Pith. sign in

REVIEW 6 major objections 5 minor 2 cited by

Quantum Adaptive Excitation Network with Variational Quantum Circuits for Channel Attention

T0 review · 6 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper claims that swapping a neural network's classical channel-attention block for a small trainable quantum circuit improves image classification on MNIST, FashionMNIST, and CIFAR-10, with larger gains at greater circuit depth.

desk verdict The architecture is a real but incremental step; the headline CIFAR-10 result is not supported because the SENet baseline looks under-optimized and there are no error bars, seeds, or code. read the letter →

arxiv 2507.11217 v1 pith:YFCPI7WN submitted 2025-07-15 quant-ph

Yu-Chao Hsu , Kuan-Cheng Chen , Tai-Yue Li , Nan-Yow Chen This is my paper
classification quant-ph
keywords QuantumMachineLearningVariationalCircuitsChannelAttentionHybridQuantum-ClassicalSystemsConvolutionalNeuralNetworksSqueeze-and-Excitationimageclassification
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

QAE-Net asks whether a variational quantum circuit can take over the "excitation" step of a Squeeze-and-Excitation channel-attention module inside a CNN. The authors replace the two fully connected layers that normally produce channel weights with a four-qubit circuit: the pooled channel descriptor is angle-encoded into qubit rotations, entangled with CNOT gates, and read out through Pauli-$Z$ measurements; a small linear layer and sigmoid then rescale the feature maps. Under what they describe as identical architecture and training settings, they report that QAE-Net beats SENet on all three datasets and that the CIFAR-10 gap is large: 89.08% versus 76.72%, rising to 90.10% with two variational layers and 92.30% with three. If these numbers hold, a near-term quantum circuit can serve as a drop-in attention component rather than a standalone quantum classifier.

What carries the argument

The load-bearing component is the VQC-based excitation module. After global average pooling produces a channel descriptor $z$, the descriptor is split into triples, each triple angle-encodes one qubit via an $\mathrm{SU}(2)$ rotation, a ring of CNOT gates entangles neighboring qubits, and $L$ layers of trainable single-qubit rotations with entanglement evolve the state. Pauli-$Z$ expectation values on the $n$ qubits form a quantum feature vector, and a single linear layer with sigmoid maps it to per-channel attention scales. The parameter-shift rule — evaluating the circuit at parameter shifts of $\pm\pi/2$ and taking half the difference — makes these quantum expectations differentiable, so the whole network trains end-to-end by backpropagation.

What would settle it

Train the same small CNN with the classical Squeeze-and-Excitation (SE) block on CIFAR-10 under a systematic hyperparameter sweep, data augmentation, and repeated seeds. If the best classical baseline reaches or exceeds 89% accuracy, the reported quantum gain is an artifact of an undertrained comparison.

Watch

Extended reading notes

Core claim

The paper's central claim is that quantum-derived channel attention is more expressive than classical channel attention in the regime tested. Specifically, replacing the SENet excitation block with a shallow variational quantum circuit yields higher classification accuracy on MNIST, FashionMNIST, and CIFAR-10 while keeping the parameter count nearly unchanged, and accuracy improves monotonically as variational layers are added. The largest reported effect is on three-channel CIFAR-10, where accuracy jumps from 76.72% (SENet) to 89.08% (one variational layer), 90.10% (two layers), and 92.30% (three layers).

Load-bearing premise

The load-bearing premise is that the classical SENet baseline was trained and tuned as seriously as the quantum model; if the 76.72% CIFAR-10 figure reflects an undertuned classical control, the reported quantum advantage is an artifact of that baseline.

Editorial extensions

If this is right

  • A four-qubit, three-layer variational circuit can replace a two-layer classical excitation block while keeping the total parameter count essentially unchanged.
  • On the datasets tested, deeper variational circuits give monotonically higher accuracy (89.08% to 90.10% to 92.30% on CIFAR-10), so circuit depth acts as a useful resource in this setting.
  • The hybrid module is trainable with standard optimizers because the parameter-shift rule supplies analytic gradients for the quantum measurements.
  • Because the design uses only four qubits, shallow depth, and ring entanglement, it stays within the reach of near-term quantum hardware.
  • The paper's largest gains occur on three-channel RGB inputs, which it attributes to the quantum circuit capturing richer inter-channel dependencies.

