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

Template-based procedures for neural network interpretation.

J A. Alexander1, M C. Mozer

  • 1Department of Computer Science, University of Colorado, Boulder, USA

Neural Networks : the Official Journal of the International Neural Network Society
|March 29, 2003
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 journal

Exploiting audio-visual modalities in videos: Object detection via multi-stage bilateral coupling network.

Neural networks : the official journal of the International Neural Network Society·2026
Same journal

Reliability-aware modality completion with cross-modal distillation for federated learning with missing modalities.

Neural networks : the official journal of the International Neural Network Society·2026
Same journal

IGFD-Net: Illumination-guided frequency decoupling for polarization image fusion.

Neural networks : the official journal of the International Neural Network Society·2026
Same journal

Multiple-Strategies dung beetle optimizer and its applications in engineering optimization and bankruptcy prediction.

Neural networks : the official journal of the International Neural Network Society·2026
Same journal

Aggregating global-scale pixel-wise forgery cues within a graph.

Neural networks : the official journal of the International Neural Network Society·2026
Same journal

Finite-Time intermittent control for secure synchronization of Neutral-Type stochastic delayed neural networks under aperiodic DoS attacks.

Neural networks : the official journal of the International Neural Network Society·2026
See all related articles

This study introduces a new method for extracting symbolic rules from neural networks, making their decision-making processes understandable. This approach enhances trust and knowledge transfer in artificial intelligence systems.

Area of Science:

  • Artificial Intelligence
  • Machine Learning
  • Connectionism

Background:

  • Neural networks exhibit strong performance but suffer from opacity, hindering understanding and knowledge transfer.
  • Traditional rule-oriented AI offers transparency, a contrast to opaque neural network models.
  • Understanding neural network behavior is crucial for confidence and broader application of learned knowledge.

Purpose of the Study:

  • To develop a principled approach for symbolic rule extraction from multilayer feedforward networks.
  • To address the opacity challenge in neural networks by translating weights into symbolic terms.
  • To improve the interpretability and applicability of neural network models.

Main Methods:

  • Utilized a novel 'weight templates' approach, defining parameterized regions in weight space.

Related Experiment Videos

  • Developed efficient methods for identifying and instantiating template parameters to match network weights.
  • Explored different representations to accommodate various logical expressions (disjunctions, conjunctions, n-of-m expressions) with varying computational complexity.

    • Main Results:

    • The proposed method offers a principled and efficient way to extract symbolic rules from neural networks.
    • Demonstrated flexibility in handling different classes of symbolic expressions with defined complexities.
    • Simulation results validated the application and highlighted the strengths and limitations of the rule extraction technique.

    Conclusions:

    • The developed rule extraction method provides a significant advancement in understanding neural network internals.
    • This approach offers advantages in simplicity, computational efficiency, and flexibility compared to existing techniques.
    • The ability to translate neural network weights into symbolic rules facilitates greater trust and knowledge transfer in AI.