Semi-Supervised Few-Shot Incremental Learning with k-Probabilistic Principal Component Analysis

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Please find the review comments attached. Thanks.

Comments for author File: Comments.pdf

Author Response

Thank you for your comments and suggestions. Our responses are attached for your review.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Line 72: “aims to learn a unified clas-” should be “aims to teach a unified clas-”.

Line 121: “i.i.d” misses a dot after d and the acronym must be explained (independent and identifiable distribution).

Line 122: comma after “Then” or remove “Then”.

Lines 224-226: a reformulation is needed to make the phrase clearer.

Section 2.5. complexity analysis can be improved. This section appears a bit sketchy to me, with only some "numbers" (formulas) presented, without any comments or insights. The final formula may lead to wrong interpretations regarding the growth factor, in absence of authors' insights. I feel the authors should make this section more clear to the readers.

Section 3.1. Datasets: for 2 datasets the size of the images is mentioned, but not for the other 2.

Tables: the header should have the full word “accuracy” instead of “acc.”.

Line 250: shouldn’t is be “schemes” instead of “schedules”?

Lines 310-311: a reformulation is needed.

Author Response

Thank you for your comments and suggestions. Our responses are attached for your review.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

 

In the manuscript titled "Semi-Supervised Few-Shot Incremental Learning with k-Probabilistic PCAs", No.: electronics-3331143, authors r introduces a novel method for Semi-Supervised Few-Shot Class Incremental Learning (SSFSCIL) that exhibits virtually no catastrophic forgetting. It consists of uses a generic feature extractor a classifier. This paper conducts extensive experiments on benchmark datasets, achieved significant improvements compared to state-of-the-art (SOTA) methods. However, I have some concerns outlined below:

1.      Please open-source your code so that other researchers can follow your work.

2.      The authors state that " However, due to the scarcity of costly labeled data in practice, there is an emerging interest in promoting the utility of existing labels." However, in reality, data with noisy labels can be obtained at a lower cost, but such data can significantly degrade the performance of deep neural networks (DNNs). Does the class-incremental learning method proposed in this paper take into account the impact of noisy labels on model performance?

3.      Take a careful look at your cited papers. Make sure it is current, and cites recent work. Please do not cite large groups of papers without individually commenting on them, e.g. "… [1-4] …".

4.      It is suggested that the authors provide a detailed introduction to the experimental environment, including hardware equipment, Torch version, and random number settings, etc..

5.      The experiments in Tables 1 to 3 primarily focus on residual networks (ResNet). Given that current mainstream image encoders, such as CLIP, use Vision Transformer (ViT) architectures, i.e., ViT-B/32, ViT-L/14@336px, it is indeed necessary for the authors to expand the scope of their experiments to include these models.

Overall, the paper has several issues that need to be revised, including formatting, method description, and experimental comparisons.

 

Author Response

Thank you for your comments and suggestions. Our responses are attached for your review.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for the response and revision. Good work.

Reviewer 3 Report

Comments and Suggestions for Authors

All my concerns have been addressed.