Class-aware prototype augmentation and decoupled feature distillation for class-incremental learning

doi:10.1016/j.patcog.2025.112692

Class-aware prototype augmentation and decoupled feature distillation for class-incremental learning

2025-11-07

0

PRE

AI

DOI：10.1016/j.patcog.2025.112692

Original

Original request for help

Save

Abstract

En 中文

• Facilitating class-incremental learning without saving exemplars of old tasks. • Diverse prototype augmentation based on the distributional characteristics of different classes. • Aligning different views of the same instance in the new task to better learn the new knowledge. • Decoupling feature distillation thus reduces the interference of the magnitude term on the angle term in the distillation.

Class-aware prototype augmentation and decoupled feature distillation for class-incremental learning

Abstract

Journal

Organization