Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Whole-Body Bilateral Teleoperation with Multi-Stage Object Parameter Estimation for Wheeled Humanoid Locomanipulation

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read Pure-state quantum generative adversarial networks are fundamentally limited: even a perfectly trained generator's best possible output is the data ensemble's dominant eigenstate, so high-rank distributions cannot be learned.

desk verdict The arXiv metadata says teleoperation, but the actual paper is a QGAN limitations study; the fidelity bound is correct, but the experiments don't actually test it because they measure and mix the state. read the letter →

arxiv 2508.09846 v1 pith:OI2W663M submitted 2025-08-13 cs.RO

Donghoon Baek , Amartya Purushottam , Jason J. Choi , Joao Ramos This is my paper
classification cs.RO MSC 68Q1281P16 PACS 03.67.-a03.67.Lx
keywords quantumgenerativeadversarialnetworkspure-stategeneratorsfidelityboundsmodecollapsegeneralizationlimitsimagegenerationmachinelearningMNIST
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 asks whether fully quantum GANs—where both generator and discriminator are variational quantum circuits—actually learn data distributions or merely reproduce dominant features of the training data. Through tests of two leading architectures (QuGAN and IQGAN) on MNIST and CIFAR-10, it finds the latter: generated images resemble class averages, FID scores match a random-sampling baseline, and without PCA preprocessing the circuits collapse to noise. The paper then proves why. When a generator outputs a single pure quantum state, its fidelity to the target data ensemble is capped at $\sqrt{\lambda_{\max}(\rho_{\mathrm{data}})}$, the square root of the data density matrix's largest eigenvalue. So for high-rank, diverse data, no pure-state generator can approximate the full distribution even in the best case; the observed collapse is a theoretical ceiling, not a training deficiency. This matters because it separates architecture-specific failures from a fundamental limitation of an entire class of quantum generative models.

What carries the argument

The central object is the fidelity between a pure-state generator and a mixed data ensemble: $F(\rho_{\mathrm{data}}, |\gamma\rangle\langle\gamma|) = \sqrt{\langle\gamma|\rho_{\mathrm{data}}|\gamma\rangle}$, maximized by the principal eigenvector of $\rho_{\mathrm{data}}$. This connects a state-discrimination bound (the Helstrom measurement) to generalization: if the data state has rank $r$ and a roughly flat spectrum, then $\lambda_{\max}(\rho_{\mathrm{data}}) \ge 1/r$, so the maximum achievable fidelity scales as $1/\sqrt{r}$. The mechanism therefore converts spectral concentration—how peaked the data's eigenvalue spectrum is—into a hard ceiling on what any pure-state quantum generator can

What would settle it

Directly estimate the trained generator's quantum state without sampling to images—for instance by state tomography or by computing its fidelity to $\rho_{\mathrm{data}}$—on a high-rank dataset. If the state's fidelity is near $\sqrt{\lambda_{\max}(\rho_{\mathrm{data}})}$ while the decoded sampled-image FID remains no better than random, then the measurement/decoding pipeline, not the pure-state bound, explains the observed collapse. Conversely, a mixed-state generator using ancilla-induced decoherence that beats the pure-state ceiling on sampled FID would confirm the bound is binding.

Watch

Extended reading notes

Core claim

The central claim is analytic: for a QGAN whose generator produces a pure state $\rho_G=|\gamma\rangle\langle\gamma|$ and whose discriminator is optimal, the maximum possible fidelity between the generated state and the data ensemble $\rho_{\mathrm{data}}$ is $F_{\max} = \sqrt{\lambda_{\max}(\rho_{\mathrm{data}})}$, where $\lambda_{\max}$ is the leading eigenvalue of the data density operator. Combined with the Helstrom bound and the Fuchs–van de Graaf inequalities, this yields the best-case discrimination bound $P_{\mathrm{succ}} \ge 1 - \tfrac{1}{2}\sqrt{\lambda_{\max}(\rho_{\mathrm{data}})}$. The paper interprets this as a fundamental constraint: a pure-state generator is optimal when it

Load-bearing premise

The load-bearing premise is that the poor images observed after measurement and inverse-PCA decoding are governed by the pure-state fidelity bound, rather than by the measurement-induced classical mixture or the aggressive PCA compression that the evaluation pipeline itself introduces.

Editorial extensions

If this is right

  • Any QGAN with a pure-state generator can represent at most one dominant mode of the data; training can only choose which principal direction to emit, not the shape of the distribution.
  • The fidelity cap applies to any generative model that outputs a single pure quantum state without sampling or post-selection, including quantum circuit Born machines and variational state-preparation circuits.
  • For high-rank or high-entropy datasets, pure-state generators will inevitably underperform, so benchmark comparisons should report or control $\lambda_{\max}(\rho_{\mathrm{data}})$ when interpreting image-quality metrics.
  • Escaping the ceiling requires mixed-state outputs—via ancillary qubits, mid-circuit measurements, or classical sampling over circuit parameters—since the proof only constrains pure-state generators.
  • A near-random FID score becomes the expected signature of a well-trained pure-state QGAN rather than evidence of poor optimization.

Reading between the lines

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

  • A quantitative design rule follows: before training a pure-state quantum generator, compute $\lambda_{\max}$ of the intended data encoding; if it is small, no amount of circuit depth or gradient tuning can make the model generalize, so resources should go to mixed-state architectures.
  • A sharper test of the paper's mechanism would be to train on synthetic data with a controlled eigenvalue spectrum and check whether sampled FID degrades exactly as the bound predicts; a mismatch would indicate that optimization failure or measurement noise, not the pure-state ceiling, drives the reported results.
  • The same bound should appear in other fully quantum generative models whenever they are evaluated after projective measurement, because the measured output is a classical mixture rather than the pure state the theorem constrains.
  • If the bound is the true cause, then retaining more PCA components—which increases the effective rank of $\rho_{\mathrm{data}}$—should quantitatively tighten the pure-state ceiling, making the collapse worse even as the encoding becomes more expressive.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The submitted full text is a quantum machine learning paper rather than the teleoperation paper announced in the title and abstract. It evaluates two fully quantum GAN architectures, QuGAN and IQGAN, on MNIST and CIFAR-10 image generation. Qualitatively, the authors report that the generated images resemble class averages; quantitatively, they report FID values close to a uniform-random latent-sampling baseline. The paper then derives an analytical bound for QGANs whose generator outputs a single pure state: the optimal discriminator success probability satisfies P_succ >= 1 - (1/2) sqrt(lambda_max(rho_data)), where rho_data is the data ensemble. The authors argue that this bound explains the observed poor generalization as an intrinsic pure-state limitation.

Significance. If the empirical connection were established, the pure-state fidelity bound would be a valuable conceptual result: it gives a clean spectral criterion for when a pure-state generator cannot represent a mixed data distribution, and it is derived correctly from standard quantum distinguishability tools (Helstrom bound, Fuchs-van de Graaf inequality, spectral decomposition). The paper also makes a useful distinction between true generative learning and deterministic reproduction of a dominant mode. However, the manuscript as submitted does not establish that connection. The FID protocol involves projective measurements and classical decoding, while the theorem applies to the unmeasured pure state; no quantitative fidelity measurement is reported. The mismatch between the announced subject and the actual full text is a separate, serious editorial problem. The paper also does not provide code or data for reproducibility.

major comments (4)
  1. [Title/Abstract vs. Full Text] The front matter advertises arXiv:2508.09846 (cs.RO), 'Whole-Body Bilateral Teleoperation with Multi-Stage Object Parameter Estimation for Wheeled Humanoid Locomanipulation', and an abstract describing a teleoperation system. The full text is a quantum-GAN paper, arXiv:2508.09844v2, 'On the Generalization Limits of Quantum Generative Adversarial Networks with Pure State Generators'. These are different papers. As submitted, the manuscript is not a coherent contribution: the abstract makes claims about a system that does not appear in the body, and the body's claims are not represented in the abstract. This must be resolved by the authors and the editor before any technical review can be finalized.
  2. [Section 2, FID protocol; Section 3, Eq. (11)] The load-bearing transfer from the pure-state theorem to the experimental FID results is not established. The paper states that 'there is no stochastic measurement-induced distribution' and that the generator represents the entire model distribution by a single deterministic quantum state. Yet the FID evaluation explicitly performs projective measurements on the generator qubits, obtains binary bitstrings, and decodes them via inverse normalization and inverse PCA. The resulting empirical distribution is a classical mixture rho_mix = sum_b |<b|gamma>|^2 |b><b|, not the pure state |gamma><gamma|. Fidelity is monotone under the dephasing CPTP map, so F(rho_mix, rho_data) can be larger than F(|gamma>, rho_data); Eq. (11) does not bound the mixture. The paper never proves that the measured/decoded distribution inherits the bound of Eq. (12). Thus the near-random FID values do not, by themsel
  3. [Section 3, Fig. 4] The claim that QuGAN and IQGAN 'converge toward this principal eigenstate' is supported only by visual resemblance between generated images and the plotted first principal components. No quantitative overlap <gamma|rho_data|gamma>, no estimate of lambda_max(rho_data), and no direct fidelity comparison between the trained generator state and the data ensemble are reported. Without such measurements, the theoretical bound is not connected to the empirical failure. The FID being close to the uniform-random baseline could equally be caused by the aggressive PCA compression, the measurement-induced mixing, or the decoding pipeline. The authors should compute the actual overlap of the generator's pure state with rho_data, or otherwise directly test Eq. (11).
  4. [Section 3, Eqs. (2)-(6)] The analytic bound is derived for the optimal Helstrom measurement in a balanced prior setting. However, the numerical experiments use IQGAN's fixed swap-test discriminator and QuGAN's trained variational discriminator, and the evaluation metric is FID, not discriminator success probability. The paper does not show that either architecture's training objective is governed by the Helstrom success probability, nor that the FID metric is monotone in the quantities appearing in Eq. (12). Consequently, the conclusion that the observed failure 'stems from' the pure-state fidelity limit is an inference beyond what the theorem proves. The authors should either test the bound directly or explicitly soften the causal claim.
minor comments (5)
  1. [Section 2, Fig. 1] Figure 1(b)(v) is referenced twice for two different claims: once for the random-number-generator baseline and once for the amplitude-encoding experiment without PCA. The figure caption identifies (v) only as the random-input inverse-PCA image. The citation should be corrected and the panels renumbered.
  2. [Section 2, IQGAN-784] The notation 'IQGAN-784' is used without definition. State explicitly that this is the 784-qubit circuit with one qubit per pixel and angle embedding, as described in the preceding paragraph.
  3. [Appendix B] The training details report optimizer, learning rate, batch size, and epochs, but do not mention random seeds, initialization strategy, or the number of independent runs. Given that the central empirical claim is that trained QGANs behave like random sampling, some estimate of run-to-run variability is needed.
  4. [Data and Code availability] The paper states that data and code will be made available on reasonable request, but no repository is provided. For a paper whose conclusions depend on numerical simulations and specific circuit architectures, a public code/data release would substantially strengthen verifiability.
  5. [General text] There are several typographical issues, including 'CIF AR-10' for CIFAR-10, a missing space in 'Fr´echet', and inconsistent use of 'QUGAN' vs 'QuGAN'. These are minor but should be cleaned up.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Eq. (12) follows from standard state-discrimination theorems and the spectral decomposition of rho_data; no prediction reduces to a fitted input.

full rationale

The paper's central derivation (Section 3) starts from the Helstrom bound (Eq. 2), the Fuchs-van de Graaf inequality (Eq. 7), and the definition of fidelity for a pure generator (Eq. 9). Maximizing <gamma|rho_data|gamma> over pure states yields lambda_max(rho_data) by the spectral theorem, and Eq. (12) is an algebraic consequence of these ingredients, not an assumed conclusion. No parameter is fitted to the data whose prediction is then reported. The empirical claim that IQGAN/QuGAN converge to the principal eigenstate is offered as a visual post-hoc interpretation (Fig. 4), not as a derived prediction, and the bound does not depend on that observation. The only self-citation, [51], supports a general contextual remark ('which in general is a problem of current QNN') and is not load-bearing for the theorem or the main conclusion. The Appendix C caveat that the classical comparison is 'not strictly fair', and the gap between the pure-state bound and the measured/decoded FID distribution, are validity concerns rather than circularity: they do not show that Eq. (12) assumes what it proves. Therefore the derivation chain is self-contained and non-circular.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The theoretical result is parameter-free: it is a corollary of the spectral theorem, Fuchs-van de Graaf inequalities, and the Helstrom bound. No quantity is fitted to data. The numerical study uses hand-set hyperparameters (10 epochs, Adam lr 0.001, batch 32; Appendix B) and a swept PCA dimension k; none enters the bound. No new entities are postulated. The load-bearing domain assumptions concern the reduction of the analyzed architectures to fidelity-based pure-state generators; the authors themselves note IQGAN "aligns more closely with a Quantum Circuit Born Machine", and the FID protocol samples bitstrings, in tension with the pure-state assumption.

free parameters (3)
  • Training epochs = 10
    Appendix B: "kept fixed since the generator loss and fidelity empirically converge within this range". Hand-selected; affects only the numerical demonstration, not the analytic bound.
  • Optimizer learning rate and batch size = Adam, lr=0.001, batch=32
    Appendix B; standard choices, not fitted to the bound.
  • Number of PCA components / generator qubits k = 2 to 16 (swept in Fig. 3)
    Swept dimension controlling data compression; the FID comparison depends on it, the analytic bound does not.
assumptions (5)
  • standard math Fuchs-van de Graaf inequalities: 1 - F(rho,sigma) <= (1/2)||rho - sigma||_1 (Eq. 7)
    Standard quantum information inequality linking fidelity to trace distance; used to convert the Helstrom success probability into the fidelity-based bound (Eq. 8).
  • standard math The Helstrom measurement is the optimal binary state-discrimination strategy (Eq. 2)
    Textbook result; used to model the quantum discriminator's best-case output as P_succ = 1/2(1 + (1/2)||rho_data - rho_G||_1).
  • standard math max over pure states of <gamma|rho_data|gamma> equals lambda_max(rho_data) (Rayleigh-Ritz, Eq. 11)
    Spectral theorem; the core step behind Eq. (11) and hence the main bound.
  • domain assumption The analyzed generators emit a single pure state with no sampling, no classical mixing, and no measurement-induced distribution
    Section 2 states this explicitly; it is required for the bound to describe the models, but it is in tension with the FID protocol, which samples bitstrings.
  • domain assumption A swap-test discriminator induces a fidelity-based objective, so the QGAN reduces to a QCBM-like optimization
    Section 2 argues IQGAN "aligns more closely with a Quantum Circuit Born Machine"; for QuGAN, whose discriminator is trainable, the reduction is approximate and the bound applies only loosely.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Whole-Body Bilateral Teleoperation with Multi-Stage Object Parameter Estimation for Wheeled Humanoid Locomanipulation." pith.science (2026). https://pith.science/paper/OI2W663M

@misc{pith2026250809846,
  author       = {Pith},
  title        = {Pith review of: Whole-Body Bilateral Teleoperation with Multi-Stage Object Parameter Estimation for Wheeled Humanoid Locomanipulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OI2W663M}},
  note         = {Machine review of arXiv:2508.09846}
}
read the original abstract

This paper presents an object-aware whole-body bilateral teleoperation framework for wheeled humanoid loco-manipulation. This framework combines whole-body bilateral teleoperation with an online multi-stage object inertial parameter estimation module, which is the core technical contribution of this work. The multi-stage process sequentially integrates a vision-based object size estimator, an initial parameter guess generated by a large vision-language model (VLM), and a decoupled hierarchical sampling strategy. The visual size estimate and VLM prior offer a strong initial guess of the object's inertial parameters, significantly reducing the search space for sampling-based refinement and improving the overall estimation speed. A hierarchical strategy first estimates mass and center of mass, then infers inertia from object size to ensure physically feasible parameters, while a decoupled multi-hypothesis scheme enhances robustness to VLM prior errors. Our estimator operates in parallel with high-fidelity simulation and hardware, enabling real-time online updates. The estimated parameters are then used to update the wheeled humanoid's equilibrium point, allowing the operator to focus more on locomotion and manipulation. This integration improves the haptic force feedback for dynamic synchronization, enabling more dynamic whole-body teleoperation. By compensating for object dynamics using the estimated parameters, the framework also improves manipulation tracking while preserving compliant behavior. We validate the system on a customized wheeled humanoid with a robotic gripper and human-machine interface, demonstrating real-time execution of lifting, delivering, and releasing tasks with a payload weighing approximately one-third of the robot's body weight.

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Switch4EAI: Leveraging Console Game Platform for Benchmarking Robotic Athletics

    cs.RO 2025-08 conditional novelty 5.0 of 10

    Switch4EAI uses Just Dance on a Nintendo Switch as a benchmark: a Unitree G1 robot scored 5,707 versus a human's 9,361 on three routines.

Reference graph

Works this paper leans on

62 extracted references · 58 canonical work pages · cited by 1 Pith paper

  1. [1]

    A comprehensive survey and analysis of generative models in machine learning

    G. M. Harshvardhan, M. K. Gourisaria, M. Pandey, and S. S. Rautaray,“A comprehensive survey and analysis of generative models in machine learning”, Comput. Sci. Rev.38, 100285 (2020). 15

    work page 2020

  2. [2]

    Scaling laws for neural language models

    J. Kaplan, S. McCandlish, T. Henighan, T. B. Brown, B. Chess, R. Child, S. Gray, A. Radford, J. Wu, and D. Amodei,“Scaling laws for neural language models”, arXiv:2001.08361 (2020)

    arXiv 2001

  3. [3]

    Deep learning: a comprehensive overview on techniques, taxonomy, applications and research directions

    I. H. Sarker,“Deep learning: a comprehensive overview on techniques, taxonomy, applications and research directions”, SN Comput. Sci.2, 1–20 (2021)

    work page 2021

  4. [4]

    A comprehensive review of quantum machine learning: from NISQ to fault tolerance

    Y. Wang and J. Liu,“A comprehensive review of quantum machine learning: from NISQ to fault tolerance”, Rep. Prog. Phys.87, 116402 (2024)

    work page 2024

  5. [5]

    Quantum machine learning on near- term quantum devices: Current state of supervised and unsupervised techniques for real-world applications

    Y. Gujju, A. Matsuo, and R. Raymond,“Quantum machine learning on near- term quantum devices: Current state of supervised and unsupervised techniques for real-world applications”, Phys. Rev. Appl.21, 067001 (2024)

    work page 2024

  6. [6]

    Noisy intermediate-scale quantum algorithms

    K. Bharti, A. Cervera-Lierta, T. H. Kyaw, T. Haug, S. Alperin-Lea, A. Anand, M. Degroote, H. Heimonen, J. S. Kottmann, T. Menke, and others,“Noisy intermediate-scale quantum algorithms”, Rev. Mod. Phys.94, 015004 (2022)

    work page 2022

  7. [7]

    Machine learning applications for noisy intermediate-scale quantum computers

    B. Coyle,“Machine learning applications for noisy intermediate-scale quantum computers”, arXiv:2205.09414 (2022)

    arXiv 2022

  8. [8]

    Fisher information in noisy intermediate-scale quantum applica- tions

    J. J. Meyer,“Fisher information in noisy intermediate-scale quantum applica- tions”, Quantum5, 539 (2021)

    work page 2021

Show all 62 references
  1. [9]

    Detecting quantum entanglement with unsupervised learning

    Y. Chen, Y. Pan, G. Zhang, and, S. Cheng,“Detecting quantum entanglement with unsupervised learning”, Quantum Sci. Technol.7, 015005 (2021)

    2021

  2. [10]

    Transforming Bell’s inequalities into state classifiers with machine learning

    Y. C. Ma, and, M. H. Yung,“Transforming Bell’s inequalities into state classifiers with machine learning”, npj Quantum Inf.4, 34 (2018)

    2018

  3. [11]

    A rigorous and robust quantum speed- up in supervised machine learning

    Y. Liu, S. Arunachalam, and, K. Temme,“A rigorous and robust quantum speed- up in supervised machine learning”, Nat. Phys.17, 1013-1017 (2021)

    2021

  4. [12]

    Practical advantage of quantum machine learning in ghost imaging

    T. Xiao, X. Zhai, X. Wu, et al.,“Practical advantage of quantum machine learning in ghost imaging”, Commun. Phys.6, 171 (2023)

    2023

  5. [13]

    Toward Heisenberg Limit without Critical Slowing Down via Quantum Reinforcement Learning

    H. Xu, T. Xiao, J. Huang, M. He, J. Fan, and, G. Zeng,“Toward Heisenberg Limit without Critical Slowing Down via Quantum Reinforcement Learning”, Phys. Rev. Lett.134, 120803 (2021)

    2021

  6. [14]

    Fully unsupervised diversity denoising with convolutional variational autoencoders

    M. Prakash, A. Krull, and F. Jug,“Fully unsupervised diversity denoising with convolutional variational autoencoders”, arXiv:2006.06072 (2020)

    2006 arXiv

  7. [15]

    Variational autoencoder based anomaly detection using reconstruction probability

    J. An and S. Cho,“Variational autoencoder based anomaly detection using reconstruction probability”, Special Lecture on IE2, 1–18 (2015)

    2015

  8. [16]

    Generative adversarial nets

    I. J. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Bengio,“Generative adversarial nets”, Adv. Neural Inf. 16 Process. Syst.27(2014)

    2014

  9. [17]

    Alias-free generative adversarial networks

    T. Karras, M. Aittala, S. Laine, E. H¨ ark¨ onen, J. Hellsten, J. Lehtinen, and T. Aila,“Alias-free generative adversarial networks”, Adv. Neural Inf. Process. Syst. 34, 852–863 (2021)

    2021

  10. [18]

    StyleGAN-T: Unlocking the power of GANs for fast large-scale text-to-image synthesis

    A. Sauer, T. Karras, S. Laine, A. Geiger, and T. Aila,“StyleGAN-T: Unlocking the power of GANs for fast large-scale text-to-image synthesis”, Int. Conf. Mach. Learn. ICML 30105–30118 (2023)

    2023

  11. [19]

    StarGAN: Unified generative adversarial networks for multi-domain image-to-image translation

    Y. Choi, M. Choi, M. Kim, J.-W. Ha, S. Kim, and J. Choo,“StarGAN: Unified generative adversarial networks for multi-domain image-to-image translation ”, Proc. IEEE Int. Conf. Comput. Vis. 8789–8797 (2018)

    2018

  12. [20]

    Encoding in style: A StyleGAN encoder for image-to-image transla- tion

    E. Richardson, Y. Alaluf, O. Patashnik, Y. Nitzan, Y. Azar, S. Shapiro, and D. Cohen-Or,“Encoding in style: A StyleGAN encoder for image-to-image transla- tion ”,Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit. 2287–2296 (2021)

    2021

  13. [21]

    Vector quantized diffusion model for text-to-image synthesis

    S. Gu, D. Chen, J. Bao, F. Wen, B. Zhang, D. Chen, L. Yuan, and B. Guo,“Vector quantized diffusion model for text-to-image synthesis”, Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit. 10696–10706 (2022)

    2022

  14. [22]

    High-resolution image synthesis with latent diffusion models

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer,“High-resolution image synthesis with latent diffusion models”, Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit. 10684–10695 (2022)

    2022

  15. [23]

    Quantum variational autoencoder

    A. Khoshaman, W. Vinci, B. Denis, E. Andriyash, H. Sadeghi, and M. H. Amin, “Quantum variational autoencoder”, Quantum Sci. Technol.4, 014001 (2018)

    2018

  16. [24]

    Quantum generative adversarial learning

    S. Lloyd and C. Weedbrook,“Quantum generative adversarial learning”, Phys. Rev. Lett.,121040502 (2018)

    2018

  17. [25]

    Quantum-Noise-Driven Generative Diffusion Models

    M. Parigi, S. Martina, and F. Caruso,“Quantum-Noise-Driven Generative Diffusion Models”, Adv. Quantum Technol. 2300401 (2024)

    2024

  18. [26]

    Quantum generative adversarial net- works

    P.-L. Dallaire-Demers and N. Killoran,“Quantum generative adversarial net- works”, Phys. Rev. A98, 012324 (2018)

    2018

  19. [27]

    Quantum generative adversarial network for generating discrete distribution

    H. Situ, Z. He, Y. Wang, L. Li, and S. Zheng,“Quantum generative adversarial network for generating discrete distribution ”, Inf. Sci.538, 193–208 (2020)

    2020

  20. [28]

    A characterization of quantum generative models

    C. A. Riofrio, O. Mitevski, C. Jones, F. Krellner, A. Vuckovic, J. Doetsch, J. Klepsch, T. Ehmer, and A. Luckow,“A characterization of quantum generative models”, ACM T. Quantum Comput.5, 12 (2024)

    2024

  21. [29]

    Variational quantum generators: Genera- tive adversarial quantum machine learning for continuous distributions

    J. Romero and A. Aspuru-Guzik,“Variational quantum generators: Genera- tive adversarial quantum machine learning for continuous distributions”, Adv. 17 Quantum Technol.4, 2000003 (2021)

    2021

  22. [30]

    Quantum generative adversarial networks for learning and loading random distributions

    C. Zoufal, A. Lucchi, and S. Woerner,“Quantum generative adversarial networks for learning and loading random distributions”, npj Quantum Inf.5, 103 (2019)

    2019

  23. [31]

    Experimental quantum generative adversarial networks for image generation

    H.-L. Huang, Y. Du, M. Gong, Y. Zhao, Y. Wu, C. Wang, S. Li, F. Liang, J. Lin, Y. Xu, and others,“Experimental quantum generative adversarial networks for image generation ”, Phys. Rev. Appl.16, 024051 (2021)

    2021

  24. [32]

    Quantum generative models for small molecule drug discovery

    J. Li, R. O. Topaloglu, and S. Ghosh,“Quantum generative models for small molecule drug discovery”, IEEE Trans. Quantum Eng.2, 1–8 (2021)

    2021

  25. [33]

    Hybrid quantum–classical generative adversarial networks for image generation via learning discrete distri- bution

    N.-R. Zhou, T.-F. Zhang, X.-W. Xie, and J.-Y. Wu,“Hybrid quantum–classical generative adversarial networks for image generation via learning discrete distri- bution ”, Signal Process.: Image Commun.110, 116891 (2023)

    2023

  26. [34]

    Hybrid quantum– classical generative adversarial network for high-resolution image generation

    S. L. Tsang, M. T. West, S. M. Erfani, and M. Usman,“Hybrid quantum– classical generative adversarial network for high-resolution image generation ”, IEEE Trans. Quantum Eng.4, 1–19 (2023)

    2023

  27. [35]

    LSTM-QGAN: Scalable NISQ generative adversarial network

    C. Chu, A. Hastak, and F. Chen,“LSTM-QGAN: Scalable NISQ generative adversarial network”, arXiv:2409.02212 [quant-ph] (2025)

    2025 arXiv

  28. [36]

    Entangling quantum generative adversarial networks

    M. Y. Niu, A. Zlokapa, M. Broughton, S. Boixo, M. Mohseni, V. Smelyanskyi, and H. Neven,“Entangling quantum generative adversarial networks”, Phys. Rev. Lett.128, 220505 (2022)

    2022

  29. [37]

    Hamiltonian quantum generative adversarial networks

    L. Kim, S. Lloyd, and M. Marvian,“Hamiltonian quantum generative adversarial networks”, Phys. Rev. Res.6, 033019 (2024)

    2024

  30. [38]

    Quantum Wasserstein generative adversarial networks

    S. Chakrabarti, Y. Huang, T. Li, S. Feizi, and X. Wu,“Quantum Wasserstein generative adversarial networks”, Adv. Neural Inf. Process. Syst.32(2019)

    2019

  31. [39]

    Towards a scalable discrete quantum generative adversarial neural network

    S. Chaudhary, P. Huembeli, I. MacCormack, T. L. Patti, J. Kossaifi, and A. Galda, “Towards a scalable discrete quantum generative adversarial neural network”, Quantum Sci. Technol.8, 035002 (2023)

    2023

  32. [40]

    Quantum generative adversarial learning in a superconducting quantum circuit

    L. Hu, S.-H. Wu, W. Cai, Y. Ma, X. Mu, Y. Xu, H. Wang, Y. Song, D.-L. Deng, C.-L. Zou, and others,“Quantum generative adversarial learning in a superconducting quantum circuit”, Sci. Adv.5, eaav2761 (2019)

    2019

  33. [41]

    Quantum generative adversarial networks with multiple superconducting qubits

    K. Huang, Z. A. Wang, C. Song et al.,“Quantum generative adversarial networks with multiple superconducting qubits”, npj Quantum Inf. , 165 (2021)

    2021

  34. [42]

    QUGAN: A quantum state fidelity based generative adversarial network

    S. A. Stein, B. Baheri, D. Chen, Y. Mao, Q. Guan, A. Li, B. Fang, and S. Xu, “QUGAN: A quantum state fidelity based generative adversarial network”, IEEE International Conference on Quantum Computing and Engineering (QCE) 71–81 (2021). 18

    2021

  35. [43]

    IQGAN: Robust quantum generative adversarial network for image synthesis on NISQ devices

    C. Chu, G. Skipper, M. Swany, and F. Chen,“IQGAN: Robust quantum generative adversarial network for image synthesis on NISQ devices”, IEEE Inter- national Conference on Acoustics, Speech and Signal Processing (ICASSP) 1–5 (2023)

    2023

  36. [44]

    Classification of digits using scikit-learn

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and ´E. Duchesnay,“Classification of digits using scikit-learn ”, Online example (2024)

    2024

  37. [45]

    Better than classical? The subtle art of benchmarking quantum machine learning models

    J. Bowles, S. Ahmed, and M. Schuld,“Better than classical? The subtle art of benchmarking quantum machine learning models”, arXiv:2403.07059 [quant-ph] (2024)

