Machine Learning as Applied to EM - Trends, Advances, and Applications

Machine learning (ML) has paved the way for development of algorithms and statistical models that enable computers to perform tasks without being explicitly programmed. The availability of voluminous amount of data and high throughput graphics processing units (GPU) has led to the promising field of Machine Learning. The primary goal of machine learning is to enable computers to learn from data. Electromagnetic (EM) algorithms solve Maxwell's equations under various boundary conditions and for a variety of materials. EM algorithms serve as the kernel of EM simulation. Traditional EM simulators are commonly based upon the Finite Difference Method (FDM), Finite Element Method (FEM), Finite Difference Time Domain (FDTD), Method of Moments (MOM). Such simulations consume large amounts of CPU time and memory. Thus full-wave computational EM solvers are prohibitive for applications requiring a real-time response. Machine learning steps in as a game-changer. Machine learning techniques provides a platform to implement an end-to-end EM simulation driven solely by data functioning as a fast real-time EM simulator. This paper describes the progress in the field of electromagnetics contributed by penetration of computationally powerful machine learning methodologies.