Prism: Policy Reuse via Interpretable Strategy Mapping in Reinforcement Learning
arXiv:2604.02353v1 Announce Type: new
Abstract: We present PRISM (Policy Reuse via Interpretable Strategy Mapping), a framework that grounds reinforcement learning agents’ decisions in discrete, causally validated concepts and uses those concepts as a…