Style-Debiased DPO: Updating LLM Knowledge with Factuality-Aware Synthetic Preference Data

• URL: http://arxiv.org/abs/2609.16532v1
• Date: Tue, 15 Sep 2026 02:26:19 GMT
• Title: Style-Debiased DPO: Updating LLM Knowledge with Factuality-Aware Synthetic Preference Data
• Abstract summary: We study the eliciting side rather than the storing side.<n>We use preference optimization, which learns from pairs of a preferred (chosen) and a dispreferred (rejected) response.<n>We propose style-debiased DPO (SD-DPO), which scores whether the rejected response of each pair is factually correct, inverts the preference of such pairs, and weights them so that the learning signal due to differences in style cancels out as a whole.
• Score: 11.157564521901634
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Continued pretraining (CPT) with data augmentation such as paraphrasing can store inside a large language model (LLM) the knowledge of a small source corpus. The stored knowledge, however, is not always retrieved correctly. We study the eliciting side rather than the storing side: we use preference optimization, which learns from pairs of a preferred (chosen) and a dispreferred (rejected) response, so that the model elicits its stored knowledge more accurately. One proposed approach takes the model's own erroneous response as rejected and the gold answer as chosen, so as to suppress the error. When the target knowledge is partially known, however, most of these rejected responses are factually correct. Using direct preference optimization (DPO) then pushes down rejected responses that contain correct knowledge and differ from the chosen answer only in style, such as length and wording. We propose style-debiased DPO (SD-DPO), which scores whether the rejected response of each pair is factually correct, inverts the preference of such pairs, and weights them so that the learning signal due to differences in style cancels out as a whole. We first test whether, on top of EntiGraph, a representative storing-side method that runs CPT on text synthesized from the corpus, our method adds accuracy efficiently. On QuALITY, the reading-comprehension QA benchmark on which EntiGraph was evaluated, SD-DPO exceeds a baseline we CPT on EntiGraph's synthetic data from the same base model and evaluate with the same procedure. The training tokens this requires are a few dozen times fewer than the additional CPT needed for the same gain. For knowledge updating, the main goal of this work, we use AToKE, a knowledge-editing benchmark for facts that change over time. There, SD-DPO reaches an overall accuracy of 0.982 and answers with the new or the old fact according to the queried period.

Related papers

• SP^2DPO: An LLM-assisted Semantic Per-Pair DPO Generalization [38.55012111778011]
  We introduce SP2DPO (Semantic Per-Pair DPO), a generalization that replaces the global temperature with an instance-specific schedule beta_i.<n>We focus our empirical study on AlpacaEval 2.0, reporting both raw win rate and length-controlled win rate.
  arXiv  Detail & Related papers  (2026-01-29T22:46:11Z)
• Reinforcement Learning via Self-Distillation [37.078107691613155]
  Large language models are increasingly post-trained with reinforcement learning in verifiable domains such as code and math.<n>Current methods for reinforcement learning with verifiable rewards (RLVR) learn only from a scalar outcome reward per attempt, creating a severe credit-assignment bottleneck.<n>We formalize this setting as reinforcement learning with rich feedback and introduce Self-Distillation Policy Optimization (SDPO)<n>SDPO converts tokenized feedback into a dense learning signal without any external teacher or explicit reward model.
  arXiv  Detail & Related papers  (2026-01-28T17:45:12Z)
• Sharpe Ratio-Guided Active Learning for Preference Optimization in RLHF [67.48004037550064]
  We propose an active learning approach to efficiently select prompt and preference pairs.<n>Our method evaluates the gradients of all potential preference annotations to assess their impact on model updates.<n> Experimental results demonstrate that our method outperforms the baseline by up to 5% in win rates against the chosen completion.
  arXiv  Detail & Related papers  (2025-03-28T04:22:53Z)
• 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)
• Knowledge Editing in Language Models via Adapted Direct Preference Optimization [50.616875565173274]
  Large Language Models (LLMs) can become outdated over time. Knowledge Editing aims to overcome this challenge using weight updates that do not require expensive retraining.
  arXiv  Detail & Related papers  (2024-06-14T11:02:21Z)
• Preference Learning Algorithms Do Not Learn Preference Rankings [62.335733662381884]
  We study the conventional wisdom that preference learning trains models to assign higher likelihoods to more preferred outputs than less preferred outputs. We find that most state-of-the-art preference-tuned models achieve a ranking accuracy of less than 60% on common preference datasets.
  arXiv  Detail & Related papers  (2024-05-29T21:29:44Z)
• Ethicist: Targeted Training Data Extraction Through Loss Smoothed Soft Prompting and Calibrated Confidence Estimation [56.57532238195446]
  We propose a method named Ethicist for targeted training data extraction. To elicit memorization, we tune soft prompt embeddings while keeping the model fixed. We show that Ethicist significantly improves the extraction performance on a recently proposed public benchmark.
  arXiv  Detail & Related papers  (2023-07-10T08:03:41Z)
• Source-Free Domain Adaptation for SSVEP-based Brain-Computer Interfaces [0.45880283710344055]
  SSVEP-based BCI spellers assist individuals experiencing speech difficulties by enabling them to communicate at a fast rate.<n> achieving a high information transfer rate (ITR) in most prominent methods requires an extensive calibration period before using the system.<n>We address this issue by proposing a novel method that adapts a powerful deep neural network (DNN) pre-trained on data from source domains.
  arXiv  Detail & Related papers  (2023-05-27T08:02:46Z)
• AdaNPC: Exploring Non-Parametric Classifier for Test-Time Adaptation [64.9230895853942]
  Domain generalization can be arbitrarily hard without exploiting target domain information. Test-time adaptive (TTA) methods are proposed to address this issue. In this work, we adopt Non-Parametric to perform the test-time Adaptation (AdaNPC)
  arXiv  Detail & Related papers  (2023-04-25T04:23:13Z)
• Pre-training Is (Almost) All You Need: An Application to Commonsense Reasoning [61.32992639292889]
  Fine-tuning of pre-trained transformer models has become the standard approach for solving common NLP tasks. We introduce a new scoring method that casts a plausibility ranking task in a full-text format. We show that our method provides a much more stable training phase across random restarts.
  arXiv  Detail & Related papers  (2020-04-29T10:54:40Z)

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.