Correlating metasurface spectra with a generation-elimination framework.

Area of Science:

• Optics and Photonics
• Materials Science
• Artificial Intelligence

Background:

• Inferring optical responses from correlated optical data is crucial for applications like biological imaging and material analysis.
• Spectra-to-spectra design, distinct from forward and inverse problems, faces challenges due to complex many-to-many correspondences.
• Existing methods for optical response prediction are limited, necessitating new approaches.

Purpose of the Study:

• To develop a novel framework for spectra-to-spectra optical response inference.
• To address the challenges of many-to-many correspondences in optical spectra correlation.
• To demonstrate the framework's capability in forecasting inaccessible spectra using terahertz metasurfaces.

Main Methods:

• A generation-elimination framework with stochastic sampling capabilities was proposed.
• The framework automatically generates diverse candidates and eliminates inferior ones.
• Dimensionality reduction was employed to visualize latent space representations of spectral data.

Main Results:

• The proposed framework achieved 98.77% accuracy in forecasting inaccessible reflection spectra for terahertz metasurfaces.
• The method successfully correlated reflection spectra across different frequency ranges without structural information.
• The dimensionality reduction approach provided interpretable insights into the deep learning model's processing.

Conclusions:

• The developed generation-elimination framework offers a powerful solution for spectra-to-spectra optical response inference.
• This approach provides explicable perspectives for deep learning in complex physical processes.
• The findings facilitate versatile applications requiring cross-wavelength information correlation.