Automated Model Generation Method With Electromagnetic Sensitivity for Microwave Components

Artificial neural networks (ANNs) have become effective tools for modeling and optimization in the microwave field. By learning from electromagnetic (EM) simulation data, ANNs provide fast and accurate predictions, reducing the need for time-consuming EM simulations. To streamline the ANN development process, automated model generation (AMG) methods have been introduced. This article reviews the AMG framework, with a focus on its recent enhancements that incorporate EM sensitivities to enable more efficient adaptive sampling and improve the accuracy of ANN modeling for microwave applications. These sensitivities, which can be efficiently computed using adjoint methods, are used both to guide sample selection and to enhance model training.