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 canonical correlation neural network for multicollinearity and functional data.

Zhenkun Gou1, Colin Fyfe

  • 1Applied Computational Intelligence Research Unit, The University of Paisley, Paisley, Scotland PA1 2BE, UK. zhen-kun.gou@paisley.ac.uk

Neural Networks : the Official Journal of the International Neural Network Society
|March 24, 2004
PubMed
Summary

We introduce robust neural network models for Canonical Correlation Analysis (CCA) and Partial Least Squares (PLS) regression. These models effectively handle multicollinearity and improve interpretability, offering a flexible approach to data analysis.

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

Novel Machine Learning HIV Intervention for Sexual and Gender Minority Young People Who Have Sex With Men (uTECH): Protocol for a Randomized Comparison Trial.

JMIR research protocols·2024
Same author

Resting frontal gamma power at 16, 24 and 36 months predicts individual differences in language and cognition at 4 and 5 years.

Behavioural brain research·2011
Same author

Introduction.

International journal of neural systems·2010
Same author

On the Accuracy of Sequence-Based Computational Inference of Protein Residues Involved in Interactions with DNA.

Trends in applied sciences research·2010
Same author

Topological mappings of video and audio data.

International journal of neural systems·2009
Same author

Early cognitive and language skills are linked to resting frontal gamma power across the first 3 years.

Behavioural brain research·2008

Area of Science:

  • Machine Learning
  • Computational Neuroscience
  • Statistical Analysis

Background:

  • Canonical Correlation Analysis (CCA) is a statistical method for finding relationships between two sets of variables.
  • Existing neural implementations of CCA may lack robustness, particularly with multicollinear data.
  • Ridge Regression offers insights into improving algorithm robustness.

Purpose of the Study:

  • To develop a robust neural implementation of Canonical Correlation Analysis.
  • To create a flexible model capable of performing both Partial Least Squares (PLS) regression and CCA.
  • To enhance the interpretability of weight vectors in neural network models.

Main Methods:

  • Reviewing a recent neural implementation of CCA.
  • Applying concepts from Ridge Regression to enhance robustness.

Related Experiment Videos

  • Developing a novel neural network model with a tunable parameter.
  • Introducing a second penalty term for smoother weight vectors.

    • Main Results:

    • The proposed models demonstrate robustness on datasets with multicollinearity.
    • The flexible model successfully performs a spectrum of operations from PLS to CCA.
    • Parameter tuning is crucial for multicollinear data but less so for general datasets.
    • The new penalty term results in smoother and more interpretable weight vectors.

    Conclusions:

    • The developed neural network models offer robust and flexible solutions for analyzing multicollinear and general datasets.
    • The tunable parameter provides a unified framework for CCA and PLS regression.
    • The robustification techniques significantly improve the interpretability of model weights.