Efficient Representation of Low-Dimensional Manifolds using Deep Networks

Ronen Basri, David Jacobs

We consider the ability of deep neural networks to represent data that lies near a low-dimensional manifold in a high-dimensional space. We show that deep networks can efficiently…