Next-Generation Quantum Neural Networks: Enhancing Efficiency, Security, and Privacy

Alberto Marchisio; Mohamed Bennai; Muhammad Kashif; Muhammad Shafique; Nouhaila Innan

REVIEW 3 major objections 5 minor 31 references

Next-Generation Quantum Neural Networks: Enhancing Efficiency, Security, and Privacy

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper argues that combining existing training, robustness, and privacy techniques yields a reliable and secure QNN pipeline.

desk verdict A clean synthesis of the authors' own prior QNN results, but the claimed new integrated pipeline is untested, so as a research contribution it does not stand. read the letter →

arxiv 2507.20537 v1 pith:UAYY5G6N submitted 2025-07-28 quant-ph

Nouhaila Innan , Muhammad Kashif , Alberto Marchisio , Mohamed Bennai , Muhammad Shafique This is my paper · ORCID

classification quant-ph

keywords quantumneuralnetworksbarrenplateausparameterinitializationresidualconnectionscircuitcuttingadversarialrobustnessfederatedlearningfullyhomomorphicencryption

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper seeks to establish that the main obstacles to practical quantum neural networks—untrainable landscapes, hardware noise, adversarial inputs, and data privacy—can be addressed together by combining techniques that were previously studied separately. It argues that careful parameter initialization and residual connections curb barren plateaus; circuit cutting lets large circuits run on few qubits; noise-aware design and observable selection improve resilience; adversarial testing guides robust architecture choices; and quantum federated learning with fully homomorphic encryption preserves privacy during distributed training. A sympathetic reader would care because each obstacle currently blocks practical QNN deployment, and the paper's recipe is composed of already-published pieces that could be acted on immediately if the integration holds. The paper is a synthesis and perspective rather than a single new experiment, so the combined pipeline is asserted rather than measured.

What carries the argument

The central object is the integrated cross-layer pipeline, illustrated as an overview in the paper's Fig. 1. It is composed of five mechanisms: parameter initialization (Xavier or narrower uniform ranges) and residual connections to preserve gradient flow and avoid barren plateaus; quantum circuit cutting to decompose large circuits into smaller subcircuits that fit limited-qubit devices; noise-aware design informed by studies of phase-flip, bit-flip, depolarizing, and amplitude-damping channels and by selecting custom Hermitian observables; preemptive adversarial testing with attacks like FGSM and PGD to identify robust ansatze; and quantum federated learning combined with fully homomorphic encryption for privacy-preserving aggregation. Each mechanism is assigned one failure mode of QNN deployment, and the paper's claim is that the mechanisms interoperate as a single pipeline.

What would settle it

Take one benchmark, such as MNIST classification with a 6-qubit QNN under depolarizing noise and an FGSM attack, and compare four configurations: the full pipeline, the pipeline minus initialization, the pipeline minus adversarial testing, and the pipeline without the privacy layer. The claim that the methods form a robust and secure development pipeline is falsified if removing a component does not measurably degrade accuracy or robustness, or if the full pipeline performs no better than its best single component.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that an end-to-end, cross-layer methodology can deliver reliable and secure QNNs in the NISQ era. The methodology integrates efficient parameter initialization, residual quantum circuit connections, systematic quantum architecture exploration, quantum circuit cutting, noise-aware design with careful observable selection, preemptive adversarial robustness testing, and quantum federated learning augmented with fully homomorphic encryption. The authors state that 'Together, these methods form a robust and secure QNN development pipeline,' and they support this by citing separate studies for each layer. The paper establishes a framework or roadmap rather than presenting a unified experimental validation of the combined system.

Load-bearing premise

The load-bearing premise is that each technique, validated separately in earlier studies, still works and fits together when combined in one pipeline; the paper asserts this integration but reports no experiment on the combined system.

