Pith. sign in

REVIEW 4 major objections 5 minor 72 references

Unsupervised Feature Extraction and Reconstruction Using Parameterized Quantum Circuits

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

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

Pith's one-line read This paper claims that a quantum autoencoder with a QCNN encoder classifies MNIST 0 versus 1 at 97.59% average accuracy from a single compressed qubit, while full reconstruction remains poor.

desk verdict Incremental but checkable QAE benchmark; the headline accuracy may be in-sample because the SVM readout split is never specified. read the letter →

arxiv 2502.07667 v1 pith:CYI4Y2EP submitted 2025-02-11 quant-ph

Li-An Lo , Li-Yi Hsu , En-Jui Kuo This is my paper · ORCID
classification quant-ph PACS 03.67.-a03.67.Lx
keywords quantumautoencoderconvolutionalneuralnetworkunsupervisedfeatureextractionMNISTclassificationangleencodingBlochspheremeasurementvariationalcircuits
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 aims to show that a quantum autoencoder (QAE) can perform unsupervised feature extraction on classical image data, not just compress quantum states or find anomalies. Using a quantum convolutional neural network as the encoder, the authors compress eight angle-encoded qubits of MNIST data into one qubit and train with a reconstruction-style loss that never sees class labels. After training, the single compressed qubit's Bloch-sphere position separates the digits 0 and 1 well enough that an SVM-rotated measurement reaches 97.59% average accuracy. The same encoder paired with its inverse as decoder reconstructs the images much less successfully, and a run on real quantum hardware degrades with a bias toward 0. The paper's contribution is evidence that QAE-QCNN is a viable unsupervised feature extractor, with reconstruction quality as the current bottleneck.

What carries the argument

The load-bearing object is the ansatz-based quantum convolutional neural network (AQCNN) used as the encoder: two-qubit convolutional ansatze taken from a reference family (Type 1, 2, 3, with Type 1 giving the best results) interleaved with pooling operations that discard trash qubits, reducing n qubits to one compressed qubit. The decoder is the unitary inverse of the same circuit, so the full autoencoder needs c+2t qubits. The classification readout is a Bloch-sphere mechanism: measure the compressed qubit in all three Pauli directions, fit a support-vector-machine plane that separates the two digit classes, rotate the state so that plane becomes the equator, and then a single z-measurement decides the label. Training uses either mean-squared-error or binary-cross-entropy loss on the probability that the trash qubits are measured in a target state, and reconstruction quality is assessed by fidelity or trace distance.

What would settle it

Re-run the same training with the SVM readout plane fit only on data excluded from the 400-image test set, or replace the learned plane with one fixed axis; if accuracy falls well below 97.59%, the headline result is an artifact of fitting the readout to the test data.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that a quantum autoencoder with an ansatz-based QCNN encoder produces a one-qubit latent space that supports high-accuracy binary classification of handwritten digits. In the main experiment, 8-qubit angle-encoded MNIST images are compressed to a single qubit by three convolution-pooling layers; training minimizes a loss on the trash-qubit register without label information. Measuring the compressed qubit along x, y, and z reveals two digit-dependent clusters on the Bloch sphere, and after an SVM finds a separating plane and a rotation makes that plane the z-axis, a single measurement classifies the digit. Averaged over eight training runs, this gives 97.59% accuracy, beating the 88.40% and 88.78% of two generic programmable circuit architectures under identical conditions. The paper also reports that full encoder-decoder reconstruction is comparatively poor, that angle encoding outperforms amplitude encoding in this setting, and that on real quantum hardware accuracy drops and leans toward digit 0.

Load-bearing premise

The reported 97.59% accuracy assumes the SVM-fit Bloch-sphere measurement axis is chosen on data separate from the 400 test images used for the accuracy number, and the paper does not state such a split.

Editorial extensions

If this is right

  • A quantum autoencoder can act as a label-free feature extractor for classical image data: the 0-versus-1 MNIST task reaches 97.59% average accuracy with a QCNN encoder and one compressed qubit.
  • The QCNN encoder's convolutional inductive bias matters: it beats two generic programmable circuit architectures (88.40% and 88.78%) under the same 8-to-1 compression.
  • Angle encoding is the better embedding for this task, reversing the ranking reported for the supervised QCNN study the paper builds on.
  • The full encoder-decoder reconstruction is markedly weaker than feature extraction, so current QAE designs do not yet balance compression and reconstruction.
  • On real NISQ hardware the method degrades and develops a systematic zero bias, meaning the headline accuracy is a noiseless-simulation result rather than a device result.

Reading between the lines

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

  • Because the decoder is fixed as the inverse of the encoder and compression is 8-to-1, weak reconstruction is information-theoretically unsurprising: one qubit cannot carry enough information to reproduce arbitrary 8-qubit MNIST states, and the reconstruction loss may mostly shape the latent geometry rather than enable faithful decoding.
  • A natural follow-up the paper does not report is whether the same Bloch-sphere separation transfers across digit pairs or random train/test splits; if the SVM plane is fit on the evaluation set, the 97.59% figure is an in-sample measurement.
  • The angle-encoding advantage is reported at 8 qubits; with larger qubit counts or different circuit families, amplitude encoding's exponential capacity could plausibly reverse the ranking.
  • The zero bias seen on real hardware is consistent with state-preparation or measurement noise; running the trained circuits under simulated depolarizing noise would test whether that bias is a noise signature rather than a property of the learned latent space.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper studies quantum autoencoders (QAEs) with parameterized quantum circuits for unsupervised feature extraction and reconstruction on MNIST. The authors propose a QCNN-based encoder (with convolutional and pooling ansatze) and compare it against two universal programmable circuits. Binary classification is performed without supervised training: the QAE is trained with a reconstruction loss, and a single compressed qubit is measured along x, y, and z; an SVM is then used to find a separating plane on the Bloch sphere, which is rotated to the z-axis for final classification. The central numerical claim is that the QCNN-based QAE reaches 97.59% average accuracy on binary MNIST (0 vs 1), outperforming the two comparison architectures. The paper also evaluates reconstruction success rates across architectures and tests different encodings, loss functions, pooling/convolutions, multi-class settings, and a real IonQ device.

Significance. If the accuracy claim is validated, the paper would provide a useful benchmark for unsupervised quantum feature extraction with QAEs and QCNNs, complementing supervised QCNN studies. The work is explicitly comparative, includes multiple architecture variants, reports parameter counts and CNOT counts, and makes code and data publicly available on GitHub. A genuine strength is that the QAE is trained with a reconstruction objective, so the classification pipeline is unsupervised in the training phase. However, the central accuracy figure is currently not supported as an out-of-sample estimate because the manuscript does not specify whether the SVM readout plane is fit on data disjoint from the 400 samples used for scoring. The paper also contains a derivational gap in the perfect-reconstruction condition and lacks variance information for the main accuracy comparison. These issues are fixable but are load-bearing for the paper's main claim.

major comments (4)
  1. [Sec. VII.B and VII.C] The reported 97.59% accuracy is not an out-of-sample estimate as written. Section VII.B states that 400 samples are randomly selected from the test dataset to calculate accuracy, and Section VII.C describes fitting an SVM plane on the Bloch sphere to separate 0 and 1 and then applying RX/RY rotations before measuring. The manuscript nowhere states that the SVM plane is trained on a set disjoint from the 400 samples used for the accuracy number. If the same 400 samples are used both to fit the three-parameter rotation plane and to score the result, Table I reports in-sample training performance, which is not a valid estimate of generalization. Please specify the readout train/test split or provide cross-validated accuracy with error bars.
  2. [Sec. III, Eq. (7)] Equation (7) does not follow from Eq. (6) without an additional assumption. From Eq. (6), the partial trace of E(ρin,θ) equals the reduced state of E(ρin,θ') up to the reference mapping, but equality of reduced states does not imply equality of the full encoder outputs in Eq. (7). The step appears to assume that the encoder output factorizes into compressed and trash subsystems and that the partial trace determines the full state, which is not generally true. As written, the perfect-reconstruction characterization leading to θ=θ' and |a⟩=|t⟩ is not justified; please state the required assumption or revise the derivation.
  3. [Sec. VII.B, Table I] The claim that QCNN significantly outperforms Architectures A and B rests on an average over 8 training runs, but no standard deviation, confidence interval, or significance test is reported. With 400 test samples and 8 runs, the differences between 97.59%, 88.40%, and 88.78% need variance information to support the statement that QCNN 'significantly outperforms' the other architectures. Please report per-run accuracies, the mean and standard deviation, and a statistical significance test.
  4. [Table II caption] The reconstruction success rate as written, ⟨D(Trt(E(|ϕ(x)⟩,θ1)⊗|a⟩,θ2)|ϕ(x)⟩, is not a well-defined success rate: it is a complex amplitude without modulus squared, and no averaging over the dataset is specified. Please define the success rate as a probability (e.g., |⟨·|·⟩|² or measurement frequency) and state how it is computed over the test set.
minor comments (5)
  1. [Throughout] There are numerous typographical errors, including 'applicaitons,' 'atterntion,' 'leraning,' 'renferece,' 'T raining,' and 'DA T A A V AILIBILITY,' which should be corrected before publication.
  2. [Sec. VII.D.3 and Sec. VII.B] Section VII.D.3 states that MSE is more suitable than BCE, while the main result in Section VII.B uses BCE; this apparent inconsistency should be resolved or explicitly discussed.
  3. [Sec. III, Eq. (2)] Equation (2) contains a stray θ' after the closing parenthesis: D(ρcomp ⊗ ρa, θ') should not be followed by a second θ'.
  4. [Sec. VI, Eqs. (10)-(11)] The MSE and BCE losses sum over T, described as the number of trash qubit states, but the index of summation is not defined; the projection probabilities for |0⟩ and |1⟩ should be written as probabilities rather than squared norms of projector expressions.
  5. [Tables] In the version I reviewed, the table bodies for Tables I, III, IV, V, VI, and VII are absent, leaving only captions; please ensure that all tables with numerical entries appear in the final PDF.

