Efficient extreme learning machine with transfer functions for filter design

This paper proposes a model based on a machine learning algorithm, extreme learning machine (ELM), and the pole-residue-based transfer function (TF) for parametric modeling of electromagnetic behavior of microwave components. Compared with the model based on the artificial neural network, the proposed ELM model can obtain the accurate results for microwave passive component design with the small training datasets due to its good iterative learning ability. The validity and efficiency of this proposed model is confirmed by a triple-mode filter.