paper-with-me

Papers

Directional Alignment Mitigates Reward Hacking in Reinforcement Learning for Language Models

2026-05-24 · Wenlong Deng, Jiaji Huang, Kaan Ozkara, Yushu Li, Christos Thrampoulidis, Xiaoxiao Li, Youngsuk Park arxiv

Reward hacking arises when a model improves a proxy reward by exploiting shortcuts rather than solving the intended task. We study this failure mode through the geometry of reinforcement learning updates in language models and argue that hacking emerges when optimization drifts away from a stable low-dimensional learning trajectory. We analyze this drift through dominant singular directions of parameter updates and show that reward-hacking runs exhibit substantially larger directional change than clean runs. Motivated by this observation, we introduce trusted-direction projection, which constrains gradients to remain within a clean reference subspace. Across reward-hacking experiments on mathematical reasoning, the proposed approach delays shortcut exploitation and better preserves task performance.

📄 PDF Abstract BibTeX arXiv:2605.25189

Code (0)

등록된 구현이 없습니다.

Tasks

Reinforcement LearningMathematical Reasoning

Similar Papers 제목 키워드 기반

Taming Preference Mode Collapse via Directional Decoupling Alignment in Diffusion Reinforcement Learning

2025-12-30 · Chubin Chen, Sujie Hu, Jiashu Zhu, Meiqi Wu 외 arxiv

Recent studies have demonstrated significant progress in aligning text-to-image diffusion models with human preference via Reinforcement Learning from Human Feedback. However, while existing methods achieve high scores o…

Reinforcement Learning

f-GRPO and Beyond: Divergence-Based Reinforcement Learning Algorithms for General LLM Alignment

2026-02-05 · Rajdeep Haldar, Lantao Mei, Guang Lin, Yue Xing 외 arxiv

Recent work shows that preference alignment objectives can be interpreted as divergence estimators between aligned (preferred) & unaligned (less-preferred) distributions, yielding a principled recipe for designing alignm…

Reinforcement Learning

PromptLoop: Plug-and-Play Prompt Refinement via Latent Feedback for Diffusion Model Alignment

2025-10-01 · Suhyeon Lee, Jong Chul Ye arxiv

Despite recent progress, reinforcement learning (RL)-based fine-tuning of diffusion models often struggles with generalization, composability, and robustness against reward hacking. Recent studies have explored prompt re…

Reinforcement Learning

IR$^3$: Contrastive Inverse Reinforcement Learning for Interpretable Detection and Mitigation of Reward Hacking

2026-02-23 · Mohammad Beigi, Ming Jin, Junshan Zhang, Jiaxin Zhang 외 arxiv

Reinforcement Learning from Human Feedback (RLHF) enables powerful LLM alignment but can introduce reward hacking - models exploit spurious correlations in proxy rewards without genuine alignment. Compounding this, the o…

Reinforcement Learning

Correlated Proxies: A New Definition and Improved Mitigation for Reward Hacking

2024-03-05 · Cassidy Laidlaw, Shivam Singhal, Anca Dragan

Because it is difficult to precisely specify complex objectives, reinforcement learning policies are often optimized using proxy reward functions that only approximate the true goal. However, optimizing proxy rewards fre…

reinforcement-learningReinforcement Learning