Would it kill the whole concept to encapsulate the weight and learning rate within the link between 2 nodes? I think(very unsurely) that it would make it more versatile, and enable nodes to seem to adapt to individual connections to other nodes rather than as a whole.
I've been thinking about NNs a lot lately, and my brains kinda overflowing with ideas. As I stupid-check them, if they pass, I'll dump them here.