Noisy Gradient Descent Converges to Flat Minima for Nonconvex Matrix Factorization

Open in new window