Pith. sign in

REVIEW 4 major objections 6 minor 42 references

A hybrid QGAN can learn and store the byte distribution of a post-quantum signature scheme.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 03:54 UTC pith:M7SBHGPE

load-bearing objection A clean but overclaimed QGAN demonstration on a likely-uniform target distribution; the capability conclusion is not supported. the 4 major comments →

arxiv 2607.13722 v1 pith:M7SBHGPE submitted 2026-07-15 quant-ph cs.CRcs.LG

Towards quantum machine learning for assessing the resilience of post-quantum cryptography

Jaros{\l}aw A. Miszczak This is my paper
classification quant-ph cs.CRcs.LG
keywords quantum machine learningquantum generative adversarial networkspost-quantum cryptographySPHINCS+distribution loadingvariational quantum circuitscryptanalysisNISQ devices
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper attempts to show that a hybrid quantum-classical generative adversarial network (QGAN) can learn the probability distribution of the first byte of signatures produced by SPHINCS+, a standardized hash-based post-quantum signature scheme, and load that distribution into the amplitudes of a 16-qubit quantum state. The motivation is that being able to encode real post-quantum data into quantum memory is a necessary first step toward using near-term quantum computers to probe post-quantum protocols for weaknesses. The experiments compare three quantum generator circuit topologies and report that the hardware-efficient ansatz yields the best Kullback-Leibler divergence to the observed distribution, with the real-amplitudes ansatz close behind. A sympathetic reader would take the contribution as evidence that near-term hybrid methods have enough expressibility to reproduce at least one byte-level distribution of a real post-quantum signature stream.

Core claim

The paper's central claim is that a QGAN with a quantum circuit as generator and a classical neural network as discriminator can be trained to approximate the distribution of nibble pairs encoding the first byte of SPHINCS+ signatures, and that the resulting parameterized circuit acts as a quantum memory that stores the empirical distribution. The authors report convergence of the loss functions and quantify the approximation quality with KL divergence; the hardware-efficient ansatz gives the closest match, while the real-amplitudes ansatz performs surprisingly well despite fewer parameters. The authors conclude that these results confirm the generalization capability of QGANs needed to repr

What carries the argument

The central mechanism is the hybrid QGAN: a parameterized quantum circuit (the generator) produces an n-qubit state whose measurement probabilities are meant to match the target empirical distribution, while a classical neural network (the discriminator) tries to separate real samples from generator samples; the two are trained adversarially with binary cross-entropy loss. The target data are the first byte of each SPHINCS+ signature, split into two four-bit nibbles, giving a 2D 16×16 histogram. Three ansatze — N-local, real-amplitudes, and hardware-efficient — serve as generator circuits, and the paper evaluates the learned distribution via the Kullback-Leibler divergence. The trained gener

Load-bearing premise

The paper's central claim depends on the assumption that the empirical distribution of the first byte of SPHINCS+ signatures, sampled from 2^13 signatures of a single plaintext, is non-trivial and representative enough that approximating it demonstrates a capability relevant to cryptanalysis; if those bytes are nearly uniform, the task reduces to trivial state preparation.

What would settle it

Sample a large number of SPHINCS+ signatures and compute the KL divergence (or chi-square statistic) between the empirical first-byte distribution and the uniform distribution on 256 values. If the divergence is within sampling noise of zero — for instance, a chi-square test with p-value above 0.01 — then the first byte carries no distinguishing structure, and the QGAN's success would not demonstrate an ability to reproduce a cryptographically meaningful distribution. Alternatively, run the same QGAN pipeline on genuinely uniform random bytes; if the KL curves are statistically indistinguishab

