paper-with-me

홈 › Papers

Beyond Binary Rewards: Training LMs to Reason About Their Uncertainty

2025-07-22 · Mehul Damani, Isha Puri, Stewart Slocum, Idan Shenfeld, Leshem Choshen, Yoon Kim, Jacob Andreas

When language models (LMs) are trained via reinforcement learning (RL) to generate natural language "reasoning chains", their performance improves on a variety of difficult question answering tasks. Today, almost all successful applications of RL for reasoning use binary reward functions that evaluate the correctness of LM outputs. Because such reward functions do not penalize guessing or low-confidence outputs, they often have the unintended side-effect of degrading calibration and increasing the rate at which LMs generate incorrect responses (or "hallucinate") in other problem domains. This paper describes RLCR (Reinforcement Learning with Calibration Rewards), an approach to training reasoning models that jointly improves accuracy and calibrated confidence estimation. During RLCR, LMs generate both predictions and numerical confidence estimates after reasoning. They are trained to optimize a reward function that augments a binary correctness score with a Brier score -- a scoring rule for confidence estimates that incentivizes calibrated prediction. We first prove that this reward function (or any analogous reward function that uses a bounded, proper scoring rule) yields models whose predictions are both accurate and well-calibrated. We next show that across diverse datasets, RLCR substantially improves calibration with no loss in accuracy, on both in-domain and out-of-domain evaluations -- outperforming both ordinary RL training and classifiers trained to assign post-hoc confidence scores. While ordinary RL hurts calibration, RLCR improves it. Finally, we demonstrate that verbalized confidence can be leveraged at test time to improve accuracy and calibration via confidence-weighted scaling methods. Our results show that explicitly optimizing for calibration can produce more generally reliable reasoning models.

📄 PDF Abstract BibTeX arXiv:2507.16806

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

BeamPERL: Parameter-Efficient RL with Verifiable Rewards Specializes Compact LLMs for Structured Beam Mechanics Reasoning

2026-03-04 · Tarjei Paule Hage, Markus J. Buehler arxiv

Can reinforcement learning with hard, verifiable rewards teach a compact language model to reason about physics, or does it primarily learn to pattern-match toward correct answers? We study this question by training a 1.…

Reinforcement Learning

Forge: Quality-Aware Reinforcement Learning for NP-Hard Optimization in LLMs

2026-05-09 · Xiaozhe Li, Xinyu Fang, Shengyuan Ding, Yang Li 외 arxiv

Large Language Models (LLMs) have achieved remarkable success on reasoning benchmarks through Reinforcement Learning with Verifiable Rewards (RLVR), excelling at tasks such as math, coding, logic, and puzzles. However, e…

Reinforcement LearningInstruction Following

Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains

2025-07-23 · Anisha Gunjal, Anthony Wang, Elaine Lau, Vaskar Nath 외 arxiv

Reinforcement Learning with Verifiable Rewards (RLVR) has proven effective for complex reasoning tasks with clear correctness signals such as math and coding. However, extending it to real-world reasoning tasks is challe…

Reinforcement Learning

Rubric-Grounded RL: Structured Judge Rewards for Generalizable Reasoning

2026-05-08 · Manish Bhattarai, Ismael Boureima, Nishath Rajiv Ranasinghe, Scott Pakin 외 arxiv

We argue that decomposing reward into weighted, verifiable criteria and using an LLM judge to score them provides a partial-credit optimization signal: instead of a binary outcome or a single holistic score, each respons…

Reinforcement Learning

Beyond Reasoning: Reinforcement Learning Unlocks Parametric Knowledge in LLMs

2026-05-08 · Wanli Yang, Hongyu Zang, Junwei Zhang, Wenjie Shi 외 arxiv

Reinforcement learning (RL) has achieved remarkable success in LLM reasoning, but whether it can also improve direct recall of parametric knowledge remains an open question. We study this question in a controlled zero-sh…

Reinforcement Learning