Pith. sign in

REVIEW 4 major objections 6 minor 43 references

Selective Feature Re-Encoded Quantum Convolutional Neural Network with Joint Optimization for Image Classification

T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read Selective re-encoding of top PCA features after each pooling layer, plus joint training of two parallel QCNNs, improves binary image classification accuracy on MNIST and Fashion-MNIST.

desk verdict A plausible QCNN architecture whose 'selective' PCA re-encoding claim is not yet separated from generic data re-uploading; worth a serious referee but needs ablations and statistics. read the letter →

arxiv 2507.02086 v1 pith:AEBOADIQ submitted 2025-07-02 quant-ph cs.LG

classification quant-phcs.LG PACS 03.67.Lx
keywords quantumconvolutionalneuralnetworkfeaturere-encodingprincipalcomponentanalysisautoencoderjointoptimizationensemblelearningimageclassificationNISQdevices
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 argues that a quantum convolutional neural network (QCNN) loses useful information each time a pooling layer halves the number of qubits, and that re-injecting the most informative classical features after pooling helps the circuit find better solutions. It proposes a selective feature re-encoding layer that takes the top principal components from a PCA feature vector and applies them as rotation angles on the remaining qubits. On the same logic, the paper combines two parallel QCNNs, one fed by PCA features and one by autoencoder features, and trains them jointly through a quantum interaction block rather than fusing their outputs after separate training. The reported experiments on binary MNIST and Fashion-MNIST tasks show accuracy gains from re-encoding and higher accuracy for the jointly optimized model than for either individual model or a decision-fusion ensemble. If these results hold, the method offers a way to strengthen NISQ-era classifiers without adding qubits.

What carries the argument

The carrying mechanism is the selective feature re-encoding layer, defined in Eq. (4) as the unitary operation $\psi'' = \bigotimes_{i=1}^k R_y(x_i)|\psi'_i\rangle$, which applies the top $k$ PCA components as $y$-axis rotation angles to the qubits that survive pooling. Because PCA orders components by decreasing variance, the first $k$ components are the ones re-encoded after each pooling step. A second mechanism is the interaction block that joins the two final qubits of Model-1 and Model-2 with six $R_y$ gates and two controlled-$R_x$ gates, creating bidirectional entanglement before measurement. The joint model measures Pauli-$Z$ expectations on both qubits and feeds them through softmax for a shared cross-entropy loss, so the gradients move parameters in both circuits together.

What would settle it

Ablation experiment: on the same binary tasks, replace the top-$k$ PCA components in the re-encoding layers with the $k$ least significant components or with fixed random angles; if accuracy stays the same, the selective content of the re-encoded features is not the source of the reported gains.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that selectively re-encoding classical features into a QCNN at intermediate depths materially improves classification. In addition to the usual one-time encoding at circuit entry, after each pooling layer that leaves $k$ qubits the first $k$ components of the PCA-reduced input are re-encoded via $R_y$ rotations. Across six binary tasks, the re-encoded QCNN outperforms the conventional QCNN in every reported accuracy comparison, with larger margins on the harder Fashion-MNIST pairs. The paper further claims that a parallel architecture joining this re-encoded PCA-QCNN (Model-1) with an autoencoder-feature QCNN (Model-2) through a parameterized interaction block and a single shared loss (Model-3) consistently beats both component models and the sum-rule ensemble (Model-4), setting reported benchmark marks such as 99.74% on MNIST 0 vs 1 and 96.50% on Fashion-MNIST T-shirt vs Trouser.

Load-bearing premise

The load-bearing assumption is that rotating the surviving qubits by the top PCA components after pooling guides the network toward a better solution rather than overwriting correlations the circuit has already learned.

Editorial extensions

If this is right

  • On the harder Fashion-MNIST pairs, the accuracy gain from re-encoding is larger than on MNIST, which the paper reads as evidence that re-encoding matters more when pooling destroys more of the input's distinguishing information.
  • The jointly optimized Model-3 beats the equally weighted sum-rule ensemble Model-4 in every reported binary task, so learning to interact before measurement is more effective than fusing independently trained outputs.
  • The pattern holds under both convolutional ansatzes, indicating the result does not depend on one particular circuit parametrization.
  • For the reported binary tasks, Model-3's accuracies exceed those listed for the comparison methods in the paper's benchmark table.

Reading between the lines

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

  • A natural test of the mechanism is to re-encode the least significant rather than the most significant PCA components; if accuracy does not drop, the ordering by variance is not what matters.
  • The interaction block is generic and small, so the joint-optimization scheme could plausibly be extended to more than two feature extractors, though the paper does not test whether the gains saturate or reverse with more branches.
  • All experiments appear to use ideal state-vector simulation; on real noisy hardware the added re-encoding and interaction gates introduce extra error, so the reported margins may narrow.
  • The re-encoding layers also add trainable capacity, and the paper does not ablate whether the gain comes from the selected feature content or simply from extra rotation gates inserted at the same depth.
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 / 6 minor

Summary. The manuscript proposes two modifications to quantum convolutional neural networks (QCNNs) for binary image classification. First, a 'selective feature re-encoding' layer re-encodes the top-k principal components of PCA-reduced features into the remaining qubits after each pooling layer, using Ry rotations (Section 2.2, Eq. (4)). Second, a parallel architecture (Model-3) couples two QCNNs—one using PCA features with re-encoding (Model-1) and one using autoencoder features with amplitude encoding (Model-2)—through a trainable interaction block and a shared loss, and compares this against an ensemble of independently trained models (Model-4). Experiments on three binary tasks from MNIST and three from Fashion-MNIST, with two convolutional ansatzes, report accuracy, precision, recall, and F1, and claim that re-encoding improves accuracy over conventional QCNNs and that Model-3 outperforms both individual models and the ensemble approach.

Significance. The proposed re-encoding strategy is simple, clearly specified, and plausible: reintroducing high-variance classical features at intermediate circuit depths could counteract information loss during pooling, and the joint-optimization architecture is a reasonable way to combine complementary classical feature extractors. If the empirical claims are confirmed with proper controls, the paper would offer a practical, reproducible improvement over earlier QCNN benchmarks (Easom, Hur, Kim, Mahmud) and a concrete design choice for NISQ-era classifiers. The circuit definitions are explicit (Figs. 3-9), and the decision to average over five independently initialized runs is a good practice. However, the paper does not yet establish the 'selective' mechanism—no ablation against alternative re-encoding choices is provided—and it does not report uncertainty estimates or significance tests, which is essential because many reported margins are smaller than 0.5 percentage points.

