Pith. sign in

REVIEW 4 major objections 6 minor 45 references

High label accuracy is not enough for quantum error correction: only confidence-gated hybrid recovery keeps large planar codes near teacher logical failure rates.

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 ↗

Meta-decoders transfer across codes and noise, yet only confidence-gated hybrid recovery—not raw teacher-label accuracy—keeps Planar5×5 logical failure near the teacher.

T0 review reviewed 2026-07-14 challenge →

load-bearing objection Integrated multi-code meta-decoding with a clear, quantified punchline: high teacher-label accuracy is not logical reliability on Planar5×5, and confidence-gated fallback mostly closes the gap at fixed operating points. the 4 major comments →

arxiv 2607.10707 v1 pith:NHFBLGYG submitted 2026-07-12 quant-ph cs.AIcs.ARcs.LG

MDQEC-QAS: Meta-Decoding for Quantum Error Correction with Hardware-Aware VQC Search and Confidence-Gated Recovery

Prashant Kumar Choudhary , Nouhaila Innan , Muhammad Shafique , Rajeev Singh This is my paper
classification quant-ph cs.AIcs.ARcs.LG
keywords quantum error correctionmeta-decodingstabilizer codesplanar codesvariational quantum circuitsquantum architecture searchconfidence-gated recoverylogical failure
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

This paper argues that one shared meta-decoder can learn syndrome-to-recovery maps across several stabilizer codes and noise models, instead of training a separate decoder for each setting. Classical multilayer perceptrons and compact variational quantum circuits both reach nontrivial teacher-label accuracy under interpolation and noise-parameter transfer, but both degrade when the code structure or size is new. On the hardest planar lattice, even near-perfect label accuracy still produces large logical failure relative to a trusted teacher. The authors therefore keep the learned model only on high-confidence cases and fall back to the teacher otherwise. That selective hybrid rule, not raw accuracy, is what brings logical performance close to the teacher and defines the intended use of learned decoding.

Core claim

Across a multi-code, multi-noise benchmark, pooled meta-decoders learn transferable teacher labels, yet on Planar5×5 high supervised accuracy is still far from teacher-level logical reliability; confidence-gated fallback reduces the logical-failure ratios relative to the teacher from 12.08 and 25.91 down to 1.71 and 1.11 in interpolation, so the right system is selective learned assistance rather than unconditional teacher replacement.

What carries the argument

Confidence-gated hybrid recovery: use the learned recovery only when max predicted probability meets a fixed threshold, otherwise apply the code-appropriate teacher recovery; this routes only high-confidence cases through the meta-decoder and is what closes the logical gap on Planar5×5.

Load-bearing premise

The method treats fixed, decoder-specific confidence cutoffs and teacher-generated recovery labels as good enough operating points and training targets, without calibrating those cutoffs or labeling by logical equivalence.

What would settle it

On Planar5×5 Monte Carlo trials, measure whether confidence-gated hybrid recovery still keeps mean logical-failure ratios near 1 across the reported p grid and transfer splits when thresholds are recalibrated or when supervision uses logical equivalence classes instead of exact teacher recovery strings.

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

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 / 6 minor

Summary. The manuscript proposes a unified meta-decoding pipeline for stabilizer QEC that trains a single syndrome-to-recovery model across FiveQubit, Steane, Planar3×3, and Planar5×5 under four noise families, and evaluates five regimes (interpolation, unseen-p, unseen-noise, few-shot unseen-code, few-shot held-out-size). A classical Meta-MLP is compared with compact VQC meta-decoders chosen by hardware-aware quantum architecture search over qubit count, depth, and entanglement. The main empirical claim is that high teacher-label accuracy is not equivalent to logical reliability on Planar5×5: raw logical-failure ratios relative to the teacher remain large (e.g., 12.08 for Meta-MLP and 25.91 for the hardware-aware VQC in interpolation), while confidence-gated teacher fallback reduces them (to 1.71 and 1.11), supporting selective learned recovery rather than unconditional teacher replacement.

