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

Distributed Learning, Recognition, and Prediction by ART and ARTMAP Neural Networks.

Gail A. Carpenter1

  • 1Boston University, Boston, U.S.A.

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

New distributed Adaptive Resonance Theory (ART) models offer stable, fast learning without forgetting. These models combine ART

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

Distributed ARTMAP: a neural network for fast distributed supervised learning.

Neural networks : the official journal of the International Neural Network Society·2003
See all related articles

Area of Science:

  • Artificial Intelligence
  • Computational Neuroscience

Background:

  • Traditional Adaptive Resonance Theory (ART) models face limitations with code representations.
  • Winner-take-all ART systems offer stable learning but lack noise tolerance.
  • Multilayer perceptrons provide noise tolerance and code compression but can suffer from catastrophic forgetting.

Purpose of the Study:

  • Introduce a novel class of distributed Adaptive Resonance Theory (ART) models.
  • Enhance learning, recognition, and prediction capabilities using distributed code representations.
  • Combine the strengths of winner-take-all ART and multilayer perceptrons.

Main Methods:

  • Developed distributed ART neural networks (dART and dARTMAP) with distributed code representations.
  • Replaced traditional path weights with dynamic weights based on coding node activation and adaptive thresholds.

Related Experiment Videos

  • Implemented a parallel distributed match-reset-search process for memory stabilization.

    • Main Results:

    • Distributed ART models enable fast and slow learning without catastrophic forgetting by apportioning learned changes.
    • Dynamic weights exhibit behavior resembling long-term potentiation (LTP) and long-term depression (LTD) based on input frequency.
    • The match-reset-search system stabilizes memory, preventing dART from becoming a simple distributed competitive learning network.

    Conclusions:

    • Distributed ART models offer a robust framework for learning, recognition, and prediction with enhanced stability and flexibility.
    • The novel dynamic weight mechanism and adaptive thresholds contribute to overcoming limitations of previous ART architectures.
    • These models present a significant advancement in artificial neural networks for complex cognitive tasks.