Multi-view Learning Via Using Statistical Invariant

Bo Xu

doi:10.1016/j.patcog.2026.114180

Multi-view Learning Via Using Statistical Invariant

2026-09-29

0

PRE

AI

DOI：10.1016/j.patcog.2026.114180

Original

Original request for help

Save

Abstract

En 中文

Multi-view learning has been widely applied in real-world scenarios. However, most existing multi-view methods rely solely on the strong mode of convergence to achieve consensus and complementary, failing to fully leverage the rich information embedded in different views. In this paper, we propose a novel learning framework that incorporates both strong and weak modes of convergence in a Hilbert space as Multi-view Learning via Using Statistical Invariant (MvLUSI). The strong mode of convergence learns view learners with least-squares probabilistic-output from the admissible function set, while the weak mode of convergence achieves consensus and complementary by constructing knowledge-based predicates with corresponding statistical invariants under the weak mode of convergence from other views. Moreover, MvLUSI dedicates a view predicate selection algorithm for identifying the beneficial predicate set for appropriate knowledge. Extensive experiments on UCI, Hd, and Corel datasets demonstrate the superiority of the proposed method.

Keywords:

The weak mode of convergence

The strong mode of convergence

Multi-view learning

Predicates

Statistical invariant

Support vector machine

Multi-view Learning Via Using Statistical Invariant

Abstract

Journal

Organization

Cited Papers