Editorial extensions

If this is right

QNN developers can adopt a ready checklist: initialize parameters carefully, add residual connections, cut circuits when qubits run short, choose observables and ansatze with noise and adversarial testing in mind, and train distributively with encryption.
Barren-plateau mitigation and circuit cutting together should let deeper or larger QNNs train on NISQ hardware than is currently routine.
Adversarial robustness can be engineered at the architecture level rather than patched after training; the cited studies report up to 60% greater robustness on MNIST and Fashion-MNIST at low perturbation levels.
Privacy-sensitive sectors can train on distributed quantum nodes with encrypted updates while retaining over 70% test accuracy across the datasets reported in the paper.
The methodology provides a template for translating QNN research into applications in finance, healthcare, and intelligent transportation, where performance, privacy, and resilience are all required at once.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The paper does not report an experiment on the combined pipeline; a natural next step its own logic implies is a component-wise ablation study to see whether the benefits add, overlap, or cancel.
If the integration works as claimed, the practical payoff may be less a new algorithm than a standardized engineering recipe for QNNs; this is an inference, since the paper only asserts the recipe.
The reported FHE runtimes (thousands of seconds per dataset) suggest that cost, not accuracy, may be the binding constraint for privacy-preserving quantum federated learning; this is an inference from the paper's Table I.
The same synthesis could plausibly extend beyond QNNs to other variational quantum algorithms, but the paper does not claim this.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Dear colleague,

Short take: this is a survey of the authors' own prior work dressed as a new integrated methodology. The integration claim is the whole paper, and it is untested.

The paper's actual content is a readable, well-organized synthesis of a half-dozen results from the same group: parameter initialization, residual connections, circuit cutting, noise-aware design, adversarial robustness, and QFL+FHE. Each component is described accurately and credited to its source. If you want to see how these pieces fit together conceptually, this is a fine place to look. For a reader unfamiliar with the group's earlier papers, the figures and Table I conveniently summarize a lot of work.

The problem is that the paper presents itself as a 'framework' and makes the load-bearing claim in Section II-C that combining these methods 'form a robust and secure QNN development pipeline.' There is no experiment, derivation, or analysis in which all components are applied together. The individual results come from different settings: Xavier initialization on hybrid circuits, cutting on a 6-qubit circuit, adversarial metrics on MNIST/Fashion-MNIST, FHE on CIFAR-10/DNA/MRI/PCOS/RAVDESS. The compatibility of these techniques in a single pipeline is asserted, not tested. The stress-test note points out concrete risks: circuit cutting changes the topology and introduces mid-circuit measurements, which could break adversarial-robustness results; FHE encrypts model updates, potentially preventing the gradient/observable access needed for preemptive adversarial testing; and the custom observable from the noise-resilience study may not match what the adversarial-defense work assumes. These are real, unaddressed concerns.

The citation pattern is worth noting: the paper is built almost entirely on the authors' own prior publications, and the displayed results are reproduced from those papers. That is not a fatal flaw—own-work aggregation can be useful—but here the 'new' contribution reduces to the integration narrative, and that narrative is never validated.

Who is this for? A reader who wants a compact summary of this group's recent QNN work might find it useful. It is not a research contribution, and it should not be treated as one.

For peer review: I would not send this to referees as a research paper. It needs to be reframed as a survey or perspective, with the integration claim downgraded accordingly. If the authors are willing to do that, it could be a reasonable review article.

Referee Report

3 major / 5 minor

Summary. The paper proposes an integrated framework for developing Quantum Neural Networks (QNNs) in the NISQ era, combining parameter initialization, residual connections, circuit cutting, noise-aware design, adversarial robustness, and quantum federated learning with fully homomorphic encryption. It presents a high-level methodology diagram and summarizes results from several of the authors' prior publications, including figures and a table of reported accuracy and runtime numbers. The stated central claim, made at the end of Section II-C, is that these methods 'form a robust and secure QNN development pipeline.'

Significance. If the proposed integrated pipeline were actually validated, it would offer a practically useful recipe for building trainable, noise-resilient, and privacy-preserving QNNs. The paper is clearly organized and the figures faithfully illustrate results from the authors' previous work, which addresses relevant and timely problems. However, the manuscript as submitted is a position/perspective compilation rather than a technical contribution: it contains no new experiments, no derivations, no pseudocode, and no analysis of how the individual components interact. The central integration claim is asserted, not demonstrated, and the only quantitative evidence is reused from prior same-group publications.

major comments (3)

