Deep linear networks for regression are implicitly regularized towards flat minima Lénaïc Chizat Institute of Mathematics

Open in new window