A Novel quasi-Newton with Momentum Training for Microwave Circuit Models using Neural Networks

This paper describes a new quasi-Newton (QN) based technique to accelerate the training of neural networks. Microwave circuits have the input and output properties with strong nonlinearities to themselves and need a robust training algorithm for their neural network models. QN was normally used for these purposes. On the other hand, the steepest gradient method such as Back-propagation is utilized for training of neural networks and accelerated by a momentum term. In this research, we verify the effectiveness of the momentum term in QN for microwave circuit modeling with high-nonlinearities and propose a novel training algorithm in which a momentum coefficient is adaptively given in each iteration. The proposed algorithm is demonstrated through the modeling of two microwave circuits.