REMIND: Rethinking Medical High-Modality Learning under Missingness--A Long-Tailed Distribution Perspective

Researchers propose REMIND, a framework for medical multi-modal AI learning that addresses the challenge of missing data across multiple modalities. The solution uses a Mixture-of-Experts architecture to handle long-tail distributions of modality combinations and shows superior performance on real-world medical datasets.

• →Medical AI systems struggle with missing data when integrating multiple modalities, creating exponentially growing combinations with long-tail distributions.
• →The REMIND framework uses group-specialized Mixture-of-Experts architecture to learn fusion functions for arbitrary modality combinations.
• →The approach addresses gradient inconsistency and concept shifts that cause underperformance in tail modality groups.
• →Extensive testing on real-world medical datasets shows consistent outperformance over existing state-of-the-art methods.
• →The framework demonstrates robust generalization across various medical multi-modal learning applications.