Process Rewards with Learned Reliability

Jinyuan Li, Langlin Huang, Chengsong Huang, Shaoyang Xu, Donghong Cai, Yuyi Yang, Wenxuan Zhang, Jiaxin Huang

arXiv:2605.15529 · 2026-05-18 공개 · arXiv · PDF

token-efficiency reasoning-benchmarks best-of-n process-reward-models reliability-signal continuation-estimation beta-prm adaptive-computation-allocation

Abstract

Process Reward Models (PRMs) provide step-level feedback for reasoning, but current PRMs usually output only a single reward score for each step. Downstream methods must therefore treat imperfect step-level reward predictions as reliable decision signals, with no indication of when these predictions should be trusted. We propose BetaPRM, a distributional PRM that predicts both a step-level success probability and the reliability of that prediction. Given step-success supervision from Monte Carlo continuations, BetaPRM learns a Beta belief that explains the observed number of successful continuations through a Beta-Binomial likelihood, rather than regressing to the finite-sample success ratio as a point target. This learned reliability signal indicates when a step reward should be trusted, enabling downstream applications to distinguish reliable rewards from uncertain ones. As one application, we introduce Adaptive Computation Allocation (ACA) for PRM-guided Best-of-N reasoning. ACA uses the learned reliability signal to stop when a high-reward solution is reliable and to spend additional computation on uncertain candidate prefixes. Experiments across four backbones and four reasoning benchmarks show that BetaPRM improves PRM-guided Best-of-N selection while preserving standard step-level error detection. Built on this signal, ACA improves the accuracy--token tradeoff over fixed-budget Best-of-16, reducing token usage by up to 33.57% while improving final-answer accuracy.

한국어 요약

📋 한 줄 요약

**[LLM 추론 / Process Reward]** 단계별 보상과 함께 그 신뢰도를 학습하는 분포형 PRM과 적응적 연산 할당.

🎯 핵심 기여도

💡 핵심 아이디어

기존 PRM은 단계별로 단일 보상 점수만 출력해 신뢰 여부를 알 수 없다. 다운스트림은 불완전한 예측을 그대로 신뢰 신호로 사용한다. BetaPRM은 Beta 신념을 학습해 "보상이 얼마나 확실한가"를 명시적으로 모델링함으로써 자원 할당을 최적화한다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: PRM의 핵심 결함(보상의 불확실성 미표현)을 분포형 학습으로 해소, 추론 비용-정확도 트레이드오프를 동적으로 개선한다. **한계**: Beta 분포 가정의 적합성은 단계 수와 도메인에 따라 다를 수 있으며, MC continuation 라벨 수집의 비용은 여전히 존재한다.

🚀 실용적 활용