Theory-Guided Deep Learning Algorithms: An Experimental Evaluation

Round 1

Reviewer 1 Report

Review the Manuscript# 1886397

Theory-guided deep learning algorithms: an experimental evaluation

Comments

The main objective of the paper is to combine both data driven and the theory-guided techniques in one unified framework to achieve better results in data modeling.

The topic is interesting in this period, where data science and data analysis are at the peak.  The manuscript is well written and well organized.

-          The introduction presented the subject matter in a clear and sufficient manner, identified the main stream of the research problem, and research methodology.

-          Review of literature gives a brief state of the art applications and algorithms of both theory guided and data driven methods, with enough and recent references, if needed for more details.

-          The manuscript discusses the strengths and weakness of each methodology to establish the need for a third method, which utilizes the best of both current methods.  

-          Experimental results are illustrated through three examples :

Lake temperature, convicted movement in climate modeling, and Climate prediction.

For theory-guided method the authors, have selected the best papers that give both code and dataset for comparisons.

-          Comparison between GNN and CNN are made using time series graphs for each of the three problems.

-          The subjective comments, in the conclusion, substantiate the need for more work to be done in future research.

-          Thirty-five up to date relevant references are cited in the manuscript.

Although the paper, as I mentioned earlier, is well written and provides necessary but not sufficient evidence to promote the proposed method. More solid evidence, via results obtained by well-defined quantitative statistical techniques and measures are required to identify the significance of the proposed methods

Comments for author File: Comments.pdf

Author Response

We are very grateful for your comments and suggestions, which allowed us to improve the quality of our work. In the following, we address your concerns point by point. Updates in the revised manuscript are highlighted in blue.

Reviewer Point 1.1 — Although the paper, as I mentioned earlier, is well written and provides necessary but not sufficient evidence to promote the proposed method. More solid evidence, via results obtained by well-defined quantitative statistical techniques and measures are required to identify the significance of the proposed methods.

Reply: To provide statistical significance to the comparison of the results, all the experiments have been iterated over 10 repetitions with random initial seeds. Statistical variabilities are reported in the figures. To better emphasize such experimental setting, we revised the paper in Section 3 (Datasets and Methods) and Section 4 (Experimental results) to clarify the experimental pipeline.
Moreover, we extended the experimental result discussion to point out whether the comparisons are statistically significant or not

Author Response File: Author Response.pdf

Reviewer 2 Report

No comments -- paper is very well written and content is detailed. 

Author Response

We are very grateful for your evaluation.

Author Response File: Author Response.pdf

Reviewer 3 Report

Please see the attachment file.

Comments for author File: Comments.pdf

Author Response

We are very grateful for your comments and suggestions, which allowed us to improve the quality of our work. In the following, we address your concerns point by point. Updates in the revised manuscript are highlighted in blue.

Reviewer Point 3.1 — In lines 107 and 134, what is abbreviation of DL.
Reply: DL stays for ”Deep Learning”. The abbreviation was introduced on its first occurrence, in the first sentence of Section 1 (Introduction).
Reviewer Point 3.2 — What is the meaning of γ and λ in equation (1). Are they real numbers?
Reply: Yes, these values are real numbers that can be used as weights for the different contributions of the loss function. This remark has been added to the text.
Reviewer Point 3.3 — What is the meaning Re in equation (2).
Reply: The physical loss of equation (2) is intended to emphasize all density differences that are greater than zero, neglecting the others. To enforce this, ReLU activation functions (i.e. x if x > 0, 0 otherwise) are applied to all the elements of the summation. Such reasoning has been made more explicit in the text.
Reviewer Point 3.4 — What is the meaning ∇² in equation (5).
Reply: ∇ is the symbol of the gradient operator, which is applied to the velocities. This means that
this term is a second-time derivative.
Reviewer Point 3.5 — In reference 11, von Rueden should be Von Rueden.
Reply: The reference has been updated as suggested

Author Response File: Author Response.pdf

Reviewer 4 Report

Review report on “Theory-guided deep learning algorithms: an experimental evaluation”

In my opinion, the paper presents some details.

 A minor revision is required to make the document significance publishing.
Presently, I think this is a nice work. I urge the publication of the article after the following suggestions

1.     Justify the “has had” you have written in the first line of the abstract.

2.     Check the format of the journal and made all the references according
to the journal style.

3.     Please check the police stop and commas in the whole manuscript once again.

4.     "Conclusion" section should be added in a more highlighting, argumentative way.

 

5.     The authors should clearly emphasize the contribution of the study.

Comments for author File: Comments.pdf

Author Response

We are very grateful for your comments and suggestions, which allowed us to improve the quality of our work. In the following, we address your concerns point by point. Updates in the revised manuscript are highlighted in blue.
Reviewer Point 4.1 — Justify the “has had” you have written in the first line of the abstract.
Reply: We used the “has had” form since we initially intended to focus on past contributions. However, in the revised paper we fixed the sentence by removing the “had”, since such effects are still ongoing. Reviewer Point 4.2 — Check the format of the journal and made all the references according to the journal style. Reply: We performed an extra check on the references and fixed the issues, to provide a coherent format.
Reviewer Point 4.3 — Please check the police stop and commas in the whole manuscript once again.
Reply: A general proofreading of the paper has been performed and issues have been fixed.
Reviewer Point 4.4 — ”Conclusion” section should be added in a more highlighting, argumentative way.
Reply: We extended Section 5 (Conclusions) to make more evident the outcomes of the paper. A whole new paragraph has been added.
Reviewer Point 4.5 — The authors should clearly emphasize the contribution of the study.
Reply: We better emphasized the contribution of our work at the end of Section 1 (Introduction).

Author Response File: Author Response.pdf