Significance. If the results hold under stronger reliability analysis, the paper contributes a useful system-level framing for learned QEC: multi-code multi-noise meta-decoding, an explicit transfer hierarchy (noise/p shifts easier than code-structure shifts), and especially the demonstration that supervised imitation accuracy can mislead on larger planar codes. The separation of teacher-label accuracy from Monte Carlo logical failure (20k trials per point), the raw-vs-fallback ratios in Table 10, and the coverage statistics in Table 11 are concrete and falsifiable. Hardware-aware VQC selection within a documented 18-architecture search space is a secondary but clear contribution for hybrid decoder design, even though VQCs do not beat the Meta-MLP in raw accuracy. The work is more systems/benchmark-oriented than a new decoding algorithm, but that is a legitimate and timely niche.

major comments (4)
  1. Sec. 3.5 and Table 3 fix decoder-specific confidence thresholds (τ=0.80 Meta-MLP, 0.75 VQC) as operating points without calibration, ROC/threshold sweeps, or risk-coverage curves. The central selective-recovery claim (abstract; Sec. 4.8; Table 10) is therefore tied to these choices. Please add a threshold sensitivity analysis on Planar5×5 (at least for interpolation and one transfer regime) showing how R_logic and coverage trade off with τ, and state how τ would be chosen in practice.
  2. Table 10 vs Table 11: under interpolation the Meta-MLP retains ~99.9% Planar5×5 coverage yet still reports fallback ratio 1.71 (raw 12.08), while VQCs reach ~1.04–1.11 largely with much lower coverage (often ~0.16–0.33 in transfer). The abstract and Sec. 4.8 present the reduced ratios as support for confidence-aware selective recovery without clearly separating residual logical cost of confident mistakes from heavy teacher routing. Please revise the interpretation so that residual ratio at high coverage and near-teacher ratio at low coverage are discussed as distinct regimes, and report confident-case logical failure (not only teacher-label confident error in Table 11).
  3. Sec. 3.1–3.2 train and score against teacher recovery labels (NaiveDecoder for algebraic codes; matching for planar codes) while noting that distinct recoveries can be logically equivalent. Logical evaluation uses e'=e⊕r, which is correct operationally, but the supervised objective and confident-error metric remain teacher-string imitation. This weakens claims that confidence isolates logically safe recoveries. At minimum, quantify how often teacher-label errors are still logically correct (or vice versa) on Planar5×5, or discuss logical-equivalence-aware labeling as a limitation with a concrete estimate of its impact.
  4. Eq. (22) and Sec. 3.4: the hardware-aware score S(a) depends on unspecified λ1–λ4 penalties and FakeManila cost proxies, yet the paper treats best-hw vs best-acc selection as a substantive finding. Report the λ values used, justify or ablate them, and show that the hardware-selected architecture remains competitive under modest λ variation; otherwise the hardware-aware claim is under-specified relative to its prominence.
minor comments (6)
  1. Table 4 and abstract report five accuracy numbers for Meta-MLP and hardware-aware VQC; also report accuracy-selected VQC consistently in the abstract for parity with Table 4.
  2. Fig. 3–4 axis labels and panel titles appear partially garbled in the manuscript text (encoding/font issues); ensure publication-quality figures with readable legends for all five regimes.
  3. Clarify global label-space construction (15,582 labels) and padding (S_max=40, 2Q_max=82) earlier when first introducing the pooled dataset; Table 2 is helpful but the semantics of LabelMap(c, r~) could be stated more formally.
  4. Sec. 3.3 sample weights and Planar5×5 oversampling factor 3 are free design choices; a short ablation or sensitivity note would help readers judge robustness of the transfer hierarchy.
  5. Related-work positioning (Table 1) is useful; consider citing recent practical surface-code neural/decoder systems more carefully when claiming novelty of confidence-gated hybrid recovery, which is standard in ML but still underexplored in QEC.
  6. Notation: p_L(p) and R_logic(p) are clear; ensure consistent use of Planar3×3/Planar5×5 vs Planar3x3/Planar5x5 across abstract, tables, and figures.

Circularity Check

0 steps flagged

Empirical supervised meta-decoding and hybrid fallback; no derivation reduces to its inputs by construction.

full rationale

This paper is a multi-regime experimental benchmark of classical and VQC meta-decoders for stabilizer-code recovery, not a first-principles derivation. Models are trained to imitate code-appropriate teacher recoveries (NaiveDecoder / matching), then scored on teacher-label accuracy and Monte Carlo logical failure; the authors explicitly separate those metrics and show that high label accuracy need not imply teacher-level p_L on Planar5×5. Confidence-gated recovery (Eq. 23–24) and the hardware-aware score S(a) (Eq. 22) use fixed operating choices (τ, λ), but those are stated experimental settings, not fitted parameters re-presented as independent predictions of the same quantities. Self-citations to the authors’ other QML/QAS work are peripheral and not load-bearing for the QEC logical-failure claims. No uniqueness theorem, ansatz smuggled via self-citation, or self-definitional identity forces the headline ratios. Residual logical ratios >1 and uncalibrated τ are robustness/correctness issues, not circularity. Score 0 with empty steps is the honest finding.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 1 invented entities

The central claim rests on standard stabilizer decoding plus several hand-chosen experimental knobs (thresholds, weights, search penalties, compact VQC space) and the modeling choice that teacher recoveries are the right supervised targets. No new physical entities; free parameters are operational, not fitted physical constants.

free parameters (4)
  • confidence thresholds τ = 0.80 / 0.75
    Fixed operating points τ=0.80 (Meta-MLP) and τ=0.75 (VQC) control when learned recovery is used; fallback results depend on these choices.
  • Planar5×5 oversampling and sample weights = oversample×3; code weights 1.0/1.0/1.2/3.0
    Oversampling factor 3 and code/noise weights (up to 3.0 / 1.1) rebalance training toward hard regimes and affect supervised accuracy.
  • hardware-aware score penalties λ1–λ4
    Coefficients in S(a) trade validation accuracy against raw/transpiled depth and two-qubit gates; they select the 'best-hw' circuit.
  • VQC search space bounds = 18 architectures
    n_q∈{4,5,6}, L∈{1,2}, entangling∈{chain,ring,full} (18 candidates) limits which quantum models can win.
axioms (4)
  • domain assumption Stabilizer syndrome decoding: s=He^T (mod 2); success iff e⊕r is logically trivial.
    Standard Gottesman/stabilizer formalism used throughout Sec. 3.1–3.2 as the evaluation ground truth.
  • domain assumption Code-appropriate teachers (NaiveDecoder for FiveQubit/Steane; matching for planar) supply correct supervised recovery labels.
    All meta-training targets are teacher recoveries; logical ratios are relative to the same teachers (Sec. 3.2, 3.5).
  • ad hoc to paper Pooled global recovery-label space with code-aware logit masking preserves code-specific recovery semantics.
    Construction in Sec. 3.2–3.3 enables one classifier head; validity of transfer metrics depends on this labeling choice.
  • ad hoc to paper Angle-embedded VQCs with classical projection and linear head are adequate compact meta-decoders for the studied regimes.
    Architecture class fixed before search (Sec. 3.3–3.4); conclusions about VQC competitiveness are relative to this ansatz family.
invented entities (1)
  • MDQEC-QAS hybrid meta-decoding pipeline no independent evidence
    purpose: Unify multi-code meta-training, hardware-aware VQC search, and confidence-gated teacher fallback into one evaluation system.
    System-level construct rather than a new physical object; independent evidence is the reported transfer and logical metrics only.

reviewed 2026-07-14 · how reviews work

0 comments
Cite this review

Pith. "Pith review of MDQEC-QAS: Meta-Decoding for Quantum Error Correction with Hardware-Aware VQC Search and Confidence-Gated Recovery." pith.science (2026). https://pith.science/paper/NHFBLGYG

@misc{pith2026260710707,
  author       = {Pith},
  title        = {Pith review of: MDQEC-QAS: Meta-Decoding for Quantum Error Correction with Hardware-Aware VQC Search and Confidence-Gated Recovery},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NHFBLGYG}},
  note         = {Machine review of arXiv:2607.10707}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

We propose a unified meta-decoding framework for quantum error correction that learns syndrome-to-recovery mappings across multiple stabilizer codes and noise settings, without requiring separate decoders for each configuration. The benchmark includes FiveQubit, Steane, Planar3x3, and Planar5x5 codes, four noise families, and five evaluation regimes: interpolation, unseen-p transfer, unseen-noise transfer, few-shot unseen-code adaptation, and few-shot held-out-size adaptation. We compare a classical Meta-MLP teacher-trained baseline with variational quantum circuit (VQC) meta-decoders selected through hardware-aware quantum architecture search over qubit count, circuit depth, and entangling topology. The Meta-MLP achieves teacher-label accuracies of 0.9993, 0.9118, 0.9342, 0.6304, and 0.7548 across the five regimes, while the hardware-aware VQC achieves 0.9400, 0.8495, 0.8415, 0.5678, and 0.7143. However, logical-level evaluation shows that high teacher-label accuracy alone is insufficient in the most challenging Planar5x5 setting. During interpolation, the raw logical-failure ratios relative to the teacher are 12.08 and 25.91 for the Meta-MLP and VQC, respectively, whereas confidence-gated fallback reduces them to 1.71 and 1.11. These results support confidence-aware selective recovery rather than unconditional teacher replacement.

