Machine Learning for the Optimization of Optical Nano-Materials

A.-P. Blanchard-Dionne, O.J.F. Martin
Ecole Polytechnique Federale de Lausanne,

Keywords: deep learning, nano-materials, nanotechnology, metamaterials, nano-photonics


In this presentation I will describe the efforts made in order to use machine learning for the design and optimization of optical nano-materials. In a first part I will describe how you can include phenomenological mathematical equations inside neural networks to improve their prediction capabilities of the optical properties of such materials. The second part is dedicated to using generative networks to accomplish reverse engineering of nano-materials, i.e. to retrieve the specific shapes and parameters that can accomplish on-demand optical properties. The focus will be on Generative Adversarial Networks as well as Variational Auto-Encoders.