Related Experiment Video
Updated: Aug 13, 2026
03:14
Augmenting Large Language Models via Vector Embeddings to Improve Domain-Specific Responsiveness
Published on: December 6, 2024
CoSP: Reconfigurable Metamaterial Inverse Design via Contrastive Pretrained Large Language Model
Shujie Yang1,2,3, Yuqi Zhang1,2,3, Xuzhe Zhao1,2,3
1Institute of Data and Information, Tsinghua Shenzhen International Graduate School, Tsinghua University, Shenzhen, Guangdong, China.
Advanced Science (Weinheim, Baden-Wurttemberg, Germany)
|August 11, 2026
Summary
This study introduces CoSP, an AI method using large language models for designing reconfigurable metamaterials (RMMs). CoSP enables precise control over optical properties for advanced applications.
Area of Science:
- Optics and Photonics
- Materials Science
- Artificial Intelligence
Background:
- Designing metamaterials for light manipulation is challenging due to complex structures.
- Existing deep learning methods struggle with reconfigurable metamaterials (RMMs) that change optical properties.
Purpose of the Study:
- To develop an intelligent inverse design method for RMMs.
- To enable the design of metamaterials with switchable optical characteristics.
Main Methods:
- Introduced CoSP, an inverse design method leveraging a contrastive pretrained large language model (LLM).
- Utilized contrastive pretraining on multi-state spectra to create a spectrum encoder.
- Coupled the encoder with a GPT-style decoder trained end-to-end.
- Enabled natural language descriptions of material structures for desired optical properties.
Main Results:
- CoSP successfully designed RMM structures for multi-state, multi-band optical responses.
- Demonstrated the method's capability to achieve switchable optical characteristics.
Conclusions:
- CoSP offers a novel approach for designing reconfigurable metamaterials.
- The method shows significant potential for applications in thermal management, optical computation, and telecommunications.