Unifying Stable Optimization and Reference Regularization in RLHF

• URL: http://arxiv.org/abs/2602.11523v1
• Date: Thu, 12 Feb 2026 03:31:19 GMT
• Title: Unifying Stable Optimization and Reference Regularization in RLHF
• Authors: Li He, Qiang Qu, He Zhao, Stephen Wan, Dadong Wang, Lina Yao, Tongliang Liu,
• Abstract summary: This paper introduces a unified regularization approach that balances objectives of preventing reward hacking and maintaining stable policy updates.<n>Our simple yet principled alignment objective yields a weighted supervised fine-tuning loss with a superior trade-off, which demonstrably improves both alignment results and implementation complexity.
• Score: 64.16830602324345
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Reinforcement Learning from Human Feedback (RLHF) has advanced alignment capabilities significantly but remains hindered by two core challenges: \textbf{reward hacking} and \textbf{stable optimization}. Current solutions independently address these issues through separate regularization strategies, specifically a KL-divergence penalty against a supervised fine-tuned model ($π_0$) to mitigate reward hacking, and policy ratio clipping towards the current policy ($π_t$) to promote stable alignment. However, the implicit trade-off arising from simultaneously regularizing towards both $π_0$ and $π_t$ remains under-explored. In this paper, we introduce a unified regularization approach that explicitly balances the objectives of preventing reward hacking and maintaining stable policy updates. Our simple yet principled alignment objective yields a weighted supervised fine-tuning loss with a superior trade-off, which demonstrably improves both alignment results and implementation complexity. Extensive experiments across diverse benchmarks validate that our method consistently outperforms RLHF and online preference learning methods, achieving enhanced alignment performance and stability.

Related papers

• Stabilizing Policy Optimization via Logits Convexity [59.242732612484474]
  We show that the convexity of the supervised fine-tuning loss with respect to model logits plays a key role in enabling stable training.<n>Motivated by this observation, we propose Logits Convex Optimization (LCO), a simple yet effective policy optimization framework.
  arXiv  Detail & Related papers  (2026-03-01T07:40:12Z)
• Ratio-Variance Regularized Policy Optimization for Efficient LLM Fine-tuning [48.34492357368989]
  We propose a primal-dual framework that supports stable on-policy learning and enables principled off-policy data reuse.<n>$R2VPO$ achieves superior performance with average relative gains of up to 17% over strong clipping-based baselines.
  arXiv  Detail & Related papers  (2026-01-06T14:01:42Z)
• The Realignment Problem: When Right becomes Wrong in LLMs [6.8304813545377]
  The alignment of Large Language Models with human values is central to their safe deployment, yet current models fail to keep pace with evolving norms and policies.<n>Existing unlearning methods act as blunt instruments that erode utility rather than enable precise policy updates.<n>We introduce TRACE, a framework for principled unlearning that reconceives realignment as a programmatic policy problem.
  arXiv  Detail & Related papers  (2025-11-04T14:52:58Z)
• GRPO-Guard: Mitigating Implicit Over-Optimization in Flow Matching via Regulated Clipping [63.33669214116784]
  GRPO-Guard is a simple yet effective enhancement to existing GRPO frameworks.<n>It restores a balanced and step-consistent importance ratio, ensuring that PPO clipping properly constrains harmful updates.<n>It substantially mitigates implicit over-optimization without relying on heavy KL regularization.
  arXiv  Detail & Related papers  (2025-10-25T14:51:17Z)
• BAPO: Stabilizing Off-Policy Reinforcement Learning for LLMs via Balanced Policy Optimization with Adaptive Clipping [69.74252624161652]
  We propose BAlanced Policy Optimization with Adaptive Clipping (BAPO)<n>BAPO dynamically adjusts clipping bounds to adaptively re-balance positive and negative contributions, preserve entropy, and stabilize RL optimization.<n>On AIME 2024 and AIME 2025 benchmarks, our 7B BAPO model surpasses open-source counterparts such as SkyWork-OR1-7B.
  arXiv  Detail & Related papers  (2025-10-21T12:55:04Z)
• ADARL: Adaptive Low-Rank Structures for Robust Policy Learning under Uncertainty [28.291179179647795]
  We propose textbfAdaptive Rank Representation (AdaRL), a bi-level optimization framework that improves robustness.<n>At the lower level, AdaRL performs policy optimization under fixed-rank constraints with dynamics sampled from a Wasserstein ball around a centroid model.<n>At the upper level, it adaptively adjusts the rank to balance the bias--variance trade-off, projecting policy parameters onto a low-rank manifold.
  arXiv  Detail & Related papers  (2025-10-13T20:05:34Z)
• Simplify RLHF as Reward-Weighted SFT: A Variational Method [34.222095430239555]
  Reinforcement Learning from Human Feedback (RLHF) is crucial for aligning Large Language Models (LLMs) with human values.<n>We propose a novel simplification of RLHF from the perspective of variational inference.<n>We transform the alignment objective into a reward-driven supervised fine-tuning form to obtain noticeable improvement on training stability and effectiveness.
  arXiv  Detail & Related papers  (2025-02-16T07:22:00Z)
• Correcting the Mythos of KL-Regularization: Direct Alignment without Overoptimization via Chi-Squared Preference Optimization [78.82586283794886]
  $chi2$-Preference Optimization ($chi$PO) is an efficient offline alignment algorithm provably robust to overoptimization.<n>$chi$PO implements the principle of pessimism in the face of uncertainty via regularization.<n>$chi$PO's simplicity and strong guarantees make it the first practical and general-purpose offline alignment algorithm provably robust to overoptimization.
  arXiv  Detail & Related papers  (2024-07-18T11:08:40Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.