46 lines
3.7 KiB
Markdown
46 lines
3.7 KiB
Markdown
full-title:: AI for Low-Code for [[AI]]
|
||
type:: [[REVIEWS]]
|
||
tags:: [[Lowcode]] [[AI]]
|
||
file:: 
|
||
venue:: [[ICSE]]
|
||
year:: 2023
|
||
status:: [[DONE]]
|
||
|
||
- ### Paper summary
|
||
- The paper presents LowCoder, a low-code environment to support the development of AI pipelines. Users are provided with an AI-powered natural language interface and a visual block-based environment to create AI pipelines without the need to write code. A user study has been executed to evaluate LowCoder, and the results are encouraging; the tool helps discover operators to be employed in the pipeline under development, improve it and even gives guidance to novices with less clarity on what to do.
|
||
- ### Strengths
|
||
- Interesting and relevant paper
|
||
- Well-structured and written paper
|
||
- Encouraging experiments
|
||
- ### Weaknesses
|
||
- Too strong emphasis on the concept of "citizen developer"
|
||
- Lack of comparison with existing baselines
|
||
- ### Comments for the authors
|
||
- [Significance and Soundness] The paper is about an important problem. The proposed approach is presented clearly and convincingly. Also, the user study is convincing and fairly presents both the strengths and the limitations of LowCoder.
|
||
- My main concern about the paper is that it gives too much emphasis on the concept of "citizen developers". As the authors have mentioned in the paper, "citizen developers" are people that do not have strong IT expertise and that still need to develop some IT systems. In the context considered by the authors, I doubt inexpert users are in the position of developing AI pipelines. I would not consider novices in AI pipelines as citizen developers because they need some background in AI/ML to conceive the pipelines of interest.
|
||
- [Novelty] Another important concern is the lack of discussion and comparison with existing baselines. In particular, the authors should discuss the proposed approach with respect to [AutoML | Home](https://www.automl.org/) and possibly make a comparison with some existing implementations. Even big players like Google [Cloud AutoML Custom Machine Learning Models | Google Cloud](https://cloud.google.com/automl?hl=en) provides users with a dedicated environment to simplify the development of ML models.
|
||
- [Verifiability and Transparency] The authors make available a replication package that includes the code implementing LowCoder, further than all the datasets and scripts of the presented evaluation, and some tutorial slides.
|
||
- [Presentation] Overall, the quality of the writing is good and meets the presentation standards of ICSE. One minor issue I found is that the paper makes too many reminders to the appendix in the replication package making the paper not always self-contained.
|
||
- ### Question for authors
|
||
- Q1: How would you locate the proposed approach within the AutoML movement, and what are the strengths and limitations of Lowcoder with respect to existing AutoML implementations?
|
||
- ### NOTES
|
||
- ((63629241-cab5-4ad3-b3bf-f84db2144e87))
|
||
- I would smooth a bit the "citizen developer" aspect
|
||
- Especially for the hyper-parameter configuration
|
||
- ((63629949-0e6c-4881-b329-df61e0399d17))
|
||
- ((6362d48b-528b-4b11-9553-ad2b24829f57))
|
||
- This needs to be further understood
|
||
- ((636435bc-49c5-4fe5-b420-49598f00fb93))
|
||
- too many reminders to the supplementary material
|
||
- ((63666544-d8bf-45ec-ad10-7b06d50d8ac2))
|
||
- why not a cross-validation?
|
||
- ((63666896-e137-411e-9645-854a1139dba2))
|
||
- NOT CLEAR SO FAR
|
||
- ((636673a0-4717-4898-b049-1c7ce2b2aad3))
|
||
- How many???
|
||
- ((636675f4-f71b-47eb-b382-c362848f5384))
|
||
- Too many references to the supplementary material. Without reading it, it is difficult understand what's going on here.
|
||
- ((636677ff-5f77-4431-9dfe-66d3478f430a))
|
||
- Two versions?
|
||
-
|
||
- |