Gradient Descent Converges to Minimizers
Date:
Gave a talk in the Microsoft Research (MSR) - IISc theory reading group on the COLT 2016 paper by Lee et al., “Gradient Descent Converges to Minimizers”.
Abstract: Gradient descent converges to a local minimizer, almost surely with random initialization. This is proved by applying the Stable Manifold Theorem from dynamical systems theory.