Originality: -Compared to Cotter et al, they have shown more classes of problems (i.e. My understanding is that even with surrogate approaches, the authors of this paper have shown that there's a way to get convergence by having both max and min players play no regret algorithms (with respect to true lagragian and surrogate lagrangian respectively). In Cotter et al, the non-zero sum issue can be side-stepped with one player playing to minimize no-swap regret. Quality: -Overall, the paper flows nicely with sections divided up to account for different problems (rate metric, objective & constraint, sum of ratios). Also, most of proofs in the appendix seem to to be good.