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

Area of Science:

• Photonics and Sensor Technology
• Machine Learning Applications
• Computational Materials Science

Background:

• Bragg gratings are vital for sensing due to sensitivity and tunability.
• Designing Bragg gratings requires extensive simulation data, often scarce.
• Machine learning models need informative training data for efficient design.

Purpose of the Study:

• To develop an efficient data acquisition strategy for Bragg grating sensor design.
• To enhance machine learning model performance in data-limited scenarios.
• To optimize the design and simulation of Bragg grating sensors.

Main Methods:

• An augmented Bayesian optimization approach was employed.
• A distance-based diversity criterion was integrated to select informative data points.
• The method prioritizes data points farthest from existing datasets when acquisition values are similar.

Main Results:

• Emphasizing output diversity during data acquisition significantly improved model performance.
• The approach is particularly effective for complex optical responses in Bragg gratings.
• Different analytical fit functions (polynomial, Gaussian) were compared to assess complexity impact.

Conclusions:

• The proposed method provides a scalable framework for generating high-quality simulation data.
• This strategy is crucial for optimizing Bragg grating sensor design in data-scarce environments.
• The findings have direct implications for advancing next-generation Bragg grating-based sensing technologies.