arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2603.16284cs.CVcs.LG
收藏

定位后再稀疏化:基于归因的视觉幻觉缓解稀疏策略

Locate-then-Sparsify: Attribution Guided Sparse Strategy for Visual Hallucination Mitigation

State Key Lab. of AI Safety, Institute of Computing Technology, Chinese Academy of Sciences(中国科学院人工智能安全国家重点实验室,计算技术研究所);School of Advanced Interdisciplinary Sciences, University of Chinese Academy of Sciences(中国科学院大学先进交叉科学学院);School of Computer Science and Technology, University of Chinese Academy of Sciences(中国科学院大学计算机科学与技术学院)

Tiantian Dang, Chao Bi, Shufan Shen, Jinzhe Liu, Qingming Huang, Shuhui Wang

AI总结:

本文提出了一种名为Locate-Then-Sparsify for Feature Steering (LTS-FS)的框架,通过定位和稀疏化策略,根据每层与幻觉的相关性调整特征引导强度,从而有效缓解视觉语言模型中的幻觉问题,同时保持良好的性能。

摘要:

尽管大型视觉-语言模型(LVLMs)在技术上取得了显著进展,但其生成幻觉的倾向削弱了可靠性并限制了更广泛的实际应用。在幻觉缓解方法中,特征引导作为一种有前景的方法,能够在不增加推理成本的情况下减少LVLMs中的错误输出。然而,当前的方法在所有层上应用统一的特征引导策略。这种启发式策略忽略了层间的差异,可能会干扰与幻觉无关的层,最终导致在通用任务上的性能下降。在本文中,我们提出了一种名为Locate-Then-Sparsify for Feature Steering (LTS-FS)的即插即用框架,该框架根据每层与幻觉的相关性来控制引导强度。我们首先构建了一个包含token级和句子级幻觉案例的数据集。基于此数据集,我们引入了一种基于因果干预的归因方法,以量化每层的幻觉相关性。利用各层的归因分数,我们提出了一种逐层策略,将这些分数转换为针对单个层的特征引导强度,从而在幻觉相关的层上实现更精确的调整。在多个LVLMs和基准测试中进行的广泛实验表明,LTS-FS有效缓解了幻觉问题,同时保持了强大的性能。代码可在https://github.com/huttersadan/LTS-FS上获得。

英文摘要:

Despite the significant advancements in Large Vision-Language Models (LVLMs), their tendency to generate hallucinations undermines reliability and restricts broader practical deployment. Among the hallucination mitigation methods, feature steering emerges as a promising approach that reduces erroneous outputs in LVLMs without increasing inference costs. However, current methods apply uniform feature steering across all layers. This heuristic strategy ignores inter-layer differences, potentially disrupting layers unrelated to hallucinations and ultimately leading to performance degradation on general tasks. In this paper, we propose Locate-Then-Sparsify for Feature Steering (LTS-FS), a plug-and-play framework which controls the steering intensity according to the hallucination relevance of each layer. We first construct a dataset comprising token-level and sentence-level hallucination cases. Based on this dataset, we introduce an attribution method based on causal interventions to quantify the hallucination relevance of each layer. With the attribution scores across layers, we propose a layerwise strategy that converts these scores into feature steering intensities for individual layers, enabling more precise adjustments specifically on hallucination-relevant layers. Extensive experiments across multiple LVLMs and benchmarks demonstrate that LTS-FS effectively mitigates hallucination while preserving strong performance. Codes are available at https://github.com/huttersadan/LTS-FS.

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。

↑