Truth or Backpropaganda? An Empirical Investigation of Deep Learning Theory