Self-Guided Process Reward Optimization with Redefined Step-wise Advantage for Process Reinforcement Learning

Wu Fei, Shuxian Liang, Yibo Yang, Yang Lin, Jing Tang, Lei Chen, Xian-Sheng Hua, Hao Kong

Keywords: Large Language Models, Reinforcement Learning

TL;DR: we propose \textbf{S}elf-Guided \textbf{P}rocess \textbf{R}eward \textbf{O}ptimization (\textbf{SPRO}), a novel framework that enables process-aware RL.

Abstract:

Process Reinforcement Learning (PRL) has demonstrated considerable potential in enhancing the reasoning capabilities of Large Language Models (LLMs).
However, introducing additional process reward models incurs substantial computational overhead, and there is no unified theoretical framework for process-level advantage estimation.
To bridge this gap, we propose \textbf{S}elf-Guided \textbf{P}rocess \textbf{R}eward \textbf{O}ptimization (\textbf{SPRO}), a novel framework that enables process-aware RL through two key innovations: (1) we show that process rewards can be derived intrinsically from the policy model itself, and (2) we redefine the step-wise advantage by introducing well-defined Cumulative Process Rewards (\textbf{CPR}) and \textbf{M}asked \textbf{S}tep \textbf{A}dvantage (\textbf{MSA}), which facilitates rigorous step-wise action advantage estimation within shared-prompt sampling groups.
Our experimental results demonstrate that SPRO outperforms vaniila GRPO with 3.4× higher training efficiency and a 17.5\% test accuracy improvement.
Furthermore, SPRO maintains a stable and elevated policy entropy throughout training while reducing the average response length by approximately $1/3$, evidencing sufficient exploration and prevention of reward hacking.
Notably, SPRO incurs no additional computational overhead compared to outcome-supervised RL methods such as GRPO, which benefit industrial implementation.

Primary Area: reinforcement learning

Submission Number: 1905