Reducing the variance in online optimization by transporting past gradients

Open in new window