Pith. sign in

REVIEW 5 major objections 5 minor 56 references

Hybrid Parameterized Quantum States for Variational Quantum Learning

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper introduces Hybrid Parameterized Quantum States (HPQS), a blended family of variational states, and argues that weighting a finite-shot circuit measurement together with a classical neural estimator gives shot-efficient…

desk verdict HPQS is a clean, useful hybrid framework, but the paper overclaims parity with exact PQC and lacks the ablations needed to attribute gains to the quantum branch. read the letter →

arxiv 2505.16676 v1 pith:HWEB7MVG submitted 2025-05-22 quant-ph

Chen-Yu Liu This is my paper · ORCID
classification quant-ph
keywords hybridparameterizedquantumstatesvariationallearningneuralcircuitsfinite-shotmeasurementsquantum-trainparameteradaptationLoRAfine-tuning
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 introduces Hybrid Parameterized Quantum States (HPQS), a variational model written as a weighted sum of two estimates: a parameterized quantum circuit's finite-shot measurement probability and a classical neural estimator of the same distribution. The claimed payoff is that the blend, controlled by a single coefficient $\lambda$, keeps the expressivity of the quantum state while smoothing away the statistical noise that makes circuit-only models unreliable on near-term hardware. Across three tasks — expectation-based quantum classification, Quantum-Train parameter generation, and LoRA adapter generation for LLM fine-tuning — the paper reports that HPQS beats circuit-only models under finite shots and matches or exceeds exact-simulation circuits and pure neural quantum states. If the claim holds, HPQS is a drop-in hybrid recipe for the noisy-intermediate-scale era that remains useful on fault-tolerant hardware where shot budgets are still finite.

What carries the argument

The load-bearing object is the scalar blending identity of Eq. (6), which combines postprocessed quantum and classical estimates into one probability-like value. The quantum branch contributes empirical finite-shot probabilities $\hat p_i^{\mathrm{emp}}$; the classical branch evaluates a neural quantum state $f_\gamma(\phi_i)$; postprocessing maps $G$ and $H$ (identity, MLP, or tensor-network decoders) bring the two branches into a common representation space; and $\lambda \in [0,1]$ sets the mix. The same identity is reused verbatim for the three tasks: expectation-based QML blends quantum expectation values with classical logits, Quantum-Train blends generated network parameters $\tilde a_i^{(Q)}$ and $\tilde a_i^{(C)}$, and QPA blends chunks of LoRA adapter matrices. Because $G$ and $H$ can be tensor-network mappings, the mechanism scales to polylogarithmic trainable-variable counts.

What would settle it

Run all three benchmarks with $\lambda$ fixed to 0 while keeping the same NQS architecture, tensor-network decoders, parameter budget, and random seeds as HPQS; if this pure-classical variant matches or beats HPQS at every shot count, the quantum branch is not contributing to the reported gains.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that the interpolation formula $\hat p_i = \lambda \tilde p_i^{(Q)} + (1-\lambda) \tilde p_i^{(C)}$ defines a new family of variational quantum states whose endpoints are PQC-only ($\lambda=1$) and NQS-only ($\lambda=0$), and whose interior is where finite-shot noise gets absorbed. The neural branch supplies a smooth, differentiable surrogate for the measurement distribution; the quantum branch keeps the model tied to physically realized correlations. Empirically, the paper reports HPQS (finite) outperforming PQC (finite) in all three settings, and matching or beating PQC (exact) in expectation-based QML and in the Gemma-2 QPA case, all at the same fixed shot budgets used for the PQC baselines. In the Quantum-Train MNIST task, HPQS (finite) reaches 86–88% accuracy at $10\times$HSS shots versus 57–66% for PQC (finite) and 90.59% for PQC (exact).

Load-bearing premise

The load-bearing premise is that fixing the shot budget and letting HPQS add a classical neural branch is a fair measure of shot efficiency; if that classical branch alone is responsible for the gains, the quantum component contributes nothing, a possibility the paper's own Limitations paragraph concedes by noting performance depends on the classical estimator.