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If a QGAN can load real post-quantum signature data into a quantum state, then a quantum computer can subsequently apply variational or sampling algorithms to that encoded distribution with only polynomially many gates.
  • The comparison of ansatze suggests that circuit expressibility is not always the decisive factor for learning quality; simple real-amplitude circuits can match or beat more expressive ones, which guides circuit selection for future quantum-machine-learning attacks.
  • The result implies a concrete, implementable first stage in a future pipeline for quantum-assisted cryptanalysis: from raw post-quantum outputs to quantum-resident probability distributions.
  • The authors' success on the benchmark normal mixture shows the method works as a generic distribution-loading tool, strengthening the case that the post-quantum-specific result is not idiosyncratic to the setup.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A direct follow-up test would be to measure the KL divergence between the SPHINCS+ first-byte distribution and the uniform distribution; if the byte values are nearly uniform, the successful loading reflects trivial state preparation, and the paper's stronger claim about reproducing 'complex' post-quantum distributions would need re-benchmarking on a genuinely non-uniform distribution.
  • The approach could be extended beyond bytes to the full signature structure or to other post-quantum families (lattice-based, code-based); the nibble-pair encoding suggests a natural path to higher-dimensional distributions at the cost of more qubits and parameters.
  • If the loaded distribution is faithful, one could use the quantum state as a prior in an amplitude-amplification-style search for high-probability signatures, potentially changing the query complexity of offline forgery attempts — though this is speculative and not claimed by the paper.
  • Because the paper's experiments are simulated, a near-term hardware run of the same circuits would test whether noise at the current NISQ level preserves the learned distribution; the authors themselves list noise as a limitation.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes a hybrid quantum-classical QGAN framework for learning and loading the probability distribution of the first byte of SPHINCS+ signatures into a quantum state. The generator is implemented with three Qiskit ansatze (N-local, real-amplitudes, hardware-efficient) on a 16-qubit register, trained with a classical discriminator and ADAM. Results are reported as loss curves and KL divergences versus epoch for each ansatz. The abstract and conclusions claim that near-term hybrid quantum-classical methods 'possess capabilities required' for reproducing 'complex probability distribution generated by post-quantum cryptography schemes,' and that this is a first step toward cryptanalytic workflows.

Significance. If fully supported, the paper would provide a modest but concrete demonstration of loading a PQC-related distribution into a quantum state with a near-term-compatible circuit, potentially motivating further work on QML-assisted cryptanalysis. The paper clearly describes the circuit architectures, data encoding, and training setup, and it makes an honest comparison of three ansatze. However, the central claim is not established by the presented evidence: there is no classical baseline, no statistical characterization of the target distribution, no train/test separation, and the reported KL values are high. The significance of the work therefore remains prospective rather than demonstrated.

