A Novel Approach for Effective Multi-View Clustering with Information-Theoretic Perspective

Multi-view clustering (MVC) is a popular technique for improving clustering performance using various data sources. However, existing methods primarily focus on acquiring consistent information while often neglecting the issue of redundancy across multiple views. This study presents a new approach called Sufficient Multi-View Clustering (SUMVC) that examines the multi-view clustering framework from an information-theoretic standpoint. Our proposed method consists of two parts. Firstly, we develop a simple and reliable multi-view clustering method SCMVC (simple consistent multi-view clustering) that employs variational analysis to generate consistent information. Secondly, we propose a sufficient representation lower bound to enhance consistent information and minimise unnecessary information among views. The proposed SUMVC method offers a promising solution to the problem of multi-view clustering and provides a new perspective for analyzing multi-view data. To verify the effectiveness of our model, we conducted a theoretical analysis based on the Bayes Error Rate, and experiments on multiple multi-view datasets demonstrate the superior performance of SUMVC.

Paper

Similar papers

Peer review

Reviewer XaYq6/10 · confidence 4/52023-06-23

Summary

This paper proposes two methods for multi-view clustering, which aims at grouping data from multiple sources or perspectives. The first part, SCMVC uses a consistent variational lower bound to learn consistent information among views. The second part, SUMVC extends the information bottleneck principle to reduce redundancy and achieve sufficient representation among views. The proposed SUMVC consists of two terms: a consistent variational lower bound and a sufficient representation lower bound to enhance consistency and minimize redundancy among views. The authors leverage information bottleneck theory and variational analysis to develop the model. The paper also provides a theoretical analysis of the generalization error of the learned representations based on the Bayes error rate. The paper evaluates the proposed methods on four real-world multi-view datasets.

Strengths

1. The paper addresses an important and challenging problem of MVC with an information-theoretic perspective. This paper proposes novel lower bounds to address the issue of view redundancy and consistency. The authors utilize Bayes Error Rate to provide a theoretical explanation of the effectiveness of the proposed method. 3. The paper presents extensive experiments on four real-world multi-view datasets and demonstrates the superior performance of the proposed methods over existing methods.

Weaknesses

1. The proposed objective function in Eq. (4) shares similarities with that of VAE [1], and it would be beneficial if the author could provide further explanations on this matter. How does this loss function differ from that of VAE? Is it merely an extension of the single-view approach to multi-view data? Additionally, what is the reason for SCMVC achieving significantly better performance than VAE-based methods in Table 2? 2. In SCMVC, the fusion representation $\vec{Z}$ is directly optimized, while it is not in SUMVC. Could this approach negatively impact the performance of SUMVC? 3. The methods discussed in this article may not be effectively applied to datasets with high heterogeneity. 4. The introduction section contains some complex sentences. Simplifying the language and structure could improve clarity. 5. It would be beneficial for the authors to provide a more comprehensive discussion of some missed information-based multi-view methods [2-5] in the related work section. [1] Auto-Encoding Variational Bayes, NIPS’13 [2] COMPLETER: Incomplete Multi-view Clustering via Contrastive Prediction, CVPR'21 [3] Rethinking Minimal Sufficient Representation in Contrastive Learning, CVPR’ 22 [4] Dual Contrastive Prediction for Incomplete Multi-View Representation Learning, TPAMI'23 [5] Multi-view information-bottleneck representation learning, AAAI’21

Questions

The authors mention that the methods discussed in this article cannot be effectively applied to datasets with high heterogeneity. Could authors provide a detailed analysis?

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

3 good

Contribution

2 fair

Limitations

Please see weakness. Also, the complexity analysis is missing.

Reviewer 89XX7/10 · confidence 4/52023-07-05

Summary

The paper proposed Sufficient Multi-View Clustering , SUMVC, which is composed of two main components. The first component is a simple and reliable multi-view clustering method called SCMVC (simple consistent multi-view clustering), which utilizes variational analysis to generate consistent information. The second component proposes a lower bound on sufficient representation, aiming to enhance consistent information and reduce unnecessary information among views.

