Pith. sign in

REVIEW 3 major objections 5 minor 57 references

Quantum Multi-view Kernel Learning with Local Information

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read L-QMVKL trains one quantum kernel per data view on a hybrid global-local alignment objective and beats single-view and classical multi-kernel baselines on Mfeat.

desk verdict A plausible incremental algorithm with an unfair classical baseline and test-set hyperparameter selection; the quantum-vs-classical claim is not supported, but the multi-view local training idea is worth a serious revision. read the letter →

arxiv 2505.16484 v1 pith:6H3SBN5F submitted 2025-05-22 quant-ph

classification quant-ph
keywords quantumkernelmultiplelearningmulti-viewkernel-targetalignmentlocalinformationparameterizedcircuitssupportvectormachineMfeatdataset
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 proposes L-QMVKL, a quantum kernel method that fuses view-specific quantum kernels instead of relying on a single feature representation. It argues that training each quantum base kernel with a hybrid global-local kernel-target alignment objective, and then learning the fusion weights, lets the model capture local structure that global-only quantum kernels miss. On the Mfeat handwritten-digit benchmark the method reports large accuracy gains over single-view quantum models and a 1.44% average advantage over an untrained classical Gaussian multi-view kernel at $\lambda=0.125$, $k=8$. The claimed significance is that quantum kernel methods traditionally plateau on heterogeneous data, and this is a concrete recipe for combining multiple views and local structure inside a trainable quantum kernel.

What carries the argument

The load-bearing object is the hybrid global-local kernel-target alignment $HTA(K,K^*)=(1-\lambda)LTA(K,K^*)+\lambda TA(K,K^*)$, where $TA$ is the usual kernel-target alignment against the ideal label kernel $K^*$ and $LTA$ averages alignment over local kernel matrices built from each point's $k$ nearest neighbors. Training is sequential: first, each view's parameterized-circuit parameters $\theta^m$ are updated by gradient ascent on $HTA$ using analytic gradients from the parameter-shift rule; second, the combination weights $\eta$ are optimized by alternating a quadratic program with a dynamically updated nearest-neighbor set defined through the combined quantum kernel. This two-stage scheme is what allows the model to fuse complementary views while still fitting local structure.

What would settle it

Re-run the Mfeat comparison with the classical multi-kernel given the same training budget: optimize its bandwidth by the same hybrid global-local kernel-target alignment, select $\lambda$, $k$, and $P$ on a validation split, and compare test accuracy. If the tuned classical model reaches or exceeds 91.25%, the claimed quantum advantage is an artifact of unequal optimization effort.

Watch

Extended reading notes

Core claim

The central claim is that the bottleneck of quantum kernels on heterogeneous data is single-view representation plus global-only training, and that both can be fixed in one construction. L-QMVKL assigns each of the M views its own trainable quantum kernel $\kappa_{\theta_m}^m$, trains the circuit parameters $\theta_m$ by maximizing $HTA(K,K^*)=(1-\lambda)LTA(K,K^*)+\lambda TA(K,K^*)$, and then learns combination weights $\eta$ by the same hybrid objective with nearest-neighbor sets recomputed in the fused-kernel space. On the binary Mfeat task this reaches 91.25% mean test accuracy, against 89.81% for the classical Gaussian multi-kernel and 88.13% for the same quantum multi-kernel left untrained, and the paper reports at least 10.06% higher accuracy than the best single-view global-only quantum model at $\lambda=0.25$, $k=8$, $P=6$. The paper states this as evidence that multi-view fusion and local information together improve quantum kernel learning on real data.

Load-bearing premise

The headline quantum advantage assumes that a classical Gaussian multi-kernel with bandwidth fixed to the mean pairwise distance, and never trained or tuned, is a fair baseline for quantum kernels that receive full optimization and hyperparameter selection.