[Section II-C] The central claim that 'Together, these methods form a robust and secure QNN development pipeline' is asserted without any demonstration. The components are validated separately in different prior studies: Xavier/residual initialization in [8,13], circuit cutting on a 6-qubit circuit in [15], noise robustness on QuanNN/QCNN in [16,17], adversarial metrics on MNIST/Fashion-MNIST in [18-21], QFL on IBM backends in [22], and FHE-QFL on CIFAR-10/DNA/MRI/PCOS/RAVDESS in [23]. No experiment or mathematical analysis in this manuscript tests whether these techniques remain effective and mutually compatible when applied together. The paper also does not address concrete compatibility risks, such as circuit cutting altering the circuit topology in a way that may invalidate adversarial-robustness results obtained on the intact circuit, or FHE-encrypted updates preventing the gradient or observable access needed by the circuit-metric adversarial design in [21]. This is load-bearing because it is the only connection between the listed ingredients and the paper's stated contribution.
[Section II (Methodology)] The proposed 'framework' is never specified precisely. There is no pseudocode, no algorithm outline, no equation defining the cost function, observables, encryption scheme, or aggregation rule, and no list of hyperparameters or compatibility conditions. As a result, the pipeline is not reproducible or falsifiable; a reader cannot determine, for example, how the noise-aware observable selection of [17] is reconciled with the adversarial circuit metrics of [20,21] or with the FHE-encrypted updates of [23].
[Section II and Figs. 3-7, Table I] All displayed quantitative results are reproduced from the authors' earlier papers ([8], [15], [16], [18], [22], [23]) with no new experiments, ablations, or independent statistical analysis. The statement in Section IV that 'We demonstrated how approaches ... can significantly improve model performance' is therefore not supported by anything demonstrated in this manuscript; it is a summary of prior results. Since the only evidence for the central integration claim is the transfer of these isolated results to the integrated setting, this is a load-bearing gap, not a mere presentation issue.

minor comments (5)

[Fig. 3 caption] The caption sentence 'Random has the performance ... Uniform performancethe best training and Normal have worst training' is garbled and must be rewritten to state clearly which initialization technique performed best and which performed worst.
[Fig. 2 caption] The figure is labeled 'Demonstration of Barren Plateaus' but the caption does not state the model, dataset, number of layers, or what is plotted on the axes; without this information the figure is uninterpretable.
[Fig. 6 caption] The caption names FGSM and PGD, but the figure contains panels (a), (b), and (c) including an unexplained 'Stabilization' panel; the relationship between the panels and the two attack types should be clarified.
[Table I] The dataset name 'RAVDESS' appears as 'RA VDESS' due to a line break, and the runtime values (e.g., 9747.32±2.23) are reported without stating the number of runs used to compute the mean and standard deviation.
[Section III (Applications)] The application paragraph quotes very high accuracies (e.g., 'above 97%', '98%', '95%') without describing experimental conditions, baseline comparisons, or error bars, so a reader cannot assess the reliability of these claims.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is a synthesis of previously published, externally testable results; the central pipeline claim is asserted rather than derived, but it is not equivalent to its inputs by construction.

full rationale

The paper is an integration/survey: it assembles previously published techniques (Xavier initialization [8], residual connections [13,14], circuit cutting [15], noise-robustness analyses [16,17], adversarial robustness [18-21], QFL [22], and QFL-FHE [23]) into a proposed pipeline. I checked each load-bearing claim for reduction-to-input. There are no equations, no fitted parameters, and no new empirical predictions in this manuscript; every quantitative claim, including Table I, is explicitly attributed to prior work. The final assertion in Section II-C ('Together, these methods form a robust and secure QNN development pipeline.') is not entailed by the individual cited results and is not a restatement of any definition; it is an untested composition claim. The close relationship between the cited works and the present authors is a self-citation concern, but per the review rules self-citation is not circular when the underlying results are published, reproducible, and externally falsifiable, as these are. The absence of an integrated experiment is a real evidential gap but belongs to correctness/support risk, not circularity. Therefore no circular step can be exhibited, and the honest finding is score 0.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

No new parameters or entities are introduced. The paper relies entirely on prior results, predominantly self-cited, for each claimed benefit.

