cs.AI, cs.LG, stat.ML

Diagnosing Non-Markovian Observations in Reinforcement Learning via Prediction-Based Violation Scoring

arXiv:2603.27389v1 Announce Type: new
Abstract: Reinforcement learning algorithms assume that observations satisfy the Markov property, yet real-world sensors frequently violate this assumption through correlated noise, latency, or partial observabili…