Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Is data-efficient learning feasible with quantum models?

T0 review · 3 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 Quantum models can learn from less data — when labels match the kernel

desk verdict Useful tool paper with an honest existence proof, but the headline data-efficiency claim is weaker than the abstract implies because the quantum kernel is handed the labels and the classical baseline is untuned. read the letter →

arxiv 2508.19437 v2 pith:UZHYHGBH submitted 2025-08-26 quant-ph cs.LG

Alona Sakhnenko , Christian B. Mendl , Jeanette M. Lorenz This is my paper
classification quant-phcs.LG MSC 68Q1281P6868T05 PACS 03.67.-a
keywords quantumkernelmethodsdata-efficientlearningtargetalignmentspectralbiasgeneralizationpredictionsemi-artificialdatasetsridgeregressionmachine
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 asks whether quantum models can learn from less data than classical models on classical datasets. It argues yes: by constructing semi-artificial labels that concentrate the target function in the top eigenmodes of a quantum kernel, quantum kernel methods reach low error with fewer training points than an untuned classical RBF kernel on three clinical datasets. The paper also transfers a spectral-bias generalization formula from classical kernel regression to quantum kernels and reports that its predicted loss tracks the measured learning curves. If correct, the contribution is an existence proof that data-efficient quantum learning on classical data is possible, plus a practical tool for designing datasets that favor quantum models and for screening real datasets for such opportunities.

What carries the argument

The engine is target-alignment saturation: a label-construction rule (Eqs. 14–15) that uses the quantum kernel's eigen-decomposition to place all target power in the first n eigenmodes, making the cumulative alignment measure C(i) rise as steeply as possible. This forces the task onto the modes the quantum kernel learns fastest. The paper couples this with the spectral-bias generalization formula (Eq. 7), which predicts per-mode and total error from kernel eigenvalues γ_i, ridge parameter λ, dataset size N, and target coefficients α̂_i.

What would settle it

Take the same three datasets and build labels from the classical RBF kernel's eigenbasis instead of the quantum kernel's, keeping everything else identical; if the quantum kernel still needs fewer points than RBF, the claim is not a construction artifact, but if RBF wins on its own engineered labels, the reported gap is an artifact of label construction. A second check: tune the RBF kernel's bandwidth per dataset and see whether the quantum advantage vanishes.

Watch

Extended reading notes

Core claim

The central claim is that the data-efficiency of a quantum kernel is governed by how well the target labels align with the kernel's eigenbasis, and that this alignment can be engineered. Concretely, the paper keeps the features of three clinical datasets fixed and replaces the labels with y~ = V_γ^{-T}√ĉ, where V_γ is the eigenvector matrix of the quantum kernel and ĉ is a step function placing nonzero target power only in the top n eigenmodes. On these relabeled datasets, the quantum kernel reaches low mean squared error with fewer points than the classical RBF kernel, and the paper calls this the first empirical evidence that data-efficient quantum kernel learning on classical data exists.

Load-bearing premise

The load-bearing premise is that constructing labels directly from the quantum kernel's eigenbasis and comparing against an untuned classical kernel is a fair test of data efficiency; if labels were chosen without knowing the quantum kernel, or the classical kernel were tuned, the data-efficiency gap could shrink or disappear.

Editorial extensions

If this is right

  • If the claim holds, quantum kernel methods can beat classical kernels on data efficiency only when the target labels align with the quantum kernel's high-eigenvalue modes; alignment, not raw expressivity, is the operative resource.
  • The data-generation tool yields tunable semi-artificial datasets (via cutoff n and scale x), enabling controlled studies of which dataset characteristics favor quantum models.
  • The validated Eq. (7) predictor lets practitioners estimate quantum kernel learning curves from a kernel spectrum and target coefficients before committing to large training runs.
  • Because supervised quantum models can be viewed as kernel methods, the same spectral-bias analysis may extend to quantum neural networks via neural tangent kernels, as the paper notes.
  • The existence proof redirects the search for quantum advantage from ad hoc benchmarks toward principled dataset design and label-kernel alignment screening.

