Compact latent-space encoding of WGM microlaser spectra for fast and accurate sensing

The synergy between resonant photonics and artificial intelligence is emerging as a new paradigm capable of overcoming long-standing limitations of conventional methods used to analyze measurement outcomes. Automated data analysis is a crucial step in the development of fast, reliable, and robust sensors. Neural-network autoencoders are now considered excellent candidates for spectral analysis, and significant results have been achieved when selected features are used as input parameters. Here, we propose an autoencoder that uses entire emission spectra from whispering-gallery microlasers as input. From the bottleneck layer, we extract a compact latent-space representation that encodes the spectral changes induced by variations in the external refractive index. The resulting latent parameters accurately reproduce experimentally observed wavelength shifts and mode-intensity variations, providing a low-dimensional and physically meaningful description of the sensor response. These findings demonstrate the potential of AI-assisted spectral analysis to enhance the sensitivity and robustness of next-generation microlaser-based photonic sensors.