Unconventional Surrogate-Assisted Approaches to EM-Driven Antenna Design. Modeling and Optimization: Global, Multi-Objective, Statistical

Design of contemporary antenna systems relies heavily on full-wave electromagnetic (EM) simulation tools. EM analysis ensures reliability but tends to be computationally expensive, which becomes a serious issue in the context of design automation. In recent years, utilization of surrogate modeling methods has been fostered to alleviate the cost-related difficulties. Yet, the existing procedures typically follow a few major approaches, e.g., different architectural variations of deep neural networks, hyper-parameter/model adjustments in Bayesian optimization, or more or less standard machine learning frameworks. This paper reviews several unconventional surrogate-assisted techniques that employ less common algorithmic tools such as the response feature technology, domain confinement, dimensionality reduction, or supplementary inverse predictors. We demonstrate how these tools can be incorporated into practical procedures for global and multi-criterial optimization, statistical analysis, and design-oriented behavioral modeling. A discussion of the algorithm operating principles is supplemented by antenna design cases.