Reading between the lines

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

  • The gap is manufactured in the paper's favor: labels are built from the quantum kernel's eigenbasis while the classical RBF baseline is left untuned, so the experiment demonstrates feasibility, not a generic advantage.
  • A natural next test would repeat the construction blind to both kernels, or with a tuned classical kernel; if the quantum gap survives, the conclusion strengthens, and if not, the observed data efficiency is partly an artifact of the construction.
  • The metric's predictive power is not yet general: Appendix A shows predicted loss diverging from empirical loss under the geometric-difference relabeling, so Eq. (7) needs validation on naturally labeled datasets before being used to screen for quantum advantage.
  • Inverting the tool suggests a practical screening procedure: compute the alignment measure C(i) on a real dataset and identify cases where a large fraction of target power sits in the top quantum-kernel eigenmodes; those are the datasets worth testing for quantum data efficiency.
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

3 major / 5 minor

Summary. The paper addresses whether quantum kernel methods (QKMs) can be more data-efficient than classical kernel methods on classical data. The authors adapt a spectral-bias generalization metric from Canatar et al. [8] to QKMs, and use it to construct 'semi-artificial' datasets: classical features are kept fixed, while new labels are generated as y~ = V_γ^{-T}√ĉ (Eq. 15), where V_γ is the eigenvector matrix of the quantum kernel Gram matrix and ĉ is a step function concentrating target power in the top n eigenmodes. On three clinical datasets, they report that QKMs achieve lower error with fewer training points than an RBF kernel, and that the predicted loss from Eq. (7) tracks empirical loss. The paper claims this is clear empirical evidence that data-efficient learning with quantum models is possible on classical data, and positions the label-generation procedure as a tool for systematic dataset design.

Significance. If the empirical claims held under an unbiased comparison, the paper would make a useful contribution: it would provide a concrete, tunable construction of datasets where quantum kernels appear to require fewer samples, and it would offer a validation of a practical generalization metric for QKMs. The paper also has strengths: it engages seriously with the spectral-bias literature, it clearly states its protocol, and it is unusually candid about limitations, including the admission in Section V that the procedure 'did not explicitly account for a classical kernel (RBF)' and the divergence between predicted and empirical loss in Appendix A. These candid statements, however, highlight that the central claim is currently not supported by the experimental design. The label-generation procedure forces the target to align with the QKM eigenbasis, so the observed advantage is expected from Eq. (10) and does not demonstrate a quantum-specific data-efficiency property.

