pith. machine review for the scientific record. sign in

arxiv: 2601.11942 · v3 · submitted 2026-01-17 · 💻 cs.LG · quant-ph

Recognition: 2 theorem links

· Lean Theorem

Geometric Preconditioning and Curriculum Optimization for Trainable Variational Quantum Regression

Qingyu Meng , Yangshuai Wang
Authors on Pith no claims yet

Pith reviewed 2026-05-16 13:10 UTC · model grok-4.3

classification 💻 cs.LG quant-ph
keywords variational quantum circuitsquantum regressiongeometric preconditioningcurriculum learninghybrid quantum-classical modelsdata reuploadingtrainabilityGram matrix
0
0 comments X

The pith

A capacity-controlled classical embedding acts as a learnable geometric preconditioner to improve trainability of variational quantum circuits for regression.

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

The paper examines why variational quantum circuits struggle to train on regression tasks when global losses, finite shots, and growing circuit depth weaken gradient signals. It introduces a hybrid architecture in which a classical embedding layer reshapes the input distribution seen by a data-reuploading quantum circuit while keeping a low-dimensional quantum bottleneck. This embedding is shown to modify the empirical Gram matrix that governs residual contraction in the linearized parameter dynamics. A curriculum that progressively deepens the circuit and switches from SPSA exploration to Adam fine-tuning is added to stabilize the process. Finite-statevector experiments on PDE regression benchmarks and small tabular data sets report lower errors for the hybrid model than for pure quantum baselines under identical quantum resource budgets.

Core claim

The central claim is that a capacity-controlled classical embedding functions as a learnable geometric preconditioner for data-reuploading variational circuits: by reshaping the input distribution it alters the empirical Gram matrix that controls one-step loss decrease in the linearized quantum-parameter dynamics, and when this design is paired with a progressive curriculum that grows circuit depth and switches optimizers, the resulting hybrid model achieves lower regression error than matched pure quantum baselines on the tested benchmarks.

What carries the argument

The capacity-controlled classical embedding that reshapes the empirical Gram matrix in the local quantum-tangent contraction analysis.

If this is right

  • The hybrid model produces lower test error than pure quantum networks on PDE regression and small-data tabular tasks under fixed quantum budget.
  • Progressive depth growth combined with an optimizer switch from SPSA to Adam stabilizes training of the data-reuploading circuit.
  • The observed benefit is attributed to improved trainability rather than to any absolute performance edge over strong classical regressors.
  • The local quantum-tangent contraction statement supplies a formal link between the embedding's effect on the Gram matrix and measured loss decrease.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same embedding-plus-curriculum pattern could be tested on classification or generative tasks that share similar gradient-conditioning problems.
  • If the Gram-matrix reshaping is the dominant mechanism, replacing the classical embedding with other differentiable preconditioners should produce comparable gains.
  • The design may extend naturally to noisy intermediate-scale hardware once the statevector audits are repeated under realistic noise models.

Load-bearing premise

The classical embedding successfully reshapes the empirical Gram matrix to improve residual contraction in the quantum-parameter dynamics without introducing offsetting ill-conditioning.

What would settle it

A statevector experiment on the same PDE-informed benchmarks in which the hybrid model shows no reduction in error relative to the pure quantum baseline despite the embedding layer being present and the measured Gram matrix change being negligible.

Figures

Figures reproduced from arXiv: 2601.11942 by Qingyu Meng, Yangshuai Wang.

