SPRO redefines step-level advantage by grouping cumulative policy-reference log-ratio rewards within same-step groups, reporting accuracy and efficiency gains over GRPO and PRIME.
Learning to reason with LLMs
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.LG 1years
2025 1verdicts
REJECT 1roles
background 1polarities
support 1representative citing papers
citing papers explorer
-
Self-Guided Process Reward Optimization with Redefined Step-wise Advantage for Process Reinforcement Learning
SPRO redefines step-level advantage by grouping cumulative policy-reference log-ratio rewards within same-step groups, reporting accuracy and efficiency gains over GRPO and PRIME.