Graybox Modeling and Control of Multiqubit Spatiotemporally Correlated Noise

Precise control of quantum systems is essential for achieving high-fidelity quantum computation. However, such control is often hindered by noise arising from system-environment interactions and imperfections in the control itself. A common strategy for mitigating noise is to design control schemes informed by noise models. Yet, developing models with sufficient predictive power remains challenging in the presence of uncertainties. To address this, we employ physics-informed machine learning—specifically, the graybox formalism—to construct multiqubit noise models that combine predictive accuracy with robustness to uncertainty. Our approach enables the learning of spatiotemporally correlated noise, a particularly harmful form of noise prevalent in today's quantum processors. By capturing these correlations, the models can be harnessed to devise optimized control strategies using genetic algorithms. We demonstrate the feasibility of this method for noise sources relevant to superconducting qubit devices. Overall, our results highlight the promise of the graybox formalism for modeling and controlling noisy multiqubit systems.