Global Optimality Beyond Two Layers: Training Deep ReLU Networks via
               Convex Programs

                       Tolga Ergen 1 Mert Pilanci 1

            Abstract                             Output
   Understanding the fundamental mechanism be-
   hind the success of deep neural networks is one
   of the key challenges in the modern machine
   learning literature. Despite numerous attempts, a
   solid theoretical analysis is yet to be developed.
   In this paper, we develop a novel u ...