CAVGO: Class adaptive variance-guided gradient optimization for robust domain generalization

doi:10.1016/j.neucom.2026.132623

CAVGO: Class adaptive variance-guided gradient optimization for robust domain generalization

2026-01-05

0

PRE

AI

DOI：10.1016/j.neucom.2026.132623

Original

Original request for help

Save

Abstract

En 中文

• Analyzes SAM’s convergence bias from a domain-class perspective and enhances its robustness for domain generalization. • Proposes Variance-Guided SAM to adapt perturbation directions by minimizing domain-class loss variance for cross-domain alignment. • Introduces Class Adaptive Label Smoothing in the second SAM step to mitigate hard-sample bias and improve training stability.

CAVGO: Class adaptive variance-guided gradient optimization for robust domain generalization

Abstract

Journal

Organization