On Convergence of Training Loss Without Reaching Stationary Points

Open in new window