major comments (3)
  1. [Section IV-A, Eq. (15), and Fig. 3] The central empirical claim is largely forced by construction. Equations (14)-(15) generate labels ỹ = V_γ^{-T}√ĉ, where V_γ are the eigenvectors of the quantum kernel Gram matrix and ĉ places all target power on the top n eigenmodes. Equation (10) states that modes with large γ_i are learned rapidly as N grows, so the QKM's low-error-at-small-N behavior is a direct consequence of the construction, not an emergent property of quantum models. The classical RBF baseline is used with fixed hyperparameters and, as the authors admit in Section V, the procedure 'did not explicitly account for a classical kernel (RBF)'. To support the paper's claim, the authors must provide a control where the same relabeling construction is applied to a classical kernel (e.g., labels built from the RBF eigenbasis), or where the classical kernel is tuned or otherwise given comparable target alignment. Without s
  2. [Section V and Appendix A] The validation of Eq. (7) as a predictive tool is also not yet convincing. The predicted loss is computed from the same γ and α that define the labels, so agreement between Eq. (7) and the QKM empirical loss is partly tautological. More importantly, Appendix A shows that for the alternative geometric-difference relabeling of Huang et al., the predicted loss 'shows a notable divergence from the empirical loss' (Fig. 4). Thus the metric's agreement in Fig. 3 is not evidence of general predictive validity for QKMs. The authors should either quantify the agreement (e.g., error bars, correlation, or normalized error) in the main construction, or explain why the divergence in Appendix A does not undermine the claim that Eq. (7) 'remains consistent with empirical behavior.'
  3. [Algorithm 1 and Fig. 3] The experimental reporting is under-specified. Algorithm 1 uses 'while n < size(X)' but n is not initialized, and the same symbol n is used for the hyperparameter in ĉ_x^n, creating ambiguity. The main plots (Fig. 3) show no error bars or confidence intervals despite the text stating that the experiment is repeated 50 times. Since the data-efficiency gap is the central quantitative claim, the spread across repetitions must be shown; otherwise the reader cannot assess whether the gaps in Fig. 3(a)-(d) are statistically meaningful.
minor comments (5)
  1. [Eq. (14)] The text says 'n and v are tunable hyperparameters', but the step function uses x, not v. The notation should be corrected (and the same for the description of Fig. 2).
  2. [Section III-B] Equation (11) defines C(i) as a cumulative distribution, but the text does not state whether the eigenvalues γ_i are sorted in descending order before this definition is applied. Since the construction of ĉ depends on this ordering, it should be made explicit.
  3. [Appendix B] The description of Fig. 5 says 'Relabeled - KM' and 'Relabeled - qKM' but the text states that the correlation values for ĉ_1^20 and ĉ_10^20 are identical. If the figure shows overlapping curves, this should be stated clearly, and the claim that x controls only scale is presented without quantitative support.
  4. [Abstract and Section I] The phrase 'one of the first evidence' (Section I) is grammatically awkward; more importantly, the abstract's 'clear empirical evidence' overstates what the current protocol demonstrates (see major comment above).
  5. [References] Reference [31] is formatted with the data-set contributor names as authors ('Kolby Nottingham Markelle Kelly Rachel Longjohn'). This should be corrected to the standard UCI repository citation.

Circularity Check

2 steps flagged · score 7.0 of 10

Quantum data-efficiency gap is by construction: labels are generated from the QKM eigenbasis (Eq. 15) and the predicted loss (Eq. 7) is evaluated on the same eigen-decomposition used to build the labels.

  1. self definitional [Section IV-A, Eqs. (14)-(15)]
    "Our method seeks to create an artificial target function that saturate the target-alignment measure C(i) from Eq. (11) for quantum kernel method. ... we create an artificial ˆc, which we define as a step-function: ˆc n x(i) = x if i ≤ n else 0 ... The new labels are assigned as ˜y = V −T γ √ ˆc."

    The labels are defined from the quantum kernel's Gram matrix eigenbasis Vγ and eigenvalues γ via the step-function ĉ. Since C(i) (Eq. 11) is the cumulative distribution of γ_i α_i^2, this construction places all target power in the top n QKM eigenmodes by definition. Eq. (10) then guarantees those modes are learned rapidly, so the QKM's low-error-at-small-N behavior is a direct consequence of the label construction, not an empirical discovery about quantum models. The classical RBF baseline is not part of the construction, and the paper admits in Section V that the procedure 'did not explicitly account for a classical kernel (RBF)', so the observed QKM-vs-RBF gap is an artifact of aligning labels to one kernel only.

  2. fitted input called prediction [Section V (Fig. 3) and Algorithm 1]
    "Epredicted ← Eq. (7)(λ, K,˜y) ... The overall trends of predicted loss from Eq. (7) remain consistent with empirical behavior."

