Flexible multi-view feature selection with semi-supervised label semantic alignment

Han Zhang

doi:10.1016/j.patcog.2026.113386

Flexible multi-view feature selection with semi-supervised label semantic alignment

2026-03-01

0

PRE

AI

DOI：10.1016/j.patcog.2026.113386

Original

Original request for help

Save

Abstract

En 中文

• We raise a semi-supervised learning framework containing the multi-view feature regression and multi-view graph clustering modules. The supervised and unsupervised labels are bidirectional optimized and coupled through a semantic alignment module. • We present a multi-view feature selection model to optimize the heterogeneous projection matrices, which is guided by the label semantic information learned from nonlinear multi-view data regression and discrete multi-view graph partitioning. • We design an elastic lδ-norm regularization term for multi-view feature selection, which smoothly collaborates the l2,0-norm and l2,1-norm. It could flexibly distinguish different views in multi-view scenarios by assigning the view-specific parameter δv.

Keywords:

multi-view feature selection

semi-supervised learning

label semantic alignment

graph clustering

elastic lδ-norm regularization

Flexible multi-view feature selection with semi-supervised label semantic alignment

Abstract

Journal

Organization