NeurIPS 2020

Penalized Langevin dynamics with vanishing penalty for smooth and log-concave targets

Meta Review

The paper was heavily discussed post-rebuttals. All the reviewers appreciate the clean mathematical analysis and writing, though on a technical level, the analysis is quite similar to prior work. However, the lack of discretization analysis was considered a serious flaw by all reviewers -- as it's unclear what are the algorithmic implications for the problem studied (furthermore, given prior work on discretizing Langevin, everyone felt this wouldn't be too hard and wouldn't harm the clarity that much).