    2024 arXiv

  38. [46]

    Quantum fingerprinting

    H. Buhrman, R. Cleve, J. Watrous, and R. De Wolf,“Quantum fingerprinting”, Phys. Rev. Lett.87, 167902 (2001)

    2001

  39. [47]

    Machine learning with quantum computers

    M. Schuld and F. Petruccione,“Machine learning with quantum computers”, Springer (2021)

    2021

  40. [48]

    Differentiable learning of quantum circuit Born machines

    J.-G. Liu and L. Wang,“Differentiable learning of quantum circuit Born machines”, Phys. Rev. A98, 062324 (2018)

    2018

  41. [49]

    Generative adversarial networks

    I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Bengio,“Generative adversarial networks”, Commun. ACM 63, 139–144 (2020)

    2020

  42. [50]

    Learning multiple layers of features from tiny images

    A. Krizhevsky, G. Hinton, and others,“Learning multiple layers of features from tiny images”, Toronto, ON, Canada (2009)

    2009

  43. [51]

    DisQu: Investigating the impact of disorder in quantum generative models

    Y. Werner, J. Frkatovic, V. F. Rey, M. Tsch¨ ope, S. Suh, P. Lukowicz, N. Palaiodi- mopoulos, and M. Kiefer-Emmanouilidis,“DisQu: Investigating the impact of disorder in quantum generative models”, arXiv:2409.16180 [cond-mat.dis-nn] (2025)

    2025

  44. [52]

    Mosaiq: Quantum generative adversarial networks for image generation on nisq comput- ers

    D. Silver, T. Patel, W. Cutler, A. Ranjan, H. Gandhi, and D. Tiwari,“Mosaiq: Quantum generative adversarial networks for image generation on nisq comput- ers”, Proc. IEEE Int. Conf. Comput. Vis, 7030–7039, (2023)

    2023

  45. [53]

    Quantum architecture search for optimizing quantum generators in quantum GAN

    Q. Ma, C. Hao, N. Si, and D. Qu,“Quantum architecture search for optimizing quantum generators in quantum GAN”, Mach. Learn.: Sci. Technol.6, 035061 (2025)

    2025

  46. [54]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, and S. Hochreiter,“Gans trained by a two time-scale update rule converge to a local nash equilibrium”, NeurIPS,30, (2017). 19

    2017

  47. [55]

    Quantum detection and estimation theory

    C. W. Helstrom,“Quantum detection and estimation theory”, J. Stat. Phys.1, 231–252 (1969)

    1969

  48. [56]

    Quantum information

    S. Barnett,“Quantum information ”, Oxford University Press (2009)

    2009

  49. [57]

    Cryptographic distinguishability measures for quantum-mechanical states

    C. A. Fuchs, and J. Van De Graaf,“Cryptographic distinguishability measures for quantum-mechanical states”, IEEE Trans. Inf. Theory45, 1216–1227 (2002)

    2002

  50. [58]

    No scratch quantum computing by reducing qubit overhead for efficient arithmetics

    O. Faizy, N. Wehn, P. Lukowicz, and M. Kiefer-Emmanouilidis,“No scratch quantum computing by reducing qubit overhead for efficient arithmetics”, arXiv:2506.17135 [quant-ph] (2025)

    2025 arXiv

  51. [59]

    Quantum convolutional neural networks

    I. Cong, S. Choi, and M. D. Lukin,“Quantum convolutional neural networks”, Nat. Phys.15, 1273–1278 (2019)

    2019

  52. [60]

    Expressibility and entangling capa- bility of parameterized quantum circuits for hybrid quantum-classical algorithms

    S. Sim, P. D. Johnson, and A. Aspuru–Guzik,“Expressibility and entangling capa- bility of parameterized quantum circuits for hybrid quantum-classical algorithms”, Adv. Quantum Technol.2, 1900070 (2019)

    2019

  53. [61]

    Connecting ansatz express- ibility to gradient magnitudes and barren plateaus

    Z. Holmes, K. Sharma, M. Cerezo, and P. J. Coles,“Connecting ansatz express- ibility to gradient magnitudes and barren plateaus”, PRX Quantum3, 010313 (2022)

    2022

  54. [62]

    Quantum boltzmann machine

    M. H. Amin, E. Andriyash, J. Rofle, B. Kulchytskyy, and R. Melko,“Quantum boltzmann machine”, PRX Quantum8, 021050 (2018). 20

    2018

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.