ProAS: A Progress-Aware Advantage Shaping Framework for Multi-Turn Agentic Reinforcement Learning
Progress-Aware Advantage Shaping (ProAS) is proposed, a plug-and-play framework that leverages a dynamically evolving Agentic Reference Graph (ARG) to achieve fine-grained credit assignment via Masking Unreasonable Incentives (MUI) and Asymmetric Penalty Redistribution (APR).
Bowen Ding, Ming-Da Wan, Yu-Han Chen et al.
· 0 citations