NoProp: Training neural networks without back-propagation or forward-propagation

"Whenever these kind of papers come out I skim it looking for where they actually do backprop.

Check the pseudo code of their algorithms.

"Update using gradient based optimizations""