PAL: Prompting analytic learning with missing modality for multi-modal class-incremental learning

doi:10.1016/j.patcog.2026.113467

PAL: Prompting analytic learning with missing modality for multi-modal class-incremental learning

2026-03-14

0

PRE

AI

庄辉平 (Huiping Zhuang) *

H

Haizhou Li

DOI：10.1016/j.patcog.2026.113467

Original

Original request for help

Save

Abstract

En 中文

• We propose PAL – a novel exemplar-free technique to address the missing modality problem during the multi-modal CIL procedure. • PAL seamlessly incorporates modality-specific prompts into analytic learning, redefining the MMCIL with missing modality into an RLS task and resolving the intrinsic under-fitting limitations via prompt tuning. • Experimental results demonstrate that PAL outperforms recent state-of-the-art methods, showcasing its robustness in addressing the missing modality problem and its ability to maintain high classification accuracy during CIL procedures.

Keywords:

PAL

missing modality

multi-modal CIL

prompt tuning

class-incremental learning

PAL: Prompting analytic learning with missing modality for multi-modal class-incremental learning

Abstract

Journal

Organization