Federated class-incremental learning with prompting

Xin Luo

doi:10.1016/j.eswa.2025.129416

Federated class-incremental learning with prompting

2025-08-20

0

PRE

AI

DOI：10.1016/j.eswa.2025.129416

Original

Original request for help

Save

Abstract

En 中文

• Proposes FCILPT, the first prompt-based method for federated class-incremental learning. • Introduces three prompt types to prevent catastrophic forgetting in FCIL. • Aligns client prompt pools to address non-iid data distribution issues. • Eliminates need for memory buffers while maintaining privacy. • Achieves SOTA performance on three benchmark datasets.

Keywords:

FCILPT

prompt-based method

federated class-incremental learning

catastrophic forgetting

non-iid data distribution

Federated class-incremental learning with prompting

Abstract

Journal

Organization