NeurIPS 2019
Sun Dec 8th through Sat the 14th, 2019 at Vancouver Convention Center

### Reviewer 1

Originality: Though it seems that the proof for bounds obtained in the paper follows along the lines of [Srikant & Ying, 2019], I don't think the extension is trivial. The key difference being the construction of a Lyapunov function that is motivated by singular perturbation theory. The second contribution on an algorithm that adaptively schedules the learning rate is novel, and not something I have seen before. Quality and Clarity: Except for minor typos, the paper was well-written, and easy to read. Specifically, the intuition for Lemma 1 by studying the associated ODE's was very useful. Significance: As stated above, the paper provides good foundations as well as directions for future research. ** After reading the rebuttal ** The authors addressed my concerns well, and I would like to still recommend acceptance of the paper.