Vocabulary Hijacking in LVLMs: Unveiling Critical Attention Heads by Excluding Inert Tokens to Mitigate Hallucination

• URL: http://arxiv.org/abs/2605.10622v1
• Date: Mon, 11 May 2026 14:16:25 GMT
• Title: Vocabulary Hijacking in LVLMs: Unveiling Critical Attention Heads by Excluding Inert Tokens to Mitigate Hallucination
• Abstract summary: We analyze the attention mechanisms via the logit lens, uncovering a distinct anomaly we term Vocabulary Hijacking.<n>We propose Hijacking-Aware Visual Attention Enhancement (HAVAE), a training-free intervention that selectively strengthens the focus of these identified heads on salient visual content.
• Score: 25.643533134113607
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large Vision-Language Models (LVLMs) have achieved remarkable progress in multimodal tasks, yet their reliability is persistently undermined by hallucinations-generating text that contradicts visual input. Recent studies often attribute these errors to inadequate visual attention. In this work, we analyze the attention mechanisms via the logit lens, uncovering a distinct anomaly we term Vocabulary Hijacking. We discover that specific visual tokens, defined as Inert Tokens, disproportionately attract attention. Crucially, when their intermediate hidden states are projected into the vocabulary space, they consistently decode to a fixed set of unrelated words (termed Hijacking Anchors) across layers, revealing a rigid semantic collapse. Leveraging this semantic rigidity, we propose Hijacking Anchor-Based Identification (HABI), a robust strategy to accurately localize these Inert Tokens. To quantify the impact of this phenomenon, we introduce the Non-Hijacked Visual Attention Ratio (NHAR), a novel metric designed to identify attention heads that remain resilient to hijacking and are critical for factual accuracy. Building on these insights, we propose Hijacking-Aware Visual Attention Enhancement (HAVAE), a training-free intervention that selectively strengthens the focus of these identified heads on salient visual content. Extensive experiments across multiple benchmarks demonstrate that HAVAE significantly mitigates hallucinations with no additional computational overhead, while preserving the model's general capabilities. Our code is publicly available at https://github.com/lab-klc/HAVAE.

Related papers

• Disentangling Semantic Attention from Structural Bias in the Attention Manifold [61.68044165974505]
  Multimodal Large Language Models (MLLMs) exhibit disproportionate attention toward certain semantically uninformative visual tokens.<n>We introduce Saliency-guided Purification and Adaptive Redistribution (SPAR), a training-free, plug-and-play intervention.
  arXiv  Detail & Related papers  (2026-07-27T05:30:19Z)
• LASER: A Corrective Lens for LVLMs via Visual Attention Preservation and Sink Suppression [50.454633334898396]
  Large vision-language models (LVLMs) exhibit strong reasoning ability but suffer from visual forgetting during long-horizon decoding.<n>We propose LASER, a post-training framework that regulates both the visual attention trajectory and intra-visual token attention distribution during reasoning.<n>Experiments on eight benchmark datasets demonstrate that LASER consistently outperforms strong baselines.
  arXiv  Detail & Related papers  (2026-07-02T04:59:56Z)
• See Only When Needed: Context-Aware Attention Intervention for Mitigating Hallucinations in LVLMs [81.71384150441163]
  We propose Context-aware Attention Intervention (CAI) for large vision-language models (LVLMs)<n>CAI enforces a see only when needed principle via two-axis selectivity: where to look and when to intervene.<n> experiments across multiple LVLM backbones and benchmarks show that CAI achieves state-of-the-art hallucination mitigation.
  arXiv  Detail & Related papers  (2026-06-29T06:35:47Z)
• Focus Matters: Phase-Aware Suppression for Hallucination in Vision-Language Models [8.304027910542446]
  We investigate the internal attention dynamics of vision encoders in Large Vision-Language Models (LVLMs)<n>Our analysis reveals that hallucination behavior is particularly sensitive to tokens receiving low attention during the focus phase.<n>We propose a lightweight inference-time intervention that selectively suppresses such tokens during the focus phase.
  arXiv  Detail & Related papers  (2026-04-04T02:46:58Z)
• SAGE: Sink-Aware Grounded Decoding for Multimodal Hallucination Mitigation [33.381194425912234]
  Large vision-language models (VLMs) frequently suffer from hallucinations, generating content that is inconsistent with visual inputs.<n>We introduce SAGE, a Sink-Aware Grounded Decoding framework that mitigates hallucinations by dynamically modulating self-attention during generation.<n>Our method achieves an average relative improvement of 10.65% on MSCOCO and 7.19% on AMBER across diverse VLM architectures.
  arXiv  Detail & Related papers  (2026-03-29T22:52:03Z)
• Attention to details, logits to truth: visual-aware attention and logits enhancement to mitigate hallucinations in LVLMs [12.578567672069601]
  We propose a training free attentional intervention algorithm to enhance the attention of task-relevant tokens.<n>To enhance the contribution of visual tokens, we inject visual attention values into the beam search decoding to identify solutions with higher visual attention.
  arXiv  Detail & Related papers  (2026-02-10T08:26:50Z)
• CAI: Caption-Sensitive Attention Intervention for Mitigating Object Hallucination in Large Vision-Language Models [60.0300765815417]
  Large Vision-Language Models (LVLMs) frequently produce content that deviates from visual information, leading to object hallucination.<n>We propose Caption-sensitive Attention Intervention (CAI), a training-free, plug-and-play hallucination mitigation method.
  arXiv  Detail & Related papers  (2025-06-30T07:52:36Z)
• Uncertainty-Aware Attention Heads: Efficient Unsupervised Uncertainty Quantification for LLMs [129.79394562739705]
  Large language models (LLMs) exhibit impressive fluency, but often produce critical errors known as "hallucinations"<n>We propose RAUQ (Recurrent Attention-based Uncertainty Quantification), an unsupervised approach that leverages intrinsic attention patterns in transformers to detect hallucinations efficiently.<n> Experiments across 4 LLMs and 12 question answering, summarization, and translation tasks demonstrate that RAUQ yields excellent results.
  arXiv  Detail & Related papers  (2025-05-26T14:28:37Z)
• Attention Reallocation: Towards Zero-cost and Controllable Hallucination Mitigation of MLLMs [62.9348974370985]
  We propose attention reallocation (AttnReal) to mitigate hallucinations with nearly zero extra cost.<n>Our approach is motivated by the key observations that, MLLM's unreasonable attention distribution causes features to be dominated by historical output tokens.<n>Based on the observations, AttnReal recycles excessive attention from output tokens and reallocates it to visual tokens, which reduces MLLM's reliance on language priors.
  arXiv  Detail & Related papers  (2025-03-11T11:52:37Z)
• Attention Hijackers: Detect and Disentangle Attention Hijacking in LVLMs for Hallucination Mitigation [123.54980913741828]
  Large Vision-Language Models (LVLMs) remain vulnerable to hallucinations.<n>We propose a novel, training-free strategy namely Attention HIjackers Detection and Disentanglement (AID)<n>AID identifies Attention Hijackers by calculating instruction-driven visual salience.<n>Next, Attention Disentanglement mechanism is proposed to mask the visual attention of these identified Hijackers.<n>Re-Disentanglement recalculates the balance between instruction-driven and image-driven visual salience to avoid over-masking effects.
  arXiv  Detail & Related papers  (2025-03-11T09:35:55Z)
• PAINT: Paying Attention to INformed Tokens to Mitigate Hallucination in Large Vision-Language Model [0.0]
  hallucinations often arise from the progressive weakening of attention weight to visual tokens.<n>textbfPAINT (textbfPaying textbfAttention to textbfINformed textbfTokens) is a plug-and-play framework that intervenes in the self-attention mechanism of the Large Vision Language Models.
  arXiv  Detail & Related papers  (2025-01-21T15:22:31Z)
• From Uncertainty to Trust: Enhancing Reliability in Vision-Language Models with Uncertainty-Guided Dropout Decoding [6.115580421973011]
  Large vision-language models (LVLMs) demonstrate remarkable capabilities in multimodal tasks but are prone to misinterpreting visual inputs, often resulting in hallucinations and unreliable outputs.<n>We propose Dropout Decoding, a novel inference-time approach that quantifies the uncertainty of visual tokens and selectively masks uncertain tokens to improve decoding.<n> Evaluations on benchmarks including CHAIR, THRONE, and MMBench demonstrate that Dropout Decoding significantly reduces object hallucinations (OH) and enhances both reliability and quality of LVLM outputs.
  arXiv  Detail & Related papers  (2024-12-09T13:21:07Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.