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 Video

Updated: Jul 7, 2026

Automated Analysis of C. elegans Fluorescence Images using SegElegans
06:27

Automated Analysis of C. elegans Fluorescence Images using SegElegans

Published on: October 10, 2025

A new EM-based training algorithm for RBF networks.

Marcelino Lázaro1, Ignacio Santamaría, Carlos Pantaleón

  • 1Departamento Ingeniería de Comunicaciones, ETSIIT, Universidad de Cantabria, Av. Los Castros s/n, 39005, Santander, Spain.

Neural Networks : the Official Journal of the International Neural Network Society
|February 11, 2003
PubMed
Summary

This study introduces an improved Expectation-Maximization (EM) algorithm for faster training of Radial Basis Function (RBF) networks. The new method enhances convergence by adaptively decomposing residuals, improving performance in nonlinear modeling 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

One-step Bayesian example-dependent cost classification: The OsC-MLP method.

Neural networks : the official journal of the International Neural Network Society·2024
Same author

Information-Theoretic Analysis of a Family of Improper Discrete Constellations.

Entropy (Basel, Switzerland)·2020
Same author

A Bayes Risk Minimization Machine for Example-Dependent Cost Classification.

IEEE transactions on cybernetics·2019
Same author

Kernel recursive least-squares tracker for time-varying regression.

IEEE transactions on neural networks and learning systems·2014
Same author

A learning algorithm for adaptive canonical correlation analysis of several data sets.

Neural networks : the official journal of the International Neural Network Society·2006
Same author

A spectral clustering approach to underdetermined postnonlinear blind source separation of sparse sources.

IEEE transactions on neural networks·2006

Area of Science:

  • Machine Learning
  • Artificial Neural Networks
  • Computational Science

Background:

  • Training feedforward networks with local activation functions, like Radial Basis Function (RBF) networks, can be slow.
  • Existing Expectation-Maximization (EM) algorithms often use inefficient residual decomposition methods, hindering convergence.

Purpose of the Study:

  • To develop a novel Expectation-Maximization (EM) algorithm that accelerates the training of feedforward networks with local activation functions.
  • To improve the convergence speed and efficiency of RBF network training.

Main Methods:

  • A new Expectation-Maximization (EM) algorithm is proposed, featuring a soft decomposition of the residual in the E-step.
  • Decoupling variables are estimated using the posterior probability of a component given an input-output pattern.

More Related Videos

Utilizing vmTracking to Improve the Accuracy of Multi-Animal Pose Estimation in Rodent Social Behavior Studies
07:34

Utilizing vmTracking to Improve the Accuracy of Multi-Animal Pose Estimation in Rodent Social Behavior Studies

Published on: November 7, 2025

Related Experiment Videos

Last Updated: Jul 7, 2026

Automated Analysis of C. elegans Fluorescence Images using SegElegans
06:27

Automated Analysis of C. elegans Fluorescence Images using SegElegans

Published on: October 10, 2025

Utilizing vmTracking to Improve the Accuracy of Multi-Animal Pose Estimation in Rodent Social Behavior Studies
07:34

Utilizing vmTracking to Improve the Accuracy of Multi-Animal Pose Estimation in Rodent Social Behavior Studies

Published on: November 7, 2025

  • The adaptive decomposition accounts for the local nature of activation functions, enabling RBF units to specialize in distinct input space subregions.

    • Main Results:

    • The proposed EM algorithm demonstrates improved convergence rates for networks with local activation units.
    • The adaptive decomposition strategy effectively utilizes the localized nature of RBF units.
    • Successful application to the nonlinear modeling of a MESFET transistor validates the algorithm's efficacy.

    Conclusions:

    • The novel EM algorithm offers a significant speed-up in training feedforward networks with local activation functions.
    • Adaptive residual decomposition enhances the training efficiency and convergence of RBF networks.
    • This approach provides a robust method for complex nonlinear modeling tasks.