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

📅 2025-07-02
📈 Citations: 0
Influential: 0
📄 PDF
🤖 AI Summary
To address two key challenges in Process Reinforcement Learning (PRL)—high computational overhead from external reward models and the absence of a unified theoretical framework for process-level advantage estimation—this paper proposes a Self-Guided Process Reward Optimization (SG-PRO) framework. Methodologically, it theoretically proves that the policy model itself can intrinsically generate process rewards; introduces Cumulative Process Reward and Masked Stepwise Advantage (MSA) to enable fine-grained, step-level advantage estimation under shared prompts, eliminating the need for auxiliary reward models; and integrates policy entropy regularization with response-length constraints. Experiments demonstrate that SG-PRO achieves 3.4× higher training efficiency and 17.5% higher test accuracy than GRPO, reduces average response length by ~33%, and maintains stable training entropy—effectively mitigating reward hacking.

Technology Category

Application Category

📝 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 extbf{S}elf-Guided extbf{P}rocess extbf{R}eward extbf{O}ptimization~( extbf{SPRO}), a novel framework that enables process-aware RL through two key innovations: (1) we first theoretically demonstrate that process rewards can be derived intrinsically from the policy model itself, and (2) we introduce well-defined cumulative process rewards and extbf{M}asked extbf{S}tep extbf{A}dvantage ( extbf{MSA}), which facilitates rigorous step-wise action advantage estimation within shared-prompt sampling groups. Our experimental results demonstrate that SPRO outperforms vaniila GRPO with 3.4x 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.
Problem

Research questions and friction points this paper is trying to address.

Reducing computational overhead in process reward models
Unifying theoretical framework for process-level advantage estimation
Enhancing reasoning capabilities of Large Language Models
Innovation

Methods, ideas, or system contributions that make the work stand out.

Self-guided intrinsic process reward derivation
Masked Step Advantage for step-wise estimation
No additional computational overhead design
🔎 Similar Papers
No similar papers found.
W
Wu Fei
Terminus Group, The Hong Kong University of Science and Technology (Guangzhou)
H
Hao Kong
Terminus Group
S
Shuxian Liang
Terminus Group
Y
Yang Lin
Terminus Group
Y
Yibo Yang
King Abdullah University of Science and Technology
J
Jing Tang
The Hong Kong University of Science and Technology (Guangzhou), The Hong Kong University of Science and Technology
L
Lei Chen
The Hong Kong University of Science and Technology (Guangzhou), The Hong Kong University of Science and Technology
X
Xiansheng Hua
Terminus Group