    The 'predicted' loss is evaluated with the same K, γ, Vγ and ỹ that the label-generation step uses: ỹ = Vγ^{-T}√ĉ is defined from eig(K). Eq. (7) is a function of γ_i and α_i, exactly the quantities the construction saturates. Therefore the match between predicted and empirical loss largely restates the input alignment rather than validating the metric on independent data. Appendix A supplies a control experiment using Huang et al.'s geometric-difference relabeling in which the predicted loss 'shows a notable divergence from the empirical loss', confirming that the good agreement in Fig. 3 is tied to the self-aligned construction.

full rationale

The central result is not independent of its construction. The artificial labels (Eq. 15) are generated from the quantum kernel's own eigenvectors and eigenvalues so as to saturate C(i) (Eq. 11), and Eq. (10) then guarantees rapid learning of those modes; hence the QKM's data efficiency in Fig. 3 is by construction. The paper's own Section V concedes the procedure did not account for the RBF kernel, so the comparison does not control for the alignment mechanism. The predicted-loss validation is likewise self-referential: Eq. (7) is evaluated on the same K, γ and ỹ used to build the labels, so agreement restates the input. Appendix A shows the predicted loss diverges under Huang et al.'s geometric-difference labels, a limitation the paper acknowledges. No load-bearing self-citation issue was found; [8] and [24] are external/prior work and not used to forbid alternatives. Weighted by the fact that the paper's stated goal is a data-generation tool and an existence proof of aligned datasets, the core 'quantum data-efficiency' claim reduces to the construction, giving a circularity score of 7.

Assumptions & free parameters 5 free parameters · 5 assumptions · 1 invented entities

The central claim rests on the eigen-decomposition of the quantum kernel matrix and on the imported spectral-bias theory. Counted free parameters: the construction's step width n and height x, the ridge λ, the feature-map evolution time t, and the PCA dimension. Counted axioms: Mercer's spectral representation; the transfer of Eqs. (7)-(9) from [8] to finite QKM gram matrices; the approximation of the population operator by the empirical spectrum; the choice of Hamiltonian-evolution embedding; and the one-hot regression surrogate. One invented construct: the semi-artificial relabeled dataset; it carries no independent falsifiable handle, since its definition is the paper's own tool.

free parameters (5)
  • n (step width of target alignment ĉ) = 4 / 20 / 50
    Hyperparameter in Eq. (14) controlling how many top quantum-kernel eigenmodes carry the constructed target; directly controls the difficulty of the generated dataset and hence the size of the quantum-classical gap (Fig. 3).
  • x (step height of target alignment ĉ) = 1 / 1.5 / 10
    Hyperparameter in Eq. (14) scaling the target power; the paper notes x 'seems to be controlling the scale' of the loss curves (Section V).
  • evolution time / bandwidth t = 0.5 (main); 1.0 (Appendix C)
    Parameter of the Hamiltonian evolution feature map (Eq. 13), fixed without tuning; Appendix C shows it changes small-N error rates, contradicting the authors' earlier bandwidth study [24].
  • ridge parameter λ = 0.0001
    Regularization constant in the KRR objective (Eq. 5); chosen, not optimized.
  • PCA-reduced feature count = 8
    All three datasets reduced to 8 features before embedding; this choice affects the kernel spectrum, the qubit count (9 qubits for 8 features), and the comparison.
assumptions (5)
  • standard math Mercer's theorem and existence of the spectral decomposition of the integral kernel operator (Eq. 3)
    Used in Section III-A to represent the kernel and target functions as eigenfunction expansions.
  • domain assumption Canatar et al. generalization error formulas (Eqs. 7-9) are valid and transfer to quantum kernels on finite classical datasets
    The paper imports Eqs. (7)-(9) from [8] without re-derivation; the empirical agreement is the evidence for transfer, but Appendix A shows a counter-example (geometric-difference labels) where predicted and empirical loss diverge.
  • domain assumption The finite-sample Gram matrix spectrum approximates the population kernel operator spectrum
    Predicted loss in Algorithm 1 uses eigenvalues of the full-data kernel matrix to predict error at smaller N; the theory in [8] concerns the population operator with measure p(x).
  • domain assumption Hamiltonian evolution feature map (Eq. 13) with Haar-random initial states is a suitable QKM embedding
    Chosen based on [5, 26, 28]; the paper does not derive properties of this map. Appendix C shows bandwidth dependence, implying sensitivity to this choice.
  • domain assumption One-hot encoding of labels and squared-error loss is an appropriate regression surrogate for classification
    Adopted from [8] (Section IV-B); classification datasets are treated with regression theory, while the empirical protocol mixes SVM language (Section IV-B) with KRR theory (Section III-A).
invented entities (1)
  • Semi-artificial dataset (relabeled classical data with step-function target alignment ĉ)
    purpose: Construct datasets whose labels saturate the target-alignment measure for a QKM, to probe whether QKMs can be data-efficient
    The construction (Eqs. 14-15) is the paper's main tool; it is a methodological device, not a physical object, and it provides no falsifiable prediction outside the paper. The advantage it produces is defined in terms of the quantum kernel's own spectrum.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Is data-efficient learning feasible with quantum models?." pith.science (2026). https://pith.science/paper/UZHYHGBH

