New method evaluates offline reinforcement learning without completeness assumptions

Original: Fitted Occupancy-Ratio Evaluation without Bellman Completeness

Source: arxiv.org ↗