Bayesian Neural Networks Enable Probabilistic Graybox Characterization of Quantum Devices with Uncertainty Quantification

Characterising the performance of quantum devices presents a significant challenge, requiring methods that accurately capture both expected behaviour and inherent uncertainties. Poramet Pathumsoot, Michal Hajdušek, and Rodney Van Meter from Keio University address this need with a new probabilistic approach to ‘Graybox’ characterisation, a technique that combines known system dynamics with unknown transformations. Their work overcomes a key limitation of traditional Graybox methods by incorporating uncertainty quantification, allowing researchers to make more informed decisions based on experimental data. The team demonstrates that their model, built using Bayesian Neural Networks, accurately captures the distribution of observed data, outperforming the original Graybox method by up to 1. 9times and offering a flexible tool for device characterisation even when detailed knowledge of noise models is lacking. This advancement promises to improve the reliability and interpretability of quantum experiments, paving the way for more robust quantum technologies.

Quantum Noise, Machine Learning, Probabilistic Programming

Research at the intersection of quantum computing and machine learning is rapidly advancing, driven by the need to overcome challenges in building reliable quantum technologies. This work focuses on characterizing and mitigating noise in quantum systems, a critical step towards realizing the potential of quantum computation. Scientists are increasingly employing machine learning techniques, particularly Bayesian Neural Networks and probabilistic programming, to improve quantum control, characterization, and optimization. These approaches allow researchers to model complex quantum systems and predict their behavior with greater accuracy.

Key areas of investigation include dynamical decoupling, comb-based spectral estimation, and the characterization of non-Gaussian noise. Researchers are leveraging technologies like JAX and Qiskit, alongside uncertainty quantification packages such as UQpy, to develop sophisticated models. This interdisciplinary field demonstrates a growing trend of applying machine learning to address fundamental problems in quantum computing and vice versa, highlighting the complexity and collaborative nature of this research area.

Bayesian Graybox Model Characterises Quantum Device Uncertainty

Scientists have developed a probabilistic Graybox model to improve the characterization of quantum devices, addressing limitations in quantifying uncertainty. This new method combines known system dynamics with unknown transformations, leveraging the strengths of both “Whitebox” and “Blackbox” modeling approaches. By employing a single experimental procedure across multiple qubit realizations, the team trained a machine learning model to approximate unknown processes within the system, forming a predictive Graybox model. To account for stochastic noise, common in superconducting, nuclear-spin, and trapped-ion qubits, the team implemented Bayesian Neural Networks for the “Blackbox” component.

These networks perform inference of model parameters using Bayesian inference, resulting in predictions represented as distributions rather than single values, naturally quantifying prediction uncertainty. Experiments utilizing binary measurement outcomes streamline data analysis and enhance efficiency. Results demonstrate the probabilistic Graybox model outperforms previous models by up to 1. 9times in capturing observed data distributions, significantly improving predictive accuracy and uncertainty quantification, offering a robust approach for local experiment development.

Probabilistic Graybox Model Captures Qubit Noise

Scientists have developed a probabilistic Graybox model to enhance the characterization of quantum devices, addressing the limitations of existing methods in quantifying prediction uncertainty. The work focuses on accurately modeling quantum systems subject to stochastic noise, which introduces variability in device performance. Researchers analyzed the data-generating process of a qubit, revealing that stochastic noise causes the expectation value of a quantum observable to become a distribution rather than a single value, meaning capturing variance is crucial for accurate modeling. Experiments demonstrate that the newly developed Probabilistic Graybox Model outperforms previous models by up to 1.

9times in capturing observed data distributions. This improvement stems from the model’s ability to accurately predict the expected value of the shifted expectation value, a key metric influenced by stochastic noise. The team found that the performance of any predictive model is fundamentally limited by its ability to capture this expected value accurately. This research provides an enhanced Graybox characterization method, offering better uncertainty estimation and serving as a valuable tool for constructing reliable predictive models for quantum systems.

Probabilistic Graybox Improves Quantum Gate Calibration

This research successfully augments the Graybox characterization method with robust uncertainty quantification, a crucial advancement for reliable device analysis. Results demonstrate that this probabilistic model significantly improves the capture of observed data distributions, outperforming previous models by up to 1. 9times. The team further demonstrated the practical application of the model by reframing control calibration as a maximum likelihood estimation problem.

Applying the model to calibrate a quantum gate, the √X gate, yielded control parameters closer to the globally optimal solution compared to statistical counterparts. Analysis revealed that the model’s performance hinges on the accurate prediction of expected values, highlighting a key factor for effective quantum device characterization. The code developed for this research is publicly available, facilitating further investigation and application of this valuable tool for understanding and calibrating quantum devices.