通过针对幻觉的直接偏好优化缓解多模态大语言模型中的幻觉
Mitigating Hallucination in Multimodal Large Language Model via Hallucination-targeted Direct Preference Optimization
AI总结：
针对多模态大语言模型的幻觉问题，提出HDPO方法，通过构建针对视觉能力不足、长上下文生成和多模态冲突三类成因的偏好对数据进行偏好优化，有效缓解幻觉并在多个评估数据集上超越多数SOTA方法。
摘要：
多模态大语言模型（MLLMs）存在幻觉问题，这限制了其实际应用。近期研究尝试应用直接偏好优化（DPO）来提升MLLMs性能，但在缓解幻觉方面效果不稳定。为更有效地解决此问题，我们提出了针对幻觉的直接偏好优化（HDPO）以减少MLLMs中的幻觉。与以往方法不同，我们的方法从幻觉的多种形式和成因入手。具体而言，我们针对以下MLLM幻觉成因开发了三类偏好对数据：（1）视觉能力不足，（2）长上下文生成，以及（3）多模态冲突。实验结果表明，我们的方法在多个幻觉评估数据集上取得卓越性能，超越大多数最先进（SOTA）方法，凸显了本方法的潜力。消融实验与深入分析进一步证实了方法的有效性，并表明通过扩大规模具有进一步改进的潜力。
英文摘要：
Multimodal Large Language Models (MLLMs) are known to hallucinate, which limits their practical applications. Recent works have attempted to apply Direct Preference Optimization (DPO) to enhance the performance of MLLMs, but have shown inconsistent improvements in mitigating hallucinations. To address this issue more effectively, we introduce Hallucination-targeted Direct Preference Optimization (HDPO) to reduce hallucinations in MLLMs. Unlike previous approaches, our method tackles hallucinations from their diverse forms and causes. Specifically, we develop three types of preference pair data targeting the following causes of MLLM hallucinations: (1) insufficient visual capabilities, (2) long context generation, and (3) multimodal conflicts. Experimental results demonstrate that our method achieves superior performance across multiple hallucination evaluation datasets, surpassing most state-of-the-art (SOTA) methods and highlighting the potential of our approach. Ablation studies and in-depth analyses further confirm the effectiveness of our method and suggest the potential for further improvements through scaling up.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