Figure 1
Figure 1. Figure 1: Hybrid Quantum Regression Framework. A lightweight classical embedding transforms inputs before quantum encoding to improve conditioning for the variational circuit. Training follows a curriculum that gradually increases circuit depth and transitions from SPSA-based exploration to Adam-based refinement. p = nq for single-qubit encoding gates). We view this map￾ping as a learnable geometric preconditioner: … view at source ↗
Figure 2
Figure 2. Figure 2: Numerical validation on PDE benchmarks. We show ground truth and absolute error distributions for four benchmarks: (A) 2D Poisson, (B) 2D Convection–Diffusion, (C) 2D Nonlinear Poisson, and (D) 3D Modified Helmholtz. Under the fixed training protocol, the Hybrid QNN yields lower and less structured error patterns compared to the considered baselines [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Optimization dynamics on Yacht (representative fold). Training loss trajectories for SPSA-only, Adam-only, and the pro￾posed two-stage schedule. A.2 Impact of Quantum Natural Gradient We further investigate geometry-aware updates for the quan￾tum parameters on the Concrete dataset. In variational quan￾tum circuits, the parameter space induced by the model [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
read the original abstract

Variational quantum circuits are increasingly studied as continuous-function approximators, but quantum regression remains difficult to train when global losses, finite-shot stochasticity, and circuit-depth growth combine to produce weak or ill-conditioned gradient signals. We study this trainability problem in a controlled hybrid quantum--classical regression design. The central ingredient is a capacity-controlled classical embedding that acts as a learnable geometric preconditioner: it reshapes the input distribution seen by a data-reuploading variational circuit while preserving a low-dimensional quantum bottleneck. We pair this representation design with a curriculum protocol that grows circuit depth progressively and switches from SPSA-based stochastic exploration to Adam-based analytic-gradient fine-tuning. We formalize the mechanism through a local quantum-tangent contraction statement: in the linearized quantum-parameter dynamics, the embedding changes the empirical Gram matrix that controls residual contraction and one-step loss decrease. Across finite-size statevector audits on PDE-informed regression benchmarks and small-data tabular tasks, the Hybrid QNN lowers error relative to Pure QNN baselines under matched quantum-model budgets. Strong classical references remain competitive, and in several cases are better in absolute error; the evidence therefore supports a trainability claim for the hybrid QNN design rather than a claim of classical or hardware quantum advantage.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript proposes a hybrid quantum-classical regression architecture in which a capacity-controlled classical embedding acts as a learnable geometric preconditioner for a data-reuploading variational quantum circuit. The design is paired with a curriculum that progressively deepens the circuit and switches from SPSA to Adam optimization. The central formal claim is a local quantum-tangent contraction statement asserting that the embedding reshapes the empirical Gram matrix governing one-step residual contraction in the linearized quantum-parameter dynamics. Finite-size statevector simulations on PDE-informed and small-data tabular benchmarks are reported to show lower error for the hybrid model relative to pure-QNN baselines under matched quantum budgets.

Significance. If the geometric-preconditioning mechanism can be directly verified, the work would provide a concrete, resource-efficient route to improving trainability of variational quantum regressors. The curriculum protocol and explicit separation of classical capacity from the quantum bottleneck are positive design choices. However, the present evidence consists solely of final error values; without diagnostics on the Gram matrix, condition numbers, or measured contraction rates, the significance remains provisional and the trainability claim is not yet load-bearing.

major comments (3)
  1. [Formalization of local quantum-tangent contraction] The local quantum-tangent contraction statement (invoked to explain the mechanism) is presented without derivation or numerical verification. No eigenvalues of the empirical Gram matrix G, condition numbers, or one-step loss-decrease rates are shown for matched Pure-QNN versus Hybrid-QNN configurations, leaving open the possibility that observed error reductions arise from added classical capacity or the curriculum schedule rather than the claimed geometric effect.
  2. [Benchmark results and experimental protocol] The abstract and results sections assert that the Hybrid QNN lowers error relative to Pure QNN baselines, yet supply no quantitative metrics, error bars, statistical significance tests, or details on data exclusion or hyper-parameter matching. This absence makes it impossible to assess whether the reported improvements are robust or merely within noise.
  3. [Capacity-control analysis] The weakest-assumption paragraph notes that the classical embedding must reshape the Gram matrix without introducing offsetting ill-conditioning; the manuscript provides no diagnostic (e.g., condition-number trajectories or eigenvalue spectra) that would confirm this balance is achieved.
minor comments (2)
  1. [Theoretical development] Notation for the empirical Gram matrix G and the linearized update should be introduced with an explicit equation reference rather than by name only.
  2. [Experimental setup] The manuscript should clarify whether the reported statevector audits use the same random seeds and initialization distributions for Pure and Hybrid runs; otherwise the comparison is not strictly controlled.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the thoughtful review and constructive suggestions. We provide point-by-point responses to the major comments and outline the revisions we will make to address the concerns raised.

