What limits accurate expectation value prediction from complex quantum systems. A powerful machine-learning technique, quantum Gaussian process regression, has extended beyond ideal scenarios to encompass general quantum channels representing realistic pathways describing information change within noisy devices. This achievement allows for accurate outcome predictions even when dealing with up to sixty-four qubits, previously hampered by limitations inherent in modelling larger systems.

Techniques utilising quantum Gaussian processes, statistical tools estimating outcomes based on limited information, now encompass more realistic scenarios in quantum computing. The advancement addresses challenges arising from noise within devices; these methods struggled when applied beyond ideal conditions or larger systems containing many qubits. Handling up to sixty-four qubits enables better result prediction despite incomplete data due to imperfections inherent in current technology.

Predictions from complex quantum systems improve using limited measurements addressing challenges posed by imperfections in current technology. Quantum channels describe how information changes within devices and can be thought of as a black box transforming quantum states like an electronic circuit altering a signal. The team employed a Bayesian optimisation surrogate, essentially using a rough map instead of detailed terrain data during calculations, to simplify computations enabling efficient finding of optimal settings.

Quantum Gaussian process regression enables prediction across expanding qubit numbers with noisy

Faithful extrapolation in simulations achieved using up to 64 qubits; previously, accurate prediction beyond small subsystems proved impossible because exponential suppression hindered learning with standard methods. Outputs converged towards a QGP under a uniform prior measure over all possible channels, enabling derivation of an associated kernel reflecting state overlaps and channel behaviour, important for predicting expectation values from limited measurements.

A rescaled kernel incorporating a learnable parameter restored predictive power for global 64-qubit channels as measurement data increased. Simulations utilising up to 64 qubits demonstrated faithful extrapolation; the method employed this learnable parameter within a rescaled kernel restoring predictive power when analysing global channels, representing a sharp improvement over previous limitations hindering learning with standard techniques.

Validation extended beyond purely unitary dynamics to encompass general quantum channels mirroring noisy devices, confirming outputs converge towards a quantum Gaussian process under a uniform prior measure across all possible channels. Further tests using actual hardware confirmed durability even under experimental conditions, validating its use as a Bayesian-optimisation surrogate for complex tasks like state preparation involving noisy XXZ dynamics.

Accurate simulation of larger qubit numbers via enhanced Gaussian process regression

Ever more precise measurements demand when predicting quantum behaviour; however, obtaining these reliably becomes exponentially harder as systems grow beyond a few entangled particles. This advance relies on an ‘empirical Bayes heuristic’, effectively a clever workaround rather than a fully justified statistical foundation.

Despite acknowledging that this ‘empirical Bayes heuristic’ isn’t a complete statistical solution, its practical value remains significant; it allows prediction of quantum behaviour using fewer measurements than previously possible. Extending the technique beyond simple systems to sixty-four qubits represents a step forward in modelling complex quantum processes. Accurate prediction of quantum behaviour is now achievable with reduced measurement requirements compared to previous methods.

Extended quantum Gaussian process regression successfully modelled up to sixty-four qubits, although it currently relies on an approximate statistical method. A statistical technique called quantum Gaussian process regression extended beyond idealised scenarios, enabling prediction of expectation values, the average result of a measurement, from unknown quantum evolutions using limited data. Proving convergence towards a quantum Gaussian process even when systems aren’t perfectly isolated allowed researchers to derive a new mathematical kernel reflecting how states and channels interact during computation. This advancement addresses limitations caused by increasing system size which previously hindered accurate predictions with standard methods as subsystems grew larger due to required precision levels.

Researchers demonstrated that quantum behaviour can be predicted accurately for systems involving up to sixty-four qubits utilising an enhanced form of quantum Gaussian process regression. This is important because obtaining precise measurements becomes increasingly difficult as quantum systems grow in complexity; the method requires fewer measurements than were previously needed. The technique incorporates an empirical Bayes heuristic, allowing learnability even when modelling global behaviours across many qubits. Authors suggest this approach maintains a strong ability to predict outcomes from limited data and exhibits robustness under experimental conditions.

👉 More information
🗞 Quantum Gaussian processes for prediction of channel observations
✍️ Jonas Jäger, Yaroslav Khmelnitskiy, Paolo Braccia, Artur Miroszewski, Diego García-Martín, M. Cerezo and Piotr Czarnik
🧠 ArXiv: https://arxiv.org/abs/2608.19306

More like this