major comments (4)
  1. [§3.2, Fig. 4(a), Abstract] The target distribution is the first byte of SPHINCS+ signatures. For a hash-based signature scheme, the leading bytes are expected to be statistically uniform (the randomizer is chosen uniformly). The paper provides no uniformity test, no justification that this byte is representative of a 'complex probability distribution,' and no comparison against a uniform baseline. With only 2^13 samples, the histogram in Fig. 4(a) may be sampling noise around 1/256. If the distribution is near-uniform, the learning task is trivial state preparation and the abstract's central claim is vacuous. The authors should add a chi-square or KL-vs-uniform test, compare against a fixed uniform state and a classical histogram, and ideally use a genuinely non-uniform, multi-byte or multi-plaintext distribution.
  2. [§5 and §4 (evaluation methodology)] The conclusion that QGANs 'possess generalization capabilities required to reproduce complex probability distribution' is unsupported because the model is trained and evaluated on the same empirical distribution. There is no train/test split, no held-out sample, and no cross-validation. The reported KL divergence measures similarity to the training set, not predictive generalization. A valid generalization claim requires evaluating the trained generator on data not seen during training, e.g., signatures generated with different plaintexts or from a held-out portion of the signature stream.
  3. [§4, Figs. 5–7] All experimental curves appear to be single runs with no error bars, no random seed reporting, and no statistical significance testing. Moreover, no classical baseline is provided. The KL values of 0.4–1.4 (Fig. 6) are high for a 256-bin distribution and cannot be interpreted without knowing the KL of a trivial uniform distribution or a classical histogram (which would be near zero). The authors should report mean ± standard deviation over multiple independent trainings and compare against at least (i) the empirical-vs-uniform KL and (ii) a classical GAN or histogram baseline.
  4. [§4, parameter counts] There is an internal inconsistency in the number of N-local ansatz parameters: the experimental setup states 98 parameters, but the discussion in the following paragraph says 96 trainable parameters. This discrepancy should be corrected, as it affects the comparison of ansatz expressibility.
minor comments (6)
  1. [Abstract] Typo: 'posses' should be 'possess'.
  2. [Throughout] The word 'ansatz' is repeatedly misspelled as 'anzatz' (e.g., Figs. 2, 5 and text).
  3. [Fig. 1] Typo: 'meeasurement' should be 'measurement'.
  4. [Acknowledgments] Typo: 'constrictive remarks' should be 'constructive remarks'.
  5. [Eq. (2)] The binary cross-entropy loss is written with x_j and y_j, but the roles of predictions and labels are ambiguous; the standard convention is y log(p) + (1-y) log(1-p), where p is the predicted probability. Please clarify the notation.
  6. [§4, benchmark description] The text says 'From the results in Fig. 4(b) one can note...' but Fig. 4(b) shows the raw data, not the learning curves; the intended reference is likely Fig. 7.

Circularity Check

1 steps flagged

Central 'generalization' claim rests on a training-set fit: the DKL metric is evaluated against the same empirical distribution used to define the target state, with no held-out data.

specific steps
  1. fitted input called prediction [Section 5 (Conclusions), with Eq. (1) in Section 3.1 and Eq. (3)/Fig. 6 in Section 4]
    "The presented results confirm that Quantum Generative Adversarial Networks posses generalization capabilities required to reproduce complex probability distribution generated by the post-quantum cryptography schemes."

    The only PQC 'complex probability distribution' is the empirical histogram of the first byte of SPHINCS+ signatures (Section 3.2, Fig. 4a). Equation (1) defines the target state directly from the training-data probabilities p_j, and Eq. (3)/Fig. 6 evaluate DKL(P||Q) with P being the same empirical p_j. There is no held-out split or independent test distribution. Therefore the reported low DKL measures how well the generator fits its own training target; calling this 'generalization capabilities' is a training-set fit renamed as a prediction. The capability conclusion is self-referential rather than evidence of generalization to unseen signature data.

full rationale

No self-citation chain is load-bearing: the paper cites external libraries and prior QGAN work [4,36,41], not the author's own results. The numerical pipeline is otherwise a standard QGAN training loop with a classical discriminator and Qiskit ansatze, and the KL curves are honestly reported as training progress. The circularity is concentrated in the interpretation: Section 3.1 builds the target state from empirical p_j, Section 4 evaluates DKL against the same empirical distribution, and Section 5 converts this into a claim about 'generalization capabilities.' Since no train/test split or out-of-sample evaluation is provided, the central capability claim is supported only by a fit to its own ground truth. A further correctness risk, not itself circularity, is that the chosen target (first byte of SPHINCS+ signatures) may be near-uniform by construction, which would make the 'complex distribution' claim vacuous; but that is an external validity concern, not a derivation-chain circularity.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The ledger is mostly about experimental design choices rather than fitted physical parameters. The two substantive load-bearing assumptions are the representativeness of the first-byte distribution and the transferability of noiseless simulation to hardware.

