Machine learning-based methods for frequency selective surfaces and metasurfaces

In this paper, we present a comprehensive review of machine learning-based inverse design techniques for frequency selective surfaces (FSSs) and metasurfaces. These surfaces are useful for controlling electromagnetic waves in different applications. Conventional design methods of metasurface and FSS mainly use forward prediction algorithms, such as finite difference time domain (FDTD) with manual parameter optimization. Although conventional approaches produce accurate results, they suffer from performance and computational efficiency issues. We explore the integration of cutting-edge machine learning methods, such as deep learning, into the inverse design framework for FSSs and metasurfaces. The basics of machine learning and different algorithms like variational autoencoder (VAE), generative adversarial network (GAN), deep convolutional generative adversarial networks (DCGAN), etc., based on metasurfaces and FSSs are presented. In this paper, we review recent developments in transfer learning for phase predictions, designing FSS using a vector-graph-feature-extraction-deep-neural network (VGFE-DNN), and hybrid AI-optimization frameworks integrating particle swarm optimization (PSO) and genetic algorithms (GA), etc. These algorithms not only reduce the simulation time to design FSSs and metasurfaces, but also generate unconventional structures with improved performances.