Gradient-Gated DPO: Stabilizing Preference Optimization in Language Models

• URL: http://arxiv.org/abs/2605.02626v1
• Date: Mon, 04 May 2026 14:15:24 GMT
• Title: Gradient-Gated DPO: Stabilizing Preference Optimization in Language Models
• Abstract summary: We introduce Gradient-Gated Preference Optimization (Gate-DPO)<n>Gate-DPO stabilizes training by modulating rejected gradients according to the model's probability geometry.<n>Experiments across multiple architectures and preference datasets show that Gate-DPO consistently reduces squeezing and improves chosen-response likelihood.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Preference optimization has become a central paradigm for aligning large language models with human feedback. Direct Preference Optimization (DPO) simplifies reinforcement learning from human feedback by directly optimizing pairwise preferences, removing the need for reward modeling and policy optimization. However, recent work shows that DPO exhibits a squeezing effect, where negative gradients applied to rejected responses concentrate probability mass on high-confidence predictions while suppressing alternative responses. This phenomenon arises even in simple softmax models and can lead to systematic probability collapse during training. We introduce Gradient-Gated Preference Optimization (Gate-DPO), a method that stabilizes training by modulating rejected gradients according to the model's probability geometry. When updates target extremely low-probability responses, the gate attenuates harmful gradients while preserving standard optimization behavior. Gate-DPO addresses this optimization pathology without modifying the underlying preference objective and is complementary to existing methods such as extended SFT, IPO, and Cal-DPO. Experiments across multiple architectures and preference datasets show that Gate-DPO consistently reduces squeezing and improves chosen-response likelihood. Mass-dynamics analysis further reveals healthier optimization behavior, with improved preferred responses and reduced suppression of the overall distribution. Notably, smaller gated models can exhibit stronger chosen-response improvements than larger ungated models, suggesting that controlling gradient dynamics, rather than scale alone, is key to stable and efficient alignment.

Related papers

• TPMM-DPO: Trajectory-aware Preference-guided Model Merging for Iterative Direct Preference Optimization [0.0]
  We propose TPMM-DPO, a trajectory-aware preference-guided model merging method.<n>In contrast to conventional iterative DPO, TPMM-DPO effectively mitigates error accumulation induced by noisy preferences.
  arXiv  Detail & Related papers  (2026-05-22T09:11:20Z)
• Model Agnostic Preference Optimization for Medical Image Segmentation [5.289507655906182]
  Preference optimization offers a scalable supervision paradigm based on relative preference signals.<n>We propose MAPO (Model-A Preference Optimization), a training framework that utilizes Dropout-driven segmentation hypotheses.<n> MAPO is fully dimensionality-agnostic, supporting 2D/3D CNN and Transformer-based segmentation pipelines.
  arXiv  Detail & Related papers  (2025-12-17T01:50:52Z)
• From Noisy Traces to Stable Gradients: Bias-Variance Optimized Preference Optimization for Aligning Large Reasoning Models [90.45197506653341]
  Large reasoning models generate intermediate reasoning traces before producing final answers.<n> aligning LRMs with human preferences, a crucial prerequisite for model deployment, remains underexplored.<n>A common workaround optimized a single sampled trajectory, which introduces substantial gradient variance from trace sampling.
  arXiv  Detail & Related papers  (2025-10-06T17:58:01Z)
• TCPO: Thought-Centric Preference Optimization for Effective Embodied Decision-making [75.29820290660065]
  This paper proposes Thought-Centric Preference Optimization ( TCPO) for effective embodied decision-making.<n>It emphasizes the alignment of the model's intermediate reasoning process, mitigating the problem of model degradation.<n>Experiments in the ALFWorld environment demonstrate an average success rate of 26.67%, achieving a 6% improvement over RL4VLM.
  arXiv  Detail & Related papers  (2025-09-10T11:16:21Z)
• Stable Preference Optimization for LLMs: A Bilevel Approach Beyond Direct Preference Optimization [2.384797824772941]
  We present a comprehensive analysis of DPO's dynamics from a probability evolution perspective.<n>We propose a theoretically grounded bilevel optimization framework that tightly integrate supervised fine-tuning with an enhanced DPO objective a.k.a. stable preference optimization.
  arXiv  Detail & Related papers  (2025-07-10T12:57:39Z)
• Divergence Minimization Preference Optimization for Diffusion Model Alignment [66.31417479052774]
  Divergence Minimization Preference Optimization (DMPO) is a principled method for aligning diffusion models by minimizing reverse KL divergence.<n>DMPO can consistently outperform or match existing techniques across different base models and test sets.
  arXiv  Detail & Related papers  (2025-07-10T07:57:30Z)
• Refining Alignment Framework for Diffusion Models with Intermediate-Step Preference Ranking [50.325021634589596]
  We propose a Tailored Optimization Preference (TailorPO) framework for aligning diffusion models with human preference.<n>Our approach directly ranks intermediate noisy samples based on their step-wise reward, and effectively resolves the gradient direction issues.<n> Experimental results demonstrate that our method significantly improves the model's ability to generate aesthetically pleasing and human-preferred images.
  arXiv  Detail & Related papers  (2025-02-01T16:08:43Z)
• 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)
• 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)
• 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.