@misc{pith2026250819437,
  author       = {Pith},
  title        = {Pith review of: Is data-efficient learning feasible with quantum models?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UZHYHGBH}},
  note         = {Machine review of arXiv:2508.19437}
}
read the original abstract

The importance of analyzing nontrivial datasets when testing quantum machine learning (QML) models is becoming increasingly prominent in literature, yet a cohesive framework for understanding dataset characteristics remains elusive. In this work, we introduce a data-generation tool that allows to construct semi-artificial classical datasets tailored to quantum kernel methods (QKMs). Using this tool, we show that on fully classical datasets, QKMs can require fewer training examples than classical kernels to reach comparable error, providing clear empirical evidence that data-efficient learning with quantum models is possible on classical data. The main motivation behind this tool is to enable the community to perform controlled studies to figure out which dataset characteristics are particularly fitting for quantum models by tuning the data-generation procedure. Additionally, our study brings a spectral-bias-based generalization metric from classical kernel methods into the QML domain and show that the performance predicted by this metric aligns closely with empirical results, thereby closing an important gap between theory and practice in QML generalization. Our tool paves the way for a systematic exploration of dataset complexities. This could potentially contribute to a deeper understanding of the generalization benefits of QKM models (extendable to a broader family of QML models) and shifts the search for quantum advantage from ad hoc benchmark hunting to principled dataset design.

Figures

Figures reproduced from arXiv: 2508.19437 by the authors.

Figure 1
Figure 1. Comprehensive overview of the interconnections among [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. The difference between unchanged c and cˆ (Eq. (14)) with different hyperparameter settings. and classical models can be (artificially) made significant. If so, it opens opportunities to analyze the properties of such datasets and discuss their potential real-world origins. We then train and test the models on modified datasets and compute their empirical accuracy. Additionally, we compute the predicted accuracy and… view at source ↗
Figure 3
Figure 3. Square mean error loss E difference between classical and quantum kernels that depends on N on dataset relabeled to saturate target alignment Notably, our method did not alter the features of the dataset, and therefore had no impact on the eigenvalues. An intriguing direction for further research is to explore how dataset features and the architectural properties of quantum models affect eigenvalues of the resulting… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Square mean error loss E difference between classical and quantum kernels that depends on N on dataset relabeled to saturate geometric difference dataset aimed at saturating the geometric difference between these kernels; however, this gap is less pronounced as in the …
Figure 5
Figure 5. Figure 5: Correlation comparison between true / generated labels, relabeled for classical and quantum kernel methods (KMs), and [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Effects of bandwidth t hyperparameter on convergence of error rate tested on semi-artificial dataset with cˆ 20 1 . to t = 0.5. Here, we explore whether changing the bandwidth to t = 1.0 (effectively removing it) has any significant impact on the performance. As illust…

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. Lund Plane to Bloch (LP2B) Encoding for Object and Polarization Tagging with Quantum Jet Substructure

    quant-ph 2026-04 unverdicted novelty 7.0 of 10

    LP2B encoding converts Lund plane jet representations into Bloch sphere qubit states, enabling a QTTN that matches classical LundNet performance on polarization tagging and W/top tagging with three orders of magnitude...