Strengths

- Analyze the effectiveness of multi-view clustering from the information-theoretic perspective is interesting and valuable. - The proposed method outperforms the state of the arts. The paper has good reproducibility with the provided codes. - The paper is very well written. The theoretical and empirical analyzes are convincing.

Weaknesses

- There are few comparative methods, and more comparative methods need to be selected reasonably to illustrate the effectiveness of the proposed methods. - The theoretical analysis (based on the Bayes Error Rate) should be elaborated more. The key findings or insights from this analysis should also be emphasized. - Did the experimental results demonstrate any limitations or potential challenges of the proposed method?

Questions

see above.

Rating

7: Accept: Technically solid paper, with high impact on at least one sub-area, or moderate-to-high impact on more than one areas, with good-to-excellent evaluation, resources, reproducibility, and no unaddressed ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

It can be found in Appendix of this paper.

Reviewer ugfB6/10 · confidence 4/52023-07-05

Summary

This work introduces a new approach called sufficient multi-view clustering (SUMVC) to improve clustering performance using multiple data sources. Existing methods often focus on acquiring consistent information while neglecting the issue of redundancy across multiple views. By contrast, the proposed SUMVC provides a promising solution to the problem of multi-view clustering and offers a new perspective for analyzing multi-view data. The effectiveness of the model is verified through theoretical analysis and experiments on multiple multi-view datasets, showing superior performance compared to other methods.

Strengths

1. By examining the multi-view clustering framework from an information-theoretic standpoint, SUMVC offers a promising solution to the problem of multi-view clustering and provides a new perspective for analyzing multi-view data. 2. The effectiveness of the proposed model is demonstrated through theoretical analysis based on the Bayes Error Rate and experiments on multiple multi-view datasets, highlighting its superior performance compared to other methods.

Weaknesses

1. This paper is lack of in-depth discussion or evaluation of the proposed approach's limitations. 2. The authors could provide more insights into the computational complexity of the proposed method and its scalability to large-scale multi-view datasets.

Questions

The paper mentions that the superiority of SUMVC is demonstrated through experiments on multiple multi-view datasets. However, there is no detailed discussion or analysis provided on the characteristics of these datasets or how they were selected. What are the criteria for dataset selection, and how representative are these datasets?

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

No potential negative societal impact of this work exists.

Reviewer NC947/10 · confidence 4/52023-07-05

Summary

This paper considers the problem of multi-view clustering from an information theoretic perspective. It focuses on representation learning, and optimizes said representation to improve down-stream clustering performance (with k-means). It introduces an Information Bottleneck based loss function, which considers consistency between views, redundancy and sufficiency of representations in addition to the traditional likelihood based reconstruction loss. It introduces 2 methods SCMVC and SUMVC based on some or all of these additional loss terms. The paper then analyzes the model effectiveness by establishing a connection to the Bayes Error Rate, and also shows good experimental results across some multi-view datasets.

Strengths

- Introduces a novel approach for representation learning for Mmulti-View Clustering (MVC) based on information theoretic criteria. - Proposes two separate methods, both of which show good experimental performance. - Has mathematical rigor in both breaking down the loss function, as well as in the theoretical analysis that follows. - Compares against multiple state-of-the-art MVC methods and shows superior performance on the chosen datasets. - Conducts a wide range of experiments, including ablation studies and parameter sensitivity analysis.

Weaknesses

- The presentation of the mathematical parts of the preliminaries, discussion and analysis is quite opaque. Variables and terms are often explained only after they have already been used in equations. The equations tend to be quite cluttered and difficult to parse. - The datasets chosen are less than ideal, and are mostly derivative from single-view datasets. There are other multi-view datasets such as NUS-WIDE and 3-Source News which are potential candidates for publicly available MV datasets. - The assumptions (and thus applicability) of the methods are restrictive. I.e. assuming mutual redundancy across all views. As they mention in the supplementary, their methods perform poorly on heterogeneous MV data, limiting their applicability. - The paper was not self-contained; the supplementary material was essentially required to understand the a lot of the details. For eg. the limitations of the methods were only mentioned in the supplementary material. The supplementary material must not be used as additional pages, and the paper itself should be able to stand alone.

