Pith. sign in

REVIEW 4 major objections 4 minor 89 references

Data-Dependent Generalization Bounds for Parameterized Quantum Models Under Noise

T0 review · 4 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read For a noisy parameterized quantum model, the generalization gap is bounded by a term built from the parameter-space volume and the quantum Fisher information matrix, with local and effective-dimension refinements that tighten the bound…

desk verdict Standard covering-number bound with a QFIM term, likely repairable, but the advertised data-dependent local refinement does not yield a valid PAC statement because the neighborhood is chosen from the same training data. read the letter →

arxiv 2412.11451 v3 pith:ZCEF5MZ5 submitted 2024-12-16 cs.LG

Bikram Khanal , Pablo Rivas This is my paper · ORCID
classification cs.LG MSC 68Q3281P68
keywords quantummachinelearninggeneralizationboundFisherinformationmatrixnoisychanneldepolarizingnoiseeffectivedimensionRademachercomplexityparameterizedcircuits
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper seeks to establish that the generalization error of a noisy quantum machine learning model is controlled by the geometry of its parameter space, not by the raw number of parameters alone. The main result, Theorem 5.1, bounds the expected risk by the empirical risk plus two terms that decay like $1/\sqrt{N}$: a complexity term built from the parameter-space volume $V_\Theta$, a lower bound $m$ on $\sqrt{\det F(\theta)}$ of the quantum Fisher information matrix, and a gradient bound $L_f^p$, together with the standard $3\sqrt{\log(2/\delta)/(2N)}$ confidence term. If the theorem is right, it gives near-term quantum models a quantitative, noise-aware handle on generalization, and the paper's local and effective-dimension refinements promise materially tighter bounds once training has settled into a well-conditioned region. The key quantity is the quantum Fisher information matrix: its determinant lower bound controls how many small balls are needed to cover the parameter space, which in turn bounds the model's Rademacher complexity.

What carries the argument

The object carrying the argument is the quantum Fisher information matrix (QFIM) and its determinant lower bound $m$. The proof chain has four links: a lower bound $\sqrt{\det F(\theta)}\ge m>0$ makes each Euclidean $\epsilon$-ball contain at least Fisher-metric volume $V_d\epsilon^d m$, which yields the covering-number estimate $\log N(\epsilon,\Theta)\le \log V_\Theta-\log V_d-\log m-d\log\epsilon$; Lipschitz continuity of the model in $\theta$ transfers Euclidean covers of $\Theta$ to covers of the function class; an entropy-integral argument converts covering numbers into a bound on empirical Rademacher complexity; and the standard Rademacher generalization inequality converts that complexity into a risk bound. The local refinement repeats the chain over a data-determined neighborhood with local constants $m_{\rm loc}$, $L^p_{f,\rm loc}$, and volume $V_{\Theta_{\rm loc}}$, and the effective-dimension refinement restricts attention to the subspace of QFIM eigenvalues above $\alpha$.

What would settle it

Compute the rank of the quantum Fisher information matrix along the training trajectory for the paper's 12-parameter circuit under depolarizing noise; if any trained parameter point has rank $< d$, the global uniform bound of Theorem 5.1 cannot hold over the full parameter space. A second direct check is numerical: for a grid of $\epsilon$ values, count the minimal number of Fisher-metric balls needed to cover $[-2\pi,2\pi]^{12}$ and compare with Lemma A.1's formula $\log N \le \log V_\Theta-\log V_d-\log m - d\log\epsilon$.

Watch

Extended reading notes

Core claim

The central claim is that for a $d$-parameter quantum model class whose noisy output takes the form $f_{\theta,p}(x)=\eta(p)f_\theta(x)$ with $\eta(0)=1$, under a $1$-Lipschitz loss, the expected risk $R(\theta)$ is uniformly bounded, with probability at least $1-\delta$, by $\widehat{R}_N(\theta) + 12\sqrt{\pi d}\,\exp(C'/d)/\sqrt{N} + 3\sqrt{\log(2/\delta)/(2N)}$, where $C'=\log V_\Theta-\log V_d-\log m+d\log L_f^p$. The paper reads this as a structured complexity measure: the parameter-space volume and the QFIM determinant lower bound determine the metric entropy of the model class, and the bound improves as data grows. After training, Corollary 5.2 restricts the same argument to a local neighborhood of the trained parameters where the QFIM stays well-conditioned, and Corollary 5.3 replaces $d$ by the effective dimension $d_{\rm eff}(\alpha)$, the number of QFIM eigenvalues above a threshold $\alpha$, yielding a tighter bound when many parameter directions are nearly inert. The numerical experiments with depolarizing noise on the Iris and reduced Digits binary tasks show the local bound tracking the observed generalization gap more closely than the global one, supporting the paper's conclusion that noise can stabilize the local Fisher geometry and thereby reduce effective complexity.

Load-bearing premise

The proof's load-bearing premise is that $\sqrt{\det F(\theta)}$ is bounded below by a positive constant $m$ uniformly over the whole parameter space; the paper enforces this only by shrinking to a data-dependent local neighborhood around the trained parameters, so the global version of the bound rests on a geometric condition that is not established for the full circuit.

