On the Expressive Power of Deep Learning: A Tensor Analysis

tldr, anyone? I'm interested in the field and have been learning about it, but I don't really understand this paper.