Pith. sign in

REVIEW 4 major objections 6 minor 62 references

HQFNN: A Compact Quantum-Fuzzy Neural Network for Accurate Image Classification

T0 review · 4 major / 6 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 The paper claims a full fuzzy inference chain can run inside a shallow quantum circuit and beat six image classifiers.

desk verdict Architecture is a legitimate but incremental hybrid; the central superiority claim is unsupported by missing ablations and the robustness numbers never measure classification accuracy. read the letter →

arxiv 2506.11146 v1 pith:CY7DINUN submitted 2025-06-11 quant-ph cs.CVcs.LG

Jianhong Yao , Yangming Guo This is my paper · ORCID
classification quant-phcs.CVcs.LG
keywords quantumfuzzyneuralnetworkimageclassificationparameterizedcircuitsmembershipfunctiondefuzzificationdatare-uploadingnoiserobustnesshybridquantum-classicallearning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that the complete fuzzy inference chain—membership evaluation, rule aggregation, and defuzzification—can be implemented inside a shallow parameterized quantum circuit, fused with a lightweight CNN feature extractor, and trained end to end. The authors build HQFNN and claim it beats classical, fuzzy-enhanced, and quantum-only baselines on MNIST, Fashion-MNIST, Dirty-MNIST, JAFFE, and COVID-19 while using orders of magnitude fewer trainable weights. A sympathetic reader would care because a working quantum-native fuzzy classifier would combine graded, transparent reasoning with a compact circuit and a plausible path to larger images. The paper supports the claim with accuracy tables, simulated noise-channel fidelity measurements, expressibility and entanglement scores, and a gate-count analysis.

What carries the argument

The load-bearing object is the quantum fuzzy composition module, and its job is to keep the entire fuzzy pipeline differentiable and compact. Membership is built by treating each feature as a rotation angle $R_y(x_i+b_i)$ and re-uploading it on a single qubit, so the measured Pauli-Z expectation is rescaled to a fuzzy degree $\mu_i=(z_i+1)/2$; the rule layer then slides a shared convolutional kernel along the membership axis, which the paper reads as a first-order Taylor surrogate of the product T-norm; the defuzzifier linearly maps the refined memberships to angles, prepares $R_x$ rotations, entangles them in triangular CNOT clusters, and returns the average of two projected halves. This module replaces the classical rule-aggregation and defuzzification stages that earlier quantum-fuzzy hybrids kept outside the circuit, which is exactly the gap the paper claims to close.

What would settle it

Train HQFNN's exact two-layer CNN stem and two-layer classifier without the quantum-fuzzy branch on MNIST and Fashion-MNIST using the same hyperparameters; if that classical model reaches or exceeds HQFNN's 99.40% and 92.81%, the claimed quantum-fuzzy advantage is not supported. A second decisive check replaces the trained quantum membership rotations with fixed random angles; if accuracy does not drop, the learned quantum memberships are not doing the work.

Watch

Extended reading notes

Core claim

The central discovery, on the paper's own terms, is that fuzzy logic need not be split between a quantum encoder and a classical inference head. HQFNN defines a single differentiable quantum fuzzy composition module: a membership function that angle-encodes each CNN-extracted feature into a single qubit through repeated $R_y(x_i+b_i)$ re-uploading; a rule layer that treats a 1-D convolution over the membership axis as a trainable surrogate for the product T-norm; and a defuzzifier that projects memberships to $R_x$ rotation angles, entangles the qubits with clustered CNOTs, maps measurements back to $[0,1]$, and averages two projected halves into one crisp scalar. That scalar is concatenated with the classical CNN embedding and passed to a linear classifier, so gradients flow from the loss back through the defuzzifier, rule layer, and membership circuit. The paper reports 99.40% accuracy on MNIST and 92.81% on Fashion-MNIST, beating the six baselines in Table III, and reports gains over quantum-fuzzy baselines on JAFFE and COVID-19, with a statistical tie on Dirty-MNIST. It also reports that circuit depth grows sublinearly with input dimension and that fidelity under amplitude-damping and phase-flip noise stays above 0.94 at 10% noise strength.

Load-bearing premise

The load-bearing premise is that the six baselines in Table III were configured with fair, comparably tuned hyperparameters and similar feature extractors, and, as the paper's own conclusion notes, that results from an ideal simulator carry over meaningfully to real quantum hardware.

Editorial extensions

If this is right

  • End-to-end quantum-native fuzzy inference becomes a practical architecture, because backpropagation no longer stops at a classical defuzzification stage.
  • The reported parameter efficiency, if reproduced, would make hybrid quantum-fuzzy classifiers attractive in memory-constrained or data-limited settings.
  • Sublinear growth of circuit depth with input dimension would keep the quantum block tractable for higher-resolution images.
  • Sustained high fidelity under amplitude-damping and phase-flip noise would imply the model needs no explicit error mitigation on near-term devices.

Reading between the lines

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

  • A reader could settle the quantum branch's contribution by ablating it: replace the QMF with fixed random rotations or classical sigmoid memberships while keeping the CNN stem identical; the accuracy gap would separate quantum-fuzzy effects from classical feature learning. This test is not in the paper.
  • Because the membership stage uses one qubit per rule and only the defuzzifier entangles qubits, the entire pipeline is classically simulable; if a classical simulation with the same functional form matches HQFNN, the claimed advantage may reside in the parameterization rather than in quantum entanglement.
  • The expressibility and entanglement scores are circuit-level statistics, not task-level predictions; connecting them to held-out accuracy would test whether the reported trade-off actually explains generalization.
  • The hyperparameter analysis shows the learning rate dominates while activation function and qubit count barely matter, which suggests the classical stem is small; HQFNN might retain most of its accuracy with an even lighter classical backbone, something the paper does not explore.
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 paper proposes HQFNN, a hybrid quantum-classical network for image classification in which a lightweight CNN stem produces features that are fed into a quantum fuzzy membership module, a classical rule-aggregation layer, a quantum defuzzification circuit, and finally a fully connected classifier. The authors claim that this design realizes the full fuzzy pipeline inside a shallow quantum circuit, consistently outperforms classical, fuzzy-enhanced, and quantum-only baselines on MNIST, Fashion-MNIST, Dirty-MNIST, JAFFE, and COVID-19, uses orders of magnitude fewer trainable weights, and tolerates gate-level noise. They also report expressibility, entangling-capability, hyperparameter-sensitivity, and complexity analyses. The architecture is described in detail and the experiments cover several datasets, but the central empirical claims are not established by the present evidence: there is no classical-only ablation, no error bars or repeated runs, the robustness section reports state fidelity rather than classification accuracy under noise, and the claimed parameter savings are not compared with baseline parameter counts.

Significance. If the claims were fully supported, HQFNN would be a useful compact hybrid quantum-fuzzy classifier with an end-to-end trainable pipeline and standard expressibility diagnostics. The paper has strengths: the circuit construction is clearly specified in Section III and Algorithm 1, the expressibility and entanglement analysis uses established measures (Sim et al.), and the complexity discussion in Section IV.F is a good start. However, the load-bearing evidence for consistent superiority over baselines is missing key controls, and the robustness claim in the abstract is not matched by the experiments actually reported. The contribution is therefore currently an architectural proposal with promising but unverified performance claims, rather than a demonstrated advance.

major comments (4)
  1. [§IV.B, Tables III–IV] The central claim that HQFNN 'consistently surpasses classical, fuzzy enhanced and quantum only baselines' is not supported because no plain CNN using the same two-layer stem and classifier is evaluated. HQFNN always couples its quantum-fuzzy branch with a classical two-layer CNN feature extractor, a classical rule layer, and a fully connected classifier, so the reported gains over QA-HFNN could plausibly come from these classical components alone. The paper must add (i) a CNN-only baseline with the identical stem, fusion, and classifier; (ii) an HQFNN ablation with the quantum branch removed; and (iii) repeated-run statistics with seeds, means, and error bars, since the reported margins are small (e.g., Dirty-MNIST is exactly tied at 84.00, JAFFE +1.04, COVID-19 +0.83).
  2. [§IV.C, Table V] The abstract and Section IV.C state that HQFNN's 'accuracy degrades only marginally' under depolarizing and amplitude damping noise, but the reported robustness evaluation computes the state fidelity F(ρ,σ) between ideal and noisy circuit outputs, not the classification accuracy of the full model. Table V and Fig. 5 therefore do not support the accuracy-robustness claim. The authors should report classification accuracy on the test sets as a function of noise probability for the same four channels, with the noisy channels applied to the full quantum blocks during inference.
  3. [§III, Rule Layer (Eqs. (6)–(10)) and Algorithm 1] The abstract's claim that HQFNN 'realises the entire fuzzy pipeline inside a shallow quantum circuit' is contradicted by the rule layer, which is a classical 1-D convolution with ReLU (Eq. (6)-(8)), and by the defuzzification stage's classical linear projection (Eq. (11)). The paper calls the rule layer part of the 'quantum fuzzy logic module,' but the equations and Algorithm 1 (lines 6-7) show only the membership and part of the defuzzification are quantum. The authors should either revise the 'fully quantum-native' claim or move the rule aggregation into the quantum circuit and demonstrate that the architecture is actually quantum-native end to end.
  4. [§IV.F and Table III] The claim that HQFNN uses 'several orders of magnitude fewer trainable weights' than the baselines is unsubstantiated because Table III reports only accuracy/precision/recall/F1 and no parameter counts for any baseline. The complexity analysis in §IV.F derives O(Bdm) as the 'end-to-end training complexity' but omits the CNN stem cost O(BCHW) and the classifier terms from the final complexity statement, and it does not compare against the parameter counts of QCNN, PP-HFNN, QFFL, QNN, QFNN, or QA-HFNN. The authors should report a parameter-count comparison table and correct the complexity summary.
minor comments (6)
  1. [Abstract] There is a grammar error: 'a innovative Highly Quantized Fuzzy Neural Network' should be 'an innovative...'.
  2. [§IV.A, Table II] The text says 'See Table 1 for default settings,' but the hyperparameter table is Table II; please correct the cross-reference.
  3. [§IV.A and code availability] The code link in the abstract is a placeholder ('httpXXXXX'); a working repository link should be provided if the reproducibility claim is to be credible.
  4. [§III, Eq. (6)] The tensor notation is inconsistent: the text defines M ∈ R^{B×m×d×q} but Eq. (6) uses H ∈ R^{B×m×d}; please align the symbols for the membership tensor and the rule-layer input throughout Section III.
  5. [§IV.D] The expressibility-entangling analysis selects the default QMF=4, QD=6 based on a trade-off between two circuit metrics, but no classification accuracies for the alternative configurations (Table VI) are reported; the link between these metrics and task accuracy is asserted rather than demonstrated.
  6. [§IV.E, Table VII] The hyperparameter analysis reports 'average change' in accuracy without stating how many runs or seeds were used; a single sentence on the number of trials and variance would make the sensitivity claims interpretable.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reported accuracies are empirical test-set measurements, the expressibility analysis uses the external Haar-random benchmark, the robustness numbers are computed from defined noise channels, and the reference list contains no self-citations.

full rationale

The paper's central claims do not reduce to their own inputs. The classification results in Tables III and IV are empirical test-set accuracies measured against held-out benchmark data; they are not derived from the model's equations, so no fitted parameter is being renamed as a prediction. The expressibility analysis (Section IV.D, Eq. 25) uses the external Haar-random fidelity distribution defined by Sim et al. [40] as a fixed benchmark, and the QMF=4 / QD=6 defaults are selected by comparing the circuit's own fidelity distribution with that external target; this is self-referential design tuning but not circular, since the Haar benchmark is parameter-free and independent of the paper's claims. The robustness evaluation (Section IV.C, Eqs. 20-24) applies standard amplitude-damping, depolarizing, bit-flip, and phase-flip channels to every single-qubit gate and computes the state fidelity between ideal and noisy outputs; the values in Table V (0.7757-0.9961) are measured consequences of the channel definitions and the circuit, not results forced by construction. No self-citation chain exists: references [1]-[62] contain no prior work by Yao or Guo, so the self-citation-load-bearing and uniqueness-imported-from-authors patterns do not apply. The architectural choices (Ry angle re-uploading, the Conv1d rule layer, clustered CNOT defuzzifier) are presented as design decisions supported by the expressibility and ablation experiments, not as conclusions imported from the authors' own prior papers. Separately, the paper has genuine evidence gaps that belong under correctness risk rather than circularity: no CNN-only or quantum-ablated control, so the claim that the quantum-fuzzy branch causes the observed accuracy is not established; the abstract's 'accuracy degrades only marginally' is supported only by state-fidelity measurements, never by classification accuracy under noise; and baseline fairness, seeds, and error bars are absent. These gaps weaken the conclusions but do not make any derivation equivalent to its inputs.

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

The central claims rest on standard quantum mechanics, on the assumption that the Conv1d rule layer implements a fuzzy t-norm surrogate, and on the choice of hyperparameters through the paper's own expressibility analysis. No novel physical entities are postulated.

free parameters (5)
  • QMF circuit layers = 4
    Chosen from expressibility and entanglement trade-off in Section IV.D, at which the highest performance is reported.
  • QD qubits = 6
    Chosen as the best balance between expressibility and entanglement in Section IV.D.
  • Learning rate = 0.001
    Hand-chosen in Table II; hyperparameter analysis identifies LR as the most sensitive setting.
  • Batch size = 500
    Hand-chosen in Table II.
  • Training epochs = 200
    Used in the MNIST convergence analysis; default for other datasets is not specified.
assumptions (4)
  • standard math Standard quantum mechanics postulates and Pauli rotation definitions in Eqs. (1)-(5).
    The definitions of Rx, Ry, Rz and the Born rule are used without proof.
  • domain assumption Kraus operator representation of amplitude damping, depolarizing, bit-flip, and phase-flip channels in Eqs. (20)-(23).
    Assumed to model realistic hardware noise, but no hardware validation is provided.
  • ad hoc to paper The 1D convolution in Eq. (8) is a valid surrogate of the product t-norm for fuzzy rule aggregation as in Eq. (9).
    Introduced to justify the Conv1d rule layer; the logarithmic Taylor approximation is not shown to hold for the learned memberships.
  • domain assumption The fuzzy membership values measured from Pauli-Z expectations in Eq. (5) faithfully represent fuzzy set membership degrees.
    The non-linear mapping may not correspond to semantically meaningful fuzzy memberships; the paper does not validate interpretability.

how reviews work

0 comments
Cite this review

Pith. "Pith review of HQFNN: A Compact Quantum-Fuzzy Neural Network for Accurate Image Classification." pith.science (2026). https://pith.science/paper/CY7DINUN

@misc{pith2026250611146,
  author       = {Pith},
  title        = {Pith review of: HQFNN: A Compact Quantum-Fuzzy Neural Network for Accurate Image Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CY7DINUN}},
  note         = {Machine review of arXiv:2506.11146}
}
read the original abstract

Deep learning vision systems excel at pattern recognition yet falter when inputs are noisy or the model must explain its own confidence. Fuzzy inference, with its graded memberships and rule transparency, offers a remedy, while parameterized quantum circuits can embed features in richly entangled Hilbert spaces with striking parameter efficiency. Bridging these ideas, this study introduces a innovative Highly Quantized Fuzzy Neural Network (HQFNN) that realises the entire fuzzy pipeline inside a shallow quantum circuit and couples the resulting quantum signal to a lightweight CNN feature extractor. Each image feature is first mapped to a single qubit membership state through repeated angle reuploading. Then a compact rule layer refines these amplitudes, and a clustered CNOT defuzzifier collapses them into one crisp value that is fused with classical features before classification. Evaluated on standard image benchmarks, HQFNN consistently surpasses classical, fuzzy enhanced and quantum only baselines while using several orders of magnitude fewer trainable weights, and its accuracy degrades only marginally under simulated depolarizing and amplitude damping noise, evidence of intrinsic robustness. Gate count analysis further shows that circuit depth grows sublinearly with input dimension, confirming the model's practicality for larger images. These results position the model as a compact, interpretable and noise tolerant alternative to conventional vision backbones and provide a template for future quantum native fuzzy learning frameworks.

Figures

Figures reproduced from arXiv: 2506.11146 by the authors.

Figure 1
Figure 1. The architecture of HQFNN. It consists of three parts: quantum membership function, fuzzy rule layer and quantum defuzzification layer. [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Visualization workflow of the HQFNN model. QG represents a trainable quantum gate. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Quantum circuit diagram of the whole quantum block. [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Training convergence of HQFNN on MNIST. operations, and measured. The measurement outcomes are subsequently re-scaled and averaged to yield a crisp quantum￾informed feature. Finally, concatenating this quantum-derived scalar with the classical feature embedding and pas…
Figure 5
Figure 5. Figure 5: The fidelity F of the output quantum state, as a function of the input parameter I and the noise probability P, quantifies the similarity between the [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Expressibility comparison of QNN in HQFNN. [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Hyperparameter analysis. (CL), and Qubits Number (QN) are chosen to implement the sensitivity analysis. Only one parameter is tweaked at a time, in the same time, the other numbers remain at their default values, which are also the configuration values at which the hig…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

62 extracted references · 37 canonical work pages

  1. [1]

    Explaining and harnessing adversarial examples,

    I. J. Goodfellow, J. Shlens, and C. Szegedy, “Explaining and harnessing adversarial examples,”arXiv preprint arXiv:1412.6572, 2014

    arXiv 2014

  2. [2]

    Towards deep learning models resistant to adversarial attacks,

    A. Mkadry, A. Makelov, L. Schmidt, D. Tsipras, and A. Vladu, “Towards deep learning models resistant to adversarial attacks,”stat, vol. 1050, no. 9, 2017

    work page 2017

  3. [3]

    Towards certification of uncertainty calibration under adversarial attacks,

    C. Emde, F. Pinto, T. Lukasiewicz, P. H. Torr, and A. Bibi, “Towards certification of uncertainty calibration under adversarial attacks,”arXiv preprint arXiv:2405.13922, 2024. 13

    arXiv 2024

  4. [4]

    A simple fuzzy rule-based system through vector membership and kernel-based granulation,

    P. Angelov and R. Yager, “A simple fuzzy rule-based system through vector membership and kernel-based granulation,” in2010 5th IEEE International Conference Intelligent Systems. IEEE, 2010, pp. 349– 354

    work page 2010

  5. [5]

    A Deep Neuro-Fuzzy Network for Image Classification

    O. Yazdanbakhsh and S. Dick, “A deep neuro-fuzzy network for image classification,”arXiv preprint arXiv:2001.01686, 2019

    work page Pith review arXiv 2001

  6. [6]

    Classification via deep fuzzy c-means clustering,

    M. Yeganejou and S. Dick, “Classification via deep fuzzy c-means clustering,” in2018 IEEE international conference on fuzzy systems (FUZZ-IEEE). IEEE, 2018, pp. 1–6

    work page 2018

  7. [7]

    Parameterized quantum circuits as machine learning models,

    M. Benedetti, E. Lloyd, S. Sack, and M. Fiorentini, “Parameterized quantum circuits as machine learning models,”Quantum science and technology, vol. 4, no. 4, p. 043001, 2019

    2019

  8. [8]

    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

Show all 62 references
  1. [9]

    An introduction to quantum machine learning,

    M. Schuld, I. Sinayskiy, and F. Petruccione, “An introduction to quantum machine learning,”Contemporary Physics, vol. 56, no. 2, pp. 172–185, 2015

    2015

  2. [10]

    A hi- erarchical fused quantum fuzzy neural network for image classification,

    S.-Y . Wu, R.-Z. Li, Y .-Q. Song, S.-J. Qin, Q.-Y . Wen, and F. Gao, “A hi- erarchical fused quantum fuzzy neural network for image classification,” arXiv preprint arXiv:2403.09318, 2024

    2024 arXiv

  3. [11]

    Distinctive image features from scale-invariant keypoints,

    D. G. Lowe, “Distinctive image features from scale-invariant keypoints,” International Journal of Computer Vision, vol. 60, no. 2, pp. 91–110, 2004

    2004

  4. [12]

    Histograms of oriented gradients for human detection,

    N. Dalal and B. Triggs, “Histograms of oriented gradients for human detection,” inProceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR), vol. 1, 2005, pp. 886–893

    2005

  5. [13]

    Support-vector networks,

    C. Cortes and V . Vapnik, “Support-vector networks,”Machine Learning, vol. 20, no. 3, pp. 273–297, 1995

    1995

  6. [14]

    Beyond bags of features: Spatial pyramid matching for recognizing natural scene categories,

    S. Lazebnik, C. Schmid, and J. Ponce, “Beyond bags of features: Spatial pyramid matching for recognizing natural scene categories,”Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR), pp. 2169–2178, 2006

    2006

  7. [15]

    Gradient-based learning applied to document recognition,

    Y . LeCun, L. Bottou, Y . Bengio, and P. Haffner, “Gradient-based learning applied to document recognition,” inProceedings of the IEEE, vol. 86, no. 11, 1998, pp. 2278–2324

    1998

  8. [16]

    Imagenet classification with deep convolutional neural networks,

    A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification with deep convolutional neural networks,” inAdvances in Neural Infor- mation Processing Systems, vol. 25. Curran Associates, Inc., 2012, pp. 1097–1105

    2012

  9. [17]

    Very deep convolutional networks for large-scale image recognition,

    K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” inInternational Conference on Learning Representations (ICLR), 2015, arXiv:1409.1556

    2015 arXiv

  10. [18]

    Going deeper with convolutions,

    C. Szegedy, W. Liu, Y . Jia, P. Sermanet, S. E. Reed, D. Anguelov, D. Erhan, V . Vanhoucke, and A. Rabinovich, “Going deeper with convolutions,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015, pp. 1–9

    2015

  11. [19]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770–778

    2016

  12. [20]

    Efficientnet: Rethinking model scaling for convolutional neural networks,

    M. Tan and Q. V . Le, “Efficientnet: Rethinking model scaling for convolutional neural networks,” inProceedings of the 36th International Conference on Machine Learning (ICML). PMLR, 2019, pp. 6105– 6114

    2019

  13. [21]

    An image is worth 16x16 words: Trans- formers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby, “An image is worth 16x16 words: Trans- formers for image recognition at scale,” inInternational Conference on Learnin...

    2021

  14. [22]

    A convnet for the 2020s,

    Z. Liu, H. Mao, C.-Y . Wu, C. Feichtenhofer, T. Darrell, and S. Xie, “A convnet for the 2020s,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp. 11 966– 11 976

    2022

  15. [23]

    Fuzzy sets,

    L. A. Zadeh, “Fuzzy sets,”Information and Control, vol. 8, no. 3, pp. 338–353, 1965

    1965

  16. [24]

    An experiment in linguistic synthesis with a fuzzy logic controller,

    E. H. Mamdani and S. F. Assilian, “An experiment in linguistic synthesis with a fuzzy logic controller,”IEEE Transactions on Computers, vol. C- 24, no. 12, pp. 1585–1588, 1975

    1975

  17. [25]

    G. J. Klir and B. Yuan,Fuzzy Sets and Fuzzy Logic: Theory and Applications. Prentice Hall PTR, 1995

    1995

  18. [26]

    Fuzzy identification of systems and its application to modeling and control,

    T. Takagi and M. Sugeno, “Fuzzy identification of systems and its application to modeling and control,”IEEE Transactions on Systems, Man, and Cybernetics, vol. SMC-15, no. 1, pp. 116–132, 1985

    1985

  19. [27]

    Anfis: Adaptive-network-based fuzzy inference system,

    J.-S. R. Jang, “Anfis: Adaptive-network-based fuzzy inference system,” IEEE Transactions on Systems, Man, and Cybernetics, vol. 23, no. 3, pp. 665–685, 1993

    1993

  20. [28]

    Interval type-2 fuzzy logic systems: Theory and design,

    Q. Liang and J. M. Mendel, “Interval type-2 fuzzy logic systems: Theory and design,”IEEE Transactions on Fuzzy Systems, vol. 8, no. 5, pp. 535–550, Oct 2000

    2000

  21. [29]

    Interval type-2 fuzzy logic systems made simple,

    J. M. Mendel, R. I. John, and F. Liu, “Interval type-2 fuzzy logic systems made simple,”IEEE Transactions on Fuzzy Systems, vol. 14, no. 6, pp. 808–824, Dec 2006

    2006

  22. [30]

    Deep neuro-fuzzy system application trends, challenges, and future perspectives: A systematic survey,

    N. Talpur, S. J. Abdulkadir, H. Alhussian, M. H. Hasan, N. Aziz, and A. Bamhdi, “Deep neuro-fuzzy system application trends, challenges, and future perspectives: A systematic survey,”Artificial Intelligence Review, vol. 56, no. 2, pp. 865–913, Apr 2022

    2022

  23. [31]

    Quantum neural computing,

    S. C. Kak, “Quantum neural computing,”Advances in Imaging and Electron Physics, vol. 94, pp. 259–313, 1995

    1995

  24. [32]

    Quantum optimiza- tion for training support vector machines,

    D. Anguita, S. Ridella, F. Rivieccio, and R. Zunino, “Quantum optimiza- tion for training support vector machines,”Neural Networks, vol. 16, no. 5–6, pp. 763–770, 2003

    2003

  25. [33]

    Quantum algorithm for data fitting,

    N. Wiebe, D. Braun, and S. Lloyd, “Quantum algorithm for data fitting,” Physical Review Letters, vol. 109, no. 5, p. 050505, 2012

    2012

  26. [34]

    Quantum algorithms for supervised and unsupervised machine learning,

    S. Lloyd, M. Mohseni, and P. Rebentrost, “Quantum algorithms for supervised and unsupervised machine learning,” arXiv preprint arXiv:1307.0411, 2013

    2013 arXiv

  27. [35]

    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

  28. [36]

    Quantum circuit learning,

    K. Mitarai, M. Negoro, M. Kitagawa, and K. Fujii, “Quantum circuit learning,”Physical Review A, vol. 98, no. 3, p. 032309, 2018

    2018

  29. [37]

    Quantum generative adversarial learning,

    S. Lloyd and C. Weedbrook, “Quantum generative adversarial learning,” Physical Review Letters, vol. 121, no. 4, p. 040502, 2018

    2018

  30. [38]

    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

  31. [39]

    q-means: A quantum algorithm for unsupervised machine learning,

    I. Kerenidis, J. Landman, A. Luongo, and A. Prakash, “q-means: A quantum algorithm for unsupervised machine learning,” inAdvances in Neural Information Processing Systems, vol. 32, 2019

    2019

  32. [40]

    Expressibility and entan- gling capability of parameterized quantum circuits for hybrid quantum- classical algorithms,

    S. Sim, P. D. Johnson, and A. Aspuru-Guzik, “Expressibility and entan- gling capability of parameterized quantum circuits for hybrid quantum- classical algorithms,”Advanced Quantum Technologies, vol. 2, no. 12, p. 1900070, 2019

    2019

  33. [41]

    Data re-uploading for a universal quantum classifier,

    A. P ´erez-Salinas, A. Cervera-Lierta, E. Gil-Fuster, and J. I. Latorre, “Data re-uploading for a universal quantum classifier,”Quantum, vol. 4, p. 226, 2020

    2020

  34. [42]

    Circuit-centric quantum classifiers,

    M. Schuld, A. Bocharov, K. M. Svore, and N. Wiebe, “Circuit-centric quantum classifiers,”Physical Review A, vol. 101, no. 3, p. 032308, 2020

    2020

  35. [43]

    Absence of barren plateaus in quantum convolutional neural networks,

    A. Pesah, M. Cerezo, S. Wang, T. V olkoff, A. T. Sornborger, and P. J. Coles, “Absence of barren plateaus in quantum convolutional neural networks,”Physical Review X, vol. 11, no. 4, p. 041011, 2021

    2021

  36. [44]

    Quantum noise protects quantum classifiers against adversaries,

    Y . Du, M.-H. Hsieh, T. Liu, D. Tao, and N. Liu, “Quantum noise protects quantum classifiers against adversaries,”Physical Review Research, vol. 3, no. 2, p. 023153, 2021

    2021

  37. [45]

    Temporal-spatial quantum graph con- volutional neural network based on schr ¨odinger approach for traffic congestion prediction,

    Z. Qu, X. Liu, and M. Zheng, “Temporal-spatial quantum graph con- volutional neural network based on schr ¨odinger approach for traffic congestion prediction,”IEEE Transactions on Intelligent Transportation Systems, 2022

    2022

  38. [46]

    QuantumNAS: Noise-Adaptive Search for Robust Quantum Circuits,

    H. Wang, Y . Ding, J. Gu, Y . Lin, D. Z. Pan, F. T. Chong, and S. Han, “QuantumNAS: Noise-Adaptive Search for Robust Quantum Circuits,” in2022 IEEE International Symposium on High-Performance Computer Architecture (HPCA). Seoul, Korea, Republic of: IEEE, Apr. 2022, pp. 692–708

    2022

  39. [47]

    Robustness of quantum reinforcement learning under hardware errors,

    A. Skolik, S. Mangini, T. B ¨ack, C. Macchiavello, and V . Dunjko, “Robustness of quantum reinforcement learning under hardware errors,” EPJ Quantum Technology, vol. 10, p. 43, 2023

    2023

  40. [48]

    Finding key nodes in complex networks through quantum deep reinforcement learning,

    J. Xiong, X.-L. Ren, and L. L ¨u, “Finding key nodes in complex networks through quantum deep reinforcement learning,”Entropy, vol. 27, no. 4, p. 382, 2025

    2025

  41. [49]

    Fuzzy set ideas in quantum logics,

    J. Pykacz, “Fuzzy set ideas in quantum logics,”International Journal of Theoretical Physics, vol. 31, no. 9, pp. 1767–1783, 1992

    1992

  42. [50]

    Fuzzy quantum logics,

    M. L. Dalla Chiara and R. Giuntini, “Fuzzy quantum logics,”Mathware and Soft Computing, vol. 3, no. 1-2, pp. 83–91, 1996

    1996

  43. [51]

    Quantum decision, quantum logic, and fuzzy sets,

    G. Melnichenko, “Quantum decision, quantum logic, and fuzzy sets,” arXiv preprint arXiv:0711.1437, 2007

    2007 arXiv

  44. [52]

    Quantum fuzzy neural network for multimodal sentiment and sarcasm detection,

    P. Tiwari, L. Zhang, Z. Qu, and G. Muhammad, “Quantum fuzzy neural network for multimodal sentiment and sarcasm detection,”Information Fusion, vol. 103, p. 102085, 2024

    2024

  45. [53]

    Quantum fuzzy federated learning for privacy protection in intelligent information processing,

    Z. Qu, L. Zhang, and P. Tiwari, “Quantum fuzzy federated learning for privacy protection in intelligent information processing,”IEEE Transactions on Fuzzy Systems, 2024. 14

    2024

  46. [54]

    Quantum assisted hierarchical fuzzy neural network for image classification,

    S. Wu, R. Li, Y . Song, S. Qin, Q. Wen, and F. Gao, “Quantum assisted hierarchical fuzzy neural network for image classification,” IEEE Transactions on Fuzzy Systems, 2024

    2024

  47. [55]

    Hierarchical graph contrastive learning framework based on quantum neural networks for sentiment analysis,

    K. Jia, F. Meng, and J. Liang, “Hierarchical graph contrastive learning framework based on quantum neural networks for sentiment analysis,” Information Sciences, vol. 690, p. 121543, 2025

    2025

  48. [56]

    Digit recognizer,

    AstroDave and W. Cukierski, “Digit recognizer,” https://kaggle.com/ competitions/digit-recognizer, 2012, kaggle

    2012

  49. [57]

    Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms,

    H. Xiao, K. Rasul, and R. V ollgraf, “Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms,” 2017. [Online]. Available: https://arxiv.org/abs/1708.07747

    2017 arXiv

  50. [58]

    Deterministic neural networks with appropriate inductive biases capture epistemic and aleatoric uncertainty,

    J. Mukhoti, A. Kirsch, J. van Amersfoort, P. H. Torr, and Y . Gal, “Deterministic neural networks with appropriate inductive biases capture epistemic and aleatoric uncertainty,”arXiv preprint arXiv:2102.11582, 2021

    2021 arXiv

  51. [59]

    ”excavating ai

    M. J. Lyons, “”excavating ai” re-excavated: Debunking a fallacious account of the jaffe dataset,” 2021. [Online]. Available: https: //arxiv.org/abs/2107.13998

    2021 arXiv

  52. [60]

    Coding facial expressions with gabor wavelets,

    M. Lyons, S. Akamatsu, M. Kamachi, and J. Gyoba, “Coding facial expressions with gabor wavelets,” inProceedings Third IEEE Interna- tional Conference on Automatic Face and Gesture Recognition, 1998, pp. 200–205

    1998

  53. [61]

    Can ai help in screening viral and covid-19 pneumonia?

    M. E. Chowdhury, T. Rahman, A. Khandakar, R. Mazhar, M. A. Kadir, Z. B. Mahbub, K. R. Islam, M. S. Khan, A. Iqbal, N. Al Emadiet al., “Can ai help in screening viral and covid-19 pneumonia?”Ieee Access, vol. 8, pp. 132 665–132 676, 2020

    2020

  54. [62]

    M. A. Nielsen and I. L. Chuang,Quantum computation and quantum information. Cambridge university press, 2010

    2010

Pith tools

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