- Rating
- 3
- Confidence
- 3
- Informativeness
- 3
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper presents FEMVAE, a method for incomplete multi-view clustering. It uses normalizing flows to model view-specific posterior distributions and combines the available representations through adaptive fusion. A Gaussian mixture prior is introduced to capture cluster structure, while pairwise optimal transport is used to reduce distributional differences across views. The model also reconstructs missing views using inverse flows and view-specific decoders. The authors evaluate the method on five datasets with different missing rates and conduct ablation studies to examine the contributions of its main components.
Critical strengths
- The integration of view-conditioned normalizing flows into variational inference enhances the expressiveness of posterior distributions, allowing the model to capture more complex latent structures than conventional Gaussian assumptions.
- The pairwise optimal transport strategy promotes consistency among view-specific latent distributions without requiring complete cross-view sample correspondences, making it suitable for incomplete multi-view settings.
- The proposed framework combines adaptive fusion, a Gaussian mixture prior, and inverse flow-based reconstruction to integrate complementary information, encourage cluster-aware latent representations, and facilitate missing-view recovery.
Critical weaknesses
- Theorem 2.1 establishes a pairwise upper bound based on a weighted mixture of view-specific distributions. However, this mixture reference does not necessarily coincide with the distribution induced by the adaptive fusion mechanism in Eq. (3), leaving a potential gap between the theoretical analysis and the actual fusion process.
- The proposed method employs pairwise optimal transport to align view-specific latent distributions, rather than directly aligning corresponding sample representations as in conventional instance-level alignment methods. Although this strategy avoids the requirement for complete cross-view sample correspondences, the distinctions between distribution-level alignment and instance-level alignment, as well as their implications for incomplete multi-view clustering, are not sufficiently discussed.
- Although the ablation study demonstrates the contribution of normalizing flows, the influence of flow depth on clustering performance has not been examined. It remains unclear whether increasing posterior expressiveness consistently benefits the proposed framework.
- The mathematical formulation introduces numerous symbols, superscripts, subscripts, and distribution-related notations across different sections, making the methodology relatively difficult to follow. In particular, the notation for view-specific and unified latent variables, probability distributions, fusion weights, and loss terms increases the complexity of understanding the proposed framework.
- Several notation inconsistencies and minor formatting errors remain in the manuscript. For example, Eq. (2) uses inconsistent encoder notation, Eq. (8) contains mismatched brackets in the KL divergence, and Eq. (11) employs a view-indexing range inconsistent with the observed-view setting. Additionally, Figure 1 contains garbled or improperly rendered mathematical symbols, which affects its readability.
Questions for authors
- The authors should clarify how the theoretical upper bound in Theorem 2.1 relates to the actual adaptive fusion process in Eq. (3), given that the weighted mixture reference distribution does not necessarily coincide with the distribution of the unified latent representation.
- The authors should further clarify the motivation for adopting distribution-level alignment instead of conventional instance-level alignment, and explain its potential advantages in handling incomplete observations and heterogeneous view availability.
- The authors should conduct additional experiments with different numbers of flow transformations to examine how flow depth affects clustering performance and whether deeper flows consistently lead to improvements.
- The authors should consider including a notation table summarizing the key mathematical symbols and their definitions to improve readability and help readers better understand the relationships among different components of the framework.
- The authors should carefully review and standardize the mathematical notation throughout the manuscript and correct the rendering issues in Figure 1 to improve clarity and presentation quality.
3 comments
To assess how flow depth affects clustering performance, we varied the number of affine coupling transformations from one to eight on CUB and Handwritten. Table R3.2 reports the ACC, NMI, and ARI at missing rates .
Clustering performance varies nonmonotonically with flow depth. On CUB, three layers achieve the highest scores on all three metrics at and , whereas one layer achieves the highest scores at . On Handwritten, two layers achieve the highest scores on all three metrics at , , and . At , increasing depth from four to eight reduces ACC from 94.15% to 87.68%. Across the eight combinations of dataset and missing rate, the default four-layer configuration achieves an ACC within 2.64 percentage points of the highest ACC among the tested depths. These results support four layers as a practical default for ACC on the two evaluated datasets and show that the best-performing depth depends on the dataset, missing rate, and metric.
In the revised manuscript, we summarize these findings in Section 3.3 and include the full results and detailed analysis in Appendix A.10.
Table R3.2. Sensitivity of clustering performance to flow depth on CUB and Handwritten. Each entry reports ACC/NMI/ARI (%). Boldface marks the highest score across the tested depths for each dataset, missing rate, and metric.
