Reflective Preference Optimization (RPO): Enhancing On-Policy Alignment via Hint-Guided Reflection

• URL: http://arxiv.org/abs/2512.13240v1
• Date: Mon, 15 Dec 2025 11:55:55 GMT
• Title: Reflective Preference Optimization (RPO): Enhancing On-Policy Alignment via Hint-Guided Reflection
• Authors: Zihui Zhao, Zechang Li,
• Abstract summary: We introduce Reflective Preference Optimization (RPO), a new framework that incorporates hint-guided reflection into the DPO paradigm.<n>RPO uses external models to identify hallucination sources and generate concise reflective hints, enabling the construction of on-policy preference pairs with stronger contrastiveness and clearer preference signals.<n> Empirically, RPO achieves superior alignment with fewer training samples and iterations, substantially reducing hallucination rates and delivering state-of-the-art performance across multimodal benchmarks.
• Score: 0.8287206589886881
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Direct Preference Optimization (DPO) has emerged as a lightweight and effective alternative to Reinforcement Learning from Human Feedback (RLHF) and Reinforcement Learning with AI Feedback (RLAIF) for aligning large language and vision-language models. However, the standard DPO formulation, in which both the chosen and rejected responses are generated by the same policy, suffers from a weak learning signal because the two responses often share similar errors and exhibit small Kullback-Leibler (KL) divergence. This leads to slow and unstable convergence. To address this limitation, we introduce Reflective Preference Optimization (RPO), a new framework that incorporates hint-guided reflection into the DPO paradigm. RPO uses external models to identify hallucination sources and generate concise reflective hints, enabling the construction of on-policy preference pairs with stronger contrastiveness and clearer preference signals. We theoretically show that conditioning on hints increases the expected preference margin through mutual information and improves sample efficiency while remaining within the policy distribution family. Empirically, RPO achieves superior alignment with fewer training samples and iterations, substantially reducing hallucination rates and delivering state-of-the-art performance across multimodal benchmarks.

Related papers

• Rethinking the Trust Region in LLM Reinforcement Learning [72.25890308541334]
  Proximal Policy Optimization (PPO) serves as the de facto standard algorithm for Large Language Models (LLMs)<n>We propose Divergence Proximal Policy Optimization (DPPO), which substitutes clipping with a more principled constraint.<n>DPPO achieves superior training and efficiency compared to existing methods, offering a more robust foundation for RL-based fine-tuning.
  arXiv  Detail & Related papers  (2026-02-04T18:59:04Z)
• Coverage Improvement and Fast Convergence of On-policy Preference Learning [67.36750525893514]
  Online on-policy preference learning algorithms for language model alignment can significantly outperform their offline counterparts.<n>We analyze how the sampling policy's coverage evolves throughout on-policy training.<n>We develop principled on-policy schemes for reward distillation in the general function class setting.
  arXiv  Detail & Related papers  (2026-01-13T10:46:06Z)
• ARF-RLHF: Adaptive Reward-Following for RLHF through Emotion-Driven Self-Supervision and Trace-Biased Dynamic Optimization [6.472219867780061]
  We introduce Adaptive Reward-Following (ARF), which converts natural feedback into continuous preference trajectories.<n>ARF consistently outperforms PPO and DPO, improving alignment by up to 7.6%.<n>Our results demonstrate that continuous reward modeling provides a scalable path toward personalized and theoretically grounded RLHF.
  arXiv  Detail & Related papers  (2025-07-03T17:59:26Z)
• Value-Free Policy Optimization via Reward Partitioning [0.08192907805418585]
  We introduce Reward Partitioning Optimization (RPO), a new method for single-trajectory reinforcement learning.<n>RPO normalizes observed rewards using a approach estimated directly from data.<n>We validate RPO on scalar-feedback language modeling tasks using Flan-T5 encoder-decoder models.
  arXiv  Detail & Related papers  (2025-06-16T17:06:27Z)
• Understanding the Performance Gap in Preference Learning: A Dichotomy of RLHF and DPO [53.914481648817066]
  We decompose the performance gap between reinforcement learning from human feedback and direct preference optimization under a representation gap.<n>We show that RLHF, DPO, or online DPO can outperform one another depending on type of model mis-specifications.
  arXiv  Detail & Related papers  (2025-05-26T09:54:02Z)
• Multi-Preference Optimization: Generalizing DPO via Set-Level Contrasts [17.243429150450886]
  We propose $textbfMulti-Preference Optimization (MPO) to optimize over entire sets of responses.<n>MPO employs deviation-based weighting, which emphasizes outlier responses that deviate most from the mean reward.<n>We theoretically prove that MPO reduces alignment bias at a rate of $mathcalOleft(frac1sqrtnright)$ with respect to the number of responses per query.
  arXiv  Detail & Related papers  (2024-12-05T21:50:22Z)
• Bridging and Modeling Correlations in Pairwise Data for Direct Preference Optimization [75.1240295759264]
  We propose an effective framework for Bridging and Modeling Correlations in pairwise data, named BMC.<n>We increase the consistency and informativeness of the pairwise preference signals through targeted modifications.<n>We identify that DPO alone is insufficient to model these correlations and capture nuanced variations.
  arXiv  Detail & Related papers  (2024-08-14T11:29:47Z)
• Provably Mitigating Overoptimization in RLHF: Your SFT Loss is Implicitly an Adversarial Regularizer [52.09480867526656]
  We identify the source of misalignment as a form of distributional shift and uncertainty in learning human preferences.<n>To mitigate overoptimization, we first propose a theoretical algorithm that chooses the best policy for an adversarially chosen reward model.<n>Using the equivalence between reward models and the corresponding optimal policy, the algorithm features a simple objective that combines a preference optimization loss and a supervised learning loss.
  arXiv  Detail & Related papers  (2024-05-26T05:38:50Z)
• REBEL: Reinforcement Learning via Regressing Relative Rewards [59.68420022466047]
  We propose REBEL, a minimalist RL algorithm for the era of generative models.<n>In theory, we prove that fundamental RL algorithms like Natural Policy Gradient can be seen as variants of REBEL.<n>We find that REBEL provides a unified approach to language modeling and image generation with stronger or similar performance as PPO and DPO.
  arXiv  Detail & Related papers  (2024-04-25T17:20:45Z)
• RS-DPO: A Hybrid Rejection Sampling and Direct Preference Optimization Method for Alignment of Large Language Models [7.676477609461592]
  Reinforcement learning from human feedback (RLHF) has been extensively employed to align large language models with user intent. DPO relies on contrastive responses generated from human annotator and alternative LLM, instead of the policy model. In this paper, we address both challenges by systematically combining sampling rejection (RS) and DPO. Our proposed method effectively fine-tunes LLMs with limited resource environments, leading to improved alignment with user intent.
  arXiv  Detail & Related papers  (2024-02-15T16:00:58Z)
• Statistical Rejection Sampling Improves Preference Optimization [42.57245965632205]
  We introduce a novel approach to source preference data from the target optimal policy using rejection sampling. We also propose a unified framework that enhances the loss functions used in both Sequence Likelihood (SLiC) and Direct Preference Optimization (DPO) from a preference modeling standpoint.
  arXiv  Detail & Related papers  (2023-09-13T01:07:25Z)

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.