Next Article in Journal
On a Hypothetical Model with Second Kind Chebyshev’s Polynomial–Correction: Type of Limit Cycles, Simulations, and Possible Applications
Previous Article in Journal
Mixed Alternating Projections with Application to Hankel Low-Rank Approximation
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Algorithms
Volume 15
Issue 12
10.3390/a15120461
Review Report
algorithms-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles
Article
Peer-Review Record

A Mask-Based Adversarial Defense Scheme

Algorithms 2022, 15(12), 461; https://doi.org/10.3390/a15120461
by Weizhen Xu, Chenyi Zhang *, Fangzhen Zhao and Liangda Fang
Reviewer 1:
Savitri Bevinakoppa
Reviewer 2: Anonymous
Algorithms 2022, 15(12), 461; https://doi.org/10.3390/a15120461
Submission received: 9 November 2022 / Revised: 29 November 2022 / Accepted: 2 December 2022 / Published: 6 December 2022
(This article belongs to the Topic Advances in Artificial Neural Networks)

Round 1

Reviewer 1 Report

Paper provides a good introduction on Deep Neural Networks (DNN) and the need for a new Mask-based Adversarial Defense scheme (MAD) to mitigate teh negative effect from teh attacks. Based on experimental results, proposed system on multiple copies of potential adversarial images, does not need any additional denoising structure or any change to a DNN’s architectural design.

Authors have written that future work will be on improvement of study of mask image. However, it would be good to expand the paragraph by providing more information on future work.

Make sure to have all references according to the MDPI specification.

Author Response

Thank you so much for your helpful suggestions. We have extended the future work section with additional information, and checked the references again to make sure that they meet MDPI specification.

Reviewer 2 Report

This paper proposes a new method (Mask-Based Adversarial Defense) to protect Deep Neural Networks from adversarial attacks. The proposed method is based on the random masking of potential adversarial images. The testing results show the effectiveness of the proposed method. As there is no noise added to the images, MAD does not need a denoising process compared with related works. The paper is interesting, well-written, and easy to follow. Furthermore, the proposed idea seems novel and valid. However, I have the following minor comments:

1- The paper can be organized better. It would be better to move mathematical concepts from the Introduction section to the background section. Moreover, a related work section can be added to include related research studies in more detail.  

2- The formal analysis of the proposed method is weak. I suggest authors try to prove the effectiveness of their proposed method mathematically. The current version discusses the claims only based on the experiments. 

 

Author Response

Thank you so much for your helpful suggestions. We have reorganized the structure of the paper, so that the submission now has a separate "related work" section, with more details supplied. Note that the intuition on how the proposed method works has been separately discussed in a semi-formal style at the beginning of Section 5. We also adjusted the position of some pictures and tables to avoid large blank sections. Given that it is difficult to formalize the structure of deep neural networks, we mainly prove the effectiveness of the proposed method through experiments, and leave the formal proof as a future work.

Back to TopTop