| Dataset | Flow layers | ||||
|---|---|---|---|---|---|
| CUB | 1 | 75.63/73.56/61.06 | 74.93/71.32/58.30 | 72.87/68.41/55.03 | 69.47/65.67/52.39 |
| 2 | 76.03/74.30/61.67 | 73.77/70.93/57.93 | 70.57/67.23/53.25 | 69.00/64.81/51.38 | |
| 3 | 76.80/74.58/62.31 | 73.90/70.51/57.87 | 73.57/69.26/56.91 | 69.27/64.22/51.51 | |
| 4 (default) | 75.08/73.55/62.10 | 76.00/71.11/59.75 | 72.37/66.46/54.36 | 66.83/63.02/50.08 | |
| 5 | 74.63/73.19/60.15 | 74.60/70.72/57.91 | 72.27/68.30/54.92 | 66.07/64.34/49.80 | |
| 6 | 76.03/73.37/61.43 | 75.17/71.85/59.11 | 72.00/68.10/54.59 | 66.67/63.47/49.57 | |
| 7 | 76.27/74.34/62.10 | 75.33/71.09/58.95 | 71.07/67.64/54.06 | 69.17/64.26/50.38 | |
| 8 | 76.37/74.26/62.15 | 75.00/70.95/58.20 | 72.50/67.36/53.94 | 66.43/64.04/50.05 | |
| Handwritten | 1 | 80.44/75.48/69.66 | 78.57/69.35/62.73 | 85.30/75.33/71.51 | 82.88/73.21/68.83 |
| 2 | 93.61/86.81/86.40 | 93.46/86.42/86.10 | 92.25/84.49/83.70 | 91.35/82.88/81.88 | |
| 3 | 82.85/78.03/72.56 | 79.57/74.05/68.13 | 87.18/77.65/74.63 | 87.63/77.49/74.85 | |
| 4 (default) | 94.15/88.61/83.92 | 92.04/84.87/84.66 | 91.46/83.96/82.39 | 90.68/82.22/80.68 | |
| 5 | 90.88/83.16/81.36 | 89.49/80.97/78.80 | 86.54/79.34/75.81 | 86.03/76.29/72.90 | |
| 6 | 87.01/83.77/80.17 | 90.21/81.90/79.82 | 89.28/80.98/79.08 | 88.95/79.60/77.14 | |
| 7 | 88.66/82.49/79.81 | 86.55/80.44/76.63 | 89.06/79.38/77.53 | 86.65/77.80/74.23 | |
| 8 | 87.68/82.93/79.26 | 86.86/79.34/76.20 | 87.37/79.88/76.38 | 86.80/76.84/73.61 |
We have added a notation table in the revised manuscript to summarize the key mathematical symbols, their meanings, and their spaces. An excerpt is provided below; the complete notation table can be found in Table 7 of the revised manuscript.
Table R3.3 Excerpt from the notation table.
| Symbol | Meaning | Space |
|---|---|---|
| Observed data in view | ||
| -th instance in view | ||
| A generic sample when the sample index is clear | ||
| View-availability matrix | ||
| Availability indicator of view for instance | ||
| , | Set of observed views for instance | |
| , | Base noise variable of instance in view | |
| , | View-specific latent representation of instance | |
| , | Unified latent representation of instance | |
| Matrix of view-specific latent representations | ||
| Matrix of unified latent representations |
We have corrected the notation and formatting in Eqs. (2), (8), and (11), as well as Fig. 1 and its caption. Specifically, we have clarified the view-specific encoder as , corrected the KL divergence brackets, and restricted the posterior product to observed views. We have also corrected the dataset and fusion notation in Fig. 1. For the reconstruction loss, we have corrected the signs in Eqs. (8) and (11) and included the omitted term in Eq. (11). Finally, we have checked the use of and , along with the relevant equations, figures, and captions, for consistency throughout the manuscript.
All of my issues have been addressed. I will maintain my positive score in support of this paper.
We sincerely thank you for your constructive comments. Your suggestions have helped us clarify the theoretical motivation for alignment, expand the experimental analysis of alignment methods and flow depth, and improve the consistency and readability of the mathematical notation. All revisions in the manuscript are highlighted in blue. Q1, W1: Relationship between Theorem 2.1 and Adaptive Fusion.
We agree that the weighted mixture reference distribution in Theorem 2.1 does not generally coincide with the distribution of the unified latent representation produced by Eq. (3). Specifically, the former is a mixture of view-specific distributions weighted by the expected fusion weights, whereas the latter is induced by the weighted sum of view-specific latent representations.
Theorem 2.1 establishes that the discrepancy between each view-specific distribution and the mixture reference is bounded by pairwise discrepancies among the view-specific distributions. It therefore motivates the pairwise alignment term as a way to reduce marginal distributional differences among the representations being fused. The theorem does not directly bound the discrepancy between a view-specific distribution and the distribution of the unified latent representation.
We have revised the discussion following Theorem 2.1 in Section 2.4 to clarify the distinction between the mixture reference and the fused distribution, as well as the role of the theorem in motivating pairwise alignment.
Q2, W2: Distribution-Level vs. Instance-Level Alignment.
We adopt distribution-level alignment because it does not require matched cross-view observations for every sample. Unlike instance-level alignment, which relies on available cross-view pairs, it compares view-specific distributions using the observed samples from each view, including those without a corresponding observation in the other view.
We further compared distribution-level alignment with paired Euclidean and contrastive alignment on CUB and Handwritten at four missing rates. As shown in Table R3.1, distribution-level alignment achieves the highest ACC, NMI, and ARI across all evaluated settings. These results support its effectiveness under incomplete observations.
We have clarified the motivation for distribution-level alignment in Section 2.4 and added comparisons with instance-level alignment methods to the ablation study in Section 3.4, with detailed results and analysis in Appendix A.11.
Table R3.1. Comparison of alignment methods on CUB and Handwritten. Each entry reports ACC/NMI/ARI (%).