Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

arXiv:2609.17896

QEMScore: How Much Does the Measurement Add to Learned Quantum Error Mitigation?

Yue Zhao, Huayue Gu, Yushun Dong, Xiyang Hu

quant-phcs.LGSubmitted Sep 15, 2026• Announced Sep 17, 2026• 0 citations
arXivHTMLPDF

Abstract

How much does the noisy measurement add to learned quantum error mitigation? An accuracy table cannot say, because a model handed circuit structure can score well without reading the measurement at all. QEMScore adds the comparison that can. Each simulated circuit carries an exact ideal answer. The learned mitigator is scored beside a capacity-matched control, a model just as flexible that reads the same circuit description but never the measurement. Each method's measurement spend is accounted and not equalized. We run a controlled campaign on simulated circuits and reanalyze two published learned mitigators, Q-LEAR and QRAFT, from their released hardware data. Three findings stand out. First, under familiar within-family conditions (S0) evaluated across two spin-chain families and three seeds, continuous couplings identify the target, and the control that never reads the measurement matches 87.7 to 100.5 percent of the mitigator's gain over an affine fit to the circuit description. A plain polynomial in the coupling parameters, fitted after the campaign, beats the mitigator on all six evaluations, reflecting the selected learners' capacity. Second, for these selected learners, matching most of the gain is not matching the accuracy: on five of six evaluations the mitigator removes 19.5 to 74.5 percent of the error the capacity-matched control leaves, a learner-specific gap rather than a measurement requirement. Third, on released hardware data where descriptors only partially identify queries, the findings differ: flexible models of the descriptors show negligible mean gain over affine fits in Q-LEAR, and measurement inputs carry predictive gains in both Q-LEAR and QRAFT. These comparisons reflect representation- and protocol-specific behavior rather than an isolated cross-regime difference. A learned mitigator's accuracy should therefore be reported beside such controls.

AI Summary

Learned quantum error mitigation can predict ideal outputs from circuit descriptors without using noisy measurements. QEMScore measures the additional predictive benefit of measurement inputs by comparing a full learner with a descriptor-only control using the same candidate models and selection rule. It also accounts for measurement costs, without equalizing budgets. The controlled campaign uses ten-qubit transverse-field Ising and Heisenberg circuits, depolarizing-plus-readout noise, exact simulation labels, and held-out macro mean absolute error. Across six primary family–seed evaluations, capacity-matched descriptor controls reproduce 87.7–100.5% of the full learner’s improvement over an affine descriptor baseline. Nevertheless, measurements reduce the control’s remaining error by 19.5–74.5% on five evaluations; the sixth gap spans zero. These intervals bootstrap physical circuits conditional on fitted models. Because the sampled couplings determine the ideal target, these gaps are learner-specific, not irreducible measurement benefits. A post-campaign degree-five polynomial using couplings alone beats the full learner on all six evaluations. Released-data reanalyses show measurement-associated gains: Q-LEAR’s nonlinear descriptor control adds negligible mean improvement over its affine counterpart, while base statistics and depth-cut features reduce Hellinger error by 0.1969 and 0.0222, respectively. QRAFT’s descriptor-only error is roughly six times higher than with forward probabilities. Limitations include testing only familiar-condition generalization, two circuit families and one noise family; omitting executable CDR/vnCDR comparisons; and not matching measurement budgets. Hardware reanalyses condition on execution-selected output support and released splits, do not establish unseen-circuit generalization, and do not reproduce QRAFT’s published objective or results.

Likely largely AI-generatedIncrementalSmall numerics onlyMissing comparison

Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.

Comments
55 pages, including references and appendices

Related papers

Machine Learning for Quantum Error Mitigation: A Systematic Review of Techniques, Evidence, and Limits

Mohammad Faridul Haque Siddiqui

Quantum error mitigation (QEM) reduces the bias that hardware noise introduces into estimated observables, at the cost of additional circuit repetitions. Machine learning has been applied to it extensively since 2020, with reported reductions in that overhead. This systematic review synthesises that literature for the period January 2020 to September 2026. The review addresses three questions. First, whether the literature divides more informatively by what a model learns — a noise model, an extrapolation map, a direct noisy-to-ideal correction, or a circuit-level policy — than by its architecture. Second, what fraction of published work reports accuracy validation outside the classically simulable regime, given that supervised mitigation requires ideal reference values obtainable only where a circuit is classically tractable. Third, whether reported gains can be placed on a common scale, and how they stand against established information-theoretic lower bounds on the sampling cost of mitigation. Conducted per PRISMA 2020 with documented adaptations for a non-clinical computing literature.

•Sep 19, 2026•0 citations

Comparing Simulated versus Hardware Training Data for Learning-Based Quantum Error Mitigation

Seyed Mohamad Ali Tousi, Guilherme N. DeSouza

Learning-based Quantum Error Mitigation (QEM) has emerged as a promising approach for reducing noise-induced bias on near-term Noisy Intermediate-Scale Quantum (NISQ) hardware. While simulated noise models enable scalable training of learning-based mitigators, real hardware data more accurately reflect device-specific error mechanisms, but are expensive and sparse. Understanding how these two data regimes interact is therefore critical for the practical deployment of learning-based QEM. In this paper, we present a systematic and empirical analysis on the nature of the data (simulated versus hardware noise) for learning-based QEM. Using a recently developed attention-based graph transformer for small- and large-scale quantum error mitigation, we compare various training regimes, starting from exclusively simulated noise and progressively integrating more measurements from real quantum hardware. Our study spans two distinct circuit families: 1) randomly generated circuits with relatively homogeneous structure; and 2) targeted quantum circuits derived from a specific problem, in this case the one-dimensional Burgers’ equation. Across five-fold cross validation, we analyze how hardware data affects generalization performance as measured by mean absolute error (MAE). Our results demonstrate that the impact of hardware data strongly depends on circuit structure and diversity. For random circuits, simulated data already guarantees lower absolute error, while hardware data yields consistent, but moderate improvements. In contrast, for targeted circuits, like the structured Burgers’ circuits, simulated noise can be less representative of the actual (testing) quantum error, and incorporating hardware data produces substantially larger relative reductions in MAE – despite the intrinsically higher MAE. These findings indicate the value of the nature of the training data in learning-based QEM, which is governed not only by the data’s ability to approximate reality – as expected – but also by how difficult it may be to capture circuit-induced error mechanisms in quantum simulations. In general, this work highlights the urgency of quantum circuit coverage in designing scalable and cost-effective learning-based QEM pipelines for devices in this new NISQ-era.

International Joint Conference on Neural Networks (IJCNN)•Jun 1, 2026•0 citations

Accuracy, Decision Risk, and Sampling Cost in Quantum Error Mitigation

Vicenzo Scavino Alfaro

Quantum error mitigation estimates expectation values that are often used to select or rank candidates. We study how estimation accuracy, decision risk and sampling cost are related in these tasks. The joint law of residual differences determines the risk of a shift-invariant decision; in Gaussian regimes, effective margins and their covariance provide a computable representation. This connects mitigation maps, shared measurement records and shot allocation, and yields an extrapolation design criterion under certified margins. Finite-shot QAOA-MaxCut controls illustrate two forms of this relationship: a positive affine CDR correction reduces mean-squared error while preserving rankings, whereas signed PEC sampling reduces mean-squared error and increases selection failures at equal evaluation cost. A retrospective, oracle-assisted study of 48 landscapes compares Gaussian-risk and MSE-based method selection using separate fitting, selection and evaluation samples. At the primary total shot budget, with learning amortized over 20 decisions per landscape, failure rates are $0.3083$ for Gaussian-risk selection, $0.4719$ for MSE selection and $0.2333$ for unmitigated estimation. These comparisons describe how uncertainty in candidate differences and the allocation of measurements between learning and evaluation jointly enter the assessment of mitigation.

quant-pharXiv:2607.02888•Jul 6, 2026•2 citations

Weakly Supervised Quantum Error Mitigation

Seyed Mohamad Ali Tousi, G. N. DeSouza

Supervised approaches to quantum error mitigation learn a map from noisy circuit outputs to ideal ones, and therefore require the ideal outputs. Producing those ideal outputs demands noiseless classical simulation, whose cost grows exponentially with system size, so supervision is unavailable in exactly the regime where mitigation matters most. We ask whether cheap, individually unreliable signals drawn from circuit structure and hardware calibration can take the place of ideal labels. We assemble sixteen heuristic labeling functions (stabilizer and parity constraints, relaxation and readout characteristics, local depth, gate counts, and neighboring activity), reconcile their disagreements with a probabilistic label model, and read the resulting per-qubit error probabilities as a readout channel whose inverse mitigates the measured distribution. No ideal output enters the training path. On $147{,}000$ five-qubit circuits executed on two IBM devices, the method removes $24.3\%$ (Algiers) and $28.8\%$ (Hanoi) of the Kullback-Leibler divergence to the ideal distribution, against $15.4\%$ and $21.5\%$ for the strongest published analytical baseline, a margin that holds on both devices and lies far outside its bootstrap interval. Supervised neural models trained on ideal distributions remain stronger where such labels exist, and we quantify that gap rather than setting it aside; the method's claim is to the regime where they do not, since the labels they require cannot be computed for the circuits mitigation is needed for. The codes will be released shortly.

arXiv:2609.25555•Sep 22, 2026•0 citations

Scalable Quantum Error Mitigation with Neighbor-Informed Learning

Zhenyu Chen, Bin Cheng, Minbo Gao +4 more

Noise in quantum hardware is the primary obstacle to realizing the transformative potential of quantum computing. Quantum error mitigation (QEM) offers a promising pathway to enhance computational accuracy on near-term devices, yet existing methods face a difficult trade-off between performance, resource overhead, and theoretical guarantees. In this work, we introduce neighbor-informed learning (NIL), a versatile and scalable QEM framework that unifies and strengthens existing methods such as zero-noise extrapolation (ZNE) and probabilistic error cancellation (PEC), while offering improved flexibility, accuracy, efficiency, and robustness. NIL learns to predict the ideal output of a target quantum circuit from the noisy outputs of its structurally related ``neighbor'' circuits. A key innovation is our 2-design training method, which generates training data for our machine learning model. In contrast to conventional learning-based QEM protocols that create training circuits by replacing non-Clifford gates with uniformly random Clifford gates, our approach achieves higher accuracy and efficiency, as demonstrated by both theoretical analysis and numerical simulation. Furthermore, we prove that the required size of the training set scales only \emph{logarithmically} with the total number of neighbor circuits, enabling NIL to be applied to problems involving large-scale quantum circuits. Our work establishes a theoretically grounded and practically efficient framework for QEM, paving a viable path toward achieving quantum advantage on noisy hardware.

quant-pharXiv:2512.12578•Dec 16, 2025•1 citation

Deep Learning Approaches to Quantum Error Mitigation

Leonardo Placidi, Ifan Williams, Enrico Rinaldi +4 more

We present a systematic investigation of deep learning methods applied to quantum error mitigation of noisy output probability distributions from measured quantum circuits. We compare different architectures, from fully connected neural networks to transformers, and we test different design/training modalities, identifying sequence-to-sequence, attention-based models as the most effective on our datasets. These models consistently produce mitigated distributions that are closer to the ideal outputs when tested on both simulated and real device data obtained from IBM superconducting quantum processing units (QPU) up to five qubits. Across several different circuit depths, our approach outperforms other baseline error mitigation techniques. We perform a series of ablation studies to examine: how different input features (circuit, device properties, noisy output statistics) affect performance; cross-dataset generalization across circuit families; and transfer learning to a different IBM QPU. We observe that generalization performance across similar devices with the same architecture works effectively, without needing to fully retrain models.

quant-pharXiv:2601.14226•Jan 21, 2026•1 citation
How much does the noisy measurement add to learned quantum error mitigation? An accuracy table cannot say, because a model handed circuit structure can score well without reading the measurement at all. QEMScore adds the comparison that can. Each simulated circuit carries an exact ideal answer. The learned mitigator is scored beside a capacity-matched control, a model just as flexible that reads the same circuit description but never the measurement. Each method's measurement spend is accounted and not equalized. We run a controlled campaign on simulated circuits and reanalyze two published learned mitigators, Q-LEAR and QRAFT, from their released hardware data. Three findings stand out. First, under familiar within-family conditions (S0) evaluated across two spin-chain families and three seeds, continuous couplings identify the target, and the control that never reads the measurement matches 87.7 to 100.5 percent of the mitigator's gain over an affine fit to the circuit description. A plain polynomial in the coupling parameters, fitted after the campaign, beats the mitigator on all six evaluations, reflecting the selected learners' capacity. Second, for these selected learners, matching most of the gain is not matching the accuracy: on five of six evaluations the mitigator removes 19.5 to 74.5 percent of the error the capacity-matched control leaves, a learner-specific gap rather than a measurement requirement. Third, on released hardware data where descriptors only partially identify queries, the findings differ: flexible models of the descriptors show negligible mean gain over affine fits in Q-LEAR, and measurement inputs carry predictive gains in both Q-LEAR and QRAFT. These comparisons reflect representation- and protocol-specific behavior rather than an isolated cross-regime difference. A learned mitigator's accuracy should therefore be reported beside such controls.
Learned quantum error mitigation can predict ideal outputs from circuit descriptors without using noisy measurements. QEMScore measures the additional predictive benefit of measurement inputs by comparing a full learner with a descriptor-only control using the same candidate models and selection rule. It also accounts for measurement costs, without equalizing budgets.
The controlled campaign uses ten-qubit transverse-field Ising and Heisenberg circuits, depolarizing-plus-readout noise, exact simulation labels, and held-out macro mean absolute error. Across six primary family–seed evaluations, capacity-matched descriptor controls reproduce 87.7–100.5% of the full learner’s improvement over an affine descriptor baseline. Nevertheless, measurements reduce the control’s remaining error by 19.5–74.5% on five evaluations; the sixth gap spans zero. These intervals bootstrap physical circuits conditional on fitted models. Because the sampled couplings determine the ideal target, these gaps are learner-specific, not irreducible measurement benefits. A post-campaign degree-five polynomial using couplings alone beats the full learner on all six evaluations.
Released-data reanalyses show measurement-associated gains: Q-LEAR’s nonlinear descriptor control adds negligible mean improvement over its affine counterpart, while base statistics and depth-cut features reduce Hellinger error by 0.1969 and 0.0222, respectively. QRAFT’s descriptor-only error is roughly six times higher than with forward probabilities.
Limitations include testing only familiar-condition generalization, two circuit families and one noise family; omitting executable CDR/vnCDR comparisons; and not matching measurement budgets. Hardware reanalyses condition on execution-selected output support and released splits, do not establish unseen-circuit generalization, and do not reproduce QRAFT’s published objective or results.
Quantum error mitigation (QEM) reduces the bias that hardware noise introduces into estimated observables, at the cost of additional circuit repetitions. Machine learning has been applied to it extensively since 2020, with reported reductions in that overhead. This systematic review synthesises that literature for the period January 2020 to September 2026. The review addresses three questions. First, whether the literature divides more informatively by what a model learns — a noise model, an extrapolation map, a direct noisy-to-ideal correction, or a circuit-level policy — than by its architecture. Second, what fraction of published work reports accuracy validation outside the classically simulable regime, given that supervised mitigation requires ideal reference values obtainable only where a circuit is classically tractable. Third, whether reported gains can be placed on a common scale, and how they stand against established information-theoretic lower bounds on the sampling cost of mitigation. Conducted per PRISMA 2020 with documented adaptations for a non-clinical computing literature.
Learning-based Quantum Error Mitigation (QEM) has emerged as a promising approach for reducing noise-induced bias on near-term Noisy Intermediate-Scale Quantum (NISQ) hardware. While simulated noise models enable scalable training of learning-based mitigators, real hardware data more accurately reflect device-specific error mechanisms, but are expensive and sparse. Understanding how these two data regimes interact is therefore critical for the practical deployment of learning-based QEM. In this paper, we present a systematic and empirical analysis on the nature of the data (simulated versus hardware noise) for learning-based QEM. Using a recently developed attention-based graph transformer for small- and large-scale quantum error mitigation, we compare various training regimes, starting from exclusively simulated noise and progressively integrating more measurements from real quantum hardware. Our study spans two distinct circuit families: 1) randomly generated circuits with relatively homogeneous structure; and 2) targeted quantum circuits derived from a specific problem, in this case the one-dimensional Burgers’ equation. Across five-fold cross validation, we analyze how hardware data affects generalization performance as measured by mean absolute error (MAE). Our results demonstrate that the impact of hardware data strongly depends on circuit structure and diversity. For random circuits, simulated data already guarantees lower absolute error, while hardware data yields consistent, but moderate improvements. In contrast, for targeted circuits, like the structured Burgers’ circuits, simulated noise can be less representative of the actual (testing) quantum error, and incorporating hardware data produces substantially larger relative reductions in MAE – despite the intrinsically higher MAE. These findings indicate the value of the nature of the training data in learning-based QEM, which is governed not only by the data’s ability to approximate reality – as expected – but also by how difficult it may be to capture circuit-induced error mechanisms in quantum simulations. In general, this work highlights the urgency of quantum circuit coverage in designing scalable and cost-effective learning-based QEM pipelines for devices in this new NISQ-era.
Quantum error mitigation estimates expectation values that are often used to select or rank candidates. We study how estimation accuracy, decision risk and sampling cost are related in these tasks. The joint law of residual differences determines the risk of a shift-invariant decision; in Gaussian regimes, effective margins and their covariance provide a computable representation. This connects mitigation maps, shared measurement records and shot allocation, and yields an extrapolation design criterion under certified margins. Finite-shot QAOA-MaxCut controls illustrate two forms of this relationship: a positive affine CDR correction reduces mean-squared error while preserving rankings, whereas signed PEC sampling reduces mean-squared error and increases selection failures at equal evaluation cost. A retrospective, oracle-assisted study of 48 landscapes compares Gaussian-risk and MSE-based method selection using separate fitting, selection and evaluation samples. At the primary total shot budget, with learning amortized over 20 decisions per landscape, failure rates are 0.30830.30830.3083 for Gaussian-risk selection, 0.47190.47190.4719 for MSE selection and 0.23330.23330.2333 for unmitigated estimation. These comparisons describe how uncertainty in candidate differences and the allocation of measurements between learning and evaluation jointly enter the assessment of mitigation.
Supervised approaches to quantum error mitigation learn a map from noisy circuit outputs to ideal ones, and therefore require the ideal outputs. Producing those ideal outputs demands noiseless classical simulation, whose cost grows exponentially with system size, so supervision is unavailable in exactly the regime where mitigation matters most. We ask whether cheap, individually unreliable signals drawn from circuit structure and hardware calibration can take the place of ideal labels. We assemble sixteen heuristic labeling functions (stabilizer and parity constraints, relaxation and readout characteristics, local depth, gate counts, and neighboring activity), reconcile their disagreements with a probabilistic label model, and read the resulting per-qubit error probabilities as a readout channel whose inverse mitigates the measured distribution. No ideal output enters the training path. On 147,000147{,}000147,000 five-qubit circuits executed on two IBM devices, the method removes 24.3%24.3\%24.3% (Algiers) and 28.8%28.8\%28.8% (Hanoi) of the Kullback-Leibler divergence to the ideal distribution, against 15.4%15.4\%15.4% and 21.5%21.5\%21.5% for the strongest published analytical baseline, a margin that holds on both devices and lies far outside its bootstrap interval. Supervised neural models trained on ideal distributions remain stronger where such labels exist, and we quantify that gap rather than setting it aside; the method's claim is to the regime where they do not, since the labels they require cannot be computed for the circuits mitigation is needed for. The codes will be released shortly.
Noise in quantum hardware is the primary obstacle to realizing the transformative potential of quantum computing. Quantum error mitigation (QEM) offers a promising pathway to enhance computational accuracy on near-term devices, yet existing methods face a difficult trade-off between performance, resource overhead, and theoretical guarantees. In this work, we introduce neighbor-informed learning (NIL), a versatile and scalable QEM framework that unifies and strengthens existing methods such as zero-noise extrapolation (ZNE) and probabilistic error cancellation (PEC), while offering improved flexibility, accuracy, efficiency, and robustness. NIL learns to predict the ideal output of a target quantum circuit from the noisy outputs of its structurally related ``neighbor'' circuits. A key innovation is our 2-design training method, which generates training data for our machine learning model. In contrast to conventional learning-based QEM protocols that create training circuits by replacing non-Clifford gates with uniformly random Clifford gates, our approach achieves higher accuracy and efficiency, as demonstrated by both theoretical analysis and numerical simulation. Furthermore, we prove that the required size of the training set scales only \emph{logarithmically} with the total number of neighbor circuits, enabling NIL to be applied to problems involving large-scale quantum circuits. Our work establishes a theoretically grounded and practically efficient framework for QEM, paving a viable path toward achieving quantum advantage on noisy hardware.
We present a systematic investigation of deep learning methods applied to quantum error mitigation of noisy output probability distributions from measured quantum circuits. We compare different architectures, from fully connected neural networks to transformers, and we test different design/training modalities, identifying sequence-to-sequence, attention-based models as the most effective on our datasets. These models consistently produce mitigated distributions that are closer to the ideal outputs when tested on both simulated and real device data obtained from IBM superconducting quantum processing units (QPU) up to five qubits. Across several different circuit depths, our approach outperforms other baseline error mitigation techniques. We perform a series of ablation studies to examine: how different input features (circuit, device properties, noisy output statistics) affect performance; cross-dataset generalization across circuit families; and transfer learning to a different IBM QPU. We observe that generalization performance across similar devices with the same architecture works effectively, without needing to fully retrain models.