NeurIPS 2020

On the Ergodicity, Bias and Asymptotic Normality of Randomized Midpoint Sampling Method

Meta Review

The paper deals with the computation of an unknown expectation w.r.t. a strongly log concave distribution \pi using MCMC methods. The referees found the results to be generally solid, if not surprising, and of limited impact. From the discussion: "I still believe that the authors can clarify their paper, especially discussing the CLTs obtained (which is the main contribution). Moreover, they do not have surprising results like rates improvements, except in a specific case, as far as I understand. However, they are tackling an interesting question and the fact that they provided CLT, which can be seen as upper bounds and lower bounds at the same time, satisfies me. I am also satisfied by some of their answers."