VIB-Probe: Detecting and Mitigating Hallucinations in Vision-Language Models via Variational Information Bottleneck

TL;DR

VIB-Probe introduces a novel framework using Variational Information Bottleneck to detect and mitigate hallucinations in vision-language models by analyzing internal attention mechanisms.

Contribution

It proposes a new method leveraging the VIB theory to identify and intervene on attention heads responsible for hallucinations in VLMs.

Findings

VIB-Probe outperforms existing baselines in hallucination detection.

The method effectively identifies attention heads causally linked to hallucinations.

Intervention strategies reduce hallucinations during inference.

Abstract

Vision-Language Models (VLMs) have demonstrated remarkable progress in multimodal tasks, but remain susceptible to hallucinations, where generated text deviates from the underlying visual content. Existing hallucination detection methods primarily rely on output logits or external verification tools, often overlooking their internal mechanisms. In this work, we investigate the outputs of internal attention heads, postulating that specific heads carry the primary signals for truthful generation.However, directly probing these high-dimensional states is challenging due to the entanglement of visual-linguistic syntax and noise. To address this, we propose VIB-Probe, a novel hallucination detection and mitigation framework leveraging the Variational Information Bottleneck (VIB) theory. Our method extracts discriminative patterns across layers and heads while filtering out semantic nuisances through the information bottleneck principle. Furthermore, by leveraging the gradients of our VIB probe, we identify attention heads with strong causal influence on hallucinations and introduce an inference-time intervention strategy for hallucination mitigation. Extensive experiments across diverse benchmarks demonstrate that VIB-Probe significantly outperforms existing baselines in both settings. Our code will be made publicly available.