read point-by-point responses

  1. Referee: [Formalization of local quantum-tangent contraction] The local quantum-tangent contraction statement (invoked to explain the mechanism) is presented without derivation or numerical verification. No eigenvalues of the empirical Gram matrix G, condition numbers, or one-step loss-decrease rates are shown for matched Pure-QNN versus Hybrid-QNN configurations, leaving open the possibility that observed error reductions arise from added classical capacity or the curriculum schedule rather than the claimed geometric effect.

    Authors: We acknowledge that the derivation of the local quantum-tangent contraction was not included in the original submission to maintain focus on the main results. In the revised manuscript, we will add a detailed derivation in the supplementary material, starting from the linearized quantum-parameter dynamics and showing how the embedding modifies the empirical Gram matrix to enhance residual contraction. Furthermore, we will include new numerical experiments displaying the eigenvalues of G, condition numbers, and one-step loss decrease rates for both Pure-QNN and Hybrid-QNN setups under matched conditions. These additions will help substantiate that the observed improvements are due to the geometric preconditioning effect. revision: yes

  2. Referee: [Benchmark results and experimental protocol] The abstract and results sections assert that the Hybrid QNN lowers error relative to Pure QNN baselines, yet supply no quantitative metrics, error bars, statistical significance tests, or details on data exclusion or hyper-parameter matching. This absence makes it impossible to assess whether the reported improvements are robust or merely within noise.

    Authors: We agree that more detailed reporting is necessary for a thorough evaluation. The revised manuscript will include comprehensive quantitative metrics, including mean errors with standard deviations from repeated runs, error bars on all relevant figures, and statistical significance tests such as Wilcoxon signed-rank tests or t-tests to compare the Hybrid QNN against Pure QNN baselines. We will also provide explicit details on hyper-parameter selection, matching procedures, data preprocessing, and any exclusion criteria to allow readers to assess the robustness of the results. revision: yes

  3. Referee: [Capacity-control analysis] The weakest-assumption paragraph notes that the classical embedding must reshape the Gram matrix without introducing offsetting ill-conditioning; the manuscript provides no diagnostic (e.g., condition-number trajectories or eigenvalue spectra) that would confirm this balance is achieved.

    Authors: The capacity-control analysis is based on the premise that the classical embedding can be tuned to improve the conditioning of the effective Gram matrix. To address this, we will add in the revision a set of diagnostic analyses, including plots of condition-number trajectories over training and eigenvalue spectra of the Gram matrix for the hybrid configurations. These will demonstrate that the embedding achieves the desired reshaping without detrimental ill-conditioning in the reported experiments. revision: yes

Circularity Check

1 steps flagged

The local quantum-tangent contraction statement restates the embedding design as geometric preconditioning without independent derivation of Gram-matrix effects.

specific steps
  1. self definitional [Abstract]
    "The central ingredient is a capacity-controlled classical embedding that acts as a learnable geometric preconditioner: it reshapes the input distribution seen by a data-reuploading variational circuit while preserving a low-dimensional quantum bottleneck. ... We formalize the mechanism through a local quantum-tangent contraction statement: in the linearized quantum-parameter dynamics, the embedding changes the empirical Gram matrix that controls residual contraction and one-step loss decrease."

