ELM not only learns much faster
with a higher generalization performance than the traditional
gradient-based learning algorithms but it also avoids many difficulties
faced by gradient-based learning methods such as stopping criteria,
learning rate, learning epochs, local minima, and the over-tuned
problems