Energy Efficiency Evaluation of Artificial Intelligence Algorithms†
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThe authors of this paper focus on energy efficiency evaluation of artificial intelligence algorithms and the related influence on green computing, which can be demonstrated by an improved empirical investigation on heuristic methods. The topic of this paper appears to be interesting and timely. However, there are several issues that need to be addressed.
1. After reading the paper, it is not clear to me what is the major contribution.
2. The motivation of the proposed method needs further explanation.
3. A clear logical clue is needed in the discussion of related work which should be expanded with more recent works.
4. The paper should read smoothly for the benefit of the reader.
5. More experiments and comparisons are needed which are in accordance with the engineering practice.
6. The conclusions should be expanded.
7. Some references are too old to effectively represent the latest research progress in this field, such as the references [4], [13], [24], and [25].
Comments on the Quality of English LanguageQuality of English language needs to be improved.
Author Response
Response to reviewer 1
Comments and Suggestions for Authors
The authors of this paper focus on energy efficiency evaluation of artificial intelligence algorithms and the related influence on green computing, which can be demonstrated by an improved empirical investigation on heuristic methods. The topic of this paper appears to be interesting and timely. However, there are several issues that need to be addressed.
- After reading the paper, it is not clear to me what is the major contribution.
Response
The article attempts to contribute to the discussion of computer energy efficiency drawing holistic generic abstraction based on key points in computing, namely, fundamental physical limitations, historical changes and improvement of hardware performance, requirements for software performance, illustration how different software could resolve the same tasks using different volume of resources and energy based on comparison of intelligent algorithms, discussion on intelligent algorithms, pointing out how efficiency can be improved on conceptual level.
Reflected within the article sections Abstract, Materials, Tools and Methods, and Conclusion.
- The motivation of the proposed method needs further explanation.
Response
Additional explanation is added in the beginning of section of section 2.
- A clear logical clue is needed in the discussion of related work which should be expanded with more recent works.
Response
Revised.
- The paper should read smoothly for the benefit of the reader.
Response
Revised.
- More experiments and comparisons are needed which are in accordance with the engineering practice.
Response
Main principle for this study is minimal use energy and other resources.
Comparison with other methods could be the subject of further research depending on available resources.
- The conclusions should be expanded.
Response
Conclusion is expanded.
- Some references are too old to effectively represent the latest research progress in this field, such as the references [4], [13], [24], and [25].
[13] is original classical source, which deserves attention
[24] and [25] provide ground and detailed information related with this publication, their absence may reflect negatively on correct apprehension of this study.
More recent publications are already added. These are: [27][39][45]
Comments on the Quality of English Language
Quality of English language needs to be improved.
Response
Revised.
Reviewer 2 Report
Comments and Suggestions for AuthorsThe title and Abstract are too generic; furthermore, the Abstract is poorly written and contains several grammatical errors. Some keywords are unnecessarily capitalized.
The Introduction, besides containing many grammatical errors, is relatively poor and generic and does not clearly state the specific objectives of the work.
In the Literature Survey, although 29 references are cited, only 9 of them are articles published in JCR-indexed journals. The literature review, in addition to being very poorly written, contains numerous grammatical errors, as well as incorrect placement of citations after the period in several paragraphs. Moreover, at the end of this section, research questions are posed that are incompatible with the objective defined in the Abstract and with the (unclear) focus mentioned at the end of the Introduction.
The Materials and Methods section was written equally poorly, and some sentences seem to make little sense. Additionally, it is written in a rather schematic manner, leaving several questions unanswered. Only seven test optimization problems were chosen, with no clear rationale for their selection and for there being only seven, as well as the reason for transforming minimization problems into maximization problems for their resolution by the three chosen metaheuristic algorithms; this choice was also not clearly justified. Also unjustified is the choice of the limit in each computational experiment being 100,000 iterations, which seems excessively high and somewhat unrealistic as a metaheuristic algorithm that requires such a large number of iterations to determine an approximation with a given accuracy (assuming the use of double-precision floating-point arithmetic) for the solution of such nonlinear optimization problems, even with dimension 100, should be considered rather inefficient; moreover, the accuracy of the approximations generated by the algorithms, i.e., the quality of the results, is not taken into account in the computational experiments. The choice of the Green Software Measurement Model is also not sufficiently justified. The manual recording of the start time, as well as the justification given for it, also seems out of place. Also not indicated (and justified) is the choice of programming language used in the implementation of the algorithms.