Reference graph

Works this paper leans on

32 extracted references · 12 canonical work pages · cited by 1 Pith paper

  1. [1]

    Quantum Convolutional Neural Networks are (Effectively) Classically Simulable

    Pablo Bermejo et al. Quantum Convolutional Neural Networks are (Effectively) Classically Simulable . 2024. arXiv: 2408.12739 [quant-ph]. URL: https://arxiv. org/abs/2408.12739

    arXiv 2024

  2. [2]

    On the similarity of bandwidth-tuned quantum kernels and classical kernels

    Roberto Flórez Ablan, Marco Roth, and Jan Schn- abel. On the similarity of bandwidth-tuned quantum kernels and classical kernels . 2025. arXiv: 2503.05602 [quant-ph]. URL: https://arxiv.org/abs/2503.05602

    work page Pith review arXiv 2025

  3. [3]

    Binary classifiers for noisy datasets: a comparative study of existing quantum machine learning frameworks and some new approaches

    N. Schetakis et al. Binary classifiers for noisy datasets: a comparative study of existing quantum machine learning frameworks and some new approaches . 2021. arXiv: 2111.03372 [quant-ph]. URL: https://arxiv.org/abs/ 2111.03372

    work page Pith review arXiv 2021

  4. [4]

    Exploiting Symmetry in Variational Quantum Machine Learning

    Johannes Jakob Meyer et al. “Exploiting Symmetry in Variational Quantum Machine Learning”. In: PRX Quantum 4.1 (Mar. 2023). ISSN : 2691-3399. DOI: 10. 1103/prxquantum.4.010328. URL: http://dx.doi.org/10. 1103/PRXQuantum.4.010328

    work page 2023

  5. [5]

    Power of data in quantum machine learning

    Hsin-Yuan Huang et al. “Power of data in quantum machine learning”. In: Nature Communications 12.1 (2021). DOI: 10.1038/s41467-021-22539-9. URL: https: //doi.org/10.1038%2Fs41467-021-22539-9

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

  6. [6]

    Neural Tangent Kernel: Convergence and Generalization in Neural Networks

    Arthur Jacot, Franck Gabriel, and Clément Hongler. “Neural Tangent Kernel: Convergence and Generalization in Neural Networks”. In: (2018). DOI: 10.48550/ARXIV . 1806.07572. URL: https://arxiv.org/abs/1806.07572

    arXiv doi:10.48550/arxiv.1806.07572 2018

  7. [7]

    Supervised quantum machine learning models are kernel methods

    Maria Schuld. Supervised quantum machine learning models are kernel methods. 2021. DOI: 10.48550/ARXIV . 2101.11020. URL: https://arxiv.org/abs/2101.11020

    arXiv doi:10.48550/arxiv.2101.11020 2021

  8. [9]

    Barren plateaus in quantum neural network training landscapes

    Jarrod R. McClean et al. “Barren plateaus in quantum neural network training landscapes”. In: Nature Commu- nications 9.1 (Nov. 2018). DOI: 10.1038/s41467-018- 07090-4. URL: https://doi.org/10.1038%2Fs41467-018- 07090-4

    doi:10.1038/s41467-018- 2018

Show all 32 references
  1. [10]

    Exponential concentration in quantum kernel methods

    Supanut Thanasilp et al. “Exponential concentration in quantum kernel methods”. In: Nature Communications 15.5200 (2024). DOI: 10.1038/s41467-024-49287-w

    2024 doi

  2. [11]

    Generalization in quantum machine learning from few training data

    Matthias C. Caro et al. “Generalization in quantum machine learning from few training data”. In: Nature Communications 13.1 (Aug. 2022). ISSN : 2041-1723. DOI: 10.1038/s41467-022-32550-3. URL: http://dx.doi. org/10.1038/s41467-022-32550-3

    2022 doi

  3. [12]

