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

1. 1Define behavioural dependency as variation in action selection with respect to internal information under fixed observations
2. 2Formalize probe-relative notion of ε-behavioural equivalence
3. 3Analyze within-policy transformation to identify limitations in preserving epistemic behavior
4. 4Apply findings to improve RL agent design and performance under partial observability

ML researchers and AI engineers working on reinforcement learning and partial observability benefit from understanding the structural non-preservation of epistemic behavior, as it impacts the design of more effective RL agents

💡 Epistemic behavior in RL agents is not structurally preserved under policy transformation, limiting their effectiveness in complex environments

🤖 RL agents under partial observability struggle to preserve epistemic behavior under policy transformation #AI #RL