Efficient Algorithm for Training Neural Networks with one Hidden Layer

The LM algorithm is now considered as the most efficient[6]. It
combines the speed of the Newton algorithm with the
stability of the steepest decent method.