Introduction to the special issue on empirical evaluations in reinforcement learning

摘要

We believe that the question of how best to conduct empirical evaluations in reinforcement learning remains a critical research topic. The research described in this special issue shows that promising initial progress has been made towards the development of principled answers. The results of the empirical evaluations presented by these articles not only offer new insights about the strengths and weaknesses of current reinforcement-learning methods, but also underscore the capacity of careful evaluations to generate such insights. We hope that this special issue will serve as a foundation for and stimulus to further developments in reinforcement-learning software, evaluation methodologies, benchmark tasks, and empirical comparisons.