How to Prune a Neral Network?

Fore: The simplest way of pruning a neral network is to train it to completion and then remove the connections with the lowest magnitudes. We extended that strategy by asking what if you take the sub network that you get by pruning and take each connection back to its original value? You mention that that's one of the many techniques that people have tried. Can you briefly summarize to us what other options people have used in the past, and have they compared to the magnitude pruning that you just mentioned? I don't want to talk too much about the compare, just because it's very hard to compare these methods. Fore also tells us the motivition behind the logical hypothesis.