SENSIPLUS-LM: A Low-Cost EIS-Enabled Microchip Enhanced with an Open-Source Tiny Machine Learning Toolchain

Round 1

Reviewer 1 Report

The authors should explain better how difficult it is to transform the models in C code. Is it possible for all the ML models?

The lack of the paper could be a limited example of usage in a real case. Did the authors think about applying the toolchain for anomaly detection on board? How are the performances in that case?

- Check missing references and sentences at "The main open issues are reported in ?? and ?? 89 First of all there is a leak in benchmarking Secondly there is a leak in security Integra- 90 tion with API libraries for SPC control"

Author Response

Thanks to all editors and reviewers for their precious work in dealing with our manuscript. We answered point by point and provided the manuscript modification in red. Furthermore, in the following, we reported all issues raised by reviewers and proposed a response for each of them. As a general comment, we are pleased that reviewers appreciated the topic and the work itself. We believe that modifications following their comments will improve the overall quality of the manuscript.

REVIEWER 1

Issue 1: The authors should explain better how difficult it is to transform the models in C code. Is it possible for all the ML models?

Response 1:  Thank you for your comment. More information on the choices made for implementing the toolchain system and some considerations on implementing models in C for embedded devices have been added in the introduction.

Issue 2: The lack of the paper could be a limited example of usage in a real case. Did the authors think about applying the toolchain for anomaly detection on board? How are the performances in that case?

Response 2: Thank you for your comment. To give a response to your questions, It should be noted that embedded devices are not suitable for implementing any machine learning model due to their limited computational resources. For example, it is difficult to implement on low-cost MCU deep neural networks or, more in general, algorithms that require a high quantity of memory or computing power.

In the current version, different algorithms have been included in the toolchain and are available for C code generation. It's important to highlight that the software structure of the toolchain allows to easily add new ML models simply by extending a Java class, as better explained in the paper (Section 3.1.2).

The algorithms implemented in the current version have been designed only for classification, but the toolchain could also include regression or anomaly detection algorithms as an extension. In this case, from a performance point of view, we don't expect significant differences compared to the behavior of the software generated by the toolchain for a classifier. Of course, this statement depends on the selected algorithm for anomaly detection.

Furthermore, the authors have already developed a system composed of anomaly detection plus classifier stages, and its description has been published in [10]. As a future step, we plan to integrate our current toolchain version with the already developed and tested anomaly detection module.

[10] Luca Gerevini, Gianni Cerro, Alessandro Bria, Claudio Marrocco, Luigi Ferrigno, Michele Vitelli, Andrea Ria, Mario Molinara, An end-to-end real-time pollutants spilling recognition in wastewater based on the IoT-ready SENSIPLUS platform, Journal of King Saud University - Computer and Information Sciences, 2023,ISSN 1319-1578, https://doi.org/10.1016/j.jksuci.2022.12.018.

Issue 3:  Check missing references and sentences at "The main open issues are reported in ?? and ?? First of all there is a leak in benchmarking Secondly there is a leak in security Integration with API libraries for SPC control"

Response 3: Thanks for the comment. The sentences were unintentionally left in draft mode. We have rewritten that period to make it readable and integrated with the remaining part of the subsection.

Reviewer 2 Report

The article is interesting and presents a useful Arduino library to facilitate sensing, data collection, and processing on an edge device.

However, due to the repeated writeup in several self-cited references, it makes the style of the writing suit a whitepaper (or a conference paper) rather than a technical journal article that requires depth and breath.

I have detected 13 references cited in the article that belong to the authors out of 37 references. That is by far the highest self-citation exercise I have ever come across.

I understand some references like [33] and [34] are necessary to cite, but many others are cited redundantly. 

Section 3 can be removed completely as the detailed description of the chip is already presented in [33]. A reference mention in passing to the interested readers is sufficient. 

The presentation can be improved by placing Table 3 closer to where it is cited in the text and bringing Appendix 1 as part of the article body.

Section 2 comprises subsection 2 and 2.1. I think they can be merged.

Author Response

Thanks to all editors and reviewers for their precious work in dealing with our manuscript. We answered point by point and provided the manuscript modification in red. Furthermore, in the following, we reported all issues raised by reviewers and proposed a response for each of them. As a general comment, we are pleased that reviewers appreciated the topic and work. Modifications following their comments improved the overall quality of the manuscript.

Issue 1: The article is interesting and presents a useful Arduino library to facilitate sensing, data collection, and processing on an edge device. However, due to the repeated writeup in several self-cited references, it makes the style of the writing suit a whitepaper (or a conference paper) rather than a technical journal article that requires depth and breath. I have detected 13 references cited in the article that belong to the authors out of 37 references. That is by far the highest self-citation exercise I have ever come across. I understand some references like [33] and [34] are necessary to cite, but many others are cited redundantly. 

Response 1: Thanks for this important comment. Actually, we added all references that, during the writing process, appeared as pertinent to the work we were proposing. Since we have worked for some years on the topic, especially concerning the SENSIPLUS platform, most of them derived from our previous efforts on the same subject. Nevertheless, we perfectly understand the reviewer’s concern about the impact of the numerosity of our previous work over the total number of citations. For these reasons, we reduced it to 7 (from 13), just keeping what we really believe unavoidable to cite for a better understanding of the context where the work is placed.

Issue 2: Section 3 can be removed completely as the detailed description of the chip is already presented in [33]. A reference mention in passing to the interested readers is sufficient. 

Response 2: We agree with the reviewer that an entire Section for the microchip description is too much, but we also believe that having some minimal information about the adopted instruments and tools in the manuscript can be useful for the reader, even though it can be retrieved from the literature. Therefore, to find a good trade-off between such needs, we decided to remove Section 3 and rename previous Section 4 (Toolchain architecture) in the current Section 3 (The Sensiplus Microchip and the toolchain architecture), where a minimized version of the previous description is reported at the beginning of the section. This could be a viable solution to fulfill the request.

Issue 3: The presentation can be improved by placing Table 3 closer to where it is cited in the text and bringing Appendix 1 as part of the article body.

Response 3: Thank you for the comment. Table 3 has been edited and moved next to the citing text. To make the content more readable, figure 8 has been incorporated into figure 7 (in subfigure d) and completed by adding an appropriate description. Concerning Appendix 1, since the GUI is an addition that is not strictly necessary to the execution of the toolchain system, we would like to consider it only as an extension. Therefore, we would leave it in its current place. Nevertheless, we can move it in the article’s body whenever the reviewer believes it is fundamental for increasing the paper's readability.

Issue 4: Section 2 comprises subsection 2 and 2.1. I think they can be merged.

Response 4: Section 2 and subsection 2.1 have been merged to make related work easier to read.

Reviewer 3 Report

This paper proposes a system based on a proprietary platform named SENSIPLUS, which is a multi-sensor platform especially devoted to performing Electrical Impedance Spectroscopy (EIS) on an 11-wide frequency interval. The system capabilities of 14 such a platform in this work are exploited for water quality assessment. The joint system, composed of 15 of the measurement platform and the developed toolchain, is named SENSIPLUS-LM, standing 16 for SENSIPLUS Learning Machine. The introduction of the toolchain empowers the SENSIPLUS 17 platform moving the inference phase of the machine learning algorithm to the edge, limiting the 18 needs of external computing platforms. The software part, i.e., the developed toolchain, is available 19 for free download from GitLab, as reported in the following. 

Overall, this paper is well-written and well-motivated. Although I am not an expert in this area, I can understand the contributions and the importance of the proposed approach. But there are still some issues: 

1. The introduction can be revised to add more background and highlight the motivations as well as contributions.

2. Could you please explain more details about Figure 9? Figure A 1 is vague.

3. The experimental analysis in 5.2 is a bit hard to understand; could you please give a more detailed analysis?

Author Response

Thanks to all editors and reviewers for their precious work in dealing with our manuscript. We answered point by point and provided the manuscript modification in red. Furthermore, in the following, we reported all issues raised by reviewers and proposed a response for each of them. As a general comment, we are pleased that reviewers appreciated the topic and work. Modifications following their comments improved the overall quality of the manuscript.

This paper proposes a system based on a proprietary platform named SENSIPLUS, which is a multi-sensor platform especially devoted to performing Electrical Impedance Spectroscopy (EIS) on an wide frequency interval. The system capabilities of such a platform in this work are exploited for water quality assessment. The joint system, composed of the measurement platform and the developed toolchain, is named SENSIPLUS-LM, standing for SENSIPLUS Learning Machine. The introduction of the toolchain empowers the SENSIPLUS platform moving the inference phase of the machine learning algorithm to the edge, limiting the needs of external computing platforms. The software part, i.e., the developed toolchain, is available for free download from GitLab, as reported in the following. 

Overall, this paper is well-written and well-motivated. Although I am not an expert in this area, I can understand the contributions and the importance of the proposed approach. But there are still some issues: 

Issue 1. The introduction can be revised to add more background and highlight the motivations as well as contributions.

Response 1: Thank you for the comment. The introduction has been revised, and motivations, limitations, and possible use cases have been clarified. Furthermore, the novel contributions have been highlighted in the related works section.

Issue 2. Could you please explain more details about Figure 9? Figure A 1 is vague.

Response 2: We added more details and explanations about the cited figures. Thanks for the comment.

Issue 3. The experimental analysis in 5.2 is a bit hard to understand; could you please give a more detailed analysis?

Response 3: Following what has been done for Issue 2, we detailed the experimental analysis and its obtained results in the text. Thanks for the observation.

Round 2

Reviewer 2 Report

I have no objection to publishing the manuscript in its current form.

Author Response

Thanks to all editors and reviewers for their precious work in dealing with our manuscript. 

Reviewer 3 Report

This paper proposes a system based on a proprietary platform named SENSIPLUS, which is devoted to performing Electrical Impedance Spectroscopy (EIS) on a wide frequency interval. After revision, some of my concerns have been addressed.

Author Response

Thanks to all editors and reviewers for their precious work in dealing with our manuscript.