In almost all shallow analytic neural network optimization landscapes, efficient minimizers have strongly convex neighborhoods

Open in new window