In the Results section, whose writing is equally poor, what stands out and demands justification is the fact that the PSO algorithm takes too much time when compared to the Free Search (FS) algorithm, proposed by the first author in a 2008 conference paper, the same being valid for the execution times of Differential Evolution (DE) compared to those of the FS algorithm.
The Discussion is also written in a rather succinct and equally poor manner, with an overuse of quotations, already observed in previous sections. Also to be clarified is the curious reference to a search space with a very specific dimension, 101,000,000.
The manuscript ends with a Conclusions section that is too brief and says nothing concrete about the objective mentioned in the Abstract, deviates from the focus mentioned in the Introduction section, and does not answer the research questions that appear at the end of the Literature Survey. By clearly stating that the experimental results illustrate the strengths and limitations of the three algorithms used, it seems to give the idea that this is the focus of the work carried out.
Finally, the references need to be carefully revised, standardized, and completed, as titles of some journals are written in full and others are abbreviated; incorrect abbreviations, a lack of standardization in the use of italics and capitalization, as well as incomplete references, are observed.
The manuscript is poorly written, with unclear and confusing sentences alongside numerous grammatical errors. These include issues with punctuation, concordance, tense consistency, verb forms, and typographical errors. The manuscript must be carefully reviewed by someone fluent in English.
Author Response
Response to reviewer 2
The title and Abstract are too generic; furthermore, the Abstract is poorly written and contains several grammatical errors. Some keywords are unnecessarily capitalized.
Response
Revised.
The Introduction, besides containing many grammatical errors, is relatively poor and generic and does not clearly state the specific objectives of the work.
Response
Revised.
In the Literature Survey, although 29 references are cited, only 9 of them are articles published in JCR-indexed journals. The literature review, in addition to being very poorly written, contains numerous grammatical errors, as well as incorrect placement of citations after the period in several paragraphs.
Response
Revised.
Moreover, at the end of this section, research questions are posed that are incompatible with the objective defined in the Abstract and with the (unclear) focus mentioned at the end of the Introduction.
Response
Revised.
The Materials and Methods section was written equally poorly, and some sentences seem to make little sense. Additionally, it is written in a rather schematic manner, leaving several questions unanswered.
Response
This section is revised with more content added.
Only seven test optimization problems were chosen, with no clear rationale for their selection and for there being only seven, as well as the reason for transforming minimization problems into maximization problems for their resolution by the three chosen metaheuristic algorithms; this choice was also not clearly justified. Also unjustified is the choice of the limit in each computational experiment being 100,000 iterations, which seems excessively high and somewhat unrealistic as a metaheuristic algorithm that requires such a large number of iterations to determine an approximation with a given accuracy (assuming the use of double-precision floating-point arithmetic) for the solution of such nonlinear optimization problems, even with dimension 100, should be considered rather inefficient; moreover, the accuracy of the approximations generated by the algorithms, i.e., the quality of the results, is not taken into account in the computational experiments.
Response
Section 2 is revised and more clarity added.
The choice of the Green Software Measurement Model is also not sufficiently justified. The manual recording of the start time, as well as the justification given for it, also seems out of place. Also not indicated (and justified) is the choice of programming language used in the implementation of the algorithms.
Response
Section is revised accordingly.
In the Results section, whose writing is equally poor, what stands out and demands justification the fact that the PSO algorithm takes too much time when compared to the Free Search (FS)algorithm, proposed by the first author in a 2008 conference paper, the same being valid for the execution times of Differential Evolution (DE) compared to those of the FS algorithm. The Discussion is also written in a rather succinct and equally poor manner, with an overuse of quotations, already observed in previous sections.
Response
Section is revised accordingly.
Also to be clarified is the curious reference toa search space with a very specific dimension, 101000000.
Response
Revised. It should be 101000000
The manuscript ends with a Conclusions section that is too brief and says nothing concrete about the objective mentioned in the Abstract, deviates from the focus mentioned in the Introduction section, and does not answer the research questions that appear at the end of the Literature Survey. By clearly stating that the experimental results illustrate the strengths and limitations of the three algorithms used, it seems to give the idea that this is the focus of the work carried out.
Response
Section is revised accordingly.
Finally, the references need to be carefully revised, standardized, and completed, as titles of some journals are written in full and others are abbreviated; incorrect abbreviations, a lack of standardization in the use of italics and capitalization, as well as incomplete references, are observed.
