<ul><li>Reinforcement learning agents can achieve superhuman performance, but their decisions are often difficult to interpret.</li><li>A theoretical framework has been developed to explain reinforcement learning through the influence of state features.</li><li>Three core elements of the agent-environment interaction benefiting from explanation are identified: behavior, performance, and value estimation.</li><li>The framework uses Shapley values from cooperative game theory to identify the influence of each state feature, offering mathematically grounded explanations with clear semantics.</li></ul>

A Theoretical Framework for Explaining Reinforcement Learning with Shapley Values

Discover more