Pith. sign in

REVIEW 4 major objections 5 minor 61 references

Application of quantum machine learning using variational quantum classifier in accelerator physics

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a variational quantum classifier can predict long-term particle survival in a storage ring from one-turn trajectory features, reaching 95.3% accuracy and beating a classical neural network when training data are…

desk verdict First VQC application to dynamic aperture is a real proof-of-concept, but the claimed advantage over ANN is a 0.4-point gap with no error bars on an untuned baseline. read the letter →

arxiv 2506.06662 v1 pith:2V4AXMB6 submitted 2025-06-07 hep-lat

He-Xing Yin , Zhi-Yuan Hu , Huan-Huan Zeng , Jia-Bao Guan , Ji-ke Wang This is my paper
classification hep-lat
keywords variationalquantumclassifiermachinelearningdynamicaperturestorageringparticletrackingnoisyintermediate-scalegeneralizationacceleratorphysics
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that a variational quantum classifier (VQC) can replace expensive long-term particle tracking when determining the dynamic aperture of a storage ring. It trains the VQC on cheap one-turn trajectory data to predict which particles survive 1000 turns, then compares it with a classical artificial neural network of matched parameter count. The authors report that the VQC reaches high accuracy from fewer training samples, is at least as accurate as the ANN at every sample size tested, and retains most of its accuracy when simulated device noise is added. If true, this would cut one of the most computationally expensive steps in storage-ring lattice design, since dynamic aperture evaluation normally requires dense long-term tracking. The practical payoff is that quantum classifiers could be used where collecting training data is slow or costly, such as online accelerator tuning.

What carries the argument

The variational quantum classifier (VQC) is a parameterized quantum circuit that encodes classical features into qubit rotations, applies several strongly entangling layers made of generic single-qubit rotations and CNOT gates in a circular topology, and reads out the probability of the first qubit being in the state $|0\rangle$. The paper uses angle encoding (12 qubits, 108 parameters) and amplitude encoding (4 qubits, 36 parameters), and trains the circuit with the gradient-free COBYLA optimizer. The theoretical device used to explain the small-data advantage is a generalization bound $\mathrm{gen}(\alpha') \in O\!\left(\sqrt{T \log T / N}\right)$ for a QML model with $T$ independently parameterized gates trained on $N$ samples, which the paper contrasts with classical generalization error of order $O(\sqrt{d/N})$ where $d$ is the VC dimension.

What would settle it

Train a range of classical classifiers (for example, gradient-boosted trees, kernel methods, and multi-layer perceptrons of varying width and depth) on the same 2000-sample dataset; if any of them reaches or exceeds 95.3% accuracy while the VQC stays at 95.3%, the claim that the VQC is always at least as accurate would be contradicted. A second check is to rerun the comparison at much larger sample sizes, where the paper itself predicts the gap narrows but the VQC should remain at least as accurate as the ANN.

Watch

Extended reading notes

Core claim

On the low-energy ring of a fourth-generation light source, a VQC with angle encoding and three strongly entangling layers, trained on the transverse coordinates at the end points of the first, fifth, and last cells of a single turn, classifies 1000-turn survival with accuracy up to 95.3% at 2000 training samples, versus 94.9% for a classical ANN with the same number of parameters. The VQC reaches its plateau accuracy with fewer training samples than the ANN, and under the noise models of three different quantum devices its accuracy drops only slightly from the ideal-simulator value on the simpler dataset. The paper interprets these results as evidence that the VQC is a promising low-data tool for dynamic aperture evaluation, and that its advantage over classical models is strongest when the number of training samples is small.

Load-bearing premise

The comparison assumes that the two-hidden-layer, three-node artificial neural network with 108 parameters is an adequate classical baseline; if a properly tuned or stronger classical model closes the accuracy gap, the claim that the VQC outperforms the ANN loses its force.

Editorial extensions

If this is right

  • If the VQC generalizes from few samples as reported, dynamic aperture evaluation could be done with only about 10% of particles tracked over the full 1000 turns, cutting the dominant simulation cost.
  • The same classifier could be embedded in Bayesian optimization loops for online accelerator tuning, where each additional training sample is expensive.
  • The reported closeness of noisy-device accuracy to ideal-simulator accuracy suggests that current noisy intermediate-scale quantum devices may be usable for this classification task without error mitigation.
  • The accuracy gap between VQC and ANN narrows as training samples grow (95.3% versus 94.9% at 2000 samples), so the claimed advantage is specifically a small-data advantage rather than a ceiling in accuracy.

Reading between the lines

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

  • The comparison depends on the chosen classical baseline, so the claim that the VQC is always more accurate would be tested by trying a wider range of tuned classical classifiers on the same dataset.
  • The generalization argument via the VC dimension is not a proof for this specific ANN, since the ANN's empirical VC dimension or effective capacity is not measured; a fairer test would compare generalization curves at matched capacity.
  • The feature selection, using $[x, x', y, y']$ at only three cell endpoints, raises the question of which measurement locations carry the most predictive information and whether classical feature engineering alone explains part of the VQC advantage.
  • One testable extension is to apply the method to storage rings with different damping characteristics: the VQC advantage should shrink where survival is more predictable from short trajectories and grow where the boundary is more nonlinear.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper reports the first application of a variational quantum classifier (VQC) to dynamic aperture evaluation in accelerator physics. Using one-turn trajectory features from a diffraction-limited storage ring (the WHPS low-energy ring), the authors train VQC models with angle and amplitude encoding to classify whether particles survive 1000 turns, following the method of Jiao et al. They compare VQC accuracy against a small classical ANN, study the effect of strongly entangled layers, and test noise resilience using IBM quantum computer noise models. The main claims are that VQC converges to good accuracy with fewer training samples, that VQC accuracy is 'always higher' than ANN accuracy (e.g., 95.3% vs 94.9% at 2000 training samples), and that VQC performance remains robust under realistic device noise.

Significance. This is a novel and timely application of quantum machine learning to a concrete accelerator-physics problem where training data are expensive because they come from long-term particle tracking. The method follows an established classical approach and uses an external generalization bound, which is a strength. If the sample-efficiency and robustness claims were statistically secured, the result would be a useful demonstration with practical implications. However, the paper does not provide code, data are 'available on request,' and the small reported accuracy gap together with the high run-to-run variance in Table 1 substantially weaken the empirical support. The conceptual framework and experimental setup are sound enough that additional experiments could strengthen the case.

major comments (4)
  1. [Section IV.A, Fig. 7] The central claim that 'the accuracy of VQC is consistently higher than that of ANN' rests on a 0.4 percentage-point gap (95.3% vs 94.9%) at 2000 training samples, yet no error bars, confidence intervals, or significance tests are reported even though the text states that accuracy is a mean over ten independent training sessions. The variability implied by Table 1 (ideal-simulator accuracy 0.717 ± 0.223) shows that run-to-run standard deviations can be tens of percentage points; if similar variance applies to the complex dataset, the observed gap is far smaller than the noise scale. The authors should report per-sample-size means and error bars and, if possible, paired statistical tests across the ten sessions.
  2. [Section IV.A, Eq. (9)] The generalization bound gen(α') ∈ O(sqrt(T log T / N)) is invoked to explain why VQC outperforms ANN when N is small, but the ANN baseline has T = 108 parameters, the same order as the VQC's 108 parameters, and the argument contrasts VQC with 'complex models' with large VC dimension. No bound on the ANN's VC dimension is given, so Eq. (9) does not apply to the chosen baseline. The theoretical explanation should either be removed or the baseline complexity should be characterized.
  3. [Section IV.A, Fig. 5] The classical baseline is a two-hidden-layer, three-node-per-layer MLP with 108 parameters, chosen solely to match the VQC parameter count, with no hyperparameter tuning, regularization, or early stopping. This is an unusually weak classical model. The conclusion that VQC is superior to classical machine learning for this task requires comparison against a reasonably tuned classical model (e.g., a larger MLP, a support vector machine, or a random forest); without such a baseline, the sample-efficiency advantage is not established.
  4. [Section IV.B, Table 1] The claim of robustness to noise is not supported by the reported numbers. The no-noise accuracy is 0.717 ± 0.223, while the noisy models give 0.688 ± 0.022, 0.686 ± 0.028, and 0.692 ± 0.018; the mean differences are much smaller than the no-noise standard deviation, and the error bars overlap. The text says the difference 'is not significant,' but with these statistics one cannot even assert a small decrease. The authors should report paired results on the same ten training configurations or use a statistical test, and they should discuss why the no-noise standard deviation is an order of magnitude larger than those of the noisy models.
minor comments (5)
  1. [Section II.A, Fig. 4 caption] There are typographical errors: 'combinating' should be 'combining' in Section II.A, and the Fig. 4 caption contains a duplicated 'the better' in 'the better the better objective value.'
  2. [References [36] and [38]] Reference [36] ('Heuristic ansatze and trade-offs (2023)') lacks author and journal information, and reference [38] (Cheng et al.) should include volume and page numbers.
  3. [Equation (8)] The definition of p_i is ambiguous: for the standard binary cross-entropy form, p_i should be explicitly defined as the model's predicted probability that Y_i = 1 given the features.
  4. [Section IV.A, paragraph after Fig. 7] The phrase 'The current researches suggest' is grammatically incorrect and should be 'Current research suggests'; the same paragraph would benefit from a more precise statement about which phenomenon has been observed in the cited studies.
  5. [Abstract and Conclusion] The phrase 'the accuracy of the variational quantum classifier is always higher than that of an artificial neural network' is too strong given the statistical concerns in the major comments; it should be qualified to reflect the observed point estimate and its uncertainty.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the VQC accuracy is measured against independently tracked long-term survival labels, and the generalization bound is an external result.

full rationale

The paper's central claim—that a VQC trained on one-turn trajectory features classifies 1000-turn survival with reported accuracy up to 95.3%—is evaluated against independently generated long-term tracking labels, so the target quantity is not defined through the model's fitted parameters. The data-encoding, variational-circuit, and measurement steps (Eqs. 1–7) are standard constructions and do not contain the final accuracy as an input. The generalization bound in Eq. (9) is imported from the external, independently published result of Caro et al. [57]; the paper does not fit that bound to the measured accuracies, nor does it use the bound to define the labels. The comparison with the ANN uses parameter counts matched following Refs. [20,21,28], but the weakness or untuned nature of the ANN baseline is a fairness or correctness concern, not a circularity: the VQC's reported accuracy is not forced by how the ANN was constructed. The selection of three strongly entangled layers and the three-interval sampling densities was tuned on the same data, which can inflate claimed performance, but this is dataset reuse and hyperparameter selection rather than a reduction of the prediction to its inputs. The only self-citations (e.g., Ref. [10], whose author list overlaps with the present corresponding author) are ordinary literature references and carry no load-bearing justificatory weight. No step in the paper reduces by construction to a fitted parameter, a self-citation chain, or a renamed known result. The skeptical concern about the 0.4-percentage-point gap and missing error bars in Fig. 7 is a statistical-evidence issue, not circularity.

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

The paper introduces no new physical entities or conserved quantities. Its central claims rest on external tools and cited methods: the trajectory-classification scheme of Jiao et al. [46], Qiskit simulators and noise models, Accelerator Toolbox tracking, and a small set of hand-chosen hyperparameters. The adequacy of the ANN baseline and the transferability of the one-turn-feature assumption are the main unverified premises.

free parameters (4)
  • Number of strongly entangled layers = 3
    Chosen after an accuracy sweep on the simple dataset (Figure 6), then used for the complex dataset and noise studies; this trades off accuracy and simulation time and shapes the reported results.
  • Radial interval sampling densities = Three intervals with increasing density, surviving 2546 and lost 2454
    Initial particle densities were adjusted so that the number of surviving particles is close to the number of lost particles, balancing the label ratio and affecting the classification boundary.
  • ANN architecture = Two hidden layers, three nodes each, 108 trainable variables
    The classical baseline was selected to match the VQC's number of training parameters; the comparison outcome depends on this particular untuned architecture.
  • Measurement shots = 1024
    Each quantum circuit outcome was estimated from 1024 measurement shots, which sets the statistical precision of the inferred output probabilities.
assumptions (4)
  • domain assumption One-turn short-term trajectory features [x, x', y, y'] at selected cell endpoints are sufficient to predict 1000-turn stability.
    Introduced in Section IV setup by restricting short-term tracking to one turn and selecting transverse endpoint coordinates as features; the validity of this shortcut is assumed from Jiao et al. [46].
  • domain assumption The generalization bound in Eq. (9), cited from Caro et al. [57], applies to this VQC and supports the small-sample advantage over ANN.
    Invoked in Section IV.A to argue that VQC generalizes better when N is small; the paper does not verify the bound's assumptions for its specific circuit or quantify the ANN's VC dimension.
  • domain assumption Accelerator Toolbox particle tracking provides an adequate model of WHPS nonlinear beam dynamics.
    All long-term and short-term tracking uses AT [43], and the paper does not benchmark AT predictions against measured or other simulation data for the WHPS ring.
  • domain assumption Qiskit Aer noise models for ibmq_lima, ibmq_belem, and ibm_lagos faithfully represent real NISQ device errors.
    The noise-robustness conclusion in Section IV.B is based on these simulated models, with only one illustrative comparison of a separate 3-qubit circuit on ibmq_belem.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Application of quantum machine learning using variational quantum classifier in accelerator physics." pith.science (2026). https://pith.science/paper/2V4AXMB6

@misc{pith2026250606662,
  author       = {Pith},
  title        = {Pith review of: Application of quantum machine learning using variational quantum classifier in accelerator physics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2V4AXMB6}},
  note         = {Machine review of arXiv:2506.06662}
}
read the original abstract

Quantum machine learning algorithms aim to take advantage of quantum computing to improve classical machine learning algorithms. In this paper, we have applied a quantum machine learning algorithm, the variational quantum classifier for the first time in accelerator physics. Specifically, we utilized the variational quantum classifier to evaluate the dynamic aperture of a diffraction-limited storage ring. It has been demonstrated that the variational quantum classifier can achieve good accuracy much faster than the classical artificial neural network, with the statistics of training samples increasing. And the accuracy of the variational quantum classifier is always higher than that of an artificial neural network, although they are very close when the statistics of training samples reach high. Furthermore, we have investigated the impact of noise on the variational quantum classifier, and found that the variational quantum classifier maintains robust performance even in the presence of noise.

Figures

Figures reproduced from arXiv: 2506.06662 by the authors.

Figure 1
Figure 1. Schematic diagram of the VQC model. X represents the dataset, Φ (⃗x)represents the quantum gate operation that maps clas￾sical data to the quantum state of qubits, W  ⃗θ  represents the opti￾mizable quantum computing layer consisting of quantum gates with adjustable parameters, which allows the parameters to be continu￾ously adjusted during the training process to minimize the loss func￾tion. Yˆ represents the cla… view at source ↗
Figure 2
Figure 2. Circuit structural diagram of a VQC model with angle en [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Schematic diagram of using ML models to predict the dy [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: The schematic diagram of COBYLA in a two-dimensional [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 7
Figure 7. Figure 7: Accuracy of angle encoding (red), amplitude encoding [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 9
Figure 9. Figure 9: Comparison of the results of a 3-qubits circuit on an ideal [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 8
Figure 8. Figure 8: The results of VQC and ANN trained with different train [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

61 extracted references · 36 canonical work pages

  1. [1]

    M. Song, X. Huang, L. Spentzouris, and Z. Zhang, Stor- age ring nonlinear dynamics optimization with multi-objective multi-generation Gaussian process optimizer.Nucl. Instrum. Methods Phys. Res. Sect. A976, 164273 (2020). doi: 10.1016/j.nima.2020.164273

    work page arXiv 2020

  2. [2]

    Wang, K.-M

    G.-L. Wang, K.-M. Chen, S.-W. Wang, et al., Beam based alignment using a neural network.Nuclear Science and Tech- niques.35, 75 (2024). doi: 10.1007/s41365-024-01436-y

    work page doi:10.1007/s41365-024-01436-y 2024

  3. [3]

    Y . Jiao, F. Chen, P. He, et al., Modification and optimization of the storage ring lattice of the High Energy Photon Source.Ra- diation Detection Technology and Methods.4, 415-424 (2020). doi: 10.1007/s41605-020-00189-7

    work page doi:10.1007/s41605-020-00189-7 2020

  4. [4]

    J. Wan, P. Chu, and Y . Jiao, Neural network-based mul- tiobjective optimization algorithm for nonlinear beam dy- namics.Phys. Rev. Accel. Beams23, 081601 (2020). doi: 10.1103/PhysRevAccelBeams.23.081601

    work page doi:10.1103/physrevaccelbeams.23.081601 2020

  5. [5]

    Lu et al., Demonstration of machine learning-enhanced multi-objective optimization of ultrahigh-brightness lattices for 4th-generation synchrotron light sources.Nucl

    Y . Lu et al., Demonstration of machine learning-enhanced multi-objective optimization of ultrahigh-brightness lattices for 4th-generation synchrotron light sources.Nucl. Instrum. Methods Phys. Res. Sect. A1050, 168192 (2023). doi: 10.1016/j.nima.2023.168192

    work page arXiv 2023

  6. [6]

    Zhang, M

    Z. Zhang, M. Song, and X. Huang, Online accelerator op- timization with a machine learning-based stochastic algo- rithm.Mach. Learn.: Sci. Technol.2, 015014 (2021). doi: 10.1088/2632-2153/abc81e

    work page doi:10.1088/2632-2153/abc81e 2021

  7. [7]

    J. Li, X. Huang, P. Pianetta, and Y . Liu, Machine-and-data in- telligence for synchrotron science.Nat. Rev. Phys.3, 766-768 (2021). doi: 10.1038/s42254-021-00397-0

    doi:10.1038/s42254-021-00397-0 2021

  8. [8]

    S. C. Leemann et al., Demonstration of Machine Learning- Based Model-Independent Stabilization of Source Properties in Synchrotron Light Sources.Phys. Rev. Lett.123, 194801 (2019). doi: 10.1103/PhysRevLett.123.194801

    work page doi:10.1103/physrevlett.123.194801 2019

Show all 61 references
  1. [9]

    Scheinker, Adaptive Machine Learning for Robust Di- agnostics and Control of Time-Varying Particle Accelerator Components and Beams.Information12, 161 (2021)

    A. Scheinker, Adaptive Machine Learning for Robust Di- agnostics and Control of Time-Varying Particle Accelerator Components and Beams.Information12, 161 (2021). doi: 10.3390/info12040161

    2021 doi

  2. [10]

    P. Wang, K. Ye, X. Hao, and J. Wang, Combining multi- objective genetic algorithm and neural network dynamically for the complex optimization problems in physics.Sci. Rep. 13, 880 (2023). doi: 10.1038/s41598-023-27478-7

    2023 doi

  3. [11]

    L.-Y . Zhou, H. Zha, J.-R. Shi, et al., A non-invasive diagnos- tic method of cavity detuning based on a convolutional neural network.Nuclear Science and Techniques.33, 94 (2022). doi: 10.1007/s41365-022-01069-z

    2022 doi

  4. [12]

    Z. An, D.L. Zhou, Deep reinforcement learning for quan- tum gate control.Europhysics Letters.126, 60002 (2019). doi: 10.1209/0295-5075/126/60002

    2019 doi

  5. [13]

    B. Duan, J. Yuan, C.-H. Yu, et al., A survey on HHL algorithm: From theory to application in quantum ma- chine learning.Physics Letters A384, 126595 (2020). doi: 10.1016/j.physleta.2020.126595

    2020

  6. [14]

    S. L. Wu and S. Yoo, Challenges and opportunities in quantum machine learning for high-energy physics.Nat. Rev. Phys.4, 143-144 (2022). doi: 10.1038/s42254-022-00425-7

    2022 doi

  7. [15]

    Batra et al., Quantum Machine Learning Algorithms for Drug Discovery Applications.JCIM61, 2641-2647 (2021)

    K. Batra et al., Quantum Machine Learning Algorithms for Drug Discovery Applications.JCIM61, 2641-2647 (2021). doi: 10.1021/acs.jcim.1c00166

    2021 doi

  8. [16]

    O. A. V on Lilienfeld, Quantum machine learning in chemical compound space.Angew. Chem. Int. Ed.57, 4164-4169 (2018). doi: 10.1002/anie.201709686

    2018 doi

  9. [17]

    Zhang, J

    R. Zhang, J. Wang, N. Jiang, H. Li, and Z. Wang, Quan- tum support vector machine based on regularized New- ton method.Neural Networks151, 376-384 (2022). doi: 10.1016/j.neunet.2022.03.043

    2022 doi

  10. [18]

    Z. Li, X. Liu, N. Xu, et al., Experimental Realization of a Quantum Support Vector Machine.Physical Review Letters 114, 140504 (2015). doi: 10.1103/PhysRevLett.114.140504

    2015 doi

  11. [19]

    Li, D.-L

    W. Li, D.-L. Deng., Recent advances for quantum classifiers, Science China Physics.Mechanics & Astronomy65, 220301 (2021). doi: 10.1007/s11433-021-1793-6

    2021 doi

  12. [20]

    Blance and M

    A. Blance and M. Spannowsky, Quantum machine learn- ing for particle physics using a variational quantum clas- sifier.J. High Energy Phys.2021, 1-20 (2021). doi: 10.1007/JHEP02(2021)212

    2021 doi

  13. [21]

    Gianelle et al., Quantum Machine Learning for b-jet charge identification.J

    A. Gianelle et al., Quantum Machine Learning for b-jet charge identification.J. High Energy Phys.2022, 14 (2022). doi: 10.1007/JHEP08(2022)014

    2022 doi

  14. [22]

    S. L. Wu et al., Application of quantum machine learning using the quantum kernel algorithm on high energy physics analysis at the LHC.Phys. Rev. Res.3, 033221 (2021). doi: 10.1103/PhysRevResearch.3.033221

    2021 doi

  15. [23]

    Liu, P., Rebentrost, Quantum machine learning for quantum anomaly detection.Physical Review A97, 042315 (2018)

    N. Liu, P., Rebentrost, Quantum machine learning for quantum anomaly detection.Physical Review A97, 042315 (2018). doi: 10.1103/PhysRevA.97.042315

    2018 doi

  16. [24]

    D. C. Ciresan, U. Meier, and J. Schmidhuber, Multi-column deep neural networks for image classification.2012 IEEE Con- ference on Computer Vision and Pattern Recognition.3642- 3649(2012). doi: 10.1109/CVPR.2012.6248110

    2012

  17. [25]

    Krizhevsky, I

    A. Krizhevsky, I. Sutskever, and G. E. Hinton, Imagenet clas- sification with deep convolutional neural networks.Commun. ACM60, 84-90 (2017), doi: 10.1145/3065386

    2017 doi

  18. [26]

    Simonyan and A

    K. Simonyan and A. Zisserman, Very deep convolutional networks for large-scale image recognition.arXiv preprint arXiv:1409.1556 (2014), doi: 10.48550/arXiv.1409.1556

    2014 doi arXiv

  19. [27]

    LeCun, Y

    Y . LeCun, Y . Bengio, and G. Hinton, Deep learning.Nature 521, 436-444 (2015), doi: 10.1038/nature14539

    2015 doi

  20. [28]

    S. Y .-C. Chen, T.-C. Wei, C. Zhang, H. Yu, and S. Yoo, Quan- tum convolutional neural networks for high energy physics data analysis.Phys. Rev. Res.4, 013231 (2022), doi: 10.1103/Phys- RevResearch.4.013231

    2022 doi

  21. [29]

    Gao, Z.-Y

    X. Gao, Z.-Y . Zhang, and L.-M. Duan, A quantum machine learning algorithm based on generative models.Sci. Adv.4, eaat9004 (2018), doi: 10.1126/sciadv.aat9004

    2018 doi

  22. [30]

    Benedetti, E

    M. Benedetti, E. Lloyd, S. Sack, and M. Fiorentini, Pa- rameterized quantum circuits as machine learning models. Quantum Sci. Technol.4, 043001 (2019), doi: 10.1088/2058- 9565/ab4eb5

    2019 doi

  23. [31]

    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.Phys. Rev. A103, 032430 (2021), doi: 10.1103/PhysRevA.103.032430

    2021 doi

  24. [32]

    Hu, et al., A theoretical framework for quantum image rep- resentation and data loading scheme.Science China Informa- tion Sciences57, 1-11 (2014), doi: 10.1007/s11432-013-4866- x

    B. Hu, et al., A theoretical framework for quantum image rep- resentation and data loading scheme.Science China Informa- tion Sciences57, 1-11 (2014), doi: 10.1007/s11432-013-4866- x

    2014 doi

  25. [33]

    Huang, et al., Power of data in quantum machine learn- ing.Nat

    H.-Y . Huang, et al., Power of data in quantum machine learn- ing.Nat. Commun.12, 2631 (2021), doi: 10.1038/s41467-021- 22539-9

    2021 doi

  26. [34]

    S. Sim, P. D. Johnson, and A. Aspuru-Guzik, Expressibility and Entangling Capability of Parameterized Quantum Circuits for Hybrid Quantum-Classical Algorithms.Adv. Quantum Tech- 10 nol.2, 1900070 (2019), doi: 10.1002/qute.201900070

    2019 doi

  27. [35]

    R. D. M. Simões, et al., Experimental Evaluation of Quantum Machine Learning Algorithms.IEEE Access11, 6197-6208 (2023), doi: 10.1109/ACCESS.2023.3236409

    2023

  28. [36]

    Heuristic ansatze and trade-offs (2023)

    2023

  29. [37]

    A CNOT gate be- tween multiphoton qubits encoded in two cavities.Nature com- munications,9, (1): 652 (2018)

    Rosenblum S, Gao Y Y , Reinhold P, et al. A CNOT gate be- tween multiphoton qubits encoded in two cavities.Nature com- munications,9, (1): 652 (2018)

    2018

  30. [38]

    Noisy intermediate-scale quantum computers.Frontiers of Physics,18, (2): 21308 (2023)

    Cheng B, Deng X H, Gu X, et al. Noisy intermediate-scale quantum computers.Frontiers of Physics,18, (2): 21308 (2023)

    2023

  31. [39]

    Wille, R

    R. Wille, R. V . Meter, and Y . Naveh, IBM’s Qiskit Tool Chain: Working with and Developing for Real Quantum Computers. 2019 Design, Automation & Test in Europe Conference & Ex- hibition (DATE),1234-1240, (2019)

    2019

  32. [40]

    Acasiete, F

    F. Acasiete, F. P. Agostini, J. K. Moqadam, and R. Por- tugal, Implementation of quantum walks on IBM quan- tum computers.Quantum Inf. Process.19, 426 (2020), doi: 10.1007/s11128-020-02938-5

    2020 doi

  33. [41]

    L. Yang, Y . Li, W. Guo, and S. Krinsky, Multiobjective opti- mization of dynamic aperture.Phys. Rev. ST Accel. Beams14, 054001 (2011)

    2011

  34. [42]

    Tan, S.-Q

    L.-Y . Tan, S.-Q. Tian, X.-Z. Liu, et al., ID effects on beam dy- namics in the SSRF-U storage ring.Nuclear Science and Tech- niques.34, 174 (2023).doi: 10.1007/s41365-023-01322-z

    2023 doi

  35. [43]

    Terebilo, Accelerator Toolbox for MATLAB

    A. Terebilo, Accelerator Toolbox for MATLAB. Report No. SLAC-PUB-8732, (2001)

    2001

  36. [44]

    Kranj ˇcevi´c, B

    M. Kranj ˇcevi´c, B. Riemann, A. Adelmann, et al., Multiob- jective optimization of the dynamic aperture using surrogate models based on artificial neural networks.Phys. Rev. Ac- cel. Beams24, 014601 (2021).doi: 10.1103/PhysRevAccel- Beams.24.014601

    2021 doi

  37. [45]

    Arpaia, et al., Machine learning for beam dynamics stud- ies at the CERN Large Hadron Collider.Nucl

    P. Arpaia, et al., Machine learning for beam dynamics stud- ies at the CERN Large Hadron Collider.Nucl. Instrum. Methods Phys. Res., Sect. A985, 164652 (2021), doi: 10.1016/j.nima.2020.164652

    2021

  38. [46]

    Wan and Y

    J. Wan and Y . Jiao, Machine learning enabled fast evaluation of dynamic aperture for storage ring accelerators.New J. Phys. 24, 063030 (2022), doi: 10.1088/1367-2630/ac77ac

    2022 doi

  39. [47]

    Y . Zhou, X. Wang, M. Zhang, et al., MPCE: A Maximum Prob- ability Based Cross Entropy Loss Function for Neural Net- work Classification.IEEE Access7, 146331-146341 (2020), doi: 10.1109/ACCESS.2019.2946264

    2020

  40. [48]

    A. A. K. Ismaeel, E. H. Houssein, D. Oliva, and M. Said, Gradient-Based Optimizer for Parameter Extraction in Pho- tovoltaic Models.IEEE Access9, 13403-13416 (2021), doi: 10.1109/ACCESS.2021.3052153

    2021

  41. [49]

    D. P. Kingma and J. Ba, Adam: A method for stochas- tic optimization.arXiv preprint,1412.6980,(2014)doi: 10.48550/arXiv.1412.6980

    2014 doi arXiv

  42. [50]

    Zhang, Z.-G

    L.-Y . Zhang, Z.-G. Gui, P.-C. Zhang, et al., Aperture shape optimization in intensity-modulated radiation therapy plan- ning,Nuclear Science and Techniques34, 140 (2023) doi: 10.1007/s41365-023-01300-5

    2023 doi

  43. [51]

    M. J. D. Powell, Direct search algorithms for optimiza- tion calculations.Acta Numer .7, 287-336, (1998) doi: 10.1017/S0962492900002841

    1998 doi

  44. [52]

    Li, et al., Project of Wuhan Photon Source inProceedings of the 12th International Particle Accelerator Conference (IPAC 2021), Online Conference, 346-349 (2021)

    H. Li, et al., Project of Wuhan Photon Source inProceedings of the 12th International Particle Accelerator Conference (IPAC 2021), Online Conference, 346-349 (2021)

    2021

  45. [53]

    Buitinck, et al., API design for machine learning soft- ware: experiences from the scikit-learn project.arXiv preprint 1309.0238(2013), doi: 10.48550/arXiv.1309.0238

    L. Buitinck, et al., API design for machine learning soft- ware: experiences from the scikit-learn project.arXiv preprint 1309.0238(2013), doi: 10.48550/arXiv.1309.0238

    2013 doi arXiv

  46. [54]

    The VC-Dimension

    S. Shalev-Shwartz and S. Ben-David, Understanding Machine Learning: From Theory to Algorithms. Cambridge University Press,Chapter 6: "The VC-Dimension", pp. 67–85;(2014)

    2014

  47. [55]

    Model Assessment and Selection

    T. Hastie, R. Tibshirani and J. Friedman, The Elements of Statistical Learning: Data Mining, Inference, and Prediction. Springer,Chapter 7: "Model Assessment and Selection", pp. 219–259. (2009)

    2009

  48. [56]

    Guan, et al., Quantum machine learning in high energy physics.Mach

    W. Guan, et al., Quantum machine learning in high energy physics.Mach. Learn.: Sci. Technol.2, 011003, (2021) doi: 10.1088/2632-2153/abc17d

    2021 doi

  49. [57]

    M. C. Caro, et al., Generalization in quantum machine learning from few training data.Nat. Commun.13, 4919 (2022), doi: 10.1038/s41467-022-32550-3

    2022 doi

  50. [58]

    A. W. Cross, L. S. Bishop, S. Sheldon, P. D. Nation, and J. M. Gambetta, Validating quantum computers using random- ized model circuits.Phys. Rev. A100, 032328 (2019), doi: 10.1103/PhysRevA.100.032328

    2019 doi

  51. [59]

    Roussel, et al., Bayesian optimization algorithms for accel- erator physics.Phys

    R. Roussel, et al., Bayesian optimization algorithms for accel- erator physics.Phys. Rev. Accel. Beams27, 084801 (2024), doi: 10.1103/PhysRevAccelBeams.27.084801

    2024 doi

  52. [60]

    Roussel, et al., Multiobjective Bayesian optimization for on- line accelerator tuning.Phys

    R. Roussel, et al., Multiobjective Bayesian optimization for on- line accelerator tuning.Phys. Rev. Accel. Beams24, 062801 (2021), doi: 10.1103/PhysRevAccelBeams.24.062801

    2021 doi

  53. [61]

    Morita, et al., Accelerator tuning method using autoencoder and Bayesian optimization.Nucl

    Y . Morita, et al., Accelerator tuning method using autoencoder and Bayesian optimization.Nucl. Instrum. Methods Phys. Res. A1057, 168730 (2023), doi: 10.1016/j.nima.2023.168730

    2023

Pith tools

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