Problem
Research questions and friction points this paper is trying to address.
Multimodal Large Language Models
hallucinations
information distribution drift
MLLMs Hallucinate when Information Distribution Drifts in Synergy Heads
🤖 AI Summary
论文提出HEAL方法，通过干预多头输出和信息校准减少MLLMs的幻觉问题，提高模型可信度。
📝 Abstract
Multimodal Large Language Models (MLLMs) often struggle with hallucinations, thus hindering their reliable practical applications. Existing attention-based mitigation methods mainly rely on indirect signals (e.g., attention weights) that fail to accurately reflect the actual information shift underlying hallucination generation. In this paper, we propose HEAL, Head-lEvel information disentAnglement and caLibration for identifying and mitigating hallucinations. HEAL first employs causal noise intervention on multi-head outputs to filter out causally redundant heads. Subsequently, it disentangles information distribution within the remaining heads via the counterfactual Difference-in-Differences, categorizing heads into four types. Through analysis, we observe: hallucinations happen when information distribution drifts away from a healthy equilibrium in synergy heads, not strongly correlated with the quantity or strength of modality-specific heads. Motivated by this insight, HEAL injects dynamic information calibration factors into the value vectors of synergy heads, and actively regulates visual-language dependencies, steering the output distribution towards factual evidence. Extensive experiments demonstrate that HEAL effectively reduces hallucinations across multiple MLLMs, offering a simple and interpretable pathway to enhance model trustworthiness.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Head-lEvel information disentAnglement and caLibration
causal noise intervention
counterfactual Difference-in-Differences
dynamic information calibration factors
🔎 Similar Papers
No similar papers found.
