Automated Design of the Deep Neural Network Pipeline

Round 1

Reviewer 1 Report

The information shown in section 3.1.2 can be better presented using a table, also the information can be better organized and less tired to read.

The acronyms must be defined before their usage for readers not familiar with deep learning.

On page 10 it is not necessary to repeat after the subtitle "Experiment 2" and then again "Experiment 2"

More explanation about the SPHH will help to reproduce similar results.

 

 

Author Response

Please see the attachment. 

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors employed a selection perturbative hyper-heuristic for the automated design of the deep neural network pipeline. The approach was evaluated for sentiment analysis, spam detection, maize disease detection and oral lesion detection. All tests used available datasets.

 They argue that, from the current state of the literature, no research has been done pertaining to the use of transfer learning in automated design, making this work the first study to examine the effect of transfer learning on the automated design of the deep neural network pipeline. This seems to be plausible considering the presented paper and the extant literature cited.  

Please see below my main concerns:

 

- There are some important statements with no references. I attached a file then you can revise all of them;

- The generated designs were not reusable. The authors should explain in detail this limitation, as well describing how a general approach could be developed;

- Review all acronyms used in your text. Even the more obvious should be described the first time they are presented;

- The subsections of “3” are purely descriptive. Especially, a figure to represent SPHH is required in 3.1. Bear this will clarify the author’s approach for a more general audience.

- Minor revisions: please see the attached file.

Comments for author File: Comments.pdf

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Now it is easier to follow your ideas than before, and it is possible to see your effort to make the corrections suggested.