IE-Net: Information-Enhanced Binary Neural Networks for Accurate Classification
Round 1
Reviewer 1 Report
See my attached report.
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
The manuscript proposes a binary neural network called information enhanced binary neural network, designed to reduce the loss of information generally caused by the binarized weights and activations. The paper is well written and the contribution, despite being incremental, is clearly stated. Experiments show a slight performance improvement of the proposed solution over the existing methods to which the authors compare, without increasing the computational complexity. There are a few changes which need to be done before publication, and the comments are provided below.
- Please check the caption of Fig. 8; there seems to be an error in the description of the curves.
- There are some writing errors throughout the paper, which must be corrected before publication (a few examples: the abbreviation SOTA is defined twice, in section 1 and again in section 3; the abbreviation FLOPs is defined three times, in sections 1, 2, and 3; the first keyword is “binary”, etc.).
- In addition, the paper has numerous minor English language flaws. Hence, a thorough analysis is required, in order to provide the necessary corrections.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors have not added any of the references that I suggested and that I consider needed in this paper. I will not accept this paper till have a new version of it.