Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

Quantum Kernel-Based Long Short-term Memory

T0 review · 4 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

Pith's one-line read Replacing an LSTM's linear gate operations with quantum kernel evaluations produces a smaller model that matches the classical network's training accuracy on a part-of-speech tagging task.

desk verdict New QK-LSTM architecture, but the parity claim rests on training curves for two sentences; the idea is worth exploring, the evidence isn't. read the letter →

arxiv 2411.13225 v1 pith:VACBLHHF submitted 2024-11-20 quant-ph cs.AI

classification quant-phcs.AI
keywords quantumkernelmethodslongshort-termmemorysequencemodelingmodelcompressionpart-of-speechtaggingmachinelearningNISQ-eradevices
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 sets out to show that quantum kernel functions can stand in for the linear transformations inside an LSTM cell without sacrificing sequence-modeling performance. It introduces QK-LSTM, in which each gate's usual weighted sum is replaced by a weighted sum of kernel similarities between the current input-hidden concatenation and a set of reference vectors. On a part-of-speech tagging exercise, the paper reports that QK-LSTM reaches training accuracy and loss curves comparable to classical LSTM and to a variational-circuit QLSTM while using 183 trainable parameters versus 477. The point of the exercise is model compression: if quantum feature spaces can carry the same expressive load with fewer parameters, sequence models become more feasible on NISQ-era and edge devices.

What carries the argument

The central object is the quantum kernel function $k(v_t,v_j)=|\langle\phi(v_t)|\phi(v_j)\rangle|^2$, a similarity measure in the quantum feature space induced by the circuit $U(v)$. It carries the argument by replacing the linear transformations in each LSTM gate with a kernel-weighted sum over reference vectors, so the model's non-linearity and expressiveness come from the quantum feature map rather than from a large weight matrix. The parameter-shift rule makes this kernel differentiable, allowing the same backpropagation-through-time training loop as a classical LSTM.

What would settle it

Train QK-LSTM and a classical LSTM on a standard part-of-speech corpus with a held-out test split and compare test accuracy at matched parameter counts; if QK-LSTM falls materially below LSTM, the on-par claim is refuted.

Watch

Extended reading notes

Core claim

QK-LSTM replaces the linear map $W[h_{t-1},x_t]+b$ in the forget, input, cell, and output gates of a standard LSTM with a sum $\sum_j \alpha_j k(v_t, v_j)+b$, where $v_t=[h_{t-1},x_t]$ and $k(v_t,v_j)=|\langle\phi(v_t)|\phi(v_j)\rangle|^2$ is a quantum kernel evaluated by a small parameterized circuit with Hadamard initialization, $R_y/R_z$ data encoding, and CNOT entanglement. The trainable weights $\alpha_j$ and biases are updated by backpropagation through time, and circuit-level gradients use the parameter-shift rule. On a two-sentence part-of-speech tagging benchmark, the model reaches training accuracy and loss curves comparable to classical LSTM and QLSTM, with 183 trainable parameters against the classical LSTM's 477. The paper's central claim is that quantum kernel evaluations are expressive enough to replace the linear gate transformations, compressing the model while keeping its learning dynamics intact.

Load-bearing premise

The parity claim leans on training curves from two manually chosen sentences with no held-out test set, so the reported accuracy could reflect fitting those particular sentences rather than general sequence-modeling ability.

Editorial extensions

If this is right

  • Sequence models can be compressed to roughly a third of the trainable parameters while preserving training performance on the reported task.
  • The standard LSTM training pipeline remains intact, so QK-LSTM can reuse backpropagation-through-time and parameter-shift updates without new optimization machinery.
  • Shallow kernel circuits, rather than deep variational ansatze, may be sufficient for quantum-enhanced sequence modeling, lowering the hardware requirement.
  • The approach extends in principle to other sequential tasks named in the paper, including time-series forecasting and signal classification.

Reading between the lines

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

  • The reported parity is measured on training curves for two hand-picked sentences, so the paper does not yet establish generalization; a held-out evaluation on a larger POS-tagged corpus would test whether the quantum kernel's inductive bias helps or hurts.
  • If the circuit parameters are also trained (the paper lists them as parameters), the true parameter count and quantum-resource cost could exceed 183, so the compression ratio depends on how the kernel circuit parameters are counted.
  • The reference vectors $v_j$ can be chosen from training data or learned; a natural test is to vary $N$ and measure how the accuracy-parameter trade-off changes, which would show how much of the compression is due to the quantum feature map versus the choice of reference points.
  • A classical kernel LSTM with the same kernel-weighted gate structure would isolate the quantum contribution; without that baseline, the improvement cannot be attributed to quantum resources alone.
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 Quantum Kernel-Based Long Short-Term Memory (QK-LSTM), which replaces the linear transformations in each LSTM gate with weighted sums of quantum kernel functions evaluated between the concatenated input and a set of reference vectors. The authors report an application to Part-of-Speech tagging on two manually selected sentences, showing training accuracy and loss curves that they interpret as achieving performance on par with classical LSTM and QLSTM while using fewer trainable parameters (183 versus 477). The manuscript also discusses parameter-shift-rule gradients for the quantum circuit and positions the architecture for NISQ/edge deployment.

Significance. The conceptual idea of substituting learned matrix multiplications with fixed quantum kernel expansions is worth exploring, and the paper identifies a relevant application area in sequence modeling. The explicit parameter-count comparison is a useful starting point, but the current manuscript does not provide a trustworthy evaluation: the only empirical evidence is in-sample fitting on two sentences, and the model equations contain an unresolved dimensional inconsistency. The contribution is therefore preliminary. No code, data, or machine-checked proofs are provided, so the claims cannot be independently verified.

major comments (4)
  1. [III-A, III-B, Fig. 3] The central claim that QK-LSTM 'achieves performance on par with classical LSTM models' is supported only by training accuracy and loss curves on two hand-picked sentences ('The dog eat the ice' and 'Everybody read that book'). There is no held-out test set, no repeated initialization, and no error bars. Reporting the training loss after fitting the model on the same two sentences demonstrates curve fitting, not generalization, so the parity claim in the abstract and Section III-B is not established. A standard POS benchmark with a train/test split and multiple seeded runs is required.
  2. [II-B-2, Eqs. (3a)-(3f), Table I] The gate equations are dimensionally inconsistent. Since k(vt,vj) in Eq. (8) is a scalar and the sums in Eqs. (3a)-(3e) are over scalar weights times scalar kernels, each gate activation is a scalar, not a vector. However, an LSTM hidden state ht and cell state Ct must be vectors, and Table I specifies a hidden dimension of 6. As written, Eq. (3f) produces a scalar hidden state, so the architecture cannot represent the multi-dimensional state that the LSTM framework requires. The authors need to define vector-valued kernels, per-output-dimension weights, or some other mechanism that preserves the hidden dimension.
  3. [II-B-4-c, Eqs. (9)-(10)] The parameter-shift rule is stated for 'a circuit parameter θ', but no trainable circuit parameter is defined in the feature map. In Eq. (7), Uenc(v) contains rotation angles that are functions of the input v and Uent is a fixed CNOT entangler; neither contains a learned parameter. Consequently, Eq. (13) updates an undefined θ, and the gradient formula in Eqs. (9)-(10) has no object to act on. If the quantum kernel is meant to be fixed after data encoding, the parameter-shift discussion should be removed and training should be over α and b only; if trainable circuit parameters are intended, they must be introduced explicitly in U(v) and the parameter-shift derivation must be repeated for that specific parameterization.
  4. [II-B-2, Table I] The compression claim is not reproducible because N, the number of reference vectors, is never specified. The trainable parameters in the QK-LSTM include the weights α_j^(gate) for each of the four gates, and the parameter count therefore scales with N; without N and an explicit counting formula, the reported value of 183 trainable parameters cannot be verified. The authors should state N and provide the exact parameter-count expressions for QK-LSTM and LSTM, including how biases and output layers are counted.
minor comments (6)
  1. [III-A] POS labels are given for 'The dog eat the ice' but not for the second sentence 'Everybody read that book'; please provide the full labeling or clarify that the second sentence is used without explicit labels.
  2. [Fig. 2(b)] The caption introduces U(x_i, w) and a parameter w that are not defined in the text; the notation should be consistent with Eq. (7).
  3. [III-B] The heading 'Performance Benchmmarking' contains a typo; it should be 'Performance Benchmarking'.
  4. [Eq. (3)] The four kernel functions k(f), k(i), k(C), and k(o) are called gate-specific, but Eq. (7) defines a single feature map U(v) for all gates; please explain how the four kernels differ or use one common kernel with gate-specific weights.
  5. [References] Reference [40] cites the PennyLane documentation; the parameter-shift rule should be attributed to the original works (e.g., Mitarai et al. or Schuld et al.) if the rule is retained.
  6. [Abstract and III-B] Phrases such as 'robust loss minimization' and 'efficient convergence' are overstatements when based on a single training run on two sentences; statistical support is needed.

Circularity Check

1 steps flagged · score 6.0 of 10

Parity claim is a training-set fit: no held-out evaluation exists, so 'performance on par' reduces to the optimized training curves rather than an independent benchmark.

  1. fitted input called prediction [Section III-A (Data Preprocessing) and Section III-B (Performance Benchmmarking), Fig. 3, Table I]
    "For illustrative purposes, we select two sentences—'The dog eat the ice' and 'Everybody read that book'—and manually assign POS tags to each word. ... Fig. 3(a) illustrates that the QK-LSTM attains accuracy levels comparable to the classical LSTM and QLSTM, with a similar rate of convergence despite the reduced parameter set."

    The abstract's claim that 'QK-LSTM achieves performance on par with classical LSTM models' is supported only by training accuracy and loss curves on the two manually selected sentences used to fit the model. The paper never describes a train/test split, held-out sentence, or generalization evaluation. Therefore the reported parity is the optimized training objective evaluated on the training data—the fitted value of the loss/accuracy—not an independent benchmark prediction. The empirical result reduces to reporting the fit itself.

full rationale

The main architectural derivation—replacing LSTM linear maps with quantum kernel-weighted sums in Eqs. (3a)-(3f)—is self-contained and not definitionally circular; the kernel expressions are new quantities and are not constructed from the LSTM output being predicted. The parameter-count comparison in Table I is also not circular, although it depends on the unspecified number of reference vectors N and chosen dimensions. The central circularity is confined to the empirical claim of 'performance on par': Section III-A fixes two training sentences, Section III-B reports Fig. 3 training accuracy and loss, and the abstract elevates this to 'benchmark comparisons reveal' parity. Since no held-out evaluation exists, the parity claim is the fitted training objective presented as an evaluation result, which fits the fitted-input-called-prediction pattern. Self-citations in the reference list (e.g., [12], [16], [24], [32]-[34]) are not load-bearing for the QK-LSTM derivation, so they do not raise the score. Separately, Eqs. (9)-(10) invoke a parameter-shift gradient with respect to θ, but the feature map U(v) in Eq. (7) has no trainable θ; this is a correctness gap, not a circularity.

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

The central claim rests on the standard LSTM framework, the validity of quantum kernels as similarity measures, and the unproven assumption that a small set of kernel-weighted references can replace learned linear transformations. The paper does not introduce new physical entities, but it does rely on several hand-chosen hyperparameters and an undefined trainable parameter scheme. The absence of a held-out test set means the reported performance is a direct fit to the training data.

free parameters (4)
  • Number of reference vectors N = Not specified
    The paper introduces N reference vectors in Eq. (3) but never states its value. This choice controls the parameter count and expressiveness of the model.
  • Hidden dimension = 6
    Table I sets the hidden dimension to 6, a hand-chosen value that affects model capacity and the reported parameter count.
  • Embedding dimension = 8
    Table I sets the embedding dimension to 8, another hand-chosen value influencing the input size to the LSTM.
  • Number of qubits = 4
    Table I lists 4 qubits for the quantum kernel circuit; this is a design choice affecting the dimension of the quantum feature space.
assumptions (4)
  • standard math Standard LSTM update equations (Eq. 1) are taken as the classical baseline.
    The paper relies on the standard LSTM formulation without proof.
  • domain assumption The quantum kernel k(vt, vj) = |<phi(vt)|phi(vj)>|^2 is a valid similarity measure for sequence data.
    This is a standard assumption in quantum kernel methods, but the paper provides no evidence that this particular feature map captures linguistic structure.
  • ad hoc to paper The parameter-shift rule applies to trainable circuit parameters in the quantum kernel, but such parameters are never defined.
    Section II-B-4-c invokes the parameter-shift rule for circuit parameters theta, yet Eq. (5) defines theta and phi as functions of input data, not as free parameters. The paper never introduces separate trainable rotation parameters.
  • ad hoc to paper Two handpicked sentences are representative of sequence modeling tasks.
    Section III-A selects two toy sentences without any justification that they represent the difficulty or diversity of real POS tagging datasets.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantum Kernel-Based Long Short-term Memory." pith.science (2026). https://pith.science/paper/VACBLHHF

