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 pre-image problem in kernel methods.

James Tin-yau Kwok1, Ivor Wai-hung Tsang

  • 1Department of Computer Science, The Hong Kong University of Science and Technology, Clear Water Bay, Hong Kong. jamesk@cs.ust.hk

IEEE Transactions on Neural Networks
|November 30, 2004
PubMed
Summary

This study introduces a novel, non-iterative method for finding kernel feature pre-images, crucial for applications like kernel principal component analysis (PCA) in image denoising. The new approach uses linear algebra for stable, efficient pre-image localization, outperforming traditional optimization techniques.

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

GBNet: Gated Boundary-Aware Network for Camouflaged Object Detection.

IEEE transactions on image processing : a publication of the IEEE Signal Processing Society·2026
Same author

Exploring the Coordination of Frequency and Attention in Masked Image Modeling.

IEEE transactions on image processing : a publication of the IEEE Signal Processing Society·2025
Same author

Constructing Diverse Inlier Consistency for Partial Point Cloud Registration.

IEEE transactions on image processing : a publication of the IEEE Signal Processing Society·2024
Same author

Learning the Relation Between Similarity Loss and Clustering Loss in Self-Supervised Learning.

IEEE transactions on image processing : a publication of the IEEE Signal Processing Society·2023
Same author

Domain Adaption via Feature Selection on Explicit Feature Map.

IEEE transactions on neural networks and learning systems·2018
Same author

A Note on the Unification of Adaptive Online Learning.

IEEE transactions on neural networks and learning systems·2016

Area of Science:

  • Machine Learning
  • Kernel Methods
  • Computer Vision

Background:

  • Kernel methods map data to high-dimensional feature spaces.
  • Finding pre-images is essential for interpreting kernel-based models.
  • Traditional pre-image methods use nonlinear optimization, facing issues like local minima and instability.

Purpose of the Study:

  • To develop a direct, stable, and efficient method for kernel feature pre-image computation.
  • To address limitations of traditional nonlinear optimization approaches for pre-image problems.

Main Methods:

  • A novel approach directly computes pre-images using distance constraints in feature space.
  • The method relies solely on linear algebra, avoiding iterative processes.
  • Noniterative, numerically stable computation.

Related Experiment Videos

Main Results:

  • The proposed method successfully finds pre-images without local minima or numerical instability.
  • Significantly improved performance in kernel principal component analysis (PCA) for image denoising.
  • Enhanced performance in kernel clustering tasks on the USPS dataset.

Conclusions:

  • The direct pre-image method offers a robust and efficient alternative to nonlinear optimization.
  • This technique enhances the applicability and performance of kernel-based methods in various domains.
  • The findings pave the way for more reliable kernel PCA and clustering applications.