Machine Learning-Based Real-Time Metasurface Reconfiguration

Reconfiguration of a programmable coupled resonator metasurface is challenging. Due to its complexity, scalability to real-world applications using known techniques is not feasible. In this paper, we explore this challenge using a machine learning approach. We investigate two well-known machine learning regression models (random forest and neural network), as well as a combination of the two using stacked generalization, in order to predict the inputs required to generate a desired far-field radiation pattern of a metasurface. Preliminary results indicate that a random forest and a neural network in a stacked generalization ensemble outperforms separate implementations of those models.