Multiview learning with twin parametric margin SVM.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Multiview learning (MVL) utilizes diverse data perspectives for improved information extraction.
• Existing twin support vector machine-based MVL (MvTSVM) models show promise but suffer from computational complexity and limitations with non-linear data and heteroscedastic noise.
• The assumption of uniform noise in MvTSVM is particularly problematic for datasets with varying error structures.

Purpose of the Study:

• To propose a novel multiview learning model, the multiview twin parametric margin support vector machine (MvTPMSVM).
• To overcome the computational complexity and noise handling limitations of traditional MvTSVM models.
• To effectively manage heteroscedastic noise structures within training data.

Main Methods:

• The MvTPMSVM model constructs parametric margin hyperplanes for both classes.
• It regulates the impact of heteroscedastic noise by avoiding explicit matrix inversions in its dual formulation.
• The model's efficiency is improved by eliminating the need for matrix inversion computations.

Main Results:

• Extensive assessments were conducted on benchmark datasets including UCI, KEEL, synthetic, and Animals with Attributes (AwA).
• Rigorous statistical analyses confirmed the proposed MvTPMSVM model's effectiveness.
• The MvTPMSVM model demonstrated superior generalization capabilities compared to baseline models.

Conclusions:

• The MvTPMSVM model offers enhanced computational efficiency.
• It effectively addresses the challenge of heteroscedastic noise in multiview learning.
• The proposed model exhibits superior performance and generalization abilities in various learning tasks.