Offline Reinforcement Learning: Fundamental Barriers for Value Function Approximation
This work addresses a foundational problem in machine learning for researchers and practitioners in RL, revealing inherent limitations in offline learning that are not incremental but rather establish new theoretical barriers.
The paper tackles the problem of offline reinforcement learning (RL) with value function approximation, proving that even under standard assumptions of concentrability and realizability, sample complexity is polynomial in state space size, making sample-efficient learning infeasible without stricter conditions. This resolves a long-standing conjecture and shows a fundamental barrier, with implications such as an arbitrarily large separation between online and offline RL in linear settings.
We consider the offline reinforcement learning problem, where the aim is to learn a decision making policy from logged data. Offline RL -- particularly when coupled with (value) function approximation to allow for generalization in large or continuous state spaces -- is becoming increasingly relevant in practice, because it avoids costly and time-consuming online data collection and is well suited to safety-critical domains. Existing sample complexity guarantees for offline value function approximation methods typically require both (1) distributional assumptions (i.e., good coverage) and (2) representational assumptions (i.e., ability to represent some or all $Q$-value functions) stronger than what is required for supervised learning. However, the necessity of these conditions and the fundamental limits of offline RL are not well understood in spite of decades of research. This led Chen and Jiang (2019) to conjecture that concentrability (the most standard notion of coverage) and realizability (the weakest representation condition) alone are not sufficient for sample-efficient offline RL. We resolve this conjecture in the positive by proving that in general, even if both concentrability and realizability are satisfied, any algorithm requires sample complexity polynomial in the size of the state space to learn a non-trivial policy. Our results show that sample-efficient offline reinforcement learning requires either restrictive coverage conditions or representation conditions that go beyond supervised learning, and highlight a phenomenon called over-coverage which serves as a fundamental barrier for offline value function approximation methods. A consequence of our results for reinforcement learning with linear function approximation is that the separation between online and offline RL can be arbitrarily large, even in constant dimension.