… Figure 1: Left: Offline policy selection attempts to choose the best policy from a set of policies,
… In offline policy selection (OPS) the task is to select the policy with the highest value from a …

… We formally consider offline policy selection as learning preferences over a set of policy …
evaluation (OPE) is a highly active area of research. While the original motivation for OPE …

… We select the classic BC [36] and state-of-the-art offline RL methods as baselines. For
methods based on dynamic programming, we compare to AWAC [33], Onestep RL [2], TD3+BC [7]…

… policies without any costly or dangerous active exploration. However, it is also challenging,
due to the distributional shift between the offline … can be selected from a larger set of policies. …

… policy due to the overly conservative constraints, offline RL confronts challenges in active …
Then, we consider all possible actions and select them through the action-selection calculation …

… In the first part of the dissertation, we study the offline policy selection problem. Policy
selection is a … We expect active research topics will be (1) to identify suitable conditions on the …

… in offline RL finding a better policy than the one used to generate the data. Although selecting
the … ]) can work well in the offline RL setting since we can avoid active data collection and …

… offline datasets, thereby disallowing the agent from actively … A challenge encountered
within Offline RL is offline policy … performance of policies relying exclusively on offline data [5, …

… Training and evaluation of deep offline RL algorithms is still in active development, and …
As the EOP incorporates offline policy selection, we can also use it to compare how well OPS …

… offline AC … a mild policy evaluation (MPE) by constraining the difference between the $Q$
values of actions supported by the target policy and those of actions contained within the offline …