Aligning with Your Own Voice: Self-Corrected Preference Learning for Hallucination Mitigation in LVLMs

2026-04-27

AI Summary

This work addresses the hallucination problem in large vision-language models (LVLMs), which often arises from a mismatch between training preference data and the model’s internal data distribution. To tackle this, the authors propose AVES-DPO, a novel framework that, for the first time, constructs in-distribution preference data using only the target model itself. By employing a consensus verification mechanism to detect hallucinations and guide self-correction, the method generates high-quality preference pairs for direct preference optimization (DPO). Notably, AVES-DPO requires no external or proprietary models and achieves significant performance gains over existing baselines with merely 5.2k samples, demonstrating superior effectiveness in mitigating hallucinations across multiple evaluation metrics.

Technology Category

• Computer Vision: Large Vision Models
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Large Vision-Language Models (LVLMs) frequently suffer from hallucinations. Existing preference learning-based approaches largely rely on proprietary models to construct preference datasets. We identify that this reliance introduces a distributional mismatch between the proprietary and target models that hinders efficient alignment. To address this, we propose Alignment via VErified Self-correction DPO (AVES-DPO), a framework that aligns LVLMs using in-distribution data derived from the model's intrinsic knowledge. Our approach employs a consensus-based verification mechanism to diagnose diverse hallucinations and guides the model to self-correct, thereby generating preference pairs strictly compatible with its internal distribution. Extensive experiments demonstrate that AVES-DPO surpasses existing baselines in hallucination mitigation while requiring only 5.2k samples.

Problem

• hallucination
• preference learning
• distributional mismatch
• Large Vision-Language Models
• alignment

Innovation

• self-corrected preference learning
• hallucination mitigation
• distribution alignment
• verified self-correction
• LVLMs