Efficient Meta-Learning through Task-Specific Pseudo Labelling

Round 1

Reviewer 1 Report

The submitted research addresses an interesting topic in AI, and that is why I did not hesitate to accept the review of the paper.

In general, meta-learning algorithms aim to learn from past experiences in order to improve their performance on new tasks. They might use techniques like transfer learning or few-shot learning to quickly adapt to new situations. This work introduces a method for transductive meta-learning called task-adaptive pseudo labeling, which addresses the problem of sample bias in adaptation through label propagation. The proposed scenario is based on the use of both labeled support sets and unlabeled query sets in the adaptation process. The main contribution of this work is to apply task-adaptation to pseudo-labeling for transductive meta-learning. This approach assigns pseudo labels on unlabeled query sets in an adaptive way to the target task.

Overall, the submitted work is of scientific interest and value, and the outcomes are useful and interesting for the community. However, my main concern is with the language of the paper, which needs thorough correction. There are numerous typos, grammatical errors, and unclear contexts throughout the paper, so I suggest a comprehensive revision. Nonetheless, the innovation level, paper structure, and methodology are acceptable. Here are my specific comments:

-         Please proofread the manuscript and correct any typos, grammatical errors, and unclear sentences.

-         The related work section is too short and not comprehensive, it may be helpful to expand it to provide a more thorough overview of the existing research in the field.

-         There is a typo in Figure 1. Adatation should be changed into ‘Adaptation’.

-         In the caption of Figure 1, please change doesn’t into does not. I would, however, suggest rewriting the whole sentence.

-         While using acronyms or abbreviations can be helpful to save space or improve readability, using "SOTA" instead of "state-of-the-art" does not offer significant benefits, especially in the Abstract.

-         In the last sentence of the Abstract, I believe you meant 1-shot and/or 5-shot, rather than 1-shot 5-shot?

-         In the caption of Figure 2, I would use ‘represent’ rather than ‘denote’.

-         Please avoid beginning your sentences with words such as ‘So’ (Line 76), or ‘Also’ (Line 183), as they may detract from the clarity and professionalism of your writing.

Please proofread the manuscript and correct any typos, grammatical errors, and unclear sentences.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Please check as following comments.

1.The research results can be detailed in the abstract with more sentences to help readers in understanding key contributions quickly.

2.There is no quantitative data in abstract section. Please provide quantitative data to assist readers in understanding main contributions quickly.

3.Please provide an overall framework to assist readers in understanding this manuscript easily and quickly.

4.There is no more detail description in Algorithm 1. What is the logic decision for “2: while not done do” ? Please state this algorithm for more details because of this algorithm is an important process in this research. 

5.Please state experimental configures for more details (such as hardware, software, application platform, dataset…).

6.There is no discussion on the cost effectiveness of the proposed method and other models. What is the computational complexity? What is the runtime? Please include such discussions.

7.Please provide more explanations of the conclusion from corresponding studies. It is very important to show key value of this research.

8.How do the results support the conclusion? How does the conclusion align with the aim?

9.A total of 7 pages of content is not enough to detailed description this topic. Please expend contents of this manuscript.

10.Please state what is the relevant to electronics for this manuscript.

Moderate editing of English language

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

This manuscript proposes a method to generate pseudo-labels for unlabeled query sets in support sets through label propagation to improve the performance of meta-learning. This work is positive for the meta-learning field, however, there are also the following issues to note in this manuscript:

1.       The abstract should identify the research gap and propose the academic contribution of the paper, showing numerical results. The number of words should be controlled to about 250.

2.       Some of the recent important literature needs to be added to the introduction of the relevant work, and some of the missing literature is listed below: PeerJ Computer Science, 7, e613; Applied Sciences, 12(8), 4059; PeerJ Computer Science, 8, e908; IEEE Access, 9, 91476-91486; ICVS 2019, Thessaloniki, Greece, September 23–25, 2019, Proceedings (pp. 343-352).

3.       It is recommended to add a flowchart of the model to make it easier for readers to understand.

4.       The structure of the paper is confusing, and it is recommended to organize the manuscript according to the structure of the introduction, suggested methods, experiments, results, discussions, and conclusions. And each part should be fully described and discussed.

5.       Table 1 should appear as a result, but is included in the suggested methodology.

6.       If there is only 2.1 under Chapter 2, is it necessary to set up subchapters?

7.       The part of discussion needs to be added, clarifying the main academic contributions of the manuscript, and comparing the differences with existing research results, without shying away from the shortcomings of the manuscript's methodology and possible paths for improvement.

8.       Conclusion is too limited. The authors may give the details of their paper's novelty with short descriptions. Please clearly highlight how the work advances the field from the present state of knowledge and the authors should provide a clear justification for the work. The impact or advancement of the work can also appear in the conclusion.

Moderate editing of English language

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have made satisfactory attempts to address most of my minor comments. However, I still have a major concern regarding the related work section, which remains short and limited in scope. I strongly recommend conducting a more comprehensive review of related works. This review should not only involve citing them briefly, but also explaining the methodologies employed in those works. It is crucial to discuss the strengths and weaknesses, similarities and differences between those works and your proposed method, as well as any other relevant content. The reader should be provided with an understanding of the existing research in the field and the bottlenecks that motivated the development of your own solution. Furthermore, it is important to explain how your proposed method can effectively address these challenges. 

There are still several unclear sentences throughout the manuscript that need to be clarified and corrected.

Author Response

I enclose the revised version according to the reviewers' comments. 

Reviewer 2 Report

Thanks for your reply.

 Moderate editing of English language required

Author Response

I enclose the revised version according to the reviewers' comments. 

Reviewer 3 Report

The authors have addressed all concerns of the reviewers. The current paper can be accepted for publication.

Author Response

I enclose the revised version according to the reviewers' comments.