assumptions (5)

domain assumption Xavier parameter initialization mitigates barren plateaus in parameterized quantum circuits
Invoked in Section II-A with citation [8]; the claim is accepted from the authors' prior work and not re-derived here.
domain assumption Residual connections preserve gradient flow and enable deeper QNNs
Invoked in Section II-A with citations [13], [14]; treated as established without new analysis.
domain assumption Quantum circuit cutting preserves model accuracy while reducing qubit requirements
Invoked in Section II-A with citation [15]; the figure is from [15].
ad hoc to paper The individual techniques remain effective when composed into a single integrated pipeline
The paper assumes without test that the benefits of each component add up; no integrated experiment is shown.
ad hoc to paper Adversarial robustness and FHE results from prior isolated studies transfer to the proposed QNN designs
Sections II-C.1 and II-C.2 cite [18]-[23] and reproduce their figures, but no new combined evaluation is provided.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Next-Generation Quantum Neural Networks: Enhancing Efficiency, Security, and Privacy." pith.science (2026). https://pith.science/paper/UAYY5G6N

@misc{pith2026250720537,
  author       = {Pith},
  title        = {Pith review of: Next-Generation Quantum Neural Networks: Enhancing Efficiency, Security, and Privacy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UAYY5G6N}},
  note         = {Machine review of arXiv:2507.20537}
}

read the original abstract

This paper provides an integrated perspective on addressing key challenges in developing reliable and secure Quantum Neural Networks (QNNs) in the Noisy Intermediate-Scale Quantum (NISQ) era. In this paper, we present an integrated framework that leverages and combines existing approaches to enhance QNN efficiency, security, and privacy. Specifically, established optimization strategies, including efficient parameter initialization, residual quantum circuit connections, and systematic quantum architecture exploration, are integrated to mitigate issues such as barren plateaus and error propagation. Moreover, the methodology incorporates current defensive mechanisms against adversarial attacks. Finally, Quantum Federated Learning (QFL) is adopted within this framework to facilitate privacy-preserving collaborative training across distributed quantum systems. Collectively, this synthesized approach seeks to enhance the robustness and real-world applicability of QNNs, laying the foundation for reliable quantum-enhanced machine learning applications in finance, healthcare, and cybersecurity.

Figures

Figures reproduced from arXiv: 2507.20537 by the authors.

**Figure 2.** Demonstration of Barren Plateaus. Random has the performance Xavier Uniform performance the best training and Normal have worst training [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗

**Figure 3.** Variance decay and training results with different initialization tech [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 7.** Figure 7: Execution of QNNs on real quantum hardware within a federated learn [PITH_FULL_IMAGE:figures/full_fig_p003_7.png]

**Figure 6.** Figure 6: Robustness evaluation of classical and quantum models using the ZZ [PITH_FULL_IMAGE:figures/full_fig_p003_6.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

31 extracted references · 13 canonical work pages

[15]

Cutting is all you need: Execution of large- scale quantum neural networks on limited-qubit devices,

A. Marchisio et al. , “Cutting is all you need: Execution of large- scale quantum neural networks on limited-qubit devices,” arXiv preprint arXiv:2412.04844, 2024

arXiv 2024
[22]

FedQNN: Federated learning using quantum neural networks,

N. Innan et al. , “FedQNN: Federated learning using quantum neural networks,” in IJCNN, 2024

work page 2024
[23]

MQFL-FHE: Multimodal quantum federated learn- ing framework with fully homomorphic encryption,

S. Dutta et al. , “MQFL-FHE: Multimodal quantum federated learn- ing framework with fully homomorphic encryption,” arXiv preprint arXiv:2412.01858, 2024

arXiv 2024
[21]

Designing robust quantum neural networks via optimized circuit metrics,

W. E. Maouaki et al. , “Designing robust quantum neural networks via optimized circuit metrics,” Advanced Quantum Technologies , 2025

work page 2025
[17]

Nrqnn: The role of observable selection in noise-resilient quantum neural networks,

M. Kashif and M. Shafique, “Nrqnn: The role of observable selection in noise-resilient quantum neural networks,” in World Congress in Computer Science, Computer Engineering & Applied Computing , 2024

2024
[8]

Alleviating barren plateaus in parameterized quantum machine learning circuits: Investigating advanced parameter initializa- tion strategies,

M. Kashif et al., “Alleviating barren plateaus in parameterized quantum machine learning circuits: Investigating advanced parameter initializa- tion strategies,” in DATE, 2024

work page 2024
[16]

Noisy hqnns: A comprehensive analysis of noise robustness in hybrid quantum neural networks,

T. Ahmed et al. , “Noisy hqnns: A comprehensive analysis of noise robustness in hybrid quantum neural networks,” in IJCNN, 2025

work page 2025
[18]

