- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This paper studies continual multi-view clustering, where views of the same samples arrive one after another and only the soft labels from the previous session are kept. In each session, SLT-CMVC runs entropy-regularized fuzzy C-means on the new view and uses a probability membership module to relate the current soft labels to the historical ones. It then stacks the current soft labels, the historical soft labels, and the learned consistency matrix into a third-order tensor, and regularizes it with a new Low-Bias Refined (LBR) norm. Optimization is done with ADMM. Experiments on seven datasets compare SLT-CMVC with ten single-view and incremental multi-view clustering methods.
Critical strengths
This paper considers a practical setting for continual multi-view clustering: raw features of earlier views are deleted, and only compact clustering information is kept.
By combining cross-session consistency learning with tensor-based correction, the framework gives a fairly complete way to use new views while keeping useful historical information.
Critical weaknesses
- Several update rules are derived incorrectly. In Eqs. (19)–(20), quadratic terms that depend on itself (the cross-session term and the ADMM penalty) are treated as constants, so the softmax update does not solve the subproblem. In Eq. (29), a -related term appears in the numerator but not in the denominator, so is not guaranteed. Eq. (26) updates the multiplier as , which does not match the penalty in Eq. (5), and the proof in Appendix 8.3 uses instead. I would like to see correct derivations for each subproblem and confirmation that the code follows them.
- Many equations and notations are inconsistent, making the method hard to follow and reproduce. For example, Eq. (24) leaves out , and Eq. (25) takes the derivative at , which does not agree with the MM derivation that follows. Indices are mixed up in several equations, the size and normalization constraint of change between sections, and the numerator and denominator of Eq. (29) use different terms. In the main text, Eqs. (1), (2), and (4) only require and leave out .
- Convergence is not properly proved. According to the paper, the objective is semi-algebraic because it consists of polynomials, logarithms, and exponentials, and therefore satisfies the KL property. Logarithms and exponentials are not semi-algebraic in general, so the KL property has to be shown in another way. Boundedness is also argued in a circle: is updated using , and is updated using , yet each is used to bound the other. Sufficient descent and the relative-error condition are mostly stated rather than proved, and they do not cover every alternating update. As it stands, the proof does not show convergence to a KKT stationary point.
- It is not clear that LBR offers a real advantage. Log-Sum, Schatten-, and other nonconvex surrogates are all designed to keep large singular values while shrinking small ones. In Table 6, LBR improves on them only slightly and is even worse on one dataset. The paper also does not say whether the other surrogates use their best parameter settings, so it is hard to tell whether these differences are meaningful.
Questions for authors
- Could the authors give a complete and rigorous proof that the algorithm converges to a KKT stationary point?
- View order is tested only from high to low quality and the reverse. Is the method still stable under random view orders?
3 comments
Dear Reviewer oCy5 ,
Thank you for your time and effort in reviewing our paper. Your comments have been valuable in helping us improve its quality.
As the author–reviewer discussion period is scheduled to end in approximately two days, we kindly encourage you to review our rebuttal and let us know whether your concerns have been adequately addressed.
Thank you again for your time and consideration.
Best regards, The Authors
Thank you for the detailed response and additional clarifications. The rebuttal is helpful, but my overall assessment of the paper remains unchanged. I will maintain my original score.
Q1: Thank you for requesting a rigorous convergence analysis. We have corrected the U and Q updates using KL majorization and aligned the tensor shrinkage step with the LBR penalty.
For a fixed session with finite data, positive row-normalized initial memberships, and μk=μ0ρk, where μ0>0 and ρ>1, the multiplier sequence is bounded and the primal iterates converge to a feasible limit in exact arithmetic. Let Uk=stack(H,Uk,Qk), with H fixed, and Zk=Uk+1−Ak/μk.
Since 0≤fε′(σ)≤1/ε, each singular value in the tensor update shrinks by at most 1/(εμk). Parseval's identity and the multiplier update give ∥Ak+1∥F=μk∥Mk+1−Zk∥F≤εmin(n,c). Thus, for a uniform bound C on ∥Ak∥F, ∥Mk+1−Uk+1∥F=μk∥Ak+1−Ak∥F≤μk2C→0.
The membership rows lie on the simplex and the updated centers lie in the convex hull of the data. Together with μk ∥Mk−Uk∥F≤2Cρ,
for k≥1, this bounds the non-entropy gradients in the U and Q updates. For either block, a row pk is updated by normalizing the exponentials of zak=η+τkτklogpak−gak,τk≥μk, where η≥0 is its entropy coefficient and ∥gk∥∞≤G.
Consequently, logpbk+1pak+1≤logpbkpak+μk2G. This also holds with the implementation's floor on centered logits, since clipping cannot enlarge a pairwise logit difference. As ∑kμk−1<∞, the log ratios and logpk are uniformly bounded. Hence ∥zk−logpk∥∞=O(μk−1).
The normalized exponential map is Lipschitz and maps logpk back to pk for each previously updated row, including with the logit floor. It follows that ∥pk+1−pk∥∞=O(μk−1). The row differences are summable, so Uk and Qk converge to strictly positive probability matrices. The weighted center update then gives convergence of Vk, and the primal residual bound gives M⋆=stack(H,U⋆,Q⋆).
The result establishes primal convergence and feasibility. (Gaussian Regression-Driven Tensorized Incomplete Multi-View Clustering with Dual Manifold Regularization NeurIPS-2025)
Q2: Thank you for raising the issue of random view order. We evaluated all six permutations of Synthetic3d. For each permutation, we ran 50 independent runs using the same set of 50 seeds. The table below shows the mean final ACC.
The final average ACC value was 91.43%. This indicates that even with a random view order, the proposed method is still able to maintain a stable clustering structure.