| title | Process Rewards with Learned Reliability |
|---|---|
| authors | Jinyuan Li, Langlin Huang, Chengsong Huang, Shaoyang Xu, Donghong Cai, Yuyi Yang, Wenxuan Zhang, Jiaxin Huang |
| date | 2026-05-15 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2605.15529 |
| affiliation | Huang's INTelligence lab |
| upvotes | 44 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf code |
| 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. |
cite
@misc{li2026process,
title={Process Rewards with Learned Reliability},
author={Jinyuan Li and Langlin Huang and Chengsong Huang and Shaoyang Xu and Donghong Cai and Yuyi Yang and Wenxuan Zhang and Jiaxin Huang},
year={2026},
eprint={2605.15529},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2605.15529}
}