Pith. sign in

REVIEW 4 major objections 6 minor 66 references

Deep-learned error mitigation via partially knitted circuits for the variational quantum eigensolver

T0 review · 4 major / 6 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 The paper shows that a deep neural network trained on noisy outputs of partially knitted versions of a variational circuit can predict ideal expectation values, bringing the ground-state energy error of a noisy variational quantum…

desk verdict Solid extension of the authors' DL-EM program with a genuinely useful partial-knitting trick, but the headline full-noise claim rests on one instance and thin statistics. read the letter →

arxiv 2506.04146 v1 pith:O7ROJNLQ submitted 2025-06-04 quant-ph cond-mat.dis-nn

Simone Cantori , Andrea Mari , David Vitali , Sebastiano Pilati This is my paper
classification quant-phcond-mat.dis-nn
keywords quantumerrormitigationvariationaleigensolverdeeplearningcircuitknittingmultilayerperceptronnoisyintermediate-scaleIsingmodel
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 argues that variational quantum eigensolvers can be made accurate at realistic noise levels if their errors are filtered by a deep neural network tailored to the specific circuit family. The network is trained on the fly during the optimization, using noisy expectation values and circuit descriptors as inputs and ideal expectation values as targets. What makes this practical is that the training circuits are produced by partial circuit knitting, which pins selected two-qubit gate angles to 0 or $\pi$ so the circuits become classically simulable, avoiding exponential reconstruction cost. On a random one-dimensional Ising model with a realistic superconducting-device noise model, the method reaches energy discrepancies below 1% at full noise and below 0.1% at tenfold-reduced noise, while zero-noise extrapolation and damping-factor corrections do not. The central bet is that a network trained on these slightly modified circuits transfers to the full circuit used in optimization.

What carries the argument

The load-bearing mechanism is the learned map from noisy expectation values and circuit angles to ideal expectation values, implemented by small multilayer perceptrons. The enabler is partial circuit knitting: cutting a circuit by fixing the angles of selected RZZ gates to 0 or $\pi$, with sampling overhead $O = \prod_{\theta_i \in K} (1 + 2|\sin\theta_i|)^2$, which makes training circuits classically simulable while keeping them representative of the target. The number of cuts $C$ sets the balance; the paper shows prediction error $1-R^2$ degrades as $C$ grows but stays acceptable over the tested range, and that the network can even filter shot noise in the knitted training targets.

What would settle it

Train the DL-EM network on circuits with a fixed number of cuts $C$ and test it on full circuits ($C=0$) for increasing system sizes; if $1-R^2$ rises steeply with $N$, or the VQE energy error exceeds 1% when the optimized angles fall outside the training distribution, the transfer claim is refuted.

Watch

Extended reading notes

Core claim

DL-EM trains a set of multilayer perceptrons, one per Hamiltonian term, to map the noisy expectation values of a variational circuit together with its gate angles to the corresponding ideal expectation values. The training data are generated by taking the current VQE circuit, fixing the angles of a subset of RZZ gates to 0 or $\pi$ (the partial knitting), which splits the circuit into two classically simulable halves, and computing both noisy and ideal outputs. The knotted gates are chosen so that the training circuits stay close to the target circuit, and the number of cuts $C$ controls the trade-off between classical cost and transfer accuracy. The paper reports that with $C = 12$ cuts on an 8-layer, 6-qubit circuit, DL-EM brings the energy discrepancy below 1% under the original noise model, and that the prediction error does not grow with qubit number for fixed $C$, which suggests the approach scales.

Load-bearing premise

The method assumes that a network trained on partially knitted circuits, where some two-qubit gate angles are pinned to 0 or $\pi$, learns a correction that transfers to the full variational circuit, and this transfer has no theoretical guarantee.

Editorial extensions

If this is right

  • If the transfer holds, VQE can be run on deep circuits with realistic noise and still keep energy error below the 1% threshold.
  • The method improves as hardware noise decreases, so it becomes more effective as devices improve.
  • General-purpose methods such as zero-noise extrapolation and damping-factor rescaling are outperformed on this testbed, indicating that circuit-specific training is worth its overhead.
  • Partial knitting keeps the classical cost of training-data generation manageable, allowing the method to be tested at larger qubit counts without exact simulation.

Reading between the lines

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

  • The same DL-EM recipe could be ported to other variational algorithms whose circuits admit partial knitting, such as QAOA, extending the result beyond ground-state energy estimation.
  • Because transferability is empirical, a natural stress test is to train at one noise level and evaluate at another; adding the noise strength as a network input could make the correction robust to drift.
  • The choice of which gates to cut is currently random; selecting gates with the smallest angles would likely reduce the sampling overhead while keeping training circuits closer to the target.
  • The authors' mention of transfer learning could be realized by pretraining on a pool of random Ising instances and fine-tuning on a new instance, cutting the on-the-fly training cost.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper introduces DL-EM, a supervised deep-learning error mitigation method for VQE. Multilayer perceptrons are trained on noisy expectation values plus circuit descriptors to predict ideal expectation values; training data are generated from partially knitted circuits (some RZZ angles fixed to 0 or pi) to reduce the classical simulation cost. The method is tested on a 1D random Ising model with N=6 qubits and P=8 layers, using a tunable FakeMelbourneV2 noise model benchmarked against three IBM devices. The authors report that DL-EM reaches energy discrepancies below 1% at p_noise=1 and below 0.1% at p_noise=0.1, outperforming ZNE and damping-factor-based mitigation, and they analyze the trade-off between number of cuts, training set size, and prediction error.

Significance. If the claims hold, the paper makes a useful contribution by showing that circuit-family-tailored supervised error mitigation can be trained cheaply via partial knitting and can restore VQE accuracy in deep circuits. Strengths include the realistic noise model validated against actual IBM devices (Fig. 4), the explicit resource trade-off analysis (Figs. 7-8), and the observation that the learning speed does not degrade with qubit number. The method is not circular: ideal targets come from independent classical simulations, and no variational energy is used in training. The main weakness is that the headline numerical claims rest on very limited statistical evidence, and the central generalization assumption across the knitting gap is not tested along an actual VQE trajectory.

major comments (4)
  1. [Sec. 3.1, Fig. 6] The claim that DL-EM achieves an energy error below 1% at p_noise=1 is based on a single Hamiltonian instance; the four-instance average reported in the text is only for p_noise=0.1. No error bars, standard deviations, or repeated training runs are provided for any of the displayed quantities, so the reader cannot assess the variability of the result or whether the <1% figure is robust.
  2. [Sec. 3.1, Fig. 6, DM comparison] The comparison with the damping-factor method is not made on equal footing: DM is applied only to the final circuit of a VQE run already performed with DL-EM, whereas ZNE and DL-EM are run throughout the optimization. This does not disprove the paper's comparative claim, but it leaves open the possibility that DM combined with full VQE would perform better; the authors should either run DM (and DM+ZNE) during the whole VQE or soften the claim.
  3. [Sec. 3.2, Fig. 7] The key transfer assumption, that an MLP trained on circuits with C=12 fixed-angle RZZ gates predicts accurately for the C=0 full circuit, is supported only by the aggregate 1-R^2 metric on test circuits generated from 100-step simulated VQE processes with random initialization. The text does not state whether the test circuits were disjoint from the training circuits or whether a fresh VQE trajectory was used, and the metric is per-observable prediction error rather than the final VQE energy error. Because the actual DL-EM trajectory in Sec. 3.1 uses iterative retraining every 9 steps and may explore angles outside the augmented training distribution, Fig. 7 does not establish that the generalization holds along the actual optimization path.
  4. [Sec. 3.2, Fig. 9] The shot-noise robustness experiment in Fig. 9 is performed only for C=14 circuits, i.e., the trained model is tested on circuits with the same number of cuts as the training data, not on full C=0 circuits. As a result, it does not verify that shot-noise-filtered predictions transfer across the knitting gap; the claim that DL-EM filters shot noise in the VQE setting remains only partially supported.
minor comments (6)
  1. [Fig. 6] The axis labels contain placeholder symbols such as '10□2'; please replace them with proper superscripts.
  2. [Eq. (3) and Sec. 2.1] The symbol O is used both for the sampling overhead in Eq. (3) and for a generic observable in Sec. 2.1; consider renaming one of them to avoid confusion.
  3. [Abstract and Sec. 3.1] The phrase 'below the 1% threshold' is used without specifying whether the discrepancy is per qubit or total; state this explicitly when the threshold is first introduced.
  4. [Sec. 2.1 and Fig. 2] The ansatz description references Fig. 2a, but the caption and the text do not clearly specify the total number of variational parameters (2 x N x P) or how the layer repetition is implemented; please make this explicit.
  5. [Throughout] Several typographical and rendering artifacts appear, including 'vis-´a-vis', 'M. n. Ramo', 'INNOV ATOR', and 'UEF A'; a careful proofreading pass is needed.
  6. [Reproducibility] No code or data availability statement is provided; for a numerical study of this type, a repository link or a detailed reproducibility statement would strengthen the paper.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity: DL-EM is trained on classically simulated ideal/noisy expectation values of partially knitted circuits and tested against exact full-circuit benchmarks; target energies are never used as labels.

full rationale

The derivation chain is self-contained. In Sec. 2.4, the MLPs minimize the MSE loss (Eq. 4) between predictions and ideal expectation values obtained from classical statevector simulations of partially knitted circuits; noisy inputs come from the FakeMelbourneV2 noise model with tunable p_noise (Sec. 2.2). The final energies are computed from MLP predictions of <ZiZi+1> and <Xi> and compared with exact ground-state energies. No ground-state energy or final result is used to fit the model, so the sub-1% accuracy claim is not forced by construction. Generalization across the knitting gap is tested against exactly simulated full circuits with C=0 (Fig. 7), an external benchmark independent of the fitted parameters. Self-citations provide auxiliary techniques — noise scaling [27], training heuristics [49], shot-noise robustness [52] — and are not load-bearing; the tunable-noise model is additionally validated against real IBM hardware in Fig. 4. The remaining concerns (single-instance p_noise=1 result, aggregate 1-R^2 rather than trajectory-level energy errors) are empirical robustness issues, not circularity, and do not raise the circularity score.

Assumptions & free parameters 7 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical entities. Its free parameters are the hyperparameters of the training procedure and noise model; the most significant is the untested generalization assumption that the trained MLPs transfer from partially knitted to full circuits.

free parameters (7)
  • noise scaling factor p_noise = 0.01 to 1
    Tunes the amount of noise in the simulated device; the main results are reported at p_noise=1 and p_noise=0.1.
  • random angle offset standard deviation sigma = 0.05
    Used to augment the training set with small perturbations; chosen by hand (Sec. 2.4).
  • retraining period = 9 VQE steps
    The DL-EM network is retrained every 9 optimization steps; chosen because gate parameters typically shift more than sigma after this many steps (Sec. 2.4).
  • number of cuts C = 12 for main VQE runs; varied from 2 to 16 in Sec. 3.2
    Determines the trade-off between training data cost and fidelity; selected to keep the training circuits representative.
  • training set size Ktrain = 10^4 for Fig. 7 and 8; not explicitly stated for the VQE runs
    Number of training circuits used to fit the MLPs.
  • number of shots S = 10^6 (and 10^2 in Fig. 9)
    Shots per observable for estimating noisy expectation values; also used to study shot-noise robustness.
  • MLP architecture = unspecified
    The number of layers and hidden units is not given, which is a reproducibility gap.
assumptions (4)
  • domain assumption The P=8-layer ansatz of Ref. [39] can represent the ground state of the random 1D Ising model well enough that the noiseless VQE error is negligible.
    The circuit depth is chosen based on the literature; the paper does not benchmark expressibility for the random-coupling case beyond the noiseless convergence seen in Fig. 5.
  • domain assumption The FakeMelbourneV2 noise model, with all error rates scaled by p_noise, faithfully represents the noise of real IBM devices.
    Validated in Fig. 4 against three real devices at one point in time, but device noise is assumed static and homogeneous.
  • standard math The circuit knitting overhead expression (Eq. 3) from Ref. [34] is valid for the considered RZZ cuts.
    This is a known result in circuit knitting; invoked without proof.
  • ad hoc to paper A single MLP per observable generalizes from partially knitted training circuits to the full target circuit.
    This is the central empirical assumption of the method, supported only by the measurements in Sec. 3.2.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Deep-learned error mitigation via partially knitted circuits for the variational quantum eigensolver." pith.science (2026). https://pith.science/paper/O7ROJNLQ

@misc{pith2026250604146,
  author       = {Pith},
  title        = {Pith review of: Deep-learned error mitigation via partially knitted circuits for the variational quantum eigensolver},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/O7ROJNLQ}},
  note         = {Machine review of arXiv:2506.04146}
}
read the original abstract

The variational quantum eigensolver (VQE) is generally regarded as a promising quantum algorithm for near-term noisy quantum computers. However, when implemented with the deep circuits that are in principle required for achieving a satisfactory accuracy, the algorithm is strongly limited by noise. Here, we show how to make VQE functional via a tailored error mitigation technique based on deep learning. Our method employs multilayer perceptrons trained on the fly to predict ideal expectation values from noisy outputs combined with circuit descriptors. Importantly, a circuit knitting technique with partial knitting is adopted to substantially reduce the classical computational cost of creating the training data. We also show that other popular general-purpose quantum error mitigation techniques do not reach comparable accuracies. Our findings highlight the power of deep-learned quantum error mitigation methods tailored to specific circuit families, and of the combined use of variational quantum algorithms and classical deep learning.

Figures

Figures reproduced from arXiv: 2506.04146 by the authors.

Figure 1
Figure 1. Schematic illustration of the VQE process enhanced by the deep learning [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. (a) Scheme of the first layer of the VQE ansatz. The same quantum gates, [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Layout of the IBM Melbourne quantum chip. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Energy discrepancies ∆E/N obtained by classically simulating the FakeMelbourneV2 model, as a function of the noise level pnoise. These simulations are compared with actual experiments performed on the three quantum devices indicated in the legend, denoted by the horizo…
Figure 5
Figure 5. Figure 5: Energy per qubit E/N as a function of the number of VQE optimization steps. We compare the ideal VQE execution (pnoise = 0), the noisy VQE performed with the FakeMelbourneV2 model (pnoise = 1), the noisy VQE mitigated using our DL￾EM method, and a general-purpose EM me…
Figure 6
Figure 6. Figure 6: Discrepancy per qubit ∆E0/N in the estimate of the ground state energy as a function of the noise level pnoise. The two datasets with symbols correspond to noisy VQE performed with COBYLA and with ADAM optimizers. The horizontal lines denote: DM approaches with and wit…
Figure 7
Figure 7. Figure 7: Prediction error 1 − R 2 as a function of the number of cuts C (i.e, the number of not-knitted RZZ gates) in the training circuits, for different numbers of qubits N. The training-set includes 104 circuits, while the test-set includes 103 exactly simulated full quantum…
Figure 8
Figure 8. Figure 8: Prediction error 1 − R 2 as a function of the training-set size Ktrain, for different number of qubits N. The training circuits feature C = 8 cuts, while the test-set includes 103 full quantum circuits, namely, setting C = 0. Training deep neural networks requires suff…
Figure 9
Figure 9. Figure 9: Prediction error 1−R 2 as a function of the training-set size Ktrain. The DL-EM is trained on target values obtained with the circuit knitting technique with C = 14 cuts (i.e. only two knitted gates) and S = 102 shots. The model is tested on quantum circuits with the s…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

66 extracted references · 24 canonical work pages

  1. [1]

    Deep-learned error mitigation via partially knitted circuits for the variational quantum eigensolver

    Introduction The practical utilization of quantum computing is still challenged by the inherent noise and decoherence in the currently available devices [1, 2, 3, 4]. In fact, noise is expected to be a key challenge also in the near and mid-term future [5, 6]. In this context, variational quantum algorithms [7] have emerged as promising applications that ...

    work page Pith review arXiv 2025

  2. [2]

    Variational quantum eigensolver for the 1D quantum Ising model Our testbed for the VQE algorithm is the 1D quantum Ising model with nearest-neighbor random couplings

    Methods 2.1. Variational quantum eigensolver for the 1D quantum Ising model Our testbed for the VQE algorithm is the 1D quantum Ising model with nearest-neighbor random couplings. The Hamiltonian reads: H =− X i JiZiZi+1− h X i Xi, (1) where Zi and Xi are standard Pauli matrices, the couplings are sampled from a uniform distribution in the interval Ji ∈ [...

    work page

  3. [3]

    Testing the VQE algorithm improved by DL-EM Hereafter, the ground-state energy of random quantum Ising chains featuring N = 6 qubits is determined using the VQE algorithm

    Results 3.1. Testing the VQE algorithm improved by DL-EM Hereafter, the ground-state energy of random quantum Ising chains featuring N = 6 qubits is determined using the VQE algorithm. In particular, the performance of a noisy device model, namely the FakeMelbourneV2 model, is tested using different EM methods. The VQE process for a representative instanc...

    work page

  4. [4]

    Hybrid algorithms for quantum simulators

    Conclusions Our work introduces a deep learning-based quantum error mitigation strategy (DL- EM) that significantly improves the performance of variational quantum eigensolvers in the presence of realistic noise. By training deep neural networks on hybrid quantum- classical data, combining noisy expectation values and circuit descriptors, we provide accur...

    work page 2022

  5. [5]

    Preskill

    J. Preskill. Quantum Computing in the NISQ era and beyond. Quantum, 2:79, August 2018. doi:10.22331/q-2018-08-06-79

    doi:10.22331/q-2018-08-06-79 2018

  6. [6]

    Bharti, A

    K. Bharti, A. Cervera-Lierta, T. H. Kyaw, T. Haug, S. Alperin-Lea, A. Anand, M. Degroote, H. Heimonen, J. S. Kottmann, T. Menke, W.-K. Mok, S. Sim, L.-C. Kwek, and A. Aspuru- Guzik. Noisy intermediate-scale quantum algorithms. Rev. Mod. Phys. , 94:015004, Feb

    work page

  7. [7]

    Cerezo, A

    M. Cerezo, A. Arrasmith, R. Babbush, S. C. Benjamin, S. Endo, K. Fujii, J. R. McClean, K. Mitarai, X. Yuan, L. Cincio, and P. J. Coles. Variational quantum algorithms. Nature Reviews Physics , 3(9):625–644, August 2021. URL: http://dx.doi.org/10.1038/ s42254-021-00348-9 , doi:10.1038/s42254-021-00348-9

    doi:10.1038/s42254-021-00348-9 2021

  8. [8]

    Stilck Fran¸ ca and R

    D. Stilck Fran¸ ca and R. Garc ´ ıa-Patr´ on. Limitations of optimization algorithms on noisy quantum devices. Nature Physics, 17(11):1221–1227, Nov 2021. doi:10.1038/s41567-021-01356-3

    doi:10.1038/s41567-021-01356-3 2021

Show all 66 references
  1. [9]

    Y. Zhou, E. M. Stoudenmire, and X. Waintal. What limits the simulation of quantum computers? Phys. Rev. X , 10:041038, Nov 2020. URL: https://link.aps.org/doi/10.1103/PhysRevX. 10.041038, doi:10.1103/PhysRevX.10.041038

    2020 doi

  2. [10]

    Zhang, G

    R. Zhang, G. Wang, and P. Johnson. Computing Ground State Properties with Early Fault-Tolerant Quantum Computers. Quantum, 6:761, July 2022. doi:10.22331/ q-2022-07-11-761

    2022

  3. [11]

    Preskill

    J. Preskill. Beyond NISQ: The megaquop machine. ACM Transactions on Quantum Computing , 6(3):1–7, April 2025. URL: http://dx.doi.org/10.1145/3723153, doi:10.1145/3723153

    2025 doi

  4. [12]

    W. Saib, P. Wallden, and I. Akhalwaya. The effect of noise on the performance of variational algorithms for quantum chemistry. In 2021 IEEE International Conference on Quantum Computing and Engineering (QCE) , pages 42–53, 2021. doi:10.1109/QCE52317.2021.00020

    2021

  5. [13]

    Tilly, H

    J. Tilly, H. Chen, S. Cao, D. Picozzi, K. Setia, Y. Li, E. Grant, L. Wossnig, I. Rungger, G. H. Booth, and J. Tennyson. The variational quantum eigensolver: A review of methods and best practices. Physics Reports, 986:1–128, November 2022. URL: http://dx.doi.org/10.1016/ j.phy...

    2022 doi

  6. [14]

    Bentellis, A

    A. Bentellis, A. Matic-Flierl, C. B. Mendl, and J. M. Lorenz. Benchmarking the variational quantum eigensolver using different quantum hardware. In 2023 IEEE International Conference Deep-learned error mitigation via partially knitted circuits for the variational quantum eigen...

    2023

  7. [15]

    Fontana, N

    E. Fontana, N. Fitzpatrick, D. M. n. Ramo, R. Duncan, and I. Rungger. Evaluating the noise resilience of variational quantum algorithms. Phys. Rev. A, 104:022403, Aug 2021. URL: https: //link.aps.org/doi/10.1103/PhysRevA.104.022403, doi:10.1103/PhysRevA.104.022403

    2021 doi

  8. [16]

    Singkanipa and D

    P. Singkanipa and D. A. Lidar. Beyond unital noise in variational quantum algorithms: noise- induced barren plateaus and limit sets. Quantum, 9:1617, January 2025. doi:10.22331/ q-2025-01-30-1617

    2025

  9. [17]

    Kandala, K

    A. Kandala, K. Temme, A. D. C´ orcoles, A. Mezzacapo, J. M. Chow, and J. M. Gambetta. Error mitigation extends the computational reach of a noisy quantum processor. Nature, 567(7749):491–495, March 2019. URL: http://dx.doi.org/10.1038/s41586-019-1040-7 , doi:10.1038/s41586-019-1040-7

    2019 doi

  10. [18]

    Brandhofer, S

    S. Brandhofer, S. Devitt, and I. Polian. Error analysis of the variational quantum eigensolver algorithm. In 2021 IEEE/ACM International Symposium on Nanoscale Architectures (NANOARCH), pages 1–6, 2021. doi:10.1109/NANOARCH53687.2021.9642249

    2021

  11. [19]

    Z. Cai, R. Babbush, S. C. Benjamin, S. Endo, W. J. Huggins, Y. Li, J. R. McClean, and T. E. O’Brien. Quantum error mitigation. Reviews of Modern Physics , 95(4), December 2023. URL: http://dx.doi.org/10.1103/RevModPhys.95.045005, doi:10.1103/revmodphys.95. 045005

    2023 doi

  12. [20]

    Temme, S

    K. Temme, S. Bravyi, and J. M. Gambetta. Error mitigation for short-depth quantum circuits. Phys. Rev. Lett., 119:180509, Nov 2017. URL: https://link.aps.org/doi/10.1103/ PhysRevLett.119.180509, doi:10.1103/PhysRevLett.119.180509

    2017 doi

  13. [21]

    Li and S

    Y. Li and S. C. Benjamin. Efficient variational quantum simulator incorporating active error minimization. Phys. Rev. X, 7:021050, Jun 2017. URL: https://link.aps.org/doi/10.1103/ PhysRevX.7.021050, doi:10.1103/PhysRevX.7.021050

    2017 doi

  14. [22]

    S. Wang, P. Czarnik, A. Arrasmith, M. Cerezo, L. Cincio, and P. J. Coles. Can Error Mitigation Improve Trainability of Noisy Variational Quantum Algorithms? Quantum, 8:1287, March

  15. [23]

    S. Endo, S. C. Benjamin, and Y. Li. Practical quantum error mitigation for near-future applications. Phys. Rev. X , 8:031027, Jul 2018. URL: https://link.aps.org/doi/10.1103/ PhysRevX.8.031027, doi:10.1103/PhysRevX.8.031027

    2018 doi

  16. [24]

    Zhang, Y

    S. Zhang, Y. Lu, K. Zhang, W. Chen, Y. Li, J.-N. Zhang, and K. Kim. Error-mitigated quantum gates exceeding physical fidelities in a trapped-ion system. Nature Communications, 11(1), January 2020. URL: http://dx.doi.org/10.1038/s41467-020-14376-z , doi:10. 1038/s41467-020-14376-z

    2020 doi

  17. [25]

    van den Berg, Z

    E. van den Berg, Z. K. Minev, A. Kandala, and K. Temme. Probabilistic error cancellation with sparse Pauli–Lindblad models on noisy quantum processors. Nature Physics, 19(8):1116–1121, May 2023. URL: http://dx.doi.org/10.1038/s41567-023-02042-2 , doi: 10.1038/s41567-023-02042-2

    2023 doi

  18. [26]

    Kurita, H

    T. Kurita, H. Qassim, M. Ishii, H. Oshima, S. Sato, and J. Emerson. Synergetic quantum error mitigation by randomized compiling and zero-noise extrapolation for the variational quantum eigensolver. Quantum, 7:1184, November 2023. doi:10.22331/q-2023-11-20-1184

    2023 doi

  19. [27]

    Cantori, A

    S. Cantori, A. Mari, D. Vitali, and S. Pilati. Synergy between noisy quantum computers and scalable classical deep learning for quantum error mitigation. EPJ Quantum Technology , 11(1):45, Jul 2024. doi:10.1140/epjqt/s40507-024-00256-8

    2024 doi

  20. [28]

    A. Lowe, M. H. Gordon, P. Czarnik, A. Arrasmith, P. J. Coles, and L. Cincio. Unified approach to data-driven quantum error mitigation. Phys. Rev. Res., 3:033098, Jul 2021. URL: https:// link.aps.org/doi/10.1103/PhysRevResearch.3.033098, doi:10.1103/PhysRevResearch. 3.033098

    2021 doi

  21. [29]

    Y. Quek, D. Stilck Fran¸ ca, S. Khatri, J. J. Meyer, and J. Eisert. Exponentially tighter bounds on limitations of quantum error mitigation. Nature Physics, 20(10):1648–1658, July 2024. URL: http://dx.doi.org/10.1038/s41567-024-02536-7 , doi:10.1038/s41567-024-02536-7

    2024 doi

  22. [30]

    Takagi, S

    R. Takagi, S. Endo, S. Minagawa, and M. Gu. Fundamental limits of quantum error mitigation. npj Quantum Information , 8(1), September 2022. URL: http://dx.doi.org/10.1038/ s41534-022-00618-z , doi:10.1038/s41534-022-00618-z . Deep-learned error mitigation via partially knitted...

    2022 doi

  23. [31]

    Jiang, J

    T. Jiang, J. Rogers, M. S. Frank, O. Christiansen, Y.-X. Yao, and N. Lanat` a. Error mitigation in variational quantum eigensolvers using tailored probabilistic machine learning. Phys. Rev. Res., 6:033069, Jul 2024. URL: https://link.aps.org/doi/10.1103/PhysRevResearch.6.03306...

    2024 doi

  24. [32]

    Czarnik, A

    P. Czarnik, A. Arrasmith, P. J. Coles, and L. Cincio. Error mitigation with Clifford quantum- circuit data. Quantum, 5:592, November 2021. doi:10.22331/q-2021-11-26-592

    2021 doi

  25. [33]

    S. H. Sack and D. J. Egger. Large-scale quantum approximate optimization on nonplanar graphs with machine learning noise mitigation. Phys. Rev. Res., 6:013223, Mar 2024. URL: https:// link.aps.org/doi/10.1103/PhysRevResearch.6.013223, doi:10.1103/PhysRevResearch. 6.013223

    2024 doi

  26. [34]

    Mitarai and K

    K. Mitarai and K. Fujii. Constructing a virtual two-qubit gate by sampling single-qubit operations. New Journal of Physics , 23(2):023021, feb 2021. URL: https://dx.doi.org/ 10.1088/1367-2630/abd7bc, doi:10.1088/1367-2630/abd7bc

    2021 doi

  27. [35]

    M. Liao, Y. Zhu, G. Chiribella, and Y. Yang. Noise-agnostic quantum error mitigation with data augmented neural models. npj Quantum Information , 11(1):8, Jan 2025. doi:10.1038/ s41534-025-00960-y

    2025

  28. [36]

    H. Liao, D. S. Wang, I. Sitdikov, C. Salcedo, A. Seif, and Z. K. Minev. Machine learning for practical quantum error mitigation. Nature Machine Intelligence , 6(12):1478–1486, Dec 2024. doi:10.1038/s42256-024-00927-2

    2024 doi

  29. [37]

    E. R. Bennewitz, F. Hopfmueller, B. Kulchytskyy, J. Carrasquilla, and P. Ronagh. Neural error mitigation of near-term quantum simulations. Nature Machine Intelligence , 4(7):618–624, Jul

  30. [38]

    doi:10.1038/s42256-022-00509-0

    doi

  31. [39]

    X.-Y. Xu, X. Xue, T. Chen, C. Ding, T. Li, H. Zhou, H.-L. Huang, and W.-S. Bao. Physics- inspired machine learning for quantum error mitigation, 2025. URL: https://arxiv.org/abs/ 2501.04558, arXiv:2501.04558

    2025 arXiv

  32. [40]

    Arute, K

    F. Arute, K. Arya, R. Babbush, D. Bacon, J. C. Bardin, R. Barends, R. Biswas, S. Boixo, F. G. S. L. Brandao, D. A. Buell, B. Burkett, Y. Chen, Z. Chen, B. Chiaro, R. Collins, W. Courtney, Deep-learned error mitigation via partially knitted circuits for the variational quantum ...

    2019

  33. [41]

    Y. Wang, B. Qi, C. Ferrie, and D. Dong. Trainability enhancement of parameterized quantum circuits via reduced-domain parameter initialization. Phys. Rev. Appl. , 22:054005, Nov 2024. URL: https://link.aps.org/doi/10.1103/PhysRevApplied.22.054005, doi: 10.1103/PhysRevApplied.22.054005

    2024 doi

  34. [42]

    D. P. Kingma and J. Ba. Adam: A method for stochastic optimization. 2017. arXiv:1412.6980

    2017 arXiv

  35. [43]

    J. Li, X. Yang, X. Peng, and C.-P. Sun. Hybrid quantum-classical approach to quantum optimal control. Phys. Rev. Lett., 118:150503, Apr 2017. URL: https://link.aps.org/doi/10.1103/ PhysRevLett.118.150503, doi:10.1103/PhysRevLett.118.150503

    2017 doi

  36. [44]

    Mitarai, M

    K. Mitarai, M. Negoro, M. Kitagawa, and K. Fujii. Quantum circuit learning. Phys. Rev. A, 98:032309, Sep 2018. URL: https://link.aps.org/doi/10.1103/PhysRevA.98.032309, doi:10.1103/PhysRevA.98.032309

    2018 doi

  37. [45]

    Virtanen, R

    P. Virtanen, R. Gommers, T. E. Oliphant, M. Haberland, T. Reddy, D. Cournapeau, E. Burovski, P. Peterson, W. Weckesser, J. Bright, S. J. van der Walt, M. Brett, J. Wilson, K. J. Millman, N. Mayorov, A. R. J. Nelson, E. Jones, R. Kern, E. Larson, C. J. Carey, ˙I. Polat, Y. Feng...

    2020

  38. [46]

    W. W. Ho and T. H. Hsieh. Efficient variational simulation of non-trivial quantum states. SciPost Phys. , 6:029, 2019. URL: https://scipost.org/10.21468/SciPostPhys.6.3.029, doi:10.21468/SciPostPhys.6.3.029

    2019 doi

  39. [47]

    Piveteau and D

    C. Piveteau and D. Sutter. Circuit knitting with classical communication. IEEE Transactions on Information Theory, 70(4):2734–2745, 2024. doi:10.1109/TIT.2023.3310797

    2024

  40. [48]

    Gentinetta, F

    G. Gentinetta, F. Metz, and G. Carleo. Overhead-constrained circuit knitting for variational quantum dynamics. Quantum, 8:1296, March 2024. doi:10.22331/q-2024-03-21-1296

    2024 doi

  41. [49]

    A. Khan, B. K. Clark, and N. M. Tubman. Pre-optimizing variational quantum eigensolvers with tensor networks, 2023. URL: https://arxiv.org/abs/2310.12965, arXiv:2310.12965

    2023 arXiv

  42. [50]

    H. Yu, Y. Zhao, and T.-C. Wei. Simulating large-size quantum spin chains on cloud-based superconducting quantum computers. Phys. Rev. Res. , 5:013183, Mar 2023. URL: https:// link.aps.org/doi/10.1103/PhysRevResearch.5.013183, doi:10.1103/PhysRevResearch. 5.013183

    2023 doi

  43. [51]

    Rosenberg, P

    E. Rosenberg, P. Ginsparg, and P. L. McMahon. Experimental error mitigation using linear rescaling for variational quantum eigensolving with up to 20 qubits. Quantum Science and Technology, 7(1):015024, jan 2022. URL: https://dx.doi.org/10.1088/2058-9565/ac3b37, doi:10.1088/20...

    2022 doi

  44. [52]

    Urbanek, B

    M. Urbanek, B. Nachman, V. R. Pascuzzi, A. He, C. W. Bauer, and W. A. de Jong. Mitigating depolarizing noise on quantum computers with noise-estimation circuits. Phys. Rev. Lett., 127:270502, Dec 2021. URL: https://link.aps.org/doi/10.1103/PhysRevLett.127. 270502, doi:10.1103/...

    2021 doi

  45. [53]

    Javadi-Abhari, M

    A. Javadi-Abhari, M. Treinish, K. Krsulich, C. J. Wood, J. Lishman, J. Gacon, S. Martiel, P. D. Nation, L. S. Bishop, A. W. Cross, B. R. Johnson, and J. M. Gambetta. Quantum computing with Qiskit, 2024. arXiv:2405.08810, doi:10.48550/arXiv.2405.08810

    2024 doi arXiv

  46. [54]

    Q. A. team and collaborators. qsim, September 2020. doi:10.5281/zenodo.4023103

    2020 doi

  47. [55]

    LaRose, A

    R. LaRose, A. Mari, S. Kaiser, P. J. Karalekas, A. A. Alves, P. Czarnik, M. E. Mandouh, M. H. Deep-learned error mitigation via partially knitted circuits for the variational quantum eigensolver19 Gordon, Y. Hindy, A. Robertson, P. Thakre, M. Wahl, D. Samuel, R. Mistri, M. Tre...

    2022

  48. [56]

    Cantori and S

    S. Cantori and S. Pilati. Challenges and opportunities in the supervised learning of quantum circuit outputs, 2024. URL: https://arxiv.org/abs/2402.04992, arXiv:2402.04992

    2024 arXiv

  49. [57]

    A. Mari, T. R. Bromley, J. Izaac, M. Schuld, and N. Killoran. Transfer learning in hybrid classical- quantum neural networks. Quantum, 4:340, October 2020. doi:10.22331/q-2020-10-09-340

    2020 doi

  50. [58]

    R. Zen, L. My, R. Tan, F. H´ ebert, M. Gattobigio, C. Miniatura, D. Poletti, and S. Bressan. Transfer learning for scalability of neural-network quantum states. Physical Review E , 101(5), May

  51. [60]

    Cantori, D

    S. Cantori, D. Vitali, and S. Pilati. Supervised learning of random quantum circuits via scalable neural networks. Quantum Science and Technology , 8(2):025022, mar 2023. URL: https://dx.doi.org/10.1088/2058-9565/acc4e2, doi:10.1088/2058-9565/acc4e2

    2023 doi

  52. [61]

    Saraceni, S

    N. Saraceni, S. Cantori, and S. Pilati. Scalable neural networks for the efficient learning of disordered quantum systems. Phys. Rev. E , 102:033301, Sep 2020. URL: https://link.aps. org/doi/10.1103/PhysRevE.102.033301, doi:10.1103/PhysRevE.102.033301

    2020 doi

  53. [64]

    Pilati and P

    S. Pilati and P. Pieri. Supervised machine learning of ultracold atoms with speckle disorder. Scientific Reports , 9(1), April 2019. URL: http://dx.doi.org/10.1038/ s41598-019-42125-w , doi:10.1038/s41598-019-42125-w

    2019 doi

  54. [65]

    Rolnick, A

    D. Rolnick, A. Veit, S. Belongie, and N. Shavit. Deep learning is robust to massive label noise,

  55. [2018]

    URL: https://arxiv.org/abs/1705.10694, arXiv:1705.10694

    arXiv

  56. [2020]

    101.053301

    URL: http://dx.doi.org/10.1103/PhysRevE.101.053301, doi:10.1103/physreve. 101.053301

    doi

  57. [2022]

    URL: https://link.aps.org/doi/10.1103/RevModPhys.94.015004, doi:10.1103/ RevModPhys.94.015004

    doi

  58. [2024]

    doi:10.22331/q-2024-03-14-1287

    2024 doi

Pith tools

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