Depth creates no more spurious local minima
This addresses a theoretical bottleneck in understanding optimization landscapes for deep learning, but it is incremental as it builds on existing work and focuses on linear networks.
The paper tackles the problem of spurious local minima in deep linear networks, showing that if a two-layer network has no spurious local minima for a convex differentiable loss, then this property holds for any depth, simplifying analysis and extending prior results like Kawaguchi 2016.
We show that for any convex differentiable loss, a deep linear network has no spurious local minima as long as it is true for the two layer case. This reduction greatly simplifies the study on the existence of spurious local minima in deep linear networks. When applied to the quadratic loss, our result immediately implies the powerful result in [Kawaguchi 2016]. Further, with the work in [Zhou and Liang 2018], we can remove all the assumptions in [Kawaguchi 2016]. This property holds for more general "multi-tower" linear networks too. Our proof builds on [Laurent and von Brecht 2018] and develops a new perturbation argument to show that any spurious local minimum must have full rank, a structural property which can be useful more generally.