LaANIL: ANIL with Look-Ahead Meta-Optimization and Data Parallelism

Round 1

Reviewer 1 Report

This paper proposed a Look Ahead ANIL, an advanced version of ANIL, for few-shot image classification. The presented LaANIL used data parallelism and Nesterrov momentum to solve the problem of ANIL.

Although their contributions can be valuable in demonstrating the applicability of LaANIL, I have several concerns and comments as follows.

1. The authors should highlight the structural difference of the proposed LaANIL compared to ANIL in the Abstract, so how each structure of LaANIL can solve the problem of ANIL.

2. In the Introduction, the authors should explain why they use few-shot image classification to prove the effectiveness of the proposed LaANIL.

3. In Section 4, the authors should use a pseudo-code that explains the experimental procedure for better understanding.

4. To prove the versatility LaANIL, the authors should use ResNet 12 or VGG as the fundamental network like other Meta-learning papers [1,2].

[1] Baik, S., Hong, S., & Lee, K. M. (2020). Learning to forget for meta-learning. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 2379-2387).

[2] J. Moon, E. Kim, J. Hwang, E. Hwang. "A Task-Adaptive Parameter Transformation Scheme for Model-Agnostic-Meta-Learning-Based Few-Shot Animal Sound Classification," Applied Sciences, Vol. 14, no. 3, pp. 1025, 2024.01

5. The authors only use ANIL for comparison. Other meta-learning-based methods like ProtoNet and MAML should be used for comparison methods.

6. The authors should thoroughly explain the advantages of using data parallelism and Nesterrov momentum through the ablation study. 

The authors use moderate quality of English.

Author Response

Dear Editor and Dear Reviewer,

We appreciate your valuable time and insightful feedback on our manuscript titled 'LaANIL: Thank you for your contribution. ANIL with Look-Ahead Meta-Optimization and Data Parallelism'.  Your thorough evaluation has contributed immensely to the improvement of our work, and we have carefully reviewed each of your comments and provided detailed responses and revisions as per your suggestions. If you find any further changes required for the manuscript, please let us know.

Thank you for the opportunity to revise our manuscript. We hope that the improvements made will meet the expectations of the reviewers and the editorial board

Best regards

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors proposes an algorithm ‘Look Ahead ANIL’ (LaANIL) to enhance ANIL as overcome deficiencies and increasing its effectiveness in real-world scenarios.

The Methodology is well described and the results are clearly presented. Results of tests of variance, generalization and adaptation on five datasets and comparisons with the results achieved with ANIL prove the effectiveness of the proposed solution.

Some improvements should be made before acceptance for publication. I recommend the authors to state their contribution in the Conclusion not in the Introduction section (row 44-row 63) and the literature gap at the end of the Background section.

Minor editing of English language required

Author Response

Dear Editor and Dear Reviewer,

We appreciate your valuable time and insightful feedback on our manuscript titled 'LaANIL: Thank you for your contribution. ANIL with Look-Ahead Meta-Optimization and Data Parallelism'.  Your thorough evaluation has contributed immensely to the improvement of our work, and we have carefully reviewed each of your comments and provided detailed responses and revisions as per your suggestions. If you find any further changes required for the manuscript, please let us know.

Thank you for the opportunity to revise our manuscript. We hope that the improvements made will meet the expectations of the reviewers and the editorial board.

Best regards

Author Response File: Author Response.pdf

Reviewer 3 Report

The paper introduces LaANIL, an enhancement to the ANIL model, focusing on modifying its architecture and gradient optimization method for improved learning. LaANIL utilizes parallel computing in the data input pipeline to extract informative features and optimize look-ahead gradient computation, resulting in faster adaptation. It employs data parallelism and look-ahead gradient calculations for task-specific adaptation, facilitating rapid generalization and compatibility with Edge devices. Experimental results validate LaANIL's effectiveness, showing improved accuracy and variance reduction on Meta-few-shot learning datasets. The following comments should be addressed to enhance the paper.

1. Abstract: Clearly state the contribution your research makes to the field. What new knowledge or insights does your work provide?

2. Keyword Section: Correct the typo in 'MAML' (Model Agnostic Meat-Learning).

3. Equations: Ensure that all equations are referenced in the text.

4. Datasets: Explain the rationale behind selecting specific datasets such as FC100, CIFAR-FS, Mini-ImageNet, CUBirds-200-2011, and 271 Tiered-ImageNet.

5. Model Configuration: Provide additional details on the fine-tuning procedure used to obtain the Meta-learning rate and adaption learning rate.

6. Additional clarification is needed for Sections 3 (Proposed Model) and 5 (Results and Discussion) to enhance understanding and provide comprehensive insights. For example; model generalization as compare to previous approaches

Author Response

Dear Editor and Dear Reviewer,

We appreciate your valuable time and insightful feedback on our manuscript titled 'LaANIL: Thank you for your contribution. ANIL with Look-Ahead Meta-Optimization and Data Parallelism'.  Your thorough evaluation has contributed immensely to the improvement of our work, and we have carefully reviewed each of your comments and provided detailed responses and revisions as per your suggestions. If you find any further changes required for the manuscript, please let us know.

Thank you for the opportunity to revise our manuscript. We hope that the improvements made will meet the expectations of the reviewers and the editorial board.

Best regards

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The quality of this paper is greatly improved during the revision process. All concerns are solved.

The quality of English seems descent, which can be easily written to authors.