Why should I choose to fully link?
Input to first Hidden layer? Hidden Layer to Hidden Layer? Hidden Layer to ouput?
If I don't fully link those, what are my options and how do I choose it?
Since the knowledge in those units is "hidden", machine representation, how should I expect to know which linkage would fit the most?
I'm asking because, I normally see Fully linked everywhere, but Convolutional just use subsets
[link][2 comments]