The Prediction of Chiral Metamaterial Resonance using Convolutional Neural Networks and Conventional Machine Learning Algorithms

Abstract

Electromagnetic resonance is the most important distinguishing property of metamaterials to examine many unusual phenomena. The resonant response of metamaterials can depend many parameters such as geometry, incident wave polarization. The estimation and the design of the unit cells can be challenging for the required application. The research on resonant behavior can yield promising applications. We investigate the resonance frequency of the chiral resonator as a unit of chiral metamaterial employing both traditional machine learning algorithms and convolutional deep neural networks. To our knowledge, this is the very first attempt on chiral metamaterials in that comparing the impact of various machine learning algorithms and deep learning model. The effect of geometrical parameters of the chiral resonator on the resonance frequency is studied. For this purpose, convolutional neural networks, support vector machines, naive Bayes, decision trees, random forests are employed for classification of resonance frequency. Extensive experiments are performed by varying training set percentages, epoch sizes, and data sets.