Exploring the loss landscape of regularized neural networks via convex duality

Open in new window