Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Video

Updated: Oct 14, 2025

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

685

Feature Distillation in Deep Attention Network Against Adversarial Examples.

Xin Chen, Jian Weng, Xiaoling Deng

    IEEE Transactions on Neural Networks and Learning Systems
    |November 5, 2021
    PubMed
    Summary

    Deep neural networks (DNNs) are vulnerable to adversarial examples. This study introduces attention modules that compress these perturbations by focusing on important image regions, enhancing robustness and accuracy.

    Related Concept Videos

    Survival Tree01:19

    Survival Tree

    181
    Survival trees are a non-parametric method used in survival analysis to model the relationship between a set of covariates and the time until an event of interest occurs, often referred to as the "time-to-event" or "survival time." This method is particularly useful when dealing with censored data, where the event has not occurred for some individuals by the end of the study period, or when the exact time of the event is unknown.
     Building a Survival Tree
    Constructing a...
    181

    You might also read

    Related Articles

    Articles linked to this work by shared authors, journal, and citation graph.

    Sort by
    Same author

    N-Doping Activated Presodiation Enhances Sodium-Ion Provision in Hard Carbon Anodes.

    Advanced science (Weinheim, Baden-Wurttemberg, Germany)·2026
    Same author

    Effects of Postures on Identifying Users for Selection-Based Behavioral Authentication in Virtual Reality.

    IEEE transactions on visualization and computer graphics·2026
    Same author

    EXT2 promotes sarcoma progression and immune evasion via the AKT/c-Myc/PD-L1 axis: a multi-omics and validation study.

    Journal of translational medicine·2026
    Same author

    Carnosine-modified gelatin-hyaluronic acid hydrogel comprising fenofibrate-loaded nanoparticles targeting chondrocyte ferroptosis and macrophage polarization for synergistic osteoarthritis therapy.

    International journal of biological macromolecules·2026
    Same author

    Cuproptosis-associated PDHA1 promotes sarcoma progression and immunotherapy responsiveness via the E2F1-PD-L1 axis: a multi-omics and clinical validation study.

    NPJ precision oncology·2026
    Same author

    Lithophilic yet Inert Interfaces Strategy for Stable Lithium Metal Anodes.

    ACS nano·2026

    Area of Science:

    • Computer Vision
    • Artificial Intelligence
    • Machine Learning

    Background:

    • Deep neural networks (DNNs) are susceptible to adversarial examples, posing security risks.
    • Current defenses often rely on full image information, which may not mimic human visual processing.
    • Attention mechanisms, successful in various computer tasks, offer a potential avenue for defense.

    Purpose of the Study:

    • To theoretically prove and experimentally validate that attention modules can compress adversarial perturbations in DNNs.
    • To design and evaluate novel attention modules based on frequency decomposition and reorganization for enhanced adversarial robustness.
    • To investigate the impact of these attention modules on clean image classification accuracy.

    Main Methods:

    • Theoretical analysis proving attention modules disrupt DNN linearity to compress adversarial perturbations.

    Related Experiment Videos

    Last Updated: Oct 14, 2025

    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

    685
  • Design and comparison of three attention module variants utilizing frequency decomposition and reorganization.
  • Experimental evaluation on CIFAR and ImageNet datasets, assessing robustness and classification accuracy.

    • Main Results:

    • Attention modules theoretically compress adversarial perturbations by destroying DNNs' linear characteristics.
    • Frequency-based attention modules demonstrate significant robustness against adversarial examples.
    • Proposed attention modules achieve comparable or improved classification accuracies on clean images.
    • Integration with existing defenses further enhances adversarial robustness.

    Conclusions:

    • Attention modules, particularly those employing frequency reorganization, offer an effective defense against adversarial perturbations.
    • These modules improve DNN robustness without sacrificing, and potentially enhancing, performance on clean data.
    • The proposed attention mechanisms represent a promising component for building more resilient deep learning systems.