Width is Less Important than Depth in ReLU Neural Networks.

Gal Vardi, Gilad Yehudai, Ohad Shamir

Browse the full COLT paper archive.