@misc{pith2026241113225,
  author       = {Pith},
  title        = {Pith review of: Quantum Kernel-Based Long Short-term Memory},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VACBLHHF}},
  note         = {Machine review of arXiv:2411.13225}
}
read the original abstract

The integration of quantum computing into classical machine learning architectures has emerged as a promising approach to enhance model efficiency and computational capacity. In this work, we introduce the Quantum Kernel-Based Long Short-Term Memory (QK-LSTM) network, which utilizes quantum kernel functions within the classical LSTM framework to capture complex, non-linear patterns in sequential data. By embedding input data into a high-dimensional quantum feature space, the QK-LSTM model reduces the reliance on large parameter sets, achieving effective compression while maintaining accuracy in sequence modeling tasks. This quantum-enhanced architecture demonstrates efficient convergence, robust loss minimization, and model compactness, making it suitable for deployment in edge computing environments and resource-limited quantum devices (especially in the NISQ era). Benchmark comparisons reveal that QK-LSTM achieves performance on par with classical LSTM models, yet with fewer parameters, underscoring its potential to advance quantum machine learning applications in natural language processing and other domains requiring efficient temporal data processing.

Figures

Figures reproduced from arXiv: 2411.13225 by the authors.

Figure 1
Figure 1. Schematic representation of a standard classical LSTM cell. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of the QK-LSTM Architecture. (a) The QK-LSTM cell integrates quantum kernel transformations within the conventional LSTM framework, [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Training performance comparison for QLSTM, Classical, and QK [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Quantum Kernel-Based Long Short-term Memory for Climate Time-Series Forecasting

    quant-ph 2024-12 reject novelty 3.0 of 10

    A quantum kernel-based LSTM reports 42% lower RMSE than a classical LSTM on one AQI forecasting benchmark, with far fewer parameters.

Reference graph

Works this paper leans on

41 extracted references · 11 canonical work pages · cited by 1 Pith paper

  1. [1]

    Recurrent neural networks,

    L. R. Medsker, L. Jain, et al., “Recurrent neural networks,” Design and Applications, vol. 5, no. 64-67, p. 2, 2001

  2. [2]

    Long Short-Term Memory,

    S. Hochreiter and J. Schmidhuber, “Long Short-Term Memory,” Neural Computation, vol. 9, pp. 1735–1780, 11 1997

  3. [3]

    A review of recurrent neural networks: Lstm cells and network architectures,

    Y . Yu, X. Si, C. Hu, and J. Zhang, “A review of recurrent neural networks: Lstm cells and network architectures,” Neural computation , vol. 31, no. 7, pp. 1235–1270, 2019

  4. [4]

    Power of data in quantum machine learning,

    H.-Y . Huang, M. Broughton, M. Mohseni, R. Babbush, S. Boixo, H. Neven, and J. R. McClean, “Power of data in quantum machine learning,” Nature communications, vol. 12, no. 1, p. 2631, 2021

  5. [5]

    Machine learning of high dimensional data on a noisy quantum processor,

    E. Peters, J. Caldeira, A. Ho, S. Leichenauer, M. Mohseni, H. Neven, P. Spentzouris, D. Strain, and G. N. Perdue, “Machine learning of high dimensional data on a noisy quantum processor,” npj Quantum Information, vol. 7, no. 1, p. 161, 2021

  6. [6]

    Quantum machine learning,

    J. Biamonte, P. Wittek, N. Pancotti, P. Rebentrost, N. Wiebe, and S. Lloyd, “Quantum machine learning,” Nature, vol. 549, no. 7671, pp. 195–202, 2017

  7. [7]

    Shedding light on the future: Exploring quantum neural networks through optics,

    S. Yu, Z. Jia, A. Zhang, E. Mer, Z. Li, V . Crescimanna, K.-C. Chen, R. B. Patel, I. A. Walmsley, and D. Kaszlikowski, “Shedding light on the future: Exploring quantum neural networks through optics,” Advanced Quantum Technologies, p. 2400074, 2024

  8. [8]

    Quantum machine learning in high energy physics,

    W. Guan, G. Perdue, A. Pesah, M. Schuld, K. Terashi, S. Vallecorsa, and J.-R. Vlimant, “Quantum machine learning in high energy physics,” Machine Learning: Science and Technology , vol. 2, no. 1, p. 011003, 2021

Show all 41 references
  1. [9]

    Quantum computing for high-energy physics: state of the art and challenges,

    A. Di Meglio, K. Jansen, I. Tavernelli, C. Alexandrou, S. Arunachalam, C. W. Bauer, K. Borras, S. Carrazza, A. Crippa, V . Croft, et al. , “Quantum computing for high-energy physics: state of the art and challenges,” PRX Quantum, vol. 5, no. 3, p. 037001, 2024

  2. [10]

    Application of quantum machine learning using the quantum kernel algorithm on high energy physics analysis at the lhc,

    S. L. Wu, S. Sun, W. Guan, C. Zhou, J. Chan, C. L. Cheng, T. Pham, Y . Qian, A. Z. Wang, R. Zhang,et al., “Application of quantum machine learning using the quantum kernel algorithm on high energy physics analysis at the lhc,” Physical Review Research, vol. 3, no. 3, p. 033221, 2021

  3. [11]

    Quantum computing at the frontiers of biological sciences,

    P. S. Emani, J. Warrell, A. Anticevic, S. Bekiranov, M. Gandal, M. J. McConnell, G. Sapiro, A. Aspuru-Guzik, J. T. Baker, M. Bastiani, et al., “Quantum computing at the frontiers of biological sciences,” Nature Methods, vol. 18, no. 7, pp. 701–709, 2021

  4. [12]

    Compressedmediq: Hybrid quantum machine learning pipeline for high-dimentional neu- roimaging data,

    K.-C. Chen, Y .-T. Li, T.-Y . Li, and C.-Y . Liu, “Compressedmediq: Hybrid quantum machine learning pipeline for high-dimentional neu- roimaging data,” arXiv preprint arXiv:2409.08584 , 2024

  5. [13]

    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

  6. [14]

    Variational quantum circuits for deep reinforcement learning,

    S. Y .-C. Chen, C.-H. H. Yang, J. Qi, P.-Y . Chen, X. Ma, and H.-S. Goan, “Variational quantum circuits for deep reinforcement learning,” IEEE access, vol. 8, pp. 141007–141024, 2020

  7. [15]

    Decentralizing feature extraction with quantum convolutional neural network for automatic speech recognition,

    C.-H. H. Yang, J. Qi, S. Y .-C. Chen, P.-Y . Chen, S. M. Siniscalchi, X. Ma, and C.-H. Lee, “Decentralizing feature extraction with quantum convolutional neural network for automatic speech recognition,” in ICASSP 2021-2021 IEEE International Conference on Acoustics, Speech an...

  8. [16]

    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,” arXiv preprint arXiv:2407.08617, 2024

  9. [17]

    Quantum computing for climate resilience and sustainability challenges,

    K. T. M. Ho, K.-C. Chen, L. Lee, F. Burt, S. Yu, et al. , “Quantum computing for climate resilience and sustainability challenges,” arXiv preprint arXiv:2407.16296, 2024

  10. [18]

    Quantum machine learning in climate change and sustainability: A short review,

    A. Nammouchi, A. Kassler, and A. Theocharis, “Quantum machine learning in climate change and sustainability: A short review,” Quantum, vol. 1, p. 1, 2023

  11. [19]

    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

  12. [20]

    The dawn of quantum natural language processing,

    R. Di Sipio, J.-H. Huang, S. Y .-C. Chen, S. Mangini, and M. Worring, “The dawn of quantum natural language processing,” in ICASSP 2022- 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 8612–8616, IEEE, 2022

  13. [21]

    Quantum computing for finance: Overview and prospects,

    R. Or ´us, S. Mugel, and E. Lizaso, “Quantum computing for finance: Overview and prospects,” Reviews in Physics , vol. 4, p. 100028, 2019

  14. [22]

    Linear- layer-enhanced quantum long short-term memory for carbon price forecasting,