Reading between the lines

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

  • This paper does not ablate the VQC against a fixed random circuit or an equally small classical nonlinear layer; without such an ablation, the reported gain could come from added nonlinearity or initialization rather than from entanglement.
  • All results come from a classical simulator, so hardware noise and sampling cost are not yet part of the comparison; real-device tests would be needed to support the NISQ deployment claim.
  • A stronger classical SENet baseline, tuned with hyperparameter search and data augmentation, might close much of the 12-point CIFAR-10 gap; if so, the interesting residual question is whether the quantum module offers a better accuracy-to-cost trade-off.
  • One quick check would be to replace the Pauli-$Z$ readout with a random fixed feature map of the same dimension; if accuracy stays high, the advantage is in the classical linear layer rather than the trained quantum circuit.
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

6 major / 5 minor

Summary. The paper proposes QAE-Net, a hybrid quantum-classical network that replaces the classical excitation block of a Squeeze-and-Excitation module with a variational quantum circuit (VQC) over four qubits. The authors report classification accuracy on MNIST, FashionMNIST, and CIFAR-10, claiming consistent improvements over a SENet baseline, with a particularly large gain on CIFAR-10 (76.72% to 89.08% for one VQC layer) and further gains when the number of variational layers is increased. The paper also discusses training dynamics and argues that deeper quantum circuits improve expressivity.

Significance. If the reported results were reliable, the paper would be a useful empirical contribution to the growing literature on integrating variational quantum circuits into neural network components, specifically channel attention. However, the central claim of a quantum advantage rests on a small number of single-run experiments without error bars or multiple seeds, and the headline CIFAR-10 improvement is implausibly large given that the VQC replaces a classical block with a comparable or smaller parameter count. The paper does not provide code, training details sufficient for reproduction, or control experiments that isolate quantum effects. The potential significance of the idea is real, but the current evidence does not support the claims and the architecture description contains an internal contradiction.

major comments (6)
  1. [Section IV.D] The description of the forward pass is internally contradictory. The text first states that 'the recalibrated feature maps X' are forwarded through the remaining convolutional and fully connected layers for classification' and that 'the final layer produces a logit vector z(i)', but later states that 'the logit vector z(i) is obtained directly from the measurement outcomes of the quantum circuit, where each component corresponds to an expectation value of a Pauli observable.' With n=4 qubits and K=10 classes, the latter is impossible. This contradiction must be resolved; as written, it is unclear which description is correct and whether the reported experiments actually use the described architecture.
  2. [Section V.A, Table I] The central comparison rests on a single run per configuration, with no error bars, standard deviations, or seed variation. The CIFAR-10 baseline of 76.72% appears under-optimized: a small two-convolutional-layer network with 142,634 parameters should typically achieve higher accuracy on CIFAR-10 with a properly tuned schedule and standard augmentation, and the paper gives no details on batch size, learning rate schedule, data augmentation, or optimizer hyperparameters, so the 'identical architectural and training settings' claim cannot be verified. The reported gain of 12.36 percentage points (89.08% vs. 76.72%) is far larger than what a 4-qubit VQC with fewer parameters would be expected to produce, suggesting that the baseline may be unfairly weak or that the comparison is flawed.
  3. [Section V.B, Table II] The claim that increasing the number of variational layers yields 'a consistent improvement' and, in the Discussion, 'an almost linear improvement' is based on three single-run accuracy values (89.08%, 90.10%, 92.30%) with no error bars. With a typical run-to-run variation of around 0.5-1 percentage point on CIFAR-10, these differences (roughly 1.0 and 2.2 points) are not statistically meaningful. Additionally, the Discussion in Section VI refers to 'increasing the number of variational layers from one to four,' but Table II and Figure 4 report only 1, 2, and 3 layers; no four-layer experiment is shown.
  4. [Section IV.B, V.A] The paper attributes the performance gains to 'quantum superposition and entanglement' but provides no ablation or control to support this attribution. To show that the VQC contributes beyond a classical nonlinear layer, the authors would need to compare against a classical excitation block with the same number of parameters and a similar nonlinear function, and against a VQC without entangling gates. Without such ablations, the observed (if real) improvements could be due to the extra nonlinearity introduced by the quantum circuit or to the linear mapping in equation (7), not to quantum correlations.
  5. [Abstract, Section V.A] The abstract claims 'consistent performance improvements across all datasets,' but the reported gains on MNIST (97.9% to 98.0%) and FashionMNIST (91.0% to 91.3%) are 0.1 and 0.3 percentage points, respectively. These differences are well within the expected run-to-run variation for these datasets and cannot be interpreted as consistency when only one seed is used. The body text even refers to the CIFAR-10 results as 'preliminary,' which contradicts the abstract's definitive tone.
  6. [Section VI] The discussion of barren plateaus is speculative and not supported by any numerical evidence. The authors state that the shallow circuit and limited qubit count 'mitigate this issue in practice,' but no gradient variance measurements or trainability analyses are presented. Given that the paper's main evidence is empirical, such a claim should be backed by data.
minor comments (5)
  1. [Table II] The phrase 'an testing accuracy' should be 'a testing accuracy'.
  2. [Section V.A] The sentence 'In this section, we present our approach experiments demonstrating' is ungrammatical; it should be 'we present experiments demonstrating our approach'.
  3. [Section IV.B] The encoding unitary in equation (1) uses three rotation angles per qubit, but the text says 'each qubit is assigned three real-valued inputs to parameterize a full single-qubit SU(2) rotation'; this is fine, but for CIFAR-10 the channel dimension C=12, so the reshaping into four groups of three is consistent, yet this should be stated explicitly.
  4. [Section III] The symbol Uent is defined with a typographical issue: the product notation in equation (2) is missing the CNOT ordering details, and the text 'ring entanglement pattern defined as Uent = Qn-1 k=1 CNOT(k, k+ 1)' is confusing; it should clarify whether the CNOTs connect qubit n to qubit 1.
  5. [References] Some reference entries have inconsistent formatting (e.g., [43] is a preprint rather than the published PyTorch paper, and several arXiv preprints lack proper journal identifiers).

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: QAE-Net's reported gains are measured comparisons against SENet, not consequences of its definitions, fitted parameters, or self-citations.

full rationale

The central claim of QAE-Net, that replacing SENet's excitation FC layers with a 4-qubit variational circuit improves classification accuracy, is supported by direct empirical comparison in Table I (MNIST 97.9 to 98.0, F-MNIST 91.0 to 91.3, CIFAR-10 76.72 to 89.08) and Table II (89.08, 90.10, 92.30 for one, two, and three variational layers). These numbers are measured outcomes of a training procedure, not outputs of a fitted formula, so there is no 'fitted input called prediction' pattern. The quantum attention vector is constructed as Q_i = ⟨ψ(z;θ)|Z_i|ψ(z;θ)⟩ (Eq. 6) and mapped to attention scores by S_c = σ(W·Q+b) (Eq. 7); nothing in these equations presupposes the reported accuracy or the claimed layer-depth trend. The paper cites many prior works by its own authors ([14], [15], [18], [19], [23], [27], [30], [32], [34]–[37]), but none is load-bearing for the novelty claim; the only technical mechanism invoked from the literature is the parameter-shift rule [41], which is an external standard result. There is no self-citation chain that forces the architecture or forbids alternatives. The paper does have serious experimental and internal-consistency weaknesses that belong in a correctness review, not a circularity finding: the Section IV.D statement that 'the logit vector z(i) is obtained directly from the measurement outcomes of the quantum circuit' contradicts the preceding FC-layer description and is impossible with n=4 and K=10; and the SENet CIFAR-10 baseline of 76.72% appears under-optimized with no error bars or code, so the size of the reported gain is not yet established. These concerns do not reduce any claimed result to its own inputs, so the circularity score is low.

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

The central claim depends entirely on the empirical comparison. The hand-chosen architecture (4 qubits, angle encoding, 1-3 layers) and the implicit trust in the classical simulator are the key unverified inputs. No new physical entities are introduced.

free parameters (4)
  • Number of qubits = 4
    The 12-dimensional channel descriptor is angle-encoded into 4 qubits (3 values per qubit). This hand-chosen width bounds the quantum feature vector length and is not optimized or justified.
  • Number of variational layers = 1, 2, 3 (4 mentioned in Discussion)
    Varied to demonstrate that deeper circuits improve accuracy; the monotonic trend is based on one run per depth.
  • Learning rate = 0.001
    Reported only for CIFAR-10 in Table II; no learning-rate sweep and no statement of optimizer or schedule.
  • Backbone channel counts = 12 (conv1) and 16 (conv2)
    Fixed small backbone; the low capacity likely explains the weak SENet baseline on CIFAR-10. Chosen by hand, not justified.
assumptions (4)
  • standard math The parameter-shift rule gives unbiased gradient estimates for the VQC parameters.
    Stated in Section IV.D and used to train the circuit; a standard result for variational circuits with Pauli rotations.
  • domain assumption PennyLane's lightning.gpu simulator exactly reproduces the unitary evolution and measurement statistics of the quantum circuit.
    All experiments are simulations on classical hardware (Section V), but the paper claims NISQ compatibility without testing noise models or real hardware.
  • domain assumption The four-qubit angle-encoded VQC provides a more expressive mapping from channel descriptors to attention weights than the two-layer MLP it replaces.
    This is the motivating premise of QAE-Net (Section IV.B); no theoretical or empirical evidence is given beyond the reported accuracy numbers.
  • domain assumption SENet and QAE-Net are trained to comparable optima under identical settings.
    The fairness of the comparison (Section V.A) requires that the baseline is not under-optimized; the low SENet CIFAR-10 accuracy suggests this may be violated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantum Adaptive Excitation Network with Variational Quantum Circuits for Channel Attention." pith.science (2026). https://pith.science/paper/YFCPI7WN

@misc{pith2026250711217,
  author       = {Pith},
  title        = {Pith review of: Quantum Adaptive Excitation Network with Variational Quantum Circuits for Channel Attention},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YFCPI7WN}},
  note         = {Machine review of arXiv:2507.11217}
}
read the original abstract

In this work, we introduce the Quantum Adaptive Excitation Network (QAE-Net), a hybrid quantum-classical framework designed to enhance channel attention mechanisms in Convolutional Neural Networks (CNNs). QAE-Net replaces the classical excitation block of Squeeze-and-Excitation modules with a shallow Variational Quantum Circuit (VQC), leveraging quantum superposition and entanglement to capture higher-order inter-channel dependencies that are challenging to model with purely classical approaches. We evaluate QAE-Net on benchmark image classification tasks, including MNIST, FashionMNIST, and CIFAR-10, and observe consistent performance improvements across all datasets, with particularly notable gains on tasks involving three-channel inputs. Furthermore, experimental results demonstrate that increasing the number of variational layers in the quantum circuit leads to progressively higher classification accuracy, underscoring the expressivity benefits of deeper quantum models. These findings highlight the potential of integrating VQCs into CNN architectures to improve representational capacity while maintaining compatibility with near-term quantum devices. The proposed approach is tailored for the Noisy Intermediate-Scale Quantum (NISQ) era, offering a scalable and feasible pathway for deploying quantum-enhanced attention mechanisms in practical deep learning workflows.

Figures

Figures reproduced from arXiv: 2507.11217 by the authors.

