Pith. sign in

REVIEW 3 major objections 6 minor 98 references

Qutrit neural quantum kernels beat matched quantum neural nets on binary and three-class tasks, and the SU(3) parameterization choice strongly shapes both training and accuracy.

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

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

T0 review · grok-4.5

2026-07-30 15:58 UTC pith:LBXPBFGF

load-bearing objection Solid, honest design study extending NQKs to qutrits; useful ablations, incremental novelty, and a real but overstated QNN-vs-kernel comparison caveat. the 3 major comments →

arxiv 2607.23683 v1 pith:LBXPBFGF submitted 2026-07-26 quant-ph

Qutrit-Based Neural Quantum Kernels for Classification Tasks

Camila Cristiano-Romero , Pablo Rodriguez-Grasa , Mikel Sanz This is my paper
classification quant-ph
keywords neural quantum kernelsqutritsquantum neural networksdata re-uploadingSU(3) parameterizationquantum kernel methodsmulticlass classificationqudit machine learning
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper extends neural quantum kernels from qubits to qutrits: a quantum neural network is pretrained on the classification task, then frozen and reused as a task-adapted embedding that defines a quantum kernel for a classical support-vector classifier. The authors argue that three-level systems give richer local embeddings (via SU(3) unitaries with up to eight parameters) and a natural three-class readout, and they systematically vary feature budget, number of qutrits, kernel construction (1-to-n versus n-to-n), and three different SU(3) parameterizations. Across Fashion-MNIST, HAR, MAGIC, and Covertype, the kernel models improve on the corresponding QNN baselines in nearly every setting, can gain from more features and more qutrits until gains saturate, and remain competitive with a classical RBF-SVM. An ablation shows the geometric (Lie-algebra exponential) parameterization trains more favorably and usually outperforms Euler-angle and Givens forms. A sympathetic reader cares because the work treats qudit models as designable systems whose building blocks—not only scale—decide whether the quantum embedding helps.

Core claim

On four standard benchmarks, for both binary and three-class problems, qutrit neural quantum kernels improve over the matched pretrained QNN classifiers in nearly all configurations tested, and performance can rise with encoded feature count and with system size, though gains often saturate and depend on dataset and on how SU(3) unitaries are parameterized; the unitary parameterization itself materially changes optimization behavior and final accuracy.

What carries the argument

Neural quantum kernels (NQKs): pretrain a data-reuploading qutrit QNN (optionally grown progressively), freeze its circuit as the embedding, and form an embedding quantum kernel from state overlaps; two lifts are used—1-to-n (replicate a trained single-qutrit map with fixed entanglement) and n-to-n (train the full multi-qutrit embedding).

Load-bearing premise

All reported gains come from ideal noiseless simulations with fixed circuit depth, nearest-neighbour entanglement, heavy random restarts, and PCA-compressed features, so the claimed QNN-to-kernel lift may not survive noise, finite shots, or different preprocessing.

What would settle it

Rerun the same Fashion-MNIST and Covertype binary and three-class protocols with realistic noise and finite-shot overlap estimates: if the 1-to-4 and 4-to-4 NQKs no longer beat the matched QNN baselines, or if the geometric SU(3) edge disappears, the central empirical claim fails to transfer.

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

If this is right

  • Task-adapted qutrit embeddings used as fixed kernels are a practical alternative to end-to-end variational readout at matched resources.
  • Feature budget and register size are useful but saturating knobs; more is not automatically better once p and n are moderate.
  • SU(3) parameterization must be treated as a first-class design choice, not an implementation detail, because it changes both trainability and accuracy.
  • 1-to-n and n-to-n constructions can reach similar accuracy at n=4 despite different trainable freedom, so the simpler replicated map remains competitive at moderate size.
  • Qutrit NQKs can stay competitive with classical RBF-SVMs on the same splits while offering a direct three-class computational-basis interface.

Where Pith is reading between the lines

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

  • If noise and shot noise erase the kernel lift first on n-to-n (deeper joint training) rather than 1-to-n, hardware-era designs may prefer replicated single-qutrit maps with fixed entanglers.
  • The geometric parameterization’s more symmetric generator coupling may explain its smoother feature scaling; testing whether random or learned feature-to-coordinate assignments close the gap for Euler and Givens forms would isolate inductive bias from expressivity.
  • Extending the same NQK pipeline to d>3 would test whether extra local generators buy multiclass headroom or mainly increase barren-plateau and calibration cost.
  • Because kernel construction reduced sensitivity to the pretraining loss in the three-class ablations, hybrid “train embedding, convex readout” pipelines may be a general stabilizer for qudit classifiers beyond this architecture.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The manuscript extends neural quantum kernels from qubits to qutrits. It generalizes the data-reuploading embedding to local SU(3) blocks, introduces qutrit SUM and controlled-rotation entanglers, and studies both 1-to-n and n-to-n kernel constructions for binary and three-class classification. Noiseless simulations on Fashion-MNIST, HAR, MAGIC, and Covertype examine feature budget, system size, three SU(3) parameterizations, and cost-function sensitivity. The authors report that the resulting NQKs outperform matched QNN baselines in nearly all settings, that gains from additional features and qutrits are dataset-dependent and often saturate, and that the Geometric SU(3) parameterization usually performs best and shows more favorable training diagnostics. The NQKs are broadly competitive with, but do not clearly exceed, an RBF-SVM baseline.

