Jove
Visualize
联系我们
JoVE
x logofacebook logolinkedin logoyoutube logo
关于 JoVE
概览领导团队博客JoVE 帮助中心
作者
出版流程编辑委员会范围与政策同行评审常见问题投稿
图书馆员
用户评价订阅访问资源图书馆顾问委员会常见问题
研究
JoVE JournalMethods CollectionsJoVE Encyclopedia of Experiments存档
教育
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab Manual教师资源中心教师网站
使用条款与条件
隐私政策
政策

相关实验视频

Updated: Jan 8, 2026

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
03:31

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Published on: December 15, 2023

991

在文本到图像扩散模型中用于后门检测的动态注意力分析.

Zhongqi Wang, Jie Zhang, Shiguang Shan

    IEEE transactions on pattern analysis and machine intelligence
    |December 15, 2025
    PubMed
    概括
    此摘要是机器生成的。

    相关概念视频

    Difference from Background: Limit of Detection01:05

    Difference from Background: Limit of Detection

    8.0K
    The limit of detection (LOD) is the smallest amount of analyte that can be distinguished from the background noise. The LOD value corresponds to the concentration at which the analyte signal is three times larger than the standard deviation of the blank signal. Below this value, the analyte signal cannot be differentiated from the background noise. It is calculated by dividing the calibration slope by 3 times the standard deviation of the blank signals.
    The LOD indicates the presence or absence...
    8.0K

    您也可能阅读

    相关文章

    通过共同作者、期刊和引用图与本文相关的文章。

    排序
    Same author

    Promoting contraceptive use among unmarried female migrants in one factory in Shanghai: a pilot workplace intervention.

    BMC health services research·2007
    Same author

    Apoptosis related protein 3, an ATRA-upregulated membrane protein arrests the cell cycle at G1/S phase by decreasing the expression of cyclin D1.

    Biochemical and biophysical research communications·2007
    Same author

    Efficient one-pot synthesis of highly substituted pyridin-2(1H)-ones via the Vilsmeier-Haack reaction of 1-acetyl,1-carbamoyl cyclopropanes.

    Organic letters·2007
    Same author

    Effects of Andrographitis Paniculata extracts on the expression of CD40 in endothelial cells.

    Journal of Huazhong University of Science and Technology. Medical sciences = Hua zhong ke ji da xue xue bao. Yi xue Ying De wen ban = Huazhong keji daxue xuebao. Yixue Yingdewen ban·2007
    Same author

    Theoretical and experimental investigations on the size of alginate microspheres prepared by dropping and spraying.

    Journal of microencapsulation·2007
    Same author

    Beneficial effects of losartan on vascular injury induced by advanced glycosylation end products and their receptors in spontaneous hypertension rats.

    Molecular and cellular biochemistry·2007
    Same journal

    Benchmarking the Robustness of Autonomous Driving to Environmental Illusions: A Lane Perception Perspective.

    IEEE transactions on pattern analysis and machine intelligence·2026
    Same journal

    Learning Topology-Aware Representations via Test-Time Adaptation for Anomaly Segmentation.

    IEEE transactions on pattern analysis and machine intelligence·2026
    Same journal

    TraGraph-GS: Trajectory Graph-based Gaussian Splatting for Arbitrary Large-Scale Scene Rendering.

    IEEE transactions on pattern analysis and machine intelligence·2026
    Same journal

    SWIFT: A Small-World Interaction Framework for Flow-Aware Trajectory Prediction in Autonomous Driving.

    IEEE transactions on pattern analysis and machine intelligence·2026
    Same journal

    HardFlow: Hard-Constrained Sampling for Flow-Matching Models Via Trajectory Optimization.

    IEEE transactions on pattern analysis and machine intelligence·2026
    Same journal

    Industrial Brain: Self-Evolving Neuro-Symbolic Autonomy with Causal Resilience for Cyber-Physical Systems.

    IEEE transactions on pattern analysis and machine intelligence·2026
    查看所有相关文章

    本研究引入了动态注意力分析 (DAA),通过分析动态特征演变来检测文本到图像扩散模型中的后门攻击. 通过检查注意力地图异常,DAA有效地识别恶意样本,优于现有的方法.

    科学领域:

    • 人工智能的人工智能
    • 机器学习安全 机器学习安全

    背景情况:

    • 文本到图像扩散模型容易受到使用隐藏文本触发器的后门攻击.
    • 现有的后门检测方法往往忽视了扩散模型的动态性质.

    研究的目的:

    • 引入一种新的后门检测方法,即动态注意力分析 (DAA),利用扩散模型固有的动力.
    • 为了证明动态特征与静态特征相比,在检测后门攻击方面具有优越的指标.

    主要方法:

    • 检查交叉注意力图的动态演变,以在令牌的后门样本中识别不同的特征模式.
    • 介绍DAA-I,它使用空间独立的注意力地图上的弗罗贝尼乌斯规范量化动态异常.
    • 提出DAA-S,一种基于动态系统的方法,使用基于图形的状态方程来建模注意力图的相互作用并分析稳定性.

    主要成果:

    • 与良性样本相比,后门样本在令牌上表现出独特的特征演变模式.
    • DAA-I和DAA-S有效量化了注意力图中的动态异常.
    • 提出的DAA方法在六种攻击场景中显著优于现有的检测方法.

    结论:

    • 动态注意力分析 (DAA) 为检测文本到图像扩散模型中的后门攻击提供了一个有希望的新视角.

    相关实验视频

    Last Updated: Jan 8, 2026

    Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
    03:31

    Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

    Published on: December 15, 2023

    991
  • 扩散模型的动态性为强大的后门检测提供了关键指标.
  • DAA的平均F1分数为79.27%,AUC为86.27%,达到了高性能.