Figure 1
Figure 1. These circuits consist of trainable quantum gates and [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Illustration of the Quantum Adaptive Excitation Network. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Training progression of QAE-Net as visualized through the Cross-Entropy (CE) Loss across epochs for three different [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Training progression of QAE-Net with different number [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Rank-Refined Quantum-Behaved Particle Swarm Optimization for Quantum Molecular Generation

    quant-ph 2026-07 conditional novelty 5.5 of 10

    RR-QPSO raises the validity–uniqueness product of 9-heavy-atom quantum molecular generation from 0.902 (BO) to 0.942 by rank-refined mean-best and fitness-guided swarm updates.

  2. Scaling Adaptive Non-Local Observable Quantum Super-Resolution via Matrix Product States

    quant-ph 2026-07 conditional novelty 5.0 of 10

    MPS simulation of ANO-VQCs scales image super-resolution feature extraction to 256 qubits and finds shallow depth L=1 best on 7×7→28×28 Fashion-MNIST under near-exact bond dimension χ=16.

Reference graph

Works this paper leans on

45 extracted references · 29 canonical work pages · cited by 2 Pith papers

  1. [1]

    Gradient-based learning applied to document recognition,

    Y . LeCun, L. Bottou, Y . Bengio, and P. Haffner, “Gradient-based learning applied to document recognition,” Proceedings of the IEEE , vol. 86, no. 11, pp. 2278–2324, 2002

    2002

  2. [2]

    A survey of convo- lutional neural networks: analysis, applications, and prospects,

    Z. Li, F. Liu, W. Yang, S. Peng, and J. Zhou, “A survey of convo- lutional neural networks: analysis, applications, and prospects,” IEEE transactions on neural networks and learning systems , vol. 33, no. 12, pp. 6999–7019, 2021

    work page 2021

  3. [3]

    Squeeze-and-excitation networks,

    J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 7132–7141, 2018

    2018

  4. [4]

    Genuine multipartite entanglement verification with convolutional neural networks,

    Y .-J. Luo, X. Leng, and C. Zhang, “Genuine multipartite entanglement verification with convolutional neural networks,” Physical Review A , vol. 110, no. 4, p. 042412, 2024

    work page 2024

  5. [5]

    Se-cnn based emergency control coordination strategy against voltage instability in multi-infeed hybrid ac/dc systems,

    Z. Zhang, B. Qin, X. Gao, T. Ding, Y . Zhang, and H. Wang, “Se-cnn based emergency control coordination strategy against voltage instability in multi-infeed hybrid ac/dc systems,”International Journal of Electrical Power & Energy Systems , vol. 160, p. 110082, 2024

    work page 2024

  6. [6]

    High-fidelity infrared u-shaped residual network digital holography with attention module for crystal growth observation,

    Y . Qiu, H. Huang, Y . Zhai, Z. Zheng, and D. Sun, “High-fidelity infrared u-shaped residual network digital holography with attention module for crystal growth observation,” Optics Letters , vol. 50, no. 6, pp. 1905– 1908, 2025

    work page 1905

  7. [7]

    Eca-net: Efficient channel attention for deep convolutional neural networks,

    Q. Wang, B. Wu, P. Zhu, P. Li, W. Zuo, and Q. Hu, “Eca-net: Efficient channel attention for deep convolutional neural networks,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pp. 11534–11542, 2020

    2020

  8. [8]

    Global second-order pooling convolutional networks,

    Z. Gao, J. Xie, Q. Wang, and P. Li, “Global second-order pooling convolutional networks,” in Proceedings of the IEEE/CVF Conference on computer vision and pattern recognition , pp. 3024–3033, 2019

    work page 2019

Show all 45 references
  1. [9]

    Quantum machine learning,

    M. Schuld and F. Petruccione, “Quantum machine learning,” in Ency- clopedia of Machine Learning and Data Science , pp. 1–12, Springer, 2023

    2023

  2. [10]

    Quantum machine learning for classical data,

    L. Wossnig, “Quantum machine learning for classical data,” arXiv preprint arXiv:2105.03684, 2021

    2021 arXiv

  3. [11]

    Quantum machine learning in feature hilbert spaces,

    M. Schuld and N. Killoran, “Quantum machine learning in feature hilbert spaces,” Physical review letters, vol. 122, no. 4, p. 040504, 2019

    2019

  4. [12]

    Quantum support vector machine for big data classification,

    P. Rebentrost, M. Mohseni, and S. Lloyd, “Quantum support vector machine for big data classification,” Physical review letters , vol. 113, no. 13, p. 130503, 2014

    2014

  5. [13]

    Supervised learning with quantum- enhanced feature spaces,

    V . Havl ´ıˇcek, A. D. C ´orcoles, K. Temme, A. W. Harrow, A. Kandala, J. M. Chow, and J. M. Gambetta, “Supervised learning with quantum- enhanced feature spaces,” Nature, vol. 567, no. 7747, pp. 209–212, 2019

    2019

  6. [14]

    Quantum-enhanced support vector machine for large-scale multi-class stellar classification,

    K.-C. Chen, X. Xu, H. Makhanov, H.-H. Chung, and C.-Y . Liu, “Quantum-enhanced support vector machine for large-scale multi-class stellar classification,” in International Conference on Intelligent Com- puting, pp. 155–168, Springer, 2024

    2024

  7. [15]

    Validating large-scale quantum ma- chine learning: Efficient simulation of quantum support vector machines using tensor networks,

    K.-C. Chen, T.-Y . Li, Y .-Y . Wang, S. See, C.-C. Wang, R. Wille, N.-Y . Chen, A.-C. Yang, and C.-Y . Lin, “Validating large-scale quantum ma- chine learning: Efficient simulation of quantum support vector machines using tensor networks,” Machine Learning: Science and Technol...

    2024

  8. [16]

    Robust decentralized quantum kernel learning for noisy and adversarial environment,

    W. Ma, K.-C. Chen, S. Yu, M. Liu, and R. Deng, “Robust decentralized quantum kernel learning for noisy and adversarial environment,” arXiv preprint arXiv:2504.13782, 2025

    2025 arXiv

  9. [17]

    Classification of tumor metastasis data by using quantum kernel-based algorithms,

    T.-Y . Li, V . R. Mekala, K.-L. Ng, and C.-F. Su, “Classification of tumor metastasis data by using quantum kernel-based algorithms,” in 2022 IEEE 22nd International Conference on Bioinformatics and Bioengineering (BIBE), pp. 351–354, IEEE, 2022

    2022

  10. [18]

    Quantum kernel-based long short- term memory,

    Y .-C. Hsu, T.-Y . Li, and K.-C. Chen, “Quantum kernel-based long short- term memory,” in 2025 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW) , pp. 1–5, IEEE, 2025

    2025

  11. [19]

    Quantum kernel-based long short-term memory for climate time-series forecasting,

    Y .-C. Hsu, N.-Y . Chen, T.-Y . Li, P.-H. H. Lee, and K.-C. Chen, “Quantum kernel-based long short-term memory for climate time-series forecasting,” in 2025 International Conference on Quantum Commu- nications, Networking, and Computing (QCNC) , pp. 421–426, IEEE, 2025

    2025

  12. [20]

    A tutorial on quantum convolutional neural networks (qcnn),

    S. Oh, J. Choi, and J. Kim, “A tutorial on quantum convolutional neural networks (qcnn),” in 2020 International Conference on Information and Communication Technology Convergence (ICTC) , pp. 236–239, IEEE, 2020

    2020

  13. [21]

    Quantum convolutional neural networks,

    I. Cong, S. Choi, and M. D. Lukin, “Quantum convolutional neural networks,” Nature Physics, vol. 15, no. 12, pp. 1273–1278, 2019

    2019

  14. [22]

    Quantum convolutional neural network for classical data classification,

    T. Hur, L. Kim, and D. K. Park, “Quantum convolutional neural network for classical data classification,” Quantum Machine Intelligence , vol. 4, no. 1, p. 3, 2022

    2022

  15. [23]

    Quantum pointwise convolution: A flexible and scalable approach for neural network enhancement,

    A. Ning, T.-Y . Li, and N.-Y . Chen, “Quantum pointwise convolution: A flexible and scalable approach for neural network enhancement,” in 2025 International Conference on Quantum Communications, Networking, and Computing (QCNC) , pp. 371–378, IEEE, 2025

    2025

  16. [24]

    Variational quantum algorithms,

    M. Cerezo, A. Arrasmith, R. Babbush, S. C. Benjamin, S. Endo, K. Fujii, J. R. McClean, K. Mitarai, X. Yuan, L. Cincio, et al. , “Variational quantum algorithms,” Nature Reviews Physics , vol. 3, no. 9, pp. 625– 644, 2021

    2021

  17. [25]

    Hybrid quantum-classical classifier based on tensor network and variational quantum circuit,

    S. Y .-C. Chen, C.-M. Huang, C.-W. Hsing, and Y .-J. Kao, “Hybrid quantum-classical classifier based on tensor network and variational quantum circuit,” arXiv preprint arXiv:2011.14651 , 2020

    2011 arXiv

  18. [26]

    A novel image classification framework based on variational quantum algorithms,

    Y . Chen, “A novel image classification framework based on variational quantum algorithms,” Quantum Information Processing, vol. 23, no. 10, pp. 1–28, 2024

    2024

  19. [27]

    Consensus-based distributed quantum kernel learning for speech recognition,

    K.-C. Chen, W. Ma, and X. Xu, “Consensus-based distributed quantum kernel learning for speech recognition,” in 2025 IEEE International Conference on Acoustics, Speech, and Signal Processing Workshops (ICASSPW), pp. 1–5, IEEE, 2025

    2025

  20. [28]

    Robust imple- mentation of generative modeling with parametrized quantum circuits,

    V . Leyton-Ortega, A. Perdomo-Ortiz, and O. Perdomo, “Robust imple- mentation of generative modeling with parametrized quantum circuits,” Quantum Machine Intelligence , vol. 3, no. 1, p. 17, 2021

    2021

  21. [29]

    A generative modeling approach for benchmark- ing and training shallow quantum circuits,

    M. Benedetti, D. Garcia-Pintos, O. Perdomo, V . Leyton-Ortega, Y . Nam, and A. Perdomo-Ortiz, “A generative modeling approach for benchmark- ing and training shallow quantum circuits,” npj Quantum information , vol. 5, no. 1, p. 45, 2019

    2019

  22. [30]

    Differentiable quantum architecture search in quantum-enhanced neural network parameter generation,

    S. Y .-C. Chen, C.-Y . Liu, K.-C. Chen, W.-J. Huang, Y .-J. Chang, and W.-H. Huang, “Differentiable quantum architecture search in quantum-enhanced neural network parameter generation,” arXiv preprint arXiv:2505.09653, 2025

    2025 arXiv

  23. [31]

    Quantum op- timization using variational algorithms on near-term quantum devices,

    N. Moll, P. Barkoutsos, L. S. Bishop, J. M. Chow, A. Cross, D. J. Egger, S. Filipp, A. Fuhrer, J. M. Gambetta, M. Ganzhorn, et al., “Quantum op- timization using variational algorithms on near-term quantum devices,” Quantum Science and Technology , vol. 3, no. 3, p. 030503, 2018

    2018

  24. [32]

    Learning to learn with quantum optimization via quantum neural networks,

    K.-C. Chen, H. Matsuyama, and W.-H. Huang, “Learning to learn with quantum optimization via quantum neural networks,” arXiv preprint arXiv:2505.00561, 2025

    2025 arXiv

  25. [33]

    A quantum circuit-based compression perspective for parameter-efficient learning,

    C.-Y . Liu, C.-H. H. Yang, H.-S. Goan, and M.-H. Hsieh, “A quantum circuit-based compression perspective for parameter-efficient learning,” arXiv preprint arXiv:2410.09846 , 2024

    2024 arXiv

  26. [34]

    Qtrl: Toward practical quantum reinforcement learning via quantum- train,

    C.-Y . Liu, C.-H. A. Lin, C.-H. H. Yang, K.-C. Chen, and M.-H. Hsieh, “Qtrl: Toward practical quantum reinforcement learning via quantum- train,” in 2024 IEEE International Conference on Quantum Computing and Engineering (QCE) , vol. 2, pp. 317–322, IEEE, 2024

