CS2125 Paper Review Form - Winter 2019

Reviewer: Mohammad Rashidujjaman Rifat

Paper Title: Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Images.

Author(s):

1) Is the paper technically correct?
 [✓] Yes
 [ ] Mostly (minor flaws, but mostly solid)
 [ ] No

2) Originality
 [ ] Very good (very novel, trailblazing work)
 [✓] Good 
 [ ] Marginal (very incremental)
 [ ] Poor (little or nothing that is new)

3) Technical Depth
 [ ] Very good (comparable to best conference papers)
 [✓] Good (comparable to typical conference papers)
 [ ] Marginal depth
 [ ] Little or no depth

4) Impact/Significance
 [ ] Very significant
 [✓] Significant
 [ ] Marginal significance.
 [ ] Little or no significance.

5) Presentation
 [ ] Very well written
 [ ] Generally well written
 [✓] Readable
 [ ] Needs considerable work
 [ ] Unacceptably bad

6) Overall Rating
 [ ] Strong accept (award quality)
 [✓] Accept (high quality - would argue for acceptance)
 [ ] Weak Accept (borderline, but lean towards acceptance)
 [ ] Weak Reject (not sure why this paper was published)


7) Summary of the paper's main contribution and rationale
   for your recommendation. (1-2 paragraphs)

This paper presents a study that shows an instance of false-positive decision making by a Deep Neural Network (DNN). Particularly, to show the false-positiveness for unrecognizable images, the paper chooses state-of-the-art DNN architectures including AlexNet, LetNet, and MNIST for the experiment. The contribution goes to the understanding of the differences between human and DNNs ability to vision that might create awareness of difficulties in the area of the applications of computer vision using DNNs. In addition, such awareness of fooling instances despite using strong generative DNN models will lead to the new computational techniques to overcome them.


8) List 1-3 strengths of the paper.  (1-2 sentences each,
identified as S1, S2, S3.)

S1. I find the paper technically strong and clearly articulated in the paper.
S2. The discussion of generative vs. discriminative models and their susceptibility to the fooling instances was useful.
S3.The paper points out the application areas that might be concerned (such as face and image recognition, biometric, search engine ranking, etc.) with such fooling events.

9) List 1-3 weaknesses of the paper (1-2 sentences each,
identified as W1, W2, W3.)

W1. This is a weakly written paper. The abstract is fully copied in the introduction. The subtitles 3.1 and 3.2 are exact copies of each other.