Big-Step-Little-Step: Efficient Gradient Methods for Objectives with Multiple Scales

Open in new window