Automated discovery of large-scale, noise-robust experimental designs in super-resolution microscopy

The discovery of super-resolution techniques, which circumvent the classical diffraction limit of light, represent a leap in optical microscopy. Yet, the vast space of all possible experimental configurations suggests that powerful techniques remain undiscovered. We demonstrate the automated discovery of SR microscopy techniques using XLuminA, an open-source JAX-based computational framework which demonstrates a speed-up of 4 orders of magnitude compared to well-established numerical optimization methods. We implement a highly-efficient optimization scheme that incorporates random noise sampling at each iteration to ensure robustness, leading to the discovery of a novel, noise-resilient experimental blueprint featuring sub-diffraction imaging capabilities. This work advances AI-driven discovery in optics and microscopy, emphasizing both high-performance and robustness.