Fetching the paper…

Noisy Gradient Descent Converges to Flat Minima for Nonconvex Matrix Factorization · Around