Research and Application of Generative-Adversarial-Network Attacks Defense Method Based on Federated Learning

Round 1

Reviewer 1 Report

The following references are not included in the paper:

1) Exploiting Defenses against GAN-Based Feature Inference Attacks in Federated Learning by Xianglong Zhang, University of Science and Technology Beijing Xinjian Luo, National University of Singapore

How the under-review paper differs from the above reference?

2) Federated Learning Attacks and Defenses: A Survey by Yao Chen, Yijie Gui,  Hong Lin, Wensheng Gan, and Yongdong Wu

 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper presents an interesting method of GAN attack defense for federated learning. The experimental results, obtained for some datasets, proved the method's efficiency. 

The paper is well written but may be improved to increase its quality and intelligibility:

1. The problem statement should be presented in a more readable and comprehensive way.

2. Is the method applicable only in the image recognition domain? ML is used in other problems (e.g. big data), thus discussions concerning possible applications of the proposed approach would be valuable. 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

** do not use abbreviations in title. Replace GAN by Generative-Adversarial-Network

** need a native English-speaking person to improve the manuscript

** Lines 115 – 117: rewrite statement to read better

** Line: 121: replace “build” by “builds”

** it is not clear what images the article is referring to. Clarify

** Line 203: describe how you mask pixels of top SHAP. Do you erase them?

** Line 206: replace “can” by “cannot” for statement to make sense

** Line 377: how does the accuracy of the proposed method compare with other method published in literature? Add columns to Table 1 for other models, so the reader can get a sense

** add some real-life scenarios and use-cases where the proposed method can be used. Would the method apply in finance sector, exploit software security problems, etc.

** add more recent articles to the references. You latest reference year is 2020. Add more articles published in 2021 and 2022 on same topic

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The references must include all authors. Not use et.al. abbreviation in references.  The comparison of your work with the analysis in reference [3] presented in your note for the reviewer must be included in your paper.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf