Stabilizing Instance-Conditioned Modality Fusion for Multi-Modal Entity Alignment

LiDong Yao

doi:10.1016/j.inffus.2026.104781

Stabilizing Instance-Conditioned Modality Fusion for Multi-Modal Entity Alignment

2026-09-12

0

PRE

AI

DOI：10.1016/j.inffus.2026.104781

Original

Original request for help

Save

Abstract

En 中文

• ModalityGate generates entity-level fusion scores from uni-modal representations. • DVDC adapts its penalty to control dispersion and lower weight concentration. • ICMF leads by 1.4 and 2.3 MRR points at 20% supervision. • ICMF has the fewest parameters and lowest inference latency among six methods.

Keywords:

Multi-modal entity alignment

Knowledge graph

Modality fusion

Contrastive learning

Stabilizing Instance-Conditioned Modality Fusion for Multi-Modal Entity Alignment

Abstract

Journal

Organization