Uploaded October 2021 | Updated November 2026, 4 days ago
A presentation of "Why Generalization in RL is Difficult: Epistemic POMDPs and Implicit Partial Observability", originally recorded for NeurIPS 2021.
This paper shows that when learning to generalize in RL actually requires solving an implicit partially observable problem (a POMDP), even though the original problem itself may be fully-observable (an MDP).
Ghosh*, Rahme*, Kumar, Zhang, Adams, Levine. Why Generalization in RL is Difficult: Epistemic POMDPs and Implicit Partial Observability, NeurIPS 2021
Paper: arxiv.org/abs/2107.06277
A presentation of "Why Generalization in RL is Difficult: Epistemic POMDPs and Implicit Partial Observability", originally recorded for NeurIPS 2021.
This paper shows that when learning to generalize in RL actually requires solving an implicit partially observable problem (a POMDP), even though the original problem itself may be fully-observable (an MDP).
Ghosh*, Rahme*, Kumar, Zhang, Adams, Levine. Why Generalization in RL is Difficult: Epistemic POMDPs and Implicit Partial Observability, NeurIPS 2021
Paper: arxiv.org/abs/2107.06277










