Design of Meta-Surface Using Machine Learning

The feasibility to use Machine Learning algorithm in designing meta-surface is studied on a Pattern-Responsibility System. A series of conventional Machine Learning network like Deep Convolutional Neural Network has been implemented for bench marking purposes. A new approach based on “complex number” based neural network is applied to compare with the traditional “real number” neural network. All of the performance will be discussed based on the number of data sets obtained from the traditional Maxwell-equation commercial software. Finally, we discuss how to include the recent developed physics informed neural network to explore its potential for the inverse design of meta-surfaces.