This is the third part of an advanced course on approximation theory of deep neural networks. Here we describe multiple approaches that explain why approximation by neural networks often does not seem to suffer from the so-called curse of dimension, i.e., it does not deteriorate exponentially for increasing dimensions.