Questions

Overall questions: - This paper is more about representation learning for clustering than clustering itself, right? Similar to spectral clustering. Not that that's a bad thing, it's just a little misleading. You spend a lot of the paper talking about MVC but the actual clustering is just k-means on top of a learned representation. It might help to be clear about this up front. - As I mentioned in the second point of the Limitations, it seems like this method is also weak to only partial information existing in views. How would you remedy this? Would ensemble variational methods work here as well? - For the MNIST experiments, I am confused about your choice to use pairs of the same digit (but written by different people) as two views. The data distribution of both the views are very similar, then. Also, if you pair up digits, shouldn't your dataset size be 35000 (halved) and not 70000? Or are you using both (A, B) and (B, A) for each pair? If so, this doesn't really seem like a good multi-view dataset since the views themselves are basically indistinguishable overall. - In table 3, the 3rd row seems to not have any useful information. It's clear that $L_{rec}$ would be required since you're using an auto-encoder. Instead, maybe you can have $L_{rec} + L_{suf}$ which you don't have here. That would be interesting to see. I will leave line-by-line comments here (since there doesn't seem to be a better place for this). There are a few typos here and there, but I won't bother too much with those: - [Line 32] Did you mean maximizing MI between representation* and output (not input and output) - [Line 82] "... to quantify amount of" -- sentence is not complete. - [Section 3] You use y, z without explaining/defining what they are. Clustering is an unsupervised task, so what does this mean in this context? I'm guessing it means cluster assignment but it isn't clear. - [Line 164] I may be wrong but isn't $\phi^i$ the generational parameter set, and not $\theta^i$? - [Line 180] What is a "pseudo-label?" A cluster assignment? You should define this earlier. - [Line 164] What do you mean by unique distribution? - [Line 186] This equation is very hard to parse. For cleaner appearance, you should consider using \left( and \right) to have larger parantheses. You could also remove the superscripts just for these long equations where there is no $j$, and just leave a note below. Also, shouldn't there be a conditioning on y in the first term? Lastly, $KL$ -> $D_{KL}$. - [Line 198] maximizing* - [Equation 9] There should be no expectation here, right? The $D_{KL}$ absorbs it, I think. Also, for the second term. the first distribution uses $z^j$ instead of $z^i$. - [Line 243] Should it be $z^j_m$ here? Also, what is $P^{\otimes n}$?

Rating

7: Accept: Technically solid paper, with high impact on at least one sub-area, or moderate-to-high impact on more than one areas, with good-to-excellent evaluation, resources, reproducibility, and no unaddressed ethical considerations.

Confidence

4: You are confident in your assessment, but not absolutely certain. It is unlikely, but not impossible, that you did not understand some parts of the submission or that you are unfamiliar with some pieces of related work.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

- [Author mentioned] Heterogeneity in the data (eg. in terms of dimensions of features) affects performance significantly. - Restrictive assumptions are made on mutual redundancy between views. Multi-view data often has only partial information available in each view. I.e. you might need more than one view to get the complete picture. This also seems like a weakness of the methods.

Reviewer 79CP6/10 · confidence 5/52023-07-06

Summary

A consistent variational lower bound is provided to explore the consistent information among views for multi-view clustering, based on which SCMVC (simple consistent multi-view clustering) is proposed. To enhance consistent information and minimize unnecessary information among views, a sufficient representation lower bound is further proposed.

Strengths

(1) The main idea is novel, necessary preliminary knowledge and sufficient theoretical analysis is given. (2) Experiments conducted on real multi-view data demonstrate the good performance of proposed methods. The codes are available. (3) The findings of this study have the potential to contribute to the advancement of multi-view clustering techniques.

Weaknesses

(1) It is said the proposed model does not perform well on datasets with strong heterogeneity between views, but no evidence or experiments support this statement. How does the heterogeneity affect the performance should be explained. (2) The paper shows quantitative results of the proposed SUMVC and SCMVC. However, I expect some visualization results to show the difference between these two methods. (3) A thorough analysis of potential drawbacks, and practical considerations would enhance the overall strength of the paper. (4) The tested data sets in this paper contain a small number (<=3) of views. It is suggested to add data sets with more than three views for discussion. (5) The writing should be further improved.