Editorial extensions

If this is right

  • HPQS lifts expectation-based MNIST classification with $5\times$HSS shots to 88.97% accuracy, where PQC (finite) sits at 46.21%.
  • In Quantum-Train, HPQS generates a 6690-parameter CNN with 649 trainable variables and reaches 86–88% MNIST-10 accuracy at $10\times$HSS shots, close to the 90.59% of exact simulation.
  • For GPT-2 LoRA fine-tuning with $1\times$HSS shots, HPQS lowers perplexity to 4.612 versus 6.975 for PQC (finite), while on Gemma-2 it matches exact-simulation perplexity (1.467 vs 1.427).
  • Under IBM noise models, HPQS degrades less than PQC (finite) in all three tasks, so the hybrid's advantage persists under hardware noise.
  • Because $\lambda=1$ and $\lambda=0$ recover PQC-only and NQS-only models as endpoints, HPQS provides a single architecture that interpolates from fully quantum to fully classical variational learning.

Reading between the lines

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

  • Editorial inference: A $\lambda=0$ control with the same tensor-network decoders, NQS architecture, and parameter budget is the natural next experiment; without it, the tables cannot separate the blend's benefit from the extra classical capacity.
  • Editorial inference: HPQS can be viewed as a learned denoiser for empirical quantum distributions, so combining it with classical shadows or error mitigation is a plausible untested extension rather than a competing alternative.
  • Editorial inference: On GPT-2, HPQS's 4.612 perplexity remains far from PQC-exact's 1.616, suggesting the hybrid's strength is concentrated in ultra-low-shot regimes; a testable prediction is that HPQS's perplexity approaches PQC-exact monotonically as the shot budget grows.
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

5 major / 5 minor

Summary. The paper proposes HPQS, a variational model that produces probability estimates (or parameter values) as a convex combination of a finite-shot PQC branch and an NQS classical branch, with optional task-specific postprocessors G and H. It claims shot efficiency and noise resilience, and demonstrates the framework on expectation-based QML classification (MNIST 3-vs-6), Quantum-Train full parameter generation for a CNN on MNIST-10, and QPA LoRA fine-tuning of GPT-2/Gemma-2 on WikiText-2. The paper also states a universality result (Eq. 7) and an appendix showing that several existing quantum-classical learning methods are special cases.

Significance. The hybrid framing is practically motivated and the three tasks are good stress tests for low-shot variational learning. The paper has concrete strengths: it compares HPQS against PQC exact, PQC finite, and NQS baselines, uses standard datasets, fixes quantum shot budgets, includes IBM noise-model simulations, and states that code is attached in the supplement. However, the central attribution claim is not yet supported: the experiments do not isolate the quantum branch's contribution, the 'shot-efficient' framing excludes classical computational costs, and one headline QPA result (GPT-2) directly contradicts the conclusion's 'comparable or better than PQC (exact)' statement. The formal universality result is inherited from the NQS limit rather than derived for the hybrid mechanism. Significance is therefore conditional on additional controlled experiments and fair resource accounting.

