Pith. sign in

REVIEW 3 major objections 34 references

A confidence gate on a neural surface-code decoder escalates only a few percent of hard syndromes to exact matching and recovers most of the accuracy gap.

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 ↗

Confidence-gated neural decoding escalates only ~3–6% of rotated-surface-code syndromes to MWPM and raises end-to-end accuracy from 99.21% to 99.81% at d=7 under circuit-level depolarising noise.

T0 review reviewed 2026-07-11 challenge →

load-bearing objection Solid, reproducible cascade benchmark for surface-code decoding; the co-design title is mostly roadmap, and confidence is only trustworthy in the low-noise regime the paper already flags. the 3 major comments →

arxiv 2607.05814 v2 pith:FTKCQG4K submitted 2026-07-07 quant-ph cs.ETcs.LG

Latency-Constrained Hardware-Aware Quantum Error Correction Co-Design with Adaptive Confidence-Gated Neural Decoding for the Rotated Surface Code

Sumit Chongder This is my paper
classification quant-ph cs.ETcs.LG
keywords quantum error correctionsurface codeneural decodingminimum-weight perfect matchingconfidence-aware inferencehardware-aware co-designreal-time decodingfault tolerance
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

Real-time decoding is a bottleneck for fault-tolerant quantum computing. This paper treats surface-code decoding as a two-stage control problem: a small feed-forward neural network handles the bulk of syndrome measurements with a confidence score, and only low-confidence cases are escalated to exact minimum-weight perfect matching. On rotated surface codes under circuit-level depolarising noise, raising the threshold so that roughly 3–6% of shots go to the refinement stage lifts end-to-end logical accuracy from about 99.2% to 99.8% while the neural path still carries most of the traffic and sustains high CPU throughput. The work supplies a full multi-axis benchmark of accuracy, latency, throughput and decoding-graph size, and situates the decoder as the first measured piece of a larger hardware-aware co-design loop whose remaining stages (code discovery, multi-noise ranking) are left for later work.

Core claim

For the rotated surface code under independent circuit-level depolarising noise, a confidence-gated cascade that routes only 0.36%–6.19% of syndromes from a lightweight neural fast path to MWPM refinement recovers a large share of the accuracy gap to exact matching (end-to-end accuracy rising from 99.21% to 99.81% at d=7, τ=0.95) while keeping average decoding cost dominated by the fast path.

What carries the argument

Adaptive confidence-gated decoder: a feed-forward network outputs a logical correction and a scalar confidence (maximum class probability); syndromes with confidence below a tunable threshold τ are escalated to PyMatching MWPM on the Stim decoding graph.

Load-bearing premise

The network’s maximum class probability remains a reliable enough signal of correctness for safe escalation across noise strengths and distances, even though the paper itself shows mean confidence can stay high while accuracy collapses at high noise.

What would settle it

Measure end-to-end logical accuracy and the fraction of true neural errors that fall below each τ on an independent held-out set (or under a different noise model); if the escalated tail does not concentrate most of the residual logical failures, the accuracy–cost trade-off disappears.

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

3 major / 0 minor

Summary. The manuscript presents a two-stage, confidence-gated decoder for the rotated surface code: a compact feed-forward network handles most syndromes, and only low-confidence cases are escalated to PyMatching MWPM. Under Stim circuit-level depolarising noise for d∈{3,5,7,9,11}, the authors report logical accuracy, confidence-controlled accuracy–cost trade-offs, CPU throughput/latency, and decoding-graph resource scaling. The headline empirical result (Table 2, §5.3) is that at d=7, raising the escalation threshold τ from 0.60 to 0.95 routes 0.36%–6.19% of shots to MWPM and lifts end-to-end accuracy from 99.21% to 99.81% relative to the neural-only path. The paper carefully separates this validated decoder module from a broader hardware-aware co-design roadmap (Figs. 1–2, Eq. 1) that remains architectural, and releases code, models, and raw tables.

