We know from reality that they get practically better, but theoretic intuition suggests you shouldn't see an effect after some point. This paper shows that this

We know from reality that they get practically better, but theoretic intuition suggests you shouldn't see an effect after some point. This paper shows that this intuition is wrong if you want your networks to be robust. It doesn't guarantee large networks will be though.