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: May 5, 2026

Applying Hyperspectral Reflectance Imaging to Investigate the Palettes and the Techniques of Painters
07:05

Applying Hyperspectral Reflectance Imaging to Investigate the Palettes and the Techniques of Painters

Published on: June 18, 2021

2.0K

A comparative evaluation of explainability techniques for image data.

Mykyta Skliarov1, Radwa El Shawi2, Chedia Dhaoui3

  • 1Institute of Computer Science, University of Tartu, Tartu, 51009, Estonia.

Scientific Reports
|November 25, 2025
PubMed
Summary
This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

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

Sort by
Same author

Sweet cherry pomace polyphenols disrupt Salmonella enteritidis colonization on chicken skin.

Food microbiology·2026
Same author

Low-dose Blue Light Extends the Antibacterial Spectrum of Bacitracin Against Escherichia coli.

Journal of global antimicrobial resistance·2026
Same author

Calcium hydroxide nanoparticles-induced oxidative stress and mitochondrial impairment drive genomic instability and programmed cell death in colorectal cancer cells.

Naunyn-Schmiedeberg's archives of pharmacology·2026
Same author

Yttrium oxide nanoparticles induce potent selective cytotoxicity in HeLa cervical cance cells through ROS-mediated genomic instability and mitochondrial apoptosis.

Scientific reports·2026
Same author

Suboptimal uptake of the malaria vaccine in Sudan: what's behind the blinds?

Malaria journal·2026
Same author

A manta ray-bayesian optimization approach for hyperparameter-tuned convolutional neural networks in lung cancer classification.

Scientific reports·2026

No single explainable artificial intelligence (XAI) method excels in all areas. Gradient-based techniques like Integrated Gradients and SmoothGrad offer superior fidelity and stability, but trade-offs exist between explanation quality and computational efficiency.

Area of Science:

  • Computer Vision
  • Machine Learning
  • Explainable Artificial Intelligence (XAI)

Background:

  • Deep neural networks achieve human-level performance in computer vision.
  • The rise of black-box models necessitates transparency and explainability.
  • Saliency maps are popular for image data analysis but challenging to evaluate.

Purpose of the Study:

  • To conduct a comprehensive comparative evaluation of six widely used saliency map explainability techniques.
  • To assess the strengths and limitations of LIME, SHAP, GradCAM, GradCAM++, Integrated Gradients (IntGrad), and SmoothGrad.
  • To evaluate techniques using five quantitative metrics: fidelity, stability, identity, separability, and computational time.

Main Methods:

  • Comparative evaluation of six saliency map techniques.

Related Experiment Videos

Last Updated: May 5, 2026

Applying Hyperspectral Reflectance Imaging to Investigate the Palettes and the Techniques of Painters
07:05

Applying Hyperspectral Reflectance Imaging to Investigate the Palettes and the Techniques of Painters

Published on: June 18, 2021

2.0K
  • Application of five quantitative metrics across three benchmark datasets and three deep learning architectures.
  • Statistical analysis to determine performance differences.

    • Main Results:

    • No single XAI method performed best across all metrics.
    • Integrated Gradients and SmoothGrad showed the best fidelity and stability.
    • GradCAM and GradCAM++ offered the highest computational efficiency but lower fidelity.
    • SHAP performed strongly on the SVHN dataset.

    Conclusions:

    • There are inherent trade-offs between explanation quality and computational efficiency in XAI methods.
    • Gradient-based methods generally provide better fidelity and stability.
    • Method performance varies significantly across different datasets and tasks.