Multiview Regularized Discriminant Canonical Correlation Analysis: Sequential Extraction of Relevant Features From...

Area of Science:

• Multiview data analysis
• Statistical learning
• Bioinformatics

Background:

• Extracting significant features from high-dimensional multiview data is crucial for real-life applications.
• Traditional multiset canonical correlation analysis (MCCA) methods are computationally intensive and face singularity issues with limited samples.
• Existing MCCA feature extraction algorithms are generally unsupervised.

Purpose of the Study:

• To develop a computationally efficient, supervised feature extraction algorithm for multiview data integration.
• To address the limitations of existing MCCA methods in terms of computational cost and unsupervised nature.
• To effectively handle high-dimensional data and the "curse of dimensionality".

Main Methods:

• Introduced a novel block matrix representation to reduce computational complexity in MCCA.
• Developed an analytical formulation for efficient computation of multiset canonical variables.
• Employed a supervised ridge regression optimization technique for feature extraction.

Main Results:

• The proposed method significantly reduces computational cost for feature extraction.
• The algorithm effectively handles high-dimensional data and the "curse of dimensionality".
• Demonstrated effectiveness on benchmark and real-life cancer datasets, outperforming existing methods.

Conclusions:

• The new supervised algorithm offers an efficient and effective solution for multiview data integration and feature extraction.
• This approach enhances the applicability of MCCA in big data scenarios, particularly in fields like cancer research.
• The method provides a foundation for sequential generation of relevant features with reduced computational burden.