    2024

  27. [35]

    Quantum-train long short- term memory: Application on flood prediction problem,

    C.-H. A. Lin, C.-Y . Liu, and K.-C. Chen, “Quantum-train long short- term memory: Application on flood prediction problem,” in 2024 IEEE International Conference on Quantum Computing and Engineering (QCE), vol. 2, pp. 268–273, IEEE, 2024

    2024

  28. [36]

    Quantum-enhanced parameter- efficient learning for typhoon trajectory forecasting,

    C.-Y . Liu, K.-C. Chen, Y .-C. Chen, S. Y .-C. Chen, W.-H. Huang, W.-J. Huang, and Y .-J. Chang, “Quantum-enhanced parameter- efficient learning for typhoon trajectory forecasting,” arXiv preprint arXiv:2505.09395, 2025

    2025 arXiv

  29. [37]

    Quantum-train- based distributed multi-agent reinforcement learning,

    K.-C. Chen, S. Y .-C. Chen, C.-Y . Liu, and K. K. Leung, “Quantum-train- based distributed multi-agent reinforcement learning,” in 2025 IEEE Symposium for Multidisciplinary Computational Intelligence Incubators (MCII Companion), pp. 1–5, IEEE, 2025

    2025

  30. [38]

    Gcnet: Non-local networks meet squeeze-excitation networks and beyond,

    Y . Cao, J. Xu, S. Lin, F. Wei, and H. Hu, “Gcnet: Non-local networks meet squeeze-excitation networks and beyond,” in Proceedings of the IEEE/CVF international conference on computer vision workshops , pp. 0–0, 2019

    2019

  31. [39]

    Srm: A style-based recalibration module for convolutional neural networks,

    H. Lee, H.-E. Kim, and H. Nam, “Srm: A style-based recalibration module for convolutional neural networks,” in Proceedings of the IEEE/CVF International conference on computer vision, pp. 1854–1862, 2019

    2019

  32. [40]

    Qca-net: Quantum-based channel attention for deep neural networks,

    J. Zhang, P. Cheng, Z. Li, H. Wu, W. An, and J. Zhou, “Qca-net: Quantum-based channel attention for deep neural networks,” in 2023 International Joint Conference on Neural Networks (IJCNN) , pp. 1–7, IEEE, 2023

    2023

  33. [41]

    General parameter- shift rules for quantum gradients,

    D. Wierichs, J. Izaac, C. Wang, and C. Y .-Y . Lin, “General parameter- shift rules for quantum gradients,” Quantum, vol. 6, p. 677, 2022

    2022

  34. [42]

    Pennylane: Automatic differentiation of hybrid quantum-classical com- putations,

    V . Bergholm, J. Izaac, M. Schuld, C. Gogolin, S. Ahmed, V . Ajith, M. S. Alam, G. Alonso-Linaje, B. AkashNarayanan, A. Asadi, et al. , “Pennylane: Automatic differentiation of hybrid quantum-classical com- putations,” arXiv preprint arXiv:1811.04968 , 2018

    2018 arXiv

  35. [43]

    Pytorch: An imperative style, high-performance deep learn- ing library,

    A. Paszke, “Pytorch: An imperative style, high-performance deep learn- ing library,” arXiv preprint arXiv:1912.01703 , 2019

    1912 arXiv

  36. [44]

    Barren plateaus in quantum neural network training landscapes,

    J. R. McClean, S. Boixo, V . N. Smelyanskiy, R. Babbush, and H. Neven, “Barren plateaus in quantum neural network training landscapes,”Nature communications, vol. 9, no. 1, p. 4812, 2018

    2018

  37. [45]

    Exponential con- centration in quantum kernel methods,

    S. Thanasilp, S. Wang, M. Cerezo, and Z. Holmes, “Exponential con- centration in quantum kernel methods,” Nature communications, vol. 15, no. 1, p. 5200, 2024

    2024

Pith tools

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