- Rating
- 2
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This paper addresses the problem of federated incomplete multi-view clustering with missing views. Unlike existing methods that typically treat missing patterns as mere data indices, this paper proposes that missing patterns themselves may contain information related to the underlying cluster structure. To address this issue, the authors introduce the CFIMC framework. Firstly, by modeling the missing mechanism, missing patterns are treated as additional clustering evidence, and inverse probability weighting is utilized to correct selection bias. Secondly, sample-level counterfactual imputation is achieved through cross-view shared semantic representations, mitigating the error self-reinforcement caused by traditional cluster-level imputation. Lastly, reliability assessment and a dual robust prototype are employed using distribution transfer weighting to reduce the impact of unreliable imputation results on clustering. The authors provide relevant theoretical analysis and conduct experimental validation on synthetic datasets and multiple real-world multi-view datasets.
Critical strengths
-
The research perspective is novel. Unlike existing methods that merely regard missing patterns as data indices or information requiring completion, this paper further considers the potential semantic information contained within the missing patterns themselves, providing a meaningful new perspective for federated incomplete multi-view clustering.
-
The methodological framework is complete and the logic is clear. The article explicitly points out three core issues caused by the absence of informational view, and designs corresponding solution modules for each issue. There is a relatively clear correspondence between research questions, methodological design, and theoretical analysis, which ensures good logical coherence in the overall framework.
Critical weaknesses
-
The paper takes the structural causal model as the core theoretical foundation and proposes counterfactual view interpolation. However, from the perspective of the actual algorithm, Module II mainly learns shared semantic representations through other observable views, and then utilizes a cross-view mapping network to predict missing views. This process bears strong similarities to existing shared representation learning and cross-view prediction methods. The authors need to further clarify what capabilities causal modeling brings at the algorithmic level that traditional cross-view interpolation cannot achieve.
-
Although the article provides a relatively comprehensive comparison of clustering performance and ablation experiments, there is still a lack of sufficient direct verification regarding the information provided by missing patterns, the correction effect of selection bias, and the estimation accuracy of the dual robust prototype.
Questions for authors
-
According to Theorem 3.4, when Condition (P) fails, even an accurate imputation model does not guarantee consistency unless . However, the reliability coefficient estimated by the proposed method may be smaller than one even for an optimal imputer. Under these circumstances, in what sense is the estimator doubly robust? Can the authors provide theoretical guarantees for the actual reliability-weighted and shrinkage-based prototype estimator, particularly when pseudo-labels contain errors?
-
Can the author further elaborate on the fundamental differences between their approach and existing cross-view imputation methods, and further explain what capabilities causal modeling brings at the algorithmic level that traditional cross-view imputation cannot achieve.
6 comments
Q1. In what sense is the estimator doubly robust when (P) fails, and what does it guarantee with pseudo-label errors?
Your reading of Theorem 3.4 is correct: when (P) fails, an accurate imputer yields consistency only at ρ = 1, while estimates the on the missing population, which is below one even for the optimal imputer whenever the view has private noise. We revised the paper as follows.
Terminology (Remark D.17). Eq. (13) is the AIPW estimating equation with the augmentation scaled by ρ. At ρ = 1 it is doubly robust in the classical sense. With it keeps the propensity branch, consistency under (P) for every imputer, while the imputation branch yields a quantified bias reduction rather than consistency. We now say this explicitly and corrected the sentence after Theorem 3.4.
Guarantee for the implemented estimator (new Proposition D.16). It covers the estimator we actually run: the estimated (any limit , no (P) needed), the shrinkage of Eq. (14), and estimated, possibly wrong, memberships , for any missingness mechanism.
- (i) , where is the selection bias of the visible mean of within .
- (ii) .
- (iii) Pseudo-label errors add at most for bounded embeddings.
- (iv) With a correct imputer and true memberships the bias is exactly : the visible-mean bias shrinks by the factor and is never amplified.
Hence, the error of the actual prototypes is controlled by the strength of selection within the estimated cluster, times × the spread of the view plus × the imputer residual, plus the misclassification rate.
Why rather than ρ = 1. The label-free reliability cannot distinguish an accurate but noisy imputer from a biased one. At ρ = 1 the term makes a poor imputer harmful, whereas under (P), where both terms vanish, ρ only affects the variance (Theorem D.18). Empirically (, four datasets ("MSRC-v1", "Caltech-5V", "Hdigit", "DryBean"), five seeds; parentheses: paired change vs. ):
| Coefficient ρ in Eq. (13) | MNAR ACC | MNAR NMI | MNAR_latent ACC | MNAR_latent NMI |
|---|---|---|---|---|
| ρ≡0, visible mean | 68.26 (−0.20) | 62.80 (+0.01) | 69.36 (+0.40) | 63.16 (+0.85) |
| ρ̂ (ours) | 68.47 | 62.79 | 68.97 | 62.31 |
| ρ≡1, AIPW | 69.63 (+1.16) | 63.21 (+0.43) | 68.52 (−0.44) | 61.85 (−0.46) |
AIPW performs comparably / slightly better under MNAR, where the imputer is accurate, which matches Proposition D.16(iv); ρ̂ protects against poor imputers, as at η = 0.9 or on small datasets (Table 20). The last rows of the third table in our next comment compare the observed residual of the implemented estimator with the share predicted by Proposition D.16(iv), using oracle prototypes.
