The Importance of Pessimism in Fixed-Dataset Policy Optimization
This addresses the challenge of reliable policy optimization from limited data for reinforcement learning practitioners, offering a theoretical framework and algorithms to improve robustness, though it is incremental in building on existing pessimism principles.
The paper tackles the problem of ensuring worst-case guarantees for policy optimization algorithms using fixed datasets, revealing that naive approaches require datasets informative of every policy to guarantee near-optimal performance, and shows that pessimistic algorithms can achieve good performance even with less informative datasets, validated by experiments on gridworld and MinAtar environments.
We study worst-case guarantees on the expected return of fixed-dataset policy optimization algorithms. Our core contribution is a unified conceptual and mathematical framework for the study of algorithms in this regime. This analysis reveals that for naive approaches, the possibility of erroneous value overestimation leads to a difficult-to-satisfy requirement: in order to guarantee that we select a policy which is near-optimal, we may need the dataset to be informative of the value of every policy. To avoid this, algorithms can follow the pessimism principle, which states that we should choose the policy which acts optimally in the worst possible world. We show why pessimistic algorithms can achieve good performance even when the dataset is not informative of every policy, and derive families of algorithms which follow this principle. These theoretical findings are validated by experiments on a tabular gridworld, and deep learning experiments on four MinAtar environments.