Metamaterials for Machine Learning
Brief Description of Project:
Metamaterials are artificially structured materials wherein the structure of the unit cells dictate the optical response allowing for unique optical properties, not existing in nature, to be achieved. Recently, the Valentine and Huo groups have shown how metamaterials can be used in concert with digital ML architectures to achieve optical image processing. The use of metamaterials allows one to off-load computationally expensive operations into ultrafast optical operations for large increases in processing speed while also decreasing energy consumption. The purpose of this summer research experience is to build on this work, investigating co-design of metasurfaces and machine learning (ML) algorithms for tasks such as edge detection and object classification. The student’s hybrid metasurface and digital designs will be benchmarked against traditional all-digital techniques to determine the efficacy of the approach. While the project does involve metasurface design, the student need not be an expert in this area. Past experience with ML algorithms is, however, beneficial for success of this project.
This project is best suited for individuals with interests in computer science and machine learning. Experience with numerical modeling is also beneficial. No experience is needed in metamaterial design.
Nature of Supervision:
The student will be co-advised by Prof. Valentine and Huo and a graduate student on this project.
A Brief Research Plan (period is for 10 weeks):
Weeks [1–3]: The student will familiarize themselves with hybrid metamaterial and ML techniques.
Weeks [3-8]: The student will design a hybrid system for a particular task, such as image segmentation.
Weeks [8-10] The student will benchmark their code against traditional all-digital architectures.
Number of Open Slots: 1
Name: Jason Valentine
Department: Computer Science