Editorial extensions

If this is right

  • For a fixed parameter dimension $d$, the dominant complexity term falls as $1/\sqrt{N}$, so the bound predicts that more training data shrinks the generalization gap of a noisy quantum model at a predictable rate.
  • In the large-$d$ limit $\exp(C'/d)\to 1$, so the complexity scales roughly as $\sqrt{d}$; consequently, high-dimensional circuits remain viable provided the sample size grows with $d\exp(2C'/d)$.
  • Noise enters through the perturbation function $\eta(p)$ and through the stability it induces in the local QFIM; the paper's depolarizing-noise experiments show the local bound hovering closer to the measured generalization gap than the global bound.
  • Parameter directions with QFIM eigenvalues below the threshold $\alpha$ are effectively inert, so the effective dimension $d_{\rm eff}(\alpha)$, not the raw parameter count, is the quantity that should guide capacity control for a trained noisy model.

Reading between the lines

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

  • Beyond the paper, the same covering-number argument with the classical Fisher information matrix in place of the QFIM would yield an analogous data-dependent bound for classical neural networks, connecting flat directions in the loss landscape to generalization.
  • Beyond the paper, the bound supplies a concrete model-selection heuristic: among circuits with comparable training error, prefer the one with smaller effective dimension, because the predicted gap grows with $d\exp(C'/d)$.
  • Beyond the paper, a rank-aware regularization of the lower bound, replacing the global $m$ by an $\varepsilon$-shifted $m_\varepsilon$, could be tested to see whether the local bound stays valid at the rank-deficient QFIM points that appear in the paper's own 12-parameter circuits.
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 / 4 minor

Summary. The paper proposes a generalization bound for parameterized quantum models under noise, combining the volume of the parameter space, a lower bound on the determinant of the quantum Fisher information matrix (QFIM), and a Lipschitz/gradient bound to control Rademacher complexity. It then refines the bound to a local neighborhood around the trained parameters and to an effective-dimension subspace, and it reports numerical experiments on two-class Iris and Digits tasks with depolarizing noise. The central advertised contribution is the data-dependent local refinement of the bound.

Significance. The problem addressed is important: principled, geometry-aware generalization bounds for noisy NISQ models would be a useful addition to quantum machine learning theory. The paper connects the QFIM, parameter-space volume, and Rademacher complexity in a way that, if correct, would give a structured complexity measure beyond raw parameter count. The numerical study attempts to show that the local bound is tighter than the global bound in practical settings. However, the main proof is not sound as written: the covering-number lemma is not derived, the local bound is applied to a data-dependent set without the required uniform-convergence justification, and the effective-dimension corollary is not proved. These issues are load-bearing, so the advertised results are not established by the manuscript.

major comments (4)
  1. [Appendix A.2, Lemma A.1] The covering-number bound is not derived. The proof uses the lower bound sqrt(det F(theta)) >= m to assert that every epsilon-ball has volume at least V_d epsilon^d m, and then concludes N(epsilon, Theta, ||.||) <= V_Theta / (V_d epsilon^d m). This treats the Euclidean volume V_Theta as an upper bound on the total volume of Theta under the Fisher metric. Since no upper bound on sqrt(det F(theta)) is assumed, the Fisher-geometric volume of Theta can exceed V_Theta by an arbitrarily large factor, and the volume argument is invalid. Because Lemma A.1 supplies the log-covering bound used in Lemma B.1 and Theorem 5.1, this error undermines the main theorem as stated.
  2. [Corollary 5.2 and Section 6.1] The local bound is not a valid 1-delta PAC guarantee. The set Theta_loc = {theta : ||theta - theta_hat|| <= delta} is defined using the trained parameters theta_hat, and Section 6.1 explains that delta is chosen by a continuity-based procedure using the QFIM at theta_hat, with L^p_{f,loc} obtained by sampling gradients inside Theta_loc. All of these objects depend on the training sample D. Theorem 5.1 provides a uniform-convergence bound over a fixed parameter class, so it does not apply to a class Theta_loc(D) selected after seeing D. The proof in Corollary 5.2 merely states that the global theorem is restricted to the smaller set, but without a union bound or sample-splitting argument the stated 1-delta probability guarantee does not follow. This issue remains even if Lemma A.1 were repaired.
  3. [Corollary 5.3] The effective-dimension corollary is not proved. The proof says only that 'the argument parallels that of Corollary 5.2, simply substituting the effective dimension deff(alpha)', which is not a derivation. Moreover, the definition deff(alpha) = max{r : lambda_r(theta) >= alpha for all theta in Theta} is a worst-case threshold over the whole parameter space, not the effective dimension introduced in Section 4 via rank or inverse participation ratio; it can even be zero when all eigenvalues are below alpha. No argument is given for why this quantity can replace d in the Rademacher/covering analysis, so Eq. (35) is unsupported.
  4. [Appendix B.1, Lemma B.1 proof] The entropy-integral step is not well defined as written. The proof substitutes log N(epsilon, F_Theta, ||.||_{2,D}) <= C' - d log epsilon into Dudley's inequality and integrates epsilon from 0 to 1, assuming epsilon_max = 1. No justification is provided for this epsilon_max, and when C' is negative the integrand sqrt(C' - d log epsilon) is not real for epsilon < exp(C'/d). The claim that f_{theta,p}(x) is bounded in [0,1] is not among the assumptions of Theorem 5.1 and is not implied by the loss being [0,1]-valued. These gaps make the proof of the Rademacher bound incomplete.
minor comments (4)
  1. [Theorem 5.1 statement] The theorem states delta in [0,1) but the bound contains log(2/delta), which is undefined at delta = 0; the final sentence says 'for any delta > 0', which is the correct domain and should be used consistently.
  2. [Section 6.1] The numerical experiments do not verify the global assumption sqrt(det F(theta)) >= m > 0 on Theta = [-2pi,2pi]^12, and the paper does not report any check of this condition. For the two-qubit circuit, rank deficiency of the QFIM is plausible, so the plotted global bound is not demonstrated to be an upper bound for the tested models.
  3. [Declarations] The Data availability statement says the article is a systematic literature review and that no datasets were generated or analyzed, but the paper reports experiments on the Iris and Digits datasets. This statement should be corrected.
  4. [Notation] There are several typos and notational slips, e.g., 'thetak' for theta_k in the caption of Fig. 2, the reference to 'Appendix 8' in the Introduction (the proofs are in Appendices A and B), and inconsistent use of V_Theta vs. V_Theta_loc. These should be cleaned up.

