paper-with-me

홈 › Papers

Uncalibrated Reasoning: GRPO Induces Overconfidence for Stochastic Outcomes

2025-08-15 · Michael Bereket, Jure Leskovec arxiv

Reinforcement learning (RL) has proven remarkably effective at improving the accuracy of language models in verifiable and deterministic domains like mathematics. Here, we examine if current RL methods are also effective at optimizing language models in verifiable domains with stochastic outcomes, like scientific experiments. Through applications to synthetic data and real-world biological experiments, we demonstrate that Group Relative Policy Optimization (GRPO) induces overconfident probability predictions for binary stochastic outcomes, while Proximal Policy Optimization (PPO) and REINFORCE Leave-One-Out (RLOO) yield well-calibrated models. We show that removing group standard normalization in GRPO fixes its miscalibration and provide a theoretical explanation for why normalization causes overconfidence. Our results provide new evidence against the use of standard normalization in GRPO and help pave the way for applications of RL for reasoning language models beyond deterministic domains.

📄 PDF Abstract BibTeX arXiv:2508.11800

Code (0)

등록된 구현이 없습니다.

Tasks

Reinforcement Learning

Similar Papers 제목 키워드 기반

Calibration-Aware Policy Optimization for Reasoning LLMs

2026-04-14 · Ziqi Wang, Xingzhou Lou, Meiqi Wu, Zhengqi Wen 외 arxiv

Group Relative Policy Optimization (GRPO) enhances LLM reasoning but often induces overconfidence, where incorrect responses yield lower perplexity than correct ones, degrading relative calibration as described by the Ar…

Mathematical Reasoning

C$^2$GSPG: Confidence-calibrated Group Sequence Policy Gradient towards Self-aware Reasoning

2025-09-27 · Haotian Liu, Shuo Wang, Hongteng Xu arxiv

Reinforcement Learning (RL) methods, exemplified by Group Relative Policy Optimization (GRPO) and its variants, play a central role in developing reasoning models. However, these methods often suffer from a critical over…

Reinforcement LearningMathematical Reasoning

GRPO Beyond English: A Large-Scale Study of GRPO in Non-English and Multilingual Settings

2026-08-13 · Konstantin Dobler, Federico Scozzafava, Jonathan Janke, Mohamed Ali 외 arxiv

Reinforcement Learning with Verifiable Rewards (RLVR), often optimized with Group Relative Policy Optimization (GRPO), has become a central recipe for improving the reasoning capabilities of pretrained language models bu…

Reinforcement Learning

Aligning LLM Uncertainty with Human Disagreement in Subjectivity Analysis

2026-05-11 · Junyu Lu, Deyi Ji, Xuanyi Liu, Lanyun Zhu 외 arxiv

Large language models for subjectivity analysis are typically trained with aggregated labels, which compress variations in human judgment into a single supervision signal. This paradigm overlooks the intrinsic uncertaint…

Subjectivity Analysis

Learning from Own Solutions: Self-Conditioned Credit Assignment for Reinforcement Learning with Verifiable Rewards

2026-06-17 · Yingyu Shan, Yuhang Guo, Zihao Cheng, Zeming Liu 외 arxiv

Reinforcement learning with verifiable rewards (RLVR) has driven substantial progress in training LLMs for reasoning tasks, but representative methods such as GRPO assign uniform credit across all tokens, wasting gradien…

Reinforcement LearningKnowledge Distillation