Has anyone tried taking the features that are learned at the various layers of a neural net and feeding them into something like this: https://news.ycombinator.com/item?id=12612246?

I imagine we would get some really interesting images back...