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 Videos

Evaluation of neural network robust reliability using information-gap theory.

S Gareth Pierce1, Yakov Ben-Haim, Keith Worden

  • 1Faculty of Mechanical Engineering, Technion-Israel Institute of Technology, Haifa 32000, Israel. s.g.pierce@eee.strath.ac.uk

IEEE Transactions on Neural Networks
|November 30, 2006
PubMed
Summary

This study introduces a new nonprobabilistic method to assess neural network robustness against data uncertainty. Findings show optimal neural networks for low uncertainty may not perform best with higher input uncertainty.

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

Rapid personalisation of cardiovascular models using invasively measured right ventricular pressure.

Computers in biology and medicine·2026
Same author

Machine-learning perspectives on Volterra system identification.

Philosophical transactions. Series A, Mathematical, physical, and engineering sciences·2025
Same author

Interpreting PPV and NPV of Diagnostic Tests with Uncertain Prevalence.

Rambam Maimonides medical journal·2024
Same author

Evidence and uncertainty: An info-gap analysis of uncertainty-augmenting evidence.

Risk analysis : an official publication of the Society for Risk Analysis·2024
Same author

Advancing Carbon Fiber Composite Inspection: Deep Learning-Enabled Defect Localization and Sizing via 3-D U-Net Segmentation of Ultrasonic Data.

IEEE transactions on ultrasonics, ferroelectrics, and frequency control·2024
Same author

Unsupervised machine learning for flaw detection in automated ultrasonic testing of carbon fibre reinforced plastic composites.

Ultrasonics·2024

Area of Science:

  • Artificial Intelligence
  • Machine Learning
  • Data Science

Background:

  • Evaluating neural network robustness against input data uncertainty is crucial for reliable real-world applications.
  • Existing methods often rely on probabilistic approaches, which may not capture all forms of uncertainty.
  • Nonprobabilistic methods offer an alternative perspective for uncertainty quantification in neural networks.

Purpose of the Study:

  • To present a novel nonprobabilistic technique for evaluating neural network robustness against input uncertainty.
  • To develop a theoretical framework for applying information-gap theory to neural networks.
  • To demonstrate the practical application of this technique across diverse datasets.

Main Methods:

  • Conventional optimization techniques were used to train multilayer perceptron (MLP) networks.

Related Experiment Videos

  • An information-gap model was employed for uncertainty analysis to quantify network response to input data uncertainty.
  • The proposed framework was tested on three distinct case studies: 2-D classification, aircraft wing vibration classification, and breast cancer incidence classification.

    • Main Results:

    • The study demonstrates that the optimal neural network configuration for low input uncertainty is not necessarily optimal for higher uncertainty levels.
    • The information-gap approach provides a quantifiable measure of network performance degradation under varying degrees of input uncertainty.
    • Performance variations were observed across the three tested applications, highlighting the context-dependent nature of network robustness.

    Conclusions:

    • The developed information-gap framework offers a valuable tool for assessing and understanding neural network robustness in the presence of input uncertainty.
    • This nonprobabilistic approach provides complementary insights to traditional probabilistic methods for uncertainty evaluation.
    • The findings underscore the importance of considering input uncertainty when selecting or designing neural networks for specific tasks.