Per-example gradients: a new frontier for understanding and improving optimizers

Open in new window