Noise-Agnostic Quantum Error Mitigation with Data Augmented Neural Models

TL;DR

Quantum error mitigation on near-term devices often requires knowledge of the noise model or noise-free data, which is impractical in many settings. The authors propose DAEM, a data augmentation-empowered neural approach that learns to undo noise without access to ideal data by using a fiducial process to generate augmented training data. DAEM can handle circuits, many-body dynamics, and continuous-variable dynamics across diverse noise models, including non-Markovian. Empirical results on VQE, swap test, QAOA, a 50-qubit Ising dynamics, and Kerr/ CV dynamics show superior or competitive performance relative to ZNE and CDR, with strong transferability to unseen circuits and hardware. This work reduces reliance on noise characterization and enables scalable error mitigation for NISQ-era quantum technologies.

Abstract

Quantum error mitigation, a data processing technique for recovering the statistics of target processes from their noisy version, is a crucial task for near-term quantum technologies. Most existing methods require prior knowledge of the noise model or the noise parameters. Deep neural networks have a potential to lift this requirement, but current models require training data produced by ideal processes in the absence of noise. Here we build a neural model that achieves quantum error mitigation without any prior knowledge of the noise and without training on noise-free data. To achieve this feature, we introduce a quantum augmentation technique for error mitigation. Our approach applies to quantum circuits and to the dynamics of many-body and continuous-variable quantum systems, accommodating various types of noise models. We demonstrate its effectiveness by testing it both on simulated noisy circuits and on real quantum hardware.

Figures (13)

• Figure 1: Framework of DAEM model. The entire procedure is divided into two phases. In the first phase, known as Noise-Awareness phase, we train the neural model for error mitigation with the assistance of a fiducial process responsible for data augmentation. In the second phase, known as Error-Mitigation phase, we apply the trained neural model to mitigate the errors in noisy measurement statistics collected from the noisy version of the target quantum process.
• Figure 2: Error mitigation for variational quantum eigensolvers. a. The variational ansatz for preparing the ground states of 4-qubit transverse Ising models. b. Mean Absolute Errors (MAE) between the mitigated measurement expectation values for phase damping noise model and ideal expectation values. c. Mean Absolute Errors (MAE) between the mitigated measurement expectation values for amplitude damping noise model and ideal expectation values. d. Mean Absolute Errors (MAE) between the mitigated measurement expectation values for phase damping noise model and ideal expectation values for the circuits not included in the training. e. Mean Absolute Errors (MAE) between the mitigated measurement expectation values for amplitude damping noise model and ideal expectation values for the circuits not included in the training. It is noteworthy that ZNE requires knowledge of noise parameters associated with statistics while our proposed DAEM not. Despite being under an unfair comparison, DAEM still demonstrates superior performance.
• Figure 3: Error mitigation for variational quantum eigensolvers affected by Non-Markovian Noise. a. Schematic diagram of quantum circuits affected by Non-Markovian noise. b. Mean Absolute Errors (MAE) between the mitigated measurement expectation values for considered Non-Markovian noise model and ideal expectation values.
• Figure 4: Error mitigation for variational quantum eigensolvers on the OriginQ Cloud quantum hardware. a. The variational ansatz for preparing the ground states of 4-qubit transverse Ising models. b. The structure of fiducial circuits. c. Mean Absolute Errors (MAE) between the mitigated measurement expectation values and ideal expectation values. The average MAE of w/o, DAEM, ZNE, CDR are 0.247, 0.067, 0.259, and 0.095 respectively.
• Figure 5: Error mitigation for the swap test. a. The swap test circuit for comparing two 5-qubit states. The gate within the green box is the controlled-SWAP gate. b. Mean Absolute Errors (MAE) between the mitigated fidelity values and the ground truth values.