arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.02763cs.LG
收藏

通过双目标攻击暴露时间序列分类器解释中的脆弱性

Exposing Vulnerabilities in Explanation for Time Series Classifiers via Dual-Target Attacks

University of Science and Technology of China(中国科学技术大学)

Bohan Wang, Zewen Liu, Lu Lin, Hui Liu, Li Xiong, Ming Jin, Wei Jin

AI总结:

提出TSEF双目标攻击方法,通过联合操纵分类器和解释器输出,实现目标性误分类同时保持解释与参考一致,揭示解释稳定性不能作为决策鲁棒性的可靠指标。

摘要:

可解释的时间序列深度学习系统通常通过检查解释的时间一致性来评估,隐含地将此视为鲁棒性的证据。我们表明这一假设可能失败:预测和解释可以被对抗性解耦,实现目标性误分类,同时解释保持合理并与选定的参考理由一致。我们提出TSEF(时间序列解释欺骗器),一种双目标攻击,联合操纵分类器和解释器输出。与破坏解释并广泛分散归因质量的单目标误分类攻击相比,TSEF在保持解释与参考一致的同时实现目标性预测变化。在多个数据集和解释器骨干上,我们的结果一致表明,解释稳定性是决策鲁棒性的误导性代理,并激励为可信赖的时间序列任务进行耦合感知的鲁棒性评估。

英文摘要:

Interpretable time series deep learning systems are often assessed by checking temporal consistency on explanations, implicitly treating this as evidence of robustness. We show that this assumption can fail: Predictions and explanations can be adversarially decoupled, enabling targeted misclassification while the explanation remains plausible and consistent with a chosen reference rationale. We propose TSEF (Time Series Explanation Fooler), a dual-target attack that jointly manipulates the classifier and explainer outputs. In contrast to single-objective misclassification attacks that disrupt explanation and spread attribution mass broadly, TSEF achieves targeted prediction changes while keeping explanations consistent with the reference. Across multiple datasets and explainer backbones, our results consistently reveal that explanation stability is a misleading proxy for decision robustness and motivate coupling-aware robustness evaluations for trustworthy time series tasks.

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。

↑