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

The VC dimension for mixtures of binary classifiers.

W Jiang1

  • 1Department of Statistics, Northwestern University, Evanston, IL 60208, USA.

Neural Computation
|August 10, 2000
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

Characterization and comparison of transgenic Artemisia annua GYR and wild-type NON-GYR plants in an environmental release trial.

Genetics and molecular research : GMR·2016
Same author

qKAT: a high-throughput qPCR method for KIR gene copy number and haplotype determination.

Genome medicine·2016
Same author

[Prospective evaluation on ventilator-associated events: a cohort study from eight intensive care units].

Zhonghua liu xing bing xue za zhi = Zhonghua liuxingbingxue zazhi·2016
Same author

[Technical advancement improves survival in patients with locally advanced non-small cell lung cancer (LA-NSCLC) receiving definitive radiotherapy].

Zhonghua zhong liu za zhi [Chinese journal of oncology]·2016
Same author

DNA methylation regulates gabrb2 mRNA expression: developmental variations and disruptions in l-methionine-induced zebrafish with schizophrenia-like symptoms.

Genes, brain, and behavior·2016
Same author

Effects of extracellular matrix on the growth characteristics of human dermal papillae cells in vitro.

Clinical and experimental dermatology·2016

Mixtures-of-experts (ME) classification uses logistic or Bernoulli models. The Vapnik-Chervonenkis (VC) dimension of ME is bounded by the number of experts (m) and input dimension (s), exactly equaling m for scalar inputs.

Area of Science:

  • Machine Learning
  • Statistical Modeling
  • Computational Learning Theory

Background:

  • Mixtures-of-experts (ME) models combine multiple expert models using local weights for classification tasks.
  • Understanding the generalization capability of ME architectures, particularly their Vapnik-Chervonenkis (VC) dimension, is crucial for theoretical analysis.
  • Existing research provides bounds for the VC dimension of ME, but exact values are often elusive.

Purpose of the Study:

  • To determine the Vapnik-Chervonenkis (VC) dimension of Mixtures-of-Experts (ME) architectures.
  • To establish precise bounds for the VC dimension based on the number of experts and input dimensionality.
  • To investigate the specific case of ME with Bernoulli experts and scalar inputs.

Main Methods:

  • Theoretical analysis of the VC dimension for ME architectures.

Related Experiment Videos

  • Derivation of lower and upper bounds for the VC dimension in terms of the number of experts (m) and input dimension (s).
  • Specific analysis for ME models composed of Bernoulli experts with scalar input data.

    • Main Results:

    • The VC dimension of ME is bounded below by m (number of experts) and above by O(m^4 * s^2) (where s is input dimension).
    • For ME with Bernoulli experts and scalar input, the lower bound m is precisely attained.
    • The exact VC dimension for this specific ME configuration is proven to be equal to the number of experts, m.

    Conclusions:

    • The VC dimension of ME architectures is tightly characterized, especially for simpler configurations.
    • The findings provide exact theoretical bounds for generalization in specific ME models.
    • This work contributes to a deeper understanding of the learnability and complexity of Mixtures-of-Experts models.