Quantum state readout with minimal resources is crucial for scalable quantum information processing. As a leading platform, neutral atom arrays rely on fluorescence readout, requiring short-exposure schemes to mitigate heating and atom loss. However, a fundamental challenge arises in the single-photon regime where severe overlap in state distributions causes conventional threshold discrimination to fail. Here, we report a neural-network-assisted Bayesian inference method for fluorescence readout in neutral atom arrays. We introduce a weakly anchored Bayesian scheme that requires calibration of only one state, addressing asymmetric calibration challenges common across quantum platforms. Acceleration via a permutation-invariant neural network, yields a 100-fold speedup by compressing Bayesian inference into a single forward pass. The approach achieves relative readout fidelity above 99% and 98% for histogram overlaps of 61% and 72%, respectively, enabling reliable extraction of Rabi oscillations and Ramsey interference-results unattainable with conventional threshold-based methods.
Quantum state readout with minimal resources is crucial for scalable quantum information processing. As a leading platform, neutral atom arrays rely on fluorescence readout, requiring short-exposure schemes to mitigate heating and atom loss. However, a fundamental challenge arises in the single-photon regime where severe overlap in state distributions causes conventional threshold discrimination to fail. Here, we report a neural-network-assisted Bayesian inference method for fluorescence readout in neutral atom arrays. We introduce a weakly anchored Bayesian scheme that requires calibration of only one state, addressing asymmetric calibration challenges common across quantum platforms. Acceleration via a permutation-invariant neural network, yields a 100-fold speedup by compressing Bayesian inference into a single forward pass. The approach achieves relative readout fidelity above 99% and 98% for histogram overlaps of 61% and 72%, respectively, enabling reliable extraction of Rabi oscillations and Ramsey interference-results unattainable with conventional threshold-based methods.
Quantum state readout with minimal resources is crucial for scalable quantum information processing. As a leading platform, neutral atom arrays rely on atomic fluorescence imaging for qubit readout, requiring short exposure, low photon count schemes to mitigate heating and atom loss while enabling mid-circuit feedback. However, a fundamental challenge arises in the single-photon regime where severe overlap in state distributions causes conventional threshold discrimination to fail. Here, we report an AI-accelerated Bayesian inference method for fluorescence readout in neutral atom arrays. Our approach leverages Bayesian inference to achieve reliable state detection at the single-photon level under short exposure. Specifically, we introduce a weakly anchored Bayesian scheme that requires calibration of only one state, addressing asymmetric calibration challenges common across quantum platforms. Furthermore, acceleration is achieved via a permutation-invariant neural network, which yields a 100-fold speedup by compressing iterative inference into a single forward pass. The approach achieves relative readout fidelity above 99% and 98% for histogram overlaps of 61% and 72%, respectively, enabling reliable extraction of Rabi oscillations and Ramsey interference results unattainable with conventional threshold based methods. This framework supports scalable, real-time readout of large atom arrays and paves the way toward AI-enhanced quantum technology in computation and sensing.
We demonstrate qubit-state measurements assisted by a supervised convolutional neural network (CNN) in a neutral-atom quantum processor. We present two CNN architectures for analyzing neutral-atom qubit readout data: a compact five-layer single-qubit CNN architecture and a six-layer multiqubit CNN architecture. We benchmark both architectures against a conventional Gaussian-threshold analysis method. In a sparse array (9-$\text{\ensuremath{\mu}}\mathrm{m}$ atom separation) which experiences negligible crosstalk, we have observed up to 32% and 56% error reduction for the multiqubit and single-qubit architectures, respectively, as compared to the benchmark. In a tightly spaced array (5-$\text{\ensuremath{\mu}}\mathrm{m}$ atom separation), which suffers from readout crosstalk, we have observed up to 43% and 32% error reduction in the multiqubit and single-qubit CNN architectures, respectively, as compared to the benchmark. By examining the correlation between the predicted states of neighboring qubits, we have found that the multiqubit CNN architecture reduces the crosstalk correlation by up to 78.5%. This work demonstrates a proof of concept for a CNN network to be implemented as a real-time readout-processing method on a neutral-atom quantum computer, enabling faster readout time and improved fidelity.
