Normalized Rewards for Preference Optimization

• URL: http://arxiv.org/abs/2607.16240v1
• Date: Fri, 26 Jun 2026 03:03:15 GMT
• Title: Normalized Rewards for Preference Optimization
• Abstract summary: DAAs have been observed to over-optimize their implicit reward model and decrease the likelihood of preferred responses.<n>We investigate how response likelihood changes are distributed over the tokens with and without regularization.<n>We find that a significant portion of the likelihood changes are due to a small set of outlier tokens.
• Score: 15.700605203114899
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Direct Alignment Algorithms (DAAs) such as DPO have become a common way to post-train and align LLMs with human preferences. However, DAAs have been observed to over-optimize their implicit reward model and decrease the likelihood of preferred responses. This results in a decrease in the total likelihood assigned to responses seen in the preference dataset, potentially resulting in undesirable behavior. To counteract this undesired side-effect of DAAs, we examine the effect of using objectives that add a regularization term to maintain the total length-normalized probabilities of the chosen and rejected responses. To better understand over-optimization, we investigate how response likelihood changes are distributed over the tokens with and without regularization. We find that a significant portion of the likelihood changes are due to a small set of outlier tokens, which explains how DAAs improve generation quality despite decreasing the likelihoods of chosen responses. We apply the proposed regularization to reference-based (DPO) and reference-free (SimPO) methods and find (1) improved trade-offs between generation quality and general benchmark capability and (2) improvements in reward modeling across datasets. For example, on Llama-3.1-8B-Instruct, we see both a >20% relative increase in AlpacaEval2 scores and >9% relative performance gains on general benchmarks. Additionally, we find that the added regularization term effectively mitigates the amount of displacement within preferred responses overall, and for the outlier tokens specifically, by utilizing low-likelihood tokens.

Related papers

• Distribution Preference Optimization: A Fine-grained Perspective for LLM Unlearning [26.120338506874976]
  Unlearning, which aims to remove the influence of specific data while preserving overall model utility, is becoming an important research area.<n>We derive a novel unlearning algorithm termed textbfDistribution textbfPreference textbfOptimization (DiPO)<n>DiPO attains the highest forget quality on the TOFU benchmark, and maintains leading scalability and sustainability on the MUSE benchmark.
  arXiv  Detail & Related papers  (2025-10-06T12:49:00Z)
• Towards Bridging the Reward-Generation Gap in Direct Alignment Algorithms [10.950194915191682]
  Direct Alignment Algorithms (DAAs) have emerged as efficient alternatives to Reinforcement Learning from Human Feedback (RLHF) algorithms.<n>DAAs suffer from a misalignment between optimization objectives during training and actual generation performance during inference.<n>We introduce a token-level MDP perspective of DAAs to analyze its limitations and introduce a simple yet effective approach called Prefix-Oriented Equal-length Training (POET)
  arXiv  Detail & Related papers  (2025-06-11T07:02:18Z)
• Proximalized Preference Optimization for Diverse Feedback Types: A Decomposed Perspective on DPO [19.5712961932773]
  We revisit direct preference optimization (DPO) and demonstrate that its loss theoretically admits a decomposed reformulation.<n>We introduce PRoximalized PReference Optimization (PRO), a unified method to align with diverse feeback types.
  arXiv  Detail & Related papers  (2025-05-29T10:23:22Z)
• On the Effect of Negative Gradient in Group Relative Deep Reinforcement Optimization [52.76330545825083]
  Reinforcement learning (RL) has become popular in enhancing the reasoning capabilities of large language models (LLMs)<n>We identify a previously unrecognized phenomenon we term Lazy Likelihood Displacement (LLD), wherein the likelihood of correct responses marginally increases or even decreases during training.<n>We develop a method called NTHR, which downweights penalties on tokens contributing to the LLD. Unlike prior DPO-based approaches, NTHR takes advantage of GRPO's group-based structure, using correct responses as anchors to identify influential tokens.
  arXiv  Detail & Related papers  (2025-05-24T18:58:51Z)
• Leveraging Robust Optimization for LLM Alignment under Distribution Shifts [51.74394601039711]
  Preference alignment methods are increasingly critical for steering large language models to generate outputs consistent with human values.<n>We propose a novel distribution-aware optimization framework that improves preference alignment despite such shifts.
  arXiv  Detail & Related papers  (2025-04-08T09:14:38Z)
• Uncertainty-Penalized Direct Preference Optimization [52.387088396044206]
  We develop a pessimistic framework for DPO by introducing preference uncertainty penalization schemes. The penalization serves as a correction to the loss which attenuates the loss gradient for uncertain samples. We show improved overall performance compared to vanilla DPO, as well as better completions on prompts from high-uncertainty chosen/rejected responses.
  arXiv  Detail & Related papers  (2024-10-26T14:24:37Z)
• Understanding Likelihood Over-optimisation in Direct Alignment Algorithms [20.043560907227018]
  Direct Alignment Algorithms (DAAs) have emerged as alternatives to online Reinforcement Learning from Human Feedback. These algorithms aim to increase the likelihood of generating better (preferred) completions while discouraging worse (non-preferred) ones. This work explores the relationship between completion likelihood and model performance in state-of-the-art DAAs.
  arXiv  Detail & Related papers  (2024-10-15T15:14:22Z)
• Reward-Augmented Data Enhances Direct Preference Alignment of LLMs [63.32585910975191]
  We introduce reward-conditioned Large Language Models (LLMs) that learn from the entire spectrum of response quality within the dataset.<n>We show that our approach consistently boosts DPO by a considerable margin.<n>Our method not only maximizes the utility of preference data but also mitigates the issue of unlearning, demonstrating its broad effectiveness beyond mere data expansion.
  arXiv  Detail & Related papers  (2024-10-10T16:01:51Z)
• 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)

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.