Circularity Check

1 steps flagged · score 5.0 of 10

Corollary 5.2's local bound selects its parameter neighborhood and constants from the same training data, so its 1−δ guarantee does not follow from Theorem 5.1; the experimentally reported tightness is partly a post-hoc fit rather than an independent prediction.

  1. fitted input called prediction [Section 5, Corollary 5.2 (Eq. 34); Section 6.1, Methodology]
    "Suppose after training, θ̂ is a solution, and consider a local neighborhood Θloc ⊂ Θ around θ̂ as Θloc := {θ ∈ Θ : ∥θ − θ̂∥ ≤ δ} ... Then, with probability at least 1 − δ, for all θ ∈ Θloc: ... (34). Proof. This follows from Theorem 5.1 by restricting the parameter space to the smaller, better-conditioned subset Θloc."

    Theorem 5.1's 1−δ statement is a uniform-convergence guarantee over a fixed parameter space Θ chosen before the data are drawn. Corollary 5.2 replaces Θ with Θloc, which is defined through θ̂, and θ̂ is a function of the training sample D. Section 6.1 then fixes δ by a continuity-based procedure evaluated at θ̂ and samples gradients inside Θloc to set L^p_{f,loc}; V_{Θloc}, m_loc, and L^p_{f,loc} are therefore all fitted to the same data used to compute the empirical risk. No union bound or sample-splitting argument is supplied to justify applying Theorem 5.1 to the random set Θloc(D). Consequently Eq.

full rationale

The global bound in Theorem 5.1 is derived from explicit assumptions (Lipschitz loss, gradient bound, and a lower bound on sqrt(det F(θ))) through a standard covering-number plus Rademacher/Dudley argument; no fitted parameter is renamed as a prediction in that derivation. The central circularity-type defect is confined to the advertised local refinement: Corollary 5.2 imports a data-dependent neighbourhood and data-dependent constants into a theorem that requires a fixed hypothesis class, and Section 6.1 then evaluates the bound using those fitted quantities, so the reported tightness of the local bound is partly an artifact of the fitting procedure. The paper's self-citations (refs. 6, 21, 34, 65) are used for dataset selection, noise modelling, and depolarization details, and are not load-bearing for the main theorem. Separate technical concerns, such as the direction of the volume-based covering-number inequality in Lemma A.1, are correctness issues rather than circularity and do not affect this score. Overall, the main derivation is self-contained, but the local bound and its empirical validation are partially fitted, giving a score of 5.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central derivation, if repaired by a plain Euclidean covering argument, needs no QFIM at all; as written it rests on an unflagged geometric assumption (Fisher volume bounded by Euclidean volume) and a global determinant condition that is implausible for the tested circuits and is only enforced locally. The theorem's noise model is a chosen multiplicative form that contradicts the affine depolarizing map used in the experiments. The local refinement and effective dimension are driven by parameters (delta, alpha, m_loc, L^p_f,loc) fitted to the trained model rather than derived.

