Computationally Tractable Offline Quantum Experimental Design for Nuclear Spin Detection

Abstract: The characterization of nuclear spin environments in solid-state devices plays an important role in advancing quantum technologies, yet traditional methods often demand long measurement times. To address this challenge, we extend our recently developed deep-learning-based SALI model (Signal-to-image ArtificiaL Intelligence) by introducing the surrogate information gain (SIG) to optimize the selection of data points in the measurements. This approach significantly reduces time requirements in experiments while preserving accuracy in nuclear spin detection. The SIG is a figure of merit based on the expected variance of the signal, which is more straightforward to compute than the expected information gain rooted in Bayesian estimation. We demonstrate our approach on a nitrogen-vacancy (NV) center in diamond coupled to $^{13}$C nuclei. In the high-field regime, our variance-based optimization is validated with experimental data, resulting in an 85$\%$ reduction in measurement time for a modest reduction in performance. This work also constitutes the first validation of SALI on experimental data. In the low-field regime, we explore its performance on simulated data, predicting a 60$\%$ reduction in the total experiment time by improving the temporal resolution of the measurements and applying SIG. This demonstrates the potential of integrating deep learning with optimized signal selection to enhance the efficiency of quantum sensing and nuclear spin characterization, paving the way for scaling these techniques to larger nuclear spin systems.