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

Using the EM algorithm to train neural networks: misconceptions and a new algorithm for multiclass classification.

Shu-Kay Ng1, Geoffrey John McLachlan

  • 1Department of Mathematics, University of Queensland, Brisbane QLD 4072, Australia. skn@maths.uq.edu.au

IEEE Transactions on Neural Networks
|September 24, 2004
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

Hospital-initiated community pharmacist medication review post-discharge.

Aging clinical and experimental research·2026
Same author

Unraveling intersectional risks: Postnatal adversities condition the impact of prenatal alcohol exposures on early childhood sleep outcomes.

Development and psychopathology·2026
Same author

Preeclampsia is Associated with Altered Expression of Ferroptosis Biomarkers in Placental but not Maternal Vasculature.

Reproductive sciences (Thousand Oaks, Calif.)·2025
Same author

Re: Enhancing communication equity in critical care: Multimodal and technological approaches to overcome language barriers.

Australian critical care : official journal of the Confederation of Australian Critical Care Nurses·2025
Same author

Sex differences in onset and prevalence of 108 diseases and multimorbidity across lifespan in Yichang, China: quantitative analysis of real-world linked electronic health records.

BMJ open·2025
Same author

Correction: Preadmission medications and recent falls in older inpatients: an observational study.

International journal of clinical pharmacy·2025
Same journal

Universal perceptron and DNA-like learning algorithm for binary neural networks: LSBF and PBF implementations.

IEEE transactions on neural networks·2013
Same journal

Guest editorial: special section on white box nonlinear prediction models.

IEEE transactions on neural networks·2011
Same journal

Data-based fault-tolerant control of high-speed trains with traction/braking notch nonlinearities and actuator failures.

IEEE transactions on neural networks·2011
Same journal

Guest editorial: special section on data-based control, modeling, and optimization.

IEEE transactions on neural networks·2011
Same journal

Neural network-based multiple robot simultaneous localization and mapping.

IEEE transactions on neural networks·2011
Same journal

Data-driven model-free adaptive control for a class of MIMO nonlinear discrete-time systems.

IEEE transactions on neural networks·2011
See all related articles

The expectation-maximization (EM) algorithm clarifies misconceptions in training neural networks like multilayer perceptron (MLP) and mixture of experts (ME) for multiclass classification. An expectation-conditional maximization (ECM) algorithm shows superior performance over IRLS for ME networks.

Area of Science:

  • Machine Learning
  • Artificial Intelligence
  • Pattern Recognition

Background:

  • The expectation-maximization (EM) algorithm is widely used in neural network applications.
  • Misconceptions exist regarding the EM algorithm's application to neural networks.
  • Multilayer perceptron (MLP) and mixture of experts (ME) networks are key areas of interest.

Purpose of the Study:

  • Clarify misconceptions about the EM algorithm in neural networks.
  • Investigate EM algorithm's utility for training MLP and ME networks in multiclass classification.
  • Propose and evaluate a novel algorithm for ME network training.

Main Methods:

  • Adopted the EM algorithm for training MLP and ME networks.
  • Analyzed the convergence and performance of the iteratively reweighted least squares (IRLS) algorithm within the M-step.

Related Experiment Videos

  • Developed and tested the expectation-conditional maximization (ECM) algorithm for ME networks.

    • Main Results:

    • Identified limitations for EM algorithm application in MLP training.
    • Found stable convergence and monotonic log-likelihood increase for IRLS with a learning rate < 1.
    • Demonstrated superior performance of the ECM algorithm over IRLS on simulated and real data.

    Conclusions:

    • The EM algorithm can be effectively applied to train MLP and ME networks with careful consideration of limitations.
    • The ECM algorithm offers a more robust and performant alternative for training ME networks in multiclass classification tasks.