On the Structural Non-Preservation of Epistemic Behaviour under Policy Transformation

Alexander Galozy

Reinforcement learning (RL) agents under partial observability often condition actions on internally accumulated information such as memory or inferred latent context. We formalise…