Figures

Figures reproduced from arXiv: 2607.10707 by Muhammad Shafique, Nouhaila Innan, Prashant Kumar Choudhary, Rajeev Singh.

Figure 1
Figure 1. Figure 1: Overview of the learned-assisted meta-decoding framework. Syndrome-recovery data are generated using code-appropriate teacher decoders, pooled into a global label space, used to train classical and VQC meta-decoders, and evaluated using teacher-label accuracy, confidence-gated fallback, and logical-failure analysis. a single model is trained on tuples (𝑠, 𝑐, 𝜂, 𝑝) ↦→ 𝑦, (5) where 𝑐 is the code identifier, … view at source ↗
Figure 2
Figure 2. Figure 2: Qubit-level mechanism of learned-assisted QEC meta-decoding. A logical qubit |𝜓𝐿⟩ is encoded into physical qubits, where Pauli errors generate stabilizer syndromes according to 𝑠 = 𝐻𝑒𝑇 (mod 2). The syndrome, code identity, noise identity, and normalized physical error rate are combined into the meta-feature vector 𝑥 = [𝑠˜∥onehot(𝑐) ∥onehot(𝜂) ∥𝑝˜] and projected to a compact VQC input. The VQC applies angle… view at source ↗
Figure 3
Figure 3. Figure 3: Training and validation accuracy of the Meta-MLP on the pooled multi-code, multi-noise dataset.        [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Retraining curves of the two selected VQC meta-decoders. Panels (a)–(e) correspond to interpolation, unseen-𝑝 transfer, unseen￾noise transfer, few-shot unseen-code adaptation, and few-shot held￾out-size adaptation, respectively. search-stage test accuracy, while [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Hardware-aware score of candidate VQC architectures. The score combines predictive performance with implementation￾oriented circuit-cost proxies such as depth and two-qubit gate count. q0 : RY (0) RY (2.272e − 05) RZ (6.283) • RY (0.5147) RZ (6.283) • q1 : RY (0) RY (1.714e − 05) RZ (3.142) • RY (3.156) RZ (3.142) • q2 : RY (0) RY (3.142) RZ (−8.761e − 05) • RY (6.283) RZ (0) • q3 : RY (0) RY (3.142) RZ (1… view at source ↗
Figure 5
Figure 5. Figure 5: Validation and search-stage test accuracy of candidate VQC architectures explored during quantum architecture search. This code-wise pattern is consistent with the broader group￾wise results in [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Selected VQC architectures obtained using search-stage val￾idation accuracy. Panels (a)–(e) correspond to interpolation, unseen-𝑝 transfer, unseen-noise transfer, few-shot unseen-code adaptation, and few-shot held-out-size adaptation, respectively. 8 [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Selected VQC architectures obtained using the hardware￾aware score. Panels (a)–(e) correspond to interpolation, unseen-𝑝 transfer, unseen-noise transfer, few-shot unseen-code adaptation, and few-shot held-out-size adaptation, respectively [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 10
Figure 10. Figure 10: Combined logical-failure curves under the confidence￾gated hybrid recovery protocol. is the most demanding regime: raw learned decoding re￾mains far from the teacher decoder, while confidence-gated fallback reduces the logical gap by using the learned decoder only on confident cases and routing uncertain cases to the teacher decoder (Additional Planar5×5 curves are reported in Appendix A) [PITH_FULL_IMAG… view at source ↗
Figure 11
Figure 11. Figure 11: Computed decoder-advantage summary across code families and noise models. Each cell reports the best learned-decoder advantage relative to the corresponding baseline decoder, averaged over transfer settings and physical error probabilities. Positive values indicate lower logical-failure rates for the best learned decoder, whereas negative values indicate that the baseline decoder remains superior. setting… 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

45 extracted references · 12 canonical work pages

  1. [1]

    Peter W. Shor. Scheme for reducing decoherence in quantum computer memory.Physical Review A, 52(4): 11 MDQEC-QAS Choudhary et al. Table 11.Planar5×5 selective-decoding coverage computed from the saved test-set confidence values. Meta-MLP uses𝜏=0.80 and VQC models use𝜏=0.75 , matching the reported fallback protocol. Confident error is measured against the ...

    work page doi:10.1103/physreva.52.r249 1995

  2. [2]

    Multiple-particleinterferenceandquantum errorcorrection.ProceedingsoftheRoyalSocietyA,452 (1954):2551–2577, 1996

    A.M.Steane. Multiple-particleinterferenceandquantum errorcorrection.ProceedingsoftheRoyalSocietyA,452 (1954):2551–2577, 1996. doi: 10.1098/rspa.1996.0136

    doi:10.1098/rspa.1996.0136 1954

  3. [3]

    PhD thesis, California Institute of Technol- ogy, 1997

    Daniel Gottesman.Stabilizer Codes and Quantum Error Correction. PhD thesis, California Institute of Technol- ogy, 1997. Ph.D. thesis

    1997

  4. [4]

    A. Yu. Kitaev. Fault-tolerant quantum computation by anyons.Annals of Physics, 303(1):2–30, 2003. doi: 10.1016/S0003-4916(02)00018-0

    doi:10.1016/s0003-4916(02)00018-0 2003

  5. [5]

    BarbaraM.Terhal.Quantumerrorcorrectionforquantum memories.Reviews of Modern Physics, 87(2):307–346,

  6. [6]

    doi: 10.1103/RevModPhys.87.307

    doi:10.1103/revmodphys.87.307

  7. [7]

    Topological quantum memory.Journal of Mathematical Physics, 43(9):4452–4505, 2002

    Eric Dennis, Alexei Kitaev, Andrew Landahl, and John Preskill. Topological quantum memory.Journal of Mathematical Physics, 43(9):4452–4505, 2002. doi: 10.1063/1.1499754

    doi:10.1063/1.1499754 2002

  8. [8]

    Fowler, Matteo Mariantoni, John M

    Austin G. Fowler, Matteo Mariantoni, John M. Martinis, andAndrewN.Cleland.Surfacecodes: Towardspractical large-scalequantumcomputation.PhysicalReviewA,86 (3):032324, 2012. doi: 10.1103/PhysRevA.86.032324

    doi:10.1103/physreva.86.032324 2012

  9. [9]

    Campbell, Barbara M

    Earl T. Campbell, Barbara M. Terhal, and Christophe Vuillot. Roads towards fault-tolerant universal quantum computation.Nature, 549(7671):172–179, 2017. doi: 10.1038/nature23460

    doi:10.1038/nature23460 2017

  10. [10]

    doi: 10.22331/q-2024-10-10-1498

    Antonio de Marti i Olius, Pau Fuentes, Román Orús, PereM.Crespo,andJoséEtxezarretaMartinez.Decoding algorithms for surface codes.Quantum, 8:1498, 2024. doi: 10.22331/q-2024-10-10-1498

    doi:10.22331/q-2024-10-10-1498 2024

  11. [11]

    Almost-linear timedecodingalgorithmfortopologicalcodes.Quantum, 5:595, 2021

    NicolasDelfosseandNaomiH.Nickerson. Almost-linear timedecodingalgorithmfortopologicalcodes.Quantum, 5:595, 2021. doi: 10.22331/q-2021-12-02-595

    doi:10.22331/q-2021-12-02-595 2021

  12. [12]

    Machine-learning methods for quantum error correction: A survey.AVS Quantum Science, 5(4):041101, 2023

    Christopher Chamberland, Guanyu Zhu, Tomas Jochym- O’Connor, and Aleksander Kubica. Machine-learning methods for quantum error correction: A survey.AVS Quantum Science, 5(4):041101, 2023. doi: 10.1116/5.01 66514

    work page doi:10.1116/5.01 2023

  13. [13]

    Johannes Bausch, A. W. Senior, Francisco J. H. Heras, TristanEdlich,AlexanderDavies,MaxNewman,Christo- pher Brej, Hieu Pham, Livia de Oliveira, Charlotte Papa- georgiou, Ioannis Karamcheti, Sagar Joglekar, Simon P. Chao, Mohammad Babaei, Neil Houlsby, Joshua Izaac, Stefano Carrazza, and Pushmeet Kohli. Learning high- accuracyerrordecodingforquantumprocess...

    doi:10.1038/s41586-024-08148-8 2024

  14. [14]

    Norris, Christopher Eichler, Alexandre Blais, Markus Müller, Sebastian Krinner, and Andreas Wallraff

    Ants Remm, Nathan Lacroix, Lukas Bödeker, Elie Genois,ChristophHellings,FrançoisSwiadek,GrahamJ. Norris, Christopher Eichler, Alexandre Blais, Markus Müller, Sebastian Krinner, and Andreas Wallraff. Exper- imentally informed decoding of stabilizer codes based on syndrome correlations.Phys. Rev. Res., 8:013044, Jan 2026. doi: 10.1103/z1ng-wg3k. URL https: ...

    work page doi:10.1103/z1ng-wg3k 2026

  15. [15]

    O’Brien, Brian Tarasinski, and Carlo W

    Paul Baireuther, Thomas E. O’Brien, Brian Tarasinski, and Carlo W. J. Beenakker. Machine-learning-assisted correctionofcorrelatedqubiterrorsinatopologicalcode. Quantum, 2:48, 2018. doi: 10.22331/q-2018-01-29-48

    doi:10.22331/q-2018-01-29-48 2018

  16. [16]

    Hardness results for decoding the surface code with pauli noise.Quantum, 8:1511, 2023

    Alex Fischer and Akimasa Miyake. Hardness results for decoding the surface code with pauli noise.Quantum, 8:1511, 2023. URLhttps://api.semanticschola r.org/CorpusID:262054085

    2023

  17. [17]

    doi: 10.1103/PhysRevA.105.0 42616

    Karl Hammar, Alexei Orekhov, Patrik Wallin Hybelius, Anna Katariina Wisakanto, Basudha Srivastava, An- tonFriskKockum,andMatsGranath.Error-rate-agnostic decodingoftopologicalstabilizercodes.PhysicalReview A, 105(4):042616, 2022. doi: 10.1103/PhysRevA.105.0 42616. 12 MDQEC-QAS Choudhary et al

    work page doi:10.1103/physreva.105.0 2022

  18. [18]

    Varbanov, Marc Serra-Peralta, David Byfield, and Barbara M

    Boris M. Varbanov, Marc Serra-Peralta, David Byfield, and Barbara M. Terhal. Neural network decoder for near-term surface-code experiments.Phys. Rev. Res., 7: 013029, Jan 2025. doi: 10.1103/PhysRevResearch.7.01

    work page doi:10.1103/physrevresearch.7.01 2025

  19. [19]

    URL https://link.aps.org/doi/10.1103 /PhysRevResearch.7.013029

  20. [20]

    Córcoles, Kristan Temme, Aram W

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

  21. [21]

    doi: 10.1038/s41586-019-0980-2

    doi:10.1038/s41586-019-0980-2

  22. [22]

    Svore, and NathanWiebe

    Maria Schuld, Alex Bocharov, Krysta M. Svore, and NathanWiebe. Circuit-centricquantumclassifiers.Phys- ical Review A, 101(3):032308, 2020. doi: 10.1103/Phys RevA.101.032308

    doi:10.1103/phys 2020

  23. [23]

    Quantum circuit architecture search for variational quantum algorithms.npj Quantum Informa- tion, 8(1):62, 2022

    YuxuanDu, TaoHuang, ShanYou, Min-HsiuHsieh, and Dacheng Tao. Quantum circuit architecture search for variational quantum algorithms.npj Quantum Informa- tion, 8(1):62, 2022. doi: 10.1038/s41534-022-00590-1

    work page doi:10.1038/s41534-022-00590-1 2022

  24. [24]

    Quantum architecture search: a survey

    Darya Martyniuk, Johannes Jung, and Adrian Paschke. Quantum architecture search: a survey. In2024 IEEE International Conference on Quantum Computing and Engineering (QCE), volume 1, pages 1695–1706. IEEE, 2024

    2024

  25. [25]

    Graph-based bayesian optimization for quantum circuit architecture search with uncertainty calibrated surrogates, 2025

    Prashant Kumar Choudhary, Nouhaila Innan, Muham- mad Shafique, and Rajeev Singh. Graph-based bayesian optimization for quantum circuit architecture search with uncertainty calibrated surrogates, 2025. URL https://arxiv.org/abs/2512.09586

    Pith/arXiv arXiv 2025

  26. [26]

    Hybridquantum-classicalneu- ralarchitecturesearch.arXivpreprintarXiv:2605.18345, 2026

    Alberto Marchisio, Muhammad Kashif, Nouhaila Innan, andMuhammadShafique. Hybridquantum-classicalneu- ralarchitecturesearch.arXivpreprintarXiv:2605.18345, 2026

    Pith/arXiv arXiv 2026

  27. [27]

    Faqnas: Flops-aware hybrid quantum neural architecture search using genetic algorithm

    Muhammad Kashif, Shaf Khalid, Alberto Marchisio, Nouhaila Innan, and Muhammad Shafique. Faqnas: Flops-aware hybrid quantum neural architecture search using genetic algorithm. In2026 Design, Automation & Test in Europe Conference (DATE), pages 1–7. IEEE, 2026

    2026

  28. [28]

    Qas-qtns: Curriculum reinforce- ment learning-driven quantum architecture search for quantum tensor networks

    Siddhant Dutta, Nouhaila Innan, Sadok Ben Yahia, and Muhammad Shafique. Qas-qtns: Curriculum reinforce- ment learning-driven quantum architecture search for quantum tensor networks. In2025 IEEE International Conference on Quantum Computing and Engineering (QCE), volume 1, pages 1739–1747. IEEE, 2025

    2025

  29. [29]

    Artificial intelligence for quantum error correction: Opportunities and challenges

    Jiahao Pan and Hyeon Kim. Artificial intelligence for quantum error correction: Opportunities and challenges. AVS Quantum Science, 6(3):030801, 2024. doi: 10.111 6/5.0214974

    2024

  30. [30]

    Derks, Jens Eisert, and Robert Wille

    Lucas Berent, Lukas Burgholzer, Peter-Jan H.S. Derks, Jens Eisert, and Robert Wille. Decoding quantum color codes with maxsat.Quantum, 8:1506, October 2024. ISSN 2521-327X. doi: 10.22331/q-2024-10-23-1506. URL http://dx.doi.org/10.22331/q-2024-1 0-23-1506

    work page doi:10.22331/q-2024-10-23-1506 2024

  31. [31]

    Giacomo Torlai and Roger G. Melko. Neural decoder for topological codes.Physical Review Letters, 119(3): 030501, 2017. doi: 10.1103/PhysRevLett.119.030501

    doi:10.1103/physrevlett.119.030501 2017

  32. [32]

    Decoding small surface codes with feedforward neu- ral networks.Quantum Science and Technology, 3(1): 015004, 2018

    Savvas Varsamopoulos, Ben Criger, and Koen Bertels. Decoding small surface codes with feedforward neu- ral networks.Quantum Science and Technology, 3(1): 015004, 2018. doi: 10.1088/2058-9565/aa955a

    doi:10.1088/2058-9565/aa955a 2018

  33. [33]

    Quantumerrorcorrectionforthetoric code using deep reinforcement learning.Quantum, 3: 183, 2019

    Philip Andreasson, Joel Johansson, Simon Liljestrand, andMatsGranath. Quantumerrorcorrectionforthetoric code using deep reinforcement learning.Quantum, 3: 183, 2019. doi: 10.22331/q-2019-09-02-183

    work page doi:10.22331/q-2019-09-02-183 2019

  34. [34]

    Moritz Lange, Pontus Havström, Basudha Srivastava, Valdemar Bergentall, Karl Hammar, Olivia Heuts, Evert P. L. van Nieuwenburg, and Mats Granath. Data-driven decoding of quantum error correcting codes using graph neuralnetworks.PhysicalReviewResearch,7(2):023181,

  35. [35]

    doi: 10.1103/PhysRevResearch.7.023181

    doi:10.1103/physrevresearch.7.023181

  36. [36]

    Adrián Pérez-Salinas, Alba Cervera-Lierta, Elies Gil- Fuster, and José I. Latorre. Data re-uploading for a universalquantumclassifier.Quantum, 4:226, 2020. doi: 10.22331/q-2020-02-06-226

    doi:10.22331/q-2020-02-06-226 2020

  37. [37]