Significance. If the reported comparisons hold up, this is a useful systematic design study for qudit quantum machine learning and a natural extension of NQKs beyond qubits. Its strengths are the breadth of the empirical protocol: two kernel constructions, binary and multiclass tasks, four datasets, three SU(3) parameterizations, matched circuit baselines, stratified five-fold evaluation with preprocessing fitted only on training folds, standard errors, a classical RBF-SVM comparison, and informative cost-function and training-diagnostics appendices. The claims are generally hedged and the paper does not overstate them as a quantum advantage. Its significance is primarily methodological guidance for qutrit model design under ideal simulation, rather than evidence of near-term practical superiority over classical methods.

major comments (3)
  1. [Sec. III B; Tables II and VI] The evaluation protocol does not state where the SVM regularization C—and the RBF baseline’s C and γ—are selected. If the grid search uses the held-out fold’s test accuracy, Tables II and VI and the QNN-versus-NQK comparisons are optimistically biased; if an inner training split or nested cross-validation was used, this needs to be stated explicitly. Because several reported differences are only about 1–3 percentage points, the selection protocol is load-bearing. Please use nested model selection or a fixed validation split, and preferably report paired fold-level differences in addition to fold means and standard errors.
  2. [Sec. II A; Eqs. (20)–(21); Figs. 4–7; Table II] The central QNN-to-NQK comparison is asymmetric in readout capacity. The QNN predicts from a fixed threshold on one ⟨Sz⟩ value, or an argmax over three probabilities on one qutrit, whereas the NQK receives a tuned SVM over a kernel of the full n-qutrit feature states. The manuscript itself characterizes the mechanism as replacing the QNN’s fixed measurement rule with an optimized readout. Consequently, the lift may reflect readout capacity or access to richer state information rather than the qutrit kernel construction specifically. A frozen-embedding control is needed: for the same trained circuits, train a logistic/linear or kernel SVM on all single-qutrit outcome probabilities or a richer fixed set of low-order moments, using the same nested tuning; a validation-tuned binary threshold would also be useful. If this control matches the NQK, the conclusion should be narrowed to an SVM-re
  3. [Sec. III E 1, Eqs. (10)–(12); Appendix D, Fig. 20] The parameterization ablation compares Geometric, Euler, and Givens coordinates at the same learning rate and with the same sequential assignment of the first p coordinates, but these coordinates have different nonlinear roles, effective ranges, and Jacobian scales. The larger gradient norms reported for the Geometric form may therefore indicate a different parameter-space metric or effective optimization step size rather than intrinsically more favorable optimization. To support the comparatively strong conclusion that the unitary representation affects optimization behaviour, please add a learning-rate/optimization-budget sweep or a Jacobian-normalized encoding-scale comparison, or explicitly restrict the claim to the present fixed optimizer and coordinate-scaling protocol.
minor comments (6)
  1. [Abstract and Sec. IV] Please state in the abstract that all experiments are ideal, noiseless state-vector simulations at fixed depth L=6. The limitation is appropriately acknowledged later, but including it in the abstract would prevent the performance claims from being read as hardware results.
  2. [Secs. II B 4 and III E 1] The statement that features enter “symmetrically” in the Geometric parameterization is potentially misleading because the su(3) generators do not commute. “Jointly through a linear combination before exponentiation” would be more precise.
  3. [Sec. III B] Please specify how the fixed per-class subsamples are drawn, whether the same subsamples and fold assignments are used across all models, and how many independent optimization runs are used at each progressive-growth step for n>1. A code/data availability statement with seeds would also improve reproducibility.
  4. [Figs. 4–7 and Appendix B] Several qualitative phrases such as “saturates,” “non-monotonic,” and “marginal” would be easier to assess if the captions or tables reported the numerical paired differences and uncertainties at the relevant values of n and p.
  5. [Table VI and Appendix B] Typographical issues include “97.69±00.33,” “qutrit NQK’s,” and the missing period in “Fig 15.” There is also a duplicated comma in the reference to “Sec. II A 1,,” in Sec. III B.
  6. [References] Refs. [11] and [50] appear to cite the same Nature Physics paper by Liu, Arunachalam, and Temme. Please consolidate duplicate references and check the reference list for similar redundancies.