Circularity Check

1 steps flagged · score 6.0 of 10

Reported 97.59% accuracy is an in-sample SVM-readout fit as written; otherwise the QAE derivation is non-circular.

  1. fitted input called prediction [Section VII.B 'Training' and Section VII.C 'Rotate The Bloch Sphere']
    "We randomly selected 400 samples from the test dataset to calculate accuracy. ... Therefore, to confirm the performance of our result, we measure all three directions ( x, y, and z) to determine the position of the current compressed state. Using an Support Vector Machine(SVM) [22], we find a plane equation on the Bloch Sphere that separates 0 and 1. Subsequently, we apply rotations, as shown in Figure 6, such that this plane becomes perpendicular to the z-axis."

    The 97.59% accuracy is computed on 400 test samples whose compressed Bloch-sphere coordinates are also used to fit the SVM readout plane. The paper never states that the SVM plane is fit on data disjoint from those 400 scored samples. As written, the same points determine the plane and are then scored after rotating to that plane. The reported accuracy is therefore the training accuracy of a three-parameter linear classifier on the very samples being evaluated, not an out-of-sample estimate. This makes the headline accuracy a fitted quantity relabeled as a prediction rather than an independent test of the unsupervised feature extractor.

full rationale

The core QAE construction is not circular by its own equations: the encoder is a parameterized circuit trained with a reconstruction loss, and the decoder is its inverse, so the reconstruction objective is a genuine unsupervised task. The classification claim, however, depends on a classical SVM readout whose training/test separation is unspecified. Because the same 400 test samples are the only data mentioned for both accuracy calculation and SVM plane fitting, the central 97.59% figure reduces, as written, to an in-sample fit of the readout. No load-bearing self-citation chain or uniqueness argument appears: author self-citations ([11], [60]) are not used to justify the main derivation, and the QAE/QCNN architecture is adopted from independent prior work ([30], [40], [45]). The paper also contains internal inconsistencies (e.g., Table V says MSE is more suitable, while the conclusion says BCE is optimal), but these are correctness issues rather than circularity. The missing train/readout split is the only step that makes a reported prediction statistically forced, so the score reflects partial circularity of the classification result without indicting the reconstruction derivation itself.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim depends on a handful of choices the paper does not justify independently: the SVM rotation plane is a parameter fit to measured data, the classical autoencoder that pre-reduces MNIST is a learned model whose parameters are not described, and the QCNN depth is a hand-picked hyperparameter. The theoretical reconstruction condition assumes a factorization of the encoder output and a decoder that is the exact unitary inverse. The quantum mechanics background is standard.

free parameters (3)
  • SVM readout plane (rotation angles RX, RY) = not reported
    The Bloch-sphere rotation that maps the separating plane to the z-axis is determined by an SVM fit on measured x, y, z coordinates (Sec VII.C). If this plane is fit on the evaluation set, the accuracy depends on a parameter fitted to the same data. The final rotation angles are not given.
  • Classical autoencoder latent dimension = 8 features (implied by 8 qubits angle encoding)
    MNIST images are reduced to 8 (or 16) features by a classical autoencoder before quantum encoding (Sec VII.A). The classical AE parameters are learned from the data and are not described or ablated.
  • Number of QCNN layers = 3 (C-P-C-P-C-P)
    The architecture depth is a hand-chosen hyperparameter; Table VI shows 16-qubit with 4 layers performs worse, so depth materially affects results.
assumptions (4)
  • standard math Quantum mechanics: unitary evolution, partial trace, CPTP maps, fidelity.
    Section III uses these without proof.
  • domain assumption The decoder is taken to be the unitary inverse of the encoder (D = E†).
    Section III Eqs (5)-(6) assume the decoder is exactly the inverse encoder; this is a design choice that does not guarantee optimal reconstruction.
  • domain assumption Classical autoencoder preprocessing preserves the class information needed for 0-vs-1 separation.
    Section VII.A reduces MNIST with a classical autoencoder before angle encoding; if this preprocessing destroys label-relevant structure, the quantum part cannot recover it, and if it preserves it, the quantum contribution is partly redundant.
  • ad hoc to paper The encoder output can be written as a tensor product of compressed and trash subsystems in the perfect-reconstruction condition.
    Equation (7) in Section III assumes E(ρin,θ) factorizes as Trt(E(ρin,θ)) ⊗ ρt, which is not guaranteed; the derivation of the reconstruction condition depends on this factorization.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Unsupervised Feature Extraction and Reconstruction Using Parameterized Quantum Circuits." pith.science (2026). https://pith.science/paper/CYI4Y2EP

@misc{pith2026250207667,
  author       = {Pith},
  title        = {Pith review of: Unsupervised Feature Extraction and Reconstruction Using Parameterized Quantum Circuits},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/CYI4Y2EP}},
  note         = {Machine review of arXiv:2502.07667}
}
read the original abstract

Autoencoders are fundamental tools in classical computing for unsupervised feature extraction, dimensionality reduction, and generative learning. The Quantum Autoencoder (QAE), introduced by Romero J.[2017 Quantum Sci. Technol. 2 045001], extends this concept to quantum systems and has been primarily applied to tasks like anomaly detection. Despite its potential, QAE has not been extensively evaluated in basic classification tasks such as handwritten digit classification, which could provide deeper insights into its capabilities. In this work, we systematically investigate the performance of QAE in unsupervised feature extraction and reconstruction tasks. Using various encoder and decoder architectures, we explore QAE's ability to classify MNIST and analyze its reconstruction effectiveness. Notably, with a QCNN-based encoder, QAE achieves an average accuracy of 97.59% in binary classification of MNIST (0 and 1), demonstrating the feasibility of QAE for such tasks. This result highlights the synergy between QAE and QCNN in achieving optimal feature extraction performance. However, we also identify limitations in QAE's reconstruction capability, underscoring the need for further advancements in encoder and decoder designs. Our findings provide a foundation for future research on leveraging QAE in practical quantum machine learning applications.

Figures

Figures reproduced from arXiv: 2502.07667 by the authors.

Figure 1
Figure 1. FIG. 1: Graphical representation of the quantum convolutional autoencoder. The unitary map [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. FIG. 2: Circuit implementation of the quantum convolutional autoencoder with QCNN as encoder. [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. FIG. 3: The convolutional structure we employed is illustrated here, showcasing three of the [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: FIG. 4: The pooling structure we used, utilizing two configurations. We call them ”ZX” and [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: FIG. 5: Two [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: FIG. 6: The left figure shows the results of feature extraction on Bloch spheres compressing 8 [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]
Figure 7
Figure 7. Figure 7: FIG. 7: 3 Layers (left) and 2 Layers (right): The x [PITH_FULL_IMAGE:figures/full_fig_p018_7.png]
Figure 8
Figure 8. Figure 8: FIG. 8: The results of quantum measurements with 8 qubits (left) / 16 qubit (right) and classical [PITH_FULL_IMAGE:figures/full_fig_p019_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

72 extracted references · 64 canonical work pages

  1. [1]

    Unsupervised feature extraction 14

    work page

  2. [2]

    Reconstruction Success Rate Across Different Architectures 14

    work page

  3. [3]

    Other Category Tests 15

    work page

  4. [4]

    Multi-Class Classification Results Across Different Layers 17

    work page

  5. [5]

    Conclusion 19 IX

    Real simulation on Quantum Computer 18 VIII. Conclusion 19 IX. Data Availibility 20 X. Acknowledgments 20 References 20 2 I. INTRODUCTION Unsupervised feature extraction refers to an effective method in machine learning that identifies and selects important features from unlabeled datasets, and the original high- dimensional data are transformed into info...

    work page

  6. [6]

    The encoder architecture is the same as what we discussed in Section IV

    Unsupervised feature extraction Accuracy of Binary Classification 0 vs 1 across different architectures In Table I, ”Half” refers to using only the encoder in QAE. The encoder architecture is the same as what we discussed in Section IV. This results show that the QCNN architecture achieved an average accuracy of 97.59 %, significantly outperforming archit...

    work page

  7. [7]

    This evaluation aims 14 TABLE I: Comparison of QCNN, Architecture A, and Architecture B in Unsupervised Feature Extraction, Parameter Counts, and Number of CNOT Gates

    Reconstruction Success Rate Across Different Architectures The results, shown in Table II, include both the encoder and decoder, which doubles the parameter count compared to the unsupervised feature extraction test. This evaluation aims 14 TABLE I: Comparison of QCNN, Architecture A, and Architecture B in Unsupervised Feature Extraction, Parameter Counts...

    work page

  8. [8]

    These tests included comparisons of different encoding methods, loss functions, and numbers of qubits, all conducted using the QCNN architecture with a 15 single compressed qubit

    Other Category Tests Next, we tested additional categories to determine the most suitable approach for unsu- pervised feature extraction. These tests included comparisons of different encoding methods, loss functions, and numbers of qubits, all conducted using the QCNN architecture with a 15 single compressed qubit. As shown in Table III, Angle Encoding o...

    work page

Show all 72 references
  1. [9]

    For the 3-class classification, we used the digits 0, 1, and 8, while for the 4-class classification, we included the digits 0, 1, 2, and 3

    Multi-Class Classification Results Across Different Layers In addition to binary classification of MNIST, we evaluated the performance on 3-class and 4-class classifications. For the 3-class classification, we used the digits 0, 1, and 8, while for the 4-class classification, ...

  2. [10]

    Using the best results from the 16-qubit and 8-qubit simulations, we tested 100 MNIST data points (50 zeros and 50 ones) on the IonQ quantum computer using 1000 shots

    Real simulation on Quantum Computer Finally, we evaluated the performance on a real quantum computer. Using the best results from the 16-qubit and 8-qubit simulations, we tested 100 MNIST data points (50 zeros and 50 ones) on the IonQ quantum computer using 1000 shots. The res...

  3. [11]

    Binary cross entropy with deep learning technique for image classifi- cation[J]

    Ruby U, Yendapalli V. Binary cross entropy with deep learning technique for image classifi- cation[J]. Int. J. Adv. Trends Comput. Sci. Eng, 2020, 9(10)

    2020

  4. [12]

    Romero, C

    A. Romero, C. Gatta, and G. Camps-Valls, Unsupervised Deep Feature Extraction for Remote Sensing Image Classification, IEEE Trans. Geosci. Remote Sens. 54 1349 (2016)

    2016

  5. [13]

    Riedhammer, B

    K. Riedhammer, B. Favre, and D. Hakkani-T¨ ur, Long story short–global unsupervised models for keyphrase based meeting summarization, Speech Comm. 52 801 (2010)

    2010

  6. [14]

    Taguchi, Unsupervised Feature Extraction Applied to Bioinformatics: A PCA Based and TD Based Approach (Springer International Publishing, 2019)

    Y.-H. Taguchi, Unsupervised Feature Extraction Applied to Bioinformatics: A PCA Based and TD Based Approach (Springer International Publishing, 2019)

    2019

  7. [15]

    Y. Wang, H. Yao, S. Zhao, Auto-encoder based dimensionality reduction, Neurocomputing 184, 232 (2016). 20

    2016

  8. [16]

    Yousefi-Azar, V

    M. Yousefi-Azar, V. Varadharajan, L. Hamey and U. Tupakula, Autoencoder-based feature learning for cyber security applications, In Proceedings of the 2017 International Joint Con- ference on Neural Networks (IJCNN), IEEE, 3854 (2017)

    2017

  9. [17]

    Vincent, H

    P. Vincent, H. Larochelle, Y. Bengio, and P.-A. Manzagol. Extracting and composing robust features with denoising autoencoders, In Proceedings of the 25th international conference on Machine learn- ing, 1096 (2008)

    2008

  10. [18]

    Z. Chen, C. K. Yeo, B. S. Lee and C. T. Lau, Autoencoder-based network anomaly detection, Iin Wireless Telecommunications Symposium (WTS), 1 (2018)

    2018

  11. [19]

    Parmar, D

    G. Parmar, D. Li, K. Lee, and Z. Tu, Dual contradistinctive generative autoencoder. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition 823 (2021)

    2021

  12. [20]

    P. Li, Y. Pei, J. Li, A comprehensive survey on design and application of autoencoder in deep learning, Appl. Soft Comput. 138, 110176 (2023)

    2023

  13. [21]

    Decoding conformal field theories: From supervised to unsupervised learning[J]

    Kuo E J, Seif A, Lundgren R, et al. Decoding conformal field theories: From supervised to unsupervised learning[J]. Physical Review Research, 2022, 4(4): 043031

    2022

  14. [22]

    Krizhevsky, I

    A. Krizhevsky, I. Sutskever, and G. E. Hinton, Imagenet classification with deep convolutional neural networks, Advances in neural information processing systems 3 25 (2012)

    2012

  15. [23]

    Masci, U

    J. Masci, U. Meier, D. Cireskan, and J. Schmidhuber, Stacked Convolutional Auto-Encoders for Hierarchical Feature Extraction, In International Conference on Artificial Neural Networks, Springer, Berlin, Heidelberg. 52-59 (2011)

    2011

  16. [24]

    Mean squared error[J]

    Error M S. Mean squared error[J]. MA: Springer US, 2010: 653-653

    2010

  17. [25]

    Vincent, H

    P. Vincent, H. Larochelle, Y. Bengio, and P. A. Manzagol, Extracting and composing robust features with denoising autoencoders. In Proceedings of the 25th international conference on Machine learning 1096 (2008)

    2008

  18. [26]

    D. Li, Z. Fu, J. Xu, Stacked-autoencoder-based model for COVID-19 diagnosis on CT images, Appl. Intell. 51, 2805 (2021)

    2021

  19. [27]

    M. Chen, X. Shi, Y. Zhang, D. Wu, M. Guizani, Deep feature learning for medical image analysis with convolutional autoencoder neural network, IEEE Trans. Big Data 7, 750 (2017). 21

    2017

  20. [28]

    Ribeiro, A

    M. Ribeiro, A. Lazzaretti, H. S. Lopes, A study of deep convolutional auto-encoders for anomaly detection in videos, Pattern Recognit. Lett, 105, 13 (2018)

    2018

  21. [29]

    V. A. Knyaz, O. Vygolov, V.V. Kniaz, Y. Vizilter, V. Gorbatsevich, T. Luhmann, and N. Conen, Deep learning of convolutional autoencoder for image matching and 3d object re- construction in the infrared range. In Proceedings of the IEEE International Conference on Computer Visi...

    2017

  22. [30]

    P. W. Shor, Polynomial-time algorithms for prime factorization and discrete logarithms on a quantum computer, SIAM Rev. 41 303 (1999)

    1999

  23. [31]

    L. K. Grover, Quantum mechanics helps in searching for a needle in a haystack. Phys. Rev. Lett. 79, 325 (1997)

    1997

  24. [32]

    Support vector machine[J]

    Suthaharan S, Suthaharan S. Support vector machine[J]. Machine learning models and algo- rithms for big data classification: thinking with examples for effective learning, 2016: 207-235

    2016

  25. [33]

    Arute et al., Quantum supremacy using a programmable superconducting processor, Nature 574, 505 (2019)

    F. Arute et al., Quantum supremacy using a programmable superconducting processor, Nature 574, 505 (2019)

    2019

  26. [34]

    Henderson, S

    M. Henderson, S. Shakya, S. Pradhan, and T. Cook, Quanvolutional neural networks: Pow- ering image recognition with quantum circuits, Quan. Mach. Intell. 2, 1 (2020)

    2020

  27. [35]

    Orduz, P

    J. Orduz, P. Rivas, and E. Baker, Quantum circuits for quantum convolutions: A quantum convolutional autoencoder, Trans. Comput. Sci. Comput. Intell. Springer 1 (2022)

    2022

  28. [36]

    Kea,W.-D

    K. Kea,W.-D. Chang, and Y. Han, Enhancing a convolutional autoencoder with quantum approximate optimization algorithm for image noise reduction, arXiv:2401.06367

  29. [37]

    Farhi, J

    E. Farhi, J. Goldstone, and S. Gutmann, A quantum approximate optimization algorithm. arXiv preprint arXiv:1411.4028

    arXiv

  30. [38]

    Cerezo, A

    M. Cerezo, A. Arrasmith, R. Babbush, S. C. Benjamin, S. Endo, K. Fujii, J. R. McClean, K. Mitarai, X. Yuan, L. Cincio, and P. J. Coles,Variational Quantum Algorithms, Nature Rev. Phys. 3, 625 (2021)

    2021

  31. [39]

    Sent ´ ıs, A

    G. Sent ´ ıs, A. Monr` as, R. Mu˜ noz-Tapia, J. Calsamiglia, and E. Bagan, Unsupervised classifi- cation of quantum data, Phys. Rev. X 9, 041029 (2019)

    2019

  32. [40]

    Romero, J

    J. Romero, J. P. Olson, and A. Aspuru-Guzik, Quantum autoencoders for efficient compression of quantum data, Quantum Sci. Technol. 2, 045001 (2017)

    2017

  33. [41]

    Bravo-Prieto, Quantum autoencoders with enhanced data encoding, Mach

    C. Bravo-Prieto, Quantum autoencoders with enhanced data encoding, Mach. Learn.: Sci. Technol. 2, 035028 (2021). 22

    2021

  34. [42]

    Bondarenko and P

    D. Bondarenko and P. Feldmann, Quantum autoencoders to denoise quantum data. Phys. Rev. Lett. 124 130502 (2020)

    2020

  35. [43]

    W.K. Mok, H. Zhang, T. Haug, X. Luo, G.Q. Lo, H. Cai, M.S. Kim, and A.Q. Liu, Rigorous noise reduction with quantum autoencoders, A VS Quantum Sci. 6, 023803 (2024)

    2024

  36. [44]

    M. M. Wang, Denoising quantum mixed states using quantum autoencoders, Quan. Inf. Proc. 23, 30 (2024)

    2024

  37. [45]

    Achache, L

    T. Achache, L. Horesh, and J. SmolinDenoising quantum states with Quantum Autoencoders - Theory and Applications, arXiv: 2012.14714

    2012 arXiv

  38. [46]

    Zhang, WKong, M

    X.-M. Zhang, WKong, M. U. Farooq, M.-H. Yung, G. Guo, and X. Wang, Generic detection- based error mitigation using quantum autoencoders, Phys. Rev. A 103, L040403 (2021)

    2021

  39. [47]

    D. F. Locher, L. Cardarelli, and M. M¨ uller, Quantum error correction with quantum autoen- coders, Quantum 7, 942 (2023)

    2023

  40. [48]

    Rocchetto, E

    A. Rocchetto, E. Grant, S. Strelchuk, G. Carleo, and S. Severini, Learning hard quantum distributions with variational autoencoders, npj Quantum Information, 4, 1 1 (2018)

    2018

  41. [49]

    A., Dhiraj Madan, A. Ray, D. Vinayagamurthy, M. S. Santhanam, Learning Hard Distribu- tions with Quantum-enhanced Variational Autoencoders, arXiv: 2305.01592

    arXiv

  42. [50]

    I. Cong, S. Choi, and M. D. Lukin, Quantum convolutional neural networks, Nat. Phys. 15, 1273 (2019)

    2019

  43. [51]

    MacCormack, C

    I. MacCormack, C. Delaney, A. Galda, N. Aggarwal, and P. Narang, Branching Quantum Convolutional Neural Networks, arXiv:2012.14439

    2012 arXiv

  44. [52]

    Franken and B

    L. Franken and B. Georgiev, Explorations in quantum neural networks with intermediate measurements, in Proceedings of ESANN, 2020

    2020

  45. [53]

    L. H. Gong, J. J. Pei, T. F. Zhang, and N. R. Zhou, Quantum convolutional neural network based on variational quantum circuits, Opt. Commun. 550, 129993 (2024)

    2024

  46. [54]

    Wei, Y.H

    S.J. Wei, Y.H. Chen, Z.R. Zhou, and G.L. Long, A quantum convolutional neural network on NISQ devices, AAPPS Bull. 32, 2 (2022)

    2022

  47. [55]

    T. Hur, L. Kim, D. K. Park, Quantum convolutional neural network for classical data classi- fication, Quan. Mach. Intell. 4, 3 (2022)