free parameters (5)
  • m (sqrt(det QFIM) lower bound) = not reported
    Assumed positive in Theorem 5.1; in Section 6.1 the local neighborhood is chosen so the FIM determinant stays above a fraction (alpha = 0.5) of its value at theta-hat, and m_loc is then taken from this measured quantity.
  • L^p_f / L^p_f,loc (gradient and Lipschitz bound) = not reported
    In the numerics, the local value is estimated by sampling gradients inside Theta_loc and taking the maximum norm (Section 6.1); it feeds directly into C' and C_loc.
  • delta (local neighborhood radius) = chosen by continuity-based procedure with alpha = 0.5
    Section 6.1: the radius is chosen so the QFIM determinant stays above half its value at theta-hat; the choice is data-dependent and has no closed form.
  • alpha (QFIM eigenvalue threshold for d_eff) = 0.5
    Used in Corollary 5.3 and Section 6.1 to define which eigenvalues count toward the effective dimension; the value is arbitrary.
  • eta(p) noise scaling function = depolarizing: (1-p) f + p/2^n (affine, not multiplicative)
    Modeling choice in Theorem 5.1 and Eq. (16); the multiplicative form f_theta,p = eta(p) f_theta does not match the affine depolarizing transformation used in the experiments.
assumptions (6)
  • domain assumption sqrt(det F(theta)) >= m > 0 for all theta in Theta (QFIM non-degenerate)
    Theorem 5.1 assumption; fails for rank-deficient QFIMs typical of the paper's circuits; enforced only by restricting to local regions in Section 6.1.
  • ad hoc to paper The Fisher volume of Theta is bounded by its Euclidean volume (implied by Lemma A.1's covering argument)
    Appendix A.2 uses V_Theta as the total volume to be covered while measuring ball volume with sqrt(det F) factors; this requires an unstated upper bound on QFIM eigenvalues.
  • ad hoc to paper Noisy output factorizes as f_theta,p(x) = eta(p) f_theta(x)
    Theorem 5.1 and Section 2.2.6; for depolarizing noise the correct relation is affine, p-tilde = (1-p) p + p/2^n, so the theorem's model class does not match the experimental channel.
  • standard math Loss bounded in [0,1] and Lipschitz with constant <= 1
    Theorem B.2; a standard assumption in Rademacher generalization bounds.
  • standard math Standard learning theory: Dudley's entropy integral, Rademacher generalization inequality, McDiarmid-type concentration
    Appendix B; classical results used without proof.
  • ad hoc to paper Local neighborhood Theta_loc and its constants are fixed before the 1-delta statement
    Corollary 5.2 needs Theta_loc, m_loc, and L^p_f,loc to be data-independent for the PAC statement; the paper instead selects them after training using the same data.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Data-Dependent Generalization Bounds for Parameterized Quantum Models Under Noise." pith.science (2026). https://pith.science/paper/ZCEF5MZ5

@misc{pith2026241211451,
  author       = {Pith},
  title        = {Pith review of: Data-Dependent Generalization Bounds for Parameterized Quantum Models Under Noise},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZCEF5MZ5}},
  note         = {Machine review of arXiv:2412.11451}
}
read the original abstract

Quantum machine learning offers a transformative approach to solving complex problems, but the inherent noise hinders its practical implementation in near-term quantum devices. This obstacle makes it difficult to understand the generalizability of quantum circuit models. Designing robust quantum machine learning models under noise requires a principled understanding of complexity and generalization, extending beyond classical capacity measures. This study investigates the generalization properties of parameterized quantum machine learning models under the influence of noise. We present a data-dependent generalization bound grounded in the quantum Fisher information matrix. We leverage statistical learning theory to relate the parameter space volumes and training sizes to estimate the generalization capability of the trained model. We provide a structured characterization of complexity in quantum models by integrating local parameter neighborhoods and effective dimensions defined through quantum Fisher information matrix eigenvalues. We also analyze the tightness of the bound and discuss the tradeoff between model expressiveness and generalization performance.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

89 extracted references · 58 canonical work pages

  1. [1]

    Quantum Science and Technology

    Schuld, M., Petruccione, F.: Machine Learning with Quantum Computers, 2nd edn. Quantum Science and Technology. Springer, Cham, Switzerland (2021). https://doi.org/10.1007/978-3-030-83098-4

    doi:10.1007/978-3-030-83098-4 2021

  2. [2]

    Nature 549(7671), 195–202 (2017)

    Biamonte, J., Wittek, P., Pancotti, N., Rebentrost, P., Wiebe, N., Lloyd, S.: Quantum machine learning. Nature 549(7671), 195–202 (2017)

    2017

  3. [3]

    arXiv preprint arXiv:2401.11351 (2024)

    Wang, Y., Liu, J.: A comprehensive review of quantum machine learning: from nisq to fault tolerance. arXiv preprint arXiv:2401.11351 (2024)

    arXiv 2024

  4. [4]

    Quantum 2, 79 (2018)

    Preskill, J.: Quantum computing in the nisq era and beyond. Quantum 2, 79 (2018)

    2018

  5. [5]

    Annual Review of Condensed Matter Physics 11(1), 325–344 (2020)

    Torlai, G., Melko, R.G.: Machine-learning quantum states in the nisq era. Annual Review of Condensed Matter Physics 11(1), 325–344 (2020)

    2020

  6. [6]

    Quantum Machine Intelligence 6(2), 1–20 (2024)

    Khanal, B., Rivas, P., Sanjel, A., Sooksatra, K., Quevedo, E., Rodriguez, A.: Generalization error bound for quantum machine learning in nisq era—a survey. Quantum Machine Intelligence 6(2), 1–20 (2024)

    2024

  7. [7]

    Nature 567(7747), 209–212 (2019)

    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 (2019)

    2019

  8. [8]

    : Variational quantum algorithms

    Cerezo, M., Arrasmith, A., Babbush, R., Benjamin, S.C., Endo, S., Fujii, K., McClean, J.R., Mitarai, K., Yuan, X., Cincio, L., et al. : Variational quantum algorithms. Nature Reviews Physics 3(9), 625–644 (2021)

    2021

Show all 89 references
  1. [9]

    Machine Learning and Knowledge Extraction 6(2), 944–964 (2024) 28

    Rivas, P., Orduz, J., Jui, T.D., DeCusatis, C., Khanal, B.: Quantum-enhanced representation learning: A quanvolutional autoencoder approach against ddos threats. Machine Learning and Knowledge Extraction 6(2), 944–964 (2024) 28

