ResNet with one-neuron hidden layers is a Universal Approximator

We demonstrate that a very deep ResNet with stacked modules that have one neuron per hidden layer and ReLU activation functions can uniformly approximate any Lebesgue integrable function inddimensions, i.e. l