The paper addresses policy optimization when rewards are noisy, restricting the policy to stay near a reference policy. It defines risk as the probability of ending up with a worse policy than the reference and shows that, under a Gaussian reward model, the optimal regularizer is a quadratic norm weighted by the reward covariance. Using this insight, the authors propose a new algorithm that regularizes with reward variance estimates, and experiments on synthetic bandits, large language models, and reward models demonstrate that it yields more robust alignment outcomes than existing methods.
arXiv:2607. 03248v1 Announce Type: cross Abstract: The alignment of large language models with human preferences is commonly achieved through Reinforcement Learning from Human Feedback or Direct Preference Optimization.
GVPO: Group Variance Policy Optimization is a new post‑training method for large language models that integrates the analytical solution to KL‑constrained reward maximization into its gradient weights. It guarantees a unique optimal solution and allows flexible sampling distributions, avoiding the on‑policy and importance sampling limitations of previous techniques. The approach offers a theoretical guarantee while remaining practically adaptable, positioning it as a reliable and versatile paradigm for LLM post‑training.
The paper introduces a doubly robust preference optimization algorithm for reinforcement learning from human feedback (RLHF) in large language models. It addresses sensitivity to misspecifications in the preference model, reference policy, or reward function by ensuring consistency when either the preference model or the reference policy is correctly specified. Experiments show that the method outperforms state‑of‑the‑art algorithms both theoretically and empirically.
arXiv:2606. 19818v1 Announce Type: cross Abstract: Reinforcement learning from human feedback (RLHF) aligns large language models by training reward models on preference data and optimizing policies to maximize predicted rewards.
arXiv:2606. 03962v1 Announce Type: cross Abstract: Classical reinforcement learning (RL) typically seeks a deterministic policy that maximizes the expected sum of a scalar reward.
arXiv:2506. 12529v2 Announce Type: replace-cross Abstract: Preference-based Reinforcement Learning (PbRL) entails a variety of approaches for aligning models with human intent to alleviate the burden of reward engineering.
arXiv:2602. 02572v2 Announce Type: replace-cross Abstract: Existing alignment methods directly use the reward model learned from user preference data to optimize an LLM policy, subject to KL regularization with respect to the base policy.
arXiv:2607. 26094v1 Announce Type: new Abstract: Reinforcement Learning from Human Feedback (RLHF) is the standard approach for aligning large language models with human preferences, but its quality is limited by static, task-agnostic reward models.
Classical reinforcement learning (RL) typically seeks a deterministic policy that maximizes the expected sum of a scalar reward. Yet, modern applications such as language model fine-tuning or scientific discovery demand diversity.
arXiv:2607. 07859v1 Announce Type: new Abstract: Reinforcement learning (RL) research has increasingly shifted focus towards alignment, ensuring agents learn behaviors adhering to human values.
The paper introduces Personalized Group Relative Policy Optimization (P‑GRPO), a new alignment framework for large language models that separates advantage estimation from immediate batch statistics. By normalizing advantages using preference‑group‑specific reward histories instead of the concurrent generation group, P‑GRPO maintains contrastive signals for distinct user preferences. Experiments across various tasks show that P‑GRPO converges faster and yields higher rewards than standard GRPO, improving alignment with heterogeneous human preferences while preserving general capabilities.
The paper introduces Collaborative Reward Modeling (CRM), an online framework that uses cross-model peer review and curriculum learning to identify reliable preference pairs for training reward models (RMs). By having two RMs evaluate each other's sample selections and synchronizing their learning progress, CRM reduces the impact of noisy feedback, improving RM generalization by up to 9.94 points on RewardBench under 40% preference noise. The approach also extends to implicit-reward alignment, demonstrating robustness and broad applicability for learning from imperfect feedback.