Review for NeurIPS paper: An Improved Analysis of (Variance-Reduced) Policy Gradient and Natural Policy Gradient Methods

Additional Feedback: Detailed comments on a section-by-section basis are given below. Lines 10-13 of abstract: I found this quite vaguely worded, and didn't understand which contributions in the main paper this is referring to. Section 1 "Policy gradients are usually estimated via Monte-Carlo rollouts". To me, this would suggest that critics are not used, or are used at most as baselines. However, as far as I am aware, most of the successful methods cited in the previous paragraph do use bootstrapping, and so are not pure Monte Carlo, in the sense that the term is used in RL.