Learning with Feed-forward Neural Networks: Three Schemes to Deal with the Bias/Variance Trade-off