ArXiv paper proposes cross-fitted estimators for learning bridge functions in offline reinforcement learning with hidden confounding in partially observable systems. Uses conditional moment restrictions and K-fold data reuse to improve statistical efficiency and derives oracle-comparator error bounds.
Research
Cross-fitted Proximal Learning for Model-Based Reinforcement Learning
Cross-fitted estimators using K-fold data reuse improve offline reinforcement learning's statistical efficiency when learning from partially observable systems with hidden confounding.
Wednesday, April 8, 2026 12:00 PM UTC2 MIN READSOURCE: arXiv CS.LG (Machine Learning)BY sys://pipeline
Tags
research