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

Constructive incremental learning from only local information

Schaal1, Atkeson

  • 1University of Southern California, Department of Computer Science, Los Angeles CA, US, HEDCO Neuroscience Building 103, 90089. sschal@usc.edu.

Neural Computation
|November 6, 1998
PubMed
Summary

This study presents an incremental learning system for regression using spatially localized linear models. It independently learns model parameters and receptive fields, offering a principled approach to the bias-variance dilemma.

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

RAPD variation within and between natural populations of the wild rice oryza rufipogon from china and brazil

Heredity·1999
Same author

Is imitation learning the route to humanoid robots?

Trends in cognitive sciences·1999
See all related articles

Area of Science:

  • Machine Learning
  • Artificial Intelligence
  • Statistical Modeling

Background:

  • Existing regression methods often require complex model structures or global parameter optimization.
  • Addressing the bias-variance dilemma is crucial for robust predictive modeling.

Purpose of the Study:

  • To introduce a novel constructive, incremental learning system for regression problems.
  • To enable independent learning of model parameters and receptive field characteristics.
  • To enhance resource allocation and robustness against negative interference.

Main Methods:

  • Utilizing spatially localized linear models to represent data.
  • Employing independent learning for receptive field size/shape and model parameters.
  • Minimizing a weighted local cross-validation error incrementally.

Related Experiment Videos

Main Results:

  • The system demonstrates principled handling of the bias-variance dilemma.
  • Spatial localization of models enhances robustness toward negative interference.
  • The learning system can be interpreted as a nonparametric adaptive bandwidth smoother or a mixture of isolated experts.

Conclusions:

  • Purely local learning offers significant potential for complex data modeling.
  • The proposed system provides a powerful approach to learning with receptive fields.
  • Independent learning of local models facilitates adaptive resource allocation.