    The claimed mechanism (embedding changes Gram matrix to control contraction) is presented as a formalization, yet it is identical to the prior definition of the embedding as a geometric preconditioner that reshapes the input distribution. No additional derivation or direct verification of the Gram-matrix change or contraction improvement is supplied; the statement is therefore tautological to the model architecture.

full rationale

The paper defines the capacity-controlled classical embedding as a learnable geometric preconditioner that reshapes the input distribution for the variational circuit. It then presents a 'local quantum-tangent contraction statement' that the embedding changes the empirical Gram matrix controlling residual contraction. This formalization directly follows from the architectural definition rather than from separate equations or diagnostics (no eigenvalues, condition numbers, or measured one-step contraction rates are shown). Evidence consists solely of final error reductions on statevector benchmarks, which could arise from added classical capacity or the curriculum schedule. The central trainability claim therefore reduces partially to the design choice itself, producing moderate circularity while still retaining some independent empirical content from the benchmark comparisons.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the unverified local quantum-tangent contraction statement and the assumption that the classical embedding parameters can be trained to improve conditioning without introducing offsetting instabilities. No new physical entities are postulated.

free parameters (1)
  • capacity control parameters of classical embedding
    The embedding is described as capacity-controlled and learnable, implying tunable parameters whose values are fitted during training to achieve the preconditioning effect.
axioms (1)
  • domain assumption Local quantum-tangent contraction statement holds in the linearized dynamics
    The mechanism by which the embedding improves gradient signals is formalized through this statement relating the Gram matrix to residual contraction.

pith-pipeline@v0.9.0 · 5515 in / 1384 out tokens · 72154 ms · 2026-05-16T13:10:28.472850+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

36 extracted references · 36 canonical work pages

  1. [1]

    The power of quantum neural networks.Nature Computa- tional Science, 1(6):403–409,

    [Abbaset al., 2021 ] Amira Abbas, David Sutter, Christa Zo- ufal, Aur´elien Lucchi, Alessio Figalli, and Stefan Woerner. The power of quantum neural networks.Nature Computa- tional Science, 1(6):403–409,

    work page 2021

  2. [2]

    UCI machine learning repository,

    [Asuncionet al., 2007 ] Arthur Asuncion, David Newman, et al. UCI machine learning repository,

    work page 2007

  3. [3]

    Training deep quan- tum neural networks.Nature communications, 11(1):808,

    [Beeret al., 2020 ] Kerstin Beer, Dmytro Bondarenko, Terry Farrelly, Tobias J Osborne, Robert Salzmann, Daniel Scheiermann, and Ramona Wolf. Training deep quan- tum neural networks.Nature communications, 11(1):808,

    work page 2020

  4. [4]

    Trainable embedding quantum physics informed neural networks for solving nonlinear pdes.Sci- entific Reports, 15(1):18823,

    [Bergeret al., 2025 ] Stefan Berger, Norbert Hosters, and Matthias M ¨oller. Trainable embedding quantum physics informed neural networks for solving nonlinear pdes.Sci- entific Reports, 15(1):18823,

    work page 2025

  5. [5]

    Con- textuality and inductive bias in quantum machine learning

    [Bowleset al., 2023 ] Joseph Bowles, Victoria J Wright, M´at´e Farkas, Nathan Killoran, and Maria Schuld. Con- textuality and inductive bias in quantum machine learning. arXiv preprint arXiv:2302.01365,

    work page arXiv 2023

  6. [6]

    Quantum neural ordinary and partial differential equations.arXiv preprint arXiv:2508.18326,

    [Caoet al., 2025 ] Yu Cao, Shi Jin, and Nana Liu. Quantum neural ordinary and partial differential equations.arXiv preprint arXiv:2508.18326,

    work page arXiv 2025

  7. [7]

    Higher order derivatives of quantum neural networks with barren plateaus.Quantum Science and Technology, 6(3):035006,

    [Cerezo and Coles, 2021] Marco Cerezo and Patrick J Coles. Higher order derivatives of quantum neural networks with barren plateaus.Quantum Science and Technology, 6(3):035006,

    work page 2021

  8. [8]

    Cerezo, Akira Sone, Tyler V olkoff, Łukasz Cincio, and Patrick J

    [Cerezoet al., 2021 ] M. Cerezo, Akira Sone, Tyler V olkoff, Łukasz Cincio, and Patrick J. Coles. Cost function de- pendent barren plateaus in shallow parametrized quantum circuits.Nature Communications, 12(1):1791,

    work page 2021

  9. [9]

    Investigating and mitigating barren plateaus in variational quantum circuits: a survey.arXiv preprint arXiv:2407.17706,

    [Cunningham and Zhuang, 2024] Jack Cunningham and Jun Zhuang. Investigating and mitigating barren plateaus in variational quantum circuits: a survey.arXiv preprint arXiv:2407.17706,

    work page arXiv 2024

  10. [10]

    QCPINN: quantum-classical physics-informed neural networks for solving pdes

    [Fareaet al., 2025 ] Afrah Farea, Saiful Khan, and Mustafa Serdar Celebi. QCPINN: quantum-classical physics-informed neural networks for solving pdes. Machine Learning: Science and Technology,

    work page 2025

  11. [11]

    Multiple embeddings for quantum machine learning

    [Hanet al., 2025 ] Siyu Han, Lihan Jia, and Lanzhe Guo. Multiple embeddings for quantum machine learning. arXiv preprint arXiv:2503.22758,

    work page arXiv 2025

  12. [12]

    Quan- tum algorithms for multiscale partial differential equa- tions.Multiscale Modeling & Simulation, 22(3):1030– 1067,

    [Huet al., 2024 ] Junpeng Hu, Shi Jin, and Lei Zhang. Quan- tum algorithms for multiscale partial differential equa- tions.Multiscale Modeling & Simulation, 22(3):1030– 1067,

    work page 2024

  13. [13]

    Quantum machine learning beyond ker- nel methods.Nature Communications, 14(1):517,

    [Jerbiet al., 2023 ] Sofiene Jerbi, Lukas Faehrmann, Ville Tanskanen, et al. Quantum machine learning beyond ker- nel methods.Nature Communications, 14(1):517,

    work page 2023

  14. [14]

    Quan- tum simulation of partial differential equations via schr ¨odingerization.Physical Review Letters, 133(23):230602,

    [Jinet al., 2024 ] Shi Jin, Nana Liu, and Yue Yu. Quan- tum simulation of partial differential equations via schr ¨odingerization.Physical Review Letters, 133(23):230602,

    work page 2024

  15. [15]

    Neural operator: Learning maps between function spaces with applications to pdes.Journal of Machine Learning Research, 24(89):1–97,

    [Kovachkiet al., 2023 ] Nikola Kovachki, Zongyi Li, Burigede Liu, Kamyar Azizzadenesheli, Kaushik Bhat- tacharya, Andrew Stuart, and Anima Anandkumar. Neural operator: Learning maps between function spaces with applications to pdes.Journal of Machine Learning Research, 24(89):1–97,

    work page 2023

  16. [16]

    Solving nonlinear dif- ferential equations with differentiable quantum circuits

    [Kyriienkoet al., 2021 ] Oleksandr Kyriienko, Annie E Paine, and Vincent E Elfving. Solving nonlinear dif- ferential equations with differentiable quantum circuits. Physical Review A, 103(5):052416,

    work page 2021

  17. [17]

    Theory of overparametrization in quantum neural networks.Nature Computational Science, 3(6):542–551,

    [Laroccaet al., 2023 ] Martin Larocca, Nathan Ju, Diego Garc´ıa-Mart´ın, Patrick J Coles, and M Cerezo. Theory of overparametrization in quantum neural networks.Nature Computational Science, 3(6):542–551,

    work page 2023

  18. [18]

    Symmetry-invariant quantum machine learning force fields.New Journal of Physics, 27(2):023015,

    [Leet al., 2025 ] Isabel Nha Minh Le, Oriel Kiss, Julian Schuhmacher, Ivano Tavernelli, and Francesco Tacchino. Symmetry-invariant quantum machine learning force fields.New Journal of Physics, 27(2):023015,

    work page 2025

  19. [19]

    Quantum fisher information ma- trix and multiparameter estimation.Journal of Physics A: Mathematical and Theoretical, 53(2):023001,

    [Liuet al., 2020 ] Jing Liu, Haidong Yuan, Xiao-Ming Lu, and Xiaoguang Wang. Quantum fisher information ma- trix and multiparameter estimation.Journal of Physics A: Mathematical and Theoretical, 53(2):023001,

    work page 2020

  20. [20]

    Quantum-embedded graph neural network architecture for molecular property pre- diction.Journal of Chemical Information and Modeling, 65(15):8057–8065,

    [Luet al., 2025 ] Min Lu, Lei Du, Ziwei Cui, Yiming Zhao, Qipeng Yan, Jianyu Zhao, Ye Li, Menghan Dou, Qingchun Wang, Yu-Chun Wu, et al. Quantum-embedded graph neural network architecture for molecular property pre- diction.Journal of Chemical Information and Modeling, 65(15):8057–8065,

    work page 2025

  21. [21]

    Barren plateaus in quantum neural network train- ing landscapes.Nature Communications, 9(1):4812,

    [McCleanet al., 2018 ] Jarrod R McClean, Sergio Boixo, Vadim N Smelyanskiy, Ryan Babbush, and Hartmut Neven. Barren plateaus in quantum neural network train- ing landscapes.Nature Communications, 9(1):4812,

    work page 2018

  22. [22]

    Quantum circuit learning.Physical Review A, 98(3):032309,

    [Mitaraiet al., 2018 ] Kosuke Mitarai, Makoto Negoro, Masahiro Kitagawa, and Keisuke Fujii. Quantum circuit learning.Physical Review A, 98(3):032309,

    work page 2018

  23. [23]

    Quantum ex- treme learning of molecular potential energy surfaces and force fields.Machine Learning: Science and Technology, 5(3):035014,

    [Monacoet al., 2024 ] Gabriele Lo Monaco, Marco Bertini, Salvatore Lorenzo, and G Massimo Palma. Quantum ex- treme learning of molecular potential energy surfaces and force fields.Machine Learning: Science and Technology, 5(3):035014,

    work page 2024

  24. [24]

    Data re-uploading for a universal quantum classifier.Quantum, 4:226,

    [P´erez-Salinaset al., 2020 ] Adri´an P ´erez-Salinas, Alba Cervera-Lierta, Elies Gil-Fuster, and Jos ´e I Latorre. Data re-uploading for a universal quantum classifier.Quantum, 4:226,

    work page 2020

  25. [25]

    Absence of barren plateaus in quantum convolu- tional neural networks.Physical Review X, 11(4):041011,

    [Pesahet al., 2021 ] Arthur Pesah, Marco Cerezo, Samson Wang, Tyler V olkoff, Andrew T Sornborger, and Patrick J Coles. Absence of barren plateaus in quantum convolu- tional neural networks.Physical Review X, 11(4):041011,

    work page 2021

  26. [26]

    A lie algebraic theory of barren plateaus for deep parameterized quantum circuits.Nature Communications, 15(1):7172,

    [Ragoneet al., 2024 ] Michael Ragone, Bojko N Bakalov, Fr´ed´eric Sauvage, Alexander F Kemper, Carlos Ortiz Mar- rero, Mart´ın Larocca, and Marco Cerezo. A lie algebraic theory of barren plateaus for deep parameterized quantum circuits.Nature Communications, 15(1):7172,

    work page 2024

  27. [27]

    The quest for a quantum neural network.Quantum Information Processing, 13(11):2567– 2586,

    [Schuldet al., 2014 ] Maria Schuld, Ilya Sinayskiy, and Francesco Petruccione. The quest for a quantum neural network.Quantum Information Processing, 13(11):2567– 2586,

    work page 2014

  28. [28]

    Layerwise learning for quantum neural networks.Quan- tum Machine Intelligence, 3(1):1–11,

    [Skoliket al., 2021 ] Andrea Skolik, Jarrod R McClean, Ma- soud Mohseni, Patrick van der Smagt, and Martin Leib. Layerwise learning for quantum neural networks.Quan- tum Machine Intelligence, 3(1):1–11,

    work page 2021

  29. [29]

    Error mitigation for short-depth quantum circuits.Physical review letters, 119(18):180509,

    [Temmeet al., 2017 ] Kristan Temme, Sergey Bravyi, and Jay M Gambetta. Error mitigation for short-depth quantum circuits.Physical review letters, 119(18):180509,

    work page 2017

  30. [30]

    Variational quantum regression algorithm with encoded data structure.arXiv preprint arXiv:2307.03334,

    [Wang and Bennink, 2023] C-C Joseph Wang and Ryan S Bennink. Variational quantum regression algorithm with encoded data structure.arXiv preprint arXiv:2307.03334,

    work page arXiv 2023

  31. [31]

    Quanonet: Quantum neural operator with ap- plication to differential equation

    [Wanget al., 2022 ] Ruocheng Wang, Zhuo Xia, Ge Yan, and Junchi Yan. Quanonet: Quantum neural operator with ap- plication to differential equation. InForty-second Interna- tional Conference on Machine Learning,

    work page 2022

  32. [32]

    An empirical comparison of optimizers for quantum ma- chine learning with SPSA-based gradients

    [Wiedmannet al., 2023 ] Marco Wiedmann, Marc H ¨olle, Maniraman Periyasamy, Nico Meyer, Christian Ufrecht, Daniel D Scherer, Axel Plinge, and Christopher Mutschler. An empirical comparison of optimizers for quantum ma- chine learning with SPSA-based gradients. In2023 IEEE International Conference on Quantum Computing and En- gineering (QCE), volume 1, pag...

    work page 2023

  33. [33]

    General parameter-shift rules for quantum gradients.Quantum, 6:677,

    [Wierichset al., 2022 ] David Wierichs, Josh Izaac, Cody Wang, and Cedric Yen-Yu Lin. General parameter-shift rules for quantum gradients.Quantum, 6:677,

    work page 2022

  34. [34]

    Physics-informed quan- tum neural network for solving forward and inverse prob- lems of partial differential equations.Physics of Fluids, 36(9),

    [Xiaoet al., 2024 ] Y Xiao, LM Yang, C Shu, SC Chew, BC Khoo, YD Cui, and YY Liu. Physics-informed quan- tum neural network for solving forward and inverse prob- lems of partial differential equations.Physics of Fluids, 36(9),

    work page 2024

  35. [35]

    Mindspore quantum: a user-friendly, high-performance, and ai- compatible quantum computing framework.arXiv preprint arXiv:2406.17248,

    [Xuet al., 2024 ] Xusheng Xu, Jiangyu Cui, Zidong Cui, Runhong He, Qingyu Li, Xiaowei Li, Yanling Lin, Jiale Liu, Wuxin Liu, Jiale Lu, et al. Mindspore quantum: a user-friendly, high-performance, and ai- compatible quantum computing framework.arXiv preprint arXiv:2406.17248,

    work page arXiv 2024

  36. [36]

    Power and limitations of single-qubit native quan- tum neural networks.Advances in Neural Information Processing Systems, 35:27810–27823,

    [Yuet al., 2022 ] Zhan Yu, Hongshun Yao, Mujin Li, and Xin Wang. Power and limitations of single-qubit native quan- tum neural networks.Advances in Neural Information Processing Systems, 35:27810–27823,

    work page 2022