W2. Direct verification of the information in missing patterns, of the selection-bias correction and of the DR prototypes
Because missingness is simulated, the hidden views and the labels allow us to audit each mechanism directly (new Appendix H.9: full model, four datasets ("MSRC-v1", "Caltech-5V", "Hdigit", "DryBean"), three seeds, , server coordinates).
(a) Information in the missing patterns. in bits; the accuracy of the best classifier that sees the pattern alone (majority-class rate in parentheses); the error of the estimated visibility table against the truth; the share of samples whose cluster changes when the evidence is added to the feature distance, how many of them move to their true cluster, and the net change in ACC.
| Mechanism | I(Y;R) bits | Pattern-only ACC % (majority) | Table MAE | Assignments changed by Λ (%) | Of which moved to the true cluster (%) | Net ACC gain (pts) |
|---|---|---|---|---|---|---|
| MNAR | 0.945 | 40.5 (16.2) | 0.073 | 0.53 | 32.6 | +0.26 |
| MNAR_latent | 0.435 | 31.5 (16.2) | 0.059 | 0.19 | 14.3 | -0.05 |
(b) Correction of the selection bias. Total variation distance between each client's visible cluster law and the population law, before and after our inverse-probability weights, and the recall of rarely visible clusters with and without Module I.
| Mechanism | TV(P(Y | R=1), P(Y)) unweighted | TV after IPW with | Rare-cluster recall %: CFIMC / w/o mech. | Common-cluster recall %: CFIMC / w/o mech. |
|---|---|---|---|---|
| MNAR | 0.155 | 0.099 | 76.4 / 73.6 | 84.7 / 84.7 |
| MNAR_latent | 0.092 | 0.032 | 79.2 / 83.3 | 78.4 / 80.1 |
(c) Accuracy of the DR prototypes. Distance of each estimator to the oracle prototype, the mean of the hidden-view embeddings over the cluster, divided by the within-cluster standard deviation, with true and estimated memberships. The last three rows compare the observed residual of our estimator with the share predicted by Proposition D.16(iv).
η = 0.5; error = ‖estimate − oracle mean of the cluster‖ / within-cluster sd, averaged over clusters and views (lower is better).
| Prototype estimator | MNAR: oracle memb. | MNAR: estimated memb. | MNAR_latent: oracle memb. | MNAR_latent: estimated memb. |
|---|---|---|---|---|
| Visible mean (ρ=0) | 0.093 | 0.247 | 0.121 | 0.191 |
| Completed-block mean (w/o DR) | 0.216 | 0.243 | 0.148 | 0.187 |
| Outcome regression | 0.261 | 0.304 | 0.173 | 0.227 |
| DR with ρ̂ (ours) | 0.208 | 0.248 | 0.148 | 0.188 |
| AIPW (ρ=1) | 0.224 | 0.250 | 0.152 | 0.189 |
| ‖B_sel‖ / sd (visible-mean bias) | 0.093 | 0.247 | 0.121 | 0.191 |
| DR residual / ‖B_sel‖ (observed) | 2.076 | 0.867 | 1.019 | 0.893 |
| 1 − ρ̂ (prediction of the new proposition) | 0.042 | 0.043 | 0.029 | 0.029 |
We hope these additions address your concerns, and we would be glad to clarify anything further.
W1 / Q2. What causal modeling brings beyond cross-view prediction
Thank you for the review and for pressing on what the causal view adds in practice. In response, we sharpened both the paper and the experiments.
We agree that the networks of Module II are ordinary MLPs and that its architecture resembles cross-view prediction (COMPLETER, DCP). The causal formulation does not change the network class; it changes what the networks estimate, on which population they are trained, how much they are trusted, and how they are used (new paragraph in Section 3.3):
Experiment (Table 2, Appendix H.7). Modules I and III are fixed, and Module II is replaced by sample-level associational imputers with the same capacity and cross-fitting, and by our own networks trained with plain squared error. Besides ACC and NMI, the hidden views give each imputer's oracle R2 on missing entries and its optimism (visible minus hidden R2), the quantities that points 1–3 predict. Averages over four datasets (
"MSRC-v1", "Caltech-5V", "Hdigit", "DryBean") and five seeds (η=0.5); parentheses: paired change vs. full CFIMC.