Federated Pseudo-Sample Clustering Algorithm: A Label-Personalized Federated Learning Scheme Based on Image Clustering

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1. The author must adhere to the proper format for writing articles.

2. Authors need to clarify their model LPFL-GD (what is the full name of it)?

3. It would be beneficial if the authors provide a detailed table in Section 2 (Related Work), includinging the objectives, publication years, methods, and obtained scores by previous approaches, for better understanding.

4. Title of section 4.1 needs to be polished. Additionally, would recommend to consider merging Section 4.3 with 4.1, as it contains the overall methodology in the form of algorithms. Furthermore, explain the algorithms/theorems in more detail.

5. Authors are encouraged to provide a table containing the parameters of the DBSCAN algorithm along with their tuned values. Also, highlight the tuning method employed.

6. The title of Section 5.1 should be changed from "Dataset" to "Datasets".

7. The quality of Figures, especially 4-9, needs improvement in terms of image quality, size, and text format, size, and color.

8. It would be beneficial if the authors provide limitations of the federated learning-based approach for more in-depth analyses.

Comments on the Quality of English Language

Authors need to thoroughly review the manuscript for grammatical errors.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

The authors presented an interesting article that will be appealing to a specific audience working with intelligent methods. The article is of a larger scope. Authors have managed to expand the scope of federated learning scenarios, and the training is no longer limited to a single classification criterion that all clients must adhere to. It is serious achievement, which is validated by results and compared with other closely related methods in terms of purpose

The personalized federated learning method based on image clustering (LPFL-GD) is well described by presenting the mathematical model, also providing the pseudo algorithm. The logic of experiments and results are also described deep enough. The article is written in the good English language and is easy readable.

Therefore, I suggest accepting the paper after minor revision. The revision is needed because of next notes:

- Figure 1 resolution is bed. Should be improved.

- There are quite a lot mistakes. The main types of mistakes are absence of space between words and other symbols like brackets and wrong formatting of symbols in text and formulas.

- Conclusion could be improved, describing the results more accurate.

- There is the information about the dataset. There is information that there were 10 classes. What was the overall size of dataset. However, what was the size of samples in every class in order to know, if the data between the classes was balanced? What equipment was used for training? The examples of the images could be provided.

Comments on the Quality of English Language

There are quite a lot mistakes. But the main types of mistakes are absence of space between words and other symbols like brackets and wrong formatting of symbols in text and formulas. 

Author Response

Please see the attachment

Author Response File: Author Response.docx