    2024

  2. [10]

    arXiv preprint arXiv:1307.0411 (2013)

    Lloyd, S., Rebentrost, P., Mohseni, M.: Quantum algorithms for supervised and unsupervised machine learning. arXiv preprint arXiv:1307.0411 (2013)

    2013 arXiv

  3. [11]

    In: Pro- ceedings of the 24th International Conference on Machine Learning, pp

    A ¨ ımeur, E., Brassard, G., Gambs, S.: Quantum clustering algorithms. In: Pro- ceedings of the 24th International Conference on Machine Learning, pp. 1–8 (2007)

    2007

  4. [12]

    Second Edition

    Mohri, M., Rostamizadeh, A., Talwalkar, A.: Foundations of Machine Learn- ing, (2018). Second Edition. https://mitpress.ublish.com/ebook/foundations-of- machine-learning–2-preview/7093/Cover

    2018

  5. [13]

    Abu-Mostafa, Y.S., Magdon-Ismail, M., Lin, H.T.: Learning from data: A short course (2012)

    2012

  6. [14]

    Shalev-Shwartz, S., Ben-David, S.: Understanding machine learning: From theory to algorithms (2014)

    2014

  7. [15]

    In: International Conference on Machine Learning, pp

    Emami, M., Sahraee-Ardakan, M., Pandit, P., Rangan, S., Fletcher, A.: Gener- alization error of generalized linear models in high dimensions. In: International Conference on Machine Learning, pp. 2892–2901 (2020). PMLR

    2020

  8. [16]

    In: Compressed Sensing and Its Applications: Third International MATHEON Conference 2017, pp

    Jakubovitz, D., Giryes, R., Rodrigues, M.R.: Generalization error in deep learning. In: Compressed Sensing and Its Applications: Third International MATHEON Conference 2017, pp. 153–193 (2019). Springer

    2019

  9. [17]

    Advances in neural information processing systems 12 (1999)

    Nadeau, C., Bengio, Y.: Inference for the generalization error. Advances in neural information processing systems 12 (1999)

    1999

  10. [18]

    PRX Quantum2(4), 040321 (2021)

    Banchi, L., Pereira, J., Pirandola, S.: Generalization in quantum machine learning: A quantum information standpoint. PRX Quantum2(4), 040321 (2021)

    2021

  11. [19]

    arXiv preprint arXiv:2306.13461 (2023)

    Gil-Fuster, E., Eisert, J., Bravo-Prieto, C.: Understanding quantum machine learning also requires rethinking generalization. arXiv preprint arXiv:2306.13461 (2023)

    2023 arXiv

  12. [20]

    Quantum5, 582 (2021)

    Caro, M.C., Gil-Fuster, E., Meyer, J.J., Eisert, J., Sweke, R.: Encoding-dependent generalization bounds for parametrized quantum circuits. Quantum5, 582 (2021)

    2021

  13. [21]

    In: 2023 Congress in Computer Science, Computer Engineering, & Applied Computing (CSCE) (2023)

    Khanal, B., Rivas, P.: Evaluating the impact of noise on variational quantum circuits in nisq era devices. In: 2023 Congress in Computer Science, Computer Engineering, & Applied Computing (CSCE) (2023)

    2023

  14. [22]

    Nature communications 13(1), 4919 (2022)

    Caro, M.C., Huang, H.-Y., Cerezo, M., Sharma, K., Sornborger, A., Cincio, L., Coles, P.J.: Generalization in quantum machine learning from few training data. Nature communications 13(1), 4919 (2022)

    2022

  15. [23]

    In: The Thirty 29 Seventh Annual Conference on Learning Theory, pp

    Caro, M.C., Gur, T., Rouz´ e, C., Franca, D.S., Subramanian, S.: Information- theoretic generalization bounds for learning from quantum data. In: The Thirty 29 Seventh Annual Conference on Learning Theory, pp. 775–839 (2024). PMLR

    2024

  16. [24]

    arXiv preprint arXiv:2303.13462 (2023)

    Haug, T., Kim, M.: Generalization with quantum geometry for learning unitaries. arXiv preprint arXiv:2303.13462 (2023)

    2023 arXiv

  17. [25]

    arXiv preprint arXiv:2206.06686 (2022)

    Canatar, A., Peters, E., Pehlevan, C., Wild, S.M., Shaydulin, R.: Bandwidth enables generalization in quantum kernel models. arXiv preprint arXiv:2206.06686 (2022)

    2022 arXiv

  18. [26]

    Nature Communications 14(1), 3751 (2023)

    Caro, M.C., Huang, H.-Y., Ezzell, N., Gibbs, J., Sornborger, A.T., Cincio, L., Coles, P.J., Holmes, Z.: Out-of-distribution generalization for learning quantum dynamics. Nature Communications 14(1), 3751 (2023)

    2023

  19. [27]

    Nature Computational Science1(6), 403–409 (2021)

    Abbas, A., Sutter, D., Zoufal, C., Lucchi, A., Figalli, A., Woerner, S.: The power of quantum neural networks. Nature Computational Science1(6), 403–409 (2021)

    2021

  20. [28]

    Quantum Science and Technology 8(2), 025013 (2023)

    Bu, K., Koh, D.E., Li, L., Luo, Q., Zhang, Y.: Effects of quantum resources and noise on the statistical complexity of quantum circuits. Quantum Science and Technology 8(2), 025013 (2023)

