Logo image
Gradient Descent Only Converges to Minimizers: Non-Isolated Critical Points and Invariant Regions
Conference proceeding

Gradient Descent Only Converges to Minimizers: Non-Isolated Critical Points and Invariant Regions

Ioannis Panageas and Georgios Piliouras
8TH INNOVATIONS IN THEORETICAL COMPUTER SCIENCE CONFERENCE, ITCS 2017, Vol.67
Leibniz International Proceedings in Informatics
01/01/2017

Abstract

Computer Science Computer Science, Theory & Methods Mathematics Mathematics, Applied Physical Sciences Science & Technology Technology
Given a twice continuously differentiable cost function f, we prove that the set of initial conditions so that gradient descent converges to saddle points where del(2) f has at least one strictly negative eigenvalue, has (Lebesgue) measure zero, even for cost functions f with non-isolated critical points, answering an open question in [12]. Moreover, this result extends to forward-invariant convex subspaces, allowing for weak (non-globally Lipschitz) smoothness assumptions. Finally, we produce an upper bound on the allowable step-size.

Metrics

1 Record Views

Details

Logo image