Review for NeurIPS paper: Efficient Projection-free Algorithms for Saddle Point Problems

Additional Feedback: - Eqns (2) & (3): This notation might be confusing for inexperienced readers. Min and max here can be perceived as functions rather than optimization problems. To clarify, you might either put some space between minmax and f, or simply drop minmax and write f(x,y) E[F(x,y,xi)]. You already have minmax in the model problem (1) anyway. First, we typically do not use LMO on X, but on a gradient term.