SAGE: Spuriousness-Aware Guided Prompt Exploration for Mitigating Multimodal Bias
Wenqian Ye, Di Wang, Guangtao Zheng, 刘博涵 (Bohan Liu), Aidong Zhang
摘要
此摘要由英文原文自动翻译。
如CLIP等大型视觉-语言模型,通过在共享嵌入空间中对齐图像与文本,展现出强大的零样本分类性能。然而,CLIP模型常发展出多模态虚假偏差——依赖虚假特征的不良倾向。例如,CLIP可能依据频繁共现的背景而非目标核心特征来推断图像中的目标类型。该偏差显著损害预训练CLIP模型在分布外数据上的鲁棒性——在该情形下这些跨模态关联不再成立。现有的多模态虚假偏差缓解方法通常需要对下游数据微调或预先知晓偏差,削弱了CLIP开箱即用的可用性。本文首先对零样本分类中多模态虚假偏差的影响进行理论分析。基于这一洞见,我们提出Spuriousness-Aware Guided Exploration(SAGE)——一个简单有效的方法,通过引导式提示选择缓解虚假偏差。SAGE无需训练、微调或外部标注。它探索提示模板空间,选择能在类别间诱导最大语义分离的提示,从而提升最差组鲁棒性。在四个真实世界基准数据集与五个主流骨干模型上的广泛实验证明,SAGE持续提升零样本性能与泛化,在无外部知识或模型更新的情况下超越既有的零样本方法。
原始摘要(英文)
Large vision-language models, such as CLIP, have shown strong zero-shot classification performance by aligning images and text in a shared embedding space. However, CLIP models often develop multimodal spurious biases, which is the undesirable tendency to rely on spurious features. For example, CLIP may infer object types in images based on frequently co-occurring backgrounds rather than the object's core features. This bias significantly impairs the robustness of pre-trained CLIP models on out-of-distribution data, where such cross-modal associations no longer hold. Existing methods for mitigating multimodal spurious bias typically require fine-tuning on downstream data or prior knowledge of the bias, which undermines the out-of-the-box usability of CLIP. In this paper, we first theoretically analyze the impact of multimodal spurious bias in zero-shot classification. Based on this insight, we propose Spuriousness-Aware Guided Exploration (SAGE), a simple and effective method that mitigates spurious bias through guided prompt selection. SAGE requires no training, fine-tuning, or external annotations. It explores a space of prompt templates and selects the prompts that induce the largest semantic separation between classes, thereby improving worst-group robustness. Extensive experiments on four real-world benchmark datasets and five popular backbone models demonstrate that SAGE consistently improves zero-shot performance and generalization, outperforming previous zero-shot approaches without any external knowledge or model updates.
本文贡献
提示词层面的多模态偏差防御:SAGE估计提示词的“虚假意识”程度,并探索引导式提示词变体,找出模型依赖真实视觉内容而非数据集捷径的表述方式。
BibTeX
@inproceedings{ye2026sage,
title = {SAGE: Spuriousness-Aware Guided Prompt Exploration for Mitigating Multimodal Bias},
author = {Wenqian Ye and Di Wang and Guangtao Zheng and Bohan Liu and Aidong Zhang},
booktitle = {AAAI Conference on Artificial Intelligence},
year = {2026},
url = {https://arxiv.org/abs/2511.13005}
}