    Y . Cao, X. Zhou, X. Fei, H. Zhao, W. Liu, and J. Zhao, “Linear- layer-enhanced quantum long short-term memory for carbon price forecasting,” Quantum Machine Intelligence , vol. 5, no. 2, p. 26, 2023

  15. [23]

    Quantum long short-term memory,

    S. Y .-C. Chen, S. Yoo, and Y .-L. L. Fang, “Quantum long short-term memory,” in ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pp. 8622–8626, IEEE, 2022

  16. [24]

    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,” arXiv preprint arXiv:2407.06103 , 2024

  17. [25]

    Quantum computing in the nisq era and beyond,

    J. Preskill, “Quantum computing in the nisq era and beyond,” Quantum, vol. 2, p. 79, 2018

  18. [26]

    Quantum classifier with tailored quantum kernel,

    C. Blank, D. K. Park, J.-K. K. Rhee, and F. Petruccione, “Quantum classifier with tailored quantum kernel,” npj Quantum Information , vol. 6, no. 1, p. 41, 2020

  19. [27]

    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

  20. [28]

    Experimental realization of a quantum support vector machine,

    Z. Li, X. Liu, N. Xu, and J. Du, “Experimental realization of a quantum support vector machine,” Physical review letters , vol. 114, no. 14, p. 140504, 2015

  21. [29]

    Variational quantum classifier for binary classification: Real vs synthetic dataset,

    D. Maheshwari, D. Sierra-Sosa, and B. Garcia-Zapirain, “Variational quantum classifier for binary classification: Real vs synthetic dataset,” IEEE access, vol. 10, pp. 3705–3715, 2021

  22. [30]

    The complexity of quantum support vector machines,

    G. Gentinetta, A. Thomsen, D. Sutter, and S. Woerner, “The complexity of quantum support vector machines,” Quantum, vol. 8, p. 1225, 2024

  23. [31]

    Hybrid quantum edge computing network,

    L. Ma and L. Ding, “Hybrid quantum edge computing network,” in Quantum Communications and Quantum Imaging XX , vol. 12238, pp. 83–93, SPIE, 2022

  24. [32]

    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,” arXiv preprint arXiv:2409.05770, 2024

  25. [33]

    cutn-qsvm: cutensornet-accelerated quantum support vector machine with cuquantum sdk,

    K.-C. Chen, T.-Y . Li, Y .-Y . Wang, S. See, C.-C. Wang, R. Willie, N.-Y . Chen, A.-C. Yang, and C.-Y . Lin, “cutn-qsvm: cutensornet-accelerated quantum support vector machine with cuquantum sdk,” arXiv preprint arXiv:2405.02630, 2024

  26. [34]

    Generalised circuit partitioning for distributed quantum computing,

    F. Burt, K.-C. Chen, and K. Leung, “Generalised circuit partitioning for distributed quantum computing,” arXiv preprint arXiv:2408.01424 , 2024

  27. [35]

    A machine learning approach to pos tagging,

    L. Marquez, L. Padro, and H. Rodriguez, “A machine learning approach to pos tagging,” Machine Learning, vol. 39, pp. 59–91, 2000

  28. [36]

    Is part-of-speech tagging a solved task? an evaluation of pos taggers for the german web as corpus,

    E. Giesbrecht and S. Evert, “Is part-of-speech tagging a solved task? an evaluation of pos taggers for the german web as corpus,” in Proceedings of the fifth Web as Corpus workshop , pp. 27–35, Citeseer, 2009

  29. [37]

    Pos tagging approaches: A comparison,

    D. Kumawat and V . Jain, “Pos tagging approaches: A comparison,” International Journal of Computer Applications , vol. 118, no. 6, 2015

  30. [38]

    Hidden markov models,

    S. R. Eddy, “Hidden markov models,” Current opinion in structural biology, vol. 6, no. 3, pp. 361–365, 1996

  31. [39]

    An introduction to conditional random fields,

    C. Sutton, A. McCallum, et al., “An introduction to conditional random fields,” Foundations and Trends® in Machine Learning , vol. 4, no. 4, pp. 267–373, 2012

  32. [40]

    Pennylane: Automatic differentiation of hybrid quantum-classical com- putations. arxiv 2018,

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

  33. [41]

    Pytorch: An imperative style, high-performance deep learning library,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, et al. , “Pytorch: An imperative style, high-performance deep learning library,” Advances in neural information processing systems , vol. 32, 2019

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.