Generalized multiview margin distribution learning with margin consistency

doi:10.1016/j.eswa.2026.133109

Generalized multiview margin distribution learning with margin consistency

2026-09-29

0

PRE

AI

DOI：10.1016/j.eswa.2026.133109

Original

Original request for help

Save

Abstract

En 中文

With the improvement and development of the margin distribution theory, many margin distribution methods have been proposed in recent years, showing attractive performance. Among them, multiview large margin distribution machine (MVLDM) is the first attempt to solve the margin distribution optimization problem in multiview learning. However, MVLDM is a two-view method, which is not applicable to generalized multiview scenarios and lacks effective processing of the margin distribution inside and outside the views. To solve these problems, in this paper, we propose a generalized multiview margin distribution framework and two fast methods GMVMDM-S and GMVMDM-G in practice. We construct four multiview margin distribution modules to jointly optimize the margin mean and variance within all views, and align the margin distribution between views from global margin manifold and sample margin error levels. Additionally, we use the Rademacher complexity theory to analyze the generalization ability of the proposed methods. Extensive experimental results on benchmark datasets demonstrate the effectiveness and superiority of our methods.

Keywords:

Multiview learning

Consensus and complementarity information

Margin distribution

Multiview margin manifold

Generalized multiview margin distribution learning with margin consistency

Abstract

Journal

Organization

Cited Papers