    Iris Cong, Soonwon Choi, and Mikhail D. Lukin. Quan- tum convolutional neural networks.Nature Physics, 15 (12):1273–1278, 2019. doi: 10.1038/s41567-019-064 8-8

    work page doi:10.1038/s41567-019-064 2019

  38. [38]

    Quantumstatetomography using quantum machine learning.Quantum Machine Intelligence, 6(1):28, 2024

    Nouhaila Innan, Owais Ishtiaq Siddiqui, Shivang Arora, Tamojit Ghosh, Yasemin Poyraz Koçak, Dominic Para- gas, Abdullah Al Omar Galib, Muhammad Al-Zafar Khan,andMohamedBennai. Quantumstatetomography using quantum machine learning.Quantum Machine Intelligence, 6(1):28, 2024

    2024

  39. [39]

    Qnn-vrcs: A quantum neural network forvehicleroadcooperationsystems.IEEETransactions on Intelligent Transportation Systems, 2025

    Nouhaila Innan, Bikash K Behera, Saif Al-Kuwari, and Ahmed Farouk. Qnn-vrcs: A quantum neural network forvehicleroadcooperationsystems.IEEETransactions on Intelligent Transportation Systems, 2025

    2025

  40. [40]

    Lep-qnn: Loan eligibility predictionusingquantumneuralnetworks

    Nouhaila Innan, Alberto Marchisio, Mohamed Bennai, and Muhammad Shafique. Lep-qnn: Loan eligibility predictionusingquantumneuralnetworks. In2025IEEE International Conference on Quantum Computing and Engineering (QCE), volume 1, pages 1864–1872. IEEE, 2025

    2025

  41. [41]

    Hqnn-fsp: A hybrid classical-quantumneuralnetworkforregression-basedfi- nancialstockmarketprediction.QuantumMachineIntel- ligence, 8(55), 2026

    Prashant Kumar Choudhary, Nouhaila Innan, Muham- mad Shafique, and Rajeev Singh. Hqnn-fsp: A hybrid classical-quantumneuralnetworkforregression-basedfi- nancialstockmarketprediction.QuantumMachineIntel- ligence, 8(55), 2026. doi: 10.1007/s42484-026-00390-9

    work page doi:10.1007/s42484-026-00390-9 2026

  42. [42]

    McClean, Masoud Mohseni, Patrick van der Smagt, and Martin Leib

    Andrea Skolik, Jarrod R. McClean, Masoud Mohseni, Patrick van der Smagt, and Martin Leib. Layerwise learningforquantumneuralnetworks.QuantumMachine Intelligence, 3(1):5, 2021. doi: 10.1007/s42484-020-0 0036-4

    work page doi:10.1007/s42484-020-0 2021

  43. [43]

    Cerezo, Akira Sone, Tyler Volkoff, Lukasz Cin- cio, and Patrick J

    M. Cerezo, Akira Sone, Tyler Volkoff, Lukasz Cin- cio, and Patrick J. Coles. Cost function dependent barren plateaus in shallow parametrized quantum cir- cuits.Nature Communications, 12(1):1791, 2021. doi: 10.1038/s41467-021-21728-w. 13 MDQEC-QAS Choudhary et al

    doi:10.1038/s41467-021-21728-w 2021

  44. [44]

    Scaling laws for hybrid quantum neural networks: Depth, width, and quantum-centric diagnostics.arXiv preprint arXiv:2604.06007, 2026

    Danil Vyskubov, Kirill Vyskubov, Nouhaila Innan, and Muhammad Shafique. Scaling laws for hybrid quantum neural networks: Depth, width, and quantum-centric diagnostics.arXiv preprint arXiv:2604.06007, 2026

    Pith/arXiv arXiv 2026

  45. [45]

    General-purpose quantum architecture search based on deep reinforcement learning.Phys

    Xiao-Yu Bi, Yi-Ming Yu, Ye-Hong Chen, and Zhi-Rong Zhong. General-purpose quantum architecture search based on deep reinforcement learning.Phys. Rev. A, 112: 052409, Nov 2025. doi: 10.1103/7rc4-p446. URLhttp s://link.aps.org/doi/10.1103/7rc4-p446. 14 MDQEC-QAS Choudhary et al. A Planar5×5 Logical-Failure Curves ThePlanar5 ×5curvespresentedinFigure12provid...

    work page doi:10.1103/7rc4-p446 2025

This paper was first reviewed by grok-4.5 on July 14, 2026.