Advqunn: A methodology for analyzing the adversarial robustness of quanvolutional neural networks,

W. E. Maouaki et al. , “Advqunn: A methodology for analyzing the adversarial robustness of quanvolutional neural networks,” in QSW, 2024

work page 2024

Show all 31 references

[1]

A survey on quantum machine learning: Current trends, challenges, opportunities, and the road ahead,

K. Zaman et al. , “A survey on quantum machine learning: Current trends, challenges, opportunities, and the road ahead,” arXiv preprint arXiv:2310.10315, 2023

2023 arXiv
[2]

Quantum computing in the nisq era and beyond,

J. Preskill, “Quantum computing in the nisq era and beyond,” Quantum, 2018

2018
[3]

A comparative analysis of hybrid-quantum classical neural networks,

K. Zaman et al. , “A comparative analysis of hybrid-quantum classical neural networks,” in World Congress in Computer Science, Computer Engineering & Applied Computing , 2024

2024
[4]

Computational advantage in hybrid quantum neural networks: Myth or reality?,

M. Kashif, A. Marchisio, and M. Shafique, “Computational advantage in hybrid quantum neural networks: Myth or reality?,” in DAC, 2025

2025
[5]

Design space exploration of hybrid quantum–classical neural networks,

M. Kashif and S. Al-Kuwari, “Design space exploration of hybrid quantum–classical neural networks,” Electronics, 2021

2021
[6]

Barren plateaus in quantum neural network training landscapes,

J. R. McClean et al. , “Barren plateaus in quantum neural network training landscapes,” Nature Communications, vol. 9, nov 2018

2018
[7]

Investigating the effect of noise on the training performance of hybrid quantum neural networks,

M. Kashif et al. , “Investigating the effect of noise on the training performance of hybrid quantum neural networks,” in IJCNN, 2024

2024
[9]

The impact of cost function globality and locality in hybrid quantum neural networks on nisq devices,

M. Kashif and S. Al-Kuwari, “The impact of cost function globality and locality in hybrid quantum neural networks on nisq devices,” Machine Learning: Science and Technology , vol. 4, no. 1, p. 015004, 2023

2023
[10]

Hqnet: Harnessing quantum noise for effec- tive training of quantum neural networks in nisq era,

M. Kashif and M. Shafique, “Hqnet: Harnessing quantum noise for effec- tive training of quantum neural networks in nisq era,” arXiv:2402.08475, 2024

2024 arXiv
[11]

The unified effect of data encoding, ansatz expressibility and entanglement on the trainability of hqnns,

M. Kashif and S. Al-Kuwari, “The unified effect of data encoding, ansatz expressibility and entanglement on the trainability of hqnns,” International Journal of Parallel, Emergent and Distributed Systems , vol. 38, no. 5, pp. 362–400, 2023

2023
[12]

The dilemma of random parameter initialization and barren plateaus in variational quantum algorithms,

M. Kashif and M. Shafique, “The dilemma of random parameter initialization and barren plateaus in variational quantum algorithms,” 2024

2024
[13]

Resqnets: a residual approach for mit- igating barren plateaus in quantum neural networks,

M. Kashif and S. Al-Kuwari, “Resqnets: a residual approach for mit- igating barren plateaus in quantum neural networks,” EPJ Quantum Technology, vol. 11, no. 1, p. 4, 2024

2024
[14]

Resqunns:towards enabling deep learning in quantum convolution neural networks,

M. Kashif and M. Shafique, “Resqunns:towards enabling deep learning in quantum convolution neural networks,” arXiv:2402.09146, 2024

2024 arXiv
[19]

Robqunns: A methodology for robust quanvolu- tional neural networks against adversarial attacks,

W. El Maouaki et al., “Robqunns: A methodology for robust quanvolu- tional neural networks against adversarial attacks,” in ICIPCW, 2024

2024
[20]

Designing robust quantum neural networks: Ex- ploring expressibility, entanglement, and control rotation gate selection for enhanced quantum models,

W. El Maouaki et al. , “Designing robust quantum neural networks: Ex- ploring expressibility, entanglement, and control rotation gate selection for enhanced quantum models,” arXiv preprint arXiv:2411.11870 , 2024

2024 arXiv
[24]

