paper-with-me

홈 › Papers

Reinforcement Learning without Ground-Truth Solutions can Improve LLMs

2026-06-25 · Yingyu Lin, Qiyue Gao, Nikki Lijing Kuang, Xunpeng Huang, Kun Zhou, Tongtong Liang, Zhewei Yao, Yi-An Ma, Yuxiong He arxiv

Reinforcement learning with verifiable rewards (RLVR) for training LLMs typically rely on ground-truth answers to assign rewards, limiting their applicability to tasks where the ground-truth solution is unknown. We introduce a \textbf{R}anking-\textbf{i}nduced \textbf{VER}ifiable framework (RiVER) that trains LLMs on score-based optimization tasks without ground-truth solutions, using deterministic execution feedback as continuous-valued supervision. When applying group-relative RL to such continuous rewards, we identify two key challenges: \emph{scale dominance}, where uncalibrated score magnitudes across test instances distort policy updates, and \emph{frequency dominance}, where repeatedly sampled suboptimal solutions can outweigh rare but stronger candidates. RiVER addresses these challenges with calibrated reward shaping that uses instance-wise comparisons and emphasizes top-ranked solvers while retaining bounded feedback for other valid solutions. We train on 12 AtCoder Heuristic Contest tasks and evaluate on Algorithm Engineering Benchmark (ALE-Bench), LiveCodeBench, and USACO. RiVER advances Qwen3-8B and GLM-Z1-9B-0414 by 8.9\% and 9.4\% in ALE rating rank. More importantly, despite training exclusively on score-based tasks without any ground-truth solutions, RiVER also improves the backbones across exact-solution benchmarks such as LiveCodeBench and USACO by an absolute average improvement of 2.4\% and 3.5\%. By contrast, baselines trained with raw execution scores improve ALE rating but fail to transfer to exact-solution benchmarks. These results suggest that score-based optimization tasks, combined with proper reward calibration, can serve as effective training environments for general coding ability without ground-truth solutions.

📄 PDF Abstract BibTeX arXiv:2606.27369

Code (0)

등록된 구현이 없습니다.

Tasks

Reinforcement Learning

Similar Papers 제목 키워드 기반

Test-time Recursive Thinking: Self-Improvement without External Feedback

2026-02-03 · Yufan Zhuang, Chandan Singh, Liyuan Liu, Yelong Shen 외 arxiv

Modern Large Language Models (LLMs) have shown rapid improvements in reasoning capabilities, driven largely by reinforcement learning (RL) with verifiable rewards. Here, we ask whether these LLMs can self-improve without…

Reinforcement Learning

Self Rewarding Self Improving

2025-05-12 · Toby Simonds, Kevin Lopez, Akira Yoshiyama, Dominique Garmier

We demonstrate that large language models can effectively self-improve through self-judging without requiring reference solutions, leveraging the inherent asymmetry between generating and verifying solutions. Our experim…

Question GenerationQuestion-Generationreinforcement-learningReinforcement Learning

Error Correcting Algorithms for Sparsely Correlated Regressors

2019-05-16 · ICML Workshop AMTL 2019 6 · Anonymous

Autonomy and adaptation of machines requires that they be able to measure their own errors. We consider the advantages and limitations of such an approach when a machine has to measure the error in a regression task. How…

compressed sensingregression

Error Correcting Algorithms for Sparsely Correlated Regressors

2019-06-17 · Andrés Corrada-Emmanuel, Edward Zahrebelski, Edward Pantridge

Autonomy and adaptation of machines requires that they be able to measure their own errors. We consider the advantages and limitations of such an approach when a machine has to measure the error in a regression task. How…

compressed sensingregression

Reinforcement Learning from Meta-Evaluation: Aligning Language Models Without Ground-Truth Labels

2026-01-29 · Micah Rentschler, Jesse Roberts arxiv

Most reinforcement learning (RL) methods for training large language models (LLMs) require ground-truth labels or task-specific verifiers, limiting scalability when correctness is ambiguous or expensive to obtain. We int…

Reinforcement Learning