New framework uses conditional diffusion models for multimodal federated learning

Researchers have developed a new framework called CondI to address missing data in multimodal federated learning, particularly in clinical settings. This approach uses conditional diffusion models to explicitly impute unobserved data points within a modality, leveraging available multimodal context. The framework trains modality-specific extractors and joint embedding spaces, enabling models to operate on complete semantic structures and improving resilience to data incompleteness. Experiments on clinical datasets showed CondI achieved results comparable to existing state-of-the-art methods. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Improves robustness of multimodal models in clinical settings with missing data, potentially enabling wider adoption of federated learning for sensitive data.

RANK_REASON This is a research paper published on arXiv detailing a new framework for multimodal federated learning.

Read on arXiv cs.LG →

• CondI
• PTB-XL
• SLEEP-EDF
• MIMIC-IV
• arXiv

• paper
• other