free parameters (5)
  • Ansatz choices and layer count = N-local, real-amplitudes, hardware-efficient; six layers
    Chosen by hand, no ablation; affects expressibility and performance.
  • Optimizer hyperparameters = lr=0.01, beta1=0.7, beta2=0.999
    Chosen by hand from standard settings; beta1=0.7 deviates from PyTorch default.
  • Number of qubits = 16
    Arbitrary; 16 qubits are used for a 256-bin distribution, introducing significant unused Hilbert space.
  • Training sample size = 2^13 signatures, single plaintext
    Chosen by hand; no cross-validation or multiple plaintexts.
  • Data representation = first byte as two 4-bit words
    Modeling choice limiting target to 256 bins; no justification for first-byte fixation.
axioms (5)
  • standard math Quantum state evolution is unitary and measurement follows the Born rule.
    Standard quantum mechanics used implicitly in the QGAN circuit model.
  • standard math The target distribution can be represented via amplitude encoding |ψ⟩ = Σ√p_j|j⟩.
    Eq. (1) assumes this encoding; it's valid for any probability vector.
  • ad hoc to paper The empirical first-byte distribution from 2^13 SPHINCS+ signatures for a single plaintext is representative of the signature scheme's output.
    No statistical validation that this sample is representative or non-uniform; the entire attack-relevance claim rests on this.
  • domain assumption Trained parameters generalize to unseen data.
    The paper claims 'generalization capabilities' but only evaluates on the training distribution (no train/test split); the model may simply memorize.
  • domain assumption Noiseless simulation results transfer to near-term NISQ hardware.
    Paper itself notes QGANs are noise-sensitive; no hardware experiments are performed.

pith-pipeline@v1.3.0-alltime-deepseek · 9211 in / 14144 out tokens · 136250 ms · 2026-08-02T03:54:43.432804+00:00 · methodology

0 comments
read the original abstract

The potential capabilities of quantum computers motivated the development of cryptographic protocols suitable for securing communication against adversaries with access to large fault-tolerant quantum computers. However, even though current quantum computers are limited in terms of size and precision, they can still be useful for finding loopholes and weaknesses in the post-quantum cryptographic protocols. In this work, we present an attempt to utilize the capabilities of Quantum Generative Adversarial Networks (QGANs), one of the promising architectures used in quantum machine learning, for this purpose. We describe an example application of QGAN architecture for the purpose of loading the probability distribution of the hash-based digital signatures into the memory of a quantum computer. Our results confirm that near-term hybrid quantum-classical methods possess capabilities required for this purpose. The presented approach can be used as a first step in the workflow, enabling the utilization of quantum computing for attacking post-quantum cryptographic primitives.

Figures

Figures reproduced from arXiv: 2607.13722 by Jaros{\l}aw A. Miszczak.

Figure 1
Figure 1. Figure 1: Architecture of the proposed framework based on hybrid quantum-classical QGAN from the Qiskit Machine Learning library [4, 36]. The quantum generator is trained using the feedback from the classical discriminator to generate a quantum state suitable for representing the observed real data. Details of the quantum circuit construction are described in Section 3.3. Details of the classical discriminator are d… view at source ↗
Figure 2
Figure 2. Figure 2 [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 2
Figure 2. Figure 2: Three types of ansatze used in this work as the basic blocks of the quantum generator. The circuits in this figure were built for four-qubit registers. We have (a) six, (b) eight, and (c) sixteen trainable parameters per layer. Using (a) and (c) we do not impose restrictions on the form of generated quantum states. For the case (b) the prepared quantum states will only have real amplitudes [PITH_FULL_IMAG… view at source ↗
Figure 3
Figure 3. Figure 3: Architecture of the artificial neural network used as the discriminator in the presented method. The scheme is based on the implementation described in [4, 36]. 4 Results Let us now demonstrate the results of utilizing the defined framework for the purpose of loading the probability distribution of the signature bytes. For the purpose of numerical experiments presented in this work, we used circuits built … view at source ↗
Figure 4
Figure 4. Figure 4: Sample data from (a) the post-quantum hash-based source and (b) an artificially created mixture of multivariate normal distributions. (a) Probability distribution of pairs of four-bit words generated from the first byte of the SPHINCS+ signatures. A sample obtained using 213 signatures for a single plain text. (b) Probability distribution obtained as a mixture of 16 multivariate normal distributions, with … view at source ↗
Figure 5
Figure 5. Figure 5: Value of the loss function for the generator and the discriminator during the learning process for the PQC data. obtained results one can conclude that the proposed method is able to learn the probability distribution of the signatures in the case of hardware-efficient ansatz and real-amplitudes ansatz. For the sake of completeness, in [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Distance between the learned probability distribution and the observed proba￾bility distribution for the data from SPHINCS+, a post-quantum secure, hash-based digital signature scheme from the NIST PQC suite. also note that real-amplitudes ansatz, despite using only 56 trainable parameters, was able to deliver a better approximation of the real data compared with N-local ansatz (using 96 trainable paramete… view at source ↗
Figure 7
Figure 7. Figure 7: Distance between the learned probability distribution and the observed prob￾ability distribution for the data from the distribution obtained as a mixture of 16 multivariate normal distributions. As the last observation, one can note that for both learning the PQC data and the normal distribution, the real-amplitudes ansatz provides a solid method for learning the distribution. As, in this case, the quantum… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

