Efficient and robust kernel learning with class-wise privileged information for pattern classification

doi:10.1016/j.patcog.2026.113522

Efficient and robust kernel learning with class-wise privileged information for pattern classification

2026-03-19

0

PRE

AI

DOI：10.1016/j.patcog.2026.113522

Original

Original request for help

Save

Abstract

En 中文

• CPSVM is an efficient single-stage kernel learning framework with class-wise PI. • CPSVM flexibly supports both discrete and continuous forms of privileged information. • A rescaled hinge loss is introduced to improve robustness against noisy information. • We theoretically analyze the convergence and generalization properties of CPSVM. • CPSVM achieves a favorable trade-off between predictive performance and efficiency.

Keywords:

CPSVM

privileged information

kernel learning

class-wise

robustness

Efficient and robust kernel learning with class-wise privileged information for pattern classification

Abstract

Journal

Organization