    2022

  48. [56]

    Matic, M

    A. Matic, M. Monnet, J. M. Lorenz, B. Schachtner, and T. Messerer, Quantum-classical convolutional neural networks in radiological image classification, in Proc. IEEE Int. Conf. Quantum Comput. Eng. (QCE), 56 (2022). 23

    2022

  49. [57]

    F. Fan, Y. Shi, T. Guggemos, and X.X. Zhu, Hybrid quantum-classical convolutional neural network model for image classification, IEEE Trans. Neural Netw. Learn. Syst. 18 (2023)

    2023

  50. [58]

    A. Wang, J. Hu, S. Zhang, L. Li, Shallow hybrid quantum-classical convolutional neural network model for image classification. Quan. Inf. Proc. 23, 17 (2024)

    2024

  51. [59]

    S. Shi, Z. Wang, J. Li, Y. Li, R. Shang, G. Zhong, and Y. Gu, Quantum convolutional neural networks for multiclass image classification, Quan. Inf. Proc. 23 189 (2024)

    2024

  52. [60]

    S. Y.-C. Chen, T.-C. Wei, C. Zhang, H. Yu, and S. Yoo, Quantum convolutional neural networks for high energy physics data analysis, Phys. Rev. Res. 4, 013231 (2022)

    2022

  53. [61]

    Venkatesh, K, J

    K. Venkatesh, K, J. NaikK, and A. Shankar, Quantum convolution neural network for multi- nutrient detection and stress identification in plant leaves, Multimed. Tools. Appl. 83 65663 (2024)

    2024

  54. [62]

    Bhatta and J

    S. Bhatta and J. Dang, Multiclass seismic damage detection of buildings using quantum convolutional neural network. Computer-Aided Civil and Infrastructure Engineering 39, 406 (2024)

    2024

  55. [63]

    J. Kima, J. Huh, and D. K. Park, Classical-to-quantum convolutional neural network transfer learning, Neurocomputing 555, 126643 (2023)

    2023

  56. [64]

    A ¨ ımeur, G

    E. A ¨ ımeur, G. Brassard, and S. Gambs, Quantum speed-up for unsupervised learning, Mach. Learn 90, 261(2013)

    2013

  57. [65]

    Lloyd, M

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

    arXiv

  58. [66]

    P. B. Sousa and R. V. Ramos, Universal quantum circuit for n-qubit quantum gate: A pro- grammable quantum gate, arXiv:quant-ph/0602174

    arXiv

  59. [67]

    Daskin, A

    A. Daskin, A. Grama, G. Kollias, S. Kais, Universal programmable quantum circuit schemes to emulate an operator, J. Chem. Phys. 137, 234112 (2012)

    2012

  60. [68]

    Quantum computing in the nisq era and beyond

    John Preskill. Quantum computing in the nisq era and beyond. Quantum, 2:79, 2018

    2018

  61. [69]

    Noisy intermediate-scale quantum algorithms

    Kishor Bharti, Alba Cervera-Lierta, Thi Ha Kyaw, Tobias Haug, Sumner Alperin-Lea, Ab- hinav Anand, Matthias Degroote, Hermanni Heimonen, Jakob S Kottmann, Tim Menke, et al. Noisy intermediate-scale quantum algorithms. Reviews of Modern Physics, 94(1):015004, 2022

    2022

  62. [70]

    Oracle separation be- tween noisy quantum polynomial time and the polynomial hierarchy

    Nai-Hui Chia, Min-Hsiu Hsieh, Shih-Han Hung, and En-Jui Kuo. Oracle separation be- tween noisy quantum polynomial time and the polynomial hierarchy. arXiv preprint 24 arXiv:2405.07137, 2024

    2024 arXiv

  63. [71]

    Multi-Class Quantum Convolutional Neural Net- works[C]//Proceedings of the 2024 Workshop on Quantum Search and Information Retrieval

    Mordacci M, Ferrari D, Amoretti M. Multi-Class Quantum Convolutional Neural Net- works[C]//Proceedings of the 2024 Workshop on Quantum Search and Information Retrieval. 2024: 9-16

    2024

  64. [72]

    The mnist database of handwritten digit images for machine learning research [best of the web][J]

    Deng L. The mnist database of handwritten digit images for machine learning research [best of the web][J]. IEEE signal processing magazine, 2012, 29(6): 141-142. 25

    2012

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.