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

    Yunchao Liu, Srinivasan Arunachalam, and Kristan Temme. “A rigorous and robust quantum speed-up in supervised machine learning”. In: Nature Physics 17.9 (July 2021), pp. 1013–1017. ISSN : 1745-2481. DOI: 10.1038/s41567- 021- 01287- z. URL: http://dx.doi. org/10.1038/s41567-021-01287-z

    2021 doi

  4. [13]

    Covariant quantum kernels for data with group structure

    Jennifer R. Glick et al. “Covariant quantum kernels for data with group structure”. In: Nature Physics 20.3 (Jan. 2024), pp. 479–483. ISSN : 1745-2481. DOI: 10.1038/ s41567-023-02340-9. URL: http://dx.doi.org/10.1038/ s41567-023-02340-9

    2024

  5. [14]

    The power of quantum neural networks

    Amira Abbas et al. “The power of quantum neural networks”. In: Nature Computational Science 1.6 (June 2021), pp. 403–409. DOI: 10.1038/s43588-021-00084-1. URL: https://doi.org/10.1038%2Fs43588-021-00084-1

    2021 doi

  6. [15]

    The inductive bias of quantum kernels

    Jonas Kübler, Simon Buchholz, and Bernhard Schölkopf. “The inductive bias of quantum kernels”. In: Advances in Neural Information Processing Systems 34 (2021), pp. 12661–12673

    2021

  7. [16]

    Contextuality and inductive bias in quantum machine learning

    Joseph Bowles et al. Contextuality and inductive bias in quantum machine learning . 2023. arXiv: 2302.01365 [quant-ph]. URL: https://arxiv.org/abs/2302.01365

    2023 arXiv

  8. [17]

    Effect of data encoding on the expressive power of vari- ational quantum-machine-learning models

    Maria Schuld, Ryan Sweke, and Johannes Jakob Meyer. “Effect of data encoding on the expressive power of vari- ational quantum-machine-learning models”. In: Physical Review A 103.3 (Mar. 2021). ISSN : 2469-9934. DOI: 10.1103/physreva.103.032430. URL: http://dx.doi.org/10. 1103/...

    2021 doi

  9. [18]

    Understanding deep learning (still) requires rethinking generalization

    Chiyuan Zhang et al. “Understanding deep learning (still) requires rethinking generalization”. In: Commun. ACM 64.3 (2021), pp. 107–115. ISSN : 0001-0782. DOI: 10. 1145/3446776. URL: https://doi.org/10.1145/3446776

    2021 doi

  10. [19]

    Understanding quantum machine learning also requires rethinking generalization

    Elies Gil-Fuster, Jens Eisert, and Carlos Bravo-Prieto. “Understanding quantum machine learning also requires rethinking generalization”. In: Nature Communications 15.1 (Mar. 2024). ISSN : 2041-1723. DOI: 10 . 1038 / s41467 - 024 - 45882 - z.URL: http : / / dx . doi . org / 10...

    2024

  11. [20]

    Generalization Bounds in Hybrid Quantum-Classical Machine Learning Models

    Tongyan Wu et al. Generalization Bounds in Hybrid Quantum-Classical Machine Learning Models . 2025. arXiv: 2504.08456 [quant-ph]. URL: https://arxiv. org/abs/2504.08456

    2025

  12. [21]

    Classical Surrogates for Quantum Learning Models

    Franz J. Schreiber, Jens Eisert, and Johannes Jakob Meyer. “Classical Surrogates for Quantum Learning Models”. In: Physical Review Letters 131.10 (Sept. 2023). ISSN : 1079-7114. DOI: 10.1103/physrevlett.131. 100803. URL: http://dx.doi.org/10.1103/PhysRevLett. 131.100803

    2023 doi

  13. [22]

    Connecting Ansatz Expressibility to Gradient Magnitudes and Barren Plateaus

    Zoë Holmes et al. “Connecting Ansatz Expressibility to Gradient Magnitudes and Barren Plateaus”. In: PRX Quantum 3 (1 2022), p. 010313. DOI: 10 . 1103 / PRXQuantum .3 .010313. URL: https : // link. aps .org/ doi/10.1103/PRXQuantum.3.010313

