Metamaterial Reverse Multiple Prediction Method Based on Deep Learning.

Area of Science:

• Optics and Materials Science
• Computational Physics
• Artificial Intelligence in Engineering

Background:

• Metamaterial research significantly impacts optics, yet on-demand structure design remains challenging.
• Deep learning approaches have shown promise in guiding metamaterial design.
• Existing methods often lack the capability for simultaneous reverse and multiple structure prediction.

Purpose of the Study:

• To propose a novel deep learning-based method for the reverse and multiple prediction of metamaterial structures.
• To enable the design of metamaterial structures based on desired spectral characteristics.
• To overcome limitations in current metamaterial design methodologies.

Main Methods:

• Development of a semisupervised learning model named the partially Conditional Generative Adversarial Network (pCGAN).
• Utilizing pCGAN for reverse prediction of multiple metamaterial structures from a target spectrum.
• Validation of the model's performance and generality.

Main Results:

• The pCGAN model successfully predicts multiple metamaterial structures for a given target spectrum.
• Achieved a mean average error (MAE) of 0.03, indicating high accuracy.
• Demonstrated good generality across different design requirements.

Conclusions:

• The proposed pCGAN method offers a new approach for the simultaneous reverse and multiple design of metamaterials.
• This advancement facilitates the creation of novel metamaterial structures with tailored optical properties.
• The method opens new avenues for efficient and on-demand metamaterial engineering.