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

Bayesian approach to feature selection and parameter tuning for support vector machine classifiers.

Carl Gold1, Alex Holub, Peter Sollich

  • 1Computation and Neural Systems, California Institute of Technology, 139-74, Pasadena, CA 91125, USA. carlg@caltech.edu

Neural Networks : the Official Journal of the International Neural Network Society
|August 23, 2005
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

Scaling the glassy dynamics of active particles: Tunable fragility and reentrance.

Proceedings of the National Academy of Sciences of the United States of America·2026
Same author

Static and dynamic rough energy landscapes can lead to identical diffusivity.

Proceedings of the National Academy of Sciences of the United States of America·2026
Same author

Ensemble inequivalence in the design of mixtures with super-Gibbs phase coexistence.

Physical review. E·2025
Same author

An elastoplastic model approach for the relaxation dynamics of active glasses.

Soft matter·2025
Same author

Bringing Together Two Paradigms of Nonequilibrium: Fragile versus Robust Aging in Driven Glassy Systems.

Physical review letters·2024
Same author

Ductile and brittle yielding of athermal amorphous solids: A mean-field paradigm beyond the random-field Ising model.

Physical review. E·2024

This study introduces Bayesian Support Vector Machines (SVMs) with automatic relevance determination (ARD) for efficient feature selection. ARD significantly reduces features, improving classification performance and model interpretability.

Area of Science:

  • Machine Learning
  • Computational Statistics
  • Pattern Recognition

Background:

  • Support Vector Machines (SVMs) are powerful classification tools.
  • Hyperparameter tuning and feature selection are crucial for SVM performance.
  • Traditional SVMs often lack systematic methods for feature relevance assessment.

Purpose of the Study:

  • To develop a Bayesian approach for SVMs enabling systematic hyperparameter tuning and feature selection.
  • To leverage automatic relevance determination (ARD) for identifying and removing irrelevant input features.
  • To enhance classification accuracy and model efficiency through optimized feature subsets.

Main Methods:

  • Formulated a Bayesian perspective of SVMs, defining an 'evidence' quantity for optimization.

Related Experiment Videos

  • Employed Hybrid Monte Carlo (HMC) sampling to approximate evidence gradients.
  • Utilized a Nyström approximation of the Gram matrix to accelerate HMC sampling.
  • Developed a relevance measure for systematic feature pruning based on tuned hyperparameters.

    • Main Results:

    • The Bayesian SVM with ARD demonstrated improved classification performance on datasets with irrelevant features.
    • Nyström approximation significantly reduced sampling times with minimal impact on accuracy.
    • ARD effectively identified and enabled pruning of irrelevant features, achieving up to 75% reduction.
    • While not always improving accuracy, feature elimination led to more parsimonious models.

    Conclusions:

    • Bayesian SVMs with ARD offer a robust framework for hyperparameter tuning and feature selection.
    • The Nyström approximation provides a computationally efficient method for Bayesian SVM analysis.
    • ARD facilitates effective feature reduction, leading to more interpretable and potentially more accurate models.