Significance. If the reported trade-off holds under the stated scope, the work is a useful systems contribution to real-time QEC decoding: it quantifies a cascade-style accuracy–cost curve for surface-code decoding jointly with throughput, batch scaling, and detector-resource growth, and it does so with full reproducibility (Stim/PyMatching pipeline, trained weights, CSV tables). The explicit separation of implemented results from the co-design roadmap is a strength relative to many over-scoped proposals. The contribution is incremental rather than transformative—cascade/mixture-of-experts routing and neural surface-code decoders both have antecedents—but the multi-axis characterisation and open artefacts make it a credible building block for latency-aware decoder deployment studies.

major comments (3)
  1. The central claim of the paper is the confidence-gated accuracy–cost trade-off, yet Table 2 / §5.3 reports it only at a single operating point (d=7; physical error rate p is not stated in the table caption or surrounding text). Given that §5.1 and Table 1 show neural accuracy and mean confidence degrading sharply with d at fixed p=10^{-3}, and §5.5/Table 3 show accuracy collapsing to 0.767 while mean confidence remains 0.947 at p=5×10^{-3}, d=7, the same small escalation fractions need not recover accuracy at other (d,p). Please either (i) report the τ-sweep for at least a few additional cells spanning the distance–noise grid (e.g. d=5 and d=9 at p=10^{-3}, and d=7 at p=5×10^{-3}), or (ii) clearly restrict the abstract/title claim to the single validated regime and move generalisation language to future work.
  2. §3.4 and §5.4 define confidence as c=max_k p_k and treat it as the routing signal, but the paper provides no quantitative calibration metric (reliability diagram, ECE, or temperature scaling). §5.5 and §6.1 themselves document a large overconfidence gap at high noise (accuracy 0.767 vs mean confidence 0.947). Without calibration diagnostics—or a demonstration that a fixed τ still isolates a disproportionately error-prone subset when the network is overconfident—the claim that “confidence-gated” routing safely recovers logical accuracy with bounded average cost is only weakly supported outside the well-calibrated regime of Table 2. A short calibration analysis at the Table 2 operating point and at the high-noise point of Table 3 would make the mechanism’s load-bearing assumption testable.
  3. For a latency-constrained decoder paper, two baselines needed to interpret Table 2 and §5.8 are missing: (i) pure MWPM logical accuracy on the same shot set (the accuracy ceiling under the matching-graph model), and (ii) a separate per-shot latency distribution for the MWPM refinement path alone. §7 item 5 acknowledges the second gap. Without them, “bounded increase in average decoding cost” and the claim that the fast path remains the dominant cost centre cannot be checked against worst-case (tail) latency, which is what hard real-time control budgets care about. Adding MWPM-only accuracy and a refinement-path latency histogram for the escalated subset at the Table 2 conditions would close this.

Circularity Check

0 steps flagged

Empirical cascade decoder: accuracy and escalation fraction are measured against Stim ground truth and PyMatching, not forced by construction or self-citation.

full rationale

This is a systems/benchmark paper, not a first-principles derivation. Logical accuracy is the fraction of shots whose correction matches Stim-sampled ground-truth logical flips; the refinement stage is an external classical MWPM solver (PyMatching); escalation fraction is the observed share of syndromes with c = max_k p_k below τ. None of these quantities is defined in terms of the others, fitted and then re-presented as a prediction, or justified by a load-bearing self-citation uniqueness theorem. Training on Stim detector samples and evaluating on matched Stim samples is standard supervised evaluation, not circular reduction. The hardware-aware co-design roadmap (Eq. 1, greyed components in Figs. 1–2) is explicitly scoped as unevaluated future work and does not underwrite the Table 2 accuracy–escalation numbers. Cascade/MoE framing cites Viola & Jones and Shazeer et al., not the author’s prior theorems. No self-definitional loop, fitted-input-as-prediction, or renaming of a known result as a forced derivation appears. Score 0 is the correct honest finding.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 2 invented entities

The central cascade claim rests on standard QEC simulation assumptions (Stim circuit-level depolarising noise; MWPM exactness on the matching graph), a hand-chosen neural architecture and training recipe, and the operational definition of confidence as max class probability with a free inference threshold τ. No new physical entities are postulated. The multi-objective co-design loss (Eq. 1) is an architectural proposal, not a fitted scientific law, and is not used to produce the main accuracy numbers.

free parameters (4)
  • escalation confidence threshold τ
    Inference-time control knob swept over {0.60,0.70,0.80,0.90,0.95}; end-to-end accuracy and escalated fraction are defined relative to this choice (§3.4, Table 2).
  • fast-path MLP architecture and training hyperparameters
    Fixed feed-forward capacity, cross-entropy training, data split, and epoch budget (Appendix C); no architecture search. Capacity choice is implicated in non-monotonic high-noise accuracy (§6.2).
  • training/evaluation shot budgets and batch sizes
    N_train and 5×10⁴-shot evaluation cells, plus batch sizes 1–512, set statistical resolution and throughput numbers (Sections 4–5).
  • multi-objective weights w1…w4 in Eq. (1)
    User-specified weights for the unrun co-design optimiser; not fitted here, but free if the roadmap were executed.
axioms (5)
  • domain assumption Stim’s independent circuit-level depolarising noise (gate and measurement flip probability p) is an adequate evaluation channel for the reported deployability claims.
    Stated in §3.3; coherent/leakage/correlated noise explicitly excluded (§7).
  • domain assumption MWPM on the Stim-derived decoding graph is exact for the assumed independent-error matching model and thus a valid accuracy floor for escalated syndromes.
    §3.4 refinement path; standard surface-code matching assumption.
  • ad hoc to paper Maximum predicted class probability is a usable confidence score for cascade routing.
    Defined in §3.4; effectiveness is empirical and degrades under high noise (§5.5–§6.1).
  • domain assumption Rotated surface-code memory experiments with d rounds of syndrome extraction define the logical error metric used throughout.
    §3.2 standard memory-experiment convention.
  • domain assumption Standard multilayer-perceptron training with cross-entropy yields a decoder whose errors concentrate in a low-confidence tail.
    Training Algorithm 2 / Appendix C; cascade value depends on this empirical shape (Fig. 8).
invented entities (2)
  • Adaptive confidence-gated two-tier surface-code decoder (neural fast path + MWPM refinement) independent evidence
    purpose: Operational cascade that trades a tunable escalated fraction for end-to-end logical accuracy under latency constraints.
    The paper’s main constructed system; related to classical cascades and prior neural QEC decoders but specified and benchmarked here as a full pipeline.
  • Hardware-aware closed-loop QEC co-design optimiser (Eq. 1 / grey boxes in Figs. 1–2) no independent evidence
    purpose: Proposed outer loop coupling code generation, noise models, decoding, and multi-objective hardware ranking.
    Architecturally described but not implemented or evaluated; no independent empirical handle in this paper.

reviewed 2026-07-11 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Latency-Constrained Hardware-Aware Quantum Error Correction Co-Design with Adaptive Confidence-Gated Neural Decoding for the Rotated Surface Code." pith.science (2026). https://pith.science/paper/FTKCQG4K

@misc{pith2026260705814,
  author       = {Pith},
  title        = {Pith review of: Latency-Constrained Hardware-Aware Quantum Error Correction Co-Design with Adaptive Confidence-Gated Neural Decoding for the Rotated Surface Code},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FTKCQG4K}},
  note         = {Machine review of arXiv:2607.05814}
}
Share X Bluesky LinkedIn Reddit HN
abstract

Real-time decoding is a major bottleneck in scaling quantum error correction (QEC) from noisy intermediate-scale quantum (NISQ) devices to fault-tolerant quantum computing. We present an adaptive confidence-gated decoding framework for the rotated surface code that treats decoding as a two-stage inference problem. A lightweight feed-forward neural network performs fast-path decoding for the majority of syndrome measurements, while only low-confidence predictions are escalated to a minimum-weight perfect matching (MWPM) refinement stage. We benchmark the framework on rotated surface codes with distances $d \in \{3,5,7,9,11\}$ under circuit-level depolarising noise using the Stim stabiliser simulator. The evaluation characterises logical accuracy, confidence-controlled accuracy-latency trade-offs, decoding throughput, per-shot latency, and decoding-graph resource scaling. Routing only 3.3%-6.2% of syndromes to the refinement stage improves logical accuracy from 99.21% for the neural-only baseline to 99.81% at a confidence threshold of 0.95 while incurring only a bounded increase in average decoding cost. Neural-decoder throughput saturates near $4.6 \times 10^{5}$ samples s$^{-1}$ at batch size 512 on commodity CPU hardware, indicating that the neural fast path is not the dominant throughput bottleneck beyond code distance $d=7$. We release the complete benchmarking pipeline, trained models, raw benchmark data, and source code, and explicitly distinguish the experimentally validated contributions from the broader hardware-aware QEC co-design roadmap, including hardware-constrained code discovery, GPU-accelerated inference, and multi-noise optimisation, which remain directions for future work.

Figures

Figures reproduced from arXiv: 2607.05814 by Sumit Chongder.

Figure 1
Figure 1. Figure 1: Hardware-aware quantum error correction co-design framework. Green solid boxes denote components implemented and empirically evaluated in this paper (noise model, circuit synthesis via Stim, adaptive confidence-gated decoder, and logical-metric evaluation). Grey dashed boxes denote the hardware-constrained code-generation and closed-loop latency-constrained optimisation components that are specified archit… view at source ↗
Figure 2
Figure 2. Figure 2: Hardware-aware co-design optimisation pipeline. The vertical chain (noise model, circuit compilation, syndrome simulation, adaptive decoding, evaluation) is the pipeline implemented and benchmarked in this paper. The hardware-constrained code generator and the multi-objective optimisation loop that would close the search over code candidates (right, dashed) are specified as the target architecture for the … view at source ↗
Figure 3
Figure 3. Figure 3: Decision flow of the adaptive, confidence-gated decoder. Every measured syndrome is first passed through the fast feed-forward decoder; syndromes with confidence at or above the operating threshold τ are corrected on the fast path, while low-confidence syndromes are escalated to exact minimum￾weight perfect matching refinement before recovery is applied. Algorithm 1 Adaptive confidence-gated decoding of a … view at source ↗
Figure 4
Figure 4. Figure 4: Fast-path logical accuracy versus code distance at fixed physical error rate p = 10−3 (data as in [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Decoder accuracy as a joint function of code distance and physical error probability. Full numerical values are given in [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Logical error rate versus physical error probability at d = 3, 5, 7, 9, and 11 (per-curve data in [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: End-to-end logical accuracy as a function of the escalation confidence threshold τ (data as in [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: shows the empirical distribution of fast-path confidence scores at d = 7. The distribution is heavily concentrated near c = 1.0, with a long, sparsely populated left tail extending down toward c = 0.5; it is precisely this left tail that the escalation mechanism of Section 5.3 targets. The shape of this distribution, rather than its mean, is what determines the cost-effectiveness of the routing mechanism: … view at source ↗
Figure 9
Figure 9. Figure 9: Fast-path decoder accuracy and throughput as a function of physical error probability at fixed d = 7 (data as in [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Fast-path decoder throughput as a function of inference batch size on a single CPU core (data as in [PITH_FULL_IMAGE:figures/full_fig_p016_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Structural resource scaling of the rotated surface code memory circuits used in this study, as a function of code distance (data as in [PITH_FULL_IMAGE:figures/full_fig_p017_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Estimated decoding-graph memory as a function of code distance, computed from detector and edge counts ( [PITH_FULL_IMAGE:figures/full_fig_p018_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Runtime scaling with code distance (left) and the corresponding runtime-versus-logical-error￾rate scatter across the tested distances (right). As discussed in Section 5.8, the right panel is descriptive of the single decoder configuration benchmarked here rather than a multi-candidate Pareto search. small, disproportionately error-prone subset of syndromes. The gap we observe between mean confidence and m… view at source ↗

Review history (2 revisions) →

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

34 extracted references · 34 canonical work pages · 2 internal anchors

  1. [1]

    Scheme for reducing decoherence in quantum computer memory,

    P. W. Shor, “Scheme for reducing decoherence in quantum computer memory,”Phys. Rev. A, vol. 52, pp. R2493–R2496, 1995

    work page 1995

  2. [2]

    Error correcting codes in quantum theory,

    A. M. Steane, “Error correcting codes in quantum theory,”Phys. Rev. Lett., vol. 77, pp. 793–797, 1996

    work page 1996

  3. [3]

    Fault-tolerant quantum computation by anyons,

    A. Y. Kitaev, “Fault-tolerant quantum computation by anyons,”Ann. Phys., vol. 303, no. 1, pp. 2–30, 2003

    work page 2003

  4. [4]

    Topological quantum memory,

    E. Dennis, A. Kitaev, A. Landahl, and J. Preskill, “Topological quantum memory,”J. Math. Phys., vol. 43, no. 9, pp. 4452–4505, 2002

    work page 2002

  5. [5]

    Surface codes: Towards practical large-scale quantum computation,

    A. G. Fowler, M. Mariantoni, J. M. Martinis, and A. N. Cleland, “Surface codes: Towards practical large-scale quantum computation,”Phys. Rev. A, vol. 86, p. 032324, 2012

    work page 2012

  6. [6]

    Exponential suppression of bit or phase errors with cyclic error correction,

    Google Quantum AI, “Exponential suppression of bit or phase errors with cyclic error correction,”Nature, vol. 595, pp. 383–387, 2021

    work page 2021

  7. [7]

    Suppressing quantum errors by scaling a surface code logical qubit,

    Google Quantum AI, “Suppressing quantum errors by scaling a surface code logical qubit,” Nature, vol. 614, pp. 676–681, 2023

    work page 2023

  8. [8]

    Quantum error correction for quantum memories,

    B. M. Terhal, “Quantum error correction for quantum memories,”Rev. Mod. Phys., vol. 87, pp. 307–346, 2015

    work page 2015

  9. [9]

    Roads towards fault-tolerant universal quantum computation,

    E. T. Campbell, B. M. Terhal, and C. Vuillot, “Roads towards fault-tolerant universal quantum computation,”Nature, vol. 549, pp. 172–179, 2017

    work page 2017

  10. [10]

    PyMatching: A Python package for decoding quantum codes with minimum- weight perfect matching,

    O. Higgott, “PyMatching: A Python package for decoding quantum codes with minimum- weight perfect matching,”ACM Trans. Quantum Comput., vol. 3, no. 3, pp. 1–16, 2022

    work page 2022

  11. [11]

    Sparse Blossom: correcting a million errors per core second with minimum-weight matching

    O. Higgott and C. Gidney, “Sparse Blossom: correcting a million errors per core second with minimum-weight perfect matching,” arXiv:2303.15933, 2023

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  12. [12]

    Degenerate quantum LDPC codes with good finite length performance,

    P. Panteleev and G. Kalachev, “Degenerate quantum LDPC codes with good finite length performance,”Quantum, vol. 5, p. 585, 2021

    work page 2021

  13. [13]

    Decoding across the quantum LDPC code landscape,

    J. Roffe, D. R. White, S. Burton, and E. Campbell, “Decoding across the quantum LDPC code landscape,”Phys. Rev. Res., vol. 2, p. 043423, 2020

    work page 2020

  14. [14]

    Almost-linear time decoding algorithm for topological codes,

    N. Delfosse and N. H. Nickerson, “Almost-linear time decoding algorithm for topological codes,”Quantum, vol. 5, p. 595, 2021

    work page 2021

  15. [15]

    Parallel window decoding enables scalable fault tolerant quantum computation,

    L. Skoric, D. E. Browne, K. M. Barnes, N. I. Gillespie, and E. T. Campbell, “Parallel window decoding enables scalable fault tolerant quantum computation,”Nat. Commun., vol. 14, p. 7040, 2023. 22

    work page 2023

  16. [16]

    Hierarchical decoding to reduce hardware requirements for quantum computing

    N. Delfosse, “Hierarchical decoding to reduce hardware requirements for quantum comput- ing,” arXiv:2001.11427, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2001

  17. [17]

    Scalable surface-code decoders with parallelization in time,

    X. Tan, F. Zhang, R. Chao, Y. Shi, and J. Chen, “Scalable surface-code decoders with parallelization in time,”PRX Quantum, vol. 4, p. 040344, 2023

    work page 2023

  18. [18]

    Better than worst-case decoding for quantum error correction,

    G. S. Ravi, J. Viszlai, F. Hua, K. N. Smith, J. Kim, K. Heckey, K. Thangaraj, T. Tomesh, and F. T. Chong, “Better than worst-case decoding for quantum error correction,” inProc. 28th ACM Int. Conf. Architectural Support for Programming Languages and Operating Systems (ASPLOS), pp. 88–102, 2023

    work page 2023

  19. [19]

    Neural decoder for topological codes,

    G. Torlai and R. G. Melko, “Neural decoder for topological codes,”Phys. Rev. Lett., vol. 119, p. 030501, 2017

    work page 2017

  20. [20]

    Decoding small surface codes with feedforward neural networks,

    S. Varsamopoulos, B. Criger, and K. Bertels, “Decoding small surface codes with feedforward neural networks,”Quantum Sci. Technol., vol. 3, p. 015004, 2017

    work page 2017

  21. [21]

    Machine-learning- assisted correction of correlated qubit errors in a topological code,