    2023

  21. [29]

    Physical Review A 105(6), 062431 (2022)

    Bu, K., Koh, D.E., Li, L., Luo, Q., Zhang, Y.: Statistical complexity of quantum circuits. Physical Review A 105(6), 062431 (2022)

    2022

  22. [30]

    Physical Review B 67(9), 094510 (2003)

    Martinis, J.M., Nam, S., Aumentado, J., Lang, K., Urbina, C.: Decoherence of a superconducting qubit due to bias noise. Physical Review B 67(9), 094510 (2003)

    2003

  23. [31]

    Nature communications 12(1), 6961 (2021)

    Wang, S., Fontana, E., Cerezo, M., Sharma, K., Sone, A., Cincio, L., Coles, P.J.: Noise-induced barren plateaus in variational quantum algorithms. Nature communications 12(1), 6961 (2021)

    2021

  24. [32]

    Physical Review A 106(5), 052421 (2022)

    Heyraud, V., Li, Z., Denis, Z., Le Boit´ e, A., Ciuti, C.: Noisy quantum kernel machines. Physical Review A 106(5), 052421 (2022)

    2022

  25. [33]

    Physical review A 52(4), 2493 (1995)

    Shor, P.W.: Scheme for reducing decoherence in quantum computer memory. Physical review A 52(4), 2493 (1995)

    1995

  26. [34]

    arXiv preprint arXiv:2409.07632 (2024)

    Khanal, B., Rivas, P.: Learning robust observable to address noise in quantum machine learning. arXiv preprint arXiv:2409.07632 (2024)

    2024 arXiv

  27. [35]

    Scientific Reports 13(1), 3912 (2023)

    Shaib, A., Naim, M.H., Fouda, M.E., Kanj, R., Kurdahi, F.: Efficient noise mitigation technique for quantum computing. Scientific Reports 13(1), 3912 (2023)

    2023

  28. [36]

    Quantum 8, 1410 (2024)

    Ferracin, S., Hashim, A., Ville, J.-L., Naik, R., Carignan-Dugas, A., Qassim, H., Morvan, A., Santiago, D.I., Siddiqi, I., Wallman, J.J.: Efficiently improving the performance of noisy quantum computers. Quantum 8, 1410 (2024)

    2024

  29. [37]

    Quantum 8, 1225 (2024)

    Gentinetta, G., Thomsen, A., Sutter, D., Woerner, S.: The complexity of quantum 30 support vector machines. Quantum 8, 1225 (2024)

    2024

  30. [38]

    Nature Physics 17(9), 1013–1017 (2021)

    Liu, Y., Arunachalam, S., Temme, K.: A rigorous and robust quantum speed-up in supervised machine learning. Nature Physics 17(9), 1013–1017 (2021)

    2021

  31. [39]

    Advances in Neural Information Processing Systems 34, 12661–12673 (2021)

    K¨ ubler, J., Buchholz, S., Sch¨ olkopf, B.: The inductive bias of quantum kernels. Advances in Neural Information Processing Systems 34, 12661–12673 (2021)

    2021

  32. [40]

    Nature Communications 15(1), 5200 (2024)

    Thanasilp, S., Wang, S., Cerezo, M., Holmes, Z.: Exponential concentration in quantum kernel methods. Nature Communications 15(1), 5200 (2024)

    2024

  33. [41]

    Nature communica- tions 12(1), 2631 (2021)

    Huang, H.-Y., Broughton, M., Mohseni, M., Babbush, R., Boixo, S., Neven, H., McClean, J.R.: Power of data in quantum machine learning. Nature communica- tions 12(1), 2631 (2021)

    2021

  34. [42]

    Quantum 5, 592 (2021)

    Czarnik, P., Arrasmith, A., Coles, P.J., Cincio, L.: Error mitigation with clifford quantum-circuit data. Quantum 5, 592 (2021)

    2021

  35. [43]

    Nature Computational Science 2(9), 567–576 (2022)

    Cerezo, M., Verdon, G., Huang, H.-Y., Cincio, L., Coles, P.J.: Challenges and opportunities in quantum machine learning. Nature Computational Science 2(9), 567–576 (2022)

    2022

  36. [44]

    PRX Quantum 3(1), 010313 (2022)

    Holmes, Z., Sharma, K., Cerezo, M., Coles, P.J.: Connecting ansatz expressibility to gradient magnitudes and barren plateaus. PRX Quantum 3(1), 010313 (2022)

    2022

  37. [45]

    Quantum 5, 466 (2021)

    Zhao, C., Gao, X.-S.: Analyzing the barren plateau phenomenon in training quantum neural networks with the zx-calculus. Quantum 5, 466 (2021)

    2021

  38. [46]

    Nature communications 9(1), 4812 (2018)

    McClean, J.R., Boixo, S., Smelyanskiy, V.N., Babbush, R., Neven, H.: Barren plateaus in quantum neural network training landscapes. Nature communications 9(1), 4812 (2018)

    2018

  39. [47]

    Quantum 5, 558 (2021)

    Arrasmith, A., Cerezo, M., Czarnik, P., Cincio, L., Coles, P.J.: Effect of barren plateaus on gradient-free optimization. Quantum 5, 558 (2021)

