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

Learning algorithms based on linearization

R Hahnloser1

  • 1Institute for Theoretical Physics, ETHZ, Zürich, Switzerland. rich@ini.phys.ethz.ch

Network (Bristol, England)
|December 23, 1998
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

Integrating neuromorphic action-oriented perceptual inputs to generate a navigation behaviour for a robot.

International journal of neural systems·2000
Same author

Feedback interactions between neuronal pointers and maps for attentional processing.

Nature neuroscience·1999
Same journal

Enhancing IoT security: A Creative Swagger Optimization algorithm for DDoS defence.

Network (Bristol, England)·2026
Same journal

Parametric optimization for electrical discharge diamond grinding (EDDG) system using dual approach.

Network (Bristol, England)·2025
Same journal

A novel lung cancer diagnosis model using hybrid convolution (2D/3D)-based adaptive DenseUnet with attention mechanism.

Network (Bristol, England)·2025
Same journal

Hybrid optimization enabled Eff-FDMNet for Parkinson's disease detection and classification in federated learning.

Network (Bristol, England)·2025
Same journal

AI-driven plant disease detection with tailored convolutional neural network.

Network (Bristol, England)·2025
Same journal

Layer modified residual Unet++ for speech enhancement using Aquila Black widow optimizer algorithm.

Network (Bristol, England)·2025
See all related articles

This study introduces tangential correlation (TC) algorithms for neural network learning, offering a mechanical framework for efficient weight updates. These methods demonstrate optimal scaling and successful application in supervised and unsupervised learning tasks.

Area of Science:

  • Computational Neuroscience
  • Machine Learning Theory
  • Artificial Intelligence

Background:

  • Traditional neural network learning often relies on gradient descent in weight space.
  • Existing methods can face challenges with scaling and computational complexity.
  • A need exists for alternative learning algorithms with improved efficiency and broader applicability.

Purpose of the Study:

  • To propose a novel mechanical framework for neural network learning algorithms.
  • To develop local and simple learning algorithms based on constraint satisfaction.
  • To investigate the properties and applications of tangential correlation (TC) algorithms.

Main Methods:

  • Interpreting neural networks as systems of configuration constraints.
  • Deriving unsupervised and supervised learning algorithms from these constraints.

Related Experiment Videos

  • Utilizing tangential correlation (TC) algorithms that operate in state space, not weight space.
  • Comparing TC algorithms with standard backpropagation on perceptrons and recurrent networks.

    • Main Results:

    • TC algorithms exhibit optimal scaling properties and simple weight update rules.
    • Demonstrated successful training of fixed points in recurrent networks.
    • Achieved unsupervised learning of oscillations with variable frequencies in various recurrent network architectures.
    • Showcased the flexibility of the framework for different network designs and learning tasks.

    Conclusions:

    • The proposed mechanical framework provides a viable alternative to gradient-based learning.
    • Tangential correlation (TC) algorithms offer efficient and scalable solutions for neural network training.
    • The findings are applicable to both the analysis and synthesis of learning algorithms in artificial neural networks.