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 2411.02708cs.LGcs.AIcs.CL
收藏

多模态大语言模型中的响应不确定性探究:误导场景下的实证评估

Exploring Response Uncertainty in MLLMs: An Empirical Evaluation under Misleading Scenarios

The Hong Kong University of Science and Technology (Guangzhou)(香港科技大学(广州));The Hong Kong University of Science and Technology(香港科技大学);Tsinghua University(清华大学);Ant Group(蚂蚁集团);Alibaba Group(阿里巴巴集团)

Yunkai Dang, Mengxi Gao, Yibo Yan, Xin Zou, Yanggan Gu, Jungang Li, Jingyu Wang, Peijie Jiang, Aiwei Liu, Jia Liu, Xuming Hu

AI总结:

本研究揭示MLLMs存在响应不确定性现象,提出两阶段评估流程量化其受误导的脆弱性,构建了多模态不确定性基准MUB,并通过混合指令微调有效降低了模型的误导率。

摘要:

多模态大语言模型(MLLMs)近期在视觉问答、视频理解等各类任务上取得了当前最优性能。然而现有研究主要聚焦于视觉-文本错位问题,MLLMs在面对误导信息时维持原有正确答案的能力仍未得到充分探究。我们揭示了一种响应不确定性现象:在9个标准数据集、12款当前最优的开源MLLMs中,65%的情况下模型在收到单个欺骗性提示后会推翻此前的正确答案。为系统量化这一脆弱性,我们提出两阶段评估流程:(1)获取每个模型在未受干扰输入上的原始响应;(2)注入显性(错误答案提示)和隐性(上下文矛盾)误导指令,计算误导率——即答案由对转错的比例。基于最易受误导的样本,我们构建了多模态不确定性基准(Multimodal Uncertainty Benchmark, MUB),该基准包含图像-问题对,根据其能误导12款当前最优MLLMs的数量分为低、中、高三个难度等级。对12款开源模型和5款闭源模型的广泛评估显示出高度不确定性:平均误导率超过86%,其中显性提示下超过67.19%,隐性提示下超过80.67%。为降低误导率,我们随后用一个含2000个样本的紧凑混合指令数据集对所有开源MLLMs进行微调,将误导率降至6.97%(显性)和32.77%(隐性),在高欺骗性输入上的一致性提升近29.37%,同时在标准基准上的准确率略有提升。我们的代码可在https://github.com/Yunkaidang/uncertainty获取。

英文摘要:

Multimodal large language models (MLLMs) have recently achieved state-of-the-art performance on tasks ranging from visual question answering to video understanding. However, existing studies have concentrated mainly on visual-textual misalignment, leaving largely unexplored the MLLMs' ability to preserve an originally correct answer when confronted with misleading information. We reveal a response uncertainty phenomenon: across nine standard datasets, twelve state-of-the-art open-source MLLMs overturn a previously correct answer in 65% of cases after receiving a single deceptive cue. To systematically quantify this vulnerability, we propose a two-stage evaluation pipeline: (1) elicit each model's original response on unperturbed inputs; (2) inject explicit (false-answer hints) and implicit (contextual contradictions) misleading instructions, and compute the misleading rate - the fraction of correct-to-incorrect flips. Leveraging the most susceptible examples, we curate the Multimodal Uncertainty Benchmark (MUB), a collection of image-question pairs stratified into low, medium, and high difficulty based on how many of twelve state-of-the-art MLLMs they mislead. Extensive evaluation on twelve open-source and five closed-source models reveals a high uncertainty: average misleading rates exceed 86%, with explicit cues over 67.19% and implicit cues over 80.67%. To reduce the misleading rate, we then fine-tune all open-source MLLMs on a compact 2000-sample mixed-instruction dataset, reducing misleading rates to 6.97% (explicit) and 32.77% (implicit), boosting consistency by nearly 29.37% on highly deceptive inputs, and slightly improving accuracy on standard benchmarks. Our code is available at https://github.com/Yunkaidang/uncertainty

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。

↑