Training strategy for backpropagation neural networks using input weighting

Presents a new strategy for training feedforward backpropagation neural network, this strategy is based on weighting (repeating) particular pairs of the input-output vectors. These particular pairs are the ones that produces the largest error after each training cycle, and therefore this training strategy is called "W_eighted I_nput". The proposed training strategy has been tested for three simple cases, and it is shown that it does provide savings in training time in two of the three cases, while it fails for the third case.