A robust approach to supervised learning in neural network

Most supervised neural networks (NN) are trained by minimizing the mean squared errors (MSE) of the training set. In the presence of outliers, the resulting NN model can differ significantly from the underlying system that generates the data. In order to handle outliers, this study proposes to minimize the mean log squared errors (MLSE), an approach which is easily adapted to most supervised learning algorithms. Simulation results indicate that this proposal is robust against outliers.<>