Adrian Albert / Sean Lubner / Mahmoud Elzouka Contact: firstname.lastname@example.org
Optical metamaterial design entails carefully designing geometric properties and selecting appropriate material combinations to achieve a desired optical properties for surfaces (e.g., spectral reflectivity profile). This is currently a manual process requiring extensive expert input and expertise, as well as expensive computational resources (e.g., running computationally- intensive Maxwell’s equations solvers) and costly and slow physical experimentation (e.g., thin film deposition).
In this work, we propose to develop an AI-based framework for automating and accelerating the metamaterial design process using generative adversarial networks (GANs). Our goal is to design an algorithm that takes a target reflectivity profile as an input, and generates candidate geometries (i.e., multilayer thicknesses). A promising starting point is a class of GAN models first proposed in the context of unsupervised image-to-image translation, trained using data generated via solving radiative transfer problems via a full-physics Maxwell’s equations solver. Our proposed setup solves in the same integrated learning loop both the forward problem of estimating a spectral reflectivity profile given a geometry and the inverse (design) problem of generating candidate geometries for a target reflectivity profile.
– Comfortable with Python programming (object-oriented programming, numpy, matplotlib, pandas etc.)
– Familiarity with modern machine learning frameworks such as PyTorch or TensorFlow
– Knowledge and familiarity with modern machine learning, including deep learning. At a minimum, class work and projects, but previous research work using/developing deep learning methods preferred (ideally having worked with GANs)
-Interest in scientific applications of AI including one or more of climate, energy, materials