    2021

  40. [48]

    American Association of Physics Teachers (2002)

    Nielsen, M.A., Chuang, I.: Quantum computation and quantum information. American Association of Physics Teachers (2002)

    2002

  41. [49]

    The Journal of Machine Learning Research 11, 2635– 2670 (2010)

    Shalev-Shwartz, S., Shamir, O., Srebro, N., Sridharan, K.: Learnability, stability and uniform convergence. The Journal of Machine Learning Research 11, 2635– 2670 (2010)

    2010

  42. [50]

    arXiv preprint arXiv:2012.04115 (2020)

    Valle-P´ erez, G., Louis, A.A.: Generalization bounds for deep learning. arXiv preprint arXiv:2012.04115 (2020)

    2020 arXiv

  43. [51]

    Journal of Machine Learning Research 23(166), 1–50 (2022) 31

    Johansson, F.D., Shalit, U., Kallus, N., Sontag, D.: Generalization bounds and representation learning for estimation of potential outcomes and causal effects. Journal of Machine Learning Research 23(166), 1–50 (2022) 31

    2022

  44. [52]

    Journal of Mathematical Psychology 64, 66–75 (2015)

    Pape, A.D., Kurtz, K.J., Sayama, H.: Complexity measures and concept learning. Journal of Mathematical Psychology 64, 66–75 (2015)

    2015

  45. [53]

    Neural computation 13(11), 2409–2463 (2001)

    Bialek, W., Nemenman, I., Tishby, N.: Predictability, complexity, and learning. Neural computation 13(11), 2409–2463 (2001)

    2001

  46. [54]

    Physical review letters 122(4), 040504 (2019)

    Schuld, M., Killoran, N.: Quantum machine learning in feature hilbert spaces. Physical review letters 122(4), 040504 (2019)

    2019

  47. [55]

    Physical Review A 98(3), 032309 (2018)

    Mitarai, K., Negoro, M., Kitagawa, M., Fujii, K.: Quantum circuit learning. Physical Review A 98(3), 032309 (2018)

    2018

  48. [56]

    Physical Review A 99(3), 032331 (2019)

    Schuld, M., Bergholm, V., Gogolin, C., Izaac, J., Killoran, N.: Evaluating analytic gradients on quantum hardware. Physical Review A 99(3), 032331 (2019)

    2019

  49. [57]

    Physical Review A 103(3), 032430 (2021)

    Schuld, M., Sweke, R., Meyer, J.J.: Effect of data encoding on the expres- sive power of variational quantum-machine-learning models. Physical Review A 103(3), 032430 (2021)

    2021

  50. [58]

    Quantum Science and Technology 4(4), 043001 (2019)

    Benedetti, M., Lloyd, E., Sack, S., Fiorentini, M.: Parameterized quantum cir- cuits as machine learning models. Quantum Science and Technology 4(4), 043001 (2019)

    2019

  51. [59]

    Scientific Reports 9(1), 12679 (2019)

    Gyongyosi, L., Imre, S.: Training optimization for gate-model quantum neural networks. Scientific Reports 9(1), 12679 (2019)

    2019

  52. [60]

    arXiv preprint arXiv:1905.13311 (2019)

    Crooks, G.E.: Gradients of parameterized quantum gates using the parameter- shift rule and gate decomposition. arXiv preprint arXiv:1905.13311 (2019)

    2019 arXiv

  53. [61]

    Walter de Gruyter GmbH & Co KG, ??? (2019)

    Holevo, A.S.: Quantum Systems, Channels, Information: a Mathematical Intro- duction. Walter de Gruyter GmbH & Co KG, ??? (2019)

    2019

  54. [62]

    Physical Review A—Atomic, Molecular, and Optical Physics 73(5), 052108 (2006)

    Gut ¸˘ a, M., Kahn, J.: Local asymptotic normality for qubit states. Physical Review A—Atomic, Molecular, and Optical Physics 73(5), 052108 (2006)

    2006

  55. [63]

    Physical review letters 106(18), 180504 (2011)

    Magesan, E., Gambetta, J.M., Emerson, J.: Scalable and robust randomized benchmarking of quantum processes. Physical review letters 106(18), 180504 (2011)

    2011

  56. [64]

    California institute of technology 16(1), 1–8 (1998)

    Preskill, J.: Lecture notes for physics 229: Quantum information and computation. California institute of technology 16(1), 1–8 (1998)

    1998

  57. [65]

    Mathematics 12(9), 1385 (2024)

    Khanal, B., Rivas, P.: A modified depolarization approach for efficient quantum machine learning. Mathematics 12(9), 1385 (2024)

    2024

  58. [66]

    Physical Review Letters 133(5), 050603 (2024) 32

    Haug, T., Kim, M.: Generalization of quantum machine learning models using quantum fisher information metric. Physical Review Letters 133(5), 050603 (2024) 32

    2024

  59. [67]

    Scientific reports 10(1), 11229 (2020)

    Gyongyosi, L., Imre, S.: Circuit depth reduction for gate-model quantum com- puters. Scientific reports 10(1), 11229 (2020)

    2020

  60. [68]

    Scientific reports 11(1), 5172 (2021)

    Gyongyosi, L., Imre, S.: Scalable distributed gate-model quantum computers. Scientific reports 11(1), 5172 (2021)

