All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
Dear Author,
Your paper has been revised. It has been accepted for publication in PeerJ Computer Science. Thank you for your fine contribution.
[# PeerJ Staff Note - this decision was reviewed and approved by Xiangjie Kong, a PeerJ Section Editor covering this Section #]
-
-
-
**PeerJ Staff Note:** Please ensure that all review and editorial comments are addressed in a response letter and that any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.
**Language Note:** The review process has identified that the English language must be improved. PeerJ can provide language editing services - please contact us at [email protected] for pricing (be sure to provide your manuscript number and title). Alternatively, you should make your own arrangements to improve the language quality and provide details in your response letter. – PeerJ Staff
1. The introduction is more of a repeat of the Abstract. Rewrite both sections to make them distinct in terms of coverage of CapsNets.
2. There are grammatically unclear sentences. Eg, This paper ... novel of two advanced methods...
3. It is best to define abbreviations the first time they are used. They can be used later as such.
The Related Work section should be organized thematically. Avoid jumping between CapsNet applications without providing a critical assessment of their limitations.
1. The "enhanced squash function" and "conditional PReLU" have insufficient derivation and justification. Why use beta 0.5? How does CPReLU improve routing?
2. Visualizations for Fire-CapsNet's residual connections should be provided and compared to other models with residual connections. Is there any difference in terms of structure and performance?
3. Please provide algorithms to demonstrate the workings of the newly proposed methods.
1. Adversarial robustness claims must be substantiated by simulating an attack.
2. BM dataset's imbalance is addressed with SMOTE and not the model. If this is the case, then it must be made clear such that the impression is not created that the model can address the imbalance by itself.
3. Carry out ablation tests to help narrow down to the individual portions of the proposed model that contribute most to its improved performance. Also, use the ablation study to show the level to which the enhanced squash improves performance and compare with enhanced squash + baseline CapsNet performance. Report confusion matrices for the datasets to show the per-class performances.
4. Avoid general statements such as "Fire-CapsNet outperforms all models..." (see Conclusion) without carrying out any statistical significance tests, eg, paired t-tests.
It would be good to use the Lipschitz continuity or gradient stability tests to provide theoretical grounding to the enhanced squash function
i. Clear and unambiguous, professional English used throughout: Partially Yes, the manuscript demonstrates generally professional English, but it contains repeated grammatical inconsistencies, awkward phrasing, and overly long sentences
Recommendation: Professional language editing is advised to polish sentence structure and ensure clarity across the paper
ii. Literature references, sufficient field background/context provided: Yes, OK
iii. Professional article structure, figures, tables. Raw data shared: Mostly Yes
iv. Self-contained with relevant results to hypotheses: Yes, the paper introduces two methods, which are aimed at enhancing feature extraction, improving squash functions, modifying dynamic routing, and others.
v. Formal results should include clear definitions of all terms and theorems, and detailed proofs: For completeness, more formal derivations of modified activation functions (e.g., enhanced squash, CPReLU) would strengthen the rigor.
i. Original primary research within Aims and Scope of the journal: Yes,
• The work proposes two novel enhancements to Capsule Networks (CapsNet)—Pretrained-CapsNet and Fire-CapsNet—applied to large-scale and imbalanced datasets.
• The study falls well within the scope of PeerJ Computer Science, specifically in the "Machine Learning" and "Artificial Intelligence" domains.
ii. Research question well defined, relevant & meaningful. It is stated how research fills an identified knowledge gap: Yes,
The gap is outlined with supporting literature: (i) CNNs' limitations (loss of information in pooling, lack of rotation invariance), (ii) CapsNet limitations (instability in routing, poor scalability, sensitivity to imbalanced data)
iii. Rigorous investigation performed to a high technical & ethical standard: Yes, the Authors could strengthen the ethical standards section by including a formal ethics approval code and license/terms of use for BM.
iv. Methods described with sufficient detail & information to replicate: Yes, the GitHub repo should include training logs, environment details (e.g., Python/TensorFlow versions), and ablation studies to fully support replication.
It is unclear whether random seeds were fixed or whether cross-validation was performed.
i. Impact and novelty not assessed. Meaningful replication encouraged where rationale & benefit to literature is clearly stated: Yes,
(i) The authors emphasize the weaknesses of standard CapsNet, such as performance degradation with unbalanced datasets and computational inefficiency.
(ii) They justify replication across four datasets, including real-world (BM) and benchmark datasets (MNIST, Fashion-MNIST, CIFAR-10).
ii. All underlying data have been provided; they are robust, statistically sound, & controlled: Yes, there is no mention of statistical significance testing.
iii. Conclusions are well stated, linked to original research question & limited to supporting results: Yes, the conclusion is not overstated—it acknowledges that while Fire-CapsNet consistently outperforms, even pretrained CapsNet models (GN-CapsNet, VGG-CapsNet) achieve significant gains.
Q.1.The abstract section is inadequately written and fails to clearly convey the study's key content areas and primary outcomes. What revisions to the abstract could the authors make to emphasize the key results and their significance in relation to the study's goals? What sets this research apart? What is the novelty of this research?
Q.2. The abstract's density could be alleviated by clearer distinctions between the two proposed methods and their respective contributions. The abstract is dense and could benefit from clearer separation of the two proposed methods and their individual contributions.
Q.3. The author modified the dynamic routing algorithm in Capsule Networks by incorporating a conditional PReLU activation function into the routing process. How does this differ from traditional dynamic routing algorithm network performance?
Q.4. Discuss the benefits provided by conditional PReLU in comparison to the traditional method.
Q.5. The authors introduce a new Fire-CapsNet architecture as the second method in their proposal. Q.6. What distinguishes the Fire-CapsNet from the conventional Capsule Network structure? Please provide a detailed explanation of the architectural changes.
Q.7. Explain how these modifications improve upon the performance capabilities of standard CapsNet.
Q.8. The authors employ a custom Swish activation function within the Fire-CapsNet architecture to overcome specific limitations of Capsule Networks. Please provide details.
Q.9. Most of the equations in the manuscript are not formatted correctly according to standard conventions.
Q.10. The unclear or difficult-to-distinguish suffixes and subscripts in the equations may make it hard for readers to comprehend the mathematical formulations and the suggested methods.
Q.11. The term ∥sj∥ appears in the denominator of Equation (4). When sj equals zero, the norm of sj becomes zero, resulting in an undefined expression due to division by zero. The manuscript does not discuss how this situation is managed.
Q.12. Throughout the manuscript, many sentences are poorly phrased or contain grammatical errors. Thorough proofreading is essential.
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.