Abstract Establishing low-error and fast detection methods for qubit readout is crucial for efficient quantum error correction. Here, we test neural networks to classify a collection of single-shot spin detection events, which are the readout signal of our qubit measurements. This readout signal contains a stochastic peak, for which a Bayesian inference filter including Gaussian noise is theoretically optimal. Hence, we benchmark our neural networks trained by various strategies versus this latter algorithm. Training of the network with 106 experimentally recorded single-shot readout traces does not improve the post-processing performance. A network trained by synthetically generated measurement traces performs similar in terms of the detection error and the post-processing speed compared to the Bayesian inference filter. This neural network turns out to be more robust to fluctuations in the signal offset, length and delay as well as in the signal-to-noise ratio. Notably, we find an increase of 7% in the visibility of the Rabi oscillation when we employ a network trained by synthetic readout traces combined with measured signal noise of our setup. Our contribution thus represents an example of the beneficial role which software and hardware implementation of neural networks may play in scalable spin qubit processor architectures.
Identifying single photon emitters from large, inhomogeneous candidate populations is key to realizing many quantum applications. This requires measuring the emitters second order autocorrelation function, whose statistical reliability is fundamentally limited by acquisition time. Machine learning classifiers can accelerate identification from sparse data, but their performance relative to physics-based inference has not been systematically examined. Here, we introduce sequential Bayesian inference for single photon emitter classification and benchmark it against Levenberg-Marquardt fitting and a feedforward neural network. We use synthetic training and test data calibrated against real Hanbury Brown-Twiss measurements from hexagonal boron nitride emitters, enabling evaluation against an exactly known ground-truth emitter number under realistic noise and background conditions. All three approaches achieve high, near-perfect accuracy with sufficient integration time, but differ greatly in convergence rate and robustness under sparse photon statistics. The neural network is most robust at short integration times. The Bayesian classifier reaches near-perfect accuracy fastest, while retaining full physical interpretability. Levenberg-Marquardt fitting remains a valuable, fully interpretable method, achieving the highest recall despite being the slowest to converge. These results lead to several key conclusions. No single method dominates across all performance metrics. Relying on any one metric alone can give a misleading picture of classifier performance, particularly under sparse photon statistics. Physics-based and data-driven methods are complementary rather than competing approaches. Together, these findings provide practical guidance for selecting and combining classification strategies for scalable single-photon-source screening and other quantum-emitter characterization tasks.
Quantum-enhanced sensing typically relies on preparing specific entangled states. However, engineering these states in scalable many-body systems remains challenging due to the difficulty of designing robust preparation protocols and accounting for realistic noise. We show that evolving a simple product state under native time-independent Hamiltonian dynamics in Rydberg atom arrays generates complex many-body correlations that can potentially enable quantum-enhanced parameter estimation reaching the Heisenberg limit (HL). The key challenge is then shifted to extracting the metrological information encoded in spatial correlations of measurement patterns. We introduce a neural correlation learning framework that combines Bayesian inference with neural networks trained on calibration data. Operating in a two-stage calibration-and-sensing protocol, we numerically demonstrate that this framework effectively extracts many-body correlations to saturate the Cramér--Rao bound dictated by the classical Fisher information. Consequently, it achieves quantum-enhanced sensitivity surpassing the standard quantum limit (SQL). Furthermore, the learned estimator is robust to realistic noise, and remains accurate in the absence of rare measurement patterns. Our results establish a hardware-efficient and scalable paradigm for quantum sensing in interacting many-body systems without requiring engineered entangled states.
Quantum computers require high-fidelity measurement of many qubits to achieve a quantum advantage. Traditional approaches suffer from readout crosstalk for a neutral-atom quantum processor with a tightly spaced array. Although classical machine learning algorithms based on convolutional neural networks can improve fidelity, they are computationally expensive, making it difficult to scale them to large qubit counts. We present two simpler and scalable machine learning algorithms that realize matched filters for the readout problem. One is a local model that focuses on a single qubit, and the other uses information from neighboring qubits in the array to prevent crosstalk among the qubits. We demonstrate error reductions of up to 32% and 43% for the site and array models, respectively, compared to a conventional Gaussian threshold approach. Additionally, our array model uses two orders of magnitude fewer trainable parameters and four orders of magnitude fewer multiplications and nonlinear function evaluations than a recent convolutional neural network approach, with only a minor (3.5%) increase in error across different readout times. Another strength of our approach is its physical interpretability: the learned filter can be visualized to provide insights into experimental imperfections. We also show that a convolutional neural network model for improved can be pruned to have 70x and 4000x fewer parameters, respectively, while maintaining similar errors. Our work shows that simple machine learning approaches can achieve high-fidelity qubit measurements while remaining scalable to systems with larger qubit counts.