Datasets:

CleverShovel

/

paper_reviews

like 1

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: DEEP LEARNING-BASED ELECTROCARDIOGRAM RHYTHM AND BEAT FEATURES FOR HEART ABNORMALITY CLASSIFICATION Review round: 1 Reviewer: 2

Basic reporting: In this manuscript, Darmawahyuni and colleagues evaluated the performance of 1D-CNN model on ECG classification. The authors reported that their model demonstrated high classification performance on both rhythm and beat abnormalities. The reviewer agrees with the clinical importance of simultaneously identifying ECG waveforms and rhythms, but unfortunately the authors' method fails to achieve their goal. The authors stated that they were able to classify 24 heart abnormalities (9 rhythms and 15 beats), but in reality, they only repeated the binary classification 24 times. This is the main drawback of this study. As a side note, the reviewer believes that confusion matrices, not ROC curves, should be used to evaluate the performance of multiclass classification. The definition of the classifications for cardiac rhythm is also incorrect. According to the authors, the rhythm abnormality includes “Myocardial Infarction”, Myocarditis”, “Heart Failure”, “Valvular heart disease”, “Hypertrophy” and “Cardiomyopathy”, none of these are abnormalities in heart rhythm. Furthermore, since myocardial infarction and heart failure often coexist, and valvular disease and cardiac hypertrophy can also coexist, classifying them as separate classes (based on ECG alone) is obviously a wrong approach. In such a complex task, it is impossible for all classification models to achieve AUROC 1.0, as shown in Figure 6. The reviewers speculate that these mistakes may be partly due to the lack of clinicians among the authors. This manuscript needs to be radically revised from the research design. Experimental design: Experimental design is inappropriate as described above. Validity of the findings: Findings are unreliable as described above. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: DEEP LEARNING-BASED ELECTROCARDIOGRAM RHYTHM AND BEAT FEATURES FOR HEART ABNORMALITY CLASSIFICATION Review round: 2 Reviewer: 1

Basic reporting: The manuscript is well-written, has a good presentation, and thoroughly covers the literature. Experimental design: The experiments have been correctly and thoroughly defined and conducted. Validity of the findings: Results seem good, and the experiments use plenty of databases. The comparison with the state-of-the-art remains flawed and does not suffice to assess the relative quality of the proposed method. The most promising literature approaches should be tested in the exact same conditions as the proposed method to directly compare their performance results. Additional comments: I thank the authors for their response to my comments. I believe most have been addressed, although one key concern has been avoided (and the manuscript's quality has suffered because of that). Direct comparison with the state-of-the-art is fundamental. If you have more classes than the literature, or other things are different (random seeds, etc.), then implement state-of-the-art methods and test them in your exact scenario. If there are many competing works in the literature, select at least the 1-2 best to have this direct benchmarking. This is essential to have a real fair comparison and to really understand if, where, and how your method is best.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: DEEP LEARNING-BASED ELECTROCARDIOGRAM RHYTHM AND BEAT FEATURES FOR HEART ABNORMALITY CLASSIFICATION Review round: 3 Reviewer: 1

Basic reporting: The manuscript is well-written, has a good presentation, and thoroughly covers the literature. Experimental design: The experiments have been correctly and thoroughly defined and conducted. Validity of the findings: Results seem good, and the experiments use plenty of databases. The manuscript now includes direct comparison with state-of-the-art methods to objectively illustrate the superiority of the proposed method. Additional comments: I thank the authors for addressing my comments. I believe it is now publishable. Congratulations on a good piece of research!

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: RESEARCH ON REMOTE SENSING IMAGE EXTRACTION BASED ON DEEP LEARNING Review round: 1 Reviewer: 1

Basic reporting: This paper mainly focused on the classification of remote sensing images by carrying out atmospheric calibration, band combination, image fusion and other data enhancement methods for Landsat 8 satellite remote sensing data to improve the data quality. In addition, deep learning and spatial/channel attention are applied to remote sensing image block segmentation in this paper. Although the experiments show the effectiveness of this method, there are still some major concerns that need to be carefully clarified and revised before considering a possible publication. Experimental design: 1.The proposed method consists of several important components, so please add ablation experiments to evaluate the importance of each component. How each component contributes to the final performance gain? Especially the AC-CBAM compared to CBAM. 2.“Results and discussions” section is too short. Please give more experimental results and deep discussions of the results. Especially the deep reasons for the gains of the results. Validity of the findings: 1.There exist many grammar errors throughout the paper, which severely disturbs the readability. I strongly suggest the author to carefully revise the paper for English grammar, the choices of words, the sentence structure, and the use of articles (a native speaker is strongly recommended for this task). 2.In my opinion, the novelty of this paper is somewhat marginal, so the authors need to describe the main contributions of this paper clearly in the Introduction section. 3.What is the full name of AC-CBAM? Abbreviations should be defined at first mention and used consistently thereafter. 4.The authors should clearly highlight the difference between CBAM and AC-CBAM. 5.A deep literature review should further given, especially regarding the topics involved in this paper. Therefore, the reviewer suggests discussing the advances by citing some references, e.g., “When deep learning meets metric learning: remote sensing image scene classification via learning discriminative CNNs”, “Remote sensing image scene classification meets deep learning: challenges, methods, benchmarks, and opportunities”, and “Feature enhancement network for object detection in optical remote sensing images”. 6.There is a typo in the caption of Table 6. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: RESEARCH ON REMOTE SENSING IMAGE EXTRACTION BASED ON DEEP LEARNING Review round: 1 Reviewer: 2

Basic reporting: The traditional remote sensing image segmentation technology cannot make full use of the rich spatial information of the image, the workload is too large, and the accuracy is not high enough. Aiming at these problems, this paper carried out atmospheric calibration, band combination, image fusion and other data enhancement methods for Landsat 8 satellite remote sensing data to improve the data quality. In addition, deep learning is applied to remote sensing image block segmentation in this paper. Based on the full convolutional neural network of codec structure, AC-CBAM innovative structure is proposed, and the optimization module of integrated attention and sliding window prediction method are adopted to effectively improve the segmentation accuracy. This paper has done some work, however, there are at least the following problems. 1) The format of English is nonstandard, for example, in Introduction, the first paragraph is too long, while other paragraphs are too short. In addition, in row 33,“In the experiment of test data, the models mIoU, mAcc and aAcc in this paper reach 97.34%, 98.66% and 98.67% respectively, which is 1.44% higher than DNLNet (95.9%), which provides a reference for deep learning to realize the automation of remote sensing land information extraction”, it is unusual for“ which” appearing twice continuously, etc. 2) The English expression is not accurate enough, much of the translations fails to conform to English expression and culture connotation, resulting in poor readability, and it is best to find professionals to improve the language expression. For example, in Row 50 of 4.3 Slide Window Prediction , “And many other constraints and prior information to improve the performance of semantic segmentation”. In row 290/ 291, “finally got good spatial resolution and retained the images of the multispectral image information, to ultimately improve the effect of plot extraction have good role in promoting” , etc. Experimental design: no comment Validity of the findings: no comment Additional comments: no comment

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: DNN-BASED MULTI-OUTPUT MODEL FOR PREDICTING SOCCER TEAM TACTICS Review round: 1 Reviewer: 1

Basic reporting: > The authors proposed to use Deep Neural Networks and feature engineering to predict the soccer tactics of teams, such as formations, game styles, and game outcome based on soccer dataset. > The authors stated that experimental results demonstrated significant improvements of the proposed model compared to baseline models. > The manuscript is literature supported. > A new English revision is encouraged, e.g: Line 24: “the proposed model, which obtain” , missing “s” Line 66: “We propose Multi-Output” , missing “a” Line 67: “game outcome; The” , missing ".” Line 76: “We propose deep learning” , missing “a” Line 195: “determining a team’s tactics” , remove “a” Experimental design: > Source code and data source were provided. Also, Python was used as programming language, which helps to reach a bigger audience. > “Our model trains a dataset consisting of 380 games and tests the 233 model with a dataset consisting of 38 games.” It seems that hold-out validation was applied, or a single cross-validation. However, it is well-known that there are other more robust validation methods, such as leave-one-out and k-fold cross-validation. Also, the raw data provided in https://github.com/Lee-Gunju/Multi-Output-Model-for-Soccer-based-on-DNN-and-Feature-Engineering-for-Predicting-Team-Tactics do not seem difficult to process. Can the authors extend the experimental study? For example, by adding new datasets and using a different validation method. > “Moreover, we have developed the visualization system” Apologies, during the revision, the provided system was unavailable (http://recsys.cau.ac.kr:8092/). Please check the provided figures. Validity of the findings: > After reviewing and testing the provided source code (https://github.com/kecau/Multi-Output-model-for-Soccer). Please update how the data is loaded in Line 10, since xlrd has explicitly removed support for anything other than xls files. You can follow this link https://stackoverflow.com/questions/65254535/xlrd-biffh-xlrderror-excel-xlsx-file-not-supported. > Source code is missing from the file “Multi-Output model for Soccer.py”. For example, when executing your script the following error is seen: “NameError: name 'train_dt' is not defined”. > Why the 2nd output (game styles) used a softmax instead of a sigmoid activation? > Regarding the game outcome performance, the results achieved are discrete (49.39% recall). However, the authors surpassed the baseline models. The authors should extend the experimental analysis in order to better support their proposal. > Please include the individual sensitivity rates achieved for every outcome in the analysis in order to have a complete picture of your proposal, e.g. win, draw, loss, offensive, defensive. > The conclusion should be stated after performing more experiments with new datasets and state-of-the-art models, if possible. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: DNN-BASED MULTI-OUTPUT MODEL FOR PREDICTING SOCCER TEAM TACTICS Review round: 1 Reviewer: 2

Basic reporting: The general writing style in this paper is legible and understandable but the authors sometimes make common grammar mistakes such as omitting the "s" in first person singular verbs or omitting some articles that should be present. The Introduction section is well written and describes the motivation and the objective of the study. I believe that the sentence about representing each segmented position by 0 or 1 (lines 59-61) is difficult to understand: where are these characteristics represented and how are they related to each position? The references along the article are appropriate. The article follows the common section structure in computer science (and more loosely the PeerJ standard), including a section for related work, sections to explain the proposed solution and a section for results and discussion. Figures are of good quality in general. I would suggest adjusting the darkness of the green color used in Figures 1 and 4 (make it lighter so that there is a higher contrast between the background and the foreground text) as well as making legends in Figure 4 way larger, as they are currently illegible. Figure 3 should include a description of the numbers in parentheses across the image, are they the dimension of each layer output? The colors applied to the layers should be described as well if they hold some meaning, e.g. are all red layers a dense feed forward operation? Original data appears to have been collected from a website, although there is no apparent direct download for the datasets, so the authors must have composed the dataset themselves out of the statistics available. The dataset has been published by the authors in a GitHub repository. Experimental design: This article seems to fit within the scope of journal PeerJ Computer Science, as it applies widely known machine learning methods to a concrete problem and develops improvements on the models and a specific solution to tackle the training and prediction for this kind of data. The authors aim to tackle a problem of modeling (european) football plays using more modern tools (neural networks) that have not been used yet for this objective. The fundamental part of the model description is Figure 3 which details the structure of the neural network used, but I think it could be clearer if it could somehow indicate the number of actual dense layers that are used (I believe it could be every red and yellow layer, but it should be clearer if some researcher wants to reproduce this work). I think that the learning rate used in the Adam optimizer (0.01, Table 6) is much higher than the standard 0.001, is there a reason behind this? Validity of the findings: The resulting metrics seem reasonable and the stated conclusions are supported by the results. Since the baseline model performs better than your newly proposed model (when both are without feature engineering), could it improve further than yours when the engineered features are provided? Additional comments: My initial concern with the approach of the authors was that feature engineering is supposed to be less useful with deep learning models but, in this case, the results show that this step is key in finding a well-performant model. My question now is whether the feature engineering is providing more improvements than the predictor itself and, as a result, if one could achieve even better metrics by providing the engineered metrics to another predictor (e.g. the baseline). The overall quality of the article, however, is good and I would consider it ready for acceptance after fixing the details I indicated above.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: DNN-BASED MULTI-OUTPUT MODEL FOR PREDICTING SOCCER TEAM TACTICS Review round: 2 Reviewer: 1

Basic reporting: The manuscript is acceptable in its current form. Experimental design: The authors have responded all the previous queries/revisions. Validity of the findings: The authors surpassed the baseline models and included a full performance analysis of their proposal. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: DNN-BASED MULTI-OUTPUT MODEL FOR PREDICTING SOCCER TEAM TACTICS Review round: 2 Reviewer: 2

Basic reporting: no comment Experimental design: no comment Validity of the findings: no comment Additional comments: The authors have addressed all of my concerns adequately in the rebuttal. I no longer hold any issues against the publication of the article.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: TEMPORAL NETWORK EMBEDDING FRAMEWORK WITH CAUSAL ANONYMOUS WALKS REPRESENTATIONS Review round: 1 Reviewer: 1

Basic reporting: No comments. Language is in general clear. Sufficient background is given. Terms are defined before use. Experimental design: Experimental designs make sense and cover a wide range of problems. Experimental details are sufficient. Validity of the findings: The findings are valid. Experimental results support the findings. Additional comments: N/A

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: TEMPORAL NETWORK EMBEDDING FRAMEWORK WITH CAUSAL ANONYMOUS WALKS REPRESENTATIONS Review round: 1 Reviewer: 2

Basic reporting: Easy to understand, interesting, relevant and convicing. Experimental design: well defined, relevant, meaningful and convincing. Validity of the findings: The findings are well presented, easy to understand and convincing. Also the conclusions are well stated and clearly support the assumption given before. Additional comments: As the authors correctly present, many machine learning tasks for graphs are increasingly prevalent, especially in link prediction and node classification, and indeed the challenge becomes greater when analyzing the dynamic (i.e., temporal) network. Therefore, in this paper, the authors propose a new approach to learn dynamic representations based on temporal graph networks by introducing a message generation function by extracting causal anonymous paths. To evaluate their novel approach, the authors present a benchmark pipeline for evaluating temporal network embeddings. Truly, this is a very interesting, relevant and also important approach. The authors also nicely demonstrate the applicability of their model in the real downstream graph of one of the leading European banks performing credit scoring based on transaction data - which is a good application. This reviewer is very positive and recommends the acceptance of this article and only recommends to make a few minor improvements: 1) page 4, section 3.1. line 157, indeed graph neural networks are enormously important for this approach and some very good background work has also been listed, here for completeness a very recent and already highly cited related work should be cited and that is by Holzinger et al (2021), Towards multi-modal causability with Graph Neural Networks enabling information fusion for explainable AI, https://doi.org/10.1016/j.inffus.2021 .01.008 - this reference is in fact an important indication for the interested reader of the possibility of traceability and understandability (which is currently subsumed under explainability), which is also interesting for further work in the context of this application domain (baking, finance). 2) Figure 1 is very good, but the small text in the boxes (this reviewer printed the work on paper) is practically impossible to read. (Figure 2 is perfect) 3) Table 1 is also difficult to read 4) Figure 3 needs a legible caption and for faster reader acquisition a better caption. (Figure 4 again is good) Generally a very good paper and this reviewer congratulates the authors.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: TEMPORAL NETWORK EMBEDDING FRAMEWORK WITH CAUSAL ANONYMOUS WALKS REPRESENTATIONS Review round: 1 Reviewer: 3

Basic reporting: The paper is well organized but the presentation could be improved. References can provide sufficient field background. The article structure, figures and tables are clear, and the raw data and code are also provided. In addition, there are some problems with this paper. Q1: The second half of Equation 2 is somewhat difficult to understand, and I hope to see a more intuitive definition of “the t-timed k-hop neighborhood”. Q2: The calculation of time difference t (Line 274, Page 7/18) is not given, which can be further explained. If t is calculated directly, perhaps it means that the earlier the node appears, the higher the probability of being selected? Q3: In Equation 12, there are two feature vectors, how are they obtained? In addition, what is the difference between the feature vectors at different times? Q4: In Section 3, the paper gives abbreviations for some methods (such as DeepWalk, EvolveGCN and CAW), but not for others, and I suggest consistency in formatting. Q5: There are also some new works can be introduced in Section 3, including but not limited to the following works. [1] Jiancan Wu, Xiang Wang, Fuli Feng, Xiangnan He, Liang Chen, Jianxun Lian, and Xing Xie. 2021. Self-supervised Graph Learning for Recommendation. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR '21). [2] Meng Liu and Yong Liu. 2021. Inductive Representation Learning in Temporal Networks via Mining Neighborhood and Community Influences. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR '21). [3] Menglin Yang, Min Zhou, Marcus Kalander, Zengfeng Huang, and Irwin King. 2021. Discrete-time Temporal Network Embedding via Implicit Hierarchical Learning in Hyperbolic Space. In Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining (KDD '21). [4] Hu, Linmei and Li, Chen and Shi, Chuan and Yang, Cheng and Shao, Chao. 2020. Graph neural news recommendation with long-term and short-term interest modeling. Information Processing & Management. Q6: A method abbreviation (such as DeepWalk) may help your paper spread more effectively. Experimental design: The paper's research matches the aims and scope of the journal. Research question well defined, relevant, and meaningful. The paper gives the raw data and code, and provides enough details to reproduce the model. The paper conducts experiment on several datasets and discusses the experimental results to demonstrate the validity of the work. In addition, there are some problems with this paper. Q7: The paper conducts experiment on link prediction and node classification tasks in transductive and inductive settings. I would like to see the construction details for the downstream tasks described further. Q8: In Line 348, Page 10/18, the paper mentions that “the number of negative samples is equal to the number of edges”. I am interested to see a further explanation of it. Validity of the findings: The paper presents an innovative combination of CAW and TGN methods, and the experimental results demonstrate its effectiveness. All underlying data have been provided, and conclusions are well stated. In addition, there are some problems with this paper. Q9: The paper seems to embed CAW as a module in the structure of TGN, does it lead to a high complexity? I would like to see a discussion on the complexity of the model. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: TEMPORAL NETWORK EMBEDDING FRAMEWORK WITH CAUSAL ANONYMOUS WALKS REPRESENTATIONS Review round: 2 Reviewer: 1

Basic reporting: The authors have taken into account the reviewer's comments and addressed them appropriately, and the reviewer would now recommend that the paper be accepted. Experimental design: n.a. Validity of the findings: n.a. Additional comments: Of course run the usual final spell checks when going into production.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: TEMPORAL NETWORK EMBEDDING FRAMEWORK WITH CAUSAL ANONYMOUS WALKS REPRESENTATIONS Review round: 2 Reviewer: 2

Basic reporting: All my questions have been explained in the revision. No other comments. Experimental design: Experimental designs make sense and cover a wide range of problems. Experimental details are sufficient. Validity of the findings: The findings are valid. Experimental results support the findings. Additional comments: N/A

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A DEEP LEARNING METHOD FOR THE RECOGNITION OF SOLAR RADIO BURST SPECTRUM Review round: 1 Reviewer: 1

Basic reporting: Line 31: Solar radio burst is (not originates from) radio wave signal. Solar radio bursts originate from energetic electrons. Line 32: Unit of frequency should be Hz. Meter kilometer-wave is the wavelength. Line 36: Height (or heliocentric distance) of solar radio bursts is not determined by propagation characteristics, it’s determined by emission mechanism and background plasma property. Line 40: Description not correct, the author of “Tan et al 2007” is not “Australian astronomers”. And “Tan et al 2007” in the reference list doesn’t have referred journal volume information. Line 62: [Scholars] -> [researches] or [previous works] or [works in literature]. (this problem happened multiple times in the paper, please correct them thorough) Line 71: Citation form: Author-year Line 92: gray-scale value after normalization could be out of the range [0,255], does here include change of data type (from uint8 to another type), or how to rebin the normalized value to the range of [0,255] Line 115: Redundant description of a common method widely used in computer vision, please reduce the description length and cite the original paper of morph-close [Image Analysis and Mathematical Morphology by Jean Serra (1982)] Line 140-202: Redundant description of CNN, please replace this part with some citation (e.g: ImageNet classification with deep convolutional neural networks 2012 [proceedings-neurips] A Krizhevsky et al) or any textbook introducing CNN, GRU is well described in the paper of [Cho, Kyunghyun, et al 2014] Line 217: delete line break Line 309-316: not the conclusion of this paper. [For the basic concept of solar radio bursts, I highly recommend the author to read the book: Introduction to Solar Radio Astronomy and Radio Physics] Experimental design: (1) I wouldn’t suggest the authors claim they proposed a “new” method or model, as the CNN-GRU network is already widely used, for example: [Hybrid CNN-GRU model for high efficient handwritten digit recognition 2019 Vantruong Nguyen] and [Detecting Hate Speech on Twitter Using a Convolution-GRU Based Deep Neural Network] (2) The description of the dataset is important, please elaborate on the data preparation procedure: how the data is labeled and segmented, input size of the network is 120*120 while the data size is 120*2520. Please state how 120*2520->120*120 is achieved, either one dynamic spectrum is divided into 21 standalone images or the image is rescaled? Downsampling rescale will cause the loss of resolution and the loss of fine structures. (This paper is about fine structures, hence should be sensitive to resolution, so I wouldn't suggest any sort of downsampling before the input of the network). (3) The detail of affine transformation needs to be elaborated. (4) Affine transformation may not be appropriate in the context of solar radio bursts, because the type of solar radio burst is determined by its shape and frequency drift rate, and affine transformation of the dataset may change the shape and frequency and create a dynamic spectrum that won’t exist in the real world. This will make the model perform largely differently in the authors’ dataset and fresh new data. Validity of the findings: (1) In this work, the output information is limited: [Event/No-event/Calibration]. Namely the existence of solar radio bursts. The output doesn’t include the information of the type of solar radio bursts and the position in the dynamic spectrum. But actually, the information of [what type, when, and which frequency range] is more useful. There are already methods that can obtain the type and time-frequency information of radio bursts: https://www.frontiersin.org/articles/10.3389/fphy.2021.646556/full https://www.swsc-journal.org/articles/swsc/pdf/2018/01/swsc170092.pdf https://meetingorganizer.copernicus.org/EGU2020/EGU2020-5109.html if the method can not be improved to output the type and time-frequency information, please at least discuss it in [future work] (2) There should be a discussion session (3) It would be better if the event list in this work can also be published Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A DEEP LEARNING METHOD FOR THE RECOGNITION OF SOLAR RADIO BURST SPECTRUM Review round: 1 Reviewer: 2

Basic reporting: Comments to the Author A paper documenting a deep learning technique for the fine structure recognition of solar radio spectrum burst is welcomed. Your manuscript contains much useful information, but I found some aspects requiring more thorough expressed. Here are my specific comments and suggestion： 1) The title may be propitiate to change to "solar radio burst spectrum". 2) Line 39-40, Line 309-316 have little relevance with the subject of this paper. I suggest removing them. 3) Line 192-193, eq(12)-eq(13) have bad display. 4) Line 216-217, Line 299-301 should be concatenated in one line. 5) Line 248, FN is defined twice. 6) Line 286, Line 289, it is hardly to see the where the Ref.12 locates in Table 6. 7) Some abbreviations should be fully explained when appearing in the first time. Such as CGRU, CBIGRU, TPR, FPR, TP, FN. 8) Line 319, what does the unbalanced sample distribution mean? It should be explained more thoroughly in advance. Experimental design: No comments. Validity of the findings: The author draws a conclusion that CGRU in this paper is more sensitive to small features of solar bursts than other methods. If it was drawn from Table 6, I suggest the author adds more support data such as the classification to complex features or structures in all kinds of solar bursts to make the conclusion solid. Additional comments: No comments.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A DEEP LEARNING METHOD FOR THE RECOGNITION OF SOLAR RADIO BURST SPECTRUM Review round: 1 Reviewer: 3

Basic reporting: The manuscript proposes a CGRU model for solar spectrum image analysis, especially for the recognition of burst radio spectrum. Owing to the preprocessing and GRU, experiments demonstrates that the proposed method is very efficient with high accuracy. Therefore, it gives new light on solar image understanding. However, it would be better to further explain the following: 1) Give the overview of the model architecture or/and the table of model hierarchy structure to clearly show that the model is much simple and efficiency; 2) Some typing errors need attentions, such the equation (12) and (13). The “FN” in line 249th seems be a mistake of “TN”. Experimental design: no comment Validity of the findings: no comment Additional comments: no comment

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A DEEP LEARNING METHOD FOR THE RECOGNITION OF SOLAR RADIO BURST SPECTRUM Review round: 2 Reviewer: 1

Basic reporting: Line 176, what’s the meaning of left right rotation, is it left and right polarization? Figure 1, the y-axes should be frequency (MHz GHz), x-axes should be UT time. Line 204, resolution should be MHz per pixel or sec per pixel, 120*120 is image size Caption of Figure 4 is not complete. Experimental design: False negative cases are important, please check case by case if there is event which is not marked in the original dataset. Validity of the findings: The SBRS is open public: https://nadc.china-vo.org/data/data/sbrs/ https://sun.bao.ac.cn/SHDA_data/ SBRS has no restrictions to publish data processing results, including the science results and event catalog. The event list is an important part of method validation. Additional comments: The author has addressed some of the comments, I still have some concerns about the manuscript and results

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A DEEP LEARNING METHOD FOR THE RECOGNITION OF SOLAR RADIO BURST SPECTRUM Review round: 2 Reviewer: 2

Basic reporting: I am glad to see the author gave quick responses and made appropriate revisions based on all reviewers' suggestions. The missing abbreviations and concepts are explained, and the wrong formats are fixed as well. The conclusions are re-organized and re-expressed so the tables and figures seem to justify them. There are still some typo and small grammar errors in the document. I think they could be corrected through some automation tools before published. Anyway, the article is suggested to be accepted and published. Experimental design: No comments. Validity of the findings: No comments. Additional comments: No comments.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A DEEP LEARNING METHOD FOR THE RECOGNITION OF SOLAR RADIO BURST SPECTRUM Review round: 3 Reviewer: 1

Basic reporting: The author has addressed most of my previous problems, these are some minor suggestions: Line 30: kHz to sub THz Line 31: background radiation -> quiet sun. background radiation usually refers to ‘The cosmic microwave background’ in astronomy Line 40: channel noise -> RFI (radio frequency interference) Line 92: eliminate -> reduce, instrument noise can never be perfectly removed. Line 203: limit -> rescale or map. Line 259: eliminate -> reduce, (this is used multiple times, please correct all) Line 269: the radio signal changes on the dynamic spectrogram -> the morphology of the radio bursts in dynamic spectrum I would suggest accepting the manuscript after minor revision. Experimental design: no comment Validity of the findings: no comment Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A DEEP LEARNING METHOD FOR THE RECOGNITION OF SOLAR RADIO BURST SPECTRUM Review round: 4 Reviewer: 1

Basic reporting: No further suggestions, the manuscript can be accepted Experimental design: no comment Validity of the findings: no comment Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A THREE-STAGE HEURISTIC TASK SCHEDULING FOR OPTIMIZING THE SERVICE LEVEL AGREEMENT SATISFACTION IN DEVICE-EDGE-CLOUD COOPERATIVE COMPUTING Review round: 1 Reviewer: 1

Basic reporting: No comment Experimental design: No comment Validity of the findings: No comment Additional comments: I would suggest the authors to incorporate the following very minor changes in the manuscript 1. I would suggest you to add a "motivation" sub-section and "contributions" sub-section in the "Introduction" section. 2. In line 73, the authors should add the reference of the paper from where the EDF technique is cited. 3. Similarly, on lines 245, 246, 248, 250, and 256 references to FF, FFD, EDF, BF, LSTF are missing, cite the papers from where these techniques are used in the current paper.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A THREE-STAGE HEURISTIC TASK SCHEDULING FOR OPTIMIZING THE SERVICE LEVEL AGREEMENT SATISFACTION IN DEVICE-EDGE-CLOUD COOPERATIVE COMPUTING Review round: 1 Reviewer: 2

Basic reporting: Authors present three-stage heuristic task scheduling for optimizing the service level agreement satisfaction in device-edge-cloud cooperative computing in this paper. Article is well written, and authors have done great job. However, there are some very serious question and comments which need to be addressed. • Abstract: According to authors, they proposed the technique to solve the problem in reasonable time. What will be considered reasonable time? • THREE-STAGE HEURISTIC TASK SCHEDULING: In the proposed technique, priority order is (i) clouds, (ii) edge servers, and (iii) local devices. Shouldn’t it be other way around? If they task can be executed on the device, it should run on the device. If it can’t, it should be placed to edge servers to reduce latency and overall cost (which will be paid to cloud service provider) and in the end if task cannot be placed on edge servers, due to limited resources, it should be forwarded to cloud. This mechanism will improve overall resource utilization of user device and edge server. Moreover, it will remove latency, cloud placement cost, and migration and rescheduling costs. • THREE-STAGE HEURISTIC TASK SCHEDULING: In 2nd stage, tasks are shifted to device which can be accommodated on the device. Why its not done in first place? • THREE-STAGE HEURISTIC TASK SCHEDULING: According to authors, last stage improves the resource cost. How will it improve resource cost? what about cost of such frequent migrations? why a smart technique is not devised which places the workload intelligently and reduces the cost of rescheduling and migration? Experimental design: • Experimental Design, please provides the details/specifications of tasks, VMs and physical machines in tabular form. • Experimental Design, please provide the proper reference number of all the selected techniques. How are these selected existing techniques closely related to your work? All of them are based on bin packing techniques. Why other state of the art, heuristic-based techniques are not considered for the comparison? Some of these selected techniques are very old and outdated. • Experimental Design, please provide the formula of each selected performance metric. • Results: According to authors, they achieved 59% better SLA satisfaction. How this satisfaction is achieved? And have they considered performance / SLA degradation due to migration? If yes, what is its impact? if no, why didn't you considered such an important consideration? Validity of the findings: • Results: According to authors, they achieved 59% better SLA satisfaction. How this satisfaction is achieved? And have they considered performance / SLA degradation due to migration? If yes, what is its impact? if no, why didn't you considered such an important consideration? • Results: Figures 2, 3, 4 and 5. What does y axis show in these graphs? Does it represent the graph heading or caption? Results are confusing. Values of y axis are in percentage or numbers? Please provide the proper axis titles in each graph along with the units. • Results: SLA Satisfaction, “In contrary, other methods prefer local resources or nearby edge resources, aiming at providing the best performance for each task with minimal resource costs. But these methods result in several local and nearby edge resources that are used by some tasks which can be finished by the cloud, and these resources can be reserved for processing other tasks whose demands cannot be satisfied by the cloud.”, there are techniques that place the tasks on local device or edge, but when the real-time tasks with latency limitation is to be offloaded, they are placed (and given priority) on the local device/edge and non-real time tasks hosted on those machines are moved to next layer of hierarchy. • Results: Resource utilization of existing techniques shouldn’t be high? They are utilizing local and edge resources first compared to cloud. • Results: Cost Efficiency of proposed technique should be much higher compared to existing techniques as it first places the tasks on cloud rather than utilizing local resources. • Results: Number of migration, performance degradation due to migrations and cost of migrations should be considered in performance evaluation process. • Results: Rationale behind the result discussion is very weak. Additional comments: • Related Work: limited related work. should be extended by including 2021 techniques.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A THREE-STAGE HEURISTIC TASK SCHEDULING FOR OPTIMIZING THE SERVICE LEVEL AGREEMENT SATISFACTION IN DEVICE-EDGE-CLOUD COOPERATIVE COMPUTING Review round: 1 Reviewer: 3

Basic reporting: Problem that is tackled in proposed paper is interesting and important, however proposed manuscript suffers from some drawbacks, which are listed below. According to my opinion proposed manuscript should undergo major revision before acceptance. Experimental design: Visualization of results should be improved - consider using box and whiskers diagrams, swarm plots, etc. Also, please include convergence speed graphs, since the convergence is very important indicator of system performance. Validity of the findings: Proposed method should be compared with SOTA (state-of-the-art) metaheuristics methods. Please, examine literature and include few metaheuristics approaches. Moreover, statistical tests should be executed to prove significant results improvements over other methods. Additional comments: Abstract should be completely reformulated to highlight main ideas and contributions of the proposed research. Abstract should emphasize a problem that is being solved, importance of the problem, employed methods and achieved results along with methods used in comparative analysis. Your abstract does not highlight the specifics of your research or findings but contains too much background information. Some details of your research would be nice. An abstract with some details helps show the impact of your research. To aid in this, here is one of many good articles concerning crafting an abstract https://writing.wisc.edu/handbook/assignments/writing-an-abstract-for-your-research-paper. It is not the best practice to use abbreviations in the abstract. Moreover, method names should not be capitalized. Writing paper in the 1st person does not sound scientifically. Please rephrase all sentences into the 3rd person. Introduction should be clearly presented to highlight main ideas and motivation behind the proposed research. Please include and clearly state research question and contributions of proposed study in Introduction. Literature review should be improved to include metaheuristics-based approaches. There are many literature sources where the cloud-edge task scheduling problem was solved by using metaheuristics. You may consider the following reference: https://www.mdpi.com/574168 Related works section should be included after Introduction, alternatively as a subsection within Introduction.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A THREE-STAGE HEURISTIC TASK SCHEDULING FOR OPTIMIZING THE SERVICE LEVEL AGREEMENT SATISFACTION IN DEVICE-EDGE-CLOUD COOPERATIVE COMPUTING Review round: 2 Reviewer: 1

Basic reporting: No comment Experimental design: No comment Validity of the findings: No comment Additional comments: I am satisfied with the manuscript now as the asked changes have been incorporated by the authors. Hence this manuscript may be accepted for publication.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: A THREE-STAGE HEURISTIC TASK SCHEDULING FOR OPTIMIZING THE SERVICE LEVEL AGREEMENT SATISFACTION IN DEVICE-EDGE-CLOUD COOPERATIVE COMPUTING Review round: 2 Reviewer: 2

Basic reporting: Dear Authors, thank you for addressing my comments. However, I still noticed some minor English language and technical issues, therefore please read once again your manuscript thoroughly before publication. All the best Experimental design: no comment Validity of the findings: no comment Additional comments: Dear Authors, thank you for addressing my comments. However, I still noticed some minor English language and technical issues, therefore please read once again your manuscript thoroughly before publication. All the best

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PERFORMANCE IMPACT OF PRECISION REDUCTION IN SPARSE LINEAR SYSTEMS SOLVERS Review round: 1 Reviewer: 1

Basic reporting: - The paper is clear and well-organized - Sections are of appropriate sizes - Intro. & Background are fine - Figures are relevant Experimental design: - The authors compared SpMV in single and double precisions on Intel CPU and NVidia GPU - They also compared LU using PARDISO, cuSPARSE and SuperLU in single vs. double - The fact that the authors are not performing a study for their own solver is also valuable to ensure a fair comparison Validity of the findings: - The authors wrote that subnormal numbers can be an issue to achieve performance, and they illustrate with meaningful results - They explained that such numbers can propagate during the computation and hurt the performance for numerous instructions - They demonstrate that expecting a speedup of 2 just by using float instead of double is not possible in the majority of cases - They pointed that incomplete factorization in float vs double has a lowest speedup compared to SpMV, which could motivate the implementation of a more efficient strategy Additional comments: I think the paper deserves to be published as it provides interesting results that are beneficial for the community. From my side, there are only a few points that should be addressed before having a final version. - One thing that I had faced in the past is the poor implementation of single-precision kernels (in the MKL especially). It might not be true anymore, but as the authors rely on the MKL/cuSPARSE as a "black-box", they are considering that the kernels are as optimized in both cases, which might not be true. I would like the authors to ensure that the differences they are seeing are not coming from a difference in the optimization level. With this aim, I would suggest the following. The authors could convert a small dense matrix that fit in the L1 cache in the CSR format and apply operations (such as SpMV) on it (hundreds or thousands of times) such that it will provide a good view of the raw performance of the kernels when the memory transfers are negligible. The authors could then simply add a sentence in the manuscript to state if there is a difference or not (and update the corresponding sections if needed). This is just one way to do it, and of course, the authors could use any other strategy to test that the MKL or cuSPARSE kernels are optimized similarly. - I would appreciate in "EXPERIMENTAL SETUP" to have an idea of how the numbers were extracted, for example, are they average of X executions? Or the median? - Similarly, I imagine that the first data that will be used by the kernels are not in the L1 cache. I would like the authors to confirm this point in "EXPERIMENTAL SETUP". - I do not understand this sentence "because they are usually processed at the software level," My first assumption is to disagree with this statement. If I program in assembly a code that does a dot product, it will work also with subnormal numbers without the need to add anything "at the software level". However, I think that instructions can have higher latency (need more clocks) when subnormal numbers are used (or NaN as Intel in the past, etc.). I would like the authors to clarify this point. - I would suggest replacing "CSR" with "MXCSR" in the comments of Line 200 (to ensure no ambiguity with CSR storage). I also suggest adding a number and a legend to this code in order to describe what it does (without the need to go in the main text). - I can imagine that the discussion concerning the bytes per values (12b/nnz ~ 8b/nnz) is also valid for other operations than the SpMV. I would consider that many of the symbolic/transformation stages are memory bound and would not benefit from the shift from double to single (because it only reduces the matrix memory by 1/3 and not 1/2). But after a careful read, the authors made it clear by giving explanations and stating that these stages are not related to computation (so I think it will be clear for all readers). - "rpecision." -> "precision."

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PERFORMANCE IMPACT OF PRECISION REDUCTION IN SPARSE LINEAR SYSTEMS SOLVERS Review round: 1 Reviewer: 2

Basic reporting: It would be better to add the definitions of "solve" in Figures 7 and 8, as well as Figures 9 and 10, because they seem to have different meanings. Experimental design: This paper studies the effect of replacing double-precision arithmetic in solving sparse linear system with single-precision. The investigation about the mechanism how subnormal numbers are generated and caused poor speedup is interesting. Also, the finding of the lack of parallelism and floating-point arithmetic in the analysis and reordering phases of the solvers as the reasons of much lower speedup is reasonable. However, it would be better if the authors could add some detailed analysis on the characteristics of matrices. For example, the matrix ss1 is small but shows low ratio of "Reordering & Analysis" ratio in Figure 7. This may caused the better speedup in Figure 1. So, the readers will be interested in why this matrix has shown different behavior. Also, discussions on the results for iterative solvers are limited. It would be better to add the reasons why it has shown less speedup on GPU than CPU. In addition to that, there should be more description about the reason why the speedup of ILU is not significant. Validity of the findings: No comment. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PERFORMANCE IMPACT OF PRECISION REDUCTION IN SPARSE LINEAR SYSTEMS SOLVERS Review round: 2 Reviewer: 1

Basic reporting: The authors carefully answer the questions/remarks, and I think the paper can now be accepted. Again, the paper is clear and well-written. Experimental design: The experimental part is rigorous and well done. Moreover, the code is publicly available. Validity of the findings: The study of the effect of subnormal numbers, through the linear systems solvers, is important in the context of increasing use of mixed precision by the community. Additional comments: None.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PERFORMANCE IMPACT OF PRECISION REDUCTION IN SPARSE LINEAR SYSTEMS SOLVERS Review round: 2 Reviewer: 2

Basic reporting: No comment. Experimental design: No comment. Validity of the findings: No comment. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: COMPARATIVE ANALYSIS OF MAGNETIC INDUCTION BASED COMMUNICATION TECHNIQUES FOR WIRELESS UNDERGROUND SENSOR NETWORKS Review round: 1 Reviewer: 1

Basic reporting: There are few format and spelling problems, and also unprofessional when claim "accurate predictions" without giving data. Examples are as follows: 1. "Table 1. Comparative analysis of EM wave, Direct MI, MI waveuide and 3-D Coil MI waveguide communcation methods" misspelling of "Waveguide" and "communication". 2. Line 29, WUSN, needs to be expended as it’s first time appears. 3. Line 137, “…accurate predictions of …”, define what is the (accurate) range, resolution and sensitivity of the prediction? 4. Line 306, “the required number of underground transceiver nodes may be reduced to a significant level”, at what level? 5. Line 311, “MI waveguide mechanism offers huge reduction of path loss”, what is the loss level? Experimental design: As the statements from the Introduction and from the Survey Methodology are inconsistent, the authors should decide "What is the objective of this paper?" In line 90 of the introduction, the authors described that “This literature review is needed and intended to do the comparative analysis of all the techniques so as to enable the users to find the optimal solution of MI based WUSNs for respective applications”, while in line 107 of the Survey Methodology, “The purpose of this paper is to review the usage of magnetic induction approach in comparison with conventional EM wave approach for WUSNs and subsequently review all available MI techniques used for the same. ” The authors should decide whether to compare the techniques among MI approach or compare the MI with the conventional EM wave approach? Validity of the findings: 1. This paper is a literature review of magnetic induction mechanism for WUSNs, however the authors should point out what is the novelty of this paper? 2. The authors summarized the comparision of EM wave, Direct MI, MI waveguide and 3-D Coil MI waveguide communication methods in Table 1. However, clear numbers or range for parameters should be provided instead of "Higher/ High/ Lower/ Low" as in table 1. Additional comments: This paper is a literature review of magnetic induction mechanism for WUSNs, the authors compared the three techniques in magnetic induction approach, and with the classical electromagnetic wave. However, the authors can further improve the manuscript in several aspects: 1. Your introduction needs to support the objective. At the beginning of the introduction, the authors detail discussed the classical electromagnetic wave approaches (3 paragraphs), and their disadvantages. The authors should decide what is the main objective of this paper? (Comparative analysis of magnetic induction? as in title) or (comparison of MI with classical electromagnetic wave as described in Methodology?) The introduction should work for introducing the main objective. 2. Conclusions are lack of data support. e.g. line 228-243, 301-343 3. What are the main challenges in magnetic induction based communication? 4. Mixed of capital and lower case in sentences. e.g. in title "Comparative Analysis of Magnetic induction based Communication for wireless underground sensor networks", line 19 "......Electromagnetic wave (EM)......", line 23 "......Electromagnetic wave (EM)......" line 188,...... 5. line 166. indentation of f)......

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: COMPARATIVE ANALYSIS OF MAGNETIC INDUCTION BASED COMMUNICATION TECHNIQUES FOR WIRELESS UNDERGROUND SENSOR NETWORKS Review round: 1 Reviewer: 2

Basic reporting: 1. In line 474 and 476, two references are same. 2. Figure 7 should show the 3-D structure more clearly. 3. Literature research is not sufficient. Experimental design: As a literature review paper, most references are too early, more latest should be listed. For example, from line 397 to 417, the reference of advanced 3D-MI technique are published in 2016. Validity of the findings: Lack of quantitative analysis for each technology. For example, more quantitative data should be presented in Table 1. More figures of analysis or comparasion of the results shoud be presented. Additional comments: Authors should strengthen literature review in related fields. There should be more academic content, such as proof of the novelty of the method and quantitative analysis, rather than vague written statements.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: COMPARATIVE ANALYSIS OF MAGNETIC INDUCTION BASED COMMUNICATION TECHNIQUES FOR WIRELESS UNDERGROUND SENSOR NETWORKS Review round: 2 Reviewer: 1

Basic reporting: 1. The literature was not well modified. Only a few new literatures were added without detailed interpretation. 2. Some pictures are not clear enough. It is better to use vector pictures. Experimental design: no comment Validity of the findings: no comment Additional comments: 1. Please add the latest references, and they must be interpreted in detail in the text, not simply cited. 2. Please increase the clarity of the picture. 3. The references on which key technologies are described are still too old.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: COMPARATIVE ANALYSIS OF MAGNETIC INDUCTION BASED COMMUNICATION TECHNIQUES FOR WIRELESS UNDERGROUND SENSOR NETWORKS Review round: 4 Reviewer: 1

Basic reporting: 1. Authors did not follow the previous comments and the references are still very old. This is not a qualified review paper. 2. The overviews of Figure 5 and Figure 7 are better put together. 3. In page 14, authors only added two technologies in the last 2 rows of the table without describing them. Above all, authors should focus on reviewing newer technologies. Experimental design: no comment Validity of the findings: no comment Additional comments: no comment

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: COMPARATIVE ANALYSIS OF MAGNETIC INDUCTION BASED COMMUNICATION TECHNIQUES FOR WIRELESS UNDERGROUND SENSOR NETWORKS Review round: 5 Reviewer: 1

Basic reporting: no comment Experimental design: no comment Validity of the findings: no comment Additional comments: no comment

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: THE ROLE OF SOFTWARE IN SCIENCE: A KNOWLEDGE GRAPH-BASED ANALYSIS OF SOFTWARE MENTIONS IN PUBMED CENTRAL Review round: 1 Reviewer: 1

Basic reporting: This is an excellent paper. The scope and purpose of this paper was well documented. The literature review is thorough and reflective. The selected data was expansive and appropriate for the purpose of the proposed research. Methods and data materials were made available to the general public. Experimental design: Software recognition is a complex task. The reviewer worked on several projects that intended to extract software entities in his past research and understand the challenges involved. Thus, I am very appreciative of the the method proposed and the large-scale data analytics reported in this research. The method is capable of not only identifying software entities but also several associated metadata which is quite valuable. The interpretation of results were insightful as they included discipline-, journal-, and timeline-based analysis. The improved methods were compared with baseline methods and it was clear a better performance was attained. A few minor comments: 1. It may be helpful to define what software is. Based on the results, OS, environments, plugins, and applications were all included as software. It is useful to draw theorization from STS/science of science to operationalize software for purpose of gaining meaningful results. 2. It is unclear what negative sentences meant in Table 2. 3. It is unclear what is n is Tables 7 and 8. Validity of the findings: The figures reported in the results section were informative. Journal- and discipline-level differences can be readily discerned. I wonder if there is any artifact in Table 8 in which late 1990s and early 2000s seemed to have more software mentions with developers than any other periods. Additional comments: The softwareKG is very valuable and it is likely to garner attention by fellow researchers to plan further research in this area.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: THE ROLE OF SOFTWARE IN SCIENCE: A KNOWLEDGE GRAPH-BASED ANALYSIS OF SOFTWARE MENTIONS IN PUBMED CENTRAL Review round: 1 Reviewer: 2

Basic reporting: In this paper, a data extraction process is given to extract data about scientific research which employs software. The paper is well-written and does not require any modification in terms of language use. An up-to-date background and literature review/related work are provided. The paper is well-organized and supported with appropriate figures and tables. Experimental design: The research question is defined and how this research fills the identified gap is clearly explained. Three specific contributions (which implicitly define the RQs) are formulated as (1) A large-scale analysis of software usage, (2) A comprehensive knowledge graph of software citations, and (3) Robust supervisor information extracting models for disambiguating software mentions and related knowledge. Methods to implement the aforementioned questions/contributions are properly explained and the relevant data sets were defined and cited. The data model for the Knowledge Graph was put forward and the information extraction process using the machine learning algorithms was investigated in detail. Performance of the BERT and ML,sw,opt models are compared in terms of Precision, Recall, and FScroe parameters. Validity of the findings: The results for the three contributions were discussed along with their limitations. All underlying data have been provided and appear to be robust and controlled. Conclusions are well-stated and the future work is included. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: THE ROLE OF SOFTWARE IN SCIENCE: A KNOWLEDGE GRAPH-BASED ANALYSIS OF SOFTWARE MENTIONS IN PUBMED CENTRAL Review round: 1 Reviewer: 3

Basic reporting: This work describes an approach for detecting software mentions in literature by combining named entity recognition, relation extraction and entity disambiguation techniques. As a result, the authors create a dataset of thousands of software mentions in papers, represented as a knowledge graph, and using reification to indicate the confidence of the results. The authors then proceed to analyze the role of scientific software in different domains; sharing insights on how software is becoming a first class citizen in most disciplines. Many software creators would be happy to see how thanks to this work they can assess how influential their software is. The paper is well written and easy to follow. I think this work is timely and highly relevant for PeerJ and the Open Science community. The paper is thorough, and does not omit important implementation details such as the hyperparameters used or the pointers to the tools used for the conversion into annotations for the models, which always take significant time to piece together. Therefore, I think the paper would be a nice addition to the journal, and I look forward to reusing some of the outcomes made available by the authors. I have some comments and questions that would be great to address/clarify in the final version of the paper (my "minor revision" is really minor). I add them in the "additional comments" field Experimental design: Robust and thorough experiments. Nice discussion of the results, highly related to the journal. Some questions to be addressed in the "Additional comments" field below Validity of the findings: The findings and analysis are highly interesting, build on prior work and improve the state of the art. I have found a few errors regarding the vocabularies used, please see "Additional comments" below Additional comments: - I found that the skg namespace (http://data.gesis.org/softwarekg2/) returns a 404, which will harm the reusability of the endpoint (I see examples online, but the data model documentation is what is critical for performing queries). I also see that some properties use http://data.gesis.org/softwarekg/software, which also returns a 404. - The authors present baseline results in the same table as the results of the paper. However, in the text, they insist that direct comparison is not appropriate. Then why adding tables comparing both approaches? I think it misleads the reader. - Having the ability to find whether software is created or repurposed in a publication is great. Are there areas where this happens more than others? I think I have not seen this part in the discussion. - I am surprised not to see Zenodo as a common archive for depositing software. Are authors (in general) not depositing their releases there, or it something out of the scope for this paper? I know that it's a common practice for researchers to point to their Zenodo releases through the GitHub release integration. - The relationship extraction between software is deemed as challenging in the paper. However, I have not found why. Is there an explanation for this? - I find curious some of the choices by the authors. For example, programming languages are marked as Software, and packages in those programming languages are plugins. Is this representation easier for the training part? - Also, I wonder if for the clustering/disambiguation, the plugins could play a part in the similarity metric. At least in that case the python/Python separation would be fixed. - I imagine the authors have tweaked the BERT models used in their evaluation. However, I have not found any of the tweaked models in any of the links provided by the authors. Where can the models be found? Minor: - The github repositories provided with the resources are not tagged (do not have releases), and therefore may be difficult to rerun the experiments if one changes. - Typo: Page 6: Objective of this information extraction step -> The objective. Similar thing in line 354. - Missing "." in line 802.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: THE ROLE OF SOFTWARE IN SCIENCE: A KNOWLEDGE GRAPH-BASED ANALYSIS OF SOFTWARE MENTIONS IN PUBMED CENTRAL Review round: 2 Reviewer: 1

Basic reporting: The authors addressed all my concerns. Experimental design: The authors addressed all my concerns. Validity of the findings: The authors addressed all my concerns. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: THE ROLE OF SOFTWARE IN SCIENCE: A KNOWLEDGE GRAPH-BASED ANALYSIS OF SOFTWARE MENTIONS IN PUBMED CENTRAL Review round: 2 Reviewer: 2

Basic reporting: All corrections suggested by the other two reviewers are handled with care. No additional comment! Experimental design: No comment. Validity of the findings: No comment. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: THE ROLE OF SOFTWARE IN SCIENCE: A KNOWLEDGE GRAPH-BASED ANALYSIS OF SOFTWARE MENTIONS IN PUBMED CENTRAL Review round: 2 Reviewer: 3

Basic reporting: The authors have answered all my questions (reflecting them in the paper when appropriate) and addressed all my minor concerns about the paper. I think this work will be a nice contribution to this journal. A few typos: - line 552 has a restricted overleaf link which I think it's not intended there. - Github should be GitHub Experimental design: n/a Validity of the findings: n/a Additional comments: n/a

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: HUMAN-ROBOT INTERACTION: THE IMPACT OF ROBOTIC AESTHETICS ON ANTICIPATED HUMAN TRUST Review round: 1 Reviewer: 1

Basic reporting: The language of the paper is clear. I spotted a few mistakes that could be removed with an additional edit, but they did not detract much fromt he paper: e.g. line 126 "that that". The freferences to the literature were appropriate. The tables and data were all clear. For me, a major failing of the paper was a failure to examine, even superficially, the concept of "trust", a central concept of the paper. "Do you trust this robot?" but to do what, exactly? In th eonly concrete example given int he paper, it is stated that, "[P]articipants were presented with mathematical 375problems which would be too complex for human calculation (i.e. 887x974 & 997x1066). Participants would then be asked to identify which Canbot (A-H) was displaying the correct solution upon their chest screen. This question required participants to determine the answer they deemed correct based solely on trusting the robot’s physical appearance". The logic of this appears to be that, if the calculation is too complex for a person to complete, a decision could only be based on the appearance of the robot. Even allowing that the calcualtion was "too complex for human calucaltion", a human shoul dbe able to decide whther the corect answer was odd or even, and perhaps have a fair guess at whethe rit was larger or smaller than a million, or even calculate the final digit. So the idea that it could only be the appearance of the robot that determined the outcome seems to me to be deeply flawed. Experimental design: I have already mentioned some of the shortcomings of the design, in terms of exactly what "trust" was taken to mean. However, in this section I find the question of sampling and data collection to be of more concern. Respondents to the questionnaire numbered only 74, from all over the world and with a range of different levels of experience of robots. there were obvioulsy too few participants to be able to analyse the differences between groups, but 74 seems much too few to be able to generalise the results to the whole of humanity. So this raises questions about how the sample was selected, what part self-selection palayed, and any resulting biases. I can see no details in the paper as to how the sample was selected, so it is impossible to be clear about how effectively the process was conducted. Validity of the findings: In each of the above sections I have commented on weaknesses that I think have a serious impact on the robustness of the findings. I do not feel confident tha tthe evidence provided supports the conlcusions. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: HUMAN-ROBOT INTERACTION: THE IMPACT OF ROBOTIC AESTHETICS ON ANTICIPATED HUMAN TRUST Review round: 1 Reviewer: 2

Basic reporting: This is an interesting and timely paper. It is written in clear, unambiguous language. Literature references are sufficient to provide background and context for research results. The paper includes sufficient introduction and background to demonstrate how the work fits into the broader field of knowledge. Relevant prior literature should be appropriately referenced. The structure of the article conforms to an acceptable format of ‘standard sections’. The submission is ‘self-contained,’ representing an appropriate ‘unit of publication’, including results relevant to the hypothesis. Experimental design: The authors provide the reader with original primary research inline with the within aims and scope of the journal. Research question is well defined, relevant & meaningful. It is stated how research fills an identified knowledge gap. The submission clearly defines the research question, which are relevant and meaningful. The knowledge gap investigated is identified, and statements are made explaining how the study contributes to filling that gap. Validity of the findings: All underlying data have been provided and they are robust and credible. Conclusions are well stated, linked to original research question & limited to supporting results. Additional comments: This is a timely piece that looks at a novel and interesting research area. Should the authors consider it appropriate, references to complementary disciplines, such as psychology or copyright might be included. Overall this is a good piece of academic writing, ready to be published.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: HUMAN-ROBOT INTERACTION: THE IMPACT OF ROBOTIC AESTHETICS ON ANTICIPATED HUMAN TRUST Review round: 1 Reviewer: 3

Basic reporting: In general, it is a well-written, easy-to read report with clear and sufficient theory and method. I found a few very small typographical errors (e.g. missing blank space),which I marked in the attached pdf file. However, I have two larger suggestion for revisions, both concerning the researcher reflection on what they research and in what context. 1. You seem to take for granted that users always should trust robots, and that when users don't trust robots, something is wrong. However, looking at the problem from a human-centered design perspective (or, a critical design perspective as well), we should acknowledge that sometimes technology should not be trusted. So, just an acknowledgement about this, and a reflection on why you take for granted that trust is good, might be good. 2. You have chosen to frame the discussion as a discussion about aesthetics. However, very much of what you have investigated could also be framed as design. Yet, you mention very little about design. Not necessarily an error in the research, but I think it would be good to briefly explain why you have chosen to talk about aesthetics instead of design, and how you see the relation between those two terms (for example, is aesthetics a sub-set of design, or is design a means to reach aesthetics, or even the other way around. While bringing up this, it could also be good to mention that there are very much work done within design when it comes to emotional effects such as trust (trust for brands, trust for other mechanical objects such as cars etc). There is also existing design patterns and design frameworks, for example the gestalt laws, that have been exploring similar questions for many many decades. So an awareness of this might support your results. Experimental design: In my opinion, a survey is highly limited as a research tool to make statements about trust users would feel towards a robot. The respondents will answer what they THINK they would feel in a real-life situation, or what they THINK you as researchers want them to feel. (Some of the survey questions are quite revealing, it is quite easy for a respondent to second-guess what you as researcher anticipate the respondent to answer.) If users actually interacted with real-life robots in field studies (real-life situations), then it is possible that their reactions would be quite different from what they state in this survey (as you hint at when you mention further research at the end). Of course you can't change that now, you have the data you have, and even if I question how much conclusions we can draw, I still think the result you have is worth reporting. But, I strongly suggest that you yourself bring up a discussion about these limitations of your study. Also, when you describe the results, it might be good to choose wording that emphasize that the results are limited, such as "this suggests", "this indicates", andso on. Validity of the findings: It is difficult to get a clear view of how the survey was designed and how the exact questions where phrased. Some of the Figures seem to show questions, but it is not clear what part of the question is and what is your explanation for me the reader. I have some issues with the wordings of the survey questions in Figure 4 (If this is the actual survey question, which is as noted not clear). The word “new” is used to describe the right hand image of the robot. Did the respondents see the word “new”. In that case it can have created a bias since some people probably interpret new as better. Besides this and my comment on 2. Experimental design, I have no further critique or suggestions for revision concerning validity. Additional comments: I would suggest a change of the title. It is difficult to understand the title before reading the article, and after reading the article, the title doesn't seem to match the actual content of the article (which is probably why it was hard to understand from the start). The article is not about visualization at all, so I strongly suggest taking away that word. Uncertainty is just one of the aspects of trust in robots that are discussed, so I find it strange to include that in the title. I would suggest something like: "Users attitudes concerning trust in robots - towards an understanding of how robot aesthetics impact anticipated trust". On line 88 you write "between man and machines is in how well they understand each other". It is a philosophical/ontological question maybe, but I object strongly towards the notion that machines "understand" anything. There is no proof as far as I know that any robot or AI actually understands anything, and it will take decades until general intelligence have made this possible. Could this be re-formulated some how (I don't know how it was formulated in Barnes and Jentsch (2010) ) ? Something like "....is how well they adapt to each others behavior", or so. That phrasing doesn't assume any actual understanding. I don't understand the heatmaps in Figure 6 and 7. Heat maps are used to visualize values distributed in two dimensions. Which value (data) does the heat map represent? Did respondents click on the image, and the heat map shows frequency of clicks? It doesn't make sense, but I can't come up with any other explanation. Please consider if heatmap is correctly used, and either take away or if it is kept explain what it visualize and how.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: HUMAN-ROBOT INTERACTION: THE IMPACT OF ROBOTIC AESTHETICS ON ANTICIPATED HUMAN TRUST Review round: 2 Reviewer: 1

Basic reporting: I think that the paper could still benefit from another review of the text to remove typographical errors, some of which have been introduced in the course of the modification. For example: Line 111: There is an unnecessary apostrophe after the full stop. Line 132: Guthrie should be cited rather than citied. Line 148: ‘designs’ should have an apostrophe after the ‘n’. Line 326: Is the question not whether the robot is trustworthy, rather than trusting? Line 326: ‘robots’ needs an apostrophe after the ‘t’. Experimental design: The authors have responded to the comments of the reviewers in regard to failures in the descrioption of the experiemntal design, rather than falinings in the desing itself, and this is now much better. Validity of the findings: In responding to the comments of the reviewers, the authors have now added some clarifications and caveats that strengthen the paper in this regard. Additional comments: I have to say that I am impressed by the way the authors have responded to the points of criticism raised by all the reviewers, and I would now say that the paper could be published. I still think there is something lacking in the analysis of trust, and of whether it is the robot that deserves or does not deserve trust. I trust a calculator to produce an answer that accurately reflects the buttons that have been pressed according to its own algorithm. I might doubt my competence in button pressing (fat fingers) or the ease of intuiting the order in which buttons have to be pressed, and therefore have doubts about the answer, but my trust in the calculator would not be diminished. In this study, participants were confronted by robots that had been deliberately programmed to deceive; was it the robots or the programmers who were being assessed as trustworthy? But perhaps this is work for future study.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: HUMAN-ROBOT INTERACTION: THE IMPACT OF ROBOTIC AESTHETICS ON ANTICIPATED HUMAN TRUST Review round: 2 Reviewer: 2

Basic reporting: The amendments introduced in the paper fully reflect my original comments, thank you for incorporating them. Experimental design: The paper, as amended, meets the PeerJ standards. No specific comment to be added. Validity of the findings: No further comment. Additional comments: Thank you for amending the paper, hope the comments have been useful.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: HUMAN-ROBOT INTERACTION: THE IMPACT OF ROBOTIC AESTHETICS ON ANTICIPATED HUMAN TRUST Review round: 2 Reviewer: 3

Basic reporting: The revisions that have been made is a bit shallow and a bit "quick fixes", but they are extensive enough and good enough to make the publication acceptable in my opinion. The authors seems to have understood the feedback, and made appropriate changes. Experimental design: no comment Validity of the findings: The limitations and arguments around the findings is more clear now. Additional comments: No further comments, I will suggest an accept.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: HUMAN-ROBOT INTERACTION: THE IMPACT OF ROBOTIC AESTHETICS ON ANTICIPATED HUMAN TRUST Review round: 2 Reviewer: 4

Basic reporting: Summary: This paper titled “Human-robot interaction: The impact of robotic aesthetics on anticipated human trust” examined people’s preference on some robot faces in terms of facial expression mood, blurriness, and color, as well as conflicting information on the chest display. The authors’ writing is great and clear. However, from a scientific research perspective, the rationale and clarity of the concepts, the experiment design, and the findings all have flaws that make the paper limited contribution to the trust-in-robot literature. Not any of the findings particularly adds new knowledge. Sorry, but I would not recommend publishing this paper. Background: 1. In the authors’ own words in the introduction, “As Barnes and Jentsch (2010) identified, the key to a successful relationship between man and machines is in how well they can work and adapt to each other,” esthetics of a facial expression is not part of how well they can work and adapt to each other. Trust is a process-based decision-making phenomenon. If the results that the robot gave are wrong, no matter how pretty the face looks, the robot is still untrustworthy. 2. The authors stated, “As Gurthrie citied in Daminao and Dumouchel (2018) points out, the tendency to see human faces in ambiguous shapes provides an important advantage to humans, helping them to distinguish between friend or enemies and establish an alliance.” This is not true. The same person can be a friend at one time and be an enemy at another. It is not the look of the face that changed, but the identity and mechanism changed. 3. The contents of trust have many dimensions, such as trusting the purpose/intention, trusting the process, and trusting the performance. Trust also has different phases, such as initial trust, trained/informed trust, and situational trust. The study is not clean on the direction: Are people willing to trust the robot (Canbot) to do what? If this question is not clear, how does esthetics matter? Experimental design: Experiment design: 4. The authors stated that “It is the ‘effective’ interaction which is of interest to the authors of this paper (i.e., the ability to build a trusting relationship through effective human-robot interaction).” By definition, interactions mean “mutual or reciprocal action or influence” (Marriam Webster Dictionary). However, the facial expression is just showing the options, and there is no interaction involved. The robots’ answer correctness is not the independent variable for people’s trust in the robot, but the robot’s appearance is the independent variable. The experiment design is not based on interactions. 5. The questions about trust should be more specific. The original question was, “Would you trust Canbot by the way they look?” However, what does it really mean? Trust Canbo on what? Trust Canbo is a robot? Trust Canbo can do all math problems correctly? Trust Canbo provides correct information on all questions? Trust Canbo can take care of your children? Trust Canbo has human conscience? Trust Canbo have the good intention for you? Validity of the findings: Results and findings: 6. The authors only reported the percentage of participants’ response category, without any statistical significance report at all. More advanced statistics (e.g., non-parametric statistics such as Chi-square in this case) are strongly encouraged. 7. The results that people preferred the good proportioned positive faces (or the visualization with no modifications) better than a weird/angry/blurred/mentally retarded looking is not surprising at all. Is it really trusting in a robot or the one chosen is just least displeasing? 8. The authors stated, “In situations such as trusting robots where a person [should this be a robot?]’s past behaviours and reputations are unknown, we acquire other sources of information to determine a person’s motivations (DeSteno et al., 2012).”, based on this logic, this study is only talking about *the first initial impression of the robots’ trustworthiness*. Once the robot presents one result to a question, that information itself is showing whether the robot is trustworthy. The so-called trust shall not go past the initial impression. The paper should not overgeneralize the findings on the impact of esthetics on trust in a robot. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: NETWORK INTRUSION DETECTION USING OVERSAMPLING TECHNIQUE AND MACHINE LEARNING ALGORITHMS Review round: 1 Reviewer: 1

Basic reporting: 1.The author claims in lines 70-71 that the proposed model performs better than previous related research, but there is no specific explanation and comparison in the paper, which is hard to convince. I suggest that the experiment is not just to compare the performance differences of existing Methodologies, but to compare the superiority of the detection framework over the previous research when facing the same dataset. 2.Lines 72-73 indicate that the data preprocessing technology mentioned in this paper is novel and has not been employed. However, there are abundant researches on data standardization and normalization technology and SMOTE technology for dealing with imbalances. So this statement is debatable. Readers are more concerned about the innovation of this work. Therefore, the author should not only declare the novelty of the work but also discuss and prove it in detail. This will help readers grasp the value of the work. Experimental design: 3.There are some errors in the presentation of data, tables and conclusions: 1）Since the IG method is used, the description from 89.5% to 95.1% in RF in line 358 should be modified from 89.5% to 95% in RF. 2）The decrease in ANN accuracy is incorrectly stated in lines 370 to 371 because 76.8% is the accuracy of KNN. The ANN accuracy decreases from 80.3% to 71.7%. 3）In lines 377 to 378 the conclusion is wrong, the accuracy of KNN increases from 84.0% to 84.7% after using PCA, which does not reflect the decrease in accuracy with PCA, and we note that the accuracy of ANN drops exactly 85.2% to 77.6% again after using PCA. In line 385 the conclusion is wrong, class balancing does not affect Logistic Regression and Artificial Neural Networks, not Decision Tree and Artificial Neural Networks as described in the paper. Validity of the findings: no comment Additional comments: no comment

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: NETWORK INTRUSION DETECTION USING OVERSAMPLING TECHNIQUE AND MACHINE LEARNING ALGORITHMS Review round: 1 Reviewer: 2

Basic reporting: • In general, the abstract should be about 240/260 words in length and up to the 500 words limit. It can contain short statements summarising the project scope, research significance/motivation/problem statement, aim, objectives, the method and techniques used to work towards these objectives, and the results achieved and conclusions made. It should give a reader sufficient information to decide whether or not to read the rest of your paper. Make sure to revise your abstract to address most of the above requirements, especially your motivation and aims. • The abstract keywords are not presented. • The uncommon abbreviations should be spelt out at first use only. EM algorithm in line 107 is not abbreviated, and SMOTE is abbreviated twice in lines 87 & 199. Therefore, the English language should be improved, I suggest you have a colleague who is fluent in English and familiar with the subject matter review your manuscript. • Make sure all the equations are in the middle of the line. • In line 318 the table name error. • In line 155 it’s F1-score, and not f1-score. • Line 101 – 103, the reference style in table 1 is not as the main text, also, there is no reference to what you cited in this table. Also, you should add more research from 2020 and 2021 in this table. Experimental design: • In lines 66-67, you need to clarify why you used only the UNSW-NB15 dataset and update this section as in 2021 more than 24 open access papers mention the UNSW-NB15 dataset. https://paperswithcode.com/dataset/unsw-nb15 • In line 71 you mentioned the previous research, and you claim that your result achieves better accuracy than the previous research. You should cite them here. • In lines 79-79 you should provide detailed information about your experimental setup. • A methodology diagram in the methodology section will benefit this section. • In line 316, the label has been extracted as a feature, can you clarify why? • In lines 184-188 More details and discussion are required in this section. • The percentage in table 3 is not accurate if you consider all datasets and not just the training datasets. You can refer to Table VIII in the reference given below for more details. N. Moustafa and J. Slay, "UNSW-NB15: comprehensive data set for network intrusion detection systems (UNSW-NB15 network data set)," 2015 Military Communications and Information Systems Conference (MilCIS), 2015, pp. 1-6, doi: 10.1109/MilCIS.2015.7348942. Validity of the findings: • In this paper, the authors used the SMOTE technique and feature selection techniques to address the class imbalance issue for some attack types that have very few instances on the UNSW-NB15 Dataset. They managed to achieve a better result when applying both techniques. However, the authors haven’t validated their results with the current research results. A Validation section should be included in this paper. Additional comments: • Only one dataset has been tested with the proposal framework, I recommend adding more than one dataset to evaluate their framework. • I noticed that the raw data and the code are not submitted • However, I do think the authors could further improve the papers by addressing the above comments, and adding a discussion section at the end of this paper will make it even more readable and comprehensive.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: NETWORK INTRUSION DETECTION USING OVERSAMPLING TECHNIQUE AND MACHINE LEARNING ALGORITHMS Review round: 2 Reviewer: 1

Basic reporting: The author has answered my concerns and I think it can be published. Experimental design: The author has answered my concerns and I think it can be published. Validity of the findings: The author has answered my concerns and I think it can be published. Additional comments: The author has answered my concerns and I think it can be published.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: NETWORK INTRUSION DETECTION USING OVERSAMPLING TECHNIQUE AND MACHINE LEARNING ALGORITHMS Review round: 2 Reviewer: 2

Basic reporting: I recommend accepting this paper; it is much improved from the previous version. It's now relatively easy to read from beginning to end, and the technical content of the paper has been settled, so I have no additional technical comments on this one Experimental design: . Validity of the findings: . Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: MANIFOLD-ADAPTIVE DIMENSION ESTIMATION REVISITED Review round: 1 Reviewer: 1

Basic reporting: The language is generally clear, but it can be improved in certain sections. The literature is well referenced. The structure is not conforming to PeerJ Computer Science standard (namely, the Methods section should be placed right after the Introduction, not after the Discussion). The figures are high quality, but adding more details in the corresponding captions would facilitate their comprehension. Experimental design: The article tackles a critical problem in the computer science field - dimensionality estimation. The research question, however, should be better stated in the introduction. Specifically, authors should clearly state the rationale and the aim of their study. The methods should be moved after the introduction and before the results so as to facilitate the reader following the flow of the article. Validity of the findings: The proposed approach solves certain limitations of the algorithm that is based on. The authors performed an extensive comparison of their approach with other state-of-the art algorithms on various synthetic datasets. Perhaps, an additional comparison of the proposed approach with such algorithms on neural data might be ideal. Conclusions are missing. Additional comments: The paper sets to revisit a dimensionality estimation algorithm, the manifold adaptive Farahmand-Szepesva’ri-Audibert (or FSA). The authors first computed the local probability density function following the original FSA pipeline and then e the median of such pdf to obtain the global estimate of the dimensionality. They further corrected for finite sample effect implementing a correction formula and finally compared the performances of their algorithm with those of the original FSA as well as other state-of-the-art techniques. The proposed approach outperforms the original FSA and perform similarly to other methods when applied to synthetic datasets. When applied to neural signals recorded during epileptic seizures, the authors hypothesize that low-dimensional brain regions might be potential sources for the seizure onset. The overall structure of the paper is ok to follow. However, certain parts of the manuscript can be improved and some details need to be added. Following are some specific comments. 1. Line 96. I suggest you provide more justification for your study. What is the rationale of your approach? How do you expect your result to differ from the original FSA algorithm? Also, could you be clearer when you say 'we correct the underestimation effect by an exponential formula'? 2. Line 98, end of introduction. Could you please add some description of your following section in a clearer way? 3. Authors should also compare their approach to the original FSA (or to some other methods) on the neural dataset and not the synthetic ones only. 4. The method section should be moved right after the introduction. This would allow to describe the proposed approach before showing the corresponding results. Moreover, some results are already described in this section (e.g., lines 252, 256, 274). Those sentences should be removed and included in the results section only. Also, please put the figures closer to the corresponding location in the main text where they are referred to. 5. Authors should add a conclusion section. 6. Line 167, cmFSA acronym should be defined before use. 7. Line 244, authors should justify why they chose to test those three specific values of k. 8. There are some inconsistencies related to the use of the notation for the true and the predicted dimensionality. According to line 255, D indicates the true dimensionality and d the predicted dimensionality. However, in line 270 you use d to indicate the true dimensionality and d ̂ for the predicted one. Choose one notation and stick with it. 9. The captions of each figure should be more detailed. Specifically, they should briefly describe the take-home message of the figure (one or two sentences are enough). 10. Figure 1: it is not clear to the reviewer how the histogram was obtained. Didn’t you test only one realization in this case (line 242)? 11. Table I: what are cmFSA_fr and M_DANCO_fr?

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: MANIFOLD-ADAPTIVE DIMENSION ESTIMATION REVISITED Review round: 1 Reviewer: 2

Basic reporting: The manuscript is sound and well written. References are generally exhaus- tive. The methodology is clearly explained. Experimental design: The propose method seems to be competitive with state-of-art methods, and I believe it offers some advantages with respect to some of issued of ID estimators, in particular boundary effects, and variations of the density of points in the data. I think the manuscript is a fair contribution to the field of ID estimation, and it can be of interest to researchers in this area, and more in general to researchers needing accurate ID estimation as part of their data analysis pipelines. Validity of the findings: Major comments: 1) In my opinion, the main problem with the boundary-effect correction is that it is o ptimized for uniformly-sampled hypercubes, and may lead to overestimation of the ID in cases when the data are not uniforly sampled. This is clearly visible form table I: while the estimation is nearly perfect for uniformly sampled data on linear subspaces [M2,M9,M10a-c], or gener- ally uniformly sampled data on locally at spaces [M5,M7,M13], it yields an overestimation in the case of non-uniformities, such as he Gaussian case [M12], the non-linear manifold case [M6], or the sphere [M1]. The overstimation may be even more sever for non-uniform samplings with heavy-tailed distributions, such as the Cauchy distribution used in Facco et al. 2015. The authors should extensively comment on this point. 2) Since this is a methodological work, I would recommend that the authors make publicly available the code implementing cmFSA. 3) It is not clear how the different sample sizes were included in the calibration of the correction term. It seems that the calibration term used to infer the ID of the datasets M1-M13 was inferred from the n = 2500 hypercubes. Is one going to use the same term with datasets of different n ? It seems that one should rather use a term calibrated on that specific n . The authors should comment on this point. Furthermore, why was k = 5 used for calibration, instead of k = 1 used in subsequent analyses? Minor comments in attached PDF. Additional comments: Comments in attached PDF.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: MANIFOLD-ADAPTIVE DIMENSION ESTIMATION REVISITED Review round: 2 Reviewer: 1

Basic reporting: In their revised manuscript, the Authors have satisfactorily addressed my main criticism and suggestions. The revised manuscript is clear and well-written, and its overall structure has improved. Many minor confusing points have been clarified. Experimental design: The methodology proposed is scientifically sound. Validity of the findings: The Authors have been very careful in delimiting the scope of validity of their method, its advantages and shortcomings. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: MICROSERVICE SECURITY: A SYSTEMATIC LITERATURE REVIEW Review round: 1 Reviewer: 1

Basic reporting: Microservices is an emerging paradigm for developing distributed systems. With their widespread adoption, more and more work investigated the relation between microservices and security. In this work, the authors conduct a systematic review of the field, gathering 290 relevant publications. The paper is well written and organized. However, it needs significant updations before it can b e accepted Experimental design: Good Validity of the findings: Satisfactory. Additional comments: 1. In the introduction, the authors can draw a table showing the difference between the current survey and existing surveys. 2. Some of the recent papers related to security such as the following can be discussed in the paper: "Iwendi, C., Jalil, Z., Javed, A. R., Reddy, T., Kaluri, R., Srivastava, G., & Jo, O. (2020). Keysplitwatermark: Zero watermarking algorithm for software protection against cyber-attacks. IEEE Access, 8, 72650-72660., Bhardwaj, A., Shah, S. B. H., Shankar, A., Alazab, M., Kumar, M., & Gadekallu, T. R. (2020). Penetration testing framework for smart contract Blockchain. Peer-to-Peer Networking and Applications, 1-16." 3. List out the main contributions of the survey. 4. The authors can add a section, lessons learnt and future directions to pave the way for researchers interested to work on this topic.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: MICROSERVICE SECURITY: A SYSTEMATIC LITERATURE REVIEW Review round: 1 Reviewer: 2

Basic reporting: Well written, well structured. One important issue with the use of references. (see later) Experimental design: According to the standards of a literature survey. However, I would like to see an explicit motivation on the exclusion of the grey literature. Validity of the findings: The authors conclude with a convincing call for action to help the community make steps forward. Additional comments: Summary. This paper reports on a systematic literature survey on microservice security. They analysed 290 relevant publications (published since 2014) in a quantitative and qualitative way. They complement this analysis by a thorough knowledge about the field and as such identify open areas where future research is welcome. They conclude with a convincing call for action to help the community make steps forward. Points in favour * The authors conducted the literature survey according to the guidelines for doing so. The data is publicly available as I would expect from a publication appearing in PeerJ On top of that they conducted a meta analysis with clusters of authors and a word-net of abstracts which was a cherry on the pie as far as quantitative analysis goes. * I was suspicious with the qualitative analysis driven by 20 (twenty!) research questions. However, as later explained what is actually happening is that these 20 questions serve as markers to classify whether these papers cover a certain topic or not. This allows for the correlation matrix (Table 2) which was again a nice complement. * The field is quite diverse which is a good sign for a literature survey. The publication outlets and the research communities illustrate quite well that a systematic literature survey like this one is due. * Throughout the paper I feel a sense of credibility: the authors know what they are talking about and are able to link and integrate findings from a very diverse field. The discussion and conclusion section formulated a s a series of open challenges is therefore very convincing. I especially appreciated the "React & recover Techniques" and the "DevSecOps" Suggestions for improvement (All below are considered minor. I trust that the authors will do their best to accommodate these suggestions and if they do not I can live with the paper) * They way to authors list papers in the bibliography is very annoying to look up references. In the end I resorted to a search on the digital PDF document while reading the article in paper form. Citations start with the family name of the first author followed by the year (e.g. Casale et al.[2016]) However in the bibliography the first name comes first, family name comes second followed by all other author names; title, misc and the year at the very end of the entry. (i.e. Giuliano Casale, Cristina Chesta, ... Cloud Forward, pages 34–42, 2016). The alphabetical sorting of the references is thus hard to infer and scanning the list is counterintuitive. Note that some authors have multiple first names (e.f. Mohammad Bany Taha) which makes it even more difficult to discern the alphabetical sorting of the references. The authors also split the reference section in two subsection (References + Publications from the Dataset). So scanning from the end of the bibliography was awkward. (Looking for the entry "Wuyts 538 et al. [2018]" I started from the back and did not find it on the first attempt. Only the PDF search revealed what was happening) * I would like to authors to explicitly motivate why they did not include grey literature in their survey. Especially in a field like microservices which is driven by practitioners this may provide complementary perspective. Perhaps there would be blog posts on "React & recover Techniques" and the "DevSecOps" which makes the call for action even more worthwhile. * "Methodology." Don't use the term "methodology" for what is a method, it is inflation of words. The postfix -OLOGY stands for "study of". (i.e. biology = the study of the living organisms; psychology = is the study of the human mind; geology = is the study of the earth). Thus methodology is "the study of the methods". * lines 101-102 "We analysed these publications to collect statistical and objective answers". There is no such thing as an "objective" answer. The very fact that you classified papers already implies a subjective interpretation. I would suggest to replace "objective" by this by "transparant". * Line 360-361 "This is reflected by the sharp increase in the number of publications since 2014." I am not convinced by this argument. The overall number of publications in computers science is steadily increasing, so the absolute number of papers does not say much.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: MICROSERVICE SECURITY: A SYSTEMATIC LITERATURE REVIEW Review round: 1 Reviewer: 3

Basic reporting: The authors present a systematic literature review on microservices’ security, in a well-written manuscript. In addition, the topic is certainly timely and relevant, and one such review can provide various benefits to both researchers and practitioners working on the topic. At the same time, the submitted manuscript must be revised to address various limitations it is currently suffering from, before it can be considered for publication in an international journal. I am listing such limitations in the following sections, together with suggestions hopefully helping the authors in improving their manuscript. Experimental design: As for the search for studies, some concerns should be clarified by the authors. Firstly, why did they choose to restrict their focus to “white literature” only? It is known that industry is heavily investing on microservices, with quite many solutions (also for security) being proposed and posted in industry-driven outlets (whitepapers, blog posts, YouTube channels, etc.). I am not saying that the authors should include “grey literature” as well, but they should at least clarify why they decided to keep it out from their study. (The authors actually notice this in their conclusions, hence raising the point of why they did not consider grey literature from the beginning) In addition, I am not sure on the “repeatability” of one of the exclusion criteria, viz., “We also excluded cases in which the work tangentially mentioned the satisfaction of some security aspects, without detailing the design/development of the security technologies to accomplish them”. Whilst all other exclusion criteria are objective, this criteria seem to potentially be threaten by observer bias (as an observer may a manuscript to satisfy this conditions and hence get excluded, while another may consider the treating of security “not so tangent”, so to say). The authors should clarify how they limited/avoided possible observer biases when evaluating this criteria, e.g., in a “Threats to Validity” section. Finally, the authors should conclude this section by listing all selected studies (e.g., in a table). This would help keeping the authors review self-contained, hence helping readers in better understanding their results. As for the research questions, the authors state that they adopted 20 dichotomous questions with the goal of favoring precision and objectiveness. It is however not clear whether/how they ensured such precision and objectiveness. The marking of a publication as “yes” or “no” for a research question seems indeed to be subject to observer biases and errors. How did the authors avoid/limit this? This should be discussed in a “Threats to Validity” section. Validity of the findings: The presentation of results can and must be improved, by also expanding their discussion. For instance, the authors should consider presenting the publication outlets (viz., conferences and journals) as plots, so that readers can visually observe them (rather than finding a flat list of names). Much better it would be to cluster venues according to some criteria and to show aggregated results. For instance, the authors could expand their discussion on the communities where microservices’ security is most discussed (with some graphical support). Also, I am not sure on whether what the authors call “qualitative results” actually provide “qualitative” information. For instance, the distribution of types of publication is still a “quantitative” information, with which the authors partition the different contributions in the field (here the authors use the word “survey” to denote “reviews”). This type of information is typically present in systematic literature reviews, and it is usually classified as a “quantitative result”. The same holds for the other aspects discussed in Sects. 5.2.2-5. The authors state “how many” publications were marked as “yes” for the research questions pertaining to each aspect, hence providing “quantitative information” on the distribution for such aspects. To make the things more “qualitative” the authors should better enter into the details of "what" is discussed in each publication and "how". Whilst the “what” can be easily shown with tables showing the authors’ classification of considered works (e.g., with rows associated with works, columns with research questions, and checkmarks placed in cells to denote that a research question is treated in a research work), the “how” requires the authors to expand their discussion by discussing how (clusters of) works discuss/tackle the research questions. As a minor comment, in Section 5.2.6 the authors state that they “were surprised to find many citations to blockchain technologies (as reported above) as well as the lack of more and more mainstream technologies like service mesh and serverless”. I am not sure on whether “service mesh” are more mainstream than “blockchains”. The authors should consider rewording this, or at least cite a reference showing the higher recognition/usage of “service mesh” if compared with “blockchains”. In their concluding remarks (Sect. 6), the authors draw some concluding remarks on the open challenges that emerged from their study. A reader however misses the links between the results presented in Sect. 5 and the open challenges/research directions listed in Sect. 6. The authors should try to make such links more explicit. For instance, they could introduce open challenges (e.g., in a “highlighting box”) in Sect. 5, immediately after the discussion highlighting the need/openness of such challenge. They could then retake/recap the open challenges (as they currently do) in Sect. 6. Last, but not least, systematic literature reviews, and systematic studies in general, are known to be prone to possible threats to their validity (like those I tried to highlight in my former comments above). The authors should discuss how they mitigated/avoided possible threats to the validity of their study in a devoted section. Additional comments: Other comments follow: [ “survey” vs. “review” ] I am not sure on whether the word “survey” used by the authors in the manuscript’s title and throughout the text is correct. Snyder’s guidelines, used by the authors to design their research, speak about “systematic literature reviews”. I would hence recommend the authors to revise their wording in “systematic literature review”, both in the title and all over the manuscript. Such wording is that most commonly associated with the type of studies like that presented by the authors in this manuscript, hence helping potential readers to better grab the type of content they would find in this manuscript. [ Related Work ] The related work discussion should be expanded to include other relevant related reviews on “microservices & security”, e.g., that on “microservices security smells” by Ponce et al. (https://arxiv.org/abs/2104.13303) . [ References ] The authors should cite all papers they considered in their literature review, to give credit to the authors who published such papers. [ Appendix ]  The authors present selected studies and their classification in a supplemental appendix. As noticed in my above comments, such information is not supplemental, but crucial for readers to understand which papers were considered and how they were classified to answer to the authors’ research questions. The information in the appendix should hence be included in the main text of the manuscript. *** other comments *** - “Alas” -> “Unfortunately” or “At the same time”? - p5/28: “within the” -> “up to the”

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: MICROSERVICE SECURITY: A SYSTEMATIC LITERATURE REVIEW Review round: 2 Reviewer: 1

Basic reporting: No remarks here, 2nd review Experimental design: No remarks here, 2nd review Validity of the findings: No remarks here, 2nd review Additional comments: The authors took into account all suggestions from the reviewers and complied as best as they could

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: MICROSERVICE SECURITY: A SYSTEMATIC LITERATURE REVIEW Review round: 2 Reviewer: 2

Basic reporting: The authors have thoroughly addressed all my former comments, significantly improving their paper. I would hence recommend accepting the paper in its current form. Experimental design: no comment Validity of the findings: no comment Additional comments: no comment

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: MODEL DESIGN AND PARAMETER OPTIMIZATION OF CNN FOR SIDE-CHANNEL CRYPTANALYSIS Review round: 1 Reviewer: 1

Basic reporting: The overall passage is well structured with adequate deep learning background information, which I think provides basic intuitions for researchers who want to study deep learning based side channel attack. It should be better if more background information about side channel attack is included, e.g. the meaning of the rank, how it is calculated, the relationship between rank and classification results, and so on. The language is good, easy to understand. Although there are some typos, which I point below. No problem to use the public database ASCAD, I am not sure it is necessary to publish the author’s training and testing codes, but they share the model structure, which I think is enough. As for the figures, please rename the figures to make the indexes consistent with the figure titles. Now they are confusing. Typos: In line 53, secret.key. ->secret key. In line 101, deep network The problem-> deep network. The problem In line 108, Parameter -> parameter In line 122, please rephrase your sentence In lines 282-283, x_j is not the inputs of all neurons but one neuron In line 395, please rephrase your sentence Please check the name ‘Conv’, there are many ‘Cnov’ in the passage Experimental design: I like the way the authors design the experiment, the whole experiments are designed in sequential way, and the CNNSCAnew is updated after every experiment. They analyze the model parameters first and then switch to the global parameters, and finally determine the optimal model structure and parameters. However, I think the authors miss one important thing. They should explain why the whole experiments flow is designed like this. For example, why they choose to determine the model parameters first instead of the global training parameters? and in section 2.2, why they choose to determine the number of Conv layers first instead of kernel size or the number of filters first? In addition, I know there are lots of parameters to be optimized and it is tough to go through all the combinations. So I am thinking that whether the impacts of different parameters on the model performance are independent, if yes, there is no problem to optimize the model parameter by parameter, if no, then the whole experiment should be carefully addressed in order to find the optimal model. I will appreciate it if the authors can give more details about this. Finally, please explain why choose the epoch number 75 for most of the experiments ? Have you tried other epoch numbers? And I think the authors should also add comparison results between models with SEnet and without SEnet if they want to show SEnet can reduce the gradient dispersion. Validity of the findings: Based on the conclusions, the authors provide us with a new CNNSCA structure that has better classification results and lower computational time than previous model structures (Benadj, Alex VGG). This part is strongly related to the experiment design part, other than the updated model structure and short training time, I expect more conclusions here. Additional comments: Please gives more background information about why it is important to use deep learning in side channel attacks and comparison between deep learning performance and traditional side channel attack analysis methods.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: MODEL DESIGN AND PARAMETER OPTIMIZATION OF CNN FOR SIDE-CHANNEL CRYPTANALYSIS Review round: 1 Reviewer: 2

Basic reporting: - This paper proposes a convolutional neural network-based side-channel attack model design and hyperparameter optimization to improve side-channel attack performance. The paper is mostly clear but there are numerous grammatical errors. For example, the sentences in lines 90, 124, 319-323, 339, and 639 need revision. - The paper contains sufficient references. I can only recommend the authors add a reference for the MNIST dataset in line 306 of the paper. - Table 1 is not clearly explained in the text. The authors can mention one or two sentences about remarks in table 1. - Figure 1 is not explained in detail. - For figure 2, the authors can use labels to show which data is input, kernel, and output. - In line 135, "BP algorithm" is mentioned but readers may not know the BP. What does BP stand for? Experimental design: - The performance of convolutional-based side-channel attacks has dramatically improved by researchers over the last 5 years. However, there is still room to develop it. - In this paper, there are valuable experimental results. The authors compare the performance of the proposed model with other models in the literature. The proposed method is superior to the other methods. Validity of the findings: - Although the authors did not propose a completely new method/model, they adapted existed methods well to generate a new model. This combined model is not proposed previously in the literature. The experimental results show that it is better than the other methods. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: MODEL DESIGN AND PARAMETER OPTIMIZATION OF CNN FOR SIDE-CHANNEL CRYPTANALYSIS Review round: 1 Reviewer: 3

Basic reporting: This paper aims to improve the state of art of deep learning based side channel-attacks by proposing a new model, named CNNSCAnew. The architecture of this model relies on a new kind of block of layers, the SE module, whose effectiveness has already been established in the domain of computer vision. A fine tunning of the hyperparameters is performed by studying the impact of each parameters independently in several experiences. The performance of the final model is compared with the state of the art models on the open dataset ASCAD. From the editorial side, this paper is not well written. There is a lot of approximations, confusions and contradictions, mostly due to the fact that some english technical terms are not correctly used. Some sentences are not grammatically correct. The overall meaning of the text can be understood (or somehow guessed), but it is very tedious and some typos or formulations can be easily corrected with a more careful review of the paper. Here are some examples found in the abstract and introduction (the list is not exhaustive) l12. decryption -> cryptanalysis l13. decryption -> cryptanalysis l13. Alex -> AlexNet l14. there is Model training -> the trained model l23. the SEnet module : the "SEnet" abbreviation is not yet defined l24. unnecessary parameters are maximized: do you mean "optimized" instead of maximized? l28. Optimize ... CNSSCA : the meaning of this sentence is not clear, do you mean : "We optimized the various ... CNNSCA"? l26. first-order mask data set of -> first-order masked dataset from l34. guess entropy -> guessing entropy l35. successful attack key -> successful recovery of the key l36. the performance was better -> we obtained better performance l42. and using cryptographic algorithms to leak ... harware encryption: the formulation is not correct, I would suggest something like : "by using the leakage of cryptographic algorithms during the computation of data (...) on hardware devices" l43. radiation Etc -> radiation, etc. l44. to crack : the word is a little bit slang l50: the weak -> weak l53: the secret. key: typo l54. the best decryption effect -> the best cryptanalysis attack l58. energy traces -> power consumption traces l69. At present, at home and abroad : ? l79. the best perfomance of breaking secrets -> the best cryptanalysis performance l123. declassification -> classification Experimental design: The study of new attention mechanisms for side-channel attacks is in line with the current trends in deep learning research and the efforts to use open dataset and reproductible results are appreciated. The results are interesting and the overall methodology for fine tunning the deeplearning model sounds correct. All the model parameter choices are published and the code is available, which is very valuable to the side channel community. The experiments rely on an open dataset and they are reproductible, which strenghten the confidence in the results. The hyperparameters are carefully selected and their impact are thorougly studied. The figures depicting the results are clear. The final model, named CNNSCAnew, contributes to improve the state of the art on SCA attacks based on deeplearning. Validity of the findings: Nevertheless, due to the poor level of english and typos, it is difficult to undestand the description of the experiments and sometimes the results are not clear. By example, in the section "Structural parameter optimization", experiment 1, the authors have tested three deeplearning architectures with different number of convolutional layers for each block (namely 1, 2 and 2/3). At line 416, they claim that "if [the number of layers] exceeds 2 or more, ..., the 8G[bytes] GPU memory ... will be exhausted, unable to run code". But they computed the guessing entropy for some of their architecture where the number of layer was 2 (cnov2 and cnov3), which is in contradiction with the fact that they were unable to run the code. Moreover some claims are vague or not correct. By example l62. "but it also loses effectiveness when attacking encryption with protection." : this statement is not true : in the ASCAD paper, a MLP is successfuly applied to a protected AES encrytion. Another example at l539, the authors claim that if the number of channels of the dense layer is low, then the computational complexity increase. However if n is the input dimension and c the number of channel, then the number of computation is equal to c*n and it increases with the number of channel. Finally, the comparaison with the state of the art models lacks some robustness. I recommand to perform a 10-fold crossvalidation or a validation method with a high number of different training/testing steps: the results obtained with these well-known methods will be more significant than a single training/testing evaluation and will reduce bias. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: ATTENTION ENHANCED CAPSULE NETWORK FOR TEXT CLASSIFICATION BY ENCODING SYNTACTIC DEPENDENCY TREES WITH GRAPH CONVOLUTIONAL NEURAL NETWORK Review round: 1 Reviewer: 1

Basic reporting: This paper makes a focused study on the text classification problem from the text syntactic relationship, sequence structure, and semantics. Specifically, the authors first comprehensively summarize the shortcomings of existing models and conclude the challenges for text classification. Later, the author introduced GCN to extract syntactic information for dependencies relationship representation on the one hand and built multi-head attention to encoding the different influence of words to enhance the effect of capsule networks on text classification on the other hand. In the end, the author proved that CapsNet, GCN, and multi-head attention have an integration effect for text classification through experiments. In general, the author states the research problems very clearly, and presents a very interesting and effective model named Syntax-AT-CapsNet. The paper is organized logically. Thus, I strongly suggest this work should be accepted. Experimental design: The experiment designed by the author is very sufficient, and the effectiveness of the model (Syntax-AT-CapsNet) proposed by the author can be well proved through the experiment. However, the evaluation metrics could be further enhanced. More diverse evaluation indicators could be added to evaluate the effectiveness of the model in a single-label classification task. Validity of the findings: The author verified the validity of the conclusion through theoretical description and experimental verification, and it has strong persuasiveness and credibility. Additional comments: Firstly, several typos should be corrected, such as: 1. At lines 160-161, “... (BERT)also also takes multi-head attention as its basic component.” -> (also) “... (BERT) also takes multi-head attention as its basic component.” 2. At lines 281, “... denotes that Nword vectors in the sentence are connected in series...”-> (N-word) “... denotes that N-word vectors in the sentence are connected in series...”. In addition, it is suggested that Fig.4 and Fig.6 should be more clear. Maybe it would be better to increase the resolution of the picture. Finally, the format of the formula in the paper should be unified.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: ATTENTION ENHANCED CAPSULE NETWORK FOR TEXT CLASSIFICATION BY ENCODING SYNTACTIC DEPENDENCY TREES WITH GRAPH CONVOLUTIONAL NEURAL NETWORK Review round: 1 Reviewer: 2

Basic reporting: The paper is logically organized, the research content is systematic and the structure is reasonable. The paper considers that the text syntactic relationship and word sequence are important and useful for text classification. So it analyzes the sequence structure and syntactic relations of text, combines the syntactic relationship, sequence structure, and semantics for text representation. And proposes a novel model that utilizes GCN for syntactic relationship, multi-head attention for words and corporate them in capsule network for text classification. The experimental results prove the effectiveness of the proposed method. Experimental design: The problem researched in this paper is clearly defined. The experimental part is related to the research problem, which is designed reasonably. Experiments on single classification and multi-classification tasks, Syntax module and module ablation experiment were carried out in the experimental part. Experiments prove that graph convolutional neural networks, capsule networks, and multi-head attention have an integrated effect on text classification tasks. And the module ablation experiment shows that each module has a certain role in improving the text classification effect. Methods described with sufficient detail and analyzed the results accordingly. Validity of the findings: The paper analyzes the background and research status of the problem, proposes an attention enhanced capsule network-based text classification model, which is quite innovative. The technical introduction of the model is clear. The analysis of the experimental results is reasonable, and the data charts are clear. The results of the experiment also verify the effectiveness of the proposed model. Additional comments: 1. It is better to briefly introduce the formulas in section 3 and separately add a new section to describe the details of the algorithm by pseudo-code overall. 2. It's better to give some insight into each module. Not just simply put the conclusions of others’ papers in “related work”. For example, what is the characteristics of CapsNet? Why it is better? I really want to see your own understanding with the formulas. The same as to GCN and multi-head attention. The introduction of those modules is not deep enough. 3. Elaborate on the principle of Syntactic Dependency Tree. 4. There are several mistakes about the format. Such as the label “Figure” in section 3.3 paragraph 2, misses number 3; Section 4.1 paragraph 1, the label “Table” misses number 1, section 4.2 (1), the label “Table 33” repeats number 3, etc. 6. In section 3, the authors should add citations of the formulas. The description of Equation 17 needs to be clearer. ( Alaparthi, and Mishra 2020) proposed a pre-trained model of language representation (BERT)also also takes multi-head attention as its basic component. "n-grams" should be "N-grams". 5. Why Table 6 can show that the motivation of using graph neural network to encode syntactic information and other models is correct?

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: ATTENTION ENHANCED CAPSULE NETWORK FOR TEXT CLASSIFICATION BY ENCODING SYNTACTIC DEPENDENCY TREES WITH GRAPH CONVOLUTIONAL NEURAL NETWORK Review round: 1 Reviewer: 3

Basic reporting: This article proposes a new network architecture for the text classification task. In the feature extraction phase, it uses multi-head attention to encode words' embeddings to extract long-range dependency information, uses GCN to encode Syntactic Dependency Tree to extract syntactic info, and subsequently concatenate them by add operation. Then pass it through a CapsNet for classification. Experimental design: Generally, the motivation is clear and the experimental work is solid and comprehensive. Validity of the findings: It's a combination of existed works but results in better performance than some of the previous works. Additional comments: Follows are some comments to further improve the paper: 1. (1) The English writing of the thesis needs to be further improved and the tenses be unified. Some contents are in the past tense and some are in the present tense. (2) There are a lot of errors in the cross-reference of tables and charts in the paper. There are no corresponding labels behind many tables and figures. (3) The writing of formulas also needs to be more standardized. The corresponding "," needs to be added after most formulas, and where should be lowercase without uppercase. (4) The data set used needs to be added to the corresponding reference instead of a footnote. 2. There are many kinds of methods that can extract syntactic information. Why do the authors choose the GCN instead of other models? I mean what is the benefit of GCN? 3. In addition to the experimental comparison with several baselines, it is also necessary to add the experimental results compared with some other existing algorithms.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: ATTENTION ENHANCED CAPSULE NETWORK FOR TEXT CLASSIFICATION BY ENCODING SYNTACTIC DEPENDENCY TREES WITH GRAPH CONVOLUTIONAL NEURAL NETWORK Review round: 2 Reviewer: 1

Basic reporting: The authors have addressed all my concerns and thus I think it can be accepted. Experimental design: The authors have addressed all my concerns and thus I think it can be accepted. Validity of the findings: The authors have addressed all my concerns and thus I think it can be accepted. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: DATA PRIVACY DURING PANDEMICS: A SYSTEMATIC LITERATURE REVIEW OF COVID-19 SMARTPHONE APPLICATIONS Review round: 1 Reviewer: 1

Basic reporting: This literature review has followed a comprehensive methodology to study different digital technologies such as CTAs which have been used to control the COVID-19 spread. The manuscript has been written clearly using professional English which is easy to understand. Experimental design: The article content has fallen in the Aims and Scope of PeerJ Computer Science. The study has done a thorough investigation with a high ethical standard on all kinds of contact tracing applications (CTAs) and reviewed 800+ papers published in some prestigious journals such as Science Direct, IEEE, Scopus, etc. Validity of the findings: Conclusions are well stated, and valid. The result and recommendations made are beneficial to governments and countries all over the world. Additional comments: Figure 1 is a bit ambiguous. I would suggest changing it as in the attachment.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: DATA PRIVACY DURING PANDEMICS: A SYSTEMATIC LITERATURE REVIEW OF COVID-19 SMARTPHONE APPLICATIONS Review round: 1 Reviewer: 2

Basic reporting: - The paper is very well written and in simple words. It is thus easy to understand which is a very important characteristic of any document. - The sections are titled as questions. While this is innovative and gets the point across, I prefer a more conventional approach to naming sections - Each section and each point has to be discussed in much more detail. The authors superficially discuss the ideas of various works and do so in a rather heterogeneous manner with separate paragraphs dedicated to separate papers. I would look forward to a more homogeneous discussion of ideas with disparate work smoothly falling into the discussion. Experimental design: - Very little discussion is dedicated to the users’ privacy and its breach. The authors discuss laws that are prevalent around the world, “ethical” issues, and “values”. I would prefer a much more detailed analysis of the issues around privacy and its breach first in a general context and subsequently specifically with respect to the pandemic and the tracing apps. - The authors discuss the attempts made by health authorities to protect and preserve privacy. I feel this discussion is limited. The authors should also include the efforts made by governments, health authorities, world bodies, and also those made at individual levels. Validity of the findings: - Conclusions on various aspects of the study are not drawn. The authors talk about the issues based on the contributions of various endeavours but do not draw appropriate conclusions based on these. Additional comments: - A few figures, block diagrams etc. would help the reader more effectively grasp the ideas being discussed. - The authors may want to justify the content and both ends for a better appearance.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: DATA PRIVACY DURING PANDEMICS: A SYSTEMATIC LITERATURE REVIEW OF COVID-19 SMARTPHONE APPLICATIONS Review round: 2 Reviewer: 1

Basic reporting: No coment Experimental design: No comment Validity of the findings: No comment Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: DATA PRIVACY DURING PANDEMICS: A SYSTEMATIC LITERATURE REVIEW OF COVID-19 SMARTPHONE APPLICATIONS Review round: 2 Reviewer: 2

Basic reporting: My concerns on basic reporting have been addressed. Experimental design: My concerns regarding the study design have been addressed. Validity of the findings: My concerns regarding the validity of the findings have been addressed. Additional comments: All my concerns have been addressed. I am happy to recommend that the paper be accepted.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PARTICIPATION IN WIKI COMMUNITIES: RECONSIDERING THEIR STATISTICAL CHARACTERIZATION Review round: 1 Reviewer: 1

Basic reporting: The manuscript uses appropriate language, references relevant research, and follows a standard article structure. Experimental design: The data collection and analysis strategy is well-motivated, clearly described, builds on prior work, extends into a new domain, and is rigorously applied. Validity of the findings: The data was appropriately retrieved and analyzed using constructs/methods, the findings replicate other empirical findings about power laws in social systems, discussion emphasizes truncated power law as appropriate fit and the corresponding mechanisms and interpretations that generate them. Additional comments: I would add the following references into the manuscript: Mitzenmacher, M. (2004). A Brief History of Generative Models for Power Law and Lognormal Distributions. Internet Mathematics, 1(2), 226–251. https://doi.org/10.1080/15427951.2004.10129088 Andriani, P., & McKelvey, B. (2009). Perspective—From Gaussian to Paretian Thinking: Causes and Implications of Power Laws in Organizations. Organization Science, 20(6), 1053–1071. https://doi.org/10.1287/orsc.1090.0481

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PARTICIPATION IN WIKI COMMUNITIES: RECONSIDERING THEIR STATISTICAL CHARACTERIZATION Review round: 1 Reviewer: 2

Basic reporting: I am happy to review this article, which examines the distribution of participation in online wiki communities. Overall, the language of this manuscript is adequate. There are a number of places where the language feels awkward to a native English speaker, but only rarely do these weaknesses hinder comprehension. I have marked some of the most egregious examples in the attached PDF. There were a few places, most notably at the end of the introduction, where \ref's were broken. It looks like this may have been because the style of the manuscript does not include section numbers? The paper identifies the most important relevant literature and is well-situated from a methodological perspective. I think that the authors could have done more to explain the practical importance of the research and to identify more concretely the practical implications of identifying a context as having distributions which are fit better by one function versus another. In other words, what is the scientific and theoretical benefit of applying Broido and Clauset's approach to all of these new contexts? In general, the structure of the article was effective. The figures were well done and persuasive. I found the multiple colors of Figure 1 a bit confusing. Does the cutoff point represent the mean, for example? This should be clarified, and referenced in the legend. I found Figure 2 engaging, but I wished that the edges were colored or weighted based on the percentage, to make it easier to distinguish the "winners" visually. At least in my copy, Figure 3 appears to have some compression artifacts and should be produced as a vector image (also Figures 5 and 6). Finally, I thought that Figure 4 might be more persuasive if it also included empirical data points (although it's possible that this might make the figures too noisy). Regarding the data, the data and code include appears to be adequate for running the statistical analyses. I was unable to find the code used to actually gather the edits per person, as described in lines 195-204. Nor was this raw data made available. I did not run any of the code to test it. Experimental design: The overall design of this paper is appropriate for a computer science journal, well defined, and well executed. The methods are well situated in previous work and well described. As explained above, I do think that the authors could and should do more to identify the knowledge that this approach gives us that we didn't have before, especially in the Introduction / Background section. I have only two, fairly minor suggestions for the methods section. First, the authors claim to consider only communities with 100 contributors at times, and at other times those with 100 registered users. These are different measures, and it should be clear which cutoff is the actual cutoff. If it is registered users, does that mean that unregistered users (i.e., "IP users") are also removed? If so, this should be made clear and justified. Second, the authors should explain why 8K wiki endpoints were not available. I would guess that this is because the wikis no longer existed, but that should be made clear. Validity of the findings: The findings are well-supported and the analysis of this paper appears both reasonable and statistically sound. Overall, the authors make a convincing case that wikis are typically well-described by a truncated power law. As mentioned above, I would have liked to have a more substantive understanding of what the conclusions infer for our understanding of how these communities operate, as well as what else we might be able to do with this approach. When these sorts of explanations occurred, I was often unconvinced. For example, the manuscript seems to argue that the findings show that high-volume contributors differ from low-volume contributors, but this seems like it would already be very likely. Indeed, if anything a power-law distribution would suggest that they are more different from each other (in the sense that the discrepancy in the number of edits is larger). I believe that the concluding remarks section should be restructured from bullet points to paragraphs. I found the description of generalizability at ~398-401 confusing. Additional comments: Overall, I found this paper to be convincing in showing that a truncated power law is a reasonable distribution for characterizing wikis. The analysis is narrow but well-executed and while I provided a number of suggestions for improvements to the presentation and discussion of the results, I think that the paper shows a number of strengths.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: TOWARDS UBIQUITOUS REQUIREMENTS ENGINEERING THROUGH RECOMMENDATIONS BASED ON CONTEXT HISTORIES Review round: 1 Reviewer: 1

Basic reporting: 1. the work is prepared well and having good background of the issues considered in the proposed model 2. Nhatos architecture using the Technical Architecture Module (TAM) modeling specification has been considered, which is right move Experimental design: 1. the experiments made in the work are on the following nodes: a. Relational Entity of the Nhatos Model b. Profile of Participating Teams c. Recommendations Made by Project d. Recommendations Through Analysis of Context Histories 2. The RESTFull API Application provides a communication channel between the Hybrid Application and 421 the Context Storage application has been implemented using right approach, however, more consistency in analysis is needed 3. Learning phase steps to be explained in better way Validity of the findings: findings have been validated and aps have been consistent the data used and provided is consistent, however, more comparative study is needed Additional comments: need to have relook at comparative study

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: TOWARDS UBIQUITOUS REQUIREMENTS ENGINEERING THROUGH RECOMMENDATIONS BASED ON CONTEXT HISTORIES Review round: 1 Reviewer: 2

Basic reporting: # Literature work should be improved in the paper as it is very limited. Add one or Two table of comparison along with line diagram of existing work. # The proposed methodology and design must be compared with existing work. Dedicated separate sub-section for this. Also, use block/line diagram for such proposed methodology. Experimental design: # The work is under the Aims and scope of this journal and also, presented with scientific contribution of this study as use of the similarity analysis of this work. # The article proposed "Nhatos" which is a computational model for ubiquitous requirements management . This is very appreciated. # The experiment demonstrated that the model achieved more than 70% stakeholder acceptance of the recommendations. This is justified Validity of the findings: # Conclusion of this paper can be written in systematic way Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: TOWARDS UBIQUITOUS REQUIREMENTS ENGINEERING THROUGH RECOMMENDATIONS BASED ON CONTEXT HISTORIES Review round: 1 Reviewer: 3

Basic reporting: The paper titled, "Towards ubiquitous requirements engineering through recommendations based on context histories" proposes a technique to recommend requirements on the basis of context histories of projects. The paper is generally well written and covers literature in an adequate manner. The authors also provide a comparison of related works with the proposed approach and the developed tool in tabular form (Table 1) which shows the usefulness of the Nhatos model as compared to other similar methods in the literature. The figures in the paper are generally in good shape however, Figures 1 and 3 give a blurred look when the pdf of paper is zoomed in for better reading perhaps their quality can be improved. The raw data has been shared and the interpretation of the approach and results are self contained. However, I could not open the links (https://github.com/robsonklima/nhatos api and https://github.com/robsonklima/nhatos front end) which is a requirement for this journal that a working software version should be available in an online repository. Experimental design: The research proposed in this paper is definitely within the scope of the journal and the research questions identified are well defined, relevant and meaningful filling the gap of recommending requirements on the basis of context histories of projects. However, I have the following questions for the authors: 1) The basic project information was in Portuguese and it was translated to English using the Google Cloud Translation API. The accuracy of this API is 100%, how did the authors make up for any discrepancies in the translation and how potentially this would have impacted the results currently produced by Nhatos? 2) On page 6 of the paper at line 234, the authors talk about the semantic similarity of requirements during the analysis of requirements. However, it will be great if they can shed some light on which semantic similarity measures for NLP have been used and why plus whether there was any syntactic analysis done prior to it or not? Validity of the findings: Since, I was unable to find and download from these two links (https://github.com/robsonklima/nhatos API and https://github.com/robsonklima/nhatos front end), therefore, it is difficult to comment on the replication and reproducibility of results. In the next version I would expect that the link is live and the files retrievable from these links. On page 14, Table 3, the data provide show acceptance percentages of five different projects none of which is above 68.7% the last row of the table shows the approval percent which is computed to be 75.1%. How is this approval percent computed and how does it go well above the accepted requirements percentage? Second last row of Table 3, shows an acceptance percent value of 61,6 which should be 61.6 The conclusions are well stated and linked to research questions along with providing possible future directions. However, my final question authors is how much would we save in terms of time by using Nhatos model specifically given the fact that it some times generates recommendations too many to be assessed by the team members as mentioned on page 15, lines 529-30? Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: GRAPH COLORING USING THE REDUCED QUANTUM GENETIC ALGORITHM Review round: 1 Reviewer: 1

Basic reporting: The manuscript aims to solve the graph coloring problem using the Reduced Quantum Genetic Algorithm. The general structure of the manuscript is respectable with good professional English and clear links between sections. However, there are some weaknesses in the basic reporting that must be revised, from which we can briefly cite the following points: The list of references, especially in the literature review section needs to be updated. Authors are advised to use more recent references. Some terms and definitions need to be detailed. For instance, quantum computing is not well defined and detailed although it is a fundamental concept in the solving approach. Some assumptions were cited by the authors without any explanation or justification. For example, authors said in lines 37 and 38 ‘On the other hand, quantum computation is very powerful for solving various problems due to its 38 specific properties and phenomena such as entanglement, interference, and exponential parallelism. (Need to add a reference to justify). Experimental design: The results section is very short and does not contain depth analysis to prove the efficiency of the proposed approach. No Rigorous investigation was performed in the manuscript. The results section needs to be revised with consideration to the following points: • To add a comparative study of the results (and not metrics) with similar solving approaches applied to the same problem (the graph coloring problem). • To perform a statistical analysis to validate the high performance of the proposed approach. • Apply the solving approach to more practical, complex, real-life problems and not only the graph coloring problem. Validity of the findings: No innovative points and contributions can be easily found in the proposed approaches. Authors clearly cited in the Methodology section (lines 162, and 163, 167 and 168) that they will use the Reduced Quantum Genetic Algorithm proposed by Udrescu et al. (2006), So what is the contribution of the authors? Authors do not present their contribution clearly - where is the novelty comparing to existing methods? In addition, the proposed solving algorithm must beapplied to more applicable, real-life problems to prove its effeciency and strengh. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: GRAPH COLORING USING THE REDUCED QUANTUM GENETIC ALGORITHM Review round: 1 Reviewer: 2

Basic reporting: Need Clear Definitions Experimental design: The authors have to compare the proposed algorithm with more number of existing algorithms Validity of the findings: Novelty not assesed Additional comments: Review Comments: In this work, the authors proposed the reduced quantum genetic algorithm for graph coloring. Although, the approach could be of interest but there are some major concerns which should be addressed. 1. The Figure 8 and 9 are not clear. I suggest the authors to provide the axis titles that makes more understanding of the figures for readers. 2. In Line No.281, the authors mentioned “adders need 2 qubits” and in Line No: 283, the authors mentioned “4 qubits are used for fitness value representation”. How the parameter values are set to specific threshold value? The authors have to justify the parameter values. 3. In Line No. 311, the authors mentioned that the proposed algorithm determines the chromatic number in Figures 17 and 18. But the details are not explained in the Figures description. 4. The authors didn’t mention about the system specification. 5. The authors should make available their software with documentation for its usage and examples for testing in a public repository like GitHub. 6. Limitations of the proposed study have to be included in the manuscript. 7. The authors shown the experimental results for their proposed algorithm. However, they have not compared with other recent optimization algorithms such as Modified Cuckoo algorithm, ABC optimization algorithm, Memetic Teaching–Learning-Based Optimization algorithm and so on. I suggest the authors to compare the proposed algorithm with more number of existing algorithms. • Mahmoudi, Shadi & Lotfi, Shahriar. (2015). Modified Cuckoo Optimization Algorithm (MCOA) to solve Graph Coloring problem. Applied Soft Computing. 33. 10.1016/j.asoc.2015.04.020. • Dökeroğlu, Tansel & Sevinç, Ender. (2021). Memetic Teaching-Learning-Based Optimization Algorithms for Large Graph Coloring Problems. Engineering Applications of Artificial Intelligence. 102. 10.1016/j.engappai.2021.104282. • R. S. Tomar, S. Singh, S. Verma and G. S. Tomar, "A Novel ABC Optimization Algorithm for Graph Coloring Problem," 2013 5th International Conference and Computational Intelligence and Communication Networks, 2013, pp. 257-261, doi: 10.1109/CICN.2013.61. • Aragón Artacho, F.J., Campoy, R. Solving Graph Coloring Problems with the Douglas-Rachford Algorithm. Set-Valued Var. Anal 26, 277–304 (2018). https://doi.org/10.1007/s11228-017-0461-4 8. I suggest the authors to provide the significance of the proposed algorithm to the society. This manuscript standard is not suitable for publication in this esteemed journal. Hence, the manuscript needs “major revision” for further steps.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: GRAPH COLORING USING THE REDUCED QUANTUM GENETIC ALGORITHM Review round: 2 Reviewer: 1

Basic reporting: The authors have addressed the concerns I have raised adequately (expect the third comment of the Experimental Design section) and they have improved the paper largely. I think the paper is now accepted and publishable. Experimental design: The experimental design was improved. Validity of the findings: The findings section was improved by using the comparative study. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PRESB-NET: PARAMETRIC BINARIZED NEURAL NETWORK WITH LEARNABLE ACTIVATIONS AND SHUFFLED GROUPED CONVOLUTION Review round: 1 Reviewer: 1

Basic reporting: pros: The authors present a new binary network architecture, namely PresB-Net, which utilizes channel-shuffle and group binary convolution operations to increase the capacity of binary features without introducing extra computing costs. Besides, the proposed "Biased PReLU+LN+Biased PReLU+BN" module seems to solve the imbalance between different channel groups. Overall, the approach may shed some light on the design of compact binary networks and accurate BNN training. I personally think the proposed method is simple yet effective. The technique part of this paper is easy to follow. The background and related works have been thoroughly discussed, such as group convolution, channel shuffle, and binary convolutions. cons: Since "group conv+channel shuffle" is a common setting in compact network design, I'm not sure whether PresB-Net is the first attempt to double the input channel number? The proposed "Basic block" seems to be a combination of ShuffleNet and DenseNet. Experimental design: cons: It would be better to fully evaluate the performance of PresB-Net, though it has achieved noticeable improvements over baseline results on CIFAR-100. Besides, I expect an ablation study shown in table form. The current manuscript makes it hard to understand the differences between the listed terms. Last but not least, the authors present a new architecture, however, I found no ablation study on the network design, e.g., why use "Biased PReLU+LN+Biased PReLU+BN" instead of "Biased PReLU+LN+BN"? Validity of the findings: cons: The authors focus on "fast output computation with low hardware costs" yet no real speedup over the FP32 counterpart has been reported in this paper. Note that when the mean and variance of LN are computed at inference time, the memory overhead and extra computing costs can be unacceptable in real-world applications. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PRESB-NET: PARAMETRIC BINARIZED NEURAL NETWORK WITH LEARNABLE ACTIVATIONS AND SHUFFLED GROUPED CONVOLUTION Review round: 1 Reviewer: 2

Basic reporting: This paper proposes PresB-Net, which applies shuffled grouped convolution to expand the channel with reduced computation resources. The proposed normalization approach can solve the imbalance between groups in the grouped convolution. Using the biased PreLU activation function with a learnable slope and binary activation with biases to improve performance. PresB-Net consists of stacked basic and expand blocks, the blocks have shortcuts for each binarized grouped convolution. The proposed model can enhance Top-1 final accuracy 1.6%–2.4% over existing ReActNet and AresB-Net on CIFAR. However, there are some issues in this paper which are listed as follows in the form of questions and suggestions. 1. Why is “performance degradation is the most critical problem in BNN models”. 2. The length of the paper is too small and the Section of experiments is rough. It is suggested to add some experimental contents, such as verifying your model on ImageNet, comparing with AresB-Net, ReActNet, MobileNet and ShuffleNet. 3. Whether using PreLU in ResNet can also improve performance? Whether the performance improvement of PresB-Net is related to PreLU? 4. “Therefore, the grouped binarized convolution results are normalized in both channel-wise and batch-wise manner. The proposed normalization approach allows the normalized convolutional results in both channel-wise and batch-wise manners, which can consider data correlations between groups.” Can the output of layer normalization and batch normalization be added directly? “consider data correlations between groups”, it needs to further explain how to consider relevance. 5. “solves the imbalance between groups arising from shuffled grouped convolution by considering the correlation of groups and allows the results of each group to follow a normal distribution, thereby increasing performance by balanced group and increasing the convergence.” How to solve imbalance between groups? 6. Whether can use basic and expand blocks to stack different network models? 7. Fig. 10 and 11 are too rough. 8. In the experiment, the reasons of the comparison results were not described. 9. The parameters of the model were not compared in the experiment. Experimental design: no comment Validity of the findings: no comment Additional comments: no comment

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PRESB-NET: PARAMETRIC BINARIZED NEURAL NETWORK WITH LEARNABLE ACTIVATIONS AND SHUFFLED GROUPED CONVOLUTION Review round: 2 Reviewer: 1

Basic reporting: I have carefully read the authors' feedback and comments from other reviewers. The authors have reflected the revision in the final version, hence I raise the overall score to an acceptance. Experimental design: Whether the experiment results on CIFAR-100 are consistent with ImageNet? Note that the authors mainly conduct experiments on CIFAR-100 and it would be better to evaluate the correlation coefficient between them. Validity of the findings: no comment Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PRESB-NET: PARAMETRIC BINARIZED NEURAL NETWORK WITH LEARNABLE ACTIVATIONS AND SHUFFLED GROUPED CONVOLUTION Review round: 2 Reviewer: 2

Basic reporting: This paper proposes PresB-Net, which applies shuffled grouped convolution to expand the channel with reduced computation resources. The proposed normalization approach can solve the imbalance between groups in the grouped convolution. Using the biased PreLU activation function with a learnable slope and binary activation with biases to improve performance. PresB-Net consists of stacked basic and expand blocks, the blocks have shortcuts for each binarized grouped convolution. The proposed model can enhance Top-1 final accuracy 1.6%–2.4% over existing ReActNet and AresB-Net on CIFAR. Experimental design: none Validity of the findings: none Additional comments: none

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PERSON IMAGE GENERATION THROUGH GRAPH-BASED AND APPEARANCE-DECOMPOSED GENERATIVE ADVERSARIAL NETWORK Review round: 1 Reviewer: 1

Basic reporting: There is a misspelling .....Apperance..even in the Title. It should be taken care of before submission. Experimental design: No comment Validity of the findings: No comment Additional comments: General Comments This paper is concerned with personal image generation through graph-based networks along with decomposed particular attributes. The authors put efforts to propose a generative adversarial network based on the graph. Both qualitative and quantitative experiments were performed to confirm the proposed method. The paper is interesting and will be beneficial to a sizable amount of researchers and students. However the tying error even in the title of the paper ……Apperance ….. should be taken care of before submission. Such errors are also seen in the text (For example Line 120) Particular Comments. 1. Line 78 and Lines 93-94, the two statements about the proposed method may confuse the readers. Are they the same or different contributions? 2. Lines 147-148 Can we get the 18 joint points automatically? 3. I wonder how we can know the probabilities defined in equation 11 are non-zero. The experimental results seem to be significant.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PERSON IMAGE GENERATION THROUGH GRAPH-BASED AND APPEARANCE-DECOMPOSED GENERATIVE ADVERSARIAL NETWORK Review round: 1 Reviewer: 2

Basic reporting: The authors should check the manuscript once again by themselves. - First of all, there are spelling mistakes in the title, and the capitalization rules are not consistent. - The single-byte space after the commas and periods are not consistent. - The operators that are supposed to be Hadamard products are written as e. - Section 3.1 and 3.2 have the same titles. Perhaps 3.1 should be called "Encoder". Experimental design: The method is written in a relatively clear manner, including mathematical expressions, but it does not correspond to Figure 3. - For example, there is no mention of AdaIN, which is included in the Texture Block in the figure. - Also, the text says that the Pose Block contains GCN, but the Pose Block in the figure is not drawn in such a way as to show this. There is no mention of how the three Losses are combined. If the three are to be added together, it should be clearly stated as such. Validity of the findings: no comment Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PERSON IMAGE GENERATION THROUGH GRAPH-BASED AND APPEARANCE-DECOMPOSED GENERATIVE ADVERSARIAL NETWORK Review round: 2 Reviewer: 1

Basic reporting: No comment Experimental design: No comment Validity of the findings: No comment Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: PERSON IMAGE GENERATION THROUGH GRAPH-BASED AND APPEARANCE-DECOMPOSED GENERATIVE ADVERSARIAL NETWORK Review round: 2 Reviewer: 2

Basic reporting: no comment. Experimental design: no comment. Validity of the findings: no comment. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: COMPARING SUPERVISED AND UNSUPERVISED APPROACHES TO MULTIMODAL EMOTION RECOGNITION Review round: 1 Reviewer: 1

Basic reporting: The writing of the paper manuscript is clear and well organized. But there are some problems to be addressed. The citation of some references should be complete, including pages (e.g. Srinivasan, R., & Martinez, A. M. (2018).) Some of the figures should be resized (e.g. Figure 6). Experimental design: The implementation of the methods should be described with sufficient detail for replication, including the setting of parameters. More recently state-of-the-art methods on supervised and unsupervised learning should be considered for comparison. Validity of the findings: The experimental results are provided. But the contribution and novelty of the article are not clear. Additional comments: The authors investigated the supervised and unsupervised emotion classifier performance. The comments should be addressed before further processing.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: COMPARING SUPERVISED AND UNSUPERVISED APPROACHES TO MULTIMODAL EMOTION RECOGNITION Review round: 1 Reviewer: 2

Basic reporting: The content of this article is acceptable on the general with novel research topic and abundant detailed experimental results. Yet it still requires certain revisions to meet the publication standard, which are listed below in several aspects. 1. Language description: The descriptions in this article are unclear and hard to unsderstand at multiple places, such as Lines 56-57, 105-110, 229-231, 237-239, 372-374, etc. Please check the language used in the article. 2. Literature references: This article provided various prior works and references, however, the methods compared in the experiments are not fully included. For example, the late fusion methods mentioned (maximum rule, sum rule, product rule, weight criterion, rule-based, and model-based methods) have not been referenced, and the early fusion methods have not been explicited presented. Please properly cite all evaluated works in this paper. 3. Article structure: This article is organized in clear and reasonable structure. However, the discussion section should be more compact and conclusive, highlighting the major findings and contributions of this article. Experimental design: Various experiments have been conducted for this research. Minor improvements are suggested as follows: 1. The optimal numbers of clusters for the traditional unsupervised approaches are said to be 2 or 6 through CH index and Silhouette score analysis (Lines 220-225). Experimental results should be presented to validate this statement. 2. In feature extraction and pre-processing section, certain data cleaning process has been conducted (Line 148). Please explain the reason for such process. 3. Multiple features for AUs are computed as video features (Lines 154-157), please give explicit definition or computational detail of each feature. Validity of the findings: no comment Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: COMPARING SUPERVISED AND UNSUPERVISED APPROACHES TO MULTIMODAL EMOTION RECOGNITION Review round: 1 Reviewer: 3

