拆除病理性捷径：忠实LVLM解码的因果框架
Dismantling Pathological Shortcuts: A Causal Framework for Faithful LVLM Decoding
AI总结：
提出Fox框架，通过视觉注意力熵探针定位风险中介头，利用数值logit饱和进行因果干预，并结合冲突门控协同解码，消除视觉语言模型中的对象幻觉，实现忠实解码。
摘要：
大型视觉语言模型（LVLM）展现出复杂的推理能力，但仍易受对象幻觉影响。与流行的注意力强度假设不同，我们揭示了一种更深层的动态结构错位：在决策关键步骤中，特定的注意力头作为风险中介，从视觉证据解耦并锁定语言先验，从而建立绕过视觉基础的病理性捷径。为拆除这一捷径，我们提出Fox（通过表达修正实现忠实性与观测流），一种无需训练、推理时干预的框架。Fox利用视觉注意力熵探针无监督地定位风险中介，然后通过数值logit饱和执行有针对性的因果干预，物理上切断捷径路径。最后，冲突门控协同解码策略协调干预的忠实性与观测的流畅性。大量实验表明，Fox达到最先进性能，在保持语言丰富性的同时，比SID高出29.1%。代码见：此https URL。
英文摘要：
Large Vision-Language Models (LVLMs) exhibit sophisticated reasoning but remain susceptible to object hallucination. Deviating from the prevailing attention intensity assumption, we reveal a deeper dynamic structural misalignment: hallucination is triggered at decision-critical steps where specific attention heads, acting as risky mediators, decouple from visual evidence to lock onto language priors. This establishes a pathological shortcut that bypasses visual grounding. To dismantle this, we propose Fox (Faithfulness and Observational-flow via eXpression-rectification), a training-free inference-time framework. Fox diagnoses structural misalignment using a visual attention entropy probe to localize risky mediators unsupervisedly. We then execute a targeted causal intervention via numerical logit saturation to physically sever the shortcut path. Finally, a conflict-gated cooperative decoding strategy reconciles interventional faithfulness with observational fluency. Extensive experiments demonstrate that Fox achieves SOTA performance, outperforming SID by 29.1% while preserving linguistic richness. Code is available at https://github.com/Cc2021start/Fox.
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。
