Pith. sign in

REVIEW 4 major objections 3 minor 27 references

A six-parameter Gaussian recovery layer can undo photonic loss at the observable level.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

Trainable Gaussian rotations and displacements can cancel loss-induced shifts of quadrature means in simulated photonic circuits, with modest robustness gains against modeled phase jitter.

T0 review reviewed 2026-08-03 challenge →

load-bearing objection The headline results are an artifact of a first-moment loss: a single displacement can zero the error, and the paper's own single-mode baseline proves it. the 4 major comments →

arxiv 2512.23776 v1 pith:3MDQCGED submitted 2025-12-29 quant-ph

DifGa: Differentiable Error Mitigation for Multi-Mode Gaussian and Non-Gaussian Noise in Quantum Photonic Circuits

Dennis Delali Kwesi Wayo , Rodrigo Alves Dias , Leonardo Goliatt , Sven Groppe This is my paper
classification quant-ph
keywords continuous-variable quantum photonicsdifferentiable error mitigationGaussian noisenon-Gaussian phase noisequadrature observablesMonte Carlo mixtureautomatic differentiationmulti-mode bosonic circuits
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

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 sets out to show that error mitigation for continuous-variable photonic circuits does not require non-Gaussian code states or full state correction: a shallow Gaussian recovery layer, trained by gradient descent, can restore the measured quadrature expectations of a noisy circuit. The central numerical claim is that under pure beam-splitter loss with transmissivity η≥0.5, the optimized recovery pushes reconstruction error below machine precision (10^-30), and under weak non-Gaussian phase jitter modeled by Monte-Carlo averaging, noise-aware training keeps error more than an order of magnitude lower than Gaussian-trained recovery. If this holds, near-term photonic hardware could suppress loss- and phase-noise bias in homodyne observables using only operations it already supports—phase rotations, displacements, and vacuum ancillas. The care a reader should take is that the result is specifically about first-moment observables, not full quantum state recovery, and the paper is explicit that it does not challenge no-go results on Gaussian error correction.

Core claim

The discovery is that a six-parameter Gaussian operation—local phase rotations and displacements on a signal mode and one ancilla—can be optimized end-to-end to act as an observable-level error mitigator. In analytic Gaussian-state simulation, the optimized layer reduces the mean-squared error in the signal mode's position- and momentum-like quadratures from about 10^-1 to below 10^-30 for transmissivities η≥0.5, and for non-Gaussian phase jitter included as a differentiable mixture of random phase kicks, explicitly training under that noise yields a mitigation that stays near 10^-3 at jitter amplitudes where a Gaussian-trained layer climbs to 10^-2–10^-1. The paper interprets this as task-s

What carries the argument

The load-bearing object is the trainable recovery layer R(θ)=∏_{j=0}^1 D_j(β_j) R_j(φ_j), six parameters acting as local displacements and phase rotations on the signal and ancilla modes, appended to a three-mode circuit (signal, ancilla, environment) in which loss is modeled by a beam-splitter interaction with a vacuum mode. The second essential ingredient is the differentiable Monte-Carlo mixture: non-Gaussian phase noise is implemented by averaging expectation values over K random Gaussian phase kicks, so each sample is a Gaussian circuit and gradients can be back-propagated through the average. The objective that ties them together is a quadratic loss on the signal-mode quadratures, targ

Load-bearing premise

The load-bearing premise is that the noise channel is known exactly and is used both to generate the noisy data and to train the recovery layer: any mismatch between the simulated beam-splitter loss plus Gaussian phase jitter and real photonic hardware would break the claimed hardware-compatible mitigation.

What would settle it

Measure the mitigated quadrature error on a photonic chip after training the recovery layer in simulation under one noise model, then apply it under a different independently characterized noise model; if the error at η=0.55 is not near zero, the known-channel assumption fails. For a purely numerical check, train on Gaussian phase jitter but evaluate on uniform or heavy-tailed phase noise of the same variance and compare the error growth to the paper's δ-curves.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • At transmissivity η≥0.5, the six-parameter recovery layer drives quadrature reconstruction error below 10^-30 in the Gaussian-loss simulation, meaning loss no longer biases first-moment measurements in that modeled regime.
  • Training under Monte-Carlo phase jitter extends the effective mitigation region: at jitter amplitudes 0.42–0.70, the noise-aware layer keeps error at 3×10^-3 to 5×10^-2, versus 1.3×10^-2 to 8.6×10^-2 for Gaussian-trained recovery, so noise-aware training is the right recipe when phase noise is present.
  • Adding ancilla modes helps twice: passive noise redistribution lowers baseline error from 0.171 to 0.100 at η=0.55, and active mitigation with ancillas reaches about 10^-41, demonstrating that multi-mode resources are structurally necessary.
  • Computational cost grows linearly with Monte-Carlo sample count; 8–16 samples already give solid mitigation under 10 seconds per training run, so the method is practical for repeated retraining and calibration.
  • Because only first moments are targeted, the method is explicitly complementary to, rather than in conflict with, no-go results that forbid Gaussian correction of Gaussian noise at the state level.

Where Pith is reading between the lines

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

  • The same differentiable-mixture trick could be applied to other non-Gaussian channels—random displacements or photon-number-dependent losses, for example—as long as each sample remains Gaussian; this would test how much of the improvement is specific to phase jitter rather than generic to averaging.
  • Since the recovery layer only matches first moments, a natural next step is to add a loss on second moments (covariance entries); the paper's own framing suggests this would move toward state-level correction and hit the no-go boundary, so the expected failure point is known in advance.
  • A practical consequence the paper leaves implicit: the learned parameters for fixed hardware could be used as an automated calibration routine, periodically retraining the six parameters as the noise environment drifts, with the measured runtime scaling indicating this is feasible in practice.
  • The strong performance starting from zero initialization hints that the optimization landscape is nearly convex for this observable-matching problem; if confirmed, the method might admit closed-form or one-shot least-squares solutions rather than iterative gradient descent.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 3 minor

Summary. The paper introduces DifGa, a differentiable error-mitigation framework for continuous-variable photonic circuits. It models Gaussian loss via a beam-splitter interaction with a vacuum environment and weak non-Gaussian phase noise through a Monte-Carlo mixture of random phase rotations. A trainable Gaussian recovery layer consisting of local phase rotations and displacements is appended to the circuit and optimized by gradient descent to minimize a quadratic loss on the signal-mode first moments. The authors report near-machine-precision suppression of this loss under Gaussian loss for moderate transmissivities, a comparison with single-mode baselines, and improved robustness when training is exposed to non-Gaussian phase noise.

Significance. The central numerical results, if valid, would support a hardware-compatible, all-Gaussian method for suppressing noise-induced errors in photonic quadrature measurements. However, the loss function used throughout is a first-moment quadratic error, and the recovery layer includes a signal-mode displacement. As the paper's own single-mode baseline demonstrates, a single displacement can make this loss zero exactly, independent of ancilla modes or entangling operations. The reported 'reconstruction errors' are therefore an artifact of moment matching, and the claimed necessity of multi-mode ancillary resources is contradicted by the same data. The paper's differentiable Monte-Carlo noise model and reproducible code are useful contributions, but they do not support the main claims about error mitigation.

major comments (4)
  1. [Section II.C, Eq. (5) and Section II.E, Eq. (9)] The loss L(θ) depends only on the first moments ⟨x0⟩ and ⟨p0⟩. The recovery layer includes a displacement D0(β0) on the signal mode. For any noisy state with finite mean, choosing β0 to shift the mean to the ideal value makes L exactly zero, irrespective of the ancilla, the entangler, or the remaining parameters. Thus the near-machine-precision errors in Fig. 2 and Sec. III.A are a trivial consequence of the displacement degree of freedom, not evidence of a learned inverse map. The paper explicitly notes the single-mode baseline reaches ~1e-44 and calls it 'trivial nature of moment matching' (Sec. III.B), but then concludes that 'ancillary-assisted architectures are structurally essential' — a direct contradiction.
  2. [Section III.D, Fig. 5] The non-Gaussian comparison is also explained by the same displacement effect. At a given δ, the Monte-Carlo averaged state has a δ-dependent first moment. A recovery trained at δ=0 fixes a displacement valid for δ=0; NG training adjusts the displacement to compensate the MC-averaged mean shift. Since the loss remains first-moment-only, the reported 'more than an order of magnitude improvement' is a calibration effect, not evidence that Gaussian recovery meaningfully mitigates non-Gaussian noise beyond the first moment. No second-moment or higher-order observable is evaluated.
  3. [Section III.A, Eq. (9), Figs. 2–7] The metric plotted as 'reconstruction error' is the training loss itself. Reporting that the optimized training loss is near zero is circular and provides no evidence of generalization or actual noise mitigation. A meaningful evaluation would require a held-out test set, an independent metric (e.g., state fidelity or second-moment/covariance error), or at least separation between training and evaluation noise realizations.
  4. [Sections II.D and III.D] The non-Gaussian noise model is assumed known exactly and is used both to generate noisy data and to train the recovery layer. The paper claims 'hardware-compatible' mitigation (Introduction and Conclusion), but no robustness test against channel mismatch is provided. For example, varying the jitter distribution, the correlation factor κ, or the loss model at evaluation would be needed to support that claim. As written, the results are simulation-level and the generalization claim is unsupported.
minor comments (3)
  1. [Throughout] Typographical errors: 'throgh' (Sec. III.B), 'extensins' (Sec. III.E), 'precisio' (Sec. III.F), and the garbled sentence 'the quadrature operator =sˆx j' in Sec. II. The paper would benefit from careful proofreading.
  2. [Section II.B] The ideal state in Eq. (3) is a product of squeezed and displaced states followed by a beam splitter. The acronyms in the circuit diagram (Fig. 1) are not all defined in the caption; please define 'S', 'D', 'B', and 'R' in the text or figure.
  3. [Section III.C] The phase-noise landscape (Fig. 4) uses K=16 Monte-Carlo samples, but the parameter dynamics (Sec. III.G) uses K=32. The effect of sample size on the reported error is not discussed; a small K may introduce estimator noise that is not addressed.

Circularity Check

2 steps flagged

Central 'near-machine-precision' result is the minimized loss L itself: a signal displacement can zero Eq. (9) exactly. The paper's own single-mode baseline reaches ~1e-44 and calls it 'the trivial nature of moment matching,' yet the multi-mode 'structurally essential' conclusion is drawn from that same triviality.

specific steps
  1. fitted input called prediction [Sec. II.C Eq. (5); Sec. II.E Eq. (9); Sec. III.A; Abstract]
    "The recovery operation is parameterized shown in Eq. (5) as R(θ) = ∏_{j=0}^{1} D_j(β_j) R_j(φ_j), where ... β_j ∈ C is a complex displacement. ... the loss function is defined as (see Eq. (9)) L(θ) = (⟨x0⟩ideal − ⟨x0⟩noisy(θ))^2 + (⟨p0⟩ideal − ⟨p0⟩noisy(θ))^2."

    L contains only the two first moments of the signal mode, while Eq. (5) gives the recovery layer a complex displacement D0(β0) on that mode. The two real components of β0 can always be chosen so the noisy first moments equal the ideal first moments, making L exactly zero for any loss/noise strength and for any covariance. The reported errors <10−30 (and even ~10−38 at η=0.30) are therefore the optimizer locating this displacement, not a physical suppression effect. The paper itself calls the single-mode version of this 'the trivial nature of moment matching.'

  2. fitted input called prediction [Sec. III.D, Fig. 5]
    "we contrast a recovery layer trained exclusively on Gaussian loss with one trained explicitly in the presence of non-Gaussian phase jitter. Figure 5 reports the final quadrature reconstruction error as a function of the phase jitter amplitude δ, evaluated under non-Gaussian noise for both training strategies. ... At the largest tested value, δ=0.70, the Gaussian-trained recovery reaches an error of 8.6×10−2, whereas the NG-trained recovery remains significantly lower at 5.3×10−2."

    The 'robust generalization' advantage is measured with the same first-moment loss L and the same Monte-Carlo phase-jitter model used to train the NG recovery (δsig=δ, δanc=0.6δ, K=16). Because the only trainable effect needed to reduce L under phase noise is a signal displacement compensating the MC-averaged first-moment shift, the NG-trained parameters are fitted to the test distribution itself. The reported order-of-magnitude improvement is thus an in-distribution fit, not an independent prediction about non-Gaussian mitigation.

full rationale

The central quantitative claims reduce to moment matching. L in Eq. (9) is the only metric plotted; the recovery layer in Eq. (5) contains a signal displacement, so the exact zero of L is reachable trivially; the <10^{-30} values are properties of the optimizer and loss definition, not of the multi-mode architecture. Sec. III.B reports the single-mode mitigated baseline at ~10^{-44} and calls it 'the trivial nature of moment matching,' yet the paper then concludes that 'ancillary-assisted architectures are ... structurally essential' — a conclusion unsupported by that same evidence. The non-Gaussian comparison in Sec. III.D is likewise in-distribution: the NG-trained model is evaluated on the same Monte-Carlo phase-noise model and the same first-moment loss on which it was trained, so its 'more than an order of magnitude' advantage is a calibration/fit effect rather than demonstrated generalization. No load-bearing self-citation chain or imported uniqueness theorem appears; the circularity is in the fitted-loss construction, not in the references. The runtime benchmarks and PennyLane implementation are independent, but they do not rescue the central error-suppression claim. Score 8: the headline 'near-machine-precision suppression' is forced by the definition of L and the recovery parameterization.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

The central claim rests on standard Gaussian-state formalism and specific noise models chosen by hand. No new physical entities are introduced, but the circuit parameters and noise correlation factor are arbitrary and not varied in sensitivity tests.

free parameters (4)
  • Circuit input parameters (r_s, φ_s, α, r_a, φ_a, θ, φ) = (0.60, 0.30, 0.80, 0.40, 0.10, 0.70, 0.20)
    Chosen by hand without sensitivity analysis; the reported error magnitudes depend on these values.
  • Phase-jitter correlation factor κ = 0.6
    Set to 0.6 (via δ_anc = 0.6δ) without physical justification.
  • Monte-Carlo sample count K = 16
    Small number of samples; no convergence check or variance reporting.
  • Optimization hyperparameters = lr=0.06, steps=30-60, zero init
    Chosen manually; no grid search or robustness shown.
axioms (5)
  • standard math Gaussian states are fully characterized by first and second moments
    Standard result in CV quantum optics; used throughout the methodology.
  • domain assumption Optical loss is modeled as a beam-splitter interaction with a vacuum environment
    Standard model, but assumes no other loss channels or frequency dependence; used in Eq. (4).
  • domain assumption Non-Gaussian phase noise is representable as a mixture of random Gaussian phase kicks
    Approximation; used in Eq. (7)-(8). The fidelity of this model to physical phase diffusion is not discussed.
  • domain assumption The noise model used in training is identical to the noise used in evaluation
    Training and evaluation use the same beam-splitter and MC models; no noise-robustness analysis.
  • domain assumption The PennyLane default.gaussian backend exactly simulates the physics
    No experimental verification; all results are simulation-only.

reviewed 2026-08-03 · how reviews work

0 comments
Cite this review

Pith. "Pith review of DifGa: Differentiable Error Mitigation for Multi-Mode Gaussian and Non-Gaussian Noise in Quantum Photonic Circuits." pith.science (2026). https://pith.science/paper/3MDQCGED

@misc{pith2026251223776,
  author       = {Pith},
  title        = {Pith review of: DifGa: Differentiable Error Mitigation for Multi-Mode Gaussian and Non-Gaussian Noise in Quantum Photonic Circuits},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3MDQCGED}},
  note         = {Machine review of arXiv:2512.23776}
}
Share X Bluesky LinkedIn Reddit HN
abstract

We introduce DifGa, a fully differentiable error-mitigation framework for continuous-variable (CV) quantum photonic circuits operating under Gaussian loss and weak non-Gaussian noise. The approach is demonstrated using analytic simulations with the default.gaussian backend of PennyLane, where quantum states are represented by first and second moments and optimized end-to-end via automatic differentiation. Gaussian loss is modeled as a beam splitter interaction with an environmental vacuum mode of transmissivity $\eta \in [0.3,0.95]$, while non-Gaussian phase noise is incorporated through a differentiable Monte-Carlo mixture of random phase rotations with jitter amplitudes $\delta \in [0,0.7]$. The core architecture employs a multi-mode Gaussian circuit consisting of a signal, ancilla, and environment mode. Input states are prepared using squeezing and displacement operations with parameters $(r_s,\varphi_s,\alpha)=(0.60,0.30,0.80)$ and $(r_a,\varphi_a)=(0.40,0.10)$, followed by an entangling beam splitter with angles $(\theta,\phi)=(0.70,0.20)$. Error mitigation is achieved by appending a six-parameter trainable Gaussian recovery layer comprising local phase rotations and displacements, optimized by minimizing a quadratic loss on the signal-mode quadratures $\langle \hat{x}_0\rangle$ and $\langle \hat{p}_0\rangle$ using gradient descent with fixed learning rate $0.06$ and identical initialization across experiments. Under pure Gaussian loss, the optimized recovery suppresses reconstruction error to near machine precision ($<10^{-30}$) for moderate loss ($\eta \ge 0.5$). When non-Gaussian phase noise is present, noise-aware training using Monte Carlo averaging yields robust generalization, reducing error by more than an order of magnitude compared to Gaussian-trained recovery at large phase jitter. Runtime benchmarks confirm linear scaling with the number of Monte Carlo samples.

Figures

Figures reproduced from arXiv: 2512.23776 by Dennis Delali Kwesi Wayo, Leonardo Goliatt, Rodrigo Alves Dias, Sven Groppe.

Figure 1
Figure 1. Figure 1: Overview of DifGa, a fully differentiable, Gaussian-only error mitigation framework for continuous-variable photonic [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 3
Figure 3. Figure 3: Single-mode versus multi-mode mitigation at fixed loss [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 5
Figure 5. Figure 5: Generalization under non-Gaussian phase noise. Com [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Critical phase-noise threshold analysis. Error growth [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Recovery parameter dynamics under NG training. [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Runtime scaling with Monte-Carlo samples [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: GKP-based logical quantum error correction following the framework of Noh [PITH_FULL_IMAGE:figures/full_fig_p010_10.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

27 extracted references · 3 linked inside Pith

  1. [1]

    Scal- ing and networking a modular photonic quantum computer,

    H. Aghaee Rad, T. Ainsworth, R. Alexander, B. Altieri, M. Askarani, R. Baby, L. Banchi, B. Baragiola, J. Bourassa, R. Chadwicket al., “Scal- ing and networking a modular photonic quantum computer,”Nature, vol. 638, no. 8052, pp. 912–919, 2025

    2025

  2. [2]

    Scalable photonic quantum technologies,

    H. Wang, T. C. Ralph, J. J. Renema, C.-Y . Lu, and J.-W. Pan, “Scalable photonic quantum technologies,”Nature Materials, vol. 24, no. 12, pp. 1883–1897, 2025

    2025

  3. [3]

    Quantum computing and machine learning on an integrated photonics platform,

    H. Zhu, H. Lin, S. Wu, W. Luo, H. Zhang, Y . Zhan, X. Wang, A. Liu, and L. C. Kwek, “Quantum computing and machine learning on an integrated photonics platform,”Information, vol. 15, no. 2, p. 95, 2024

    2024

  4. [4]

    Photonic source of heralded greenberger- horne-zeilinger states,

    H. Cao, L. Hansen, F. Giorgino, L. Carosini, P. Zah ´alka, F. Zilk, J. Loredo, and P. Walther, “Photonic source of heralded greenberger- horne-zeilinger states,”Physical Review Letters, vol. 132, no. 13, p. 130604, 2024

    2024

  5. [5]

    Gaussian models to non-gaussian realms of quantum photonic simulators,

    D. D. K. Wayo, R. A. Dias, M. D. Ganji, C. M. Saporetti, and L. Go- liatt, “Gaussian models to non-gaussian realms of quantum photonic simulators,”arXiv preprint arXiv:2502.05245, 2025

    Pith/arXiv arXiv 2025

  6. [6]

    Linear optics to scalable photonic quantum computing,

    D. D. K. Wayo, L. Goliatt, and D. Ganji, “Linear optics to scalable photonic quantum computing,”arXiv preprint arXiv:2501.02513, 2025

    Pith/arXiv arXiv 2025

  7. [7]

    Opportunities and challenges for data quality in the era of quantum computing,

    S. Groppe, V . Uotila, and J. Groppe, “Opportunities and challenges for data quality in the era of quantum computing,”arXiv preprint arXiv:2512.00870, 2025

    arXiv 2025

  8. [8]

    Gaussian quantum information,

    C. Weedbrook, S. Pirandola, R. Garc ´ıa-Patr´on, N. J. Cerf, T. C. Ralph, J. H. Shapiro, and S. Lloyd, “Gaussian quantum information,”Reviews of Modern Physics, vol. 84, no. 2, pp. 621–669, 2012

    2012

  9. [9]

    Quantum information with contin- uous variables,

    S. L. Braunstein and P. Van Loock, “Quantum information with contin- uous variables,”Reviews of modern physics, vol. 77, no. 2, pp. 513–577, 2005

    2005

  10. [10]

    Quantum computation over continuous variables,

    S. Lloyd and S. L. Braunstein, “Quantum computation over continuous variables,”Physical Review Letters, vol. 82, no. 8, p. 1784, 1999

    1999

  11. [11]

    Fault-tolerant measurement-based quantum comput- ing with continuous-variable cluster states,

    N. C. Menicucci, “Fault-tolerant measurement-based quantum comput- ing with continuous-variable cluster states,”Physical review letters, vol. 112, no. 12, p. 120504, 2014

    2014

  12. [12]

    Blueprint for a scalable photonic fault-tolerant quantum computer,

    J. E. Bourassa, R. N. Alexander, M. Vasmer, A. Patil, I. Tzitrin, T. Matsuura, D. Su, B. Q. Baragiola, S. Guha, G. Dauphinaiset al., “Blueprint for a scalable photonic fault-tolerant quantum computer,” Quantum, vol. 5, p. 392, 2021

    2021

  13. [13]

    Fiber-coupled epr-state generation using a single temporally multiplexed squeezed light source,

    M. V . Larsen, X. Guo, C. R. Breum, J. S. Neergaard-Nielsen, and U. L. Andersen, “Fiber-coupled epr-state generation using a single temporally multiplexed squeezed light source,”npj Quantum Information, vol. 5, no. 1, p. 46, 2019

    2019

  14. [14]

    No-go theorem for gaussian quantum error correction,

    J. Niset, J. Fiur ´aˇsek, and N. J. Cerf, “No-go theorem for gaussian quantum error correction,”Physical review letters, vol. 102, no. 12, p. 120501, 2009

    2009

  15. [15]

    Limitations of quantum comput- ing with gaussian cluster states,

    M. Ohliger, K. Kieling, and J. Eisert, “Limitations of quantum comput- ing with gaussian cluster states,”Physical Review A—Atomic, Molecular, and Optical Physics, vol. 82, no. 4, p. 042336, 2010

    2010

  16. [16]

    Positive wigner functions render classical simulation of quantum computation efficient,

    A. Mari and J. Eisert, “Positive wigner functions render classical simulation of quantum computation efficient,”Physical review letters, vol. 109, no. 23, p. 230503, 2012

    2012

  17. [17]

    Encoding a qubit in an oscillator,

    D. Gottesman, A. Kitaev, and J. Preskill, “Encoding a qubit in an oscillator,”Physical Review A, vol. 64, no. 1, p. 012310, 2001

    2001

  18. [18]

    Encoding an oscillator into many oscillators,

    K. Noh, S. Girvin, and L. Jiang, “Encoding an oscillator into many oscillators,”Physical Review Letters, vol. 125, no. 8, p. 080503, 2020

    2020

  19. [19]

    Towards scalable bosonic quantum error correction,

    B. M. Terhal, J. Conrad, and C. Vuillot, “Towards scalable bosonic quantum error correction,”Quantum Science and Technology, vol. 5, no. 4, p. 043001, 2020

    2020

  20. [20]

    Quantum error correction of continuous-variable states against gaussian noise,

    T. Ralph, “Quantum error correction of continuous-variable states against gaussian noise,”Physical Review A—Atomic, Molecular, and Optical Physics, vol. 84, no. 2, p. 022339, 2011

    2011

  21. [21]

    Machine learning for practical quantum error mitigation,

    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, vol. 6, no. 12, pp. 1478–1486, 2024

    2024

  22. [22]

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

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

    Pith/arXiv arXiv 2018

  23. [23]

    Strawberry fields: A software platform for photonic quantum computing,

    N. Killoran, J. Izaac, N. Quesada, V . Bergholm, M. Amy, and C. Weed- brook, “Strawberry fields: A software platform for photonic quantum computing,”Quantum, vol. 3, p. 129, 2019

    2019

  24. [24]

    Performance and structure of single-mode bosonic codes,

    V . V . Albert, K. Noh, K. Duivenvoorden, D. J. Young, R. Brierley, P. Reinhold, C. Vuillot, L. Li, C. Shen, S. M. Girvinet al., “Performance and structure of single-mode bosonic codes,”Physical Review A, vol. 97, no. 3, p. 032346, 2018

    2018

  25. [25]

    Dynamically protected cat-qubits: a new paradigm for universal quantum computation,

    M. Mirrahimi, Z. Leghtas, V . V . Albert, S. Touzard, R. J. Schoelkopf, L. Jiang, and M. H. Devoret, “Dynamically protected cat-qubits: a new paradigm for universal quantum computation,”New Journal of Physics, vol. 16, no. 4, p. 045014, 2014

    2014

  26. [26]

    Error mitigation for short- depth quantum circuits,

    K. Temme, S. Bravyi, and J. M. Gambetta, “Error mitigation for short- depth quantum circuits,”Physical Review Letters, vol. 119, no. 18, p. 180509, 2017

    2017

  27. [27]

    Hybrid quantum- classical algorithms and quantum error mitigation,

    S. Endo, Z. Cai, S. C. Benjamin, and X. Yuan, “Hybrid quantum- classical algorithms and quantum error mitigation,”Journal of the Physical Society of Japan, vol. 90, no. 3, p. 032001, 2021

    2021

This paper was first reviewed by deepseek-v4-flash on August 3, 2026.