Week4 proposal: scientific paper

[CoordinatesNotFound]

Assignment Proposal

Title

• MLOps for Cyber-Physical Production Systems: Challenges and Solutions

Names and KTH ID

• Tianning Liang ([email protected])
• Yinan Hu ([email protected])

Deadline

• Week 4

Category

• Scientific paper

Description

The purpose of this scientific paper presentation is to guide the audience understand the industrial MLOps of for Cyber-Physical Production Systems (CPPS), regarding 3 major challenges and corresponding solutions, as outlined in the paper ["MLOps for Cyber-Physical Production Systems: Challenges and Solutions"](https://ieeexplore-ieee-org.focus.lib.kth.se/document/10636080/keywords#keywords).

We will explain the data-related, model-related and operations-related challenges that are specific to implementing industrial MLOps for CPPS, and discuss about existing solutions. We will also go into the technical details of real applications of the solutions.

Relevance

The topics covered are particularly relevant to MLOps, as they address the MLOps in industrial production setting based on real industrial experience.

[javierron]

@CoordinatesNotFound Thanks for the proposal. It is relevant to this week's topic.

[HexuL]

Feedback

By Hexu Li and Peiyang Zheng

Code of Conduct

We certify that generative AI, incl. ChatGPT has not been used to write this feedback. Using generative AI without permission is considered academic misconduct.

High-level strengths and Weaknesses:

Strengths

• The structure of the presentation is very clear, first briefly introduces MLOps, and then separately introduces and analyzes the three challenges (Data related, Model related, Operations-Related) mentioned in the original article.

• When introducing each challenge, there is an example to help the audience better understand this part.

• Based on the original paper, There also some code snippets have been added to increase technical depth, which provides a practical demonstration of how these tools can be implemented in real-world applications.

• The presentation discusses two more related papers which are not in the bibliography.

Weakness

• Lack of Introduction to Research Approach.
• The demo mentioned three main categories of challenges: data-related, model-related, and operational-related, but the proposed solutions seemed to only touch on common problems without going into the actual difficulty or technical depth of these problems in specific industrial scenarios.
• The discussion seems to rely more on the author's experience or general descriptions than on concrete data supporting the effectiveness of the solution in practice.
• I feel that it could have dug deeper into the technical details of a particular section by citing other papers.
• It would be better if there was more introduction about CPPS.😀
• The example “OPC UA Web Platform” is not highly relevant to the challenges presented. (They removed it from their presentation)

Feedback per section

General

This presentation is great, but since the paper itself does not go into this topic in great depth and the presentation time is limited, it is difficult to discuss the relevant issues in depth.

Introduction to MLOps for CPPS:

The introduction part briefly introduces MLOps and CPPS, but it is a bit too brief, and the original paper has an "Approach" between "Industrial MLOps" and "Challenges and Solutions", transitioning from the introduction of MLOps to the challenges, and mentions that there are many challenges but the paper only discusses three of them. However, the presentation starts directly with the challenges after introducing MLOps. So it feels like some information is missing.

Challenges & Solutions of MLOps for CPPS

The three challenges introduced in the original paper are all mentioned in the presentation, and there are examples and code snippets to help understand. However, in Tianning and Yinan's early presentation, the code explanation is a bit short, and some examples are not very relevant to the challenge. There is no data either, but this is also the weakness of the original paper. If they can find supplementary data in other papers, it will make the presentation more credible and in-depth. We write a paper in additional material, which provides good concrete examples to analyze.

Reflection & Related Work

Reflection & Related Work points out some limitations of the paper and provides some supplements from other articles, which is very good.

Conclusion

The conclusion is very clear and relevant to the topic.

Additional Material

Josu Diaz-de-Arcaya, Ana I.Torre-Bastida, Gorka Zárate, Raúl Miñón, and Aitor Almeida. 2023. A Joint Study of the Challenges, Opportunities, and Roadmap of MLOps and AIOps: A Systematic Survey. ACM Comput. Surv. 56, 4, Article 84 (April 2024), 30 pages. https://doi.org/10.1145/3625289