This paper has received positive reviews and is suitable for publication without significant changes. During the discussion, Reviewer #4 pointed out that their question about preconditioned OGD was somewhat misunderstood by the authors: their point was that since the decision set in this setting is unconstrained, there may be no need for preconditioning the gradient updates at all, so omitting this step may make the algorithm and analysis simpler. The authors may consider taking this into account when preparing the final version of the paper.