Mitigating Hallucination in Multimodal Large Language Model via Hallucination-targeted Direct Preference Optimization

TL;DR

This paper introduces Hallucination-targeted Direct Preference Optimization (HDPO), a novel method to reduce hallucinations in Multimodal Large Language Models by addressing their diverse causes, leading to improved performance over existing approaches.

Contribution

The paper proposes HDPO, a new approach that specifically targets different causes of hallucinations in MLLMs, outperforming previous methods in hallucination mitigation.

Findings

HDPO surpasses most SOTA methods in hallucination reduction.

The method effectively addresses hallucinations from visual, contextual, and multimodal conflicts.

Ablation studies confirm the effectiveness and potential for further scaling.

Abstract

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.