Augmented Bayesian Data Selection: Improving Machine Learning Predictions of Bragg Grating Spectra.

Bragg gratings are fundamental components in a wide range of sensing applications due to their high sensitivity and tunability. In this work, we present an augmented Bayesian approach for efficiently acquiring limited but highly informative training data for machine learning models in the design and simulation of Bragg grating sensors. Our method integrates a distance-based diversity criterion with Bayesian optimization to identify and prioritize the most informative design points. Specifically, when multiple candidates exhibit similar acquisition values, the algorithm selects the point that is farthest from the existing dataset to enhance diversity and coverage. We apply this strategy to the Bragg grating design space, where various analytical functions are fitted to the optical response. To assess the influence of output complexity on model performance, we compare different fit functions, including polynomial models of varying orders and Gaussian functions. Results demonstrate that emphasizing output diversity during the initial stages of data acquisition significantly improves performance, especially for complex optical responses. This approach offers a scalable and efficient framework for generating high-quality simulation data in data-scarce scenarios, with direct implications for the design and optimization of next-generation Bragg grating-based sensors.