major comments (4)
  1. [Section 2.2, Eq. (4), Fig. 6] The central claim that selecting the most significant PCA features drives the improvement is not tested. The comparison in Tables 2-3 only contrasts a re-encoded QCNN against a conventional QCNN; the re-encoding layers add data-dependent rotations at intermediate depths, so the observed gains could be due to generic data re-uploading or the altered circuit structure rather than the top-k selection. Please add ablation experiments with random features, with bottom-k PCA components, and with fixed data-independent rotations. If top-k is not clearly better than these controls, the 'selective' claim should be withdrawn or substantially weakened.
  2. [Section 4.4, Tables 4-5, Fig. 11] The abstract and conclusion state that joint optimization 'consistently outperforms' the individual models and the ensemble, but the paper's own results contradict this on MNIST 0 vs 1: Model-2 achieves 99.70% (Ansatz-1) and 99.81% (Ansatz-2), while Model-3 achieves 99.58% and 99.74% respectively. The text acknowledges this exception in Section 4.4, but the unqualified abstract and conclusion claims are not supported. Please revise the claims to reflect the actual pattern, including this counterexample.
  3. [Section 4.2, Tables 2-5] The paper reports only mean values over five runs, with no standard deviations, confidence intervals, or significance tests. Given that several key comparisons are smaller than 0.5 percentage points (e.g., Model-3 vs Model-4 for Fashion MNIST T-shirt vs Trouser with Ansatz-2: 96.50 vs 96.20), the reader cannot judge whether these differences are meaningful. Please report per-run results or dispersion measures, and apply a paired statistical test (e.g., Wilcoxon signed-rank or bootstrap) across the twelve task/ansatz configurations.
  4. [Section 4.2] The experimental setup lacks several details needed for reproducibility: the learning rate and momentum hyperparameter for the Nesterov optimizer, the total number of training steps expressed in epochs (200 iterations with mini-batch 25 is less than one epoch for datasets of ~12,000 samples), how PCA and the single-epoch autoencoder are fit (on the training split only, or on the full data?), and the software/simulator used. Adding these details is important because the reported improvements are modest and may be sensitive to hyperparameter choices.
minor comments (6)
  1. [Abstract and Conclusion] The phrase 'consistently outperforms' should be replaced with a more measured statement that acknowledges the MNIST 0 vs 1 exception noted in Section 4.4.
  2. [Table 1] The header 'F ashion MNIST' contains a typo and should be 'Fashion MNIST'.
  3. [Section 2.2, end] The analogy between re-encoding and the classical self-attention mechanism is not developed or supported; consider removing it or providing a concrete connection.
  4. [Eq. (4)] The notation with triple kets |ψ′i⟩ is nonstandard and visually confusing; please rewrite the equation using standard ket notation.
  5. [Fig. 10 caption] The caption says 'Average Accuracy Improvement' but the chart displays absolute accuracies; please clarify the caption to match the plot.
  6. [Fig. 7] The in-text references to subfigures (a) and (b) should be checked against the printed order: the caption lists joint optimization as (a) and ensemble as (b), and the current layout should be consistent throughout.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central claims are empirical benchmark comparisons, not derivations from fitted targets or self-citation chains.

full rationale

The paper makes two central claims: selective feature re-encoding improves QCNN accuracy, and jointly optimized parallel QCNNs outperform individual models and ensemble fusion. Neither claim is derived from its own conclusion by construction. Equations (3)-(5) define the encoding and re-encoding maps; Eq. (4) is a circuit construction (tensor product of Ry rotations on post-pooling qubits), not a fitted quantity used to manufacture the reported accuracy. The PCA top-k selection is a fixed preprocessing rule applied before training and evaluated on held-out MNIST/Fashion-MNIST test splits, so the improvement in Tables 2-3 is an empirical measurement rather than an algebraic identity. Model-3's joint training and Model-4's fusion are likewise trained and evaluated against independent baselines. The only overlapping-author citation is Mahmud et al. [35], used as a state-of-the-art benchmark and as motivation for avoiding ancilla qubits; it is not invoked to justify the proposed mechanism or to forbid alternatives, and the comparison is externally falsifiable. The absence of an ablation isolating top-k selection from generic re-uploading is a limitation of experimental support, not circularity. Accordingly, no circular step is present.

Assumptions & free parameters 6 free parameters · 7 assumptions · 0 invented entities

The central claims rest on a small set of hand-chosen hyperparameters and on heuristic assumptions about what makes a good quantum feature representation. No new physical entities are postulated. The most fragile item is the assumption that re-encoding via Ry rotations after pooling helps rather than disrupts the learned state.