Editorial extensions

If this is right

  • Multi-view quantum kernels trained this way beat every single-view quantum kernel on Mfeat; at circuit depth $P=6$ the independent multi-view advantage is at least 9.5%.
  • Local information helps beyond global-only training: on three views (FOU, ZER, MOR) the locally trained kernels outperform global-only kernels, with a peak gain of 3.81 percentage points at $k=8$, $P=6$.
  • Trained quantum base kernels outperform the untrained classical Gaussian baseline, and the assembled L-QMVKL averages 1.44% higher accuracy than the classical multi-view model at $\lambda=0.125$, $k=8$; accuracy generally rises with $k$, reaching 90.56% at $k=12$.
  • Increasing the QAOA-style circuit depth $P$ raises the kernel-target alignment value but not accuracy monotonically: mean accuracy peaks at 90.75% for $P=7$ and falls at $P=8$.
  • The construction accepts any parameterized data-encoding operation $W(x,\theta)$, so the same recipe could be applied with other ansatz families and other kernel-based classifiers.

Reading between the lines

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

  • Inference: the Section IV.E comparison pairs trained, hyperparameter-selected quantum kernels against a classical Gaussian multi-kernel whose bandwidth is not optimized, so that asymmetry, rather than a quantum-specific effect, may explain part of the 1.44% gap.
  • Inference: Sections IV.C and IV.D select $\lambda$, $k$, and $P$ by the test accuracies reported in the tables, with no separate validation split, so the headline gains are likely optimistic for unseen data.
  • Inference: the same hybrid global-local alignment objective could be tested on other trainable kernels and on kernel-based clustering, where local structure is known to matter; those settings would show whether the mechanism, not the particular SVM-plus-Mfeat combination, is responsible.
  • Inference: the divergence between rising $HTA$ and falling accuracy at $P=8$ suggests that maximizing alignment alone may overfit the kernel to the training labels, a possibility the paper does not develop.
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 proposes L-QMVKL, a quantum multi-view kernel learning method that combines view-specific trainable quantum base kernels into a single multi-kernel and classifies with an SVM. The circuit parameters of each base kernel are trained by maximizing a hybrid global-local kernel-target alignment, and the combination weights are then learned by an alternating quadratic-programming procedure. Numerical simulations on the Mfeat dataset, with features reduced by PCA, are used to support claims that multi-view fusion, local information, and the quantum model itself improve accuracy over single-view, global-only, and classical Gaussian multi-kernel baselines.

Significance. The algorithm is coherently constructed and the two-stage training scheme is a reasonable extension of multiple kernel learning and local kernel alignment to quantum kernels. The paper provides analytic gradients, the parameter-shift rule for quantum gradient estimation, and a quadratic-programming formulation for the combination weights, which are useful methodological details. If the numerical claims were established with fair baselines and proper validation, this would be a valuable benchmark contribution to quantum kernel methods. However, the current empirical support is not yet convincing: the classical baseline is untrained, hyperparameters are selected on test accuracy, and key accuracy gaps are within the reported statistical uncertainty. The significance of the proposed contribution therefore rests on a reworked experimental evaluation.

major comments (3)
  1. [§IV.E, Table V] The claim that L-QMVKL outperforms its classical counterpart by 1.44% is not supported by the reported data. The classical multi-view baseline uses Gaussian base kernels whose bandwidth is fixed to the mean Euclidean distance, with no training or tuning of the bandwidth or of the combination weights, whereas L-QMVKL optimizes the circuit parameters through hybrid global-local alignment and learns the weights via Eq. (22). The paper itself notes that untrained quantum base kernels do not consistently beat classical kernels, so the observed gain may reflect unequal optimization effort rather than a quantum-specific benefit. Moreover, with 20 repeats the unpaired standard error of the difference between 89.81±4.63% and 91.25±4.56% is about 1.45%, so the 1.44% gap is not statistically significant; no paired test or confidence interval is reported. The comparison should be made fair by giving the classical model the same optimization budget, and the difference should be reported with paired statistics.
  2. [§IV.C–D] The headline hyperparameter settings, including λ=0.125, k=8 and P=6, are selected using test-set accuracy. The text in §IV.C states that 'the optimal hyperparameters corresponding to the maximum accuracy vary across different feature views,' which shows that model selection is performed on the test set rather than on a separate validation split. This selection bias inflates the reported gains, including the 10.06% improvement quoted in the introduction and the 1.44% classical comparison in §IV.E. The authors should either fix the hyperparameters a priori, use a validation split, or report nested cross-validation results so that the reported test accuracy corresponds to a single pre-specified configuration.
  3. [§IV, Tables II–V] The experimental conclusions are drawn from 20 repeats with standard deviations of roughly 4–7%, but no statistical significance tests are reported. For example, the claimed local-information gains of 0.19% and 0.56% at λ=0.25 in Table II are far smaller than the standard errors of the means, and the 3.81% MOR-feature improvement is also within the range that could arise from sampling noise. The paper consistently uses words such as 'significant' (abstract and Section I) without supporting tests. For the central comparisons, the authors should report paired tests or bootstrap confidence intervals rather than only mean ± standard deviation.
minor comments (5)
  1. [Table IV] In the Multi-view row at P=6, the entry '80.56.94 ± 3.59' appears to be a typo and should likely read '90.56 ± 3.59'; this should be corrected because the tables are the main evidence for the accuracy claims.
  2. [§IV.A] It is not specified whether the 20 repeats use independent random training/test splits or a single fixed split; this detail is needed to interpret the reported standard deviations.
  3. [§IV.E] The classical multi-view baseline is described only as using Gaussian base kernels; the paper does not state how the view-specific classical kernels are combined (for example, equal weights or learned weights), which is necessary for a fair and reproducible comparison.
  4. [§II.B, Eq. (24)–(25)] The text says 'σx q and σx q represent the PauliX operator and the Pauli Z operator', but the second symbol should be σ_z; the Z-operator symbol is missing a subscript.
  5. [Figure 4] The captions of Fig. 4(a) and 4(b) do not explain the 'untrain' horizontal line or the meaning of the separate single-view curves; please clarify the legend and the untrained baseline.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: L-QMVKL is an empirical training and benchmarking paper whose claims rest on held-out accuracy comparisons, not on reductions of predictions to fitted inputs or on load-bearing self-citations.

full rationale

The paper contains no derivation chain in which an output is equivalent to an input by construction. The quantum base kernel in Eq. (4), the combined kernel in Eq. (6), and the alignment objectives in Eqs. (7)-(12) are defined independently of the accuracy values they are later used to report. Training in Eqs. (13)-(22) optimizes circuit parameters and combination weights against a label-based kernel-target alignment objective, and the reported accuracy is measured on a separate test set, so the results are not fitted quantities renamed as predictions. The comparison in Section IV.E against an untrained classical Gaussian baseline is statistically fragile and may overstate the quantum advantage, but that is a fairness-of-benchmark concern, not a circularity of the paper's own argument. The self-citations in the reference list (e.g., [16], [17], [20], [21]) appear only as related-work citations and do not supply a load-bearing premise, uniqueness theorem, or ansatz; removing them would not change the method or its empirical claims. No step reduces by definition to a previous fitted value or to a cited prior result by the same authors. Accordingly, there is no substantial circularity under the stated rubric.

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

The central claim relies on standard kernel-alignment assumptions, the choice of several hyperparameters tuned on test accuracy, and the representativeness of a single benchmark dataset. No new physical entities are introduced. The free-parameter count is moderate and mostly standard for empirical quantum machine learning papers.

free parameters (5)
  • lambda (hybrid weight) = 0.125 for main comparisons; grid {0, 0.0625, 0.125, 0.25, 0.5, 1} explored
    Controls the trade-off between global and local kernel alignment. The best value is selected based on test accuracy without a validation split.
  • k (number of neighbors) = 8 for main comparisons; grid {4, 6, 8, 10, 12} explored
    Defines the local neighborhood size in the local kernel alignment. The value is chosen on test accuracy, and the paper notes accuracy rises up to k=12, so selection is post hoc.
  • P (circuit depth) = 6 for main comparisons; grid {4, 5, 6, 7, 8} explored
    Number of parameterized layers in the quantum circuit. Table IV shows P=7 gives 90.75% accuracy and P=8 drops, so the depth is tuned on the test set.
  • PCA reduced dimension dr = 6
    All six views are reduced to 6 dimensions before quantum encoding. No sensitivity analysis is provided for this dimensionality choice.
  • Training and test subset sizes = 80 training and 80 test samples per repetition
    Only 160 of the 2000 Mfeat instances are used per repetition. The sampling procedure is described only briefly and may affect generalizability.
assumptions (4)
  • domain assumption Kernel-target alignment maximization is a valid surrogate for classification accuracy.
    The method optimizes hybrid global-local kernel alignment, not SVM accuracy directly. Accuracy is only evaluated after training, so the objective is assumed to correlate with generalization.
  • domain assumption k-nearest-neighbor graphs in each view capture intrinsic local structure.
    Local kernel alignment assumes that local neighborhoods are meaningful for classification. This is inherited from classical local alignment work but not independently justified for the quantum kernels used here.
  • domain assumption Parameterized quantum kernel circuits are trainable in this setting.
    Training relies on gradient descent with the parameter-shift rule. The paper does not address known issues such as exponential concentration of quantum kernels, cited as [32], which could affect the representational power of the trained kernels.
  • domain assumption The Mfeat binary classification task is representative for evaluating quantum multi-view kernel methods.
    Only one dataset is used, with a coarse binary digit split and reduced dimensions. This narrow scope makes the reported improvements hard to generalize.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantum Multi-view Kernel Learning with Local Information." pith.science (2026). https://pith.science/paper/6H3SBN5F

@misc{pith2026250516484,
  author       = {Pith},
  title        = {Pith review of: Quantum Multi-view Kernel Learning with Local Information},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6H3SBN5F}},
  note         = {Machine review of arXiv:2505.16484}
}
read the original abstract

Kernel methods serve as powerful tools to capture nonlinear patterns behind data in machine learning. The quantum kernel, integrating kernel theory with quantum computing, has attracted widespread attention. However, existing studies encounter performance bottlenecks when processing complex data with localized structural patterns, stemming from the limitation in single-view feature representation and the exclusive reliance on global data structure. In this paper, we propose quantum multi-view kernel learning with local information, called L-QMVKL. Specifically, based on the multi-kernel learning, a representative method for multi-view data processing, we construct the quantum multi-kernel that combines view-specific quantum kernels to effectively fuse cross-view information. Further leveraging local information to capture intrinsic structural information, we design a sequential training strategy for the quantum circuit parameters and weight coefficients with the use of the hybrid global-local kernel alignment. We evaluate the effectiveness of L-QMVKL through comprehensive numerical simulations on the Mfeat dataset, demonstrating significant accuracy improvements achieved through leveraging multi-view methodology and local information. Meanwhile, the results show that L-QMVKL exhibits a higher accuracy than its classical counterpart. Our work holds promise for advancing the theoretical and practical understanding of quantum kernel methods.

Figures

Figures reproduced from arXiv: 2505.16484 by the authors.

Figure 1
Figure 1. FIG. 1. Work flow of L-QMVKL [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2. Quantum circuit of [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Adjoint approach for estimating [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: FIG. 4. Mean accuracy under different (a) hybrid parameters [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5. Performance under different circuit depth [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6. Accuracy of L-QMVKL compared to its classical [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

57 extracted references · 38 canonical work pages

  1. [1]

    LeCun, Y

    Y. LeCun, Y. Bengio, and G. Hinton, Deep learning, Na- ture 521, 436 (2015)

  2. [2]

    M. I. Jordan and T. M. Mitchell, Machine learning: Trends, perspectives, and prospects, Science 349, 255 (2015)

  3. [3]

    Hofmann, B

    T. Hofmann, B. Sch¨ olkopf, and A. Smola, Kernel meth- ods in machine learning, Ann. Stat. 36, 1171 (2008)

  4. [4]

    G¨ onen and E

    M. G¨ onen and E. Alpaydın, Multiple kernel learning al- gorithms, J. Mach. Learn. Res. 12, 2211 (2011)

  5. [5]

    Shor, Algorithms for quantum computation: discrete logarithms and factoring, in Proceedings 35th Annual Symposium on Foundations of Computer Science (1994) pp

    P. Shor, Algorithms for quantum computation: discrete logarithms and factoring, in Proceedings 35th Annual Symposium on Foundations of Computer Science (1994) pp. 124–134

  6. [6]

    L. K. Grover, Quantum mechanics helps in searching for a needle in a haystack, Phys. Rev. Lett. 79, 325 (1997)

  7. [7]

    A. W. Harrow, A. Hassidim, and S. Lloyd, Quantum al- gorithm for linear systems of equations, Phys. Rev. Lett. 103, 150502 (2009)

  8. [8]

    Biamonte, P

    J. Biamonte, P. Wittek, N. Pancotti, P. Rebentrost, N. Wiebe, and S. Lloyd, Quantum machine learning, Na- ture 549, 195 (2017)

Show all 57 references
  1. [9]

    Dunjko and H

    V. Dunjko and H. J. Briegel, Machine learning & artificial intelligence in the quantum domain: a review of recent progress, Rep. Prog. Phys. 81, 074001 (2018)

  2. [10]

    E. H. Houssein, Z. Abohashima, M. Elhoseny, and W. M. Mohamed, Machine learning in the quantum realm: The state-of-the-art, challenges, and future vision, Expert Syst. Appl. 194, 116512 (2022)

  3. [11]

    Huang, X.-Y

    H.-L. Huang, X.-Y. Xu, C. Guo, G. Tian, S.-J. Wei, X. Sun, W.-S. Bao, and G.-L. Long, Near-term quantum computing techniques: Variational quantum algorithms, error mitigation, circuit compilation, benchmarking and classical simulation, Sci. China-Phys. Mech. Astron. 66, 250302 (2023)

  4. [12]

    Rebentrost, M

    P. Rebentrost, M. Mohseni, and S. Lloyd, Quantum sup- port vector machine for big data classification, Phys. Rev. Lett. 113, 130503 (2014)

  5. [13]

    Schuld, A

    M. Schuld, A. Bocharov, K. M. Svore, and N. Wiebe, Circuit-centric quantum classifiers, Phys. Rev. A 101, 032308 (2020). 10

  6. [14]

    Abbas, D

    A. Abbas, D. Sutter, C. Zoufal, A. Lucchi, A. Figalli, and S. Woerner, The power of quantum neural networks, Nat. Comput. Sci. 1, 403 (2021)

  7. [15]

    K. Beer, D. Bondarenko, T. Farrelly, T. J. Osborne, R. Salzmann, D. Scheiermann, and R. Wolf, Training deep quantum neural networks, Nat. Commun. 11, 808 (2020)

  8. [16]

    Y. Song, Y. Wu, S. Wu, D. Li, Q. Wen, S. Qin, and F. Gao, A quantum federated learning framework for classical clients, Sci. China-Phys. Mech. Astron. 67, 250311 (2024)

  9. [17]

    S. Wu, R. Li, Y. Song, S. Qin, Q. Wen, and F. Gao, Quantum-assisted hierarchical fuzzy neural network for image classification, IEEE Trans. Fuzzy Syst. 33, 491 (2025)

  10. [18]

    Bondarenko and P

    D. Bondarenko and P. Feldmann, Quantum autoencoders to denoise quantum data, Phys. Rev. Lett. 124, 130502 (2020)

  11. [19]

    Kerenidis and J

    I. Kerenidis and J. Landman, Quantum spectral cluster- ing, Phys. Rev. A 103, 042415 (2021)

  12. [20]

    L. Li, J. Li, Y. Song, S. Qin, Q. Wen, and F. Gao, An effi- cient quantum proactive incremental learning algorithm, Sci. China-Phys. Mech. Astron. 68, 1 (2025)

  13. [21]

    Wu, Y.-Q

    S.-Y. Wu, Y.-Q. Song, R.-Z. Li, S.-J. Qin, Q.-Y. Wen, and F. Gao, Resource-efficient adaptive variational quantum algorithm for combinatorial optimization problems, Adv. Quantum Technol. , 2400484 (2025)

  14. [22]

    Havl ´ ıˇ cek, A

    V. Havl ´ ıˇ cek, A. D. C´ orcoles, K. Temme, A. W. Harrow, A. Kandala, J. M. Chow, and J. M. Gambetta, Super- vised learning with quantum-enhanced feature spaces, Nature 567, 209 (2019)

  15. [23]

    Schuld and N

    M. Schuld and N. Killoran, Quantum machine learning in feature hilbert spaces, Phys. Rev. Lett. 122, 040504 (2019)

  16. [24]

    Schuld, Supervised quantum machine learning mod- els are kernel methods, arXiv preprint arXiv:2101.11020 10.48550/arXiv.2101.11020 (2021)

    M. Schuld, Supervised quantum machine learning mod- els are kernel methods, arXiv preprint arXiv:2101.11020 10.48550/arXiv.2101.11020 (2021)

  17. [25]

    Henry, S

    L.-P. Henry, S. Thabet, C. Dalyac, and L. Henriet, Quantum evolution kernel: Machine learning on graphs with programmable arrays of qubits, Phys. Rev. A 104, 032416 (2021)

  18. [26]

    J. Liu, F. Tacchino, J. R. Glick, L. Jiang, and A. Mezza- capo, Representation learning via quantum neural tan- gent kernels, PRX Quantum 3, 030323 (2022)

  19. [27]

    Incudini, M

    M. Incudini, M. Grossi, A. Mandarino, S. Vallecorsa, A. D. Pierro, and D. Windridge, The quantum path ker- nel: A generalized neural tangent kernel for deep quan- tum machine learning, IEEE Trans. Quantum Eng. 4, 1 (2023)

  20. [28]

    Shirai, K

    N. Shirai, K. Kubo, K. Mitarai, and K. Fujii, Quan- tum tangent kernel, Physical Review Research 6, 033179 (2024)

  21. [29]

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

  22. [30]

    X. Wang, Y. Du, Y. Luo, and D. Tao, Towards under- standing the power of quantum kernels in the nisq era, Quantum 5, 531 (2021)

  23. [31]

    Hubregtsen, D

    T. Hubregtsen, D. Wierichs, E. Gil-Fuster, P.-J. H. Derks, P. K. Faehrmann, and J. J. Meyer, Training quan- tum embedding kernels on near-term quantum comput- ers, Phys. Rev. A 106, 042431 (2022)

  24. [32]

    Thanasilp, S

    S. Thanasilp, S. Wang, M. Cerezo, and Z. Holmes, Expo- nential concentration in quantum kernel methods, Nat. Commun. 15, 5200 (2024)

  25. [33]

    Miyabe, B

    S. Miyabe, B. Quanz, N. Shimada, A. Mitra, T. Ya- mamoto, V. Rastunkov, D. Alevras, M. Metcalf, D. J. M. King, M. Mamouei, M. D. Jackson, M. Brown, P. Intal- lura, and J.-E. Park, Quantum multiple kernel learning in financial classification tasks (2023), arXiv:2312.00260 [quant-ph]

  26. [34]

    Fu, X.-L

    X. Fu, X.-L. Zhang, C.-H. H. Yang, and J. Qi, Ex- ploiting a quantum multiple kernel learning approach for low-resource spoken command recognition, in ICASSP 2024 - 2024 IEEE International Conference on Acous- tics, Speech and Signal Processing (ICASSP) (2024) pp. 12931–12935

  27. [35]

    S. S. Vedaie, M. Noori, J. S. Oberoi, B. C. Sanders, and E. Zahedinejad, Quantum multiple kernel learning (2020), arXiv:2011.09694 [quant-ph]

  28. [36]

    Ghukasyan, J

    A. Ghukasyan, J. S. Baker, O. Goktas, J. Carrasquilla, and S. K. Radha, Quantum-classical multiple kernel learning (2023), arXiv:2305.17707 [quant-ph]

  29. [37]

    J. Zhao, X. Xie, X. Xu, and S. Sun, Multi-view learn- ing overview: Recent progress and new challenges, Inf. Fusion 38, 43 (2017)

  30. [38]

    Z. Yu, Z. Dong, C. Yu, K. Yang, Z. Fan, and C. P. Chen, A review on multi-view learning, Front.. Comput. Sci. 19, 197334 (2025)

  31. [39]

    Yeh, T.-C

    Y.-R. Yeh, T.-C. Lin, Y.-Y. Chung, and Y.-C. F. Wang, A novel multiple kernel learning framework for heteroge- neous feature fusion and variable selection, IEEE Trans. Multimedia 14, 563 (2012)

  32. [40]

    J. Chen, Z. Chen, Z. Chi, and H. Fu, Emotion recognition in the wild with feature fusion and multiple kernel learn- ing, in Proceedings of the 16th International Conference on Multimodal Interaction (2014) pp. 508–513

  33. [41]

    W. Yan, Y. Li, and M. Yang, Towards deeper match for multi-view oriented multiple kernel learning, Pattern Recognit. 134, 109119 (2023)

  34. [42]

    Y. Wang, X. Liu, Y. Dou, Q. Lv, and Y. Lu, Multiple ker- nel learning with hybrid kernel alignment maximization, Pattern Recognit. 70, 104 (2017)

  35. [43]

    Q. Wang, Y. Dou, X. Liu, F. Xia, Q. Lv, and K. Yang, Local kernel alignment based multi-view clustering using extreme learning machine, Neurocomputing 275, 1099 (2018)

  36. [44]

    Zhang, X

    T. Zhang, X. Liu, L. Gong, S. Wang, X. Niu, and L. Shen, Late fusion multiple kernel clustering with local kernel alignment maximization, IEEE Trans. Multimedia 25, 993 (2021)

  37. [45]

    Kandala, A

    A. Kandala, A. Mezzacapo, K. Temme, M. Takita, M. Brink, J. M. Chow, and J. M. Gambetta, Hardware- efficient variational quantum eigensolver for small molecules and quantum magnets, Nature 549, 242 (2017)

  38. [46]

    Farhi, J

    E. Farhi, J. Goldstone, and S. Gutmann, A quan- tum approximate optimization algorithm (2014), arXiv:1411.4028 [quant-ph]

  39. [47]

    Hadfield, Z

    S. Hadfield, Z. Wang, B. O’gorman, E. G. Rieffel, D. Ven- turelli, and R. Biswas, From the quantum approximate optimization algorithm to a quantum alternating opera- tor ansatz, Algorithms 12, 34 (2019)

  40. [48]

    J. A. Suykens and J. Vandewalle, Least squares support vector machine classifiers, Neural Process. Lett. 9, 293 (1999)

  41. [49]

    T. Wang, D. Zhao, and S. Tian, An overview of kernel alignment and its applications, Artif. Intell. Rev. 43, 179 11 (2015)

  42. [50]

    Ruder, An overview of gradient descent optimization algorithms (2017), arXiv:1609.04747 [cs.LG]

    S. Ruder, An overview of gradient descent optimization algorithms (2017), arXiv:1609.04747 [cs.LG]

  43. [51]

    Romero, R

    J. Romero, R. Babbush, J. R. McClean, C. Hempel, P. J. Love, and A. Aspuru-Guzik, Strategies for quantum com- puting molecular energies using the unitary coupled clus- ter ansatz, Quantum Sci. Technol. 4, 014008 (2018)

  44. [52]

    Schuld, V

    M. Schuld, V. Bergholm, C. Gogolin, J. Izaac, and N. Kil- loran, Evaluating analytic gradients on quantum hard- ware, Phys. Rev. A 99, 032331 (2019)

  45. [53]

    Duin, Multiple Features, UCI Machine Learning Repository (1998)

    R. Duin, Multiple Features, UCI Machine Learning Repository (1998)

  46. [54]

    Bergholm, J

    V. Bergholm, J. Izaac, M. Schuld, C. Gogolin, S. Ahmed, V. Ajith, M. S. Alam, G. Alonso-Linaje, B. Akash- Narayanan, A. Asadi, J. M. Arrazola, U. Azad, S. Ban- ning, C. Blank, T. R. Bromley, B. A. Cordier, J. Ceroni, A. Delgado, O. D. Matteo, A. Dusko, T. Garg, D. Guala, A. Ha...

  47. [55]

    Amari, Backpropagation and stochastic gradient de- scent method, Neurocomputing 5, 185 (1993)

    S.-i. Amari, Backpropagation and stochastic gradient de- scent method, Neurocomputing 5, 185 (1993)

  48. [56]

    Preskill, Quantum computing in the nisq era and be- yond, Quantum 2, 79 (2018)

    J. Preskill, Quantum computing in the nisq era and be- yond, Quantum 2, 79 (2018)

  49. [57]

    Z. Cai, R. Babbush, S. C. Benjamin, S. Endo, W. J. Hug- gins, Y. Li, J. R. McClean, and T. E. O’Brien, Quantum error mitigation, Rev. Mod. Phys. 95, 045005 (2023)

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.