The reviewers expressed concern on the novelty of the algorithm, but still think it is well written and should be accepted for NIPS. We encourage the authors to carefully revise the work for the final version.