    2021

  61. [69]

    arXiv preprint arXiv:2408.03418 (2024)

    Kasatkin, V., Mozgunov, E., Ezzell, N., Lidar, D.: Detecting quantum and classi- cal phase transitions via unsupervised machine learning of the fisher information metric. arXiv preprint arXiv:2408.03418 (2024)

    2024 arXiv

  62. [70]

    Quantum Views 5, 61 (2021)

    Bharti, K.: Fisher information: A crucial tool for nisq research. Quantum Views 5, 61 (2021)

    2021

  63. [71]

    Quantum 5, 539 (2021)

    Meyer, J.J.: Fisher information in noisy intermediate-scale quantum applications. Quantum 5, 539 (2021)

    2021

  64. [72]

    Neural computation 10(2), 251–276 (1998)

    Amari, S.-I.: Natural gradient works efficiently in learning. Neural computation 10(2), 251–276 (1998)

    1998

  65. [73]

    PRX Quantum 2(4), 040309 (2021)

    Haug, T., Bharti, K., Kim, M.: Capacity and quantum geometry of parametrized quantum circuits. PRX Quantum 2(4), 040309 (2021)

    2021

  66. [74]

    Journal of Machine Learning Research 21(146), 1–76 (2020)

    Martens, J.: New insights and perspectives on the natural gradient method. Journal of Machine Learning Research 21(146), 1–76 (2020)

    2020

  67. [75]

    Quantum 4, 269 (2020)

    Stokes, J., Izaac, J., Killoran, N., Carleo, G.: Quantum natural gradient. Quantum 4, 269 (2020)

    2020

  68. [76]

    New Journal of Physics15(12), 125004 (2013)

    Baumgratz, T., N¨ ußeler, A., Cramer, M., Plenio, M.B.: A scalable maximum likelihood method for quantum state tomography. New Journal of Physics15(12), 125004 (2013)

    2013

  69. [77]

    Physical Review Letters 72(22), 3439 (1994)

    Braunstein, S.L., Caves, C.M.: Statistical distance and the geometry of quantum states. Physical Review Letters 72(22), 3439 (1994)

    1994

  70. [78]

    Journal of Physics A: Mathematical and Theoretical 53(2), 023001 (2020)

    Liu, J., Yuan, H., Lu, X.-M., Wang, X.: Quantum fisher information matrix and multiparameter estimation. Journal of Physics A: Mathematical and Theoretical 53(2), 023001 (2020)

    2020

  71. [79]

    Physical Review A 63(4), 042304 (2001)

    Fujiwara, A.: Quantum channel identification problem. Physical Review A 63(4), 042304 (2001)

    2001

  72. [80]

    In: Quantum Probability and Related Topics, pp

    Petz, D., Ghinea, C.: Introduction to quantum fisher information. In: Quantum Probability and Related Topics, pp. 261–281. World Scientific, ??? (2011)

    2011

  73. [81]

    arXiv preprint arXiv:1909.05074 (2019) 33

    Yamamoto, N.: On the natural gradient for variational quantum eigensolver. arXiv preprint arXiv:1909.05074 (2019) 33

    2019 arXiv

  74. [82]

    Physica A: Statistical Mechanics and its Applications 410, 167–173 (2014)

    Liu, J., Xiong, H.-N., Song, F., Wang, X.: Fidelity susceptibility and quan- tum fisher information for density operators with arbitrary ranks. Physica A: Statistical Mechanics and its Applications 410, 167–173 (2014)

    2014

  75. [83]

    arXiv preprint arXiv:2112.04807 (2021)

    Abbas, A., Sutter, D., Figalli, A., Woerner, S.: Effective dimension of machine learning models. arXiv preprint arXiv:2112.04807 (2021)

    2021 arXiv

  76. [84]

    John Wiley & Sons google schola 2, 831– 842 (1998)

    Vapnik, V.: Statistical learning theory. John Wiley & Sons google schola 2, 831– 842 (1998)

    1998

  77. [85]

    Journal of Statistical Physics 1, 231–252 (1969)

    Helstrom, C.W.: Quantum detection and estimation theory. Journal of Statistical Physics 1, 231–252 (1969)

    1969

  78. [86]

    International Journal of Quantum Information 7(supp01), 125–137 (2009)

    Paris, M.G.: Quantum estimation for quantum technology. International Journal of Quantum Information 7(supp01), 125–137 (2009)

    2009

  79. [87]

    Journal of Machine Learning Research 3(Nov), 463–482 (2002)

    Bartlett, P.L., Mendelson, S.: Rademacher and gaussian complexities: Risk bounds and structural results. Journal of Machine Learning Research 3(Nov), 463–482 (2002)

    2002

  80. [88]

    Nature Computational Science 3(6), 542–551 (2023)

    Larocca, M., Ju, N., Garc ´ ıa-Mart ´ ın, D., Coles, P.J., Cerezo, M.: Theory of over- parametrization in quantum neural networks. Nature Computational Science 3(6), 542–551 (2023)

    2023

  81. [89]

    Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences474(2209), 20170551 (2018) 34

    Ciliberto, C., Herbster, M., Ialongo, A.D., Pontil, M., Rocchetto, A., Severini, S., Wossnig, L.: Quantum machine learning: a classical perspective. Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences474(2209), 20170551 (2018) 34

    2018

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.