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

Efficient perceptron learning using constrained steepest descent.

S J Perantonis1, V Virvilis

  • 1Institute of Informatics and Telecommunications, National Research Center Demokritos, Athens, Greece. sper@iit.demokritos.gr

Neural Networks : the Official Journal of the International Neural Network Society
|August 11, 2000
PubMed
Summary

A novel algorithm trains single-layered perceptrons efficiently by using successive steepest descent. This method guarantees convergence and provides a natural criterion for linear separability in pattern classification tasks.

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

Two highly efficient second-order algorithms for training feedforward networks.

IEEE transactions on neural networks·2008
Same author

A dynamical model for the analysis and acceleration of learning in feedforward networks.

Neural networks : the official journal of the International Neural Network Society·2001
Same author

An efficient constrained training algorithm for feedforward networks.

IEEE transactions on neural networks·1995
Same author

Translation, rotation, and scale invariant pattern recognition by high-order neural networks and moment classifiers.

IEEE transactions on neural networks·1992

Area of Science:

  • Machine Learning
  • Artificial Intelligence
  • Pattern Recognition

Background:

  • Single-layered perceptrons are fundamental in machine learning for binary classification.
  • Existing training algorithms may lack guaranteed convergence or efficiency.
  • Parameter tuning and heuristics can complicate algorithm application.

Purpose of the Study:

  • To propose a novel, parameter-free algorithm for training single-layered perceptrons.
  • To ensure convergence and provide a clear criterion for linear separability.
  • To enhance the speed and effectiveness of perceptron training.

Main Methods:

  • The algorithm employs successive steepest descent directions concerning the perceptron cost function.
  • It frames the search for optimal directions as a quadratic programming problem.

Related Experiment Videos

  • A fast and effective solution is proposed for the quadratic programming task.

    • Main Results:

    • The algorithm converges in a finite number of steps.
    • It successfully finds a separating hyperplane for linearly separable patterns.
    • Algorithm termination without full separation indicates linear inseparability.

    Conclusions:

    • The proposed algorithm offers a robust and efficient method for training single-layered perceptrons.
    • It naturally determines linear separability without free parameters or heuristics.
    • Demonstrated superior speed on benchmark classification tasks compared to existing methods.