Deep Multi-View Clustering Based on Reconstructed Self-Expressive Matrix

Round 1

Reviewer 1 Report

Title: 

*****

Deep Multi-view Clustering Based on Reconstructed Self-expressive Matrix

 

Overview:

*********

The authors introduce a deep multi-view subspace clustering solution based on reconstructing the self-expressive matrix coefficients. They integrate local and global representations of each sample using shared and specific layers, in order to better utilize information across multiple views. The paper is well organized. The proposal is well described. Experimental results are well described and seem satisfactory. 

 

Minor comments:

******************

Comment 1: When describing their multi-view convolutional encoder-decoder models, the authors mention their architecture choice (three-layer encoders with [10, 20, 30] channels correspondingly and adopt [4 × 4, 3 × 3, 4 × 4] kernel size and Mish for the non-linear activation functions). What is the rationale behind this choice. Please provide an explanation.

 

Comment 2: Different clustering approaches can benefit from the model suggested by the authors. The authors can discuss the extensibility of their solution on top of existing clustering models. They can refer to recent reviews below:

- An overview of skew distributions in model-based clustering. J. Multivar. Anal. 188: 104853 (2022)

- An overview of cluster-based image search result organization: background, techniques, and ongoing challenges. Knowl. Inf. Syst. 64(3): 589-642 (2022)

 

Comment 3: The paper’s English needs to be carefully revised. The manuscript includes many typos. 

 

The paper’s English needs to be carefully revised. The manuscript includes many typos. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

In this paper, the authors introduce a novel deep multi-view subspace clustering approach  based on reconstructing self-expressive matrix (DCRSM) that goes beyond the conventional 363

self-expressive model. 

The proposed DCRSM is well described. The details are correctly presented.   

 

Within the proposed approach the authors  build a reconstruction module to approximate the self-expressive matrix for each view of the dataset.  The goal of this approximation is to 

improve flexibility and efficiency in the training process.

Indeed the proposed approack consider 

intrinsic relationship between samples and utilizes shared and specific layers  to integrate local and global representations of each sample.  

In my opinion at this point the authors should add several works of Bennani, Sublime and the teams related on exactly the subject of the paper  that is the multi-view approach.

 

 The authors conduct experimental  results on publicly available datasets have and they 

show the superiority of the proposed DCRSM. 

 

I propose to accpet the paper with minor corrections. The minor corrections consist in the update of the References. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Authors should apply following comments in the main text to improve quality of the paper as follows:

1- Authors can add some numerical analysis of the simulation for extensive experimental results in Abstract section.

2- In section 1, after main contributions, authors can add paper organization.

3- In Section 2, authors can extend recent clustering methods for the proposed problem statement and by discussing on following case studies:

- https://ieeexplore.ieee.org/abstract/document/10053658?casa_token=2I1_sRUrQVoAAAAA:TZXZTlc1J-t9_C1peDDPLW3SISYOC-FxpS1TkAYLtM1vV3CFArW3vvFjhUOJEWx9YW8TY6c3 

-  https://link.springer.com/article/10.1007/s10586-022-03749-2

- https://www.sciencedirect.com/science/article/pii/S0020025522007836?casa_token=MCrGJs_FjH8AAAAA:eKJuM7y0ydsNyWVTJ-Cv8NxJfbOi8OBJp9OVQIvT8I7R7PrrzT4hx1szEx6lmhNlNIViVz6T

- https://link.springer.com/article/10.1007/s11063-023-11172-w

4- Authors can discuss on time complexity of Algorithm 1 as Training process of DCRSM.

5- Please carefully check grammar and English writing. There are several miss-understanding and errors in the main text.

Please carefully check grammar and English writing. There are several miss-understanding and errors in the main text.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Authors partially addressed reviewers comments in this version.

They should discuss on fillowing:

1. Please discuss existing case studies exactly in the related work section.

https://ieeexplore.ieee.org/abstract/document/10053658?casa_token=2I1_sRUrQVoA AAAA:TZXZTlc1J-t9_C1peDDPLW3SISYOCFxpS1TkAYLtM1vV3CFArW3vvFjhUOJEWx9YW8TY6c3

https://link.springer.com/article/10.1007/s10586-022-03749-2

https://www.sciencedirect.com/science/article/pii/S0020025522007836?casa_token= MCrGJs_FjH8AAAAA:eKJuM7y0ydsNyWVTJCv8NxJfbOi8OBJp9OVQIvT8I7R7PrrzT4hx1szEx6lmhNlNIViVz6T 

2. Time complexity of the proposed algorithm should be Discussed exactly. 

 

Sufficient 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf