Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Improving GANs by leveraging the quantum noise from real hardware

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read The paper claims that using bitstrings from a 16-qubit entangling circuit, especially with real hardware noise, as the latent prior of GANs consistently improves FID compared to the standard Gaussian prior and to a noiseless simulator.

desk verdict The pipeline is new and the FID gains are plausibly real, but the paper never tests a classical prior with the same covariance as the quantum pool, so the 'unique quantum correlations' attribution is unsubstantiated. read the letter →

arxiv 2507.01886 v1 pith:CJ6XEQH2 submitted 2025-07-02 quant-ph

Hongni Jin , Kenneth M. Merz Jr This is my paper
classification quant-ph
keywords quantumgenerativeadversarialnetworksnoiseasaresourcelatentpriorFrechetinceptiondistanceentanglementnoisyintermediate-scalehardwareCIFAR-10hybridrepresentation
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

The paper tries to establish that the standard i.i.d. Gaussian latent of a GAN can be replaced, in whole or in part, by a latent built from measurement bitstrings of a 16-qubit entangling circuit, and that doing so improves image quality on CIFAR-10 without touching the generator or discriminator. The construction turns blocks of repeated qubit measurements into binary fractions, applies the inverse Gaussian CDF to each coordinate, and projects the resulting 16-dimensional Gaussian vector into the full latent space with a fixed random matrix, so the only new ingredient is the joint correlation structure, or copula, inherited from the quantum circuit. Across three architectures, the hybrid quantum-classical prior produces lower FID than the Gaussian baseline, and real hardware noise outperforms the noiseless simulator, with the best results coming from running two layouts in parallel on the QPU. The paper's conclusion is that device-specific quantum noise is a structured inductive bias rather than a liability, opening a practical route for using current noisy quantum processors to enrich classical generative models.

What carries the argument

The machinery is the quantum-correlated latent pipeline: encode 16-qubit measurement bitstrings as binary fractions $u_{b,q}=\sum_{s=1}^S b_{s,q}2^{-(s+1)}$, map each coordinate through the inverse Gaussian CDF $\Phi^{-1}$ to get a marginal standard Gaussian, then multiply the $B\times 16$ matrix by a fixed row-orthonormal projection $P$ to reach the GAN's latent dimension. The identity that carries the argument is the hybrid mixture $z_{\mathrm{hyb}}=\alpha z_{\mathrm{quant}}+(1-\alpha)z_{\mathrm{class}}$, which interpolates between an unstructured Gaussian and the quantum-correlated prior while preserving off-diagonal covariance. The entanglement-generated copula, partially eroded but not destroyed by hardware decoherence and readout error, is the inductive bias the generator exploits; the dual-layout QPU-in-parallel scheme is what trades off entropy and residual correlation.

What would settle it

Train the same GANs with a classical Gaussian prior whose covariance matrix exactly equals the empirical covariance of the quantum-derived latents, using identical marginals, inverse-CDF smoothing, and the same random projection; if the FID matches the quantum-pool results, the claimed quantum-correlation advantage is not needed. A second check is to permute the bits within each qubit across samples, destroying qubit-wise correlations while preserving marginals, and see whether the FID gain disappears.

Watch

Extended reading notes

Core claim

The central discovery is that quantum-measurement noise, far from being a nuisance, can serve as a better generative prior than Gaussian white noise. The paper builds a pool of tens of millions of 16-bit strings from a single parameterized entangling circuit, sampled either on a noiseless simulator or on real hardware. Each latent vector is formed by taking $S$ repeated shots per qubit, interpreting them as a binary fraction, passing that fraction through the inverse Gaussian CDF to obtain a standard normal marginal per qubit, and applying a fixed row-orthonormal projection $P\in\mathbb{R}^{d_{\mathrm{latent}}\times 16}$. The resulting vectors have independent rows but a nonseparable joint copula inside each row, which the authors call quantum-correlated structure. Hybridizing them with the classical Gaussian via $z_{\mathrm{hyb}}=\alpha z_{\mathrm{quant}}+(1-\alpha)z_{\mathrm{class}}$ and training WGAN, SNGAN, and BigGAN on CIFAR-10 yields lower FID in every configuration tested, with hardware-derived pools beating the simulator and the dual-layout parallel QPU pool beating both.

Load-bearing premise

The load-bearing premise is that the only material difference between the quantum bitstring pool and the classical i.i.d. pool is genuine quantum correlation, so the FID gains are not side effects of a different covariance, marginal distribution, or smoothing.

Editorial extensions

If this is right

  • Any GAN that accepts a fixed-size latent vector can adopt the quantum-correlated prior as a drop-in replacement, with no changes to architecture or hyperparameters, since the bitstring pool is pre-sampled.
  • The FID curves decrease with $\alpha$, meaning the pure quantum prior performs best in the tested range.
  • Hardware noise outperforms a noiseless simulator, so on NISQ devices error suppression is not the only useful goal; retaining some controlled noise can be beneficial for generative tasks.
  • Running two distinct qubit layouts in parallel produces the best FID, up to 17% below baseline in BigGAN, so the practical bottleneck is sampling diversity and correlation control, not raw circuit fidelity.

Reading between the lines

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

  • The paper does not include a classical prior with the same covariance matrix as the quantum pool; a reader should treat the attribution to quantum correlations as provisional until such a matched control is tested.
  • The binary-fraction-plus-inverse-CDF construction is a general smoother for discrete noise: any source of bitstrings with a nontrivial copula, classical or quantum, could be fed through the same pipeline, so the method may extend beyond quantum sampling.
  • The optimal device noise level appears architecture-dependent, with the lower-error device winning for WGAN while the higher-error device wins for SNGAN, suggesting that the mixing weight and error rate should be tuned per model.
  • Extending the circuit to more qubits or to time-correlated measurement patterns could probe whether the FID gain scales with the dimensionality of the quantum copula; the paper does not test this.
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 paper proposes replacing or hybridizing the standard i.i.d. Gaussian latent prior of GANs with a quantum-correlated prior built from 16-qubit measurement bitstrings. Bitstrings are grouped into blocks, converted to binary fractions, passed through the inverse Gaussian CDF, and projected with a fixed random matrix (Eqs. 1-3). The authors evaluate this prior on WGAN, SNGAN, and BigGAN on CIFAR-10, reporting FID improvements over the Gaussian baseline and over an i.i.d. classical bitstring pool, with the largest gains from a dual-layout QPU-parallel execution on ibm-kingston. The central claim is that hardware noise provides a structured inductive bias and that the extra improvement over i.i.d. classical is due to unique quantum correlations.

Significance. The empirical finding that structured, non-Gaussian latent priors improve FID across three GAN architectures is of potential practical interest, and the paper is commendable for releasing code and data and for evaluating on a standard benchmark (CIFAR-10) with a standard metric (FID). The strongest result, the QPU-parallel prior, is clear and reproducible in its broad strokes. However, the central attribution to quantum correlations is not established by the current controls: the QPU-parallel and i.i.d.-classical pools differ in covariance, and no classical prior with matched covariance is tested. If the covariance-matched control also yields the same FID improvement, the paper's contribution reduces to using structured priors, not quantum advantage, which changes the claims substantially.

major comments (4)
  1. [Section 4, Table 2 vs Table 1, Section 3.2, Eqs. (1)-(2)] The claim that 'the extra improvement is attributed to the unique quantum correlations existing in the bitstrings' (Section 4) is not supported by the experiments as reported. Table 1 shows that the QPU-parallel pool has a nonzero mean pairwise bit correlation of 0.0019 ± 0.0052, whereas the I.I.D.-classical pool has zero correlation by construction. Through Eqs. (1)-(2), these bit correlations propagate into a nonzero covariance of X and hence of zquant; neither the Gaussian baseline N(0,I) nor the I.I.D.-classical arm reproduces this covariance. The observed FID gap between QPU-parallel and I.I.D.-classical could therefore be fully explained by a classical prior with the same low-rank covariance matrix, without invoking entanglement or hardware-specific quantum noise. Please add a control that matches the covariance of the quantum pool (e.g., a Gaussian with the empirical covariance of zquant, or a classically generated bitstring pool with the same pairwise marginals) and re-evaluate the attribution.
  2. [Section 2 and Section 3.3] The latent construction depends on the block size S in Eq. (1), the randomly initialized and fixed rotation angles of the 16-qubit circuit (Section 2), and the fixed random row-orthonormal matrix P in Eq. (2), but none of these are specified in the manuscript or the appendix. Because S controls the coarse-graining of the bitstrings and P controls the mixing of the 16 dimensions into the latent space, the exact priors used in Table 2 are not reproducible from the text. Please report S, the circuit angle seeds (or provide the circuit parameters in a supplemental file), and the construction of P, and state whether these were held fixed across all experiments.
  3. [Section 4, Figure 7, Table 2] The value of the mixing coefficient α used to produce the Table 2 results is not reported. Figure 7 shows an α-sweep only for BigGAN, with FID decreasing as α increases and the best result at α=1, but the text does not state whether WGAN and SNGAN use α=1, a single shared α, or per-model selected α. If α was chosen after inspecting FID, the reported numbers are selected optima rather than predictions; please specify the α selection protocol and, ideally, report FID for a fixed α across all models and devices.
  4. [Abstract and Section 2] The abstract describes the latent as having a 'joint copula [that] reflects genuine quantum entanglement.' No entanglement witness or Bell-type test is performed on the measured bitstrings; the nonzero pairwise correlations in Table 1 could in principle be generated by classical common-cause noise, readout crosstalk, or other device imperfections that are not genuinely quantum. At minimum, the wording should be softened to 'quantum-derived correlations' unless a quantum correlation test is added, and the covariance-matched classical control in the first major comment should be used to support the quantum-specific claim.
minor comments (5)
  1. [Section 3.3] CIFAR-10 contains 50,000 training and 10,000 test images; please clarify which set is used as the reference distribution for the reported FID values.
  2. [Section 4] The claim of a 17.20% FID improvement for BigGAN in Section 4 appears to correspond to (4.06−3.37)/4.06 ≈ 17.0%; please verify the arithmetic and report the exact computation.
  3. [Section 3.2 and Figure 3] Please report the physical qubit indices and error rates of the two mapomatic-selected layouts shown in Figure 3, since the dual-layout construction is central to the QPU-parallel result.
  4. [Section 4 and Figure 6] The t-SNE analysis is purely qualitative; please provide a quantitative distribution-distance measure (e.g., MMD) or state explicitly that the plots are illustrative only.
  5. [Section 2 and Eq. (3)] The hybrid prior is not renormalized, so its variance changes with α; please state whether any normalization was applied during training and discuss the potential effect on the comparison.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the FID results are external-benchmark observations, though attribution to quantum correlations is under-controlled.

full rationale

The paper's derivation chain is: sample bitstrings from a fixed circuit or from hardware; transform them through Eqs. (1)-(2) into a Gaussian-like latent; optionally mix with a Gaussian via Eq. (3); train standard GANs with unchanged hyperparameters; and evaluate with the external FID metric on 50,000 generated images. No step uses the FID outcome as an input, and no parameter is fitted to the benchmark: the mixing coefficient alpha is scanned over the full range, and the dual-layout QPU configuration is motivated by measured error rates and bit-correlation statistics, not by achieved FID. The relevant citations (Qiskit, StudioGAN, mapomatic) are external tooling, not self-citations carrying the argument. The claim that the QPU-parallel improvement is 'attributed to the unique quantum correlations' is causally under-supported because the I.I.D.-classical control differs from the QPU-parallel pool in covariance, which is visible in Table 1 and propagates through Eqs. (1)-(2); however, an incomplete control is a confound in interpretation, not a circular reduction of the prediction to its inputs. The FID numbers themselves remain independent empirical evidence.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim relies on four unverified premises: that Gaussianization preserves quantum correlations, that the measured correlations are uniquely quantum, that training differences are solely due to the prior, and that the rank-16 projection is not itself responsible for the gains. There are no new physical entities introduced.

free parameters (5)
  • Block size S (repeated shots per qubit) = not stated in paper
    Controls the resolution of the binary-fraction encoding in Eq. (1) and how independent the batch samples are; without a value the exact prior distribution is not reproducible.
  • Hybrid mixing coefficient alpha = alpha=1 reported as best
    Figure 7 sweeps alpha and the paper reports that the pure quantum prior yields the lowest FID; selection is made on test-set FID, not a held-out split.
  • Random circuit rotation angles (Rx, Ry, Rz) = randomly initialized once, seed not given
    Section 2: the angles define the specific correlation structure of the bitstring pool; they are fixed for all experiments but arbitrary.
  • Fixed random row-orthonormal projection matrix P = not specified
    Eq. (2): P maps the 16-dimensional Gaussianized vector into z_dim; because P is rank 16, even classical i.i.d. inputs become low-rank, a confound not controlled in the baseline comparison.
  • Choice of dual QPU layouts = not specified
    Section 3.2: mapomatic is used to pick a low-error and a high-error layout; this design choice determines the final bitstring pool and is not described with enough detail to reproduce exactly.
assumptions (4)
  • domain assumption Inverse Gaussian CDF applied to binary fractions yields Gaussian marginals while preserving quantum correlations.
    Section 2, Eq. (1)-(2): no verification that the transformed 16-dimensional vector is approximately Gaussian or that any non-classical dependence survives the nonlinearity and projection.
  • ad hoc to paper The measured bitstring correlations are genuinely quantum and cannot be reproduced by any classically generated bitstring pool.
    Section 3.1 and Table 1 are used to support this, but the paper provides no entanglement witness and no classical covariance-matched control; this is the main attribution assumption.
  • domain assumption FID differences across priors are caused by the prior alone.
    Section 3.3: all hyperparameters are kept fixed and three trials are run, but random seeds for initialization are not controlled and no significance tests are reported.
  • domain assumption A fixed random projection P preserves enough structure to benefit GAN training without dominating the comparison.
    Section 2, Eq. (2): P is row-orthonormal with rank 16, so even the classical control is a low-rank Gaussian; the paper does not analyze this effect.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Improving GANs by leveraging the quantum noise from real hardware." pith.science (2026). https://pith.science/paper/CJ6XEQH2

@misc{pith2026250701886,
  author       = {Pith},
  title        = {Pith review of: Improving GANs by leveraging the quantum noise from real hardware},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CJ6XEQH2}},
  note         = {Machine review of arXiv:2507.01886}
}
read the original abstract

We propose a novel approach to generative adversarial networks (GANs) in which the standard i.i.d. Gaussian latent prior is replaced or hybridized with a quantum-correlated prior derived from measurements of a 16-qubit entangling circuit. Each latent sample is generated by grouping repeated shots per qubit into a binary fraction, applying the inverse Gaussian CDF to obtain a 16-dimensional Gaussian vector whose joint copula reflects genuine quantum entanglement, and then projecting into the high-dimensional space via a fixed random matrix. By pre-sampling tens of millions of bitstrings, either from a noiseless simulator or from IBM hardware, we build large pools of independent but internally quantum-correlated latents. We integrate this prior into three representative architectures (WGAN, SNGAN, BigGAN) on CIFAR-10, making no changes to the neural network structure or training hyperparameters. The hybrid latent representations incorporating hardware-derived noise consistently lower the FID relative to both the classical baseline and the simulator variant, especially when the quantum component constitutes a substantial fraction of the prior. In addition, we execute on the QPU in parallel to not only save computing time but also further decrease the FID up to 17% in BigGAN. These results indicate that intrinsic quantum randomness and device-specific imperfections can provide a structured inductive bias that enhances GAN performance. Our work demonstrates a practical pipeline for leveraging noisy quantum hardware to enrich deep-generative modeling, opening a new interface between quantum information and machine learning. All code and data are available at https://github.com/Neon8988/GAN_QN.git.

Figures

Figures reproduced from arXiv: 2507.01886 by the authors.

Figure 1
Figure 1. The bistrings frequency distribution across multiple devices. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. The qubit correlation analysis across multiple devices. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. The physical layout of ibm-kingston. Red: layout 1(low error rate), blue:layout 2(high [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: The bitstring frequency distribution for QPU execution in parallel on ibm-kingston. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: The qubit correlation analysis for QPU execution in parallel on ibm-kingston. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: The t-SNE analysis across multiple devices in BigGAN. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: FID scores with different α across multiple devices in BigGAN. 8 [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

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. Quantum latent distributions in deep generative models

    cs.LG 2025-08 conditional novelty 6.0 of 10

    Quantum latent distributions from boson samplers are shown in theory to expand the output distribution class of invertible Lipschitz generators, and in GAN benchmarks on QM9 to beat Gaussian, Bernoulli, and distinguis...

Reference graph

Works this paper leans on

25 extracted references · 4 canonical work pages · cited by 1 Pith paper

  1. [5]

    Antonia Creswell, Tom White, Vincent Dumoulin, Kai Arulkumaran, Biswa Sengupta, and Anil A

    URL https://arxiv.org/abs/2005.13178. Antonia Creswell, Tom White, Vincent Dumoulin, Kai Arulkumaran, Biswa Sengupta, and Anil A. Bharath. Generative adversarial networks: An overview. IEEE Signal Processing Magazine , 35(1):53–65, January

    arXiv 2005

  2. [9]

    Christian Ledig, Lucas Theis, Ferenc Huszar, Jose Caballero, Andrew Cunningham, Alejandro Acosta, Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, and Wenzhe Shi

    URL https://arxiv.org/abs/2209.15352. Christian Ledig, Lucas Theis, Ferenc Huszar, Jose Caballero, Andrew Cunningham, Alejandro Acosta, Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, and Wenzhe Shi. Photo- realistic single image super-resolution using a generative adversarial network. arXiv preprint arXiv:1609.04802,

    arXiv

  3. [11]

    Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A

    URL https: //arxiv.org/abs/2410.20657. Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A. Efros. Unpaired image-to-image translation using cycle-consistent adversarial networks. InProceedings of the IEEE International Conference on Computer Vision (ICCV), Oct

    arXiv

  4. [13]

    Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida

    URL https: //arxiv.org/abs/1704.00028. Takeru Miyato, Toshiki Kataoka, Masanori Koyama, and Yuichi Yoshida. Spectral normalization for generative adversarial networks. arXiv preprint arXiv:1802.05957 ,

    arXiv

  5. [14]

    Andrew Brock, Jeff Donahue, and Karen Simonyan

    URL https:// arxiv.org/abs/1802.05957. Andrew Brock, Jeff Donahue, and Karen Simonyan. Large scale gan training for high fidelity natural image synthesis. arXiv preprint arXiv:1809.11096 ,

    arXiv

  6. [15]

    Tero Karras, Samuli Laine, and Timo Aila

    URL https://arxiv.org/ abs/1809.11096. Tero Karras, Samuli Laine, and Timo Aila. A style-based generator architecture for generative adversarial networks. arXiv preprint arXiv:1812.04948,

    arXiv

  7. [16]

    Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, and Timo Aila

    URL https://arxiv.org/abs/ 1812.04948. Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, and Timo Aila. Analyz- ing and improving the image quality of stylegan. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June

    arXiv

  8. [17]

    Stylegan-xl: Scaling stylegan to large di- verse datasets

    Axel Sauer, Katja Schwarz, and Andreas Geiger. Stylegan-xl: Scaling stylegan to large di- verse datasets. In ACM SIGGRAPH 2022 Conference Proceedings , SIGGRAPH ’22, New York, NY , USA,

    work page 2022

Show all 25 references
  1. [19]

    doi: 10.1103/physrevlett.121.040502

    ISSN 1079-7114. doi: 10.1103/physrevlett.121.040502. URL http: //dx.doi.org/10.1103/PhysRevLett.121.040502. Pierre-Luc Dallaire-Demers and Nathan Killoran. Quantum generative adversarial networks. Phys- ical Review A , 98(1), July

    doi

  2. [20]

    doi: 10.1103/physreva.98.012324

    ISSN 2469-9934. doi: 10.1103/physreva.98.012324. URL http://dx.doi.org/10.1103/PhysRevA.98.012324. Ling Hu, Shu-Hao Wu, Weizhou Cai, Yuwei Ma, Xianghao Mu, Yuan Xu, Haiyan Wang, Yipu Song, Dong-Ling Deng, Chang-Ling Zou, and Luyan Sun. Quantum generative adversarial learning i...

    doi

  3. [21]

    doi: 10.1126/sciadv. aav2761. URL https://www.science.org/doi/abs/10.1126/sciadv.aav2761. He-Liang Huang, Yuxuan Du, Ming Gong, Youwei Zhao, Yulin Wu, Chaoyue Wang, Shaowei Li, Futian Liang, Jin Lin, Yu Xu, Rui Yang, Tongliang Liu, Min-Hsiu Hsieh, Hui Deng, Hao Rong, Cheng-Zhi...

    doi

  4. [23]

    doi: 10.1364/ opticaq.530346

    ISSN 2837-6714. doi: 10.1364/ opticaq.530346. URL http://dx.doi.org/10.1364/OPTICAQ.530346. Samuel A. Stein, Betis Baheri, Daniel Chen, Ying Mao, Qiang Guan, Ang Li, Bo Fang, and Shuai Xu. Qugan: A quantum state fidelity based generative adversarial network. In 2021 IEEE Inter...

    doi

  5. [24]

    Manuel S Rudolph, Sacha Lerch, Supanut Thanasilp, Oriel Kiss, Oxana Shaya, Sofia Vallecorsa, Michele Grossi, and Zoë Holmes

    doi: 10.1109/QCE52317.2021.00023. Manuel S Rudolph, Sacha Lerch, Supanut Thanasilp, Oriel Kiss, Oxana Shaya, Sofia Vallecorsa, Michele Grossi, and Zoë Holmes. Trainability barriers and opportunities in quantum generative modeling. npj Quantum Information, 10(1):116,

    2021

  6. [26]

    Runqiu Shu, Xusheng Xu, Man-Hong Yung, and Wei Cui

    URL https://arxiv.org/abs/2406.02668. Runqiu Shu, Xusheng Xu, Man-Hong Yung, and Wei Cui. Variational quantum circuits enhanced generative adversarial network. arXiv preprint arXiv:2402.01791,

    arXiv

  7. [27]

    org/abs/2402.01791

    URL https://arxiv. org/abs/2402.01791. Ali Javadi-Abhari, Matthew Treinish, Kevin Krsulich, Christopher J. Wood, Jake Lishman, Julien Gacon, Simon Martiel, Paul D. Nation, Lev S. Bishop, Andrew W. Cross, Blake R. Johnson, and Jay M. Gambetta. Quantum computing with qiskit. arX...

    arXiv

  8. [28]

    URL https://arxiv.org/abs/2405.08810. Paul D. Nation and Matthew Treinish. Suppressing quantum circuit errors due to system variability. PRX Quantum , 4:010327, Mar

    arXiv

  9. [29]

    URL https:// link.aps.org/doi/10.1103/PRXQuantum.4.010327

    doi: 10.1103/PRXQuantum.4.010327. URL https:// link.aps.org/doi/10.1103/PRXQuantum.4.010327. Minguk Kang, Joonghyuk Shin, and Jaesik Park. Studiogan: a taxonomy and benchmark of gans for image synthesis. IEEE Transactions on Pattern Analysis and Machine Intelligence , 45(12): ...

    doi

  10. [2014]

    Jie Gui, Zhenan Sun, Yonggang Wen, Dacheng Tao, and Jieping Ye

    URL https://arxiv.org/abs/1406.2661. Jie Gui, Zhenan Sun, Yonggang Wen, Dacheng Tao, and Jieping Ye. A review on generative adver- sarial networks: Algorithms, theory, and applications. arXiv preprint arXiv:2001.06937 ,

    2001 arXiv

  11. [2017]

    Yintai Ma, Diego Klabjan, and Jean Utke

    URL https://arxiv.org/abs/1609.04802. Yintai Ma, Diego Klabjan, and Jean Utke. Video to video generative adversarial network for few- shot learning based on policy gradient. arXiv preprint arXiv:2410.20657 ,

    arXiv

  12. [2018]

    doi: 10.1109/msp.2017.2765202

    ISSN 1558-0792. doi: 10.1109/msp.2017.2765202. URL http: //dx.doi.org/10.1109/MSP.2017.2765202. Angona Biswas, MD Abdullah Al Nasim, Al Imran, Anika Tabassum Sejuty, Fabliha Fairooz, Sai Puppala, and Sajedul Talukder. Generative adversarial networks for data augmentation. arXi...

    2017

  13. [2019]

    9 Felix Kreuk, Gabriel Synnaeve, Adam Polyak, Uriel Singer, Alexandre Défossez, Jade Copet, Devi Parikh, Yaniv Taigman, and Yossi Adi

    URL https://arxiv.org/abs/1802.04208. 9 Felix Kreuk, Gabriel Synnaeve, Adam Polyak, Uriel Singer, Alexandre Défossez, Jade Copet, Devi Parikh, Yaniv Taigman, and Yossi Adi. Audiogen: Textually guided audio generation. arXiv preprint arXiv:2209.15352,

    arXiv

  14. [2020]

    Tanujit Chakraborty, Ujjwal Reddy K S, Shraddha M Naik, Madhurima Panja, and Bayapureddy Manvitha

    URL https://arxiv.org/abs/2001.06937. Tanujit Chakraborty, Ujjwal Reddy K S, Shraddha M Naik, Madhurima Panja, and Bayapureddy Manvitha. Ten years of generative adversarial nets (gans): a survey of the state-of-the-art. Ma- chine Learning: Science and Technology , 5(1):011001, January

    2001 arXiv

  15. [2021]

    doi: 10.1103/physrevapplied.16.024051

    ISSN 2331-7019. doi: 10.1103/physrevapplied.16.024051. URL http://dx.doi.org/10.1103/PhysRevApplied.16.024051. 10 Tigran Sedrakyan and Alexia Salavrakos. Photonic quantum generative adversarial networks for classical data. Optica Quantum , 2(6):458, December

    doi

  16. [2022]

    ISBN 9781450393379

    Association for Computing Machinery. ISBN 9781450393379. doi: 10.1145/3528233.3530738. URL https://doi.org/10.1145/3528233.3530738. Seth Lloyd and Christian Weedbrook. Quantum generative adversarial learning. Physical Review Letters, 121(4), July

  17. [2023]

    Chris Donahue, Julian McAuley, and Miller Puckette

    URL https://arxiv.org/abs/2306.02019. Chris Donahue, Julian McAuley, and Miller Puckette. Adversarial audio synthesis. arXiv preprint arXiv:1802.04208,

    arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.