free parameters (6)
  • PCA feature count for initial encoding = 8
    Hand-chosen to match the 8-qubit Model-1 circuit; this count defines both the initial encoding and the re-encoding schedule.
  • Autoencoder latent dimension = 16
    Chosen to fit the 4-qubit amplitude encoding of Model-2; no ablation shows this is optimal.
  • Number of training iterations = 200
    Fixed stopping criterion in Section 4.2; no convergence analysis or early-stopping rule is given.
  • Mini-batch size = 25
    Fixed in Section 4.2; no sensitivity study is reported.
  • Learning rate and Nesterov momentum settings
    Mentioned as the optimizer in Section 4.2 but specific values are not reported, making exact replication impossible.
  • Number of re-encoding layers = 2
    Architecture choice after pooling layers 1 and 2; no ablation compares one versus two re-encoding layers.
assumptions (7)
  • standard math PCA eigenvalue decomposition yields components sorted by variance, with the first k capturing the most significant information.
    Invoked in Section 2.2 to justify selecting x1 through xk as the most significant features for re-encoding.
  • domain assumption Angle encoding maps classical feature values to single-qubit rotation angles.
    Used in Eqs. (3) and (4) as the encoding and re-encoding mechanism; assumes this mapping preserves useful class information.
  • domain assumption A noise-free quantum simulation approximates the behavior of a NISQ device for these experiments.
    The paper reports simulations without specifying a noise model or hardware; the results may not transfer to noisy devices.
  • domain assumption QCNN architectures mitigate barren plateaus, as argued by Pesah et al.
    Cited in Section 1 as motivation for using QCNNs; the paper does not verify gradient scaling in its own circuits.
  • ad hoc to paper Re-encoding top-k PCA features after pooling guides the circuit toward a better solution rather than disrupting learned information.
    Central heuristic in Section 2.2 and Eq. (4); no proof or ablation is provided to show that additional Ry rotations preserve or enhance the processed quantum state.
  • ad hoc to paper An autoencoder trained for a single epoch with one hidden layer produces useful features for the downstream QCNN.
    Stated in Section 4.2 as a resource-saving choice; no validation is shown for the quality of these features.
  • ad hoc to paper The interaction block creates beneficial bidirectional entanglement between the two QCNN output qubits.
    Assumed in Section 3.3 and Fig. 9; the paper does not isolate the contribution of the interaction block versus simply concatenating the two qubits.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Selective Feature Re-Encoded Quantum Convolutional Neural Network with Joint Optimization for Image Classification." pith.science (2026). https://pith.science/paper/AEBOADIQ

@misc{pith2026250702086,
  author       = {Pith},
  title        = {Pith review of: Selective Feature Re-Encoded Quantum Convolutional Neural Network with Joint Optimization for Image Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AEBOADIQ}},
  note         = {Machine review of arXiv:2507.02086}
}
read the original abstract

Quantum Machine Learning (QML) has seen significant advancements, driven by recent improvements in Noisy Intermediate-Scale Quantum (NISQ) devices. Leveraging quantum principles such as entanglement and superposition, quantum convolutional neural networks (QCNNs) have demonstrated promising results in classifying both quantum and classical data. This study examines QCNNs in the context of image classification and proposes a novel strategy to enhance feature processing and a QCNN architecture for improved classification accuracy. First, a selective feature re-encoding strategy is proposed, which directs the quantum circuits to prioritize the most informative features, thereby effectively navigating the crucial regions of the Hilbert space to find the optimal solution space. Secondly, a novel parallel-mode QCNN architecture is designed to simultaneously incorporate features extracted by two classical methods, Principal Component Analysis (PCA) and Autoencoders, within a unified training scheme. The joint optimization involved in the training process allows the QCNN to benefit from complementary feature representations, enabling better mutual readjustment of model parameters. To assess these methodologies, comprehensive experiments have been performed using the widely used MNIST and Fashion MNIST datasets for binary classification tasks. Experimental findings reveal that the selective feature re-encoding method significantly improves the quantum circuit's feature processing capability and performance. Furthermore, the jointly optimized parallel QCNN architecture consistently outperforms the individual QCNN models and the traditional ensemble approach involving independent learning followed by decision fusion, confirming its superior accuracy and generalization capabilities.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

43 extracted references · 18 canonical work pages

  1. [1]

    Advances in neural information processing sys- tems 25 (2012) https://doi.org/10.1145/3065386

    Krizhevsky, A., Sutskever, I., Hinton, G.E.: ImageNet classification with deep convolutional neural networks. Advances in neural information processing sys- tems 25 (2012) https://doi.org/10.1145/3065386

  2. [2]

    IEEE Access8, 37265–37274 (2020) https://doi.org/10

    Hu, M., Lin, H., Fan, Z., Gao, W., Yang, L., Liu, C., Song, Q.: Learning to Recognize Chest-Xray Images Faster and More Efficiently Based on Multi-Kernel Depthwise Convolution. IEEE Access8, 37265–37274 (2020) https://doi.org/10. 1109/ACCESS.2020.2974242

  3. [3]

    IEEE Access 12, 26626–26635 (2024) https://doi.org/10

    Alqahtani, F.F., Mohsan, M.M., Alshamrani, K., Zeb, J., Alhamami, S., Alqarni, D.: CNX-B2: A Novel CNN-Transformer Approach For Chest X-Ray Medical Report Generation. IEEE Access 12, 26626–26635 (2024) https://doi.org/10. 1109/ACCESS.2024.3367360

  4. [4]

    IET biometrics10(6), 581–606 (2021) https://doi.org/10.1049/ bme2.12029

    Zeng, D., Veldhuis, R., Spreeuwers, L.: A survey of face recognition techniques under occlusion. IET biometrics10(6), 581–606 (2021) https://doi.org/10.1049/ bme2.12029

  5. [5]

    IEEE Access8, 45244–45253 (2020) https://doi.org/ 10.1109/ACCESS.2020.2978938

    Song, A.-P., Hu, Q., Ding, X.-H., Di, X.-Y., Song, Z.-H.: Similar face recognition using the IE-CNN model. IEEE Access8, 45244–45253 (2020) https://doi.org/ 10.1109/ACCESS.2020.2978938

  6. [6]

    In: Autonomous Systems: Sensors, Processing, and Security for Vehicles and Infrastructure 2021, vol

    Rawashdeh, N.A., Bos, J.P., Abu-Alrub, N.J.: Drivable path detection using CNN sensor fusion for autonomous driving in the snow. In: Autonomous Systems: Sensors, Processing, and Security for Vehicles and Infrastructure 2021, vol. 11748, pp. 36–45 (2021). https://doi.org/10.1117/12.2587993 . SPIE 22

  7. [7]

    In: 2020 Third International Conference on Smart Systems and Inventive Tech- nology (ICSSIT), pp

    Dandamudi, A.G.B., Vasumithra, G., Praveen, G., Giriraja, C.: CNN Based Aerial Image processing model for Women Security and Smart Surveillance. In: 2020 Third International Conference on Smart Systems and Inventive Tech- nology (ICSSIT), pp. 1009–1017 (2020). https://doi.org/10.1109/ICSSIT48917. 2020.9214143 . IEEE

  8. [8]

    2024.24.3.4033

    Urhobo, B., Ugwuegbulam, D.: What comes after Moore’s Law: A comprehensive review of emerging computing paradigms (2024) https://doi.org/10.30574/wjarr. 2024.24.3.4033

Show all 43 references
  1. [9]

    arXiv preprint arXiv:2203.17181 (2022) https://doi.org/10.48550/arXiv.2203.17181

    Fedorov, A.K., Gisin, N., Beloussov, S.M., Lvovsky, A.I.: Quantum computing at the quantum advantage threshold: a down-to-business review. arXiv preprint arXiv:2203.17181 (2022) https://doi.org/10.48550/arXiv.2203.17181

  2. [10]

    Proceed- ings of the National Academy of Sciences 95(19), 11032–11033 (1998) https: //doi.org/10.1073/pnas.95.19.11032

    Brassard, G., Chuang, I., Lloyd, S., Monroe, C.: Quantum computing. Proceed- ings of the National Academy of Sciences 95(19), 11032–11033 (1998) https: //doi.org/10.1073/pnas.95.19.11032

  3. [11]

    IEEE Access 8, 219275–219294 (2020) https://doi.org/10.1109/ACCESS.2020.3041719

    Khan, T.M., Robles-Kelly, A.: Machine Learning: Quantum vs Classical. IEEE Access 8, 219275–219294 (2020) https://doi.org/10.1109/ACCESS.2020.3041719

  4. [12]

    Nature549(7671), 195–202 (2017) https://doi.org/ 10.1038/nature23474

    Biamonte, J., Wittek, P., Pancotti, N., Rebentrost, P., Wiebe, N., Lloyd, S.: Quantum machine learning. Nature549(7671), 195–202 (2017) https://doi.org/ 10.1038/nature23474

  5. [13]

    In: 46th International Colloquium on Automata, Languages, and Pro- gramming (ICALP 2019) (2019)

    Brandão, F.G., Kalev, A., Li, T., Lin, C.Y.-Y., Svore, K.M., Wu, X.: Quan- tum SDP Solvers: Large Speed-ups, Optimality, and Applications to Quantum Learning. In: 46th International Colloquium on Automata, Languages, and Pro- gramming (ICALP 2019) (2019). https://doi.org/10.42...

  6. [14]

    Physical Review A94(2), 022342 (2016) https://doi.org/10

    Schuld, M., Sinayskiy, I., Petruccione, F.: Prediction by linear regression on a quantum computer. Physical Review A94(2), 022342 (2016) https://doi.org/10. 1103/PhysRevA.94.022342

  7. [15]

    arXiv preprint quant-ph/0107012 (2001) https://doi.org/10.48550/arXiv.quant-ph/0107012

    Altaisky, M.: Quantum neural network. arXiv preprint quant-ph/0107012 (2001) https://doi.org/10.48550/arXiv.quant-ph/0107012

  8. [16]

    Physics Letters A379(7), 660–663 (2015) https://doi.org/10.1016/j

    Schuld, M., Sinayskiy, I., Petruccione, F.: Simulating a perceptron on a quantum computer. Physics Letters A379(7), 660–663 (2015) https://doi.org/10.1016/j. physleta.2014.11.061

  9. [17]

    arXiv preprint arXiv:1711.11240 (2017) https://doi.org/10.48550/arXiv.1711.11240

    Cao, Y., Guerreschi, G.G., Aspuru-Guzik, A.: Quantum Neuron: an elemen- tary building block for machine learning on quantum computers. arXiv preprint arXiv:1711.11240 (2017) https://doi.org/10.48550/arXiv.1711.11240

  10. [18]

    npj Quantum information3(1), 36 (2017) https://doi.org/10.1038/s41534-017-0032-4

    Wan, K.H., Dahlsten, O., Kristjánsson, H., Gardner, R., Kim, M.: Quantum 23 generalisation of feedforward neural networks. npj Quantum information3(1), 36 (2017) https://doi.org/10.1038/s41534-017-0032-4

  11. [19]

    Quantum Machine Intelligence 3(1), 15 (2021) https://doi.org/10.1007/ s42484-021-00046-w

    Zhao, C., Gao, X.-S.: QDNN: deep neural networks with quantum lay- ers. Quantum Machine Intelligence 3(1), 15 (2021) https://doi.org/10.1007/ s42484-021-00046-w

  12. [20]

    : Calibra- tion of a Cross-Resonance Two-Qubit Gate Between Directly Coupled Trans- mons

    Patterson, A., Rahamim, J., Tsunoda, T., Spring, P., Jebari, S., Ratter, K., Mergenthaler, M., Tancredi, G., Vlastakis, B., Esposito, M., et al. : Calibra- tion of a Cross-Resonance Two-Qubit Gate Between Directly Coupled Trans- mons. Physical Review Applied 12(6), 064013 (201...

  13. [21]

    Nature601(7893), 348–353 (2022) https://doi.org/10.1038/s41586-021-04292-7

    Mądzik, M.T., Asaad, S., Youssry, A., Joecker, B., Rudinger, K.M., Nielsen, E., Young, K.C., Proctor, T.J., Baczewski, A.D., Laucht, A.,et al.: Precision tomog- raphy of a three-qubit donor quantum processor in silicon. Nature601(7893), 348–353 (2022) https://doi.org/10.1038/s...

  14. [22]

    Nature Reviews Physics 3(9), 625–644 (2021) https://doi.org/10.1038/ s42254-021-00348-9

    Cerezo, M., Arrasmith, A., Babbush, R., Benjamin, S.C., Endo, S., Fujii, K., McClean,J.R.,Mitarai,K.,Yuan,X.,Cincio,L., et al.:Variationalquantumalgo- rithms. Nature Reviews Physics 3(9), 625–644 (2021) https://doi.org/10.1038/ s42254-021-00348-9

  15. [23]

    Quantum Science and Technology4(4), 043001 (2019) https://doi.org/10.1088/2058-9565/ab4eb5

    Benedetti, M., Lloyd, E., Sack, S., Fiorentini, M.: Parameterized quantum cir- cuits as machine learning models. Quantum Science and Technology4(4), 043001 (2019) https://doi.org/10.1088/2058-9565/ab4eb5

  16. [24]

    IEEE Access 8, 141007–141024 (2020) https://doi.org/10.1109/ACCESS.2020.3010470

    Chen, S.Y.-C., Yang, C.-H.H., Qi, J., Chen, P.-Y., Ma, X., Goan, H.-S.: Vari- ational Quantum Circuits for Deep Reinforcement Learning. IEEE Access 8, 141007–141024 (2020) https://doi.org/10.1109/ACCESS.2020.3010470

  17. [25]

    Physical review letters118(15), 150503 (2017) https: //doi.org/10.1103/PhysRevLett.118.150503

    Li, J., Yang, X., Peng, X., Sun, C.-P.: Hybrid Quantum-Classical Approach to Quantum Optimal Control. Physical review letters118(15), 150503 (2017) https: //doi.org/10.1103/PhysRevLett.118.150503

  18. [26]

    Nature communications 9(1), 4812 (2018) https://doi.org/10.1038/s41467-018-07090-4

    McClean, J.R., Boixo, S., Smelyanskiy, V.N., Babbush, R., Neven, H.: Barren plateaus in quantum neural network training landscapes. Nature communications 9(1), 4812 (2018) https://doi.org/10.1038/s41467-018-07090-4

  19. [27]

    npj Quantum Information 4(1), 65 (2018) https://doi.org/10.1038/s41534-018-0116-9

    Grant, E., Benedetti, M., Cao, S., Hallam, A., Lockhart, J., Stojevic, V., Green, A.G., Severini, S.: Hierarchical quantum classifiers. npj Quantum Information 4(1), 65 (2018) https://doi.org/10.1038/s41534-018-0116-9

  20. [28]

    Physical Review X 11(4), 041011 (2021) https://doi.org/10.1103/PhysRevX.11.041011 24

    Pesah, A., Cerezo, M., Wang, S., Volkoff, T., Sornborger, A.T., Coles, P.J.: Absence of barren plateaus in quantum convolutional neural networks. Physical Review X 11(4), 041011 (2021) https://doi.org/10.1103/PhysRevX.11.041011 24

  21. [29]

    Nature Physics 15(12), 1273–1278 (2019) https://doi.org/10.1038/s41567-019-0648-8

    Cong, I., Choi, S., Lukin, M.D.: Quantum convolutional neural networks. Nature Physics 15(12), 1273–1278 (2019) https://doi.org/10.1038/s41567-019-0648-8

  22. [30]

    Quantum Machine Intelligence4(1), 3 (2022) https://doi.org/ 10.1007/s42484-021-00061-x

    Hur, T., Kim, L., Park, D.K.: Quantum convolutional neural network for classical data classification. Quantum Machine Intelligence4(1), 3 (2022) https://doi.org/ 10.1007/s42484-021-00061-x

  23. [31]

    Quantum Machine Intelligence 2(1), 2 (2020) https://doi.org/10.1007/s42484-020-00012-y

    Henderson, M., Shakya, S., Pradhan, S., Cook, T.: Quanvolutional neural net- works: powering image recognition with quantum circuits. Quantum Machine Intelligence 2(1), 2 (2020) https://doi.org/10.1007/s42484-020-00012-y

  24. [32]

    Quantum Information Processing 23(5), 1–16 (2024) https://doi.org/10.1007/s11128-024-04360-7

    Shi, S., Wang, Z., Li, J., Li, Y., Shang, R., Zhong, G., Gu, Y.: Quantum convolu- tional neural networks for multiclass image classification. Quantum Information Processing 23(5), 1–16 (2024) https://doi.org/10.1007/s11128-024-04360-7

  25. [33]

    IEEE Transactions on Neural Networks and Learning Systems (2022) https://doi.org/ 10.1109/TNNLS.2022.3179354

    Easom-McCaldin, P., Bouridane, A., Belatreche, A., Jiang, R., Al-Maadeed, S.: Efficient Quantum Image Classification Using Single Qubit Encoding. IEEE Transactions on Neural Networks and Learning Systems (2022) https://doi.org/ 10.1109/TNNLS.2022.3179354

  26. [34]

    Neurocomputing555, 126643 (2023) https://doi.org/10.1016/ j.neucom.2023.126643

    Kim, J., Huh, J., Park, D.K.: Classical-to-quantum convolutional neural network transfer learning. Neurocomputing555, 126643 (2023) https://doi.org/10.1016/ j.neucom.2023.126643

  27. [35]

    Quantum Machine Intelligence 6(1), 11 (2024) https://doi.org/10.1007/ s42484-024-00145-4

    Mahmud, J., Mashtura, R., Fattah, S.A., Saquib, M.: Quantum convolu- tional neural networks with interaction layers for classification of classical data. Quantum Machine Intelligence 6(1), 11 (2024) https://doi.org/10.1007/ s42484-024-00145-4

  28. [36]

    arXiv preprint arXiv:1203.0722 (2012) https://doi.org/10.48550/arXiv.1203.0722

    Wei, H.-R., Di, Y.-M.: Decomposition of orthogonal matrix and synthesis of two- qubit and three-qubit orthogonal gates. arXiv preprint arXiv:1203.0722 (2012) https://doi.org/10.48550/arXiv.1203.0722

  29. [37]

    Physical Review A69(3), 032315 (2004) https://doi.org/10.1103/PhysRevA.69

    Vatan, F., Williams, C.: Optimal quantum circuits for general two-qubit gates. Physical Review A69(3), 032315 (2004) https://doi.org/10.1103/PhysRevA.69. 032315

  30. [38]

    Physical Review A 103(3), 032430 (2021) https://doi.org/10.1103/PhysRevA.103.032430

    Schuld, M., Sweke, R., Meyer, J.J.: Effect of data encoding on the expres- sive power of variational quantum-machine-learning models. Physical Review A 103(3), 032430 (2021) https://doi.org/10.1103/PhysRevA.103.032430

  31. [39]

    Physical Review A102(3), 032420 (2020) https://doi.org/10.1103/PhysRevA.102.032420

    LaRose, R., Coyle, B.: Robust data encodings for quantum classifiers. Physical Review A102(3), 032420 (2020) https://doi.org/10.1103/PhysRevA.102.032420

  32. [40]

    IEEE signal processing magazine29(6), 141–142 (2012) https://doi.org/10.1109/MSP.2012.2211477 25

    Deng, L.: The MNIST Database of Handwritten Digit Images for Machine Learn- ing Research [Best of the Web]. IEEE signal processing magazine29(6), 141–142 (2012) https://doi.org/10.1109/MSP.2012.2211477 25

  33. [41]

    arXiv preprint arXiv:1708.07747 (2017) https://doi.org/10.48550/arXiv.1708.07747

    Xiao, H., Rasul, K., Vollgraf, R.: Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithms. arXiv preprint arXiv:1708.07747 (2017) https://doi.org/10.48550/arXiv.1708.07747

  34. [42]

    Journal of Machine Learning Research12, 2825–2830 (2011) https://doi.org/10.48550/arXiv.1201.0490

    Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Prettenhofer, P., Weiss, R., Dubourg, V., Vanderplas, J., Passos, A., Cournapeau, D., Brucher, M., Perrot, M., Duchesnay, E.: Scikit-learn: Machine learning in Python. Journal of Mach...

  35. [43]

    Doklady Akademii Nauk SSSR269(3), 543 (1983) 26

    Nesterov, Y.: A method of solving a convex programming problem with conver- gence rate O 1 k2 . Doklady Akademii Nauk SSSR269(3), 543 (1983) 26

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.