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

A two-stage linear discriminant analysis via QR-decomposition.

Jieping Ye1, Qi Li

  • 1Department of Computer Science and Engineering, University of Minnesota-Twin Cities, 4-192 EE/CSCI Bldg., 200 Union St. SE, Minneapolis, MN 55455, USA. jieping@cs.umn.edu

IEEE Transactions on Pattern Analysis and Machine Intelligence
|June 10, 2005
PubMed
Summary

This study introduces LDA/QR, a novel two-stage method for feature extraction that overcomes singularity problems in Linear Discriminant Analysis (LDA). LDA/QR offers an efficient and scalable solution for high-dimensional data classification tasks.

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

scDifformer: diffusion-based post-training for virtual cell modeling across large-scale single-cell data.

Nucleic acids research·2026
Same author

Author Correction: UKB-MDRMF: a multi-disease risk and multimorbidity framework based on UK biobank data.

Nature communications·2026
Same author

MyESL: A Software for Evolutionary Sparse Learning in Molecular Phylogenetics and Genomics.

Molecular biology and evolution·2025
Same author

Rhizosphere microbial diversity and functional roles in tea cultivars: insights from high-throughput sequencing and functional isolates.

Plant signaling & behavior·2025
Same author

LLaFS++: Few-Shot Image Segmentation With Large Language Models.

IEEE transactions on pattern analysis and machine intelligence·2025
Same author

CATI: A medical context-enhanced framework for diagnosis code assignment in the UK Biobank study.

Artificial intelligence in medicine·2025

Area of Science:

  • Machine Learning
  • Data Science
  • Pattern Recognition

Background:

  • Linear Discriminant Analysis (LDA) is a key technique for feature extraction and dimensionality reduction in high-dimensional data.
  • Classical LDA faces singularity issues, limiting its application when scatter matrices are singular.
  • Existing extensions like PCA+LDA are often computationally intensive and not scalable.

Purpose of the Study:

  • To propose a novel two-stage LDA method, LDA/QR, addressing singularity problems.
  • To enhance computational efficiency and scalability compared to existing LDA extensions.
  • To provide a robust alternative for feature extraction in image and text classification.

Main Methods:

  • A two-stage approach combining QR decomposition with LDA.

Related Experiment Videos

  • The first stage utilizes QR decomposition on a matrix of class centroids.
  • The second stage applies LDA, preceded by an efficient dimension reduction technique.

    • Main Results:

    • LDA/QR effectively overcomes singularity problems inherent in classical LDA.
    • The proposed method demonstrates superior efficiency and scalability.
    • Experimental results on face images and text documents validate its effectiveness.

    Conclusions:

    • LDA/QR presents a significant advancement in LDA methodology.
    • The algorithm offers a computationally efficient and scalable solution for high-dimensional data analysis.
    • It provides a practical and effective approach for classification tasks where traditional LDA fails.