Learning Rate Annealing Can Provably Help Generalization, Even for Convex Problems

Open in new window