Performance Evaluation of Various ANN Architectures Using Proposed Cost Function

Abstract

Traditional learning methods focus on the training of neural networks. However robustness or fault tolerance or sensitivity of its input to output which plays an important role while designing a neural network are not considered appropriately. Purpose of this paper is the optimization of various feed forward artificial neural network (FFANN) architectures with hyper-parmeter selection and using proposed cost function AvgNew.This paper, has proposed a cost function which when applied to TDFFANN(tailored deep feed forward artificial neural network) and other FFANN architectures makes it more fault-tolerant, sensitive or robust. Result obtained using variants of MLP(Multi Layer Perceptron) and FFANN(feed forward artificial neural network) proves AvgNew to be efficient.