Reviews: Using a Logarithmic Mapping to Enable Lower Discount Factors in Reinforcement Learning

The authors propose remapping value functions into a logarithmic space, leading to "logarithmic Q-learning" which is demonstrated to perform quite well in practice. This paper has by far the strongest overall scores (9, 9, 8) in my paper batch. All three reviewers are enthusiastic about the paper and its contributions and results. I am recommending that NeurIPS accept the paper for Oral presentation.