 | 49.79 | 34.70 | 44.30 | 41.07 | 22.75 | |
| OURS | 82.17 | 74.41 | 66.49 | 80.17 | 73.14 | 65.07 | 79.83 | 72.10 | 63.31 | 76.00 | 68.41 | 57.83 | |
| YouTubeFace10 | DCP | 60.85 | 65.78 | 32.38 | 62.04 | 69.81 | 63.47 | 54.28 | 63.35 | 52.33 | 56.51 | 62.43 | 58.22 |
| CPSPAN | 69.80 | 74.69 | 59.35 | 73.94 | 79.21 | 71.27 | 68.12 | 79.34 | 64.17 | 72.68 | 80.05 | 72.93 | |
| ProImp | 49.26 | 54.60 | 32.73 | 53.47 | 59.11 | 52.21 | 55.71 | 63.90 | 54.92 | 53.82 | 62.24 | 52.02 | |
| RPCIC | 70.23 | 77.09 | 68.41 | 74.45 | 79.14 | 73.05 | 73.58 | 75.37 | 74.86 | 74.26 | 74.76 | 73.64 | |
| PMIMC | 78.15 | 82.60 | 71.91 | 71.24 | 76.15 | 63.88 | 76.68 | 80.53 | 71.01 | 75.89 | 80.69 | 70.49 | |
| FreeCSL | 77.42 | 81.07 | 70.43 | 80.99 | 81.71 | 71.30 | 79.99 | 81.61 | 71.58 | 73.28 | 78.33 | 68.17 | |
| HSACC | 38.77 | 41.19 | 7.48 | 40.49 | 43.48 | 8.32 | 39.60 | 44.84 | 10.50 | 38.11 | 41.93 | 7.98 | |
| DCG | 82.82 | 80.73 | 73.01 | 84.77 | 83.30 | 76.26 | 77.87 | 81.90 | 74.63 | 71.14 | 76.56 | 64.69 | |
| CAMERA | 74.36 | 77.89 | 67.72 | 76.65 | 79.08 | 68.30 | 73.27 | 78.18 | 65.23 | 71.31 | 77.73 | 66.77 | |
| GLGC | 28.04 | 30.34 | 12.82 | 32.14 | 33.87 | 15.19 | 29.88 | 31.71 | 14.63 | 27.55 | 29.83 | 11.32 | |
| OURS | 83.18 | 82.78 | 72.27 | 84.18 | 84.43 | 74.02 | 86.72 | 84.62 | 75.35 | 83.46 | 83.47 | 72.79 |
Rebuttal Part 1: Responses to Q1–Q3
We sincerely thank you for the encouraging remarks and the valuable suggestions. Our responses are organized by each point below.
Response to Q1: We acknowledge that some of the fundamental techniques adopted in this work, such as entropy-regularized optimal transport, and bidirectional semantic supervision, have been explored in previous studies. Therefore, the contribution of this work does not lie in proposing these techniques individually, but in developing a task-oriented collaborative learning framework to address view reliability discrepancies and cross-view semantic consistency in incomplete multi-view clustering.
Specifically, the main improvements of our method lie in two aspects. First, unlike methods such as FreeCSL, which employ fixed completeness-based weights for view fusion, we introduce a learnable instance-level reliability-aware gating mechanism to adaptively adjust the fusion weights according to the feature information of different available views. This enables the learning of more reliable consensus representations and provides a more discriminative semantic foundation for subsequent shared prototype learning. Second, unlike methods that decouple view fusion from semantic prototype learning, PRSCL couples the two modules: the instance-level reliability-aware consensus representation G is used to re-anchor the shared prototypes at every epoch, while the prototypes in turn provide cluster-level semantic supervision that further refines the consensus. In FreeCSL, by contrast, the shared prototypes are derived from a fixed completeness-weighted consensus. This reliability–semantic coupling, rather than shared prototypes or optimal transport per se, constitutes our core contribution.
Response to Q2: We agree that inaccurate prototypes and pseudo-labels in the early training stage may cause cumulative errors. PRSCL mitigates this issue through several designs. First, the encoder–decoder warm-up provides meaningful initial representations before prototype-based supervision begins. Second, instead of using hard pseudo-labels, entropy-regularized optimal transport produces balanced soft assignments, so that an uncertain sample distributes its mass over multiple prototypes rather than being locked to a single possibly wrong cluster. Third, the shared prototypes are re-estimated by K-means on the current reliability-weighted consensus representation G at every epoch; early prototypes are therefore not fixed and can be progressively corrected as the representations improve. Finally, the reconstruction loss Lrec and the feature-level cross-view contrastive loss Lccl provide learning signals that are independent of the cluster assignments. These designs reduce the propagation of early errors, although they cannot completely eliminate them. We will clarify this discussion in the revised paper.
Response to Q3: Our definition of cross-view semantic consistency does not assume that each individual view is independently sufficient for accurate clustering. Instead, it requires that the semantic representations of the same sample from different available views yield consistent or approximately consistent cluster assignments when evaluated against the shared semantic prototypes.
We acknowledge that different views may have varying discriminative capabilities, and some may contain noisy or insufficient semantic information. This is precisely why PRSCL incorporates a reliability-aware gating mechanism to adaptively assess the contributions of available views, together with prototype-based bidirectional supervision to promote cross-view semantic alignment. Therefore, semantic consistency refers to the alignment of clustering semantics across views, rather than requiring every view to independently achieve reliable clustering results.
Table 1. Cross-view semantic consistency and single-view clustering sufficiency.
As verified on ALOI-100, we examine per-view quality under our cross-view semantic consistency framework. From Table 1, the learned reliability weights reflect views' discriminative ability: views yielding higher single-view ACC obtain larger weights during consensus fusion. In this way, our method aggregates complementary information while tolerating views with inferior individual clustering capacity.