major comments (5)
  1. [Section 5 / Table 3] The concluding claim that HPQS 'achieves comparable or better performance than PQC (exact)' is contradicted by the GPT-2 QPA row, where HPQS (finite) has perplexity 4.612 versus 1.616 for PQC (exact), and only approximately matched in QT (Table 1: 87.64% versus 90.59%). Because the GPT-2 result is one of the three core demonstrations, the paper should either weaken the conclusion substantially or provide an explanation and a proper statistical comparison for why this case is considered comparable.
  2. [Section 4.1 / Tables 1-3 / Appendix B] No ablation isolates the quantum branch's contribution. In QML the blending coefficient is lambda=0.1 (Appendix B.2), so 90% of the final prediction comes from the classical branch; HPQS also has more trainable parameters than the NQS baseline (65 vs 21 in Table 2; 792 vs 512 in Table 1). The paper does not hold the classical architecture and parameter count fixed while removing or randomizing the quantum input, and it does not include a standard classical LoRA fine-tuning baseline for QPA. Without such controls, the reported improvements over NQS could be due entirely to a larger or better classical model, and the quantum component could merely be a noisy regularizer.
  3. [Section 4.1 / Appendix A] The shot-efficiency claim counts only quantum measurement shots. HPQS adds an NQS branch and tensor-network postprocessors whose training and inference costs are not included in any resource comparison, so 'shot-efficient' is not a total-resource statement. The paper should either define the resource metric as quantum shots only and justify that choice, or report wall-clock/training cost comparisons so readers can judge the total overhead.
  4. [Section 4.1 / Table 3 / Appendix A] QPA results are reported as the best over three seeds with no mean or standard deviation, and on Gemma-2 the HPQS perplexity (1.467) differs from NQS (1.471) and PQC finite (1.472) by margins that are almost certainly within seed variance. Without repeated-seed statistics or significance testing, the QPA advantage is not established. In addition, Table 4 shows HPQS QML accuracy under ibm_fez (93.33%) higher than the noise-free setting (88.97%), which is unexplained and weakens the noise-resilience narrative.
  5. [Section 3.1 / Eq. (7)] The universality statement in Eq. (7) is a direct corollary of the NQS universal approximation property at lambda=0; it does not establish that the blended estimator with finite-shot quantum inputs is universal, nor does it provide any guarantee for lambda>0. The paper should either explicitly present this as an inherited property of the classical limit, or prove a nontrivial expressivity statement for the hybrid regime with finite shots and noise.
minor comments (5)
  1. [Appendix A / Table 4] The ibm_fez row for HPQS QML exceeds the noise-free row in accuracy (93.33% vs 88.97%); please clarify whether this is seed variation, a known regularizing effect of noise, or an artifact of the noise-model implementation, and consider paired comparisons.
  2. [Section 3.1 / Appendix B] The symbol lambda is described as a hyperparameter but is fixed per task (0.1 for QML, 0.5 for QT and QPA); a sensitivity analysis over lambda would help readers understand the robustness of the hybrid mechanism.
  3. [Section 3.1 / after Eq. (7)] The notation 'phi_GH' for the combined parameters is inconsistent with the earlier notation theta, gamma, G, H and should be corrected.
  4. [Appendix C] The claim that the framework of [53] is a special case with lambda=0 and H containing a learnable quantum circuit seems to conflate postprocessing with blending; this reduction should be spelled out more explicitly.
  5. [Section 4 / footnote 1] The paper states that code is attached to the supplemental material but gives no repository link in the main text; please include a link for reproducibility.

Circularity Check

1 steps flagged · score 3.0 of 10

One formal claim (universality, Eq. 7) reduces to NQS universal approximation at λ=0; the empirical comparisons are self-contained but lack a quantum-attribution control.

  1. self definitional [Section 3.1, Eq. (7) after Eq. (6); see also the bullet 'When λ = 0...']
    "Since HPQS includes an NQS component with universal approximation capacity, it inherits the ability to approximate any target probability distribution p∗(ϕi) over computational basis states. Specifically, for any ϵ >0, there exist parameters ϕGH and a blending weight λ ∈ [0, 1] such that the hybrid estimate satisfies |ˆpi − p∗(ϕi)| < ϵ, ∀ϕi ∈ {0, 1}n, (7) ... When λ = 0, the model reduces to a purely classical (NQS-based) estimation framework."

    Equation (7) is not derived from the hybrid mechanism; it is the NQS universal-approximation statement of Eq. (4) restated at the λ = 0 limiting case of HPQS. Because λ = 0 makes p̂_i = H(ϕ_i, f_γ(ϕ_i)), any distribution approximable by the NQS branch is approximable by HPQS by definition. The formal expressivity result is therefore equivalent to its input assumption, and the claimed 'generality' of HPQS is a definitional inclusion rather than an independent derivation. This does not affect the empirical benchmark results, which are measured rather than derived from this corollary.

full rationale

The paper's central contribution is empirical: HPQS is a convex combination of a finite-shot PQC measurement and an NQS prediction, and Section 4 reports measured accuracy/perplexity against PQC (exact), PQC (finite), and NQS. These measurements are self-contained and do not reduce to the formal universality claim; they are not circular. The main derivational weakness is the 'theoretical expressivity' statement of Eq. (7), which is a direct special case of the NQS universal approximation theorem at λ = 0; the framework's 'generality' as a unifier of PQC-only and NQS-only models is likewise true by construction rather than by independent argument. The reliance on the author's prior QT and QPA frameworks is not circular for the empirical claims: those frameworks are published, machine-verifiable setups with their own statements, and the paper's HPQS experiments are new evaluations against baselines, not fits of the target quantities. However, the experiments do not include an ablation that isolates the quantum branch's contribution: HPQS always has a larger or equal classical model and the QML branch uses λ = 0.1, so the empirical advantage could be attributed to the classical estimator. That is an attribution/correctness concern, not a circular-derivation concern, and is therefore reflected only in the rationale rather than as an additional circular step.

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

HPQS introduces no new physical entity; it is a model composition of existing PQC and NQS components. The central claim rests on hand-chosen hyperparameters (lambda, bond dimensions, depth, hidden width) and on three domain assumptions: the universal approximation capacity of the classical branch, the representativeness of simulated noise models, and the validity of the QT/QPA parameter-generation paradigm.

free parameters (5)
  • Blending coefficient lambda (QML) = 0.1
    Fixed by hand in Appendix B.2; controls the hybrid balance and strongly affects results, with 90% weight on the classical branch.
  • Blending coefficient lambda (QT/QPA) = 0.5
    Fixed by hand in Appendix B.3 and B.4; gives equal weight to quantum and classical branches.
  • Tensor network bond dimensions r = QT: 2/1; QPA: 10/4
    Chosen per task in Appendix B.1; controls the capacity of the tensor network decoders that map basis-probability pairs to parameters.
  • NQS hidden layer width = 32
    Used in QT/QPA classical branch as specified in Appendix B.3; determines the capacity of the classical estimator.
  • PQC depth L = QT: 1; QPA: 8; QML: 4-qubit ansatz
    Circuit depth chosen per task in Appendix B; affects quantum expressivity and noise sensitivity.
assumptions (5)
  • standard math Universal approximation theorem for feedforward neural networks
    Invoked in Eq. (4) and Eq. (7) to claim that NQS, and hence HPQS at lambda = 0, can approximate any probability distribution.
  • domain assumption Born rule and empirical frequency estimation of |<phi|psi(theta)>|^2 from finite shots
    Foundation of all PQC probability estimates in Section 3.1 and Appendix D.
  • domain assumption IBM noise models (ibm_fez, ibm_torino) are representative of real hardware
    Appendix A uses simulated IBM noise models to support the noise-resilience claims.
  • domain assumption Tensor network MPS mapping can faithfully map basis-probability pairs to useful network weights
    Assumed in Sections 3.2, 3.4, and B.1; no theoretical guarantee or ablation is provided.
  • domain assumption Parameter generation from a low-dimensional latent space preserves task performance
    Borrowed from the author's prior QT/QPA works [35,44]; HPQS relies on this transfer without re-deriving it.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hybrid Parameterized Quantum States for Variational Quantum Learning." pith.science (2026). https://pith.science/paper/HWEB7MVG

@misc{pith2026250516676,
  author       = {Pith},
  title        = {Pith review of: Hybrid Parameterized Quantum States for Variational Quantum Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/HWEB7MVG}},
  note         = {Machine review of arXiv:2505.16676}
}
read the original abstract

Variational quantum learning faces practical challenges in the noisy intermediate-scale quantum (NISQ) era. Parameterized quantum circuit (PQC) models suffer from statistical uncertainty due to finite-shot measurements and are highly sensitive to quantum noise, while purely classical approximations like neural quantum states (NQS) lack access to genuine quantum correlations and are limited in scalability. This work introduces Hybrid Parameterized Quantum States (HPQS), a general-purpose modeling framework that interpolates between quantum and classical parameterizations. HPQS combines PQC-based measurements with neural estimators via a blending mechanism and postprocessing functions, enabling enhanced, shot-efficient evaluation under hardware constraints. We demonstrate HPQS across three representative quantum learning tasks: (1) Expectation-based QML, where HPQS yields higher classification accuracy than PQC-only and NQS-only baselines under limited quantum measurements. (2) Quantum-Train, where HPQS generates the entire parameter set of classical networks using polylogarithmic trainable variables; and (3) Quantum Parameter Adaptation (QPA), where HPQS produces LoRA adapter parameters for fine-tuning large language models like GPT-2 and Gemma-2 with improved perplexity under low-shot conditions; Together, these results position HPQS as a scalable, noise-resilient approach for variational quantum learning, compatible with both current NISQ hardware and future fault-tolerant architectures.

Figures

Figures reproduced from arXiv: 2505.16676 by the authors.

Figure 1
Figure 1. Overview of the HPQS framework. (a) HPQS interpolates between a pure classical NQS and [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

56 extracted references · 38 canonical work pages

  1. [1]

    Data re- uploading for a universal quantum classifier

    Adrián Pérez-Salinas, Alba Cervera-Lierta, Elies Gil-Fuster, and José I Latorre. Data re- uploading for a universal quantum classifier. Quantum, 4:226, 2020

    work page 2020

  2. [2]

    Effect of data encoding on the expres- sive power of variational quantum-machine-learning models.Physical Review A, 103(3):032430, 2021

    Maria Schuld, Ryan Sweke, and Johannes Jakob Meyer. Effect of data encoding on the expres- sive power of variational quantum-machine-learning models.Physical Review A, 103(3):032430, 2021

    work page 2021

  3. [3]

    Quantum machine learning in feature hilbert spaces

    Maria Schuld and Nathan Killoran. Quantum machine learning in feature hilbert spaces. Physical review letters, 122(4):040504, 2019

    2019

  4. [4]

    Hybrid quantum-classical convolutional neural networks

    Junhua Liu, Kwan Hui Lim, Kristin L Wood, Wei Huang, Chu Guo, and He-Liang Huang. Hybrid quantum-classical convolutional neural networks. Science China Physics, Mechanics & Astronomy, 64(9):290311, 2021

    work page 2021

  5. [5]

    Quantum machine learning

    Jacob Biamonte, Peter Wittek, Nicola Pancotti, Patrick Rebentrost, Nathan Wiebe, and Seth Lloyd. Quantum machine learning. Nature, 549(7671):195–202, 2017

    2017

  6. [6]

    Demonstration of quantum advantage in machine learning

    Diego Ristè, Marcus P Da Silva, Colm A Ryan, Andrew W Cross, Antonio D Córcoles, John A Smolin, Jay M Gambetta, Jerry M Chow, and Blake R Johnson. Demonstration of quantum advantage in machine learning. npj Quantum Information, 3(1):16, 2017

    work page 2017

  7. [7]

    Quantum advantage in learning from experiments

    Hsin-Yuan Huang, Michael Broughton, Jordan Cotler, Sitan Chen, Jerry Li, Masoud Mohseni, Hartmut Neven, Ryan Babbush, Richard Kueng, John Preskill, et al. Quantum advantage in learning from experiments. Science, 376(6598):1182–1186, 2022

    work page 2022

  8. [8]

    Challenges and opportunities in quantum machine learning

    Marco Cerezo, Guillaume Verdon, Hsin-Yuan Huang, Lukasz Cincio, and Patrick J Coles. Challenges and opportunities in quantum machine learning. Nature Computational Science, 2(9):567–576, 2022

    2022

Show all 56 references
  1. [9]

    Few measurement shots challenge generalization in learning to classify entanglement

    Leonardo Banchi, Jason Pereira, and Marco Zamboni. Few measurement shots challenge generalization in learning to classify entanglement. arXiv preprint arXiv:2411.06600, 2024

    2024 arXiv

  2. [10]

    Quantum computing in the nisq era and beyond

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

    2018

  3. [11]

    Shadow tomography of quantum states

    Scott Aaronson. Shadow tomography of quantum states. In Proceedings of the 50th annual ACM SIGACT symposium on theory of computing, pages 325–338, 2018

    2018

  4. [12]

    Reliable quantum state tomography

    Matthias Christandl and Renato Renner. Reliable quantum state tomography. Physical Review Letters, 109(12):120403, 2012

    2012

  5. [13]

    Quantum error mitigation

    Zhenyu Cai, Ryan Babbush, Simon C Benjamin, Suguru Endo, William J Huggins, Ying Li, Jarrod R McClean, and Thomas E O’Brien. Quantum error mitigation. Reviews of Modern Physics, 95(4):045005, 2023

    2023

  6. [14]

    Practical quantum error mitigation for near- future applications

    Suguru Endo, Simon C Benjamin, and Ying Li. Practical quantum error mitigation for near- future applications. Physical Review X, 8(3):031027, 2018

    2018

  7. [15]

    Quantum error correction

    Daniel A Lidar and Todd A Brun. Quantum error correction. Cambridge university press, 2013

    2013

  8. [16]

    High-threshold and low-overhead fault-tolerant quantum memory

    Sergey Bravyi, Andrew W Cross, Jay M Gambetta, Dmitri Maslov, Patrick Rall, and Theodore J Yoder. High-threshold and low-overhead fault-tolerant quantum memory. Na- ture, 627(8005):778–782, 2024

    2024

  9. [17]

    Solving the quantum many-body problem with artificial neural networks

    Giuseppe Carleo and Matthias Troyer. Solving the quantum many-body problem with artificial neural networks. Science, 355(6325):602–606, 2017

    2017

  10. [18]

    Neural- network quantum states, string-bond states, and chiral topological states

    Ivan Glasser, Nicola Pancotti, Moritz August, Ivan D Rodriguez, and J Ignacio Cirac. Neural- network quantum states, string-bond states, and chiral topological states. Physical Review X, 8(1):011006, 2018

    2018

  11. [19]

    Neural-network approach to dissipative quantum many-body dynamics

    Michael J Hartmann and Giuseppe Carleo. Neural-network approach to dissipative quantum many-body dynamics. Physical review letters, 122(25):250502, 2019. 11

    2019

  12. [20]

    Symmetries and many- body excitations with neural-network quantum states

    Kenny Choo, Giuseppe Carleo, Nicolas Regnault, and Titus Neupert. Symmetries and many- body excitations with neural-network quantum states. Physical review letters, 121(16):167204, 2018

    2018

  13. [21]

    Neural-network quantum state tomography

    Giacomo Torlai, Guglielmo Mazzola, Juan Carrasquilla, Matthias Troyer, Roger Melko, and Giuseppe Carleo. Neural-network quantum state tomography. Nature physics, 14(5):447–450, 2018

    2018

  14. [22]

    Fermionic neural-network states for ab-initio electronic structure

    Kenny Choo, Antonio Mezzacapo, and Giuseppe Carleo. Fermionic neural-network states for ab-initio electronic structure. Nature communications, 11(1):2368, 2020

    2020

  15. [23]

    Neural-network quantum states for ultra-cold fermi gases

    Jane Kim, Gabriel Pescia, Bryce Fore, Jannes Nys, Giuseppe Carleo, Stefano Gandolfi, Morten Hjorth-Jensen, and Alessandro Lovato. Neural-network quantum states for ultra-cold fermi gases. Communications Physics, 7(1):148, 2024

    2024

  16. [24]

    Phases of two-dimensional spinless lattice fermions with first-quantized deep neural-network quantum states

    James Stokes, Javier Robledo Moreno, Eftychios A Pnevmatikakis, and Giuseppe Carleo. Phases of two-dimensional spinless lattice fermions with first-quantized deep neural-network quantum states. Physical Review B, 102(20):205122, 2020

    2020

  17. [25]

    Variational quantum circuits for deep reinforcement learning

    Samuel Yen-Chi Chen, Chao-Han Huck Yang, Jun Qi, Pin-Yu Chen, Xiaoli Ma, and Hsi- Sheng Goan. Variational quantum circuits for deep reinforcement learning. IEEE access, 8:141007–141024, 2020

    2020

  18. [26]

    Expressive power of parametrized quantum circuits

    Yuxuan Du, Min-Hsiu Hsieh, Tongliang Liu, and Dacheng Tao. Expressive power of parametrized quantum circuits. Physical Review Research, 2(3):033125, 2020

    2020

  19. [27]

    Parameterized quantum circuits as machine learning models

    Marcello Benedetti, Erika Lloyd, Stefan Sack, and Mattia Fiorentini. Parameterized quantum circuits as machine learning models. Quantum science and technology, 4(4):043001, 2019

    2019

  20. [28]

    Hardware-efficient variational quantum eigensolver for small molecules and quantum magnets

    Abhinav Kandala, Antonio Mezzacapo, Kristan Temme, Maika Takita, Markus Brink, Jerry M Chow, and Jay M Gambetta. Hardware-efficient variational quantum eigensolver for small molecules and quantum magnets. nature, 549(7671):242–246, 2017

    2017

  21. [29]

    On the practical usefulness of the hardware efficient ansatz

    Lorenzo Leone, Salvatore FE Oliviero, Lukasz Cincio, and Marco Cerezo. On the practical usefulness of the hardware efficient ansatz. Quantum, 8:1395, 2024

    2024

  22. [30]

    Efficient quantum state tomography

    Marcus Cramer, Martin B Plenio, Steven T Flammia, Rolando Somma, David Gross, Stephen D Bartlett, Olivier Landon-Cardinal, David Poulin, and Yi-Kai Liu. Efficient quantum state tomography. Nature communications, 1(1):149, 2010

    2010

  23. [31]

    Quan- tum state tomography via nonconvex riemannian gradient descent

    Ming-Chien Hsu, En-Jui Kuo, Wei-Hsuan Yu, Jian-Feng Cai, and Min-Hsiu Hsieh. Quan- tum state tomography via nonconvex riemannian gradient descent. Physical Review Letters, 132(24):240804, 2024

    2024

  24. [32]

    Error mitigation extends the computational reach of a noisy quantum processor

    Abhinav Kandala, Kristan Temme, et al. Error mitigation extends the computational reach of a noisy quantum processor. Nature, 567:491–495, 2019

    2019

  25. [33]

    Approximation by superpositions of a sigmoidal function

    George Cybenko. Approximation by superpositions of a sigmoidal function. Mathematics of control, signals and systems, 2(4):303–314, 1989

    1989

  26. [34]

    Multilayer feedforward networks are universal approximators

    Kurt Hornik, Maxwell Stinchcombe, and Halbert White. Multilayer feedforward networks are universal approximators. Neural networks, 2(5):359–366, 1989

    1989

  27. [35]

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

    Chen-Yu Liu, En-Jui Kuo, Chu-Hsuan Abraham Lin, Jason Gemsun Young, Yeong-Jar Chang, Min-Hsiu Hsieh, and Hsi-Sheng Goan. Quantum-train: Rethinking hybrid quantum-classical machine learning in the model compression perspective. arXiv preprint arXiv:2405.11304, 2024

    2024 arXiv

  28. [36]

    Training classical neural networks by quantum machine learning

    Chen-Yu Liu, En-Jui Kuo, Chu-Hsuan Abraham Lin, Sean Chen, Jason Gemsun Young, Yeong- Jar Chang, and Min-Hsiu Hsieh. Training classical neural networks by quantum machine learning. In 2024 IEEE International Conference on Quantum Computing and Engineering (QCE), volume 2, page...

    2024

  29. [37]

    Qtrl: Toward practical quantum reinforcement learning via quantum-train

    Chen-Yu Liu, Chu-Hsuan Abraham Lin, Chao-Han Huck Yang, Kuan-Cheng Chen, and Min- Hsiu Hsieh. Qtrl: Toward practical quantum reinforcement learning via quantum-train. In 2024 IEEE International Conference on Quantum Computing and Engineering (QCE) , volume 2, pages 317–322. IEEE, 2024

    2024

  30. [38]

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

    Chu-Hsuan Abraham Lin, Chen-Yu Liu, and Kuan-Cheng Chen. Quantum-train long short-term memory: Application on flood prediction problem. In 2024 IEEE International Conference on Quantum Computing and Engineering (QCE), volume 2, pages 268–273. IEEE, 2024

    2024

  31. [39]

    Quantum-train with tensor network mapping model and distributed circuit ansatz

    Chen-Yu Liu, Chu-Hsuan Abraham Lin, and Kuan-Cheng Chen. Quantum-train with tensor network mapping model and distributed circuit ansatz. InICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pages 1–4. IEEE, 2025

    2025

  32. [40]

    Quantumnas: Noise-adaptive search for robust quantum circuits

    Hanrui Wang, Yongshan Ding, Jiaqi Gu, Yujun Lin, David Z Pan, Frederic T Chong, and Song Han. Quantumnas: Noise-adaptive search for robust quantum circuits. In 2022 IEEE International Symposium on High-Performance Computer Architecture (HPCA), pages 692–708. IEEE, 2022

    2022

  33. [41]

    Language models are unsupervised multitask learners

    Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9, 2019

    2019

  34. [42]

    Gemma 2: Improving open language models at a practical size

    Gemma Team, Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupatiraju, Léonard Hussenot, Thomas Mesnard, Bobak Shahriari, Alexandre Ramé, et al. Gemma 2: Improving open language models at a practical size. arXiv preprint arXiv:2408.00118, 2024

    2024 arXiv

  35. [43]

    Lora: Low-rank adaptation of large language models

    Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, Weizhu Chen, et al. Lora: Low-rank adaptation of large language models. ICLR, 1(2):3, 2022

    2022

  36. [44]

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

    Chen-Yu Liu, Chao-Han Huck Yang, Hsi-Sheng Goan, and Min-Hsiu Hsieh. A quantum circuit- based compression perspective for parameter-efficient learning. In The Thirteenth International Conference on Learning Representations, 2025

    2025

  37. [45]

    Federated quantum-train with batched parameter gen- eration

    Chen-Yu Liu and Samuel Yen-Chi Chen. Federated quantum-train with batched parameter gen- eration. In 2024 15th International Conference on Information and Communication Technology Convergence (ICTC), pages 1133–1138. IEEE, 2024

    2024

  38. [46]

    Automatic differentiation in pytorch

    Adam Paszke, Sam Gross, Soumith Chintala, Gregory Chanan, Edward Yang, Zachary DeVito, Zeming Lin, Alban Desmaison, Luca Antiga, and Adam Lerer. Automatic differentiation in pytorch. 2017

    2017

  39. [47]

    Mnist handwritten digit database

    Yann LeCun, Corinna Cortes, and CJ Burges. Mnist handwritten digit database. ATT Labs [Online]. Available: http://yann.lecun.com/exdb/mnist, 2, 2010

    2010

  40. [48]

    Pointer sentinel mixture models, 2016

    Stephen Merity, Caiming Xiong, James Bradbury, and Richard Socher. Pointer sentinel mixture models, 2016

    2016

  41. [49]

    Supervised learning with tensor networks

    Edwin Stoudenmire and David J Schwab. Supervised learning with tensor networks. Advances in neural information processing systems, 29, 2016

    2016

  42. [50]

    Torchmps

    Jacob Miller. Torchmps. https://github.com/jemisjoky/torchmps, 2019

    2019

  43. [51]

    Variational quantum algorithms

    Marco Cerezo, Andrew Arrasmith, Ryan Babbush, Simon C Benjamin, Suguru Endo, Keisuke Fujii, Jarrod R McClean, Kosuke Mitarai, Xiao Yuan, Lukasz Cincio, et al. Variational quantum algorithms. Nature Reviews Physics, 3(9):625–644, 2021

    2021

  44. [52]

    Quanvolutional neural networks: powering image recognition with quantum circuits

    Maxwell Henderson, Samriddhi Shakya, Shashindra Pradhan, and Tristan Cook. Quanvolutional neural networks: powering image recognition with quantum circuits. Quantum Machine Intelligence, 2(1):2, 2020

    2020

  45. [53]

    Transfer learning in hybrid classical-quantum neural networks

    Andrea Mari, Thomas R Bromley, Josh Izaac, Maria Schuld, and Nathan Killoran. Transfer learning in hybrid classical-quantum neural networks. Quantum, 4:340, 2020. 13

    2020

  46. [54]

    Supervised learning with quantum enhanced feature spaces

    V ojtˇech Havlíˇcek, Antonio D Córcoles, Kristan Temme, Aram W Harrow, Abhinav Kandala, Jerry M Chow, and Jay M Gambetta. Supervised learning with quantum enhanced feature spaces. Nature, 567(7747):209–212, 2019

    2019

  47. [55]

    Learning-based quantum error mitigation

    Armands Strikis, Dayue Qin, Yanzhu Chen, Simon C Benjamin, and Ying Li. Learning-based quantum error mitigation. PRX Quantum, 2(4):040330, 2021

    2021

  48. [56]

    Evidence for the utility of quantum computing before fault tolerance

    Youngseok Kim, Andrew Eddins, Sajant Anand, Ken Xuan Wei, Ewout Van Den Berg, Sami Rosenblatt, Hasan Nayfeh, Yantao Wu, Michael Zaletel, Kristan Temme, et al. Evidence for the utility of quantum computing before fault tolerance. Nature, 618(7965):500–505, 2023. 14 A Effects of...

    2023

Pith tools

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