42 extracted references · 5 canonical work pages

  1. [1]

    Release Notes (Apr 2025), https://www.openssh.com/txt/ release-10.0

    OpenSSH 10. Release Notes (Apr 2025), https://www.openssh.com/txt/ release-10.0

    2025

  2. [2]

    Post-quantum cryptography standardization (May 2025), https://csrc.nist.gov/ pqc-standardization

    2025

  3. [3]

    pqcrypto (v.0.3.4): Post-quantum cryptography for Python (Jul 2025), https: //pypi.org/project/pqcrypto/

    2025

  4. [4]

    Qiskit machine learning (2025), https://github.com/qiskit-community/ qiskit-machine-learning

    2025

  5. [5]

    Reference Implementation of version 24 of the Java SE Platform (Mar 2025), https://openjdk.org/projects/jdk/24/

    2025

  6. [6]

    Reference Implementation of version 26 of the Java SE Platform (Mar 2026), https://openjdk.org/projects/jdk/26/

    2026

  7. [7]

    Nature574(7779), 505–510 (Oct 2019)

    Arute, F., et al.: Quantum supremacy using a programmable superconducting processor. Nature574(7779), 505–510 (Oct 2019). https://doi.org/10.1038/ s41586-019-1666-5

    2019

  8. [8]

    IEEE Access11, 54175–54187 (2023)

    Baksi, A., Breier, J., Dasu, V.A., Hou, X., Kim, H., Seo, H.: New results on machine learning-based distinguishers. IEEE Access11, 54175–54187 (2023). https: //doi.org/10.1109/access.2023.3270396

    arXiv 2023

  9. [9]

    https: //doi.org/10.48550/ARXIV.2502.08721

    Benedetti, M., Buhrman, H., Weggemans, J.: Complement sampling: Provable, verifiable and nisqable quantum advantage in sample complexity (2025). https: //doi.org/10.48550/ARXIV.2502.08721

    work page doi:10.48550/arxiv.2502.08721 2025

  10. [10]

    New Journal of Physics21(4), 043023 (Apr 2019).https://doi.org/10.1088/1367-2630/ab14b5

    Benedetti, M., Grant, E., Wossnig, L., Severini, S.: Adversarial quantum circuit learning for pure state approximation. New Journal of Physics21(4), 043023 (Apr 2019).https://doi.org/10.1088/1367-2630/ab14b5

    doi:10.1088/1367-2630/ab14b5 2019

  11. [11]

    (eds.): Post-Quantum Cryptography

    Bernstein, D.J., Buchmann, J., Dahmen, E. (eds.): Post-Quantum Cryptography. Springer Berlin Heidelberg, Berlin, Heidelberg (2009)