Financial fraud detection: a comparative study of quantum machine learning models,

N. Innan et al. , “Financial fraud detection: a comparative study of quantum machine learning models,” International Journal of Quantum Information, 2024

2024
[25]

Comparative performance analysis of quantum machine learning architectures for credit card fraud detection,

M. El Alami et al. , “Comparative performance analysis of quantum machine learning architectures for credit card fraud detection,” arXiv preprint arXiv:2412.19441, 2024

2024
[26]

QFNN-FFD: Quantum federated neural network for financial fraud detection,

N. Innan et al. , “QFNN-FFD: Quantum federated neural network for financial fraud detection,” arXiv preprint arXiv:2404.02595 , 2024

2024 arXiv
[27]

LEP-QNN: Loan eligibility prediction using quantum neural networks,

N. Innan et al. , “LEP-QNN: Loan eligibility prediction using quantum neural networks,” arXiv preprint arXiv:2412.03158 , 2024

2024
[28]

Quantum state tomography using quantum machine learning,

N. Innan et al. , “Quantum state tomography using quantum machine learning,” Quantum Machine Intelligence , 2024

2024
[29]

Mqml: Multi-omic quantum machine learning based cancer classification, biomarker identification in human lung adenocarcinoma,

M. K. Saggi and S. Kais, “Mqml: Multi-omic quantum machine learning based cancer classification, biomarker identification in human lung adenocarcinoma,” in 2024 IEEE International Conference on Quantum Computing and Engineering (QCE) , vol. 1, pp. 1713–1720, IEEE, 2024

2024
[30]

QNN-VRCS: A quantum neural network for vehicle road cooperation systems,

N. Innan et al. , “QNN-VRCS: A quantum neural network for vehicle road cooperation systems,” IEEE Transactions on Intelligent Transporta- tion Systems , 2025

2025
[31]

QUIET-SR: Quantum image enhancement transformer for single image super-resolution,

S. Dutta et al. , “QUIET-SR: Quantum image enhancement transformer for single image super-resolution,” arXiv preprint arXiv:2503.08759 , 2025

2025

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [15]

Cutting is all you need: Execution of large- scale quantum neural networks on limited-qubit devices,

A. Marchisio et al. , “Cutting is all you need: Execution of large- scale quantum neural networks on limited-qubit devices,” arXiv preprint arXiv:2412.04844, 2024

arXiv 2024

[2] [22]

FedQNN: Federated learning using quantum neural networks,

N. Innan et al. , “FedQNN: Federated learning using quantum neural networks,” in IJCNN, 2024

work page 2024

[3] [23]

MQFL-FHE: Multimodal quantum federated learn- ing framework with fully homomorphic encryption,

S. Dutta et al. , “MQFL-FHE: Multimodal quantum federated learn- ing framework with fully homomorphic encryption,” arXiv preprint arXiv:2412.01858, 2024

arXiv 2024

[4] [21]

Designing robust quantum neural networks via optimized circuit metrics,

W. E. Maouaki et al. , “Designing robust quantum neural networks via optimized circuit metrics,” Advanced Quantum Technologies , 2025

work page 2025

[5] [17]

Nrqnn: The role of observable selection in noise-resilient quantum neural networks,

M. Kashif and M. Shafique, “Nrqnn: The role of observable selection in noise-resilient quantum neural networks,” in World Congress in Computer Science, Computer Engineering & Applied Computing , 2024

2024

[6] [8]

Alleviating barren plateaus in parameterized quantum machine learning circuits: Investigating advanced parameter initializa- tion strategies,

M. Kashif et al., “Alleviating barren plateaus in parameterized quantum machine learning circuits: Investigating advanced parameter initializa- tion strategies,” in DATE, 2024

work page 2024

[7] [16]

Noisy hqnns: A comprehensive analysis of noise robustness in hybrid quantum neural networks,

T. Ahmed et al. , “Noisy hqnns: A comprehensive analysis of noise robustness in hybrid quantum neural networks,” in IJCNN, 2025

work page 2025

[8] [18]

Advqunn: A methodology for analyzing the adversarial robustness of quanvolutional neural networks,

W. E. Maouaki et al. , “Advqunn: A methodology for analyzing the adversarial robustness of quanvolutional neural networks,” in QSW, 2024

work page 2024