Beyond Tikhonov: faster learning with self-concordant losses, via iterative regularization

Open in new window