Adversarial Training for Process Reward Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Juneja, Gurusha, Nathani, Deepak, Wang, William Yang
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909931273191424
author Juneja, Gurusha
Nathani, Deepak
Wang, William Yang
author_facet Juneja, Gurusha
Nathani, Deepak
Wang, William Yang
contents Process Reward Models (PRMs) enhance reasoning ability of LLMs by providing step-level supervision. However, their widespread adoption is limited due to expensive manual step-level annotation and poor generalization of static training data to novel errors. We introduce Adversarially Trained PRMs (\texttt{APRM}), where a Generator ($G$) learns to produce reasoning errors to deceive a PRM ($R$), while $R$ concurrently learns to detect them. This interaction yields progressively harder negatives for $R$, improving its robustness and generalization to novel errors without requiring manual step-level labels. Averaged across diverse mathematical reasoning benchmarks, \texttt{APRM} improves solver accuracy by $+3.4$ percentage points (pp) over the strongest PRM baseline. \texttt{APRM} achieves gains of $+5.3$ pp on out-of-distribution tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2511_22888
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Adversarial Training for Process Reward Models
Juneja, Gurusha
Nathani, Deepak
Wang, William Yang
Machine Learning
Artificial Intelligence
Process Reward Models (PRMs) enhance reasoning ability of LLMs by providing step-level supervision. However, their widespread adoption is limited due to expensive manual step-level annotation and poor generalization of static training data to novel errors. We introduce Adversarially Trained PRMs (\texttt{APRM}), where a Generator ($G$) learns to produce reasoning errors to deceive a PRM ($R$), while $R$ concurrently learns to detect them. This interaction yields progressively harder negatives for $R$, improving its robustness and generalization to novel errors without requiring manual step-level labels. Averaged across diverse mathematical reasoning benchmarks, \texttt{APRM} improves solver accuracy by $+3.4$ percentage points (pp) over the strongest PRM baseline. \texttt{APRM} achieves gains of $+5.3$ pp on out-of-distribution tasks.
title Adversarial Training for Process Reward Models
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2511.22888