    2009

  12. [12]

    Nature549(7671), 188–194 (Sep 2017).https://doi.org/10.1038/nature23461

    Bernstein, D.J., Lange, T.: Post-quantum cryptography. Nature549(7671), 188–194 (Sep 2017).https://doi.org/10.1038/nature23461

    doi:10.1038/nature23461 2017

  13. [13]

    Nature Reviews Physics3(9), 625–644 (Aug 2021)

    Cerezo, M., Arrasmith, A., Babbush, R., Benjamin, S.C., Endo, S., Fujii, K., McClean, J.R., Mitarai, K., Yuan, X., Cincio, L., Coles, P.J.: Variational quantum algorithms. Nature Reviews Physics3(9), 625–644 (Aug 2021). https://doi.org/ 10.1038/s42254-021-00348-9

    doi:10.1038/s42254-021-00348-9 2021

  14. [14]

    International Conference on Research in Security Standardisation, pp

    Chen, L.: Standardisation of and Migration to Post-Quantum Cryptography, vol. International Conference on Research in Security Standardisation, pp. 3–13. Springer Nature (2025).https://doi.org/10.1007/978-3-031-87541-0_1

    work page doi:10.1007/978-3-031-87541-0_1 2025

  15. [15]

    Physical Review A98(1), 012324 (Jul 2018)

    Dallaire-Demers, P.L., Killoran, N.: Quantum generative adversarial networks. Physical Review A98(1), 012324 (Jul 2018). https://doi.org/10.1103/physreva. 98.012324 Title Suppressed Due to Excessive Length 13

    doi:10.1103/physreva 2018

  16. [16]

    In: Proceedings of the 10th ACM Asia Public-Key Cryptography Workshop

    Dubrova, E., Ngo, K., G¨ artner, J., Wang, R.: Breaking a fifth-order masked imple- mentation of crystals-kyber by copy-paste. In: Proceedings of the 10th ACM Asia Public-Key Cryptography Workshop. pp. 10–20. ASIA CCS ’23, ACM (Jul 2023). https://doi.org/10.1145/3591866.3593072

    arXiv 2023

  17. [17]

    https://doi

    Goodfellow, I.J., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., Bengio, Y.: Generative adversarial networks (2014). https://doi. org/10.48550/arXiv.1406.2661

    Pith/arXiv arXiv doi:10.48550/arxiv.1406.2661 2014

  18. [18]

    In: Proc

    Grover, L.K.: A fast quantum mechanical algorithm for database search. In: Proc. 28th annual ACM symposium on Theory of computing. pp. 212–219 (1996). https: //doi.org/10.1145/237814.237866

    arXiv 1996

  19. [19]

    Physica Scripta99(8), 086002 (Jul 2024)

    Hariharasitaraman, S., Mishra, N., Vishnuvardhanan, D.: QHopNN: investigating quantum advantage in cryptanalysis using a quantum hopfield neural network. Physica Scripta99(8), 086002 (Jul 2024). https://doi.org/10.1088/1402-4896/ ad5ed1

    doi:10.1088/1402-4896/ 2024

  20. [20]

    Nature 567(7747), 209–212 (Mar 2019).https://doi.org/10.1038/s41586-019-0980-2

    Havl ´ ıˇ cek, V., C´ orcoles, A.D., Temme, K., Harrow, A.W., Kandala, A., Chow, J.M., Gambetta, J.M.: Supervised learning with quantum-enhanced feature spaces. Nature 567(7747), 209–212 (Mar 2019).https://doi.org/10.1038/s41586-019-0980-2

    doi:10.1038/s41586-019-0980-2 2019

  21. [21]

    Science376(6598), 1182–1186 (Jun 2022)

