Recent Advances in Gradient-based EM Optimization using Neuro-TF Surrogate

Abstract

A surrogate model composed of neural networks and transfer functions has been presented to execute efficient electromagnetic (EM) optimization instead of using a coarse model. It is called neuro-transfer function (neuro-TF). This paper introduces the recent advances in neuro-TF methods for electromagnetic optimization, including multifeature-assisted method and adjoint sensitivity-based method. Further speedup in EM optimization is achieved using these neuro-TF surrogate optimization methods.