Questions

Please see ‘Weaknesses’.

Rating

6: Weak Accept: Technically solid, moderate-to-high impact paper, with no major concerns with respect to evaluation, resources, reproducibility, ethical considerations.

Confidence

5: You are absolutely certain about your assessment. You are very familiar with the related work and checked the math/other details carefully.

Soundness

3 good

Presentation

3 good

Contribution

3 good

Limitations

Yes

Reviewer ugfB2023-08-17

I have read the other reviewer's comments and the author's rebuttal. I will keep my original score.

I have read the other reviewer's comments and the author's rebuttal. I will keep my original score.

Reviewer NC942023-08-17

I acknowledge the author's comments, and have gone through the other reviewers' feedback. The authors' responses have answered my questions well. I believe that the presentation of the paper will be much clearer after incorporating the changes they mentioned in their rebuttal. I hope that weakness #4 above will also be addressed in their changes to the manuscript (i.e. paper not being self-contained and needing the appendix to really understand it). My main remaining concern is that the evaluations/experiments are not conducted on natural multi-view datasets, and rather on modified single-view datasets. While I understand that MNIST-MV is commonly used in literature, lacking experimental evaluations on natural multi-view datasets detracts from the impact of the contribution. It seems that most of the reviewers have similar concerns on the experimental evaluations. If additional experiments on other datasets have been conducted since then, I would like to see those evaluations as well. At this point, I intend to keep my original score.

Authorsrebuttal2023-08-19

Dear Reviewer NC94, We appreciate your response to our rebuttal and the additional questions raised. Regarding the evaluations being conducted on modified single-view datasets rather than natural multi-view datasets, we’d like to clarify that for the MULTI-COIL-10 and MULTI-COIL -20 datasets, they are not derived from single-view data, and each view represents different angles at which the photos were taken. To further address your concerns, we have conducted additional experiments using two widely used multi-view datasets: the REU dataset and the HW dataset. We have also introduced five additional comparison methods, namely IDEC (Improved Deep Embedded Clustering), CSMSC (Consistent and Specific Multi-View Subspace Clustering), FMR (Flexible Multi-View Representation Learning for Subspace Clustering), GMC (Graph-Based Multi-View Clustering), and CGMSC (Multi-View Subspace Clustering with Adaptive Locally Consistent Graph Regularization). The results are presented in the tables below. | Method (REU)| IDEC| SAMVC | GMC | DEMVC | SUMVC | | --- | --- | --- | --- | --- | --- | | ACC | 46.0|18.8 | 19.8|46.7|58.3 | | NMI| 25.2|4.6 | 13.8| 25.3 |60.0| | ARI | 18.0| 0.3| 1.3| 20.4|47.5| | Method (HW)| SAMVC | FMR |CSMSC | CGMSC | FMVACC | SUMVC | | --- | --- | --- | --- | --- | --- | --- | | ACC | 76.4 | 86.1| 89.8 |69.1|89.5| 96.4 | | NMI| 84.4 | 76.5|83.0 |81.8|86.0|93.2| | ARI | 73.9| 72.6| 79.5|69.5|85.0|92.4| It can be observed that our method performs well on these datasets. We will add new comparative methods and datasets in the final version. We hope our response has addressed your concerns.

Reviewer XaYq2023-08-17

After thoroughly reviewing the feedback provided by both the other reviewer and the author's rebuttal, I am pleased to state that my initial concerns have been adequately addressed. I hope that these discussions can be of value to the authors in their efforts to enhance their work and I decided to raise my rating.

Reviewer 79CP2023-08-22

I

Dear authors, Thanks for your detailed reply. My concerns have been addressed by your response. The addtional emperiments comparsions are hoped to add in the revised version. I would like to increse my scores into 'weak accept'.

Program Chairsdecision2023-09-21

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC