Pith. sign in

REVIEW 5 major objections 5 minor 64 references

Genetic Transformer-Assisted Quantum Neural Networks for Optimal Circuit Design

T0 review · 5 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 A transformer encoder paired with an NSGA-II genetic search finds quantum classifiers that match or exceed published accuracies while using far fewer gates.

desk verdict A reasonable hybrid-pipeline idea with clearly presented circuit results, but the headline accuracy claim rests on uncontrolled baseline comparisons that would likely not survive a fair re-run. read the letter →

arxiv 2506.09205 v1 pith:55GYJIQA submitted 2025-06-10 quant-ph cs.NE

Haiyan Wang This is my paper
classification quant-phcs.NE
keywords quantummachinelearningneuralnetworkstransformerencoderNSGA-IIgeneticalgorithmvariationalcircuitsNISQdevicesFisherinformationParetomulti-objectiveoptimization
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 claims that a classical transformer encoder, a shallow variational quantum circuit, and a multi-objective genetic search can together produce quantum classifiers that match or beat the accuracies of published quantum models on four standard benchmarks while using far fewer gates. The reason this matters is that gate count is the dominant cost on noisy intermediate-scale quantum devices: shorter circuits are more likely to run before decoherence destroys the computation. The paper further claims that Fisher information spectra of the trained hybrid models show the networks avoid the barren-plateau regime, with the strongest curvature concentrating in the quantum part of the model as qubit count grows. If correct, the framework offers a practical template for quantum machine learning on near-term hardware.

What carries the argument

The load-bearing object is the GTQNN pipeline itself. Its classical stage is a transformer encoder: each input token is linearly projected to queries, keys, and values; self-attention re-weights the tokens into a content-adaptive kernel; and average pooling followed by a linear layer produces a feature vector whose length equals the number of qubits. Its quantum stage is a shallow circuit of Hadamard gates, RY rotations whose angles come from the transformer output, and CNOT gates whose placement is encoded in a binary chromosome of length N(N−1)/2. NSGA-II evolves a population of these chromosomes under two objectives—classification accuracy and total primitive-gate count—and returns a Pareto front of circuits, of which the ten most accurate are retrained and the best retained. A complementary diagnostic partitions the empirical Fisher information matrix into transformer and QNN blocks and measures the fraction of each leading eigenvector's energy in each block, which is how the paper argues the model avoids barren plateaus.

What would settle it

Run the three baseline models cited in the paper on the exact train/test splits, class subsets, and training budgets used here; if any of them reaches the same accuracy with no more gates, the paper's central comparison fails. A complementary check is to compute the variance of parameter gradients over many random initializations of the found circuits: exponential decay in qubit count would indicate a flat training landscape even if the reported curvature eigenvalues look healthy.

Watch

Extended reading notes

Core claim

On the paper's own terms, the claim is that the GTQNN pipeline—transformer compression plus NSGA-II circuit search—finds Pareto-optimal circuits that match or exceed the best published QNN accuracies on Iris, Breast Cancer Wisconsin, MNIST (digits 1, 2, 3), and Heart Disease while using substantially fewer primitive gates. The reported numbers support the claim: a 3-qubit GTQNN reaches perfect Iris accuracy with 8 gates, Breast Cancer accuracy stays near 0.974 with 10–12 gates, MNIST accuracy stays above 0.95 with 8–64 gates, and Heart Disease accuracy reaches 0.985 with 10 gates; the cited baselines use more gates for comparable or lower accuracy. The paper also claims that the empirical Fisher spectra of the trained models have leading eigenvalues of order one and that the leading curvature directions concentrate increasingly in the QNN subspace as qubit count grows, which it reads as evidence that the models are far from barren plateaus and that the transformer has already compressed the data.

Load-bearing premise

The claim rests on the premise that the accuracy numbers quoted from other papers are fairly comparable even though those papers used different data splits, different class sets (for example, only three MNIST digits here), and different training procedures; it also rests on the premise that a single trained model's curvature spectrum can reveal whether training landscapes are flat.

Editorial extensions

If this is right

  • On near-term devices, the studied classification problems become feasible with 3 to 10 qubits and no more than a few tens of primitive gates, because the transformer absorbs the input-dimensionality burden.
  • The Pareto front makes the accuracy-versus-hardware trade-off an explicit design choice: on MNIST, raising the qubit budget from 3 to 10 multiplies the gate count by eight while accuracy moves only from 0.973 to 0.993, so cheap circuits give up almost nothing.
  • A Fisher spectrum with order-one leading eigenvalues and quantum-dominated leading modes can serve as a practical trainability check for hybrid models of this type.
  • Qubit count no longer has to scale with raw feature dimension, since the transformer output is matched to the qubit budget rather than to the input size.

Reading between the lines

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

  • A natural extension of the same architecture is to swap the accuracy fitness for regression or generative objectives, since nothing in the pipeline is classification-specific.
  • An ablation that would sharpen the claim is to compare the genetic search against randomly sampled circuits at the same gate budgets, which would show how much of the gain comes from NSGA-II as opposed to the transformer compression.
  • A complementary trainability diagnostic to the Fisher spectrum would be a direct measurement of gradient variance over random initializations; if it decays exponentially in qubit count, the found circuits would still face barren-plateau-like behavior.
  • Extending the search to hardware-aware costs, such as two-qubit-gate depth and qubit routing, would test whether the Pareto fronts persist under real device noise.
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

5 major / 5 minor

Summary. The paper introduces GTQNNs, a hybrid quantum-classical model in which a transformer encoder compresses classical features into a small vector that is fed into a shallow variational quantum circuit, whose architecture is optimized by the NSGA-II genetic algorithm under two objectives: classification accuracy and primitive gate count. Experiments are reported on Iris, Breast-Cancer Wisconsin, MNIST (restricted to digits 1-3), and Heart Disease, and the results are compared with accuracies quoted from three prior papers. The paper further analyzes the empirical Fisher information spectrum of the trained models and claims that the networks avoid barren plateaus and that curvature concentrates in the QNN subspace as the qubit count grows. The central claim is that GTQNNs match or exceed state-of-the-art quantum models while using many fewer gates.

Significance. If the central claim were properly supported, GTQNNs would be a useful contribution to resource-efficient quantum machine learning on NISQ devices: the combination of a learned classical compressor with evolutionary circuit search is a reasonable idea, and the explicit multi-objective treatment of gate count is well motivated. The paper also ships no code and gives no reproducible experimental protocol, so its value currently rests entirely on the reported tables and figures. The Fisher-spectrum diagnostic is an interesting idea but is not validated and, as written, contains a direct contradiction. The proposed framework is coherent enough that a substantially revised version with controlled baselines and rigorous statistics could be publishable, but the present evidence does not support the headline claims.

major comments (5)
  1. [§4.4, Tables 3–4] The central accuracy claim is based on uncontrolled comparisons to accuracies quoted from other papers. For MNIST, the GTQNN is trained only on digits 1, 2, and 3 (Section 4.4), whereas the EfficientSU2 baseline [4] is quoted as 0.85 with no statement of its class set, train/test split, or preprocessing; if that baseline is evaluated on the full 10-class task, the comparison is invalid. The same problem affects Tables 1, 2, and 4, where the baseline numbers from [2] and [30] come from different architectures, different feature pre-processing (e.g., PCA to 7 features), and unknown splits. Without re-running these baselines under the exact GTQNN task configuration, the abstract's claim that GTQNNs 'match or exceed state-of-the-art quantum models' is unsupported.
  2. [§4.1, Tables 1–4] No error bars, repeated-seed statistics, or statistical tests are reported anywhere. Every accuracy in Tables 1–4 is a single number from a single run of a stochastic pipeline that includes random initialization, NSGA-II variation operators, and 1024-shot QASM sampling. It is therefore impossible to know whether the reported differences, e.g., 0.9737 repeated across many qubit counts in Table 2 or the 0.8829/0.8488 drop in Table 4, are real effects or sampling noise. The authors should report means and standard deviations over multiple independent runs and, where appropriate, significance tests or confidence intervals.
  3. [§5.2] The Fisher-spectrum analysis contains an internal contradiction and an unvalidated diagnostic assumption. In the paragraph after Figures 7 and 8, the text first states that the model is 'far from a barren plateau,' then immediately writes 'so TQNN does sit in a barren plateau.' This is not a minor typo: the two statements assert opposite conclusions. More fundamentally, the empirical Fisher matrix evaluated at one trained parameter point cannot diagnose barren-plateau behavior, which is defined by the expected variance of gradients over random initializations and training trajectories. The additional claim that the leading curvature directions 'increasingly align with the quantum subspace' is computed on the trained model and used as evidence of data condensation, but no comparison to an untrained or randomly initialized model, and no theoretical link between this eigen-spectrum and barren plateaus, is provided.
  4. [§3.1] The paper states that 'TQNN-assisted NSGA-II improves prediction accuracies compared with the classical version,' but no experiment comparing GTQNN against NSGA-II without the transformer, or against a non-genetic training baseline, is reported. This sentence is therefore an unsupported claim, not a result. Such an ablation is necessary to attribute any accuracy gain to the genetic search or to the transformer component.
  5. [§3.2, Eq. (3.1)] The resource-efficiency claim is partly circular by construction: gate count is one of the two NSGA-II objectives, so the algorithm is explicitly selecting circuits that minimize gates. The comparison with baselines should therefore be framed as 'GTQNN circuits are optimized for gate count,' not as an independent finding that GTQNNs happen to require fewer gates. In addition, the tables compare GTQNN 'gate count' (primitive gates) with baseline values sometimes described as 'parameters' or 'gates (parameters),' which are not necessarily the same quantity; these differences must be clarified before the gate savings can be assessed.
minor comments (5)
  1. [Figure 6 caption] The caption contains a typo: 'Heart Diseast dataset' should be 'Heart Disease dataset.'
  2. [§4.2] The text refers to 'Table 3' and 'Table 4' when describing optimal circuits for Iris and Breast Cancer, but the circuits are shown in Figures 3 and 4; the table/figure references should be corrected.
  3. [§5.1] There is a duplicated phrase: 'the notion of the contribution of contribution' should be 'the notion of the contribution.'
  4. [§4.1] The experimental section omits essential details needed for reproducibility: the transformer architecture hyperparameters (number of layers, heads, d_model, embedding size), the exact train/test splits for each dataset, the number of independent runs, and the random seeds. The statement that 'code will be available upon request' is not a sufficient data/code availability statement for a computational paper.
  5. [References] References [19] and [63] are duplicates of the same work; one should be removed and the citation list cleaned.

Circularity Check

2 steps flagged · score 6.0 of 10

Partial circularity: the low-gate-count result is the NSGA-II objective restated, and the Fisher 'far from barren plateau' conclusion is read off the fitted model rather than the landscape; the accuracy claims retain independent empirical content.

  1. fitted input called prediction [Abstract; Section 3.2 Eq. (3.1); Tables 1–4]
    "NSGA-II searches for Pareto optimal circuits that (i) maximize classification accuracy and (ii) minimize primitive gate count—an essential constraint for noisy intermediate-scale quantum (NISQ) hardware. Experiments on four benchmarks ... show that GTQNNs match or exceed state of the art quantum models while requiring much fewer gates for most cases. ... (maximize)Fitness 1 = Classification accuracy (minimize)Fitness 2 = Gate count (3.1)"

    The low-gate-count claim is the second NSGA-II objective restated as a result. Because gate count is literally Fitness 2 in Eq. (3.1), any circuit selected by the Pareto search has been optimized for few gates; reporting 'requiring much fewer gates' is not an independent prediction but a description of the selection criterion. The accuracy part of the claim remains empirical and is not circular, so this is only partial circularity.

  2. other [Section 5.2 (Fisher spectrum experiments)]
    "Recall that a barren-plateau regime would force tr(Eθ[F(θ)]) to vanish exponentially in the number of qubits [1]. Our empirical spectrum shows λ0 = O(1), Σ λk >> 0, so TQNN does sit in a barren plateau. ... The hybrid Fisher spectrum indicates that the model is far from a barren plateau, its leading curvature is concentrated in more QNN direction as the qubit number increases, confirming that additional qubits enlarge the expressivity and that the transformer front-end has already compressed the input features into a lower-dimensional representation."

    The paper's own quoted barren-plateau diagnostic is an expectation over random parameter initializations, tr(Eθ[F(θ)]), but the experiment evaluates the empirical Fisher at one converged, NSGA-II-selected parameter vector (Eq. 5.3). A large O(1) eigenvalue at the trained point is a property of a model that has already been successfully fitted, not a measurement of the landscape; hence 'far from a barren plateau' is read off from the same fitted input used to define the model rather than derived from the definition. The printed sentence even states the opposite ('does sit in a barren plateau'), and no random-initialization average is reported, making the Fisher 'confirmation' a restatement of the fitted model rather than a predictive test.

full rationale

The paper is primarily an empirical architecture-and-optimization study, and its accuracy headline is not circular: the reported GTQNN accuracies are measured on held-out data and compared with external baselines. However, two load-bearing conclusions reduce to their own inputs. First, the resource-efficiency claim ('requiring much fewer gates') is not a discovered prediction but the direct output of Fitness 2 = Gate count in Eq. (3.1); the genetic search is explicitly asked to minimize gates, so the low gate counts in Tables 1–4 are selected for by construction. Second, the Fisher-spectrum claim that the trained networks are 'far from barren plateaus' is not a valid test of the paper's own definition: the quoted criterion is an expectation over random parameter initializations, while the experiment computes the empirical Fisher at the converged, search-selected parameters, where O(1) curvature is expected for a successfully trained model. The sentence as printed even asserts the opposite ('TQNN does sit in a barren plateau'), which further undermines the confirmation. The accuracy comparisons to quoted baselines are uncontrolled (different class subsets, splits, and preprocessing), but that is a correctness and validity concern, not circularity, because the baselines are external. The self-citations [36, 37] are used to ground the fitness-function design and are not the source of the headline numbers, so they do not materially raise the score. Overall, the central 'match or exceed state of the art' accuracy claim has independent empirical content, but the companion resource-efficiency and barren-plateau confirmations are partially circular, warranting a 6 rather than a higher score.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical laws or entities. Its central claim rests on a specific pipeline of hand-chosen transformer and genetic algorithm hyperparameters, on assumptions about the validity of the empirical Fisher computation, and on the representational sufficiency of the transformer compression. The most serious methodological issue is the inconsistent and likely incorrect statement about barren plateaus in Section 5.2.

free parameters (6)
  • Transformer architecture hyperparameters (layers, heads, d_model, embedding size)
    Not specified in the paper; chosen by hand, and they determine the representation fed to the quantum circuit.
  • NSGA-II population and offspring size = 20
    Section 4.1; evolutionary search dynamics depend on these.
  • Crossover and mutation probabilities = pc=0.90, pm=1/string length
    Section 4.1; chosen without justification.
  • Number of generations = 50 or 60
    Section 4.1; stopping criterion is generation count only.
  • Internal training epochs and outer retraining epochs = 50 inner; 100/250/400/500 outer
    Section 4.1; different values per dataset.
  • Number of selected Pareto solutions = k=10
    Section 4.1; used for final retraining.
assumptions (4)
  • domain assumption The empirical Fisher information matrix computed from 256 data samples is a consistent estimator of the true Fisher information.
    Section 5.1; the paper asserts consistency because 'samples are drawn from the true model', but they are drawn from the data distribution, and ref [9] shows the empirical Fisher is not generally consistent.
  • ad hoc to paper The Fisher spectrum evaluated at a trained parameter vector can diagnose whether the model lies in a barren-plateau regime.
    Section 5.2; barren-plateau diagnostics normally require expectation over random initializations, not the final trained point.
  • domain assumption A noiseless QASM simulation with 1024 shots is a faithful proxy for NISQ hardware.
    Section 4.1; no noise model or hardware results are provided.
  • ad hoc to paper The transformer can compress arbitrarily structured data (e.g., 784-pixel MNIST images) to n_qubits features without losing classification-relevant information.
    Section 2.3; this is the core design premise, not validated by an ablation or a classical-only baseline.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Genetic Transformer-Assisted Quantum Neural Networks for Optimal Circuit Design." pith.science (2026). https://pith.science/paper/55GYJIQA

@misc{pith2026250609205,
  author       = {Pith},
  title        = {Pith review of: Genetic Transformer-Assisted Quantum Neural Networks for Optimal Circuit Design},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/55GYJIQA}},
  note         = {Machine review of arXiv:2506.09205}
}
read the original abstract

We introduce Genetic Transformer Assisted Quantum Neural Networks (GTQNNs), a hybrid learning framework that combines a transformer encoder with a shallow variational quantum circuit and automatically fine tunes the circuit via the NSGA-II multi objective genetic algorithm. The transformer reduces high-dimensional classical data to a compact, qubit sized representation, while NSGA-II searches for Pareto optimal circuits that (i) maximize classification accuracy and (ii) minimize primitive gate count an essential constraint for noisy intermediate-scale quantum (NISQ) hardware. Experiments on four benchmarks (Iris, Breast Cancer, MNIST, and Heart Disease) show that GTQNNs match or exceed state of the art quantum models while requiring much fewer gates for most cases. A hybrid Fisher information analysis further reveals that the trained networks operate far from barren plateaus; the leading curvature directions increasingly align with the quantum subspace as the qubit budget grows, confirming that the transformer front end has effectively condensed the data. Together, these results demonstrate that GTQNNs deliver competitive performance with a quantum resource budget well suited to present-day NISQ devices.

Figures

Figures reproduced from arXiv: 2506.09205 by the authors.

Figure 1
Figure 1. Transformer-assisted quantum neural network (TQNN) [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Genetic Transformer Quantum Neural Networks (GTQNNs) [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Optimal Variational Quantum Circuits for Iris dataset [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Optimal Variational Quantum Circuits for Breast cancer dataset [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: Optimal Variational Quantum Circuits for MNIST dataset [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: Optimal Variational Quantum Circuits for Heart Diseast dataset [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Above: contribution (energy) of Transformer (blue for Transformer and [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]
Figure 8
Figure 8. Figure 8: Above: contribution (energy) of Transformer for MNIST dataset (blue for [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

64 extracted references · 34 canonical work pages

  1. [4]

    C.-Y. Liu, E. J. Kuo, C. H. A. Lin et al. Training classical neural networks by quantum machine learning. 2024 IEEE International Conference on Quantum Computing and Engineering (QCE), 2024, Volume: 2, Pages: 34-38 DOI 10.1109/QCE60285.2024.10248 arXiv:2402.16465, 2024

    arXiv 2024

  2. [2]

    Alam and S

    M. Alam and S. Ghosh. DeepQMLP: A scalable quantum–classical hybrid deep neural network architecture for classification. In Proc. 35th Int. Conf. VLSI Design & 21st Int. Conf. Embedded Systems (VLSID) , pages 275–280. IEEE, 2022

    work page 2022

  3. [30]

    Singh, A

    U. Singh, A. Z. Goldberg, K. Heshami. Coherent feed-forward quantum neural network. Quantum Machine Intelligence 6, 89 (2024)

    work page 2024

  4. [1]

    Abbas, D

    A. Abbas, D. Sutter, C. Zoufal, A. Lucchi, A. Figalli, and S. Woerner. The power of quantum neural networks. Nature Computational Science , 1(6):403– 409, 2021. DOI: 10.1038/s43588-021-00084-1

    doi:10.1038/s43588-021-00084-1 2021

  5. [3]

    Chen, Y.-J

    H.-Y. Chen, Y.-J. Chang, S.-W. Liao, and C.-R. Chang. Deep Q-learning with hybrid quantum neural network on solving maze problems. Quantum Machine Intelligence, 2024, 6:2 https://doi.org/10.1007/s42484-023-00137-w

    doi:10.1007/s42484-023-00137-w 2024

  6. [5]

    MacCormack, C

    I. MacCormack, C. Delaney, A. Galda, N. Aggarwal, and P. Narang. Branch- ing quantum convolutional neural networks. Physical Review Research , 4(1):013117, 2022. DOI: 10.1103/PhysRevResearch.4.013117

    work page doi:10.1103/physrevresearch.4.013117 2022

  7. [6]

    Do Quantum Neural Networks have Simplicity Bias?

    J. Pointing. Do quantum neural networks have simplicity bias? arXiv preprint arXiv:2407.03266, 2024. 19

    work page Pith review arXiv 2024

  8. [8]

    S.-I. Amari. Natural gradient works efficiently in learning. Neural computation, 10(2):251–276, 1998. DOI: 10.1162/089976698300017746

    doi:10.1162/089976698300017746 1998

Show all 64 references
  1. [9]

    Kunstner, P

    F. Kunstner, P. Hennig, and L. Balles. Limitations of the empirical Fisher approximation for natural-gradient descent. InAdvances in Neural Information Processing Systems 32, pages 4156–4167, 2019

    2019

  2. [10]

    B. R. Frieden. Science from Fisher Information: A Unification . Cambridge University Press, 2004. DOI: 10.1017/CBO9780511616907

    2004 doi

  3. [11]

    J. J. Rissanen. Fisher information and stochastic complexity. IEEE Transac- tions on Information Theory , 42(1):40–47, 1996. DOI: 10.1109/18.481776

    1996 doi

  4. [12]

    W. Ji, D. Liu, Y. Meng, and Y. Xue. A review of genetic-based evolutionary al- gorithms in SVM parameter optimisation. Evolutionary Intelligence, 14:1389– 1414, 2021. DOI: 10.1007/s12065-020-00439-z

    2021 doi

  5. [13]

    Lahoz-Beltra

    R. Lahoz-Beltra. Quantum genetic algorithms for computer scientists. Com- puters, 5(4):24, 2016. DOI: 10.3390/computers5040024

    2016 doi

  6. [14]

    Acampora and A

    G. Acampora and A. Vitiello. Implementing evolutionary optimisation on actual quantum processors. Information Sciences , 575:542–562, 2021. https://doi.org/10.1016/j.ins.2021.06.049

    2021 doi

  7. [15]

    Altares-L´ opez, A

    S. Altares-L´ opez, A. Ribeiro, and J. J. Garc ´ ıa-Ripoll. Automatic design of quantum feature maps. Quantum Science and Technology , 6(4):045010, 2021. https://iopscience.iop.org/article/10.1088/2058-9565/ac1ab1

    2021 doi

  8. [16]

    Chivilikhin, A

    D. Chivilikhin, A. Samarin, V. Ulyantsev, I. Iorsh, A. R. Oganov, and O. Kyri- ienko. MoG-VQE: Multi-objective genetic variational quantum eigensolver. arXiv preprint arXiv:2007.04424, 2020

    2007 arXiv

  9. [17]

    K. Deb, S. Agrawal, A. Pratap, and T. Meyarivan. A fast and elitist multi- objective genetic algorithm: NSGA-II. IEEE Transactions on Evolutionary Computation, 6(2):182–197, 2002. DOI: 10.1109/4235.996017

    2002

  10. [18]

    Miettinen

    K. Miettinen. Nonlinear Multiobjective Optimization . Kluwer Academic Pub- lishers, Boston, 1998

    1998

  11. [19]

    Alaia, I.H

    E.B. Alaia, I.H. Dridi, H. Bouchriha and P. Borne, Genetic algorithm with pareto front selection for multi-criteria optimization of multi-depots and multi- vehicle pickup and delivery problems with time windows , 2014 15th Interna- tional Conference on Sciences and Techniques ...

    2014

  12. [20]

    Bar´ an, A

    B. Bar´ an, A. Carballude and M. Villagra, Neighbor Optimization of N- Dimensional Quantum Circuits , SN Computer Science 2, 2021

    2021

  13. [21]

    Cerezo et al

    M. Cerezo et al. Challenges and opportunities in quantum machine learning. Nat. Comput. Sci. 2, 567 (2022)

    2022

  14. [22]

    Beer et al

    K. Beer et al. Training deep quantum neural networks. Nat. Commun. 11, 808 (2020)

    2020

  15. [23]

    Havl ´ ıˇ ceket al

    V. Havl ´ ıˇ ceket al. Supervised learning with quantum-enhanced feature spaces. Nature 567, 209–212 (2019)

    2019

  16. [24]

    D. P. Garc ´ ıa, J. Cruz-Benito, F. J. Garc ´ ıa-Pe˜ nalvo. Systematic literature re- view: quantum machine learning and its applications. Computer Science Re- view : 51, 100619(2024)

    2024

  17. [25]

    S. Sim, P. D. Johnson, A. Aspuru-Guzik. Expressibility and entangling capa- bility of parameterized quantum circuits. Adv. Quantum Technol. 2, 1900070 (2019)

    2019

  18. [26]

    S. X. Zhang et al. TensorCircuit: a quantum software framework for the NISQ era. Quantum 7, 912 (2023)

    2023

  19. [27]

    Farhi, H

    E. Farhi, H. Neven. Classification with quantum neural networks on near-term processors. arXiv :1802.06002 (2018)

    2018 arXiv

  20. [28]

    Sharma et al

    K. Sharma et al. Trainability of dissipative perceptron-based quantum neural networks. Phys. Rev. Lett. 128, 180505 (2022)

    2022

  21. [29]

    Sagingalieva et al

    A. Sagingalieva et al. Hybrid quantum neural network for drug response pre- diction. Cancers 15, 2705 (2023)

    2023

  22. [31]

    M. Schuld. Supervised quantum machine learning models are kernel methods. arXiv :2101.11020 (2021)

    2021 arXiv

  23. [32]

    Vaswani et al

    A. Vaswani et al. Attention is all you need. In Advances in Neural Information Processing Systems 30, pages 5998–6008, 2017

    2017

  24. [33]

    Devlin, M.-W

    J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova. BERT: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of NAACL, pages 4171–4186, 2019. https://doi.org/10.18653/v1/N19-1423

    2019 doi

  25. [34]

    Dosovitskiy et al

    A. Dosovitskiy et al. An image is worth 16 times16 words: Transformers for image recognition at scale.International Con- ference on Learning Representations, 2021. https://arxiv.org/abs/2010.11929 21

    2021 arXiv

  26. [35]

    Jennewein, Douglas M. et al. The Sol Supercomputer at Arizona State Uni- versity. PEARC ’23: Practice and Experience in Advanced Research Com- puting 2023: Computing for the Common Good Pages 296 - 301, 2023. https://doi.org/10.1145/3569951.3597573

    2023

  27. [36]

    Wang Several fitness functions and entanglement gates in quantum kernel generation

    H. Wang Several fitness functions and entanglement gates in quantum kernel generation. Quantum Mach. Intell. , 7, 7 (2025). https://doi.org/10.1007/s42484-024-00233-5

    2025 doi

  28. [37]

    Wang A novel feature selection method based on quantum support vec- tor machine

    H. Wang A novel feature selection method based on quantum support vec- tor machine. Physica Scripta , 99 056006 (2024). , Volume 99, Number 5 https://doi.org/DOI 10.1088/1402-4896/ad36ef

    2024 doi

  29. [38]

    Antonio Ben ´ ıtez-Hidalgo, Antonio J. Nebro, Jos´ e Garc ´ ıa-Nieto, Izaskun Oregi, Javier Del Ser, jMetalPy: A Python framework for multi-objective optimiza- tion with metaheuristics, Swarm and Evolutionary Computation , 51, 100598, 2019 https://doi.org/10.1016/j.swevo.2019.100598

    2019

  30. [39]

    Schuld and F

    M. Schuld and F. Petruccione. Machine Learning with Quan- tum Computers . Springer International Publishing, Cham, 2021. https://link.springer.com/book/10.1007/978-3-030-83098-4

    2021 doi

  31. [40]

    M. A. Nielsen and I. L. Chuang. Quantum Computation and Quantum Infor- mation (10th-anniv. ed.). Cambridge University Press, Cambridge, 2011

    2011

  32. [41]

    Biamonte, P

    J. Biamonte, P. Wittek, N. Pancotti, P. Rebentrost, N. Wiebe, and S. Lloyd. Quantum machine learning. Nature, 549(7671):195–202, 2017. DOI: 10.1038/nature23474

    2017 doi

  33. [42]

    Schuld, I

    M. Schuld, I. Sinayskiy, and F. Petruccione. An introduction to quan- tum machine learning. Contemporary Physics , 56(2):172–185, 2015. DOI: 10.1080/00107514.2014.964942

    2015

  34. [43]

    Havl ´ ıˇ cek, A

    V. Havl ´ ıˇ cek, A. D. C´ orcoles, K. Temme et al. Supervised learning with quantum-enhanced feature spaces. Nature, 567(7747):209–212, 2019. DOI: 10.1038/s41586-019-0980-2

    2019 doi

  35. [44]

    D. P. Garc ´ ıa, J. Cruz-Benito, and F. J. Garc ´ ıa-Pe˜ nalvo. Systematic literature review: Quantum machine learning and its applications. arXiv:2201.04093,

    arXiv

  36. [45]

    D. N. Diep. Some quantum neural networks. International Journal of Theoret- ical Physics, 59:1179–1190, 2020. DOI: 10.1007/s10773-020-04397-1

    2020 doi

  37. [46]

    Chalumuri, R

    A. Chalumuri, R. Kune, and B. S. Manoj. A hybrid classical-quantum approach for multi-class classification. Quantum Information Processing , 20:119, 2021. DOI: 10.1007/s11128-021-03029-9. 22

    2021 doi

  38. [47]

    B. Q. Chen and X. F. Niu. Quantum Neural Network with Improved Quan- tum Learning Algorithm. International Journal of Theoretical Physics , 59:1978–1990, 2020. DOI: 10.1007/s10773-020-04470-9

    1978 doi

  39. [48]

    Tacchino, S

    F. Tacchino, S. Mangini, P. K. Barkoutsoset al. Variational Learning for Quan- tum Artificial Neural Networks. IEEE Transactions on Quantum Engineering, 2:1–10, 2021. DOI: 10.1109/TQE.2021.3062494

    2021

  40. [49]

    J. Wang, Y. Chen, R. Chakraborty, and S. X. Yu. Quantum gradient descent algorithms. arXiv:1911.12207, 2019. DOI: 10.48550/arXiv.1911.12207

    1911 doi arXiv

  41. [50]

    Y. Li, R. G. Zhou, R. Xu, J. Luo, and W. Hu. A quantum deep convolu- tional neural network for image recognition. Quantum Science and Technology, 5:044003, 2020. DOI: 10.1088/2058-9565/ab9f93

    2020 doi

  42. [51]

    S. L. Wu et al. Application of quantum ensemble learning to particle- physics analysis at the LHC. Physical Review Research , 3:033221, 2021. DOI: 10.1103/PhysRevResearch.3.033221

    2021 doi

  43. [52]

    K. H. Wan, O. Dahlsten, H. Kristj´ ansson, R. Gardner, and M. S. Kim. Quan- tum generalisation of feed-forward neural networks. npj Quantum Information, 3:36, 2017. DOI: 10.1038/s41534-017-0032-4

    2017 doi

  44. [53]

    K. Beer, D. Bondarenko, T. Farrelly et al. Training deep quan- tum neural networks. Nature Communications , 11:808, 2020. DOI: 10.1038/s41467-020-14454-2

    2020 doi

  45. [54]

    I. Cong, S. Choi, and M. D. Lukin. Quantum convolutional neural networks. Nature Physics, 15:1273–1278, 2019. DOI: 10.1038/s41567-019-0648-8

    2019 doi

  46. [55]

    Sharma, M

    K. Sharma, M. Cerezo, L. Cincio, and P. J. Coles. Trainability of dissipa- tive perceptron-based quantum neural networks. Physical Review Letters , 128:180505, 2022. DOI: 10.1103/PhysRevLett.128.180505

    2022 doi

  47. [56]

    M. G. Zhou, Z. P. Liu, H. L. Yin et al. Quantum Neural Network for Quantum Neural Computing. Research, 6:0134, 2023. DOI: 10.34133/research.0134

    2023 doi

  48. [57]

    Preskill

    J. Preskill. Quantum computing in the NISQ era and beyond. Quantum, 2:79,

  49. [58]

    E. A. Cherrat, I. Kerenidis, N. Mathur, J. Landman, M. Strahm, and Y. Y. Li. Quantum vision transformers. Quantum, 8:1265, 2024. DOI: 10.22331/q-2024-02-22-1265

    2024 doi

  50. [59]

    H. Ma, H. Shang, and J. Yang. Quantum embedding method with transformer neural network quantum states for strongly correlated materials. npj Compu- tational Materials , 10:220, 2024. DOI: 10.1038/s41524-024-01406-3. 23

    2024 doi

  51. [60]

    Zhang and Q

    H. Zhang and Q. Zhao. A survey of quantum transformers: Tech- nical approaches, challenges and outlooks. arXiv:2504.03192, 2025. arXiv:2504.03192

    2025

  52. [61]

    , Wang, X

    Li, G., Zhao, X. , Wang, X. Quantum self-attention neural net- works for text classification. Sci. China Inf. Sci. 67, 142501 (2024). https://doi.org/10.1007/s11432-023-3879-7

    2024 doi

  53. [62]

    Baevski, Y

    A. Baevski, Y. Zhou, A. Mohamed, and M. Auli. wav2vec 2.0: A frame- work for self-supervised learning of speech representations. In Advances in Neural Information Processing Systems 33 , pages 12449–12460, 2020. https://arxiv.org/abs/2006.11477

    2020 arXiv

  54. [63]

    Alaia, I.H

    E.B. Alaia, I.H. Dridi, H. Bouchriha and P. Borne, Genetic algorithm with pareto front selection for multi-criteria optimization of multi-depots and multi- vehicle pickup and delivery problems with time windows , 2014 15th Interna- tional Conference on Sciences and Techniques ...

    2014

  55. [2018]

    DOI: 10.22331/q-2018-08-06-79

    2018 doi

  56. [2022]

    DOI: 10.48550/arXiv.2201.04093

    doi arXiv

Pith tools

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