Data Collection and Network Design for Deep Learning Based Metasurface Design

This paper presents novel approaches to collect balanced high degree of freedom meta-atom patterns for machine learning models and a new design of neural network for meta-atom performance evaluation.