Basic reporting: Very good paper. A few slight comments: I understand that one of the strength of this study is to investigate the variety of emotional categories and to show the importance of each features (voice + face). A little more introduction should be discussed in comparison with previous studies (see https://github.com/EvelynFan/AWESOME-MER) that adopted a multimodal approach. Knowledge of features goes to application to robots. I felt that it would be good if there was a discussion about Human-robot interaction in the Discussion section. Barrett et al (2019: Line 80-81) strictly criticizes emotion recognition in emotional categories like this paper. Citing Keltner is preferable as a position, so I recommend to cite it (e.g., https://psycnet.apa.org/record/2017-30838-004; http://emotionresearcher.com/the-great-expressions-debate/) Experimental design: no comment Validity of the findings: "actor and gender specific aspects also contributed to clustering" What are the implications of this finding? The Unsupervised method seems just to report the characteristics of the stimulus (GEMEP). I want the author to clarify that necessity. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: COMPARING SUPERVISED AND UNSUPERVISED APPROACHES TO MULTIMODAL EMOTION RECOGNITION Review round: 2 Reviewer: 1

Basic reporting: The authors have enhanced the article with more introductions and discussions. Experimental design: No comment. Validity of the findings: No comment. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: ADVERSARIAL EXAMPLE DEFENSE BASED ON IMAGE RECONSTRUCTION Review round: 1 Reviewer: 1

Basic reporting: 1. The paper is easy to understand, but the English language should be improved on grammar and tense. Some examples include lines 241-242, 273, 288-289, and the inconsistency of tense used in Section experiment results. 2. The authors missed some background information on adversarial attacks and defense methods. In the experiment, you used DeepFool attack and compared your proposed method with Pixel Defend, Feature Squeezing, and ComDefend but you didn't introduce these methods in Section Background. 3. The structure of this paper is good. Figures are relevant to the content of the paper. Some problems are in Table 3. The names of the first two columns should be "network" and "method". The caption of Table 3 doesn't provide sufficient information to help understand this table. Specifically, why for each cell there are two numbers (e.g., 93%/93%)? I also don't find the explanation in the paper. 4. I thank you for providing the source code, but you need also to provide a Readme file to describe how to use the code. In addition, there are many Chinese characters in the code and filenames, which should be replaced with English words. The authors should also provide pre-trained models to ensure the experimental results are reproducible. Experimental design: 1. The paper proposed an adversarial defense method based on image compression and reconstruction. However, the detailed structure of the image reconstruction network was not introduced. In the provided source code I saw some different network architectures used on different datasets. The authors should introduce them in the paper. 2. In lines 214-216, the authors described they generated adversarial examples and used them with clear examples to train the image reconstruction network. However, in the source code (cs-59969-MNIST-002.zip/train_turn_defense.ipynb), I found only adversarial examples were used. The authors should double-check the used method and describe it clearly and accurately. 3. The authors conducted experiments on evaluating the performances of the proposed method, the transferability of the proposed methods, and the comparison of the proposed method with other defense methods. The research questions are well defined and meaningful. However, the experiment setting should be improved in some ways. First, ImageNet or tiny-imagenet dataset should be used to evaluate the proposed method because it is the most widely used and most complicated dataset for image classification. Second, for the comparison of the proposed method and other defense methods, more similar defense methods based on image reconstruction such as HGD mentioned in the paper could be considered to compare. Validity of the findings: 1. For the experiment of comparing the proposed method with other defense methods, only result on F-MNIST was reported. The results on the other two datasets should be provided to prove the proposed method is better than other methods in general. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: ADVERSARIAL EXAMPLE DEFENSE BASED ON IMAGE RECONSTRUCTION Review round: 1 Reviewer: 2

Basic reporting: The introduction section demands to be more convincing. Try to structure the introduction section with four paragraphs as follows: i) State the motivation and clearly define the problem to be solved. ii) Make a thorough discussion of the state-of-the-art. iii) Describe your proposal in fair context to other published methods highlighting advantages and disadvantages of these methods. iv) Clearly pinpoint the novelty/contribution of your proposal and briefly describe your findings. Experimental design: The performance of CNN strongly depends on an optimum structure of a network. The training structure in Figure 4 needs to be self-contained such as number of layers, height and width of each layer. Networks with defense method show degraded performance for clean images in Table 2. How do you validate this result? Validity of the findings: Network models in supplement files were not possible to test. It is needed to provide comprehensive readme files to run and test source codes, models and dataset. Is there any limitation of the proposed methodology? Additional comments: The manuscript is overall well written. If there are weaknesses, as I have noted above which need be improved upon before publication.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: ADVERSARIAL EXAMPLE DEFENSE BASED ON IMAGE RECONSTRUCTION Review round: 2 Reviewer: 1

Basic reporting: 1. The paper proposed an adversarial defense method by combining image compression and image reconstruction models. The background of adversarial attack and defense was introduced in detail. The authors conducted comprehensive experiments to evaluate the performance of the proposed method against common adversarial attacks and compared the method with existing defenses. The experiment result shows that the proposed method achieves good performance. 2. There are a few tense inconsistencies in the Section background, which should be revised. 3. In the Section Approach, the overview of the defense method should be described more clearly. The authors could consider showing how the defense method works from taking as the input of the original image to reporting whether the image is an adversarial example in Figure 4. Now Figure 4 just shows the process until the image reconstruction. I cannot easily know how to use the output image to detect adversarial examples from the Figure. Experimental design: The authors conducted experiments on evaluating the performances of the proposed method, the transferability of the proposed methods, and the comparison of the proposed method with other defense methods. The experiment results show the proposed method outperforms other baseline defenses. In Table 4, the author should highlight (bold) the best experiment results, which is better to help compare the performances of these methods. Validity of the findings: The experiment results are well evaluated. The authors provided the source code and detailed instructions for reproducing the experiment. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: ADVERSARIAL EXAMPLE DEFENSE BASED ON IMAGE RECONSTRUCTION Review round: 2 Reviewer: 2

Basic reporting: The revised manuscript has been improved and addressed all the concerns accordingly. The usage of English language is satisfactory. Experimental design: The experiential design and analysis is satisfactory in the revised manuscript. Validity of the findings: no comment Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: ADVERSARIAL EXAMPLE DEFENSE BASED ON IMAGE RECONSTRUCTION Review round: 3 Reviewer: 1

Basic reporting: The revised manuscript has addressed my questions accordingly. Experimental design: Experiments are solid in the revised manuscript. Validity of the findings: no comments Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: WHO PARTICIPATES IN COMPUTER SCIENCE EDUCATION STUDIES? A LITERATURE REVIEW ON K-12 SUBJECTS Review round: 1 Reviewer: 1

Basic reporting: Is the review of broad and cross-disciplinary interest and within the scope of the journal? The authors have focused their literature review on three flagship conference proceedings in order to highlight the ways in which researchers describe, or not, the demographics of the students who participated in the studies. The authors maintain that this is important to study for two reasons: first, to inform policy and support replications, and second, to identify demographic attributes that are important for the field to report. As such, the literature review holds broad appeal and ought to be of interest to those working in the multidisciplinary computer science education field. Has the field been reviewed recently? If so, is there a good reason for this review (different point of view, accessible to a different audience, etc.)? The authors note that similar review work has been completed (McGill et al., 2018; Schlesinger et al., 2017), although they seem to indicate that their review is distinct in that it evaluates additional demographic information, disability status in particular. I am not certain that the authors have convinced me that this sensitive information can be widely and appropriately asked across K-12 studies, at least those that occur in the United States. Does the Introduction adequately introduce the subject and make it clear who the audience is/what the motivation is? The Introduction is well written, situates the review among related research, then appropriately narrows its focus on a brief overview of this study’s process and purpose. Experimental design: Is the Survey Methodology consistent with a comprehensive, unbiased coverage of the subject? If not, what is missing? The authors do a commendable job of detailing their process and justifying their use of the three conference proceedings. They do note in Limitations that there was no attempt at inter-rater reliability, or double-coding. This does seem potentially problematic as the authors indicated they deduced information context, rather than from details overtly stated in the studies. It may be that one researcher missed vital information or misunderstood. Such double checking is important to the validity of the authors’ claims. Additionally, the authors do not address what safeguards were in place to ensure that an intervention was not submitted to and accepted at more than one of these conferences and therefore those data were considered more than once. Are sources adequately cited? Quoted or paraphrased as appropriate? The authors cite appropriately, although I am used to alphabetizing in-text citations when there is more than one cite. Relatedly, there are several places (lines 61, 69, and 74) where the formatting of the citation is incorrect. Is the review organized logically into coherent paragraphs/subsections? The review is organized well and logically flows from Introduction to Conclusion. Validity of the findings: Is there a well developed and supported argument that meets the goals set out in the Introduction? I provide more specific information below in General Comments, but overall I am struggling to see the feasibility of including disability status as a socio-demographic metric. I understand the authors’ contention that such a characteristic needs to be considered for generalizability purposes; however, the reality of receiving such information, at least in the US, is highly complicated and would not lend itself to generalizability. This aside, the authors complete their literature review as they set out in the Introduction. I am not certain that the statement “Both for race/ethnic background and for disabilities a pattern can be seen where these categories are only reported on when deviating from the majority” is supported. Is it that these characteristics are only reported when they differ from the majority or could it be that these socio-demographic metrics are problematic to collect? Does the Conclusion identify unresolved questions / gaps / future directions? The Conclusion identifies suggestions for making consistent across CSEd studies the ways we report on demographics. What the authors need to include are real ways to balance the tension between sharing empirical findings about specific students in K-12 studies when the data we are able to collect might be at the school level only. I feel the authors could strengthen their argument and the paper overall by delving more deeply into the articles that report on disability status, as this seems to be their major interest, especially those not focused solely on a certain population of students (ie., blind students). How did these articles report on these disabilities? What was their institutional review policy? How did they obtain this information (ie., from the students, teachers, or parents)? What major findings/conclusions did they offer? The number of articles that fall into this category is quite small, so more closely exploring them may help draw out and support your contention that disability status is a metric essential to report. Given this small number, perhaps it would be wise to broaden your search beyond the flagship conferences. In other words, the authors could share their literature review findings, including the small number of articles that report on disability status, and then provide a detailed overview of articles in CSEd at large (conferences and journals) that do report on disability status. Additional comments: The following are line item/section comments regarding grammar, formatting, or general structure: Line 5 (Abstract): if space permits, it may be helpful to include the year range Lines 25, 38, and 50: authors appear to use ‘pre-college,’ ‘pre-university,’ and ‘K-12’ interchangeably. I think it would read better if one phrase were used; K-12 is most common. These phrases appear throughout the article. Line 99: it seems the RQ should end as “... in K-12 programming” and not into Line 156: grade 6 is included twice in both elementary and middle school; where did you ultimately include it? Gender: is it important to include the gender options the studies permitted the participants to select, or that --% of studies permitted students to opt out or select Other? Also, the finding that males are overrepresented is not terribly surprising, but I feel this interacts with the context of the studies. For example, did the studies where males were overrepresented take place in elective classes or optional experiences such as camps? Race/Ethnic Background Line 178: the sentence that begins on line 177 and includes “studies is represented in,” is poorly worded or missing a reference to a figure. “This is further confirmed by the large number of studies that report on mixed-racial experimental groups, as Figure 3 depicts, which does not correspond with how mixed classrooms typically are in the US (Stroub and Richards, 2013)”: what does this mean? What are mixed classrooms and how are they typically organized/studied in the US? SES and Disabilities I have combined these two subsections as my concerns with them are similar. As a US researcher, I have always used NSLP (free and reduced school lunch eligibility) as a proxy for SES. This follows because we have never found it appropriate to ask students about their family’s SES and this information is publicly available at the school level. When working with a single class or a subset of students in class, it feels inappropriate to report out the school level SES. No university IRB would approve US researchers asking K-12 students, deemed a vulnerable population, about their disability status. FERPA regulations prohibit school personnel from disclosing such information at an individual level to outside researchers. It may be that a teacher could tell a researcher something generic, such as “of the 30 students in this classroom, 5 have IEPs and 4 have 504 plans.” Line 216: “However, given the aforementioned prevalence of individuals with a disability, it seems unlikely that the participant groups of the other 128 papers did not include any of these children”: of course these classrooms included students with disabilities, but the practice of mainstreaming and not delineating students by ability, likely contributed to the small number of studies that report on this characteristic Lines 273-4: the wording of the final sentence is problematic. You have set up a juxtaposition whereby poor inner city cannot include well-educated. Don’t conflate education and income here. Something like “it could be a school in a poor inner city area or a high-income neighborhood” would be better.

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: WHO PARTICIPATES IN COMPUTER SCIENCE EDUCATION STUDIES? A LITERATURE REVIEW ON K-12 SUBJECTS Review round: 1 Reviewer: 2

Basic reporting: In the introduction, the authors adequately introduce the topic by providing sufficient background on pre-university programming education and comprehensively discussing related work on demographics in computer science education studies. They clearly distinguish their work from prior research (e.g., by McGill et al. (2018)) and identify gaps. The desideratum on which the authors build is that there is no comprehensive research on the demographics of computer science education studies between 2014 and 2018. Overall, the paper is well structured and contains detailed sections on methodology, results, and conclusions. The argumentation is comprehensive throughout and easy to follow. Relevant previous literature is referenced appropriately. Figures and tables are appropriately labelled and well described in the respective sections/paragraphs. The paper is written in professional, unambiguous English. The text is technically correct and follows professional standards of courtesy and expression. Suggestions for improvement: - In the “Pre-university Programming Education” section, two improvements should be made to the citation style: (1) (Yang et al., 2015) for example […] -> Yang et al. (2015) for example; (2) Aivaloglou and Hermans (Aivaloglou and Hermans, 2016) performed an […] -> Aivaloglou and Hermans (2016) performed an. - Authors should consider submitting the figures in a better resolution as the diagrams provided are slightly blurred. Please, add a dot after each of the figure labels. - Please, reference Figure 2 in the text. Currently, there is no reference to it in the text. - Please, use CSed abbreviation consequently. Experimental design: The paper is a literature review with a clear formulated study design. It aims to address the wide computer science education community which is in line with the aims and scope of the PeerJ journal. In the methodology section, the authors clearly articulate the research question (What are the demographics of subjects who participate in CSEd studies into K-12 programming?) and comprehensively describe the procedure for literature selection, narrowing the scope of the selected literature to studies of the pre-university target group with the focus on programming education. The selection of demographic categories is clearly reasoned, and the description of the analysis process is extensively described. The paper also includes a detailed discussion of the limitations of the literature review. All raw data is provided for better understanding and comprehension of the results, which I found helpful. Suggestions for improvement: - Although the authors refer to computer science education in the title of the paper, the literature used for the review is limited to programming education only. However, computer science education is much broader than learning programming. Please, include this information as a limitation and explain in the introduction why you focused on programming education as one of the aspects of computer science education. - In the introduction, the authors justify the selection of SIGCSE, ITiCSE and ICER by arguing that these are “flagship conferences” in the CSEd community. Please, support this argument with references. - In presenting the results in the “Race/Ethnic Background” section, the authors go beyond describing the results (“This leads us to hypothesize that reporting bias is at play here”). However, since this argument is already about evaluating the results, I recommend moving it to the “Ethical and Legal Considerations” section. Validity of the findings: The findings are appropriately structured and presented according to the categories proposed in the survey methodology section. In the conclusions, the authors reflect on the findings based on the research question and the state of the arts presented in the introduction. The authors also identify future directions relevant to the entire computer science education community, suggesting, for example, the development of standards/guidelines for collecting and reporting demographic information in empirical studies of K-12 CSed. Overall, I found the conclusions to be well articulated and linked to the original research question. The authors fully achieved the objectives stated in the introduction and survey methodology. Additional comments: I find the results of this literature review enriching for research in computer science education. Many thanks to the authors of this work!

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: WHO PARTICIPATES IN COMPUTER SCIENCE EDUCATION STUDIES? A LITERATURE REVIEW ON K-12 SUBJECTS Review round: 2 Reviewer: 1

Basic reporting: The authors have addressed all my previous concerns. Experimental design: The authors have addressed all my previous concerns. Validity of the findings: The authors have addressed all my previous concerns. Additional comments: No additional comments

You are one of the reviewers, your task is to write a review for the article. You will be given the title of the article, the number of the round in which the article is located, and your order among the reviewers.

Title: WHO PARTICIPATES IN COMPUTER SCIENCE EDUCATION STUDIES? A LITERATURE REVIEW ON K-12 SUBJECTS Review round: 2 Reviewer: 2

Basic reporting: In the introduction, the authors adequately introduce the topic by providing sufficient background on pre-university programming education and comprehensively discussing related work on demographics in computer science education studies. They clearly distinguish their work from prior research (e.g., by McGill et al. (2018)) and identify gaps. The desideratum on which the authors build is that there is no comprehensive research on the demographics of computer science education studies between 2014 and 2018. Overall, the paper is well structured and contains detailed sections on methodology, results, and conclusions. The argumentation is comprehensive throughout and easy to follow. Relevant previous literature is referenced appropriately. Figures and tables are appropriately labelled and well described in the respective sections/paragraphs. The paper is written in professional, unambiguous English. The text is technically correct and follows professional standards of courtesy and expression. The authors included suggestions for improvement from the first review. Experimental design: The paper is a literature review with a clear formulated study design. It aims to address the wide computer science education community which is in line with the aims and scope of the PeerJ journal. In the methodology section, the authors clearly articulate the research question (What are the demographics of subjects who participate in CSEd studies into K-12 programming?) and comprehensively describe the procedure for literature selection, narrowing the scope of the selected literature to studies of the pre-university target group with the focus on programming education. The selection of demographic categories is clearly reasoned, and the description of the analysis process is extensively described. The paper also includes a detailed discussion of the limitations of the literature review. All raw data is provided for better understanding and comprehension of the results, which I found helpful. The authors included suggestions for improvement from the first review. Validity of the findings: The findings are appropriately structured and presented according to the categories proposed in the survey methodology section. In the conclusions, the authors reflect on the findings based on the research question and the state of the arts presented in the introduction. The authors also identify future directions relevant to the entire computer science education community, suggesting, for example, the development of standards/guidelines for collecting and reporting demographic information in empirical studies of K-12 CSed. Overall, I found the conclusions to be well articulated and linked to the original research question. The authors fully achieved the objectives stated in the introduction and survey methodology. Additional comments: No additional comments