    Huang, H.Y., Broughton, M., Cotler, J., Chen, S., Li, J., Mohseni, M., Neven, H., Babbush, R., Kueng, R., Preskill, J., McClean, J.R.: Quantum advantage in learning from experiments. Science376(6598), 1182–1186 (Jun 2022). https: //doi.org/10.1126/science.abn7293

    doi:10.1126/science.abn7293 2022

  22. [22]

    Nature Communications 12(1) (May 2021).https://doi.org/10.1038/s41467-021-22539-9

    Huang, H.Y., Broughton, M., Mohseni, M., Babbush, R., Boixo, S., Neven, H., McClean, J.R.: Power of data in quantum machine learning. Nature Communications 12(1) (May 2021).https://doi.org/10.1038/s41467-021-22539-9

    doi:10.1038/s41467-021-22539-9 2021

  23. [23]

    https: //doi.org/10.48550/arXiv.2506.18002

    Islam, M., Turkeli, S., Ozaydin, F.: A survey of quantum generative adversarial networks: Architectures, use cases, and real-world implementations (2025). https: //doi.org/10.48550/arXiv.2506.18002

    Pith/arXiv arXiv doi:10.48550/arxiv.2506.18002 2025

  24. [24]

    In: IEEE European Symposium on Security and Privacy, EuroS&P 2022 - Workshops, Genoa, Italy, June 6-10,

    Kannwischer, M.J., Schwabe, P., Stebila, D., Wiggers, T.: Improving software quality in cryptography standardization projects. In: IEEE European Symposium on Security and Privacy, EuroS&P 2022 - Workshops, Genoa, Italy, June 6-10,

    2022

  25. [25]

    Cryptology ePrint Archive (2023), https: //ia.cr/2023/004

    Kim, H., Lim, S., Baksi, A., Kim, D., Yoon, S., Jang, K., Seo, H.: Quantum artificial intelligence on cryptanalysis. Cryptology ePrint Archive (2023), https: //ia.cr/2023/004

    2023

  26. [26]

    In: International conference on learning representations (ICLR)

    Kinga, D., Adam, J.B., et al.: A method for stochastic optimization. In: International conference on learning representations (ICLR). vol. 5. San Diego, California (2015), https://arxiv.org/abs/1412.6980

    Pith/arXiv arXiv 2015

  27. [27]

    American Institute of Aeronautics and Astronautics (Jan 2025)

    Mohammad, K.: Cyber Shield: Advances in Detection, Isolation, and Con- tainment Mechanisms. American Institute of Aeronautics and Astronautics (Jan 2025). https://doi.org/10.2514/6.2025-2724, https://arc.aiaa.org/ doi/abs/10.2514/6.2025-2724

    work page doi:10.2514/6.2025-2724 2025

  28. [28]

    https://doi.org/10.1109/msp

    Mosca, M.: Cybersecurity in an era with quantum computers: Will we be ready? IEEE Security & Privacy16(5), 38–41 (Sep 2018). https://doi.org/10.1109/msp. 2018.3761723,https://ia.cr/2015/1075

    arXiv 2018

  29. [29]

    Electronics12(4) (2023)

    Ngo, T.A., Nguyen, T., Thang, T.C.: A survey of recent advances in quantum generative adversarial networks. Electronics12(4) (2023). https://doi.org/10. 3390/electronics12040856

    2023

  30. [30]

    IEEE Commu- 14 Jaroslaw A

    Nguyen, V.L., Nguyen, L.H., Hwang, R.H., Canberk, B., Duong, T.Q.: Quantum machine learning for 6G network intelligence and adversarial threats. IEEE Commu- 14 Jaroslaw A. Miszczak nications Standards Magazine9(Sep 2025). https://doi.org/10.1109/MCOMSTD. 2025.3575261

    arXiv 2025

  31. [31]

    In: 2024 8th International Conference on Intelligent Systems, Metaheuristics & Swarm Intelligence (ISMSI)

    Nokhwal, S., Nokhwal, S., Pahune, S., Chaudhary, A.: Quantum generative adver- sarial networks: Bridging classical and quantum realms. In: 2024 8th International Conference on Intelligent Systems, Metaheuristics & Swarm Intelligence (ISMSI). pp. 105–109. ISMSI 2024, ACM (Apr 2024). https://doi.org/10.1145/3665065. 3665082

    work page doi:10.1145/3665065 2024

  32. [32]

    Mathematics12(23), 3852 (2024)

    Pajuhanfard, M., Kiani, R., Sheng, V.S.: Survey of quantum generative adversarial networks (qgan) to generate images. Mathematics12(23), 3852 (2024). https: //doi.org/10.3390/math12233852

    work page doi:10.3390/math12233852 2024

  33. [33]

    River Publishers (May 2025)

    Prasad, R., Koren, A.: Safeguarding 6G: Security and Privacy for the Next Genera- tion. River Publishers (May 2025)

    2025

  34. [34]

    Red Hat Inc.: 4 key steps to prepare for post-quantum cryp- tography (May 2025), https://www.redhat.com/en/resources/ 4-steps-for-postquantum-cryptography-checklist

    2025

  35. [35]

    https: //doi.org/10.48550/arXiv.1901.00848

    Romero, J., Aspuru-Guzik, A.: Variational quantum generators: Generative ad- versarial quantum machine learning for continuous distributions (2019). https: //doi.org/10.48550/arXiv.1901.00848

    Pith/arXiv arXiv doi:10.48550/arxiv.1901.00848 2019

  36. [36]

    Sahin, M.E., Altamura, E., Wallis, O., Wood, S.P., Dekusar, A., Millar, D.A., Imamichi, T., Matsuo, A., Mensa, S.: Qiskit machine learning: an open-source library for quantum machine learning tasks at scale on quantum hardware and classical simulators (May 2025),https://arXiv.org/abs/2505.17756

    Pith/arXiv arXiv 2025

  37. [37]

    SIAM Review41(2), 303–332 (Jan 1999)

    Shor, P.W.: Polynomial-time algorithms for prime factorization and discrete logarithms on a quantum computer. SIAM Review41(2), 303–332 (Jan 1999). https://doi.org/10.1137/s0036144598347011

    doi:10.1137/s0036144598347011 1999

  38. [38]

    Advanced Quantum Technologies2(12), 1900070 (2019)

    Sim, S., Johnson, P.D., Aspuru-Guzik, A.: Expressibility and entangling capability of parameterized quantum circuits for hybrid quantum-classical algorithms. Advanced Quantum Technologies2(12), 1900070 (2019)

    2019

  39. [39]

    https://doi.org/10.48550/arXiv

    Ueda, K., Matsuo, A.: Optimizing ansatz design in quantum generative adversarial networks using large language models (2025). https://doi.org/10.48550/arXiv. 2503.12884

    Pith/arXiv arXiv doi:10.48550/arxiv.2503.12884 2025

  40. [40]

    https://doi.org/10.48550/arXiv.2310.10315

    Zaman, K., Marchisio, A., Hanif, M.A., Shafique, M.: A survey on quantum machine learning: Current trends, challenges, opportunities, and the road ahead (2023). https://doi.org/10.48550/arXiv.2310.10315

    Pith/arXiv arXiv doi:10.48550/arxiv.2310.10315 2023

  41. [41]

    npj Quantum Information5(1) (Nov 2019).https://doi.org/10.1038/s41534-019-0223-2

    Zoufal, C., Lucchi, A., Woerner, S.: Quantum generative adversarial networks for learning and loading random distributions. npj Quantum Information5(1) (Nov 2019).https://doi.org/10.1038/s41534-019-0223-2

    doi:10.1038/s41534-019-0223-2 2019

  42. [2022]

    pp. 19–30. IEEE Computer Society, Los Alamitos, CA, USA (2022). https: //doi.org/10.1109/EuroSPW55150.2022.00010

    arXiv 2022