The interesting bit is that you don't need to tweak the subnet, only remove the other noisy connections. If you remove capacity, at some point test accuracy nee

The interesting bit is that you don't need to tweak the subnet, only remove the other noisy connections. If you remove capacity, at some point test accuracy needs to go down. The hypothesis is about the idea that we maybe don't "train" the weights, we "find" them, not about unlimited nesting (as far as I understand it, but I have high confidence. Happy to be proven wrong tho)