Paper Title (use style: paper title)

In this work, a new gradient based reverse modeling approach employing Artificial Neural Networks (ANNs) for systematic RF/microwave modeling is introduced. This approach is particularly suited to modeling scenarios, where standard ANN multi-layer perceptron (MLP) fails to deliver a satisfactory model. The proposed approach detects the simplest input-output relationship inherent to the modeling problem, which we term as the reverse model as compared to the original model (i.e., the modeling problem using standard ANN model). This reverse model is short-listed from a pool of candidate models obtained by systematically reversing the input-output variables of the original modeling problem, while retaining the ANN’s structural simplicity. The proposed reverse and the not-so-accurate original models complement each other to yield accurate models. The advantages of this approach are demonstrated via modeling transmission lines and spiral inductors. Keywords-ANN; conjugate gradient; modeling; optimization; reverse model; sensitivity; spiral inductor; transmission line