    2022 doi

  14. [23]

    Gradients and frequency profiles of quantum re-uploading models

    Alice Barthe and Adrián Pérez-Salinas. “Gradients and frequency profiles of quantum re-uploading models”. In: Quantum 8 (Nov. 2024), p. 1523. ISSN : 2521-327X. DOI: 10.22331/q-2024-11-14-1523. URL: http://dx.doi.org/10. 22331/q-2024-11-14-1523

    2024 doi

  15. [24]

    A hyperparameter study for quantum kernel methods

    Sebastian Egginger, Alona Sakhnenko, and Jeanette Miriam Lorenz. “A hyperparameter study for quantum kernel methods”. In: Quantum Machine Intelligence 6.2 (July 2024). ISSN : 2524-4914. DOI: 10.1007/s42484- 024-00172-1. URL: http://dx.doi.org/10.1007/s42484- 024-00172-1

    2024 doi

  16. [25]

    Understanding the effects of data encoding on quantum-classical convolutional neural networks

    Maureen Monnet et al. Understanding the effects of data encoding on quantum-classical convolutional neural networks. 2024. arXiv: 2405.03027 [quant-ph]. URL: https://arxiv.org/abs/2405.03027

    2024 arXiv

  17. [26]

    Bandwidth Enables General- ization in Quantum Kernel Models

    Abdulkadir Canatar et al. Bandwidth Enables General- ization in Quantum Kernel Models . 2023. arXiv: 2206. 06686 [quant-ph]. URL: https://arxiv.org/abs/2206. 06686

    2023

  18. [27]

    Quantum supremacy using a programmable superconducting processor

    Frank Arute et al. “Quantum supremacy using a programmable superconducting processor”. In: Nature 574.7779 (2019), pp. 505–510. DOI: 10.1038/s41586- 019-1666-5. URL: https://doi.org/10.1038%2Fs41586- 019-1666-5

    2019 doi

  19. [28]

    Importance of kernel bandwidth in quantum machine learning

    Ruslan Shaydulin and Stefan M. Wild. “Importance of kernel bandwidth in quantum machine learning”. In: Physical Review A 106.4 (Oct. 2022). DOI: 10.1103/ physreva.106.042407. URL: https://doi.org/10.1103% 2Fphysreva.106.042407

    2022

  20. [29]

    Training quantum embedding kernels on near-term quantum computers

    Thomas Hubregtsen et al. “Training quantum embedding kernels on near-term quantum computers”. In: Physical Review A 106.4 (Oct. 2022). ISSN : 2469-9934. DOI: 10.1103/physreva.106.042431. URL: http://dx.doi.org/10. 1103/PhysRevA.106.042431

    2022 doi

  21. [30]

    Clinical data classification with noisy intermediate scale quantum computers

    Sasan Moradi et al. “Clinical data classification with noisy intermediate scale quantum computers”. In: Scien- tific Reports 12.1 (2022), p. 1851

    2022

  22. [31]

    The UCI Machine Learning Repository

    Kolby Nottingham Markelle Kelly Rachel Longjohn. The UCI Machine Learning Repository . https://archive. ics.uci.edu

  23. [32]

    Quantum machine learning beyond kernel methods

    Sofiene Jerbi et al. “Quantum machine learning beyond kernel methods”. In: Nature Communications 14.1 (Jan. 2023). ISSN : 2041-1723. DOI: 10.1038/s41467- 023- 36159-y. URL: http://dx.doi.org/10.1038/s41467-023- 36159-y

    2023 doi

  24. [33]

    Lorena et al

    Ana C. Lorena et al. How Complex is your classifi- cation problem? A survey on measuring classification complexity. 2020. arXiv: 1808.03591 [cs.LG]. URL: https://arxiv.org/abs/1808.03591. APPENDIX A. Saturating geometric difference For completness, we compare our method with a...

    2020 arXiv

Pith tools

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