    P. Baireuther, T. E. O’Brien, B. Tarasinski, and C. W. J. Beenakker, “Machine-learning- assisted correction of correlated qubit errors in a topological code,”Quantum, vol. 2, p. 48, 2018

    work page 2018

  22. [22]

    Deep neural decoders for near term fault-tolerant experi- ments,

    C. Chamberland and P. Ronagh, “Deep neural decoders for near term fault-tolerant experi- ments,”Quantum Sci. Technol., vol. 3, p. 044002, 2018

    work page 2018

  23. [23]

    Scalable neural decoder for topological surface codes,

    K. Meinerz, C.-Y. Park, and S. Trebst, “Scalable neural decoder for topological surface codes,”Phys. Rev. Lett., vol. 128, p. 080505, 2022

    work page 2022

  24. [24]

    Data-driven decoding of quantum error correcting codes using graph neural networks,

    M. Lange, W. Havström, S. Srivastava, P. Bengtsson, M. Bergentall, K. Hammar, J. Heuts, E. van Nieuwenburg, and J. Granath, “Data-driven decoding of quantum error correcting codes using graph neural networks,”Phys. Rev. Research, vol. 7, p. 023181, 2025

    work page 2025

  25. [25]

    Neural-network decoders for quantum error correction using surface codes: A space exploration of the hardware cost-performance tradeoffs,

    R. W. J. Overwater, F. Sebastiano, and E. Charbon, “Neural-network decoders for quantum error correction using surface codes: A space exploration of the hardware cost-performance tradeoffs,”IEEE Trans. Quantum Eng., vol. 3, pp. 1–19, 2022

    work page 2022

  26. [26]

    Learning high-accuracy error decoding for quantum processors,

    J. Bausch, A. W. Senior, F. J. H. Heras, T. Edlich, A. Davies, M. Newman, C. Jones, K. Satzinger, M. Y. Niu, S. Blackwell,et al., “Learning high-accuracy error decoding for quantum processors,”Nature, vol. 635, pp. 834–840, 2024

    work page 2024

  27. [27]

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

    S. Bravyi, A. W. Cross, J. M. Gambetta, D. Maslov, P. Rall, and T. J. Yoder, “High- threshold and low-overhead fault-tolerant quantum memory,”Nature, vol. 627, pp. 778–782, 2024

    work page 2024

  28. [28]

    Optimizing quantum error correction codes with reinforcement learning,

    H. P. Nautrup, N. Delfosse, V. Dunjko, H. J. Briegel, and N. Friis, “Optimizing quantum error correction codes with reinforcement learning,”Quantum, vol. 3, p. 215, 2019

    work page 2019

  29. [29]

    Streamlining quantum error correction and application development with CUDA-QX,

    NVIDIA Corporation, “Streamlining quantum error correction and application development with CUDA-QX,” NVIDIA Developer Blog, 2024.https://developer.nvidia.com/blog/

    work page 2024

  30. [30]

    Stim: a fast stabilizer circuit simulator,

    C. Gidney, “Stim: a fast stabilizer circuit simulator,”Quantum, vol. 5, p. 497, 2021

    work page 2021

  31. [31]

    Low-distance surface codes under realistic quantum noise,

    Y. Tomita and K. M. Svore, “Low-distance surface codes under realistic quantum noise,” Phys. Rev. A, vol. 90, p. 062320, 2014

    work page 2014

  32. [32]

    Surface code quantum computing with error rates over 1%,

    D. S. Wang, A. G. Fowler, and L. C. L. Hollenberg, “Surface code quantum computing with error rates over 1%,”Phys. Rev. A, vol. 83, p. 020302(R), 2011. 23

    work page 2011

  33. [33]

    Rapid object detection using a boosted cascade of simple features,

    P. Viola and M. Jones, “Rapid object detection using a boosted cascade of simple features,” inProc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), vol. 1, pp. I–511, 2001

    work page 2001

  34. [34]

    Outra- geously large neural networks: The sparsely-gated mixture-of-experts layer,

    N. Shazeer, A. Mirhoseini, K. Maziarz, A. Davis, Q. Le, G. Hinton, and J. Dean, “Outra- geously large neural networks: The sparsely-gated mixture-of-experts layer,” inProc. Int. Conf. Learning Representations (ICLR), 2017. 24 A Additional algorithmic detail Algorithm 2 describes the training procedure used to obtain the fast-path decoder evaluated through...

    work page 2017

This paper was first reviewed by grok-4.5 on July 11, 2026.