Search papers, labs, and topics across Lattice.
This paper explores the intersection of reinforcement learning (RL) and potential theory, highlighting how potential-theoretic perspectives can enhance core RL representations and algorithms under a fixed-policy assumption. By leveraging this connection, the authors propose methods that could lead to improved sample efficiency and introduce formal constraints applicable to RL. Additionally, they extend their findings to nonlinear scenarios when the fixed-policy assumption is relaxed, suggesting broader implications for RL methodologies.
Potential theory could unlock new levels of sample efficiency in reinforcement learning algorithms.
Reinforcement learning (RL) theory fundamentally depends on probability theory through the Markov chain. There is a deep connection between probability theory and potential theory. This paper reviews that connection and explores the potential-theoretic viewpoint for core reinforcement learning representations and algorithms under a fixed-policy assumption. This viewpoint may offer a path for improved sample efficiency and formal constraints that can be applied to RL. When the fixed-policy assumption is relaxed, the linear potential theory framework can be naturally extended to the nonlinear case.
