Enhancing multiview synergy: Robust learning by exploiting the wave loss function with consensus and complementarity...

Area of Science:

• Machine Learning
• Pattern Recognition
• Data Science

Background:

• Multiview learning (MvL) enhances models using multiple data perspectives, focusing on view-consistency and view-discrepancy.
• Existing multiview support vector machine (SVM) models primarily use the consensus principle, often neglecting complementarity and lacking robustness against noisy data.

Purpose of the Study:

• To introduce Wave-MvSVM, a novel MvL framework that integrates both consensus and complementarity principles.
• To enhance model robustness against noisy, error-prone, and view-inconsistent samples in multiview datasets.
• To leverage the unique properties of the wave loss (W-loss) function for improved learning.

Main Methods:

• The proposed Wave-MvSVM framework utilizes a novel wave loss (W-loss) function, characterized by smoothness, asymmetry, and boundedness.
• It incorporates a between-view co-regularization term for view consistency and an adaptive combination weight strategy.
• Optimization is achieved through a combination of gradient descent (GD) and alternating direction method of multipliers (ADMM).

Main Results:

• Wave-MvSVM effectively harnesses both consensus and complementarity principles, offering a more comprehensive learning process.
• The W-loss function significantly mitigates the impact of noisy and outlier data, enhancing model stability and classification calibration.
• Empirical evaluations across diverse datasets show Wave-MvSVM outperforms existing benchmark models, demonstrating superior generalization ability.

Conclusions:

• Wave-MvSVM presents a robust and efficient solution for multiview learning challenges, effectively addressing limitations of previous approaches.
• The model's efficacy is further validated through its implementation on a Schizophrenia dataset, showcasing real-world applicability.
• Theoretical generalization is supported by Rademacher complexity analysis.