Response
Revised.
The manuscript is poorly written, with unclear and confusing sentences alongside numerous grammatical errors. These include issues with punctuation, concordance, tense consistency, verb forms, and typographical errors. The manuscript must be carefully reviewed by someone fluent in English.
Response
Revised.
Reviewer 3 Report
Comments and Suggestions for AuthorsAn empirical evaluation and comparison of intelligent and adaptive algorithms time and energy consumption is applied to heterogeneous numerical tests. Comparison of results is performed.
Remarks:
1. Fixing population size to 10 for all algorithms is too strong and not justified limitation. Actually, some methods perform better with less some with bigger population size. As rule the evolutionary methods can be tuned for particular problems. It is suggested to use different population sizes (say 10,20,30,40) and compare the results.
2. „Each experiment is limited to 100000 iterations“. Why 100000 iterations are used for each algorithm? Actually the aim of the algorithm is to find out optimal solution and end work when it is found, not just to compute and compute. In such way we will see which algorithm outperform others. Note that here number of iterations needed depend on population size used.
3. The widely used metric „computational complexity“ or „time complexity“ of the algorithm is not even discussed. This metric is based on number of main operations needed to apply algorithm and it is most commonly used for deterministic algorithms (see https://doi.org/10.1016/j.mtcomm.2020.101290 ) rather than metaheuristic algorithms. Iti s suggested to at least discuss metric „computational complexity“ in introduction.
Author Response
Response to reviewer 3
An empirical evaluation and comparison of intelligent and adaptive algorithms time and energy consumption is applied to heterogeneous numerical tests. Comparison of results is performed.
Remarks:
- Fixing population size to 10 for all algorithms is too strong and not justified limitation. Actually, some methods perform better with less some with bigger population size. As rule evolutionary methods can be tuned for particular problems. It is suggested to use different population sizes (say 10,20,30,40) and compare the results.
Response
The aim is to evaluate the process performance. How much energy a computer will use if it run this software. Clarification is added, in section 2.
- „Each experiment is limited to 100000 iterations“. Why 100000 iterations are used for each algorithm? Actually the aim of the algorithm is to find out optimal solution and end work when it is found, not just to compute and compute. In such way we will see which algorithm outperform others. Note that here number of iterations needed depend on population size used.
Response
The aim, is to evaluate the process performance. How much energy a computer will use if it run this software. Clarification is added in section 2.
- The widely used metric „computational complexity“ or „time complexity“ of the algorithm is not even discussed. This metric is based on number of main operations needed to apply algorithm it is most commonly used for deterministic algorithms (see https://doi.org/10.1016/j.mtcomm.2020.101290 ) rather than metaheuristic algorithms. Iti suggested to at least discuss metric „computational complexity“ in introduction.
Response
Evaluation of energy efficiency as a function of „computational complexity“ and „time complexity“ is an excellent idea and could be a subject of further research.
The scope of this comparison as illustration is limited to evaluation of software process energy efficiency.
Clarification is added in Section 2.
Reviewer 4 Report
Comments and Suggestions for AuthorsDear authors, I have some technical remarks regarding the template.
Please check the title as it spans to 2 lines, I guess it should not be interupted by break.
Line 44 - I think it is not pro-cessing but processing
Line 122 - "paters" should be patterns in my opinion
The Conclusion is a little bit short.
Author Response
Response to reviewer 4
Dear authors, I have some technical remarks regarding the template.
Please check the title as it spans to 2 lines, I guess it should not be interrupted by break.
Response
Revised, Thank you.
Line 44 - I think it is not pro-cessing but processing
Response
Revised, Thank you.
Line 122 - "paters" should be patterns in my opinion
Response
Revised, Thank you.
The Conclusion is a little bit short.
Response
Revised, Thank you.
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsThe issues raised in the first round of review were addressed adequately or acceptably by the authors. The text has been thoroughly revised, rewritten, and supplemented, with no grammatical errors present.
It is worth noting two instances of American spelling instead of British, which should be corrected: "Optimization" -> "Optimisation" on line 245, and "centers" -> "centres" on line 43.
Author Response
Comments on the Quality of English Language
It is worth noting two instances of American spelling instead of British, which should be corrected: "Optimization" -> "Optimisation" on line 245, and "centers" -> "centres" on line 43.
Response
Revised
