To reduce the computation time for training, the back-propagation algorithm has been parallelized in three ways. Before the different parallelizations are described the backpropagation algorithm is explained.