Performance of multilayer neural networks in binary-to-binary mappings under weight errors

The probability of misclassification error of a multilayer neural network used in binary-to-binary mappings is derived. The connection weights, determined through training, are assumed to be subject to an additive, random, normally distributed error. The probability of misclassification is derived through simulation. The simulation results and the theoretical results are shown to match very closely. The results give predictability to neural network performance and allow for changing neural network design parameters, such as weight vectors and number of nodes, in order to obtain a certain tolerance to weight errors