Pith. sign in

REVIEW 4 major objections 4 minor 2 cited by

Quantum-Enhanced Parameter-Efficient Learning for Typhoon Trajectory Forecasting

T0 review · 4 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A hybrid quantum-classical method, Quantum Parameter Adaptation, is claimed to cut a typhoon forecasting model's trainable parameters from 8.39 million to about 0.3 million while keeping forecast error comparable to the full model.

desk verdict A useful application demo of a known QPA method, but the '96% parameter reduction' counts training-time variables, not deployed model size, so the compression comparison is misleading. read the letter →

arxiv 2505.09395 v1 pith:57I66KVU submitted 2025-05-14 quant-ph cs.AIcs.LG

Chen-Yu Liu , Kuan-Cheng Chen , Yi-Chien Chen , Samuel Yen-Chi Chen , Wei-Hao Huang , Wei-Jia Huang , Yen-Jui Chang This is my paper · ORCID
classification quant-phcs.AIcs.LG
keywords QuantumNeuralNetworksModelCompressionMachineLearningQuantum-TrainParameterAdaptationTyphoonTrajectoryForecastingParameter-EfficientFine-TuningClimateModeling
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

Quantum Parameter Adaptation (QPA), a hybrid quantum-classical method introduced in this paper, is claimed to compress an 8.39-million-parameter typhoon trajectory model to roughly 0.3 million trainable parameters while keeping forecast error comparable to the full model. The method trains a small parameterized quantum circuit and a lightweight neural-network mapping that together generate the weights of a LoRA-style parameter-efficient fine-tuning update; at inference, the reconstructed model is purely classical, so no quantum hardware is required. The paper tests this on typhoon data from 2015 to 2018 and compares with pruning and weight sharing, reporting that QPA reaches competitive accuracy at a far smaller parameter count. If these results hold, high-performance typhoon forecasting would become substantially cheaper and more energy-efficient, and the same recipe could be applied to other large climate and machine-learning models.

What carries the argument

QPA is the central mechanism: the target network's trainable parameters are never optimized directly. Instead, a parameterized quantum circuit on N = ceil(log2(n_ch)) qubits prepares a state |psi($\theta$)>; the probabilities |<phi_i|psi>|^2 of computational basis states are paired with the binary representation of the basis index and passed through a small neural-network mapping that outputs parameter values, or, with chunk size n_mlp, a batch of values. The key identity is N = ceil(log2(ceil(m/n_mlp))), which replaces a model with m trainable parameters by a polylogarithmic quantum parameter count; for the LoRA-style adaptation here, m = r(d+k) is the number of entries in the two low-rank matrices. During training, the loss gradient is routed back through the Jacobian da/d($\theta$,b) into the circuit angles and the mapping weights, and the final classical network is instantiated from the generated vector a. This is what makes the compression hold and what makes inference quantum-free.

What would settle it

Retrain the same QPA configuration many times with different random initializations and record the spread of total average trajectory errors on the 2015-2018 test set. If the error distribution is wide enough that some runs clearly miss the full model's accuracy, or if the reduction to 0.3 million parameters is not stable across random seeds, the claim of comparable, reliable performance would be refuted.

Watch

Extended reading notes

Core claim

The central claim is that QPA works for typhoon trajectory forecasting: using a parameterized quantum circuit plus a small mapping network to generate the trainable parameters of an Attention-based Multi-ConvGRU model, the paper reduces trainable parameters from 8,399,540 in the full baseline to between about 0.2 million and 0.3 million, a reduction of more than 96 percent, while keeping total average trajectory error on the 2015-2018 test set comparable to the full model. The generated parameters represent the LoRA-style low-rank matrices applied to most layers, together with the last two linear layers of the model. The quantum circuit provides measurement probabilities of its basis states; a lightweight neural-network mapping converts each probability, together with the binary label of the basis state, into either individual weight values or, with batching, blocks of n_mlp values. Only the circuit angles and mapping weights are optimized, and the resulting classical model is deployed without quantum hardware. The paper also reports that, in the accuracy-versus-parameter trade-off, QPA sits favorably against pruning and weight sharing, and that some individual typhoon trajectories predicted by the 0.3M-parameter model match or beat the full model.

Load-bearing premise

The load-bearing premise is that the small quantum-circuit-plus-neural-network generator is expressive enough to produce the needed weight configurations—the low-rank update matrices and the last two linear layers—so that optimizing only those generator parameters recovers the full model's forecasting accuracy; the paper does not prove this expressivity and tests it only on a small number of runs without error bars.

Editorial extensions

If this is right

  • An 8.39-million-parameter typhoon forecasting model can be trained with roughly 0.3 million trainable parameters, a more than 96 percent reduction, with comparable forecast error on the 2015-2018 test set.
  • QPA occupies a better accuracy-versus-parameter operating point than pruning and weight sharing on the same test data.
  • The deployed model needs no quantum hardware at inference, so any energy savings from parameter compression carry through to real-time forecasting.
  • Because QPA compresses the low-rank adaptation matrices rather than full weights, the scheme is claimed to extend to other parameter-efficient fine-tuning tasks, not only typhoon forecasting.
  • The result is the first demonstration of quantum machine learning applied to large-scale typhoon trajectory prediction, a task that previously seemed too computationally heavy for hybrid quantum-classical training.

Reading between the lines

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

  • A natural extension not tested in the paper is transferability: if the quantum-generated parameter generator is trained on typhoon seasons from one basin, the same circuit-parameter mapping might adapt to a new basin with only a short retraining run; the paper's setup would make this straightforward to test.
  • The comparison to pruning and weight sharing would be stronger if the inference-time architecture were held fixed and only the training procedure varied; as reported, the baselines may differ in more than the compression method.
  • The experiments use simulated quantum circuits, so real-hardware noise, sampling overhead, and error-mitigation costs are not yet accounted for; the advertised efficiency gain could shrink on actual quantum processors.
  • If the observed roughly 0.2-0.3M parameter sweet spot is a general feature of QPA, then the method's quantum cost would grow only logarithmically with model size, making it a candidate for billion-parameter weather and climate models as hardware matures.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper applies Quantum Parameter Adaptation (QPA), a quantum-train-based parameter-efficient learning method, to the Attention-based Multi-ConvGRU typhoon trajectory forecasting model. QPA generates classical network weights (or PEFT parameters) from a parameterized quantum circuit and a small MLP mapping, so that only the circuit angles and mapping weights are optimized. The authors report reducing trainable parameters from 8.39 million to approximately 0.2-0.3 million while maintaining forecasting accuracy comparable to the full model on the 2015-2018 test set, and they claim advantages over classical pruning and weight sharing. The manuscript also includes hyperparameter studies and example trajectory visualizations.

Significance. If the empirical claims are verified, this would be a useful first demonstration of quantum-train-style parameter-efficient learning for a large-scale typhoon forecasting model, and the qubit-count reduction arithmetic in Eqs. (8)-(10) is sound. The work builds on an established QT/QPA line and applies it to a concrete climate application, which is a positive feature. However, the central quantitative claims are not yet supported by the evidence presented: there is no numerical accuracy table, no error bars or repeated-seed statistics, and the parameter-count comparison conflates training-time optimizer variables with deployed model size. The paper is a promising exploratory contribution, but the headline claims require substantial revision before the results can be assessed.

major comments (4)
  1. [§IV.A.1, Fig. 5, Eqs. (2)-(3)] The x-axis of Fig. 5 is labeled as the number of trainable parameters, but for QPA this count (0.2M-0.3M) refers only to the optimizer variables (θ, b), whereas the deployed model must materialize the generated 8.39M AM-ConvGRU weights plus any LoRA factors in order to make predictions. Because the generated parameter vector a is a deterministic function of (θ, b) in Eqs. (2)-(3), QPA reduces the dimension of the optimization problem, not the stored model size or inference cost. The comparison with pruning and weight sharing is therefore not apples-to-apples, and the conclusion in §V that QPA maintains a significantly smaller parameter footprint and outperforms classical compression is unsupported as stated. Please state explicitly what is counted on the x-axis and, if the deployed-footprint claim is intended, provide a separate comparison of stored model size and inference cost.
  2. [§IV.A, Figs. 5-6] The central accuracy claim is supported only by figures. There is no numerical table of total average error, no standard deviation over random initializations, and no statistical comparison between QPA, the full AM-ConvGRU model [55], pruning, and weight sharing. The text states that QPA consistently achieves competitive performance and that some configurations outperform the full model, but none of this can be checked from the figures; the two outperformance examples in Figs. 7-8 and the two underperformance examples in Figs. 9-10 are anecdotal. Please add a table with mean and standard deviation over at least 3-5 random initializations for the relevant settings, and define the reported metric (total average error over which forecast horizons) precisely.
  3. [§IV.A.1] The experimental setup is ambiguous: 'QPA is applied to the last two linear layers, while LoRA is applied to the remaining layers.' It is not specified which parameters of AM-ConvGRU are frozen, how the LoRA factors are initialized and whether they are also generated by the quantum mapping, or how the 0.2M-0.3M parameter count is computed. Since the mapping model in Table I has hidden sizes [32, 32, nmlp] with N ≤ 14 and nmlp ≤ 768, the parameter count |θ| + |b| is on the order of tens of thousands, not 0.2-0.3M; this discrepancy suggests that the reported number includes something else or is counted differently. Please define the reported number precisely and describe the exact training setup.
  4. [§II.A, Eqs. (5)-(6), §IV.2] The expressivity of the mapping G_b is assumed rather than demonstrated; no formal guarantee or controlled study shows that a small MLP applied to basis-state probabilities can represent the weight configurations of AM-ConvGRU needed for accurate typhoon prediction. The paper itself shows cases (Figs. 9-10) where QPA does not surpass the full model, so this representation error is empirically visible. A sensitivity analysis over random seeds, together with a discussion of when the mapping fails, would strengthen the claim that QPA is a generally viable approach for this task.
minor comments (4)
  1. [Abstract] The abstract contains a sentence fragment: 'Quantum-Train (QT), a hybrid quantum-classical framework that leverages quantum neural networks (QNNs) to generate trainable parameters exclusively during training, eliminating the need for quantum hardware at inference time.' This should be rewritten as a complete sentence.
  2. [§II.A, Eq. (2)] The notation ∂a/∂(θ,b) is informal; please define the Jacobian explicitly, including the dimensions of the matrix and how it is computed in the batched setting.
  3. [§IV.A, Figs. 5-6] The metric 'total average error' is used throughout the results but is never defined in the text. Please define it explicitly, for example as the mean Great Circle Distance over all test typhoons and forecast horizons, and state which horizons are included.
  4. [References] References [39] and [40] cite the same arXiv preprint and should be consolidated.

Circularity Check

1 steps flagged · score 4.0 of 10

Parameter-count reduction is definitional (only θ,b are trained), so the 96% figure is a consequence of QPA's counting convention; forecasting-accuracy comparison against the external AM-ConvGRU baseline is independent and non-circular.

  1. self definitional [Section IV.A.1 (Overall Benchmarking), Fig. 5; Eqs. (2)-(6)]
    "The classical full model consists of 8,399,540 (8.39M) trainable parameters. In contrast, when QPA is used with different configurations, as shown in Fig. 5, the number of trainable parameters is reduced to approximately 0.2M to 0.3M. Since QPA is fundamentally a parameter-efficient learning method that compresses the number of trainable parameters, it is crucial to compare it with other classical compression techniques, such as pruning and weight sharing."

    In Eqs. (2)-(3), the only variables updated during training are the PQC angles θ and the mapping-model weights b; the target-network parameter vector a is generated deterministically via Eqs. (5)-(6). Therefore the reported drop from 8.39M to 0.2-0.3M is the definitional count of optimizer variables, not an independently measured property of the trained model. The quoted passage even acknowledges this by calling QPA 'fundamentally ... parameter-efficient.' Plotting this optimizer-variable count on the same x-axis as pruning and weight sharing in Fig. 5 presents a chosen counting convention as an empirical compression result; at inference the generated a still supplies the full AM-ConvGRU weights. The 96% reduction thus follows from the method's definition, not from a derived finding.

full rationale

The only concrete circularity I can substantiate with paper text is the definitional parameter-count framing above. The accuracy side of the central claim is independently testable: QPA's forecasts are compared with the published classical AM-ConvGRU model [55] on the external CMA/ERA-Interim data, and the paper reports both cases where QPA does not beat the full model (Figs. 9-10), which is inconsistent with a result forced by construction. The method itself is reproduced in Section II rather than merely assumed, so the heavy self-citation of QT/QPA papers ([35], [36], [44], [54]) does not carry the derivation. No uniqueness theorem or forbidden-alternative argument is imported from the authors' prior work. The lack of numeric error tables, confidence intervals, and repeated-seed statistics is a reproducibility concern, not circularity. Because the parameter-reduction claim reduces by definition while the performance-comparison claim has independent external content, the appropriate score is moderate rather than maximal: the derivation is not entirely self-contained, but its scientific core is not circular.

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

The central claims rest on the previously proposed QPA method (self-authored), the AM-ConvGRU baseline, and the CMA and ERA-Interim datasets. No new free parameters beyond QPA hyperparameters are introduced, but the omitted training details and the expressivity assumption carry the load. No invented physical entities are introduced.

free parameters (4)
  • Chunk size nmlp = 32, 64, 128, 256, 512, 768 (scanned)
    Chunk size determines the number of qubits via Eq. (7) and directly sets the trainable parameter count. No validation protocol is reported for selecting it.
  • QNN depth L = 4, 8, 12, 16, 20 (scanned)
    The number of layers in the parameterized quantum circuit is a hyperparameter varied in Fig. 6; performance depends on it.
  • LoRA rank r = not reported (example uses r=4)
    The illustrative example in Eq. (10) uses rank 4, but the experiments do not state the rank used for the LoRA layers; the parameter counts and performance depend on this choice.
  • Training hyperparameters = not reported
    Learning rate, batch size, number of epochs, and optimizer settings for the QPA hybrid training are not disclosed, yet they affect convergence and final accuracy.
assumptions (4)
  • standard math Chain rule and parameter-shift rule for gradient computation (Eq. 2, refs [51], [52])
    The gradient update formula in Eq. (2) relies on standard differentiable programming through the quantum-classical mapping.
  • domain assumption CMA typhoon records and ERA-Interim reanalysis are accurate and consistently preprocessed via CLIPER and RCAB (Section III)
    The empirical results depend on the correctness of the historical typhoon data and the preprocessing pipeline described in the data section.
  • domain assumption AM-ConvGRU from Xu et al. [55] is a valid strong baseline, and its reported error levels transfer to this comparison (Section II-D)
    The paper uses AM-ConvGRU as the classical target model and baseline; if this baseline is not properly reproduced or compared, the QPA advantage claim loses meaning.
  • ad hoc to paper A PQC with parameters theta plus a small MLP mapping G_b can express the weight configurations of AM-ConvGRU needed for accurate typhoon prediction (Eqs. 5 and 6)
    The central mechanism of QPA assumes that low-dimensional quantum-generated parameters can represent the needed weights. The paper provides no expressivity analysis or formal guarantee, only empirical curves.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Quantum-Enhanced Parameter-Efficient Learning for Typhoon Trajectory Forecasting." pith.science (2026). https://pith.science/paper/57I66KVU

@misc{pith2026250509395,
  author       = {Pith},
  title        = {Pith review of: Quantum-Enhanced Parameter-Efficient Learning for Typhoon Trajectory Forecasting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/57I66KVU}},
  note         = {Machine review of arXiv:2505.09395}
}
read the original abstract

Typhoon trajectory forecasting is essential for disaster preparedness but remains computationally demanding due to the complexity of atmospheric dynamics and the resource requirements of deep learning models. Quantum-Train (QT), a hybrid quantum-classical framework that leverages quantum neural networks (QNNs) to generate trainable parameters exclusively during training, eliminating the need for quantum hardware at inference time. Building on QT's success across multiple domains, including image classification, reinforcement learning, flood prediction, and large language model (LLM) fine-tuning, we introduce Quantum Parameter Adaptation (QPA) for efficient typhoon forecasting model learning. Integrated with an Attention-based Multi-ConvGRU model, QPA enables parameter-efficient training while maintaining predictive accuracy. This work represents the first application of quantum machine learning (QML) to large-scale typhoon trajectory prediction, offering a scalable and energy-efficient approach to climate modeling. Our results demonstrate that QPA significantly reduces the number of trainable parameters while preserving performance, making high-performance forecasting more accessible and sustainable through hybrid quantum-classical learning.

Figures

Figures reproduced from arXiv: 2505.09395 by the authors.

Figure 1
Figure 1. Overview of (a) Quantum Machine Learning [ [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Overview of Quantum Parameter Adaptation [ [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Overview of our solution strategy in this work. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Visualization of typhoon trajectories from the CMA dataset. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Typhoon trajectory forecasting on testing dataset (2015-2018). [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Typhoon trajectory forecasting on testing dataset (2015-2018). (a) Fixing # of QNN layers [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 8
Figure 8. Figure 8: Typhoon trajectory forecasting results for typhoon Chan-hom, # [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Typhoon trajectory forecasting results for typhoon MANGKHUT, # [PITH_FULL_IMAGE:figures/full_fig_p009_9.png]
Figure 10
Figure 10. Figure 10: Typhoon trajectory forecasting results for typhoon JONGDARI, # [PITH_FULL_IMAGE:figures/full_fig_p009_10.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Quantum Adaptive Excitation Network with Variational Quantum Circuits for Channel Attention

    quant-ph 2025-07 reject novelty 5.0 of 10

    A hybrid CNN that uses a small trainable quantum circuit for channel attention claims large accuracy gains, but the evidence is statistically thin.

  2. Special-Unitary Parameterization for Trainable Variational Quantum Circuits

    quant-ph 2025-07 reject novelty 4.0 of 10

    SUN-VQC claims to avoid barren plateaus by using SU(4) exponential blocks, but the dynamical-Lie-algebra argument is invalid for the brick-wall circuit in the experiments.

Reference graph

Works this paper leans on

59 extracted references · 24 canonical work pages · cited by 2 Pith papers

  1. [55]

    Am- convgru: a spatio-temporal model for typhoon path prediction,

    G. Xu, D. Xian, P. Fournier-Viger, X. Li, Y . Ye, and X. Hu, “Am- convgru: a spatio-temporal model for typhoon path prediction,” Neural Computing and Applications , vol. 34, no. 8, pp. 5905–5921, 2022

    work page 2022

  2. [1]

    Increasing destructiveness of tropical cyclones over the past 30 years,

    K. Emanuel, “Increasing destructiveness of tropical cyclones over the past 30 years,” Nature, vol. 436, no. 7051, pp. 686–688, 2005

    work page 2005

  3. [2]

    Changes in tropical cyclone number, duration, and intensity in a warming envi- ronment,

    P. J. Webster, G. J. Holland, J. A. Curry, and H.-R. Chang, “Changes in tropical cyclone number, duration, and intensity in a warming envi- ronment,” Science, vol. 309, no. 5742, pp. 1844–1846, 2005

    work page 2005

  4. [3]

    Masson-Delmotte, P

    V . Masson-Delmotte, P. Zhai, S. Pirani, C. Connors, S. P ´ean, N. Berger, Y . Caud, L. Chen, M. Goldfarb, and P. M. Scheel Monteiro, “Ipcc, 2021: Summary for policymakers. in: Climate change 2021: The physical science basis. contribution of working group i to the sixth assessment report of the intergovernmental panel on climate change,” 2021

    work page 2021

  5. [4]

    Are the rich less prone to flooding? a case study on flooding in the southern taiwan during typhoon morakot and typhoon fanapi,

    Y .-L. Kuo, Y .-M. Liu, H.-J. Chu, and H.-C. Lee, “Are the rich less prone to flooding? a case study on flooding in the southern taiwan during typhoon morakot and typhoon fanapi,” Natural Hazards and Earth System Sciences Discussions , vol. 2022, pp. 1–19, 2022

    work page 2022

  6. [5]

    The quiet revolution of numerical weather prediction,

    P. Bauer, A. Thorpe, and G. Brunet, “The quiet revolution of numerical weather prediction,” Nature, vol. 525, no. 7567, pp. 47–55, 2015

    2015

  7. [6]

    Quantum algorithms: an overview,

    A. Montanaro, “Quantum algorithms: an overview,” npj Quantum Infor- mation, vol. 2, no. 1, pp. 1–8, 2016

    2016

  8. [7]

    The future of quantum computing with superconducting qubits,

    S. Bravyi, O. Dial, J. M. Gambetta, D. Gil, and Z. Nazario, “The future of quantum computing with superconducting qubits,” Journal of Applied Physics, vol. 132, no. 16, 2022

    2022

Show all 59 references
  1. [8]

    Quantum-centric supercomputing: The next wave of computing,

    J. Gambetta, “Quantum-centric supercomputing: The next wave of computing,” IBM Research Blog , 2022

    2022

  2. [9]

    Noise- aware distributed quantum approximate optimization algorithm on near- term quantum hardware,

    K.-C. Chen, X. Xu, F. Burt, C.-Y . Liu, S. Yu, and K. K. Leung, “Noise- aware distributed quantum approximate optimization algorithm on near- term quantum hardware,” in 2024 IEEE International Conference on Quantum Computing and Engineering (QCE) , vol. 2. IEEE, 2024, pp. 144–149

    2024

  3. [10]

    Hybrid gate-based and annealing quantum computing for large-size ising problems,

    C.-Y . Liu and H.-S. Goan, “Hybrid gate-based and annealing quantum computing for large-size ising problems,” Bulletin of the American Physical Society, vol. 68, 2023

    2023

  4. [11]

    Partitioning optimization problems for hybrid classical,

    M. Booth, S. P. Reinhardt, and A. Roy, “Partitioning optimization problems for hybrid classical,” quantum execution. Technical Report , pp. 01–09, 2017

    2017

  5. [12]

    Portfolio optimisation using the d- wave quantum annealer,

    F. Phillipson and H. S. Bhatia, “Portfolio optimisation using the d- wave quantum annealer,” in International Conference on Computational Science. Springer, 2021, pp. 45–59

    2021

  6. [13]

    Quantum local search for traveling salesman problem with path-slicing strategy,

    C.-Y . Liu, H. Matsuyama, W.-h. Huang, and Y . Yamashiro, “Quantum local search for traveling salesman problem with path-slicing strategy,” arXiv preprint arXiv:2407.13616 , 2024

    2024 arXiv

  7. [14]

    Hybrid quantum annealing for larger-than-qpu lattice-structured problems,

    J. Raymond, R. Stevanovic, W. Bernoudy, K. Boothby, C. C. McGeoch, A. J. Berkley, P. Farr ´e, J. Pasvolsky, and A. D. King, “Hybrid quantum annealing for larger-than-qpu lattice-structured problems,” ACM Trans- actions on Quantum Computing , vol. 4, no. 3, pp. 1–30, 2023

    2023

  8. [15]

    Imple- mentation of trained factorization machine recommendation system on quantum annealer,

    C.-Y . Liu, H.-Y . Wang, P.-Y . Liao, C.-J. Lai, and M.-H. Hsieh, “Imple- mentation of trained factorization machine recommendation system on quantum annealer,” arXiv preprint arXiv:2210.12953 , 2022

    2022 arXiv

  9. [16]

    Quantum kernel-based long short- term memory,

    Y .-C. Hsu, T.-Y . Li, and K.-C. Chen, “Quantum kernel-based long short- term memory,” arXiv preprint arXiv:2411.13225 , 2024

    2024 arXiv

  10. [17]

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

    Y .-C. Hsu, N.-Y . Chen, T.-Y . Li, K.-C. Chen et al. , “Quantum kernel- based long short-term memory for climate time-series forecasting,”arXiv preprint arXiv:2412.08851, 2024

    2024 arXiv

  11. [18]

    Robust decentralized quantum kernel learning for noisy and adversarial environment,

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

    2025 arXiv

  12. [19]

    Learning to learn with quantum optimization via quantum neural networks,

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

    2025 arXiv

  13. [20]

    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

    2024 arXiv

  14. [21]

    Compressedmediq: Hybrid quantum machine learning pipeline for high-dimensional neuroimaging data,

    K.-C. Chen, Y .-T. Li, T.-Y . Li, C.-Y . Liu, P.-H. Li, and C.-Y . Chen, “Compressedmediq: Hybrid quantum machine learning pipeline for high-dimensional neuroimaging data,” arXiv preprint arXiv:2409.08584, 2024

    2024 arXiv

  15. [22]

    Resource-efficient compilation of distributed quantum circuits for solving large-scale wireless communication network problems,

    K.-C. Chen, F. Burt, S. Yu, C.-Y . Liu, M.-H. Hsieh, and K. K. Le- ung, “Resource-efficient compilation of distributed quantum circuits for solving large-scale wireless communication network problems,” arXiv preprint arXiv:2501.10242, 2025

    2025 arXiv

  16. [23]

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

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

    2024

  17. [24]

    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. 141 007–141 024, 2020

    2020

  18. [25]

    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

  19. [26]

    Effect of data encoding on the expressive power of variational quantum-machine-learning models,

    M. Schuld, R. Sweke, and J. J. Meyer, “Effect of data encoding on the expressive power of variational quantum-machine-learning models,” Physical Review A , vol. 103, no. 3, p. 032430, 2021

    2021

  20. [27]

    Transfer learning in hybrid classical-quantum neural networks,

    A. Mari, T. R. Bromley, J. Izaac, M. Schuld, and N. Killoran, “Transfer learning in hybrid classical-quantum neural networks,” Quantum, vol. 4, p. 340, 2020

    2020

  21. [28]

    Challenges and opportunities in quantum machine learning,

    M. Cerezo, G. Verdon, H.-Y . Huang, L. Cincio, and P. J. Coles, “Challenges and opportunities in quantum machine learning,” Nature Computational Science, vol. 2, no. 9, pp. 567–576, 2022

    2022

  22. [29]

    Quantum advantage in learning from experiments,

    H.-Y . Huang, M. Broughton, J. Cotler, S. Chen, J. Li, M. Mohseni, H. Neven, R. Babbush, R. Kueng, J. Preskill et al., “Quantum advantage in learning from experiments,” Science, vol. 376, no. 6598, pp. 1182– 1186, 2022

    2022

  23. [30]

    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

    2017

  24. [31]

    Generalization in quantum machine learning from few training data,

    M. C. Caro, H.-Y . Huang, M. Cerezo, K. Sharma, A. Sornborger, L. Cincio, and P. J. Coles, “Generalization in quantum machine learning from few training data,” Nature communications, vol. 13, no. 1, pp. 1– 11, 2022

    2022

  25. [32]

    Quantum- classical-quantum workflow in quantum-hpc middleware with gpu ac- celeration,

    K.-C. Chen, X. Li, X. Xu, Y .-Y . Wang, and C.-Y . Liu, “Quantum- classical-quantum workflow in quantum-hpc middleware with gpu ac- celeration,” in 2024 International Conference on Quantum Communica- tions, Networking, and Computing (QCNC) . IEEE, 2024, pp. 304–311

    2024

  26. [33]

    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

    2021

  27. [34]

    Toward large-scale distributed quantum long short-term memory with modular quantum computers,

    K.-C. Chen, S. Y .-C. Chen, C.-Y . Liu, and K. K. Leung, “Toward large-scale distributed quantum long short-term memory with modular quantum computers,” arXiv preprint arXiv:2503.14088 , 2025

    2025 arXiv

  28. [35]

    Quantum-train: Rethinking hybrid quantum- classical machine learning in the model compression perspective,

    C.-Y . Liu, E.-J. Kuo, C.-H. A. Lin, J. G. Young, Y .-J. Chang, M.-H. Hsieh, and H.-S. Goan, “Quantum-train: Rethinking hybrid quantum- classical machine learning in the model compression perspective,” arXiv preprint arXiv:2405.11304, 2024

    2024 arXiv

  29. [36]

    Introduction to quantum-train toolkit,

    C.-Y . Liu, C.-H. A. Lin, W.-J. Huang, and M.-H. Hsieh, “Introduction to quantum-train toolkit,” in 2024 IEEE International Conference on Quantum Computing and Engineering (QCE) , vol. 2. IEEE, 2024, pp. 456–457

    2024

  30. [37]

    Quantum- trained convolutional neural network for deepfake audio detection,

    C.-H. A. Lin, C.-Y . Liu, S. Y .-C. Chen, and K.-C. Chen, “Quantum- trained convolutional neural network for deepfake audio detection,” arXiv preprint arXiv:2410.09250 , 2024

    2024 arXiv

  31. [38]

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

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

    2024

  32. [40]

    Quantum-train-based distributed multi-agent reinforcement learn- ing,

    ——, “Quantum-train-based distributed multi-agent reinforcement learn- ing,” arXiv preprint arXiv:2412.08845 , 2024

    2024 arXiv

  33. [41]

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

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

    2024

  34. [42]

    Federated quantum-train long short-term memory for gravitational wave signal,

    C.-Y . Liu, S. Y .-C. Chen, K.-C. Chen, W.-J. Huang, and Y .-J. Chang, “Federated quantum-train long short-term memory for gravitational wave signal,” arXiv preprint arXiv:2503.16049 , 2025

    2025 arXiv

  35. [43]

    Programming variational quantum circuits with quantum-train agent,

    ——, “Programming variational quantum circuits with quantum-train agent,” arXiv preprint arXiv:2412.01173 , 2024

    2024 arXiv

  36. [44]

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

    C.-Y . Liu, C.-H. H. Yang, H.-S. Goan, and M.-H. Hsieh, “A quantum circuit-based compression perspective for parameter-efficient learning,” in The Thirteenth International Conference on Learning Representa- tions, 2025

    2025

  37. [45]

    Lora: Low-rank adaptation of large language models,

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, and W. Chen, “Lora: Low-rank adaptation of large language models,” arXiv preprint arXiv:2106.09685 , 2021

    2021 arXiv

  38. [46]

    Dora: Weight-decomposed low-rank adapta- tion,

    S.-Y . Liu, C.-Y . Wang, H. Yin, P. Molchanov, Y .-C. F. Wang, K.-T. Cheng, and M.-H. Chen, “Dora: Weight-decomposed low-rank adapta- tion,” arXiv preprint arXiv:2402.09353 , 2024

    2024 arXiv

  39. [47]

    Parameter-efficient transfer learning for nlp,

    N. Houlsby, A. Giurgiu, S. Jastrzebski, B. Morrone, Q. De Laroussilhe, A. Gesmundo, M. Attariyan, and S. Gelly, “Parameter-efficient transfer learning for nlp,” in International conference on machine learning . PMLR, 2019, pp. 2790–2799

    2019

  40. [48]

    Exploring versatile generative language model via parameter-efficient transfer learning,

    Z. Lin, A. Madotto, and P. Fung, “Exploring versatile generative language model via parameter-efficient transfer learning,” arXiv preprint arXiv:2004.03829, 2020

    2004 arXiv

  41. [49]

    Prefix-tuning: Optimizing continuous prompts for generation,

    X. L. Li and P. Liang, “Prefix-tuning: Optimizing continuous prompts for generation,” arXiv preprint arXiv:2101.00190 , 2021

    2021 arXiv

  42. [50]

    V oice2series: Reprogram- ming acoustic models for time series classification,

    C.-H. H. Yang, Y .-Y . Tsai, and P.-Y . Chen, “V oice2series: Reprogram- ming acoustic models for time series classification,” in International conference on machine learning . PMLR, 2021, pp. 11 808–11 819

    2021

  43. [51]

    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

  44. [52]

    Eval- uating analytic gradients on quantum hardware,

    M. Schuld, V . Bergholm, C. Gogolin, J. Izaac, and N. Killoran, “Eval- uating analytic gradients on quantum hardware,” Physical Review A , vol. 99, no. 3, p. 032331, 2019

    2019

  45. [53]

    Choosing hardware for your qsim simulation,

    Q. A. Google, “Choosing hardware for your qsim simulation,” https: //quantumai.google/qsim/choose hw, 2024

    2024

  46. [54]

    Federated quantum-train with batched parameter generation,

    C.-Y . Liu and S. Y .-C. Chen, “Federated quantum-train with batched parameter generation,” in 2024 15th International Conference on Infor- mation and Communication Technology Convergence (ICTC) . IEEE, 2024, pp. 1133–1138

    2024

  47. [56]

    An overview of the china meteorological administration tropi- cal cyclone database,

    M. Ying, W. Zhang, H. Yu, X. Lu, J. Feng, Y . Fan, Y . Zhu, and D. Chen, “An overview of the china meteorological administration tropi- cal cyclone database,” Journal of Atmospheric and Oceanic Technology, vol. 31, no. 2, pp. 287–301, 2014

    2014

  48. [57]

    The ecmwf ensemble prediction system: Methodology and validation,

    F. Molteni, R. Buizza, T. N. Palmer, and T. Petroliagis, “The ecmwf ensemble prediction system: Methodology and validation,” Quarterly journal of the royal meteorological society , vol. 122, no. 529, pp. 73– 119, 1996

    1996

  49. [58]

    An overview of neural network compression,

    J. O. Neill, “An overview of neural network compression,” 2020. [Online]. Available: https://arxiv.org/abs/2006.03669

    2020 arXiv

  50. [59]

    What is the state of neural network pruning?

    D. Blalock, J. J. Gonzalez Ortiz, J. Frankle, and J. Guttag, “What is the state of neural network pruning?” Proceedings of machine learning and systems, vol. 2, pp. 129–146, 2020

    2020

  51. [60]

    Simplifying neural networks by soft weight sharing,

    S. J. Nowlan and G. E. Hinton, “Simplifying neural networks by soft weight sharing,” in The mathematics of generalization . CRC Press, 2018, pp. 373–394

    2018

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.