The Progress and Prospects of Bayesian-Based Automated ANN Model Generation Method for Microwave Modeling

Automated model generation (AMG) can avoid manual trial-and-errors in the artificial neural network (ANN) model development process for microwave design. This paper reviews recent advancements in using Bayesian-based AMG methods for ANN modeling in microwave design. The first method offers an efficient solution to determine the minimum number of hidden neurons required to attain maximum accuracy in a single hidden layer Multi-Layer Perceptron (MLP). The second method, which extends the first method, systematically determines the optimum configuration of an MLP model with multiple hidden layers, leading to improved accuracy with a comparable number of network parameters and computational resources. To showcase the benefits of the Bayesian-based AMG methods, two microwave filter examples are utilized.