Towards the Causal Complete Cause of Multi-Modal Representation Learning

Multi-Modal Learning (MML) aims to learn effective representations across modalities for accurate predictions. Existing methods typically focus on modality consistency and specificity to learn effective representations. However, from a causal perspective, they may lead to representa-* Equal contribution