no code implementations • 19 Jan 2023 • David A. R. Robin, Kevin Scaman, Marc Lelarge
In this paper, we present a new strategy to prove the convergence of deep learning architectures to a zero training (or even testing) loss by gradient flow.