Circularity Check

0 steps flagged

Empirical architecture study with no circular derivation: reported gains are measured, not forced by definition or self-citation.

full rationale

The paper extends Neural Quantum Kernels from qubits to qutrits and reports cross-validated classification accuracy under controlled ablations (feature budget p, system size n, 1-to-n vs n-to-n, SU(3) parameterizations) on four external benchmarks, plus a classical RBF-SVM baseline. The central claim—that qutrit NQKs improve over matched QNN baselines in nearly all settings—is an empirical comparison of two predictors built from the same pretrained embedding, not a first-principles derivation. Kernel entries are Hilbert–Schmidt overlaps of frozen feature states (Eqs. 3–4); SVM coefficients are fit on the training kernel matrix and scored on held-out folds. Nothing equates reported test accuracy to a fitted input by construction. Self-citations to the authors’ prior NQK work supply the method being generalized (progressive growth, 1-to-n / n-to-n), not the accuracy numbers or the SU(3) ablation outcomes. Hyperparameter search (C, restarts) and the acknowledged readout asymmetry (fixed QNN threshold/argmax vs SVM) are experimental-design issues, not circular reductions. No uniqueness theorem, smuggled ansatz-as-theorem, or renaming of a known law appears. Score 0 is appropriate.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 1 invented entities

The work rests on standard quantum kernel and variational-circuit assumptions plus architectural choices fixed by the authors. No new physical entities are postulated. Free parameters are training/architecture hyperparameters and classical SVM regularization chosen by grid search; results are conditional on noiseless simulation and the chosen ansatz family.

free parameters (5)
  • Circuit depth L = 6
    Fixed to L=6 re-uploading layers for all main experiments; capacity and scaling trends depend on this choice.
  • Encoded feature budget p and PCA truncation = p ≤ 8 (often fixed at 8)
    p is varied up to 8; features come from PCA fitted on training folds then rescaled to [-1,1], shaping the embedding geometry.
  • QNN optimizer hyperparameters = LR=0.006 (1-qutrit), 0.002 (growth); WD=1e-5; 70/20 epochs; 7000 restarts
    Adam LR, weight decay, batch size, epochs, and 7000 random initializations for 1-qutrit models strongly affect which embedding is frozen into the kernel.
  • SVM regularization C (and RBF γ for baseline) = C in {0.01,0.1,1,10,100}; γ in {scale,0.001,0.01,0.1,1}
    Selected by grid search on each fold; final reported NQK and classical accuracies depend on this classical fit on the fixed kernel/features.
  • Entangling ansatz angles / fixed SUM pattern = nearest-neighbour CR1·CR4·CR6 or fixed SUM3
    n-to-n uses trainable CR sequences on λ1,λ4,λ6; 1-to-n uses fixed SUM3. These hand-chosen primitives define the multi-qutrit feature map.
axioms (5)
  • domain assumption Embedding quantum kernels from pure-state overlaps k(xi,xj)=|⟨0|S(xi)†S(xj)|0⟩|² are valid PSD kernels for classical SVM training.
    Used throughout Sec. II A as the kernel definition after QNN pretraining.
  • domain assumption Noiseless, infinite-shot simulation of qutrit circuits is a meaningful proxy for comparing architectures.
    All results in Sec. III and appendices; Discussion flags noise as future work.
  • ad hoc to paper Progressive growth with zero-init on new qutrits/entanglers is a fair scalable training strategy for n-to-n embeddings.
    Inherited from prior NQK work and fixed in Sec. II A 1; alternatives are not compared.
  • ad hoc to paper Sequential assignment of the first p coordinates in each SU(3) parameterization is a reasonable feature-encoding convention.
    Sec. II B 4; authors note it may induce parameterization-dependent bias, confirmed in ablation.
  • standard math Standard linear algebra / SU(d) representation theory for qutrit unitaries and controlled gates.
    Background for Geometric, Euler, and Givens forms and SUM/CR primitives.
invented entities (1)
  • Qutrit neural quantum kernel (1-to-n and n-to-n constructions) no independent evidence
    purpose: Name the qudit extension of pretrained-QNN embeddings used as fixed quantum kernels for binary and three-class SVM classification.
    Not a new physical object; an architectural construct built from existing NQK ideas plus SU(3) blocks. Independent evidence is only the empirical benchmarks in this paper.

pith-pipeline@v1.2.0-grok45-kimik3 · 32970 in / 3714 out tokens · 71262 ms · 2026-07-30T15:58:25.224747+00:00 · methodology

0 comments
read the original abstract

Neural quantum kernels (NQKs) construct quantum kernels by pretraining a quantum neural network (QNN) and subsequently reusing the trained circuit as a task-adapted embedding. Extending this framework to qudits, with local unitaries in $\mathrm{SU}(d)$, provides a natural route to richer data embeddings through the increased local degrees of freedom and a direct interface for multiclass classification via intrinsically multi-level quantum systems. In this work, focusing on qutrits ($d=3$), we extend NQKs to the qudit setting and perform a systematic study of key design choices, including the number of encoded features, the number of qutrits, the kernel construction (1-to-$n$ and $n$-to-$n$), and the parameterization of $\mathrm{SU}(3)$ unitaries. Across binary and three-class tasks on four benchmark datasets, qutrit NQKs improve over the corresponding QNN baselines in nearly all settings considered and can benefit from scaling both the feature budget and the system size, although the magnitude of these gains may saturate, is dataset-dependent, and depends on the chosen parameterization. In particular, an ablation over $\mathrm{SU}(3)$ parameterizations shows that the unitary representation can substantially impact both optimization behaviour and classifier performance. These findings highlight the potential of qudit-based quantum models not only as a straightforward generalization of qubit-based architectures, but also as a promising means to better exploit complex data structures in quantum machine learning.

Figures

Figures reproduced from arXiv: 2607.23683 by Camila Cristiano-Romero, Mikel Sanz, Pablo Rodriguez-Grasa.

Figure 1
Figure 1. Figure 1: FIG. 1: Progressive-growth training strategy used to scale the data re-uploading QNN embedding. Training starts from a single [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: FIG. 2: Schematic description of the two neural quantum kernel constructions considered in this work. (a) [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: shows how test accuracy varies with the number of encoded features under this fixed configuration. Accu￾racy generally improves as p increases, with gains that tend to saturate at larger feature counts. This behaviour is consis￾tent with the motivation for qutrit embeddings, namely that the larger local parameter space of SU(3) enables richer local transformations and allows additional encoded features to … view at source ↗
Figure 5
Figure 5. Figure 5: FIG. 5: Test accuracy on Fashion-MNIST as a function of the [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: FIG. 6: Test accuracy on Fashion-MNIST as a function of the [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: FIG. 8: Effect of [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 10
Figure 10. Figure 10: FIG. 10: Effect of [PITH_FULL_IMAGE:figures/full_fig_p012_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: FIG. 11: Test accuracy of the [PITH_FULL_IMAGE:figures/full_fig_p018_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: FIG. 12: Test accuracy of the [PITH_FULL_IMAGE:figures/full_fig_p018_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: FIG. 13: Test accuracy of the [PITH_FULL_IMAGE:figures/full_fig_p019_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: FIG. 14: Test accuracy of the [PITH_FULL_IMAGE:figures/full_fig_p019_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: FIG. 15: Test accuracy of the [PITH_FULL_IMAGE:figures/full_fig_p020_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: FIG. 16: Test accuracy of the [PITH_FULL_IMAGE:figures/full_fig_p020_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: FIG. 17: Effect of the [PITH_FULL_IMAGE:figures/full_fig_p021_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: FIG. 18: Test accuracy of the [PITH_FULL_IMAGE:figures/full_fig_p021_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: FIG. 19: Effect of the training objective for three-class qutrit classification. Test accuracy for the [PITH_FULL_IMAGE:figures/full_fig_p023_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: FIG. 20: Training diagnostics for the [PITH_FULL_IMAGE:figures/full_fig_p023_20.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

98 extracted references · 7 canonical work pages · 1 internal anchor

  1. [1]

    Pretraining a data re-uploading QNN as a scalable quantum embedding In Neural Quantum Kernels, the quantum embedding is re- alized by a parameterized quantum circuit trained in a super- vised manner. The architecture considered in [54] follows a data re-uploading structure [21], in which the classical fea- tures are encoded repeatedly throughout the circu...

  2. [2]

    Kernel construction We consider two neural quantum kernel constructions: the n-to-nand1-to-napproaches, which are summarized in Fig

  3. [3]

    Both constructions define an EQK by (i) learning a task- adapted quantum feature map via QNN pretraining and (ii) freezing the resulting embedding to construct the EQK ac- cording to Eq. (3). a.n-to-nNQK.Following [54], then-to-nconstruction trains ann-qubit data re-uploading QNN using the progressive growth strategy described in Sec. II A 1. After traini...

  4. [4]

    The computational basis is given by the set of orthonormal states [33, 77, 78] |0⟩,|1⟩,

    Qudit preliminaries Aquditis a generalization of the qubit to ad-dimensional Hilbert spaceH d. The computational basis is given by the set of orthonormal states [33, 77, 78] |0⟩,|1⟩, . . . ,|d−1⟩. An arbitrary pure state can be expressed as |ψ⟩= d−1X k=0 ck|k⟩, wherec k ∈Cand Pd−1 k=0 |ck|2 = 1. Single-qudit unitary evolutions are described by operators U...

  5. [5]

    Add a second qubit and initialize using previous parameters

  6. [6]

    1: Progressive-growth training strategy used to scale the data re-uploading QNN embedding

    Add a third qubit and initialize using previous parameters Initialize |0⟩ Initialize Up to n Initialize θ⋆=argminfcost(θ(1)) fcost(θ) S(2) θ(x)|0⟩ fcost(θ,φ) |0⟩ θ⋆,φ⋆=argminfcost(θ,φ)|0⟩ fcost(θ,φ) |0⟩ |0⟩ S(3) θ(x) Update Update Update FIG. 1: Progressive-growth training strategy used to scale the data re-uploading QNN embedding. Training starts from a ...

  7. [7]

    Training a n-qubit QNN (b)

  8. [8]

    Training a 1-qubit QNN |0⟩ |0⟩ S(n) θ⋆,φ⋆(x)

  9. [9]

    EQK construction |0⟩ |0⟩ S(n) θ⋆,φ⋆(xi)†S(n) θ⋆,φ⋆(xj) Find |0⟩ Sθ(x) Layer 1 Find |0⟩

  10. [10]

    2: Schematic description of the two neural quantum kernel constructions considered in this work

    EQK construction |0⟩ Sθ⋆(xj) kij=P0 U(x)U(θ1) E U(x)U(θL) U(x)U(θ(1) l) U(x)U(θ(n) l) Sθ⋆(xi)† Layer L Layer l FIG. 2: Schematic description of the two neural quantum kernel constructions considered in this work. (a)n-to-n: an n-qubit QNN is trained to obtain optimized embedding parameters(θ ⋆,φ ⋆), which are then frozen and used to construct the EQK by e...

  11. [11]

    In this work we consider two families of entangling primitives, which are used in different kernel constructions

    Qutrit entangling primitives To construct multiqutrit embeddings we combine local SU(3)blocks with two-qutrit entangling operations. In this work we consider two families of entangling primitives, which are used in different kernel constructions. a. SUM (generalized C X) gate.A standard generaliza- tion of the qubit controlled-Xto qudits is the SUM gate, ...

  12. [12]

    precomputed

    From qubits to qutrits The extension from qubit-based NQKs to qutrits is con- ceptually straightforward since the overall data re-uploading architecture, the progressive growth strategy, and the subse- quent kernel construction remain unchanged, while the circuit building blocks are promoted from localSU(2)operations to localSU(3)unitaries. Concretely, fo...

  13. [13]

    ERDF A way of making Europe

    Effect ofSU(3)parameterization We next examine how the choice of single-qutritSU(3)pa- rameterization impacts performance. We compare theGeo- metric(Lie-algebra exponential) form (Eq. 10), theEuler- anglesdecomposition (Eq. 11), and theGivens-rotationform (Eq. 12), keeping the rest of the experimental setup un- changed. This choice affects performance alr...

    2020

  14. [14]

    Biamonte, P

    J. Biamonte, P. Wittek, N. Pancotti, P. Rebentrost, N. Wiebe, and S. Lloyd, Quantum machine learning, Nature549, 195 (2017)

    2017

  15. [15]

    Carleo, I

    G. Carleo, I. Cirac, K. Cranmer, L. Daudet, M. Schuld, N. Tishby, L. V ogt-Maranto, and L. Zdeborov´a, Machine learn- ing and the physical sciences, Reviews of Modern Physics91, 10.1103/revmodphys.91.045002 (2019)

    doi:10.1103/revmodphys.91.045002 2019

  16. [16]

    Dunjko and H

    V . Dunjko and H. J. Briegel, Machine learning & artificial in- telligence in the quantum domain: a review of recent progress, Reports on Progress in Physics81, 074001 (2018)

    2018

  17. [17]

    Benedetti, E

    M. Benedetti, E. Lloyd, S. Sack, and M. Fiorentini, Parame- terized quantum circuits as machine learning models, Quantum Science and Technology4, 043001 (2019)

    2019

  18. [18]

    Farhi and H

    E. Farhi and H. Neven, Classification with Quantum Neural Networks on Near Term Processors (2018)

    2018

  19. [19]

    Skolik, S

    A. Skolik, S. Jerbi, and V . Dunjko, Quantum agents in the Gym: a variational quantum algorithm for deep Q-learning, Quantum 6, 720 (2022)

    2022

  20. [20]

    Schuld, A

    M. Schuld, A. Bocharov, K. M. Svore, and N. Wiebe, Circuit- centric quantum classifiers, Physical review. A/Physical review, A101, 10.1103/physreva.101.032308 (2020)

    doi:10.1103/physreva.101.032308 2020

  21. [21]

    Gyurik and V

    C. Gyurik and V . Dunjko, On establishing learning separations between classical and quantum machine learning with classical data (2022)

    2022

  22. [22]

    Sweke, J.-P

    R. Sweke, J.-P. Seifert, D. Hangleiter, and J. Eisert, On the Quantum versus Classical Learnability of Discrete Distribu- tions, Quantum 10.22331/q-2021-03-23-417 (2021)

    doi:10.22331/q-2021-03-23-417 2021

  23. [23]

    Jerbi, L

    S. Jerbi, L. M. Trenkwalder, H. P. Nautrup, H. J. Briegel, and V . Dunjko, Quantum Enhancements for Deep Reinforcement Learning in Large Spaces, PRX Quantum2, 10.1103/prxquan- tum.2.010328 (2021)

    doi:10.1103/prxquan- 2021

  24. [26]

    Jadhav, A

    A. Jadhav, A. Rasool, and M. Gyanchandani, Quantum Ma- chine Learning: Scope for real-world problems, Procedia Com- puter Science218, 2612 (2023)

    2023

  25. [27]

    J. R. McClean, S. Boixo, V . N. Smelyanskiy, R. Babbush, and H. Neven, Barren plateaus in quantum neural network training landscapes, Nature Communications9, 4812 (2018)

    2018

  26. [28]

    Cerezo, A

    M. Cerezo, A. Sone, T. V olkoff, L. Cincio, and P. J. Coles, Cost function dependent barren plateaus in shallow parametrized 15 quantum circuits, Nature Communications12, 1791 (2021)

    2021

  27. [29]

    Holmes, K

    Z. Holmes, K. Sharma, M. Cerezo, and P. J. Coles, Connect- ing Ansatz Expressibility to Gradient Magnitudes and Bar- ren Plateaus, PRX Quantum3, 10.1103/prxquantum.3.010313 (2022)

    doi:10.1103/prxquantum.3.010313 2022

  28. [30]

    Fontana, D

    E. Fontana, D. Herman, S. Chakrabarti, N. Kumar, R. Yalovet- zky, J. Heredge, S. H. Sureshbabu, and M. Pistoia, Characteriz- ing barren plateaus in quantum ans ¨atze with the adjoint repre- sentation, Nature Communications15, 7171 (2024)

    2024

  29. [31]

    Thanasilp, S

    S. Thanasilp, S. Wang, M. Cerezo, and Z. Holmes, Exponential concentration in quantum kernel methods, Nature communica- tions15, 5200 (2024)

    2024

  30. [33]

    Ragone, B

    M. Ragone, B. N. Bakalov, F. Sauvage, A. F. Kemper, C. Or- tiz Marrero, M. Larocca, and M. Cerezo, A lie algebraic theory of barren plateaus for deep parameterized quantum circuits, Na- ture Communications15, 7172 (2024)

    2024

  31. [34]

    P ´erez-Salinas, A

    A. P ´erez-Salinas, A. Cervera-Lierta, E. Gil-Fuster, and J. I. Latorre, Data re-uploading for a universal quantum classifier, Quantum4, 226 (2020)

    2020

  32. [35]

    P ´erez-Salinas, M

    A. P ´erez-Salinas, M. Y . Rad, A. Barthe, and V . Dunjko, Univer- sal approximation of continuous functions with minimal quan- tum circuits, Physical Review Research7, 10.1103/t49h-jmty (2025)

    work page doi:10.1103/t49h-jmty 2025

  33. [36]

    Schuld, R

    M. Schuld, R. Sweke, and J. J. Meyer, Effect of data encod- ing on the expressive power of variational quantum-machine- learning models, Physical review. A/Physical review, A103, 10.1103/physreva.103.032430 (2021)

    doi:10.1103/physreva.103.032430 2021

  34. [37]

    Casas and A

    B. Casas and A. Cervera-Lierta, Multidimensional fourier se- ries with quantum circuits, Physical Review A107, 062612 (2023)

    2023

  35. [38]

    Barthe and A

    A. Barthe and A. P ´erez-Salinas, Gradients and fre- quency profiles of quantum re-uploading models (2023), arXiv:2311.10822 [quant-ph]

    Pith/arXiv arXiv 2023

  36. [39]

    De Souza Farias, L

    T. De Souza Farias, L. Friedrich, and J. Maziero, A short review on qudit quantum machine learning (2025)

    2025

  37. [40]

    Gokhale, J

    P. Gokhale, J. M. Baker, C. Duckering, N. C. Brown, K. R. Brown, and F. T. Chong,Asymptotic improvements to quantum circuits via qutrits(2019) pp. 554–566

    2019

  38. [41]

    Pavlidis and E

    A. Pavlidis and E. Floratos, Quantum-Fourier-transform-based quantum arithmetic with qudits, Physical review. A/Physical re- view, A103, 10.1103/physreva.103.032417 (2021)

    doi:10.1103/physreva.103.032417 2021

  39. [42]

    Gedik, I

    Z. Gedik, I. A. Silva, B. C ¸ akmak, G. Karpat, E. L. G. Vidoto, D. O. Soares-Pinto, E. R. deAzevedo, and F. F. Fanchini, Com- putational speed-up with a single qudit, Scientific Reports5, 14671 (2015)

    2015

  40. [43]

    Deller, S

    Y . Deller, S. Schmitt, M. Lewenstein, S. Lenk, M. Federer, F. Jendrzejewski, P. Hauke, and V . Kasper, Quantum approx- imate optimization algorithm for qudit systems, Physical re- view. A/Physical review, A107, 10.1103/physreva.107.062410 (2023)

    doi:10.1103/physreva.107.062410 2023

  41. [44]

    Roca-Jerat, J

    S. Roca-Jerat, J. Rom ´an-Roche, and D. Zueco, Qudit machine learning, Machine Learning Science and Technology5, 015057 (2024)

    2024

  42. [45]

    Anand, J

    N. Anand, J. Marshall, J. Saied, E. Rieffel, and A. Morello, Qudit Designs and Where to Find Them (2026)

    2026

  43. [46]

    Y . Wang, Z. Hu, B. C. Sanders, and S. Kais, Qudits and High-Dimensional Quantum Computing, Frontiers in Physics 8, 10.3389/fphy.2020.589504 (2020)

    arXiv 2020

  44. [47]

    Ringbauer, M

    M. Ringbauer, M. Meth, L. Postler, R. Stricker, R. Blatt, P. Schindler, and T. Monz, A universal qudit quantum processor with trapped ions, Nature Physics18, 1053 (2022)

    2022

  45. [48]

    M. S. Blok, V . V . Ramasesh, T. Schuster, K. O’Brien, J. M. Kreikebaum, D. Dahlen, A. Morvan, B. Yoshida, N. Y . Yao, and I. Siddiqi, Quantum Information Scrambling on a Super- conducting Qutrit Processor, Physical Review X11, 021010 (2021)

    2021

  46. [49]

    Mandilara, B

    A. Mandilara, B. Dellen, U. Jaekel, T. Valtinos, and D. Syvridis, Classification of data with a qudit, a geometric approach, Quantum Machine Intelligence6, 10.1007/s42484-024-00146- 3 (2024)

    work page doi:10.1007/s42484-024-00146- 2024

  47. [50]

    N. L. Wach, M. S. Rudolph, F. Jendrzejewski, and S. Schmitt, Data re-uploading with a single qudit, Quantum Machine Intel- ligence5, 10.1007/s42484-023-00125-0 (2023)

    work page doi:10.1007/s42484-023-00125-0 2023

  48. [51]

    Valtinos, A

    T. Valtinos, A. Mandilara, and D. Syvridis, The Gell-Mann fea- ture map of qutrits and its applications in classification tasks, in Quantum Computing, Communication, and Simulation IV, V ol. 12911, edited by P. R. Hemmer and A. L. Migdall, International Society for Optics and Photonics (SPIE, 2024) p. 129110O

    2024

  49. [52]

    Acar and ˙I

    E. Acar and ˙I. Yılmaz, Unlocking the high dimensional’ po- tential: Comparative analysis of qubits and qutrits in varia- tional quantum neural networks, Neurocomputing623, 129404 (2025)

    2025

  50. [53]

    L. C. Souza and R. Portugal, Single-qudit quantum neural net- works for multiclass classification, Quantum Information Pro- cessing24, 10.1007/s11128-025-04998-x (2025)

    work page doi:10.1007/s11128-025-04998-x 2025

  51. [54]

    D. H. Useche, A. Giraldo-Carvajal, H. M. Zuluaga-Bucheli, J. A. Jaramillo-Villegas, and F. A. Gonz ´alez, Quantum mea- surement classification with qudits, Quantum Information Pro- cessing21, 10.1007/s11128-021-03363-y (2021)

    work page doi:10.1007/s11128-021-03363-y 2021

  52. [55]

    M. C. Lai ˜no, V . Chobanova, and M. L. Mart ´ınez, Qutrits for physics at the LHC (2025)

    2025

  53. [56]

    Havl´ıˇcek, A

    V . Havl´ıˇcek, A. D. C´orcoles, K. Temme, A. W. Harrow, A. Kan- dala, J. M. Chow, and J. M. Gambetta, Supervised learning with quantum-enhanced feature spaces, Nature567, 209 (2019)

    2019

  54. [57]

    Schuld and N

    M. Schuld and N. Killoran, Quantum Machine Learning in Feature Hilbert Spaces, Physical Review Letters122, 040504 (2019)

    2019

  55. [58]

    Bartkiewicz, C

    K. Bartkiewicz, C. Gneiting, A. ˇCernoch, K. Jir´akov´a, K. Lemr, and F. Nori, Experimental kernel-based quantum machine learning in finite feature space, Scientific Reports10, 12356 (2020)

    2020

  56. [59]

    Huang, M

    H.-Y . Huang, M. Broughton, M. Mohseni, R. Babbush, S. Boixo, H. Neven, and J. R. McClean, Power of data in quantum machine learning, Nature Communications12, 2631 (2021)

    2021

  57. [60]

    Peters, J

    E. Peters, J. Caldeira, A. Ho, S. Leichenauer, M. Mohseni, H. Neven, P. Spentzouris, D. Strain, and G. N. Perdue, Machine learning of high dimensional data on a noisy quantum proces- sor, npj Quantum Information7, 10.1038/s41534-021-00498-9 (2021)

    doi:10.1038/s41534-021-00498-9 2021

  58. [61]

    Kusumoto, K

    T. Kusumoto, K. Mitarai, K. Fujii, M. Kitagawa, and M. Ne- goro, Experimental quantum kernel trick with nuclear spins in a solid, npj Quantum Information7, 94 (2021)

    2021

  59. [62]

    Schuld, Supervised quantum machine learning models are kernel methods (2021)

    M. Schuld, Supervised quantum machine learning models are kernel methods (2021)

    2021

  60. [63]

    Y . Liu, S. Arunachalam, and K. Temme, A rigorous and ro- bust quantum speed-up in supervised machine learning, Nature Physics17, 1013 (2021)

    2021

  61. [64]

    Kyriienko and E

    O. Kyriienko and E. B. Magnusson, Unsupervised quantum ma- chine learning for fraud detection (2022)

    2022

  62. [65]

    Y . Wu, B. Wu, J. Wang, and X. Yuan, Quantum Phase Recog- nition via Quantum Kernel Methods, Quantum7, 981 (2023)

    2023

  63. [66]

    Tomasi, S

    J. Tomasi, S. Anthoine, and H. Kadri, Benign Overfitting with Quantum Kernels (2025). 16

    2025

  64. [67]

    Rodriguez-Grasa, Y

    P. Rodriguez-Grasa, Y . Ban, and M. Sanz, Neural quantum ker- nels: Training quantum kernels with quantum neural networks, Physical Review Research7, 10.1103/xphb-x2g4 (2025)

    doi:10.1103/xphb-x2g4 2025

  65. [68]

    Rodriguez-Grasa, R

    P. Rodriguez-Grasa, R. Farzan-Rodriguez, G. Novelli, Y . Ban, and M. Sanz, Satellite image classification with neural quantum kernels, Machine Learning Science and Technology6, 015043 (2025)

    2025

  66. [69]

    Rodriguez-Grasa, P

    P. Rodriguez-Grasa, P. Zhelnin, C. A. Arg ¨uelles, and M. Sanz, Neutrino Telescope Event Classification on Quantum Comput- ers (2025)

    2025

  67. [70]

    Sch ¨olkopf and A

    B. Sch ¨olkopf and A. J. Smola,Learning with Kernels: Support Vector Machines, Regularization, Optimization, and Beyond, Adaptive Computation and Machine Learning (MIT Press, Cambridge, MA, 2002) p. 644

    2002

  68. [71]

    Schuld and F

    M. Schuld and F. Petruccione, Quantum models as kernel meth- ods, inMachine Learning with Quantum Computers(Springer International Publishing, Cham, 2021) pp. 217–245

    2021

  69. [72]

    Buhrman, R

    H. Buhrman, R. Cleve, J. Watrous, and R. De Wolf, Quantum fingerprinting, Physical Review Letters87, 167902 (2001)

    2001

  70. [73]

    Fanizza, M

    M. Fanizza, M. Rosati, M. Skotiniotis, J. Calsamiglia, and V . Giovannetti, Beyond the Swap Test: Optimal Estimation of Quantum State Overlap, Physical Review Letters124, 060503 (2020)

    2020

  71. [74]

    Cincio, Y

    L. Cincio, Y . Subas ¸ı, A. T. Sornborger, and P. J. Coles, Learn- ing the quantum algorithm for state overlap, New Journal of Physics20, 113022 (2018)

    2018

  72. [75]

    Gil-Fuster, J

    E. Gil-Fuster, J. Eisert, and V . Dunjko, On the expressivity of embedding quantum kernels, Machine Learning: Science and Technology5, 025003 (2024)

    2024

  73. [76]

    Lloyd, M

    S. Lloyd, M. Schuld, A. Ijaz, J. Izaac, and N. Killoran, Quantum embeddings for machine learning (2020)

    2020

  74. [77]

    Blank, D

    C. Blank, D. K. Park, J.-K. K. Rhee, and F. Petruccione, Quan- tum classifier with tailored quantum kernel, npj Quantum Infor- mation6, 57 (2020)

    2020

  75. [78]

    J. M. K ¨ubler, S. Buchholz, and B. Sch¨olkopf, The inductive bias of quantum kernels, inProceedings of the 35th International Conference on Neural Information Processing Systems, NIPS ’21 (Curran Associates Inc., Red Hook, NY , USA, 2021)

    2021

  76. [79]

    The Impact of Feature Embedding Placement in the Ansatz of a Quantum Kernel in QSVMs

    I. Salmenper ¨a, I. Kuhtarskis, V . De Griend Arianne Meijer, and J. K. Nurminen, The Impact of Feature Embedding Placement in the Ansatz of a Quantum Kernel in QSVMs, arXiv (Cornell University) 10.48550/arxiv.2409.13147 (2024)

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2409.13147 2024

  77. [80]

    Shirai, K

    N. Shirai, K. Kubo, K. Mitarai, and K. Fujii, Quantum tan- gent kernel, Physical Review Research6, 10.1103/physrevre- search.6.033179 (2024)

    doi:10.1103/physrevre- 2024

  78. [81]

    S. S. Vedaie, M. Noori, J. S. Oberoi, B. C. Sanders, and E. Za- hedinejad, Quantum Multiple Kernel Learning (2020)

    2020

  79. [82]

    Hubregtsen, D

    T. Hubregtsen, D. Wierichs, E. Gil-Fuster, P.-J. H. S. Derks, P. K. Faehrmann, and J. J. Meyer, Training quantum embed- ding kernels on near-term quantum computers, Physical re- view. A/Physical review, A106, 10.1103/physreva.106.042431 (2022)

    doi:10.1103/physreva.106.042431 2022

  80. [83]

    Ghukasyan, J

    A. Ghukasyan, J. S. Baker, O. Goktas, J. Carrasquilla, and